Retaining Older Workers in the Danish Labour Market
Mona Larsen Department of Economics
2
Table of contents
Chapter 1. Introduction and Summary........................................................................ 7 1. Introduction .............................................................................................................. 8 2. Subjectively assessed job factors............................................................................ 12 3. Health ..................................................................................................................... 17 4. Economic incentives to leave the labour market .................................................... 21 5. Pathways to retirement ........................................................................................... 26 6. Concluding Remarks .............................................................................................. 30 7. References .............................................................................................................. 31
Chapter 2. The Effect of Job Characteristics and Job Satisfaction on Planned Retirement Age ............................................................................................................. 33 1. Introduction ............................................................................................................ 34 2. Previous findings on the determinants of retirement.............................................. 39 2.1. Job related factors............................................................................................ 39 2.2. Individual factors............................................................................................. 41 3. Empirical model ..................................................................................................... 43 4. The Danish data ...................................................................................................... 46 5. Estimation and results for Denmark ....................................................................... 49 5.1. Results for Denmark........................................................................................ 53 5.2. Marginal effects of subjectively assessed job measures for Denmark ............ 62 6. Data and results for the U.S.................................................................................... 67 6.1. Data for the U.S. .............................................................................................. 68 6.2. Results for the U.S........................................................................................... 72 7. Concluding remarks................................................................................................ 79 8. References .............................................................................................................. 82 9. Appendix ................................................................................................................ 85 9.1. Measurement of expected income when retired .............................................. 85 9.2. Measurement of social security wealth ........................................................... 92 9.3. Tables .............................................................................................................. 93
3
Chapter 3. The Impact of Health on Individual Retirement Plans: a Panel Analysis comparing Self-reported versus Diagnostic Measures.............................................. 99 1. Introduction .......................................................................................................... 100 2. Health and retirement in Denmark ....................................................................... 106 3. Empirical model ................................................................................................... 109 4. Data and descriptives............................................................................................ 114 5. Results of estimation ............................................................................................ 122 5.1. Pooled OLS analysis...................................................................................... 122 5.2 Unobserved heterogeneity .............................................................................. 126 5.3 Workers only .................................................................................................. 130 5.4 Endogeneity and measurement error .............................................................. 132 5.5 Cohort differences .......................................................................................... 135 5.6 Discontinuities in planned retirement age ...................................................... 138 5.7 Disaggregated disease conditions and diagnoses ........................................... 142 5.8 Health changes................................................................................................ 145 6. Conclusions .......................................................................................................... 148 7. References ............................................................................................................ 151 Appendix A1. Health measures ................................................................................ 153 Appendix A2. Compensation rate ............................................................................ 156 Appendix A3. Tables................................................................................................ 162
Chapter 4. An Experimental Analysis of the Effect of an Increase in Delaying Incentives in the Post Employment Wage Program on Retirement Age............... 171 1. Introduction .......................................................................................................... 172 2. Post Employment Wage (PEW, efterløn) and the changes in 1992 ..................... 176 3. Public Employee Pension (PEP, tjenestemandspension) ..................................... 179 4. Data....................................................................................................................... 181 5. Empirical models.................................................................................................. 184 6. Results .................................................................................................................. 191 6.1. Distribution of retirement age and retirement age hazards............................ 192 6.2. DD and DDD analyses .................................................................................. 194 7. Concluding remarks.............................................................................................. 202 8. References ............................................................................................................ 204
4
9. Appendix .............................................................................................................. 206
Chapter 5. Pathways to Retirement in Denmark, 1984 – 2000 .............................. 209 1. Introduction .......................................................................................................... 210 2. Pathways and retirement programs ...................................................................... 212 2.1. Pathways........................................................................................................ 212 2.2. Retirement programs ..................................................................................... 215 3. The distribution on pathways, 1984-2000 ............................................................ 220 4. Data and empirical model..................................................................................... 230 5. Results .................................................................................................................. 235 6. Conclusions .......................................................................................................... 245 7. References ............................................................................................................ 249 8. Appendix .............................................................................................................. 251
Resume (Summary in Danish)................................................................................... 253
5
6
Chapter 1 Introduction and Summary
Mona Larsen
7
1. Introduction A fundamental public policy challenge in Denmark and other OECD countries in the coming decades will be the issue of retaining older workers in the labour market.
This is because first and foremost, demographic
changes expected in the future imply that older workers will form an increasing part of the labour force. The forecast for the Danish labour force for the next 35 years is a decline in the number of people in the active population from a ratio of 4.5 to 1 to a ratio of 3 to 1, cf. Schaumann (2001).
Second, the labour market for older workers has been subject to considerable change during recent decades, see e.g. Gruber & Wise (1999) and OECD (2000). The labour force participation rate for this group has decreased over time, cf. Figure 1 below. Certainly, the labour force participation rate for women in their fifties has increased since the beginning of the 1980’s except for a period in the middle of the nineties where it decreased due to earlier departure through the Transitional Benefit Programme (TBP, overgangsydelse). I return to this scheme in Section 4. In the same period, however, the labour force participation rate for 60-66-year-old men as well as women fell considerably although a small increase has taken place since 1999 suggesting that the recent reform of the most popular early retirement scheme Post Employment Wage (PEW, efterløn) might have had some effect. I return to this scheme as well in Section 4.
8
Figure 1. Labour force participation rate by age groups, index figures (1981=100), men and women, 1981-2000. 1a) Men
1b) Women Index figures 130 120 110 100 90 80 70 60
Index figures 130 120 110 100 90 80 70 60 1981 1984 1987 1990 1993 1996 1999 2002 16-66 50-54
55-59
60-66
1981
1984
1987 1990 16-66 55-59
1993 1996
1999 2002 50-54 60-66
Source: Statbank Denmark.1
In addition, the unemployment rate for women aged 55 or above in particular but also for men of the same age has been high since the late nineties compared to the labour force in general, cf. Figure 2 below. Especially in the case of workers in the age group 55-59 years, this difference resulted from the fact that they have longer unemployment periods, cf. Figure 3 below. This suggests that it is more difficult for older workers to return to employment after an unemployment period. Part of the explanation behind a lack of demand for older workers might be lack of qualifications. Compared to 25-64-year-old labour force participants in general, female workers aged 55 or above and male workers aged 60 or above have a lower level of education, cf. Figure 4 below.
1
StatBank Denmark, which is hosted by Statistics Denmark, contains detailed statistical information on the Danish society, see http://www.statistikbanken.dk/statbank5a/default.asp?w=1280
9
Figure 2. Unemployment rate by age groups, men and women, 1980-2002. 2a) Men
2b) Women
Per cent 25
Per cent 25
20
20
15
15
10
10
5
5
0
0 1980 1983 1986 1989 1992 1995 1998 2001 16-66
55-59
1980 1983 1986 1989 1992 1995 1998 2001
60-66
16-66
55-59
60-66
Source: Statistics Denmark (1990, 1994, 2001-2003) and own calculations.
Figure 3. Distribution of unemployment duration by age groups, men and women, 2002. 3a) Men Per cent 60 50 40 30 20 10 0 0.0010.200
3b) Women Per cent 60 50 40 30 20 10 0
0.2010.400 16-66
0.4010.600 55-59
0.6010.800 60-66
0.8011.000
0.0010.200
0.2010.400 16-66
0.4010.600 55-59
0.6010.800
0.8011.000
60-66
Source: Statbank Denmark.
Finally, pension policy has undergone many changes and new welfare schemes have been introduced. For instance, the TBP scheme was introduced in 1992 and removed four years later, the most popular early retirement scheme PEW was changed in 1992 and 1999 and the official retirement age will be lowered from 67 to 65 in 2004. I return to these changes in Section 4.
10
Figure 4. Level of education among labour force participants by age groups, men and women, 2002. 4a) Men
4b) Women
Per cent 50
Per cent 50
40
40
30
30
20
20
10
10 0
0 25-64
50-54
No q ualifying ed ucatio n
55-59
Vo catio nal training
60-64 Hig her ed ucat io n
25-64 No qualifying education
50-54
55-59
Vocational training
60-64
Higher education
Source: Statbank Denmark.
Demographic changes expected in the future and changes in the labour market for older workers imply that lack of labour supply is a future source of concern. Further, the net effect of the demographic changes is estimated to amount to 4-5 per cent of GNP. Therefore, an increasing burden on the providers is expected corresponding to a projected increase in the base tax in the range of 8-10 percentage points. Although these demographic changes are more modest than for most other OECD countries, the projected changes nonetheless are considered problematic because of the already high level of the tax/GNP ratio, the relatively high participation rate and relatively low unemployment rate, see e.g. OECD (2003), Bingley et al. (2003) and Schaumann (2001). In fact, an increase of taxation, participation or employment is not possible to a greater extent. Instead, one of the best solutions in order to diminish the problems related to both labour shortage and the increasing burden on the providers is to prolong the working life of older workers. Thereby, labour supply and the tax receipts will be enhanced, while at the same time public expenditure on social security benefits will be reduced.
11
The purpose of this PhD-dissertation is to add to the existing knowledge about the factors affecting the retirement decision in particular in order to gain better understanding of how to retain older workers in the labour market after the age of 60, which is the early retirement age in Denmark. The dissertation includes four papers treating different aspects of retirement behaviour. Pathways to retirement are mapped out. Further, both the effect of economic incentives and the impact of non-economic aspects including subjectively assessed job factors as well as health are examined. One of the motives behind focusing on non-economic factors is that retirement behaviour according to some previous studies is at least as strongly affected by preferences for leisure as financial incentives, cf. Gustman & Steinmeier (2000). In fact, improving workplace conditions might be one way to accommodate the increased preference of leisure at older ages. Also health is related to leisure in the sense that poor health can influence retirement by changing people’s perception of the utility of work versus leisure. Significant gender differences are found in the decision to retire early in Denmark as well as in other countries, see e.g. Pedersen & Smith (1996), Danø et al. (1998, 2000), Christensen & Datta Gupta (2000) and Dahl et al. (2000, 2003). Therefore, whenever possible, separate models are estimated for women as well as men in order to increase our knowledge about gender differences in retirement behaviour.
2. Subjectively assessed job factors Subjectively assessed job measures include many different aspects of working life such as job, working hours and wage satisfaction, job security and job demands and opportunities, all of which have a distribution across
12
workers. Cross-section analyses based on a representative sample of 52and 57-year-old workers show, for instance, that more than 20 per cent find it hard to satisfy job demands as regards new technology, 13 per cent find it hard to readjust to new tasks, while only 10 per cent face problems in relation to demands for further training, see Figure 5 below. Significantly more women than men find that use of new technology is a problem, while the oldest cohort makes up the majority of respondents stating that job demands requiring use of new technology and readjustment to new tasks are hard to satisfy.
Figure 5. Share of individuals that find it hard to satisfy the job demands further training, use of new technology or readjustment to new task, by gender and age, 2002. Per cent 30 25 20 15 10 5 0 Further training Men, 52
Use of new technology Men, 57
Women, 52
Readjustment to new tasks Women, 57
Source: Own calculations on a longitudinal database of elderly people. For further description of these data, see Chapter 3.
Further, half of the sample finds that their job satisfaction is diminished by stress and a tight timetable; almost 40 per cent is bothered by high speed, while about 20 per cent is disturbed by lack of influence, lack of recognition and respect and job insecurity respectively, cf. Figure 6 below. High
13
speed, stress and a tight timetable and lack of recognition and respect trouble women in particular.
Figure 6. Share of individuals that find that their job satisfaction is disturbed by high speed, stress and a tight timetable, lack of influence, lack of recognition and respect or job insecurity, by gender and age, 2002. Per cent 60 50 40 30 20 10 0 High speed
Stress and tight timetable
Men, 52
Lack of influence
Men, 57
Lack of recognition and respect
Women, 52
Job insecurity
Women, 57
Source: See Figure 5.
Figure 7. The share of individuals that are in a position to reduce number of working hours, move to a less demanding job or move to a more challenging job at their current working place, by gender and age, 2002. Per cent 50 40 30 20 10 0 Reduce number of working hours Men, 52
Source: See Figure 5.
14
Move to a less demanding job Men, 57
Women, 52
Move to a more challenging job Women, 57
One way to persuade older workers to remain in the labour market after the age of sixty might be to adjust working conditions. However, such adjustments are only possible for some older employees at their current working place, cf. Figure 7 above. For instance, 40 per cent are in a position, in which a reduction of the number of working hours is possible. Women constitute the majority of this group.
A little more than 20 per cent fill a post that enables them to move to a less demanding job. Men dominate this group. Finally, around 20 per cent have the possibility to move to a more challenging job. In this case, the youngest cohort constitutes the majority.
Usually, retirement models do not include the demand side. Part of the reason is that firm level data and information on working conditions have been in short supply, see e.g. Hakola (2003).2 However, I have access to a unique data set that includes information about a wide variety of subjectively assessed job characteristics. These data are applied in the first paper “The Effect of Job Characteristics and Job Satisfaction on Planned Retirement Age”, cf. Chapter 2. Here, the demand side is incorporated by focusing on working conditions. The purpose of the paper is to look at whether or not job characteristics and job satisfaction play a role in retaining older workers in the labour market in Denmark. The results yield important input to the ongoing debate about how to persuade older workers (50 years or
2
Some exceptions are papers by Hurd and McGarry (1993) and Friedberg (2001) in which Health and Retirement study data are used. For more information about these data, see Chapter 2.
15
more) to prolong their working life. The effect of subjectively assessed job factors on retirement planning is analysed controlling for the effect of health, education, cohabitation, financial and retirement income variables and objective job characteristics. An ordered probit model is estimated. In addition, a method for correcting for unobserved heterogeneity is applied. The Danish results are put into perspective by conducting a similar analysis based on data for the U.S. from the Health and Retirement study , as the two countries are similar in terms of background characteristics and labour force attachment of their older worker populations, yet significantly different in terms of the structure and incentives of their pension systems.
Results for Denmark suggest that subjectively assessed job measures, while not as important as for example retirement income, do play a role in retirement planning. In particular, job satisfaction, working hour satisfaction, the usual retirement age in one’s position and difficulties with satisfying job demands are important factors when individuals consider whether to retire after the early retirement age or at or before this age. However, gender differences are found in the relative importance of these factors, i.e. they are in general more important for women than for men. While removing barriers to satisfying job demands are found to be the most suitable way to persuade women to prolong their working life after the early retirement age, an increase in working hour satisfaction seems to be the most effective approach in the case of men. However, to get an appreciable effect on the retirement age, quite large changes in these factors are required. Compared to the Danish case, subjectively assessed job factors are found to be less important in the U.S. suggesting that American wage earners are more financially constrained.
16
3. Health A number of previous studies have shown that health is an important determinant of preferences for retirement. This connection between health and retirement is illustrated in Figure 8 below. Namely, while labour force participation is decreasing by age, expected remaining lifetime is decreasing as well, while the number of physician visits is increasing. This development by age is similar for men and women. However, labour force participation falls considerably from the age of 60 on. In fact, health changes do not seem to be the most important explanation of this development since no sudden changes occur around the age of 60 in either the number of physician visits or in expected remaining lifetime. However, it can be argued that these measures of health while useful for demonstrating general tendencies in overall health of the population, are only crude proxies for work incapacity, which in theory should be the factor affecting early withdrawal from the labour force.
The second paper ”The Impact of Health on Individual Retirement Plans: a Panel Analysis comparing Self-reported versus Diagnostic Measures” which is joint work with Nabanita Datta Gupta, assesses the importance of precisely measured health relative to economic factors on planned retirement behaviour, cf. Chapter 3. By using a wide array of alternative health measures including both survey self-reported measures and diagnostic measures extracted from the Danish national patient registry records, the role of subjectively versus objectively measured health is compared as a determinant of retirement planning, controlling for income, labour market, job and background characteristics. Further, the paper adds fresh evidence
17
to the debate about “justification bias”, the positive bias imparted to selfreported health measures by retired workers who cite failing health as an excuse for early withdrawal. Finally, extending the existing literature by Dwyer & Mitchell (1999) and McGarry (2003), we control for unobserved heterogeneity as well as account for endogeneity and measurement error of health in retirement, and estimate separate models for women as well as men. Our analysis is based on a panel model of the effect of health on individual retirement plans, and is estimated on a sample of older workers and retirees drawn from a Danish panel survey from 1997 to 2002 that is merged to longitudinal register data from the period 1993-2001.
Figure 8. Labour force participation rate, physician visits and calculated mean expected remaining lifetime by gender, 1993-2001. Individuals born in 1945, 1940 and 1935. a) Individuals born in 1945. Number of years/visits
LFP, per cent 100
35 30 25 20 15 10 5 0
80 60 40 20 0 48
49
50
51
52
53
54
55
56
Age
18
LFP, men
LFP, women
Visits, men
Visits, women
Remaining lifetime, men
Remaining lifetime, women
b) Individuals born in 1940. Number of years/visits
LFP, per cent 100
30 25 20 15 10 5 0
80 60 40 20 0 53
54
55
56
57
58
59
60
61
Age LFP, men
LFP, women
Visits, men
Visits, women
Remaining lifetime, men
Remaining lifetime, women
c) Individuals born in 1935. Number of years/visits
LFP, per cent 100
25
80
20
60
15
40
10
20
5
0
0 58
59
60
61
62
63
64
65
66
Age LFP, men
LFP, women
Visits, men
Visits, women
Remaining lifetime, men
Remaining lifetime, women
Sources: Statbank Denmark and own calculations on register data drawn from a longitudinal database of elderly people, which include register information about the total sample of the cohorts in question. For more information about this database, see Chapter 3.
As in the case of recent U.S. studies we find little support for the “justification bias” hypothesis and neither endogeneity nor measurement error turn out to be important sources of concern in the Danish data. Unobserved heterogeneity however, turns out to be important and estimates from ran-
19
dom effects models show that self-rated physical and mental health are important predictors of retirement planning, in fact more important than economic factors, both among men as well as among women. However, health seems to be relatively more influential in men’s retirement planning. Being in poor general health or poor mental health significantly reduces planned retirement age for both men and women. Being in poor general health reduces planned retirement age by about 1.3 years for men and 8 months for women. Being in poor mental health lowers men’s planned retirement age by about 1.2 years and women’s by about a year. Other health measures, in particular having health worse than others or having a disease for men and a reduction in working capacity for women, also lowers planned retirement age significantly. At a disaggregated level, back problems and myalgia significantly hasten male retirement, while back problems and particularly osteoporosis and depression are significant conditions triggering retirement among women. Retirement planning is in general unaffected by being hospitalised for a serious condition, except in the case of hospitalisation for heart diseases, which reduces planned retirement age for men as well as women, but only marginally so. Looking at health changes strengthens the conclusion that health is an important factor in retirement planning. In fact, health shocks seem to increase the propensity to retire earlier. However, health seems to be less important for retirement planning in Denmark compared to the U.S due to the subsidized and fully-covered health care system and the easier access to health-related exit.
20
4. Economic incentives to leave the labour market During the last decade, new welfare schemes have been introduced and pension policy has undergone many changes.
The TBP scheme, which was a sort of extension of the PEW scheme, cf. below, was introduced in 1992. Eligible persons for entry were initially 5559 years old members of unemployment insurance funds who had been unemployed for at least 12 out of the most recent 15 months. Benefits were set at 82 per cent of maximum unemployment insurance benefits and the maximum duration was until transition to PEW at the age of 60. From the beginning of 1994 the program was extended to cover the age group 50-54 years with the same labour market criteria as for the 55-59 years old group. In the beginning of 1996, the program was removed.
As mentioned, PEW is the most popular early retirement scheme in Denmark. To be entitled to PEW benefits, the worker must be 60-66 years of age (60-64 years from 2004), have been a member of an unemployment insurance fund in a certain number of years3, have been working full-time at least 52 weeks in the last three years4, and live in Denmark. In 1992, the PEW scheme was changed in order to reduce early retirement through this scheme. Namely, the incentives to delay retirement until at least the age of 63 were increased. Before 1 March 1992, PEW benefits were equivalent to unemployment benefits the first two and a half years. After two and a half 3
The number of years has been changed several times since the introduction of the scheme in 1979. 4
Before 1 January 1997, the condition was at least 26 weeks in the last three years.
21
years, PEW benefits were reduced to a maximum of 80 per cent of unemployment benefit. After 1 March 1992, the reduction in PEW benefits after two and a half years was removed for some people. That is, individuals that delayed retirement until at least age 63 received a benefit equivalent to the unemployment benefit over the whole PEW period.
In 1999 a fairly complicated reform of the PEW was enacted. The age of first eligibility was still 60, but incentives were introduced to postpone entry until the age of 62 or later. To become eligible for PEW after the reform an individual is required to have been in an unemployment insurance fund for 25 of the last 30 years compared to 20 of the last 25 years before this change. Finally, the rules were changed in 1999 to make it more attractive to continue work for a number of hours at the same time as collecting PEW.
As part of the PEW reform in 1999, it was decided to lower the official age of retirement from 67 to 65 years of age from 2004. This implies a cut in the public expenditures since PEW benefits are more costly than Old Age Pension, (OAP, folkepension).
The existence of the TBP scheme had a pronounced effect on the mean retirement age for 45-67-year-olds, cf. Figure 9 below. In fact, the mean retirement age has been relatively stable since the middle 1980’s. The level has been between 60 and 61 for men and between 59 and 60 for women. The only exception was the period in the middle 1990’s where the mean age dropped dramatically due to the existence of the TBP scheme. Men as
22
well as women retired through this scheme but the effect was most remarkable for women. While the age for men fell 1.9 years from 1993 to 1995, the fall for women was 2.4 years.
Figure 9. Mean retirement age by gender, 45-67-year-olds, 1985-1999. Age 62 61 60 59 58 57 56 1985
1987
1989
1991
1993
Men
1995
1997
1999
Women
Source: Own calculations on the basis of a 10 per cent sample of the Danish population aged 45-67 years. For further description of these data, see Chapter 5.
Figure 10. Mean gross compensation rate by gender, 45-67-year-olds, 1985-1999. Rate 0.8 0.75 0.7 0.65 0.6 0.55 0.5 1985
1987
1989
1991 Men
1993
1995
1997
1999
Women
Source: See Figure 9.
23
In general, the mean gross compensation rate has shown a downward tendency since the middle of the 1980’s, cf. Figure 10 above. The only exception is the situation in the middle of the 1990’s where a sudden rise took place. This increase suggests that part of the explanation of the popularity of the TBP scheme was a relatively high gross compensation rate for people that retired through this scheme.
The effect of the TBP scheme surpassed the effect of change of the PEW scheme in 1992 as is evident in Figure 9. I return to the effect of this change below.
As Figure 9 only includes the years up to and including 1999, the effect of reform of the PEW program in 1999 is not visible either on this figure. However, the small increase since 1999 in the labour force participation rate for individuals aged 60-66 years mentioned above might be due to these changes. Namely, the reform has so far resulted in a decline in the take up of PEW for the 60-61 years old, see Economic Council (2001). The initial impact on retirement is analysed using micro data by Jørgensen (2004), Bjørn and Larsen (2003), Quaade (2001, 2002) and Danø et al. (2000).
The TBP scheme seems to be the scheme that has had the most pronounced effect in the nineties. However, it is not straightforward to examine the effect of this scheme due to difficulties with identifying an appropriate control group. Instead, the focus of the third paper “An Experimental Analysis of the Effect of an Increase in Delaying Incentives in the Post Employment
24
Wage Program on Retirement Age” is on the changes in the PEW program in 1992, cf. Chapter 4. The purpose of this paper is to examine the effect of this change on the average retirement age. While previous Danish studies of the response to changes in social security are based on simulations, see Danø et al. (2000), Christensen and Datta Gupta (2000) and Bingley et al. (2003), an experimental analysis including difference-in-difference (DD) and triple difference-in-difference (DDD) is carried out in this paper. Experimental analysis may be useful in this regard because causal responses of retirement to policy changes can be inferred. The treatment group consists of people entitled to PEW, while people eligible for Public Employment Pension (PEP, tjenestemandspension) constitute the control group. The analyses are based on a representative two per cent sample of individuals and their spouses respectively for the period 1980-1998.
All in all, results suggest that the response of the retirement age to the change in the PEW policy in 1992 was relatively small, at least in the short run. Distributions of retirement age and retirement age hazard functions indicate that the changes did not have the intended effect. However, DDD analyses suggest that the change of the PEW policy increased the incentives to retire at age 63 and 64 as intended. But the analyses also suggest that the aggregate effect was small in the sense that transitions to retirement at age 60-62 and at age 65-66 were not affected significantly. Consequently, retirement age for people entitled to PEW did not increase on average.
25
5. Pathways to retirement Traditionally, work and retirement have been considered as distinct states, reflecting the idea of a life cycle where full-time employment in a job is followed directly at a specific and common age by entry into OAP. This view is still true for some people in the labour force. However, increasingly the transition from work to retirement can consist of many combinations of programs for early retirement, unemployment insurance benefits, sickness benefits and welfare benefits. Further, the period between full-time work and retirement at the official pension age can be bridged by part-time work or by early activation of private pension programs at an actuarial discount.
Ruhm (1990) shows that partial retirement is very widespread in the U.S. In the case of Denmark, many individuals, in particular self-employed, express a desire to retire partially at the end of their working lives, cf. Pedersen (1998), Nørregaard (1996) and Pedersen & Smith (1995). However, paradoxically, the extent of partial retirement through less working hours seems to be limited in Denmark. In fact, Nørregaard (1996) finds that only relatively few retire through part-time work, flexible working hours etc. Similarly, a survey conducted by the Danish National Institute of Social Research in 1999 shows that only 6 per cent of already retired individuals had reduced their working week gradually through part-time work. Furthermore, very few individuals have joined the part-time retirement schemes, cf. Pedersen & Smith (1995). In 1998, only 4,000 retired on the
26
part-time pension scheme while 1,600 retired on a part-time PEW scheme5, cf. Initiative Committee on Senior Policy, Ministry of Labour (1999).
Analyses of survey data confirm that the extent of shift to a part-time job from full-time employment in general is quite limited. Among 52 years old full-time employed people in 1997 only 1 per cent of men and 8 per cent of women had shifted to a part-time job in 2002, cf. Figure 11 below. The corresponding figures for 57-year-olds are 3 and 5 per cent. The fact that the share of women shifting to a part-time job is highest for the youngest cohort suggest that full-time working 57 years old women prefer to make use of the possibility to retire early while a woman that wants to retire gradually to a greater extent reduces her working hours at as early as in her middle fifties. Figure 11. Distribution of full-time workers in 1997 on full- and part-time work in 2002, a) all, b) non-retired. Individuals born in 1940 and 1945 by gender. 11a) All
11b) Non-retired
100%
100%
80%
80%
60%
60%
40%
40%
20%
20% 0%
0% Men, born Men, born Women, 1945 1940 born 1945 Full-time
Part-time
Women, born 1940
Men, born Men, born 1945 1940
Retired
Full-time
Women, born 1945
Women, born 1940
Part-time
Source: See Figure 5. Note: Part-time is defined as less than 30 hours per week.
5
The part-time PEW scheme was abolished 1 July 1999.
27
However, looking only at those remaining in the labour force in 2002 some partial retirement is found to take place among women in particular but among men as well, prolonging their working life after the age of 60. Namely, 14 and 6 per cent of 57-year-old full-time employed women and men respectively worked part-time five years later. For the younger cohort, figures were 1 and 8 per cent of men and women, respectively. That is, partial retirement takes place among men as well as women in their sixties, while only women engage in partial retirement when they are in their fifties.
The purpose of the fourth paper “Pathways to retirement in Denmark, 1984-2000”, which is joint work with Peder J. Pedersen, is to describe the multitude of pathways to exit from the labour force in an institutional setting with a very broad range of available programs, cf. Chapter 5. That is, pathways such as entry from a job, from a position as self-employed or from different types of social security benefit or combinations of these through which entry into either early retirement or OAP have occurred. We focus on the development since the middle of the 1980’s. For the moment, not much is known about the relative magnitude of these pathways. This type of knowledge is important in research and policy discussions about reforms of welfare and retirement programs. Furthermore, policies to promote higher participation and employment rates among the seniors in the labour force also depend on more solid evidence about the withdrawal process from the labour force.
The data set applied in this paper is a 10 per cent panel sample of the Danish population 45-67 years old followed during the period from 1984 and 28
on. Multinomial logit (MNL) analyses of pathways from work to retirement are estimated for selected years. The MNL model is a choice model but since the pathways in focus is a mixture of choices and risks, in this context the MNL approach is applied to conduct multivariate analyses of the characteristics of individuals that retire through different pathways compared to individuals that remain in the labour force.
We find that the transition from work to retirement is complex and far from a conventional idea of exit typically occurring from a job at the official pension age. Eight pathways from work to normal OAP or to an early retirement program are identified, which seem to fall into three groups. One group is transitions directly from employment that covers three out of four of all transitions in the sample period. Another group is the pathways dominated by UIB covering 20 per cent. The remaining 5 per cent are pathways dominated by benefits reflecting a low attachment to the labour force in the period prior to transition to retirement. The relative magnitude of these aggregated pathways is affected by the cyclical profiles over time and the temporary opening of the TBP in the middle of 1990’s. Across all three pathways, the PEW destination becomes increasingly important over time, while a very low incidence of entry directly from OAP is observed. Overall, availability and/or generosity of retirement programs seem to be very important for retirement through the employment and UIB dominated pathways, while individual background factors are of minor importance. For retirement through “other pathways”, however, personal characteristics seem to be at least as important as retirement programs.
29
6. Concluding Remarks It is important to add to our knowledge about factors affecting the retirement decision because demographic changes expected in the future imply that older workers will form an increasing part of the labour force. Further, the labour market for older workers has been subject to considerable change during recent decades. These changes imply that a shortage of labour supply and an increasing burden on the providers are future sources of concern. One of the best solutions to minimize these problems is to prolong the working life of older workers.
This PhD-dissertation deals with non-economic and economic aspects of retirement behaviour as well as pathways to retirement. The results suggest that it is important to look at the role of subjectively assessed job factors, health, changes in social security benefits and pathways in order to enhance our understanding of how to retain older workers in the labour market after the age of 60. Further, gender differences are found in the relative importance of factors affecting the retirement decision which suggest that different approaches will have to be applied in order to prolong the working life of women as well as men.
30
7. References Bingley, P., Datta Gupta, N. & Pedersen, P.J. (2003) The Impact of Incentives on Retirement in Denmark, forthcoming in Social Security and Retirement Around the World: Microestimation, (eds.) Jonathan Wise and David Gruber, NBER, 2003. Bjørn, N.H. & Larsen, M. (2003) Delay of Retirement (In Danish). Social forskning, 2003: 2. Copenhagen. Christensen, B.J. & Datta Gupta, N. (2000) The Effect of a Pension Reform on the Retirement of Danish Married Couples (in Danish). Nationaløkonomisk Tidskrift, 138, pp. 222-242. Dahl, S.Å., Nilsen, Ø.A. & Vaage, K. (2000) Work or retirement? Exit routes for Norwegian elderly. Applied Economics, 32, pp. 1865-1876. Dahl, S.Å., Nilsen, Ø.A. & Vaage, K. (2003) Gender differences in Early Retirement Behaviour, European Sociological Review, Vol. 19, No. 2, 179-198. Danø, A.M., Ejrnæs, M. & Husted, L. (1998) Gender Differences in Retirement Behaviour, Institute of Local Government Studies - Denmark, Copenhagen. Danø, A.M., Ejrnæs, M. & Husted, L. (2000) How is the Retirement Age Affected by the Reform of the Post Employment Wage program? (in Danish), Nationaløkonomisk Tidsskrift, 138, pp.205-221. Dwyer, D.S. & Mitchell, O.S. (1999): Health problems as determinants of retirement: Are self-rated measures endogenous? Journal of Health Economics 18 (1999), p. 173-193. Economic Council (2001) Danish Economy, Spring 2001. Copenhagen. Friedberg, L. (2001) The Impact of Technological Change on Older Workers: Evidence from Data on Computer use, National Bureau of Economic Research, Working Paper No. 8297. Gruber, J. & Wise, D.A. (1999) Social Security and Retirement around the World, The National Bureau of Economic Research, Chicago. Gustman, A. and Steinmeier, T.L. (2000) Retirement in Dual-Career Families: A Structural Model, Journal of Labor Economics, val 18, no. 3, p. 503-545. Hakola, T. (2003) Alternative Approaches to Model Withdrawals from the Labour Market - A Literature Review, Working Paper Series, Department of Economics, Uppsala University, No. 2003:4. Hurd, M. & McGarry, K. (1993) The Relationship Between Job Characteristics and Retirement, National Bureau of Economic Research, Working Paper, No. 4558. Initiative Committee of Senior Policy, Ministry of Labour (1999) The Seniors and the Labour Market now and in the Future (In Danish), Copenhagen. Jørgensen, M.S. (2004) The Danes postpone retirement from the labour market a bit (in Danish). Social forskning, 2004:1. Copenhagen.
31
McGarry, K. (2002): Health and Retirement: Do Changes in Health Affect Retirement Expectations? NBER Working Paper 9317. Nørregaard, C. (1996) Work and Retirement in the Nineties – and Retired Persons in the Future (in Danish). Copenhagen: Socialforskningsinstituttet. OECD (2000) Reforms for an Ageing Society, Social Issues, Paris. OECD (2003) Employment Outlook. Paris. Pedersen, P.J. (1998) The Elderly and the Labour Market, in Smith, N. (ed) Work, Work Incentives and Unemployment (in Danish), Aarhus., pp. 151-178. Pedersen, P.J. & Smith, N. (1995) The Retirement Decision, in Mogensen, G.V. (ed) Work Incentives in the Danish Welfare State, New Empirical Evidence, The Rockwool Foundation Research Unit, Aarhus. Pedersen, P.J. & Smith, N. (1996) A Duration Analysis of the Decision to Retire Early, in Wadensjö, E. (ed.) The Nordic Labour Markets in the 1990's, Amsterdam, pp. 31-68. Quaade, T. (2001) Retirement from the Labour Market (In Danish). The Danish National Institute of Social Research. Report 01: 7. Copenhagen. Quaade, T. (2002) Reform of the Post Employment Wage Programme with Limited Effect (in Danish), in: Socialforskningsinstituttet, Social forskning 2002:3, Copenhagen, 10-11. Ruhm, C.J. (1990) Bridge Jobs and Partial Retirement. Journal of Labor Economics, vol. 8, no. 4. Schaumann, A. (2001) The Aging Society. Demography - Expenditure Pressure - What can be done? The Danish Board of Technology, digital document: http://www.tekno.dk/ Statistics Denmark (1990, 1994, 2001) Statistical Yearbook, Copenhagen. Statistics Denmark (2002, 2003) Statistic Information, Labour market 2002:5 and 2003:6, Copenhagen.
32
Chapter 2 The Effect of Job Characteristics and Job Satisfaction on Planned Retirement Age Mona Larsen‡ JEL Codes: J14, J26, J28. Abstract This paper looks at whether or not job characteristics and job satisfaction play a role in retaining older workers in the labour market in Denmark. To answer this question, an ordered probit model of the determinants of planned retirement age is estimated. In addition, a method for correcting for unobserved heterogeneity is applied. I use data from a unique Danish survey that is merged with longitudinal register data. To put the results for Denmark into perspective, a parallel analysis is carried out on data for the U.S. For this purpose, data from the Health and Retirement Study is applied. Results for Denmark suggest that subjectively assessed job measures, while not as important as for example retirement income, do play a role in retirement planning. In particular, job satisfaction, working hour satisfaction, the usual retirement age in one’s position and difficulties with satisfying job demands are important factors when individuals consider whether to retire after the early retirement age or at or before this age. However, gender differences are found in the relative importance of these factors, i.e. they are in general more important for women than for men. While removing barriers to satisfying job demands are found to be the most suitable way to persuade women to prolong their working life after the early retirement age, an increase in working hour satisfaction seems be the most effective approach in the case of men. However, to get an appreciable effect on the retirement age, quite large changes in these factors are required. Compared to the Danish case, subjectively assessed job factors are found to be less important in the U.S. suggesting that American wage earners are more financially constrained.
‡
Aarhus School of Business, Department of Economics, The Danish National Institute of Social Research and
Graduate School for Integration, Production and Welfare. This work is part of the research of the Graduate School for Integration, Production and Welfare. Financial support from the Danish Social Science Research Council is gratefully acknowledged. This work benefited from comments by Peder Pedersen, Paul Bingley, Michael Rosholm, Torben Tranæs and Nabanita Datta Gupta. Thanks also to the participants at the 4th International Research Conference on Social Security 2003 in Antwerp and the PhD Workshop at the Graduate School for Integration, Production and Welfare 2003 in Odense. Finally, thanks to Patricia StClair for very helpful assistance with handling of RAND HRS data. All remaining errors are my own.
33
1. Introduction The most popular early retirement scheme in Denmark is the Post Employment Wage (PEW) or efterløn. In 2001, more than 40 per cent of the 60-66-year-olds received PEW benefits, see Department of Unemployment Insurance (2001). While early retirement schemes such as the PEW are financially attractive due to their relative generosity, see e.g. Gruber and Wise (1999) and Blöndal and Scarpetta (1998), some previous studies have shown that retirement behaviour is at least as strongly affected by preferences for leisure as financial incentives, cf. Gustman and Steinmeier (2000). In fact, for the majority of people entitled to PEW at age 606 financial circumstances are apparently not the main reason for prolongation of working life after this age. Job related factors such as "is happy at work" and "work is an important part of life" seem to be more important, see Jørgensen (1997). Consequently, since rather large changes in economic policy variables seem to be required in order to elicit substantial changes in retirement ages, see e.g. Mitchell and Fields (1984) and Leonesio (1990), another way to reverse the tendency towards earlier withdrawal from the labour market might be to improve workplace conditions in order to accommodate the increased preference of leisure at older ages. Therefore, the purpose of my study is to look at the role of subjectively assessed job measures in retaining older workers in the Danish labour market through their effect on planned retirement age. The results from such an analysis are expected to yield important input to the ongoing debate about how to persuade older workers (50 years or more) to prolong their working life.
6
The conditions of entitlement are outlined in Section 9.1 in the Appendix.
34
Labour market mobility is typically studied in terms of objective job measures such as wages, hours of work and tenure. Subjectively assessed job measures are normally omitted either because these variables are considered to be unreliable or because of lack of suitable data. However, Clark (2001) finds that subjectively assessed job characteristics are powerful predictors of workers' labour market behaviour. In fact, he states that workers do not seem to misrepresent either their preferences or their job satisfaction in cross section studies. Further, by integrating findings on the meaningfulness of answers to survey questions into a measurement-error framework, Bertrand and Mullainathan (2001) conclude that subjective variables can be useful as control variables.7 Earlier studies carried out on Danish data included little information about the relationship between retirement and job characteristics in general and no information at all about subjectively assessed job measures, see e.g. Christensen & Datta Gupta (2000), Danø et al. (1998) and Pedersen & Smith (1995, 1996). I have access to a unique data set that includes information about a wide variety of subjectively assessed job characteristics such as job demands, job satisfaction and the usual retirement age in a given position. Further details about the data set appear in Section 4.
A potential problem related to subjectively assessed job measures could be that such measures are simply capturing strong preferences for working.
7
Certainly, Bertrand and Mullainathan (2001) also state that care must be taken in interpreting answers to survey questions since the estimated coefficient does not only capture the effect of attitude but also the effect of other variables that influence how the attitude is self-reported. At the same time, however, they state that “this is closely related to the causality problem that we often encounter, even with perfectly measured variables”, cf. Bertrand and Mullainathan (2001, p. 69).
35
However, as an attempt to control for this, the individual effect retrieved from a fixed effects wage estimation is added as an additional explanatory variable in the model of planned retirement in order to capture unobserved heterogeneity in tastes for work.
Finally, another source of concern regarding the validity of subjectively assessed job measures is that employers can potentially compensate workers for poor working conditions with higher pay. However, the correlation between individual earnings and working conditions that might be considered as poor, such as having a physical demanding job, being stressed and finding it hard to satisfy job demands respectively is -0.33, 0.02 and -0.06 for men and -0.20, 0.08 and -0.01 for women. That is, the correlation is either found to be negative or only slightly positive. In fact, a negative correlation is found between working hour satisfaction and average earnings, cf. Figure 1 below. Again, however, the correlation is rather low (-0.09 for men and -0.11 for women). All in all, compensating wage differentials do not seem to be of major importance in these data.
The demographic changes expected in Denmark and internationally in the next few decades imply that older workers will form an increasing part of the population. The projection for Denmark for the next 35 years is a decline in the number of people in the active population relative to people aged 65 and above, from a ratio of 4.5 to l to a ratio of 3 to 1, see Schaumann (2001).
36
Figure 1. Log earnings (average) by working hour satisfaction score, men and women. Log earnings (average) 12.6 12.5 12.4 12.3 12.2 12.1 12.0 1
2
3
4
5
Working hour satisfaction score M en
Women
Furthermore, the labour market for aging workers has been subject to considerable change during recent decades; see e.g. Gruber & Wise (1999) and OECD (2000). In Denmark, the unemployment rate of older people became high in the late nineties compared to the average. The average retirement age fell in the period starting from the early eighties to the mid-nineties; see Ministry of Finance (2001). Pension policy has undergone many changes and at the same time new welfare schemes have been introduced. For example, the transitional benefit programme8 introduced in the mid-nineties encouraged earlier departure for long-term unemployed and in fact a lot of people used this possibility. 5 per cent of the 50-54-year-olds and 9 per cent of the 55-59-year-olds received the transitional benefit in 1996. More recently, the latest changes in the PEW programme9 have been designed to persuade workers to prolong their working life. Before these changes, most
8
The transitional benefit (overgangsydelse) was available in the period 1992 to 1996 for people 55-59 years old (from 1994 also 50-54-year-olds) who were a member of a unemployment insurance fund and had been unemployed for at least 12 out of the last 15 months. 9
These changes are outlined in Section 9.1 in the Appendix.
37
people retired when they were in their early sixties. However, at the same time it was decided to lower the official age of retirement from 67 to 65 years of age from 2004. This implies a cut in the public expenditures since PEW benefits are more costly than Old Age Pension (OAP).10 Moreover, today the PEW is only partly financed as a pay-as-you-go system. That is, it is moving toward a funded system.
The changing labour market for aging workers and the demographic changes expected have resulted in increased attention to labour supply and employment conditions of the older workforce. Furthermore, future financing problems of the welfare state are a source of concern. The demographic changes expected imply an increasing burden on the providers. The net effect of this increase is estimated to amount to 4-5 per cent of GNP, see Schaumann (2001). Consequently, it becomes increasingly important to add to our knowledge about the factors affecting the retirement decision, including the relative importance of subjectively assessed job measures to objective characteristics.
At the same time, to put the Danish results in international perspective, a parallel analysis is carried out on U.S. data, as the two countries are similar in terms of background characteristics and labour force attachment of their older worker populations, yet significantly different in terms of the structure and incentives of their pension systems. While Denmark is characterized by a universal, guaranteed old-age pension that is independent of labour market history, pension benefits in the U.S. are tied to a worker’s 10
See Section 9.1 in the Appendix for an outlining of the OAP scheme.
38
earnings record. That is, the incentives to retire early are typically lower for workers who have not built up an adequate earnings profile. Thus, I would expect that the attributes of the job played a smaller role in the U.S. where workers are typically more financially constrained. The U.S. analysis is carried out on data from the Health and Retirement study.
The remainder of the paper is organized as follows. Some relevant results of earlier empirical studies are presented in Section 2 and the empirical model is expounded in Section 3. Details of the applied Danish data are introduced in Section 4. Estimations and results for Denmark are presented in Section 5. The U.S. data and results are presented in Section 6 whereas Section 7 contains the conclusions.
2. Previous findings on the determinants of retirement The main purpose of my empirical model is to examine how, if at all, subjectively assessed job characteristics affect the early retirement decision. In addition, a wide variety of other explanatory variables are included. The variables are split up into job related and individual factors. Results of earlier empirical studies are presented, in particular studies carried out on Danish individual-level data.
2.1. Job related factors Studies carried out on Danish data have not included any information about the relationship between subjectively assessed job measures and retirement. However, Hurd & McGarry (1993) study this particular topic for the U.S. 39
They find that the physical and mental demands of the job seem to have only a modest influence on prospective retirement, whereas the usual retirement age in a particular type of work is important. Furthermore, the ability to move to a less demanding job with the same employer or to change hours of work increases the prospective retirement age. The latter result is also found in Panis et al. (2002).
Lazear (1979) offers an explanation of the use of mandatory-retirement clauses in labour contracts. He shows that this date of retirement in many cases is well scheduled. The date is negotiated in advance and is part of an optimal contractual arrangement that insures that firm-worker separation occurs at the appropriate time. At this moment, the value of the workers' marginal product is equal to their reservation wage. However, wage exceeds reservation rate at this point because the payment scheme produces a superior lifetime profile that workers prefer ex ante. Namely, it is shown that a payment scheme in which workers and firms agree to a long-term wage stream that pays workers less than the value of their marginal product when young and more than the value of their marginal product when old pays both parties.
Lazear (1979) also looks at more traditional job variables. He finds that job tenure has a negative effect on the retirement age. Workers with fewer years of job tenure have more recently made a "contract" with the employer than those with longer tenure. This new "contract" has been negotiated either because the value of the workers' leisure did not rise as rapidly as expected or because they ended up being more productive in the labour force
40
than expected. Lazear (1979) argues that in either case, it is optimal for the worker to have a later retirement date in the new "contract".
Other labour market attributes also affect the retirement age; see e.g. Danø et al. (1998), Pedersen (1998) and Pedersen & Smith (1995, 1996). Namely, spells of unemployment lower the planned and the actual retirement age. This might reflect a discouraged worker effect: If you do not believe that you can get a job, you might just as well retire. At the same time, work experience also has a positive effect on early retirement. That is, wage earners such as unskilled workers, who began their work life when they were quite young, might have had a physical demanding job that make them wish to retire early. In fact, the planned retirement age for unskilled workers is lower than for other groups of employed. On the contrary, the self-employed have a relatively high retirement age. For instance, individuals employed within agriculture etc. seem to have a rather low propensity to retire early.
2.2. Individual factors Both studies carried out in Denmark and internationally show that individual factors affect the retirement age, see for instance Bingley et al. (2003), Christensen & Datta Gupta (2000), Pozzebon & Mitchell (1989), Burtless & Moffitt (1985) and Hurd & Boskin (1984). Gender differences in retirement behaviour seem to be distinctive both in Denmark and abroad. Apparently, women retire or expect to retire earlier than men. Furthermore, factors affecting the retirement decision seem to differ between men and women. This might reflect a difference in preferences for work and leisure. 41
Women might attach greater importance than men to spend time in the home and within the family.
Most of the studies carried out on Danish data find a positive connection between education and retirement age. In addition, education has a greater impact on the retirement decision for women than for men. Bad health affects early retirement positively.11 There might be a relation between the effect of education and health. The higher the education, the less physical demanding the job and with it, the lower the risk of bad health.
Single individuals have a higher expected retirement age than individuals that are married or cohabiting. In particular, this result seems to apply to women. The result for single women might reflect that these women are self-supporting and need to work. Another explanation might be that these women have a greater preference for work than married or cohabiting women while the latter attach more importance to family life. The presence of dependent children has no discernible impact.
Individuals living outside the Copenhagen Metropolitan area withdraw or expect to withdraw from the labour market earlier than employees living within the capital area. Furthermore, home-ownership has a positive effect
11
While Christensen & Datta Gupta (2000) and Pedersen & Smith (1996) show that bad health has an overall positive impact on early retirement, Danø et al. (1998) find that bad health seems to hasten retirement for men, while the effect is insignificant for women.
42
on the retirement age. The effect of home-ownership might reflect an income effect.
Retirement age increases with labour market income, overall disposable income and wealth. On the contrary, there is a positive connection between retirement probabilities and Social Security wealth.12 In general, the provisions of the social security system in Denmark seem to play an important role in determining retirement behaviour. In particular, forward-looking incentive measures such as peak value or option value have the expected significantly negative effect on the probability of retirement (Bingley et al. forthcoming 2003). Apparently, the financial incentives to retire early are strong among low-wage earners whereas for high wage earners, the replacement rate is typically low, creating financial incentives to continue working in the 60s.
3. Empirical model The empirical specification I adopt for my analysis is an ordered probit model of planned retirement age. Thus, it presupposes a static model of retirement where people decide once and for all at what age they retire and do not change this decision afterwards. An example is the life-cycle model of consumption and labour supply that has been applied to the retirement
12
Moreover, Pedersen & Smith (1995) do not find any correlation between the expected retirement age and the expected level of compensation when retired, whereas Pedersen & Smith (1996) find a significant positive impact of the economic compensation rate on the propensity to exit to non-health related early retirement.
43
decision by Burtless & Moffitt (1985). Although more recently dynamic models of retirement have been developed (Gustman & Steinmeier (1986), Rust (1989), Stock & Wise (1989) and Berkovec & Stern (1991)), as only one observation of the core variables relating to job attributes and planned retirement age are available in the survey, a static approach is necessitated.
The starting point in the analysis is a latent regression relating the individual’s desired work propensity in old age, R*, to its determinants:
R* = β ′J + γ ′Z + ε
where ε ~ N (0,1)
J is a vector of subjectively assessed job measures, β is the coefficient vector on the job measures, Z is a vector of other variables affecting work attachment, e.g. tenure, labour market income and health and γ the parameter vector on these measures. However, R* is unobserved. What we observe is planned retirement age, R. While in principle, R varies continuously from the individual’s age at the time of the survey (52 or 57) and up, in practice, it is necessary to group it in different categories according to pension program rules that lead to 6 distinct retirement regimes (see Section 4 below).
Thus, we observe, R = 0 if R*
0.
= 1 if 0 < R*
µ1
= 2 if µ1 < R* µ2
44
. . = 5 if µ4
R*.
As the outcome variable R is discrete, either a probit or multinomial logit (MNL) specification could have been adopted. However, these models would fail to take into account the ordinal nature of the dependent variable i.e. each successive category represents an increasing retirement age/work propensity category. For example, the MNL would treat each age category as distinct but of equal rank. The µ variables are unknown threshold parameters that are estimated together with β and γ. In this way, the ordered probit specification is flexible and allows for the determinants of work intensity to vary by age category by way of the threshold parameters.13
The ordered probit likelihood is given as follows:
5
L = ∏ [Pr( Ri = j )]
Nj
j =0
4
= [ F (− X i ) N 0 ][∏ [ F ( µ j − X i ) − F ( µ j −1 − X i )] j ][1 − F ( µ 4 − X i )]N 5 N
j =1
µj are the threshold parameters to be estimated with µ0 = 0. N is the total number of observations. Of the N observations, N0 plan to retire at R = 0, N1 plan to retire at R = 1, N2 plan to retire at R = 2 and so on. That is, N =
13
Another possibility would be to use quantile regression. However the sample being small, there are not enough degrees of freedom to implement such a model and therefore, the ordered probit model which is similar in approach, is more adaptable for this purpose.
45
5
∑N
j
. F(·) is the cumulative probability distribution function of the error
j =0
terms and X i = β ´ J i + γ ´Z i .
4. The Danish data The Danish data are obtained from an existing longitudinal database of elderly people (Ældredatabasen). This database consists of two parts: survey data and register data created for administrative purpose. The survey was conducted in 1997 face-to-face in the homes of the respondents born every fifth year from 1920 to 1945. Namely, a total of six cohorts were interviewed. The questionnaire consists of a number of questions about withdrawal from the labour market, job characteristics, family, health, residence and so on. Details about both planned retirement age and a wide variety of subjectively assessed job characteristics make the survey unique. For definition of these job characteristics, see Table A.1. The survey information is combined with longitudinal data from registers compiled by Statistics Denmark from a 5-year-period (1993-1997). In particular, I can obtain information about education, income, work experience and unemployment from this source. Furthermore, I can also obtain information about spouses.
In this study, information from respondents born in 1940 and 1945 is used. That is, respondents aged approximately 52 and 57 years at the time of the interview. Two cohorts are included because I need some variation in pension program rules to identify the model, see e.g. Krueger & Pischke (1992). I focus on wage earners, including individuals that are temporary out of employment in consequence of unemployment, sickness or leave. 46
Hence, self-employed and people outside the labour market are not included.
In the analysis, the dependent variable is the individuals' planned retirement age. In fact, information on actual retirement age is also available but survey information on working conditions is not available for already retired individuals. However, an attractive feature of looking at planned retirement age is that this age is expected to better describe how employees actually behave. Obviously, this age is not necessarily coincident with the actual retirement age. However, since the purpose of this paper is to identify factors that affect wage earners' planned retirement age at a given moment in time, correspondence between planned and actual retirement age is not needed. On the contrary, the analysis only makes sense if the planned and the actual retirement age are strongly positively correlated. This seems to be the fact, see e.g. Disney & Tanner (1999). Their evidence, which is based on two waves of the UK Retirement Survey from 1988-89 and 1994, suggests that the expected age corresponds to a measure of central tendency. The measure that corresponds most closely is the median. A similar result is found in Bernheim (1989). The closer the expected retirement age is to an individual’s age at the first wave, the more likely he or she is to retire later than expected and the higher the mean actual retirement age relative to the expected, cf. Disney and Tanner (1999). Further, the highest proportions of individuals retiring after they expect are unsurprisingly individuals expecting to retire relatively early. Register information after 1997 could have useful in order to compare retirement plans with realisations in the Danish data applied in this paper. Unfortunately, by construction,
47
Ældredatabasen only contained register information up to and including 1997 at the time the analyses were conducted.
The dependent variable14 is grouped to reflect different pension policy rules at different ages. At the time of the interview, the early and the normal retirement age in Denmark was 60 and 67, respectively. Furthermore, if people waited until the age of 63 to join the PEW, they received a higher benefit, cf. Section 9.1 in the Appendix. However, the possibility to form categories that reflect the pension policy is restricted by the number of observations. Therefore, I distinguish between six categories of planned retirement age: below 60, 60, 61-62, 63, 64-66 and 67 and above. Individuals that answered either "don't know" or "as long as possible" are removed from the sample. I return to this sample selection issue below.
Withdrawal from the labour market around the early retirement age has more or less become the norm, cf. Table 1 below. Within the sample used more than half answered 60 when asked about their planned retirement age. This feature is salient among women, in particular. Furthermore, it is worth noting that only 5 per cent of these respondents (fewer women than men) plan to defer retirement until the normal retirement age or later. As mentioned above, the normal age was 67 at the time of the interview. That is,
14
The creation of the dependent variable is based on the following question: "At what age would you prefer to stop working?" The respondent could either mention some age or answer "don't know" or "as long as possible". If some age was mentioned, the respondent was then asked: "Do you expect to be able to work until this age?” If the answer was "no" or "don't know", the respondent was finally asked: "When do you expect to stop?" That is, the possible answers are some age, "as long as possible" and "don't know".
48
the official age of retirement did not constitute a usual retirement age for the work force in general. As expected, men have a higher planned retirement age men than women.
Table 1. Planned retirement age, Denmark. Men
Women
Total
6
5
5
60
54
63
58
61-62
12
15
13
63
13
10
11
64-66
10
5
8
7 67
6
3
5
Total
100
100
100
Number of observations
675
574
1249
< 60
Several explanatory variables are included in the model. For definition and descriptive statistics, see Table A.2 in the Appendix.
5. Estimation and results for Denmark An ordered probit model of the determinants of planned retirement age is estimated, cf. Section 3. As mentioned above, earlier empirical studies have shown that factors that affect the retirement decision differ between men and women. Therefore, the analyses are conducted for men and women separately.
49
There is a potential for selection bias because half of the sample are not included in the analysis, cf. Table 2 below.15
Table 2. Individuals sorted out of the sample, Denmark. Break down by causes. Number of individuals and per cent of total sample. Number
Per cent
2656
100
Self-employed/ assisting spouse
338
13
People outside the labour force
484
18
Variables with missing values
306
12
Planned retirement age: "as long as possible"
108
4
Planned retirement age: "don't know"
171
6
1407
53
Total sample
Total of individuals sorted out of the sample
Self-employed and assisting spouses are not included in the analysis because they are likely to have different retirement patterns. Furthermore, some of the subjectively assessed job measures are not relevant for this group and therefore, they were not asked about these factors in the survey. Similarly, people outside the labour force were not asked these questions and therefore, it is not possible to include them in the analysis. To be able to compare different models, nested models are required, cf. Section 5.1. Therefore, it has been necessary to exclude individuals with missing values
15
A correction for potential selection bias has been conducted using a Heckman twostep procedure. The included explanatory variables were age, health, education, experience, number of children, single, partner on the labour market, self-employed or assisting spouse most of the life and current or last industry and the two instruments unemployed the last four years and duration of unemployment the last four years. Several of these variables were significant but neither for men nor for women was the Mills ratio significant and including it did not change the results of the ordered probit analyses either. Therefore, these results are not reported in tables but are available on request.
50
for one or more variables from the analysis. People that "don't know" when to retire are sorted out because it is not possible in a reasonable way to include this group in the range of an ordered dependent variable. At first sight, people that want to continue to work "as long as possible" are of particular interest in relation to the topic of this paper. However, it is also difficult to place this group within the range of the ordered dependent variable. The most obvious thing to do seems to be to pool this group with people that answered 67 years and above. However, I hesitate to do so, as individuals citing “as long as possible” turn out to be a very heterogeneous group whose characteristics do not in general resemble those who cite 67 years and above when queried. A possible explanation is that “as long as possible” is ambiguous in the sense that the motives behind given this answer are mixed. In fact, the motivation could be health problems, financial reasons as well as enjoyment of working life. Therefore, people who want to continue to work "as long as possible" are also excluded from the analysis.16
The presence of selection effects as to who is found in the different types of jobs could be the explanation behind why physical and mental demands of the job according to Hurd & McGarry (1993) seem to have only a modest influence on prospective retirement. That is, it is not necessarily the pleas16
Nevertheless, as a check of the robustness of the results, “as long as possible” is tentatively included in the 67+ category. By doing this, the sample increases from 1249 to 1357 observations. The results are to a large extent qualitatively similar and only differ from the results of ordered probit analyses by the fact that in the case of men, higher education, wealth and tenure become significant, while for women poor health, salaried worker/ public servant, stress and wage satisfaction become significant. The results on the subjectively assessed job measures are however much the same at least for the male sample and as it appears difficult to consider the “as long as possible” combined with the “67+” as a homogenous group, I maintain this selection.
51
ant attributes themselves that makes one stay at the job longer, rather that, those who would stay on the job longer also tend to have the more challenging and/or rewarding jobs. One alternative would be to explicitly model this self-selection into different types of jobs. However, a lack of suitable instruments in my data prevents me from carrying out this type of correction.
Another consideration is that job attributes may be picking up the effect of unobserved person-specific heterogeneity. For example, high achieving, motivated individuals are found in highly demanding jobs. Consequently, the probit model where job demands enter exogenously will be compared to a model where wage determination is modelled first in a separate fixed effects model, and then the retrieved fixed effect is added as an explanatory variable to the retirement probit. It is assumed that the unobserved individual factors such as ability or earnings capacity that affect wages are the same as the one that affect the planned retirement age by leading to e.g. increased motivation and enjoyment of working life. Register information on the period 1993-1997 is used to calculate the fixed effects. The results of these analyses for men and women respectively are shown in Table A.3 in the Appendix.
Finally, in the model, it is assumed that the explanatory variables are exogenous, but in fact this may be incorrect. For instance, the length and conditions of one's working life may affect health so that expectations concerning the timing of retirement are determined jointly with health. In this paper, I do not inquire into this potential endogeneity problem.
52
5.1. Results for Denmark To examine the role of subjectively assessed job characteristics in planning of the retirement age, the analysis is conducted in five stepwise-extended models, cf. Table 3 and 4 below.17 The nested models include: (1) Individual factors, financial and retirement income variables (2) Additional retirement income variables added (3) Labour market characteristics added. (4) Subjectively assessed job measures added. (5) Fixed effects added.
In general, the coefficients in the five models are comparatively stable when new variables are added suggesting that the added variables including the subjectively assessed job measures are uncorrelated with the other characteristics.
Wald tests show that the most appropriate models are model 4 and 5 for men and women respectively. Keeping in mind that care must be taken in 17
While I choose to adopt an ordered probit framework in this paper, a linear regression produces similar results. One might argue that people do not plan in such detail as choosing between before 60, 60, 61-62, 63, 64-66 and 67 and above. Therefore, simple probit analyses are conducted in which the distinction is made between age 60 or earlier and above age 60. However, the results of these analyses are qualitatively similar and only differ from the results of ordered probit analyses by the fact that in the case of men poor health, partner at the same age or older, compensation rate and hours of work are not significant, while tenure is significant, while for women, the results only differ with respect to wage satisfaction, which is only significant in the simple probit analysis. However, as the age-groups defined coincide with eligibility to different pension program regimes which are widely known, I choose to retain the detailed ordered probit specification.
53
interpreting answers to survey questions, these results suggest that subjectively assessed job measures play a role in retaining older workers in the Danish labour market. The difference between men and women arise from the fact that the fixed effect error component is only significant for women. Thus, unobserved person-specific factors such as innate ability and motivation are apparently important for women in their planning of retirement age.
Job satisfaction and working hour satisfaction increase the planned retirement age among both men and women. A high degree of job satisfaction means that the respondent is quite sure that he would choose the current job again, cf. the definition of the job satisfaction variable in Table A.1 in the Appendix. Furthermore, if the respondent is very satisfied with the current number of working hours, he expects to retire later.
54
Table 3. Ordered probit estimates of determinants of planned retirement age, men, Denmark (standard errorsa) in parentheses). Age
Vocational training
Higher education
Poor health
Fair health
Single
Partner: same age/ older
Log earnings
Wealth
Compensation rate
Missing compensation rate
Early retirement benefit
(1)
(2)
(3)
(4)
(5)
0.219*
0.227*
0.392***
0.349**
0.352**
(0.089)
(0.089)
(0.109)
(0.111)
(0.117)
0.044
0.076
0.088
0.109
-0.095
(0.102)
(0.103)
(0.105)
(0.106)
(0.194)
0.374**
0.361**
0.209
0.229
0.250(*)
(0.135)
(0.138)
(0.151)
(0.153)
(0.145)
-0.417***
-0.377**
-0.364**
-0.288*
-0.285*
(0.118)
(0.120)
(0.121)
(0.126)
(0.130)
-0.031
-0.004
0.039
0.042
0.037
(0.118)
(0.120)
(0.122)
(0.123)
(0.117)
0.038
-0.041
-0.046
-0.029
-0.035
(0.153)
(0.156)
(0.157)
(0.159)
(0.170)
-0.209*
-0.228*
-0.217*
-0.212*
-0.220*
(0.091)
(0.093)
(0.093)
(0.094)
(0.092)
0.438**
0.616**
0.548**
0.578**
0.614**
(0.157)
(0.163)
(0.179)
(0.190)
(0.198)
-0.000(*)
-0.000
-0.000
-0.000
-0.000
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
-0.209
1.308***
1.382***
1.504***
1.456***
(0.290)
(0.408)
(0.412)
(0.421)
(0.408)
-42.004
261.605***
276.581***
300.959***
291.318***
(58.086)
(81.627)
(82.500)
(84.243)
(81.584)
-
-2.202***
-2.210***
-2.242***
-2.251***
(0.282)
(0.283)
(0.286)
(0.335)
-1.453***
-1.425***
-1.436***
Labour market pensions
-
-1.369*** (0.283)
(0.288)
(0.292)
(0.356)
Private schemes
-
-1.719***
-1.767***
-1.779***
-1.793***
(0.288)
(0.290)
(0.295)
(0.356)
-2.233***
-2.258***
-2.201***
-2.210***
(0.300)
(0.302)
(0.307)
(0.368)
-
-0.028*
-0.022(*)
-0.021(*)
(0.012)
(0.012)
(0.013)
0.035
0.012
0.032
(0.108)
(0.112)
(0.117)
-0.110
0.057
-0.047
(0.105)
(0.110)
(0.138)
-0.006
-0.006
-0.006
(0.004)
(0.004)
(0.004)
Other income
Experience
-
-
Salaried worker/ public servant
-
Private sector
-
Tenure
-
-
-
-
55
Hours of work
Physical demanding
Stress
Job satisfaction
Wage satisfaction
Working hour satisfaction
Hard to satisfy job demands
No usual retirement age
Usual retirement age
Fixed effect
(1)
(2)
(3)
(4)
(5)
-
-
0.008
0.013(*)
0.012(*)
(0.007)
(0.008)
(0.007)
-
-0.024
-0.031
(0.096)
(0.089)
-0.079
-0.071
(0.093)
(0.092)
0.091**
0.089**
(0.031)
(0.031)
-0.072
-0.077
(0.058)
(0.059)
0.140*
0.141*
(0.057)
(0.060)
0.108
0.104
(0.095)
(0.093)
17.581***
17.894***
(3.241)
(3.193)
0.106***
0.108***
(0.020)
(0.192)
-
1.187
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
(0.924) Cut 1
3.658
4.404
2.745
11.156
17.324
Cut 2
5.576
6.391
4.752
13.281
19.454
Cut 3
5.924
6.756
5.122
13.662
19.836
Cut 4
6.385
7.257
5.629
14.182
20.357
Cut 5
6.963
7.933
6.309
14.896
21.072
-923
-885
-879
-853
-852
Log likelihood Number of observations
675
Notes: a) In model 5, robust standard errors are calculated to take into account that the fixed error component is an estimate. Therefore, a log pseudo-likelihood value is reported for this model. (*) Significant at a 10% level, * significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level. Wald test: chi2 (11) = 58.62; prob > chi2 = 0.000 chi2 (4) = 73.68; prob > chi2 = 0.000 chi2 (5) = 11.96; prob > chi2 = 0.035 chi2 (8) = 50.26; prob > chi2 = 0.000 chi2 (1) = 1.65; prob > chi2 = 0.199
56
Table 4. Ordered probit estimates of determinants of planned retirement age, women, Denmark (standard errorsa) in parentheses). Age
Vocational training
Higher education
Poor health
Fair health
Single
Partner: same age/ younger
Log earnings
Wealth
Compensation rate
Missing compensation rate
Early retirement benefit
(1)
(2)
(3)
(4)
(5)
0.501***
0.512***
0.550***
0.601***
0.605***
(0.101)
(0.102)
(0.105)
(0.108)
(0.102)
0.302**
0.305**
0.290*
0.333**
0.248(*)
(0.116)
(0.117)
(0.126)
(0.130)
(0.129)
0.322*
0.253(*)
0.163
0.218
0.092
(0.138)
(0.139)
(0.152)
(0.157)
(0.155)
-0.358**
-0.315*
-0.306*
0.143
-0.148
(0.123)
(0.124)
(0.124)
(0.130)
(0.138)
-0.151
-0.148
-0.144
0.089
-0.068
(0.125)
(0.126)
(0.127)
(0.130)
(0.131)
0.619***
0.604***
0.598***
0.642***
0.620***
(0.137)
(0.138)
(0.139)
(0.143)
(0.138)
0.141
0.082
0.086
0.083
0.105
(0.115)
(0.116)
(0.117)
(0.120)
(0.118)
0.290
0.378*
0.393(*)
0.245
0.155
(0.178)
(0.180)
(0.233)
(0.244)
(0.233)
-0.000*
-0.000*
-0.000*
-0.000*
-0.000*
(0.000)
(0.000)
(0.000)
(0.000)
(0.000)
-0.391
0.229
0.189
0.308
0.239
(0.284)
(0.355)
(0.359)
(0.363)
(0.327)
-78.281
45.707
37.837
61.634
47.895
(56.853)
(71.017)
(71.795)
(72.774)
(65.428)
-
-1.686***
-1.715***
-1.932***
-1.788***
(0.375)
(0.378)
(0.382)
(0.559)
-1.131**
1.407***
-1.309*
Labour market pensions
-
-1.095** (0.375)
(0.379)
(0.384)
(0.593)
Private schemes
-
-1.547***
-1.611***
1.843***
-1.769**
(0.401)
(0.403)
(0.410)
(0.600)
-1.501***
-1.554***
-1.682***
-1.573**
(0.385)
(0.388)
(0.393)
(0.597)
-
-0.006
0.009
-0.009
(0.009)
(0.009)
(0.009)
0.119
0.057
0.039
(0.145)
(0.151)
(0.137)
-0.155
0.141
-0.193
(0.111)
(0.117)
(0.118)
-0.014*
-0.013*
-0.011*
(0.005)
(0.006)
(0.005)
Other income
Experience
-
-
Salaried worker/ public servant
-
Private sector
-
Tenure
-
-
-
-
57
Hours of work
Physical demanding
Stress
Job satisfaction
Wage satisfaction
Working hour satisfaction
Hard to satisfy job demands
No usual retirement age
Usual retirement age
Fixed effect
(1)
(2)
(3)
(4)
(5)
-
-
0.003
0.011
0.013
(0.009)
(0.010)
(0.009)
-
0.208(*)
-0.179
(0.112)
(0.109)
0.127
-0.154
(0.104)
(0.097)
0.131***
0.134***
(0.038)
(0.041)
-0.076
-0.077
(0.052)
(0.056)
0.178**
0.207***
(0.062)
(0.059)
-0.333**
-0.308**
(0.109)
(0.104)
7.117*
6.321(*)
(3.478)
(3.247)
0.044*
0.039*
(0.021)
(0.020)
-
0.822***
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
(0.236) Cut 1
1.989
1.740
1.582
3.208
5.858
Cut 2
4.323
4.090
3.954
5.729
8.418
Cut 3
4.842
4.623
4.495
6.300
8.997
Cut 4
5.363
5.169
5.050
6.886
9.596
Cut 5
5.885
5.757
5.646
7.510
10.230
-642
-629
-624
-600
-594
Log likelihood Number of observations
574
Notes: a) In model 5, robust standard errors are calculated to take into account that the fixed error component is an estimate. Therefore, a log pseudo-likelihood value is reported for this model. (*) significant at a 10% level, * significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level. Wald test: chi2 (11) = 78.09; prob > chi2 = 0.000 chi2 (4) = 25.69; prob > chi2 = 0.000 chi2 (5) = 9.63; prob > chi2 = 0.087 chi2 (8) = 47.35; prob > chi2 = 0.000 chi2 (1) = 12.11; prob > chi2 = 0.001
58
The higher the usual retirement age in one’s position, the higher the planned retirement age. This result applies also to both men and women. That is, established rules, customs and practices in one’s position are important when the timing of retirement is planned. Similar results are found in Hurd & McGarry (1993). The result is also confirmed by Duflo & Saez (2002) who find that social interactions have a strong effect on economic decisions. Finally, the planned retirement age appears to be lower for women, if they find it hard to satisfy job demands. In this context, job demands are further training, use of new technology and readjustment to new tasks. The distribution for women on these three types of job demands indicates that use of new technology is the major problem.
The effect of working in a physical demanding or stressful job on the planned retirement age is insignificant. A similar result is found in Hurd & McGarry (1993). The effect of wage satisfaction is also insignificant. Presumably, wage satisfaction is based on a relative wage. That is, if pay is considered to be fair compared to what colleagues in the same position are paid, then the individual is satisfied. I image that the insignificant result for wage satisfaction suggests an insignificant relationship between planned retirement age and relative wage satisfaction.
Labour market characteristics also affect the planned retirement age although the importance seems to be minor. Experience affect men's planned retirement age negatively. Namely, the longer the working life and the more worn-out a man is, the earlier he wants to retire. On the contrary, the number of working hours increases men's planned retirement age. This indicates that a large number of working hours reflect a greater preference for 59
work. As expected, a high degree of tenure also affects the planned retirement age negatively. However, this result only applies to women. As found in previous studies, education seems to be more important among women than among men. Namely, lower educated women retire earlier than others.
Financial circumstances are also important. However, the effect on planned retirement age differs for men and women. Namely, earnings have a significant positive effect for men, while wealth has a significant negative effect for women. The difference between men and women might reflect that men prefer work, while women prefer leisure in the sense that they attach greater importance to spending time in their home and within the family.
Expected income when retired is also important. In this context, two different measures are used. For further description of these variables, see Section 9.1 in the Appendix. People that expect that either early retirement benefits, labour market pension, private schemes or other income is the most important income when retired have a lower planned retirement age than people that consider OAP as the most important income source. This result applies to both men and women. Labour market pension and private schemes are expected income sources among high wage earners in particular and the result suggests that these groups plan to retire before the normal retirement age at which OAP is available, cf. Section 9.1 in the Appendix.
Finally, men's planned retirement age increases with increased compensation rate. This is the opposite of what was expected. The result is due to correlation between the estimated compensation rate and the expected in-
60
come sources when retired. In fact, the differences between model 1 and 2 show that the sign for compensation rate goes from positive to negative when the expected income sources are added. Further examination of this relationship shows that the estimated compensation rate is positively correlated with early retirement benefits and negatively correlated with labour market pensions and private schemes. In fact, when I force the compensation rate to have only one sign, the result becomes either insignificant or wrong. In this case, it seems to be wrong.
Other individual factors also influence retirement planning. In fact, irrespective of sex, the older a person is, the later they plan to retire.18 That is, the planned retirement age is higher among 57-year-olds than among 52year-olds. One reason is that health is taken into account. Moreover, the result might reflect that some healthy individuals in their fifties tend to adjust their planned retirement age upwards when they approach the age of 60 because they are pleased about going to work. Poor health decreases planned retirement age among men. A similar result is found in Danø et al. (1998, 2000). According to Danø et al. (2000) one possible explanation of the difference between men and women might be that men to a greater extent than women are employed in jobs that are inconsistent with poor health. Another explanation could be that compared to men a relatively larger share of women in good health expects to retire early. However, the results in general rest on the implicit assumption that will be no change. In
18
The estimates on the age variable across the models are more stable for women than for men because of a high degree of correlation between men's age and their work experience.
61
other words, it seems reasonable to believe that worsened health might affect men as well as women, see also Larsen and Datta Gupta (2004).
I also look at the effect of being single or living with a partner. To take into account the joint retirement decision of couples, differences in the ages of the partners enter into these variables. The predominant pattern within couples is that the man is older than the woman and consequently, the baselines differ. The baseline for men is "younger partner" while the baseline is "older partner" for women. The results differ as well. Planned retirement age is lower for men with a partner about the same age or older. A potential explanation is that the desire to retire is stronger when your partner is expected to retire at the same time or before yourself. Single women have a higher planned retirement age compared to other women. Apparently, this result reflects that single women have a greater preference for work since their earnings are higher. Again, however, changes such as singles finding a partner or cohabiting individuals getting divorced or becoming a widow or widower might change the retirement plans.
5.2. Marginal effects of subjectively assessed job measures for Denmark The results above suggest that subjectively assessed job measures do play a role when the retirement age is planned. However, relative to the signs of the coefficients only the signs of the first and the last category of the dependent variable are unambiguous, when ordered probit analysis is conducted. Therefore, in order to obtain more detailed information about the direction of the effects of these job measures and the size of these effects 62
compared to the effect of the other explanatory variables, marginal effects are calculated for men and women, respectively.
Table 5. Marginal effects of a unit change in explanatory variables from ordered probit estimation of the probability of planning to retire at different ages, men, Denmark. Less than 60
60
61-62
63
64-66
67 or above
Age**
-0.023
-0.112
0.020
0.041
0.048
0.026
Vocational training
-0.008
-0.035
0.007
0.013
0.015
0.008
Higher education
-0.014
-0.075
0.012
0.027
0.032
0.018
0.024
0.084
-0.020
-0.035
-0.036
-0.017
-0.003
-0.013
0.002
0.005
0.006
0.003
Single
0.002
0.009
-0.002
-0.004
-0.004
-0.002
Partner: same age/ older*
0.015
0.067
-0.013
-0.026
-0.028
-0.015
-0.040
-0.183
0.035
0.070
0.078
0.040
Poor health* Fair health
Log earnings** Wealth
0.000
0.000
0.000
0.000
0.000
0.000
Compensation rate***
-0.105
-0.475
0.091
0.183
0.203
0.104
Missing compensation rate***
-1.000
0.000
0.000
0.000
0.000
1.000
Early retirement benefit***
0.195
0.538
-0.059
-0.160
-0.255
-0.259
Labour market pensions***
0.223
0.202
-0.107
-0.143
-0.123
-0.053
Private schemes***
0.346
0.118
-0.126
-0.156
-0.128
-0.053
Other income***
0.539
-0.079
-0.139
-0.157
-0.119
-0.046
Experience(*)
0.002
0.007
-0.001
-0.003
-0.003
-0.002
Salaried worker/public servant
-0.001
-0.004
0.001
0.001
0.002
0.001
Private sector
-0.004
-0.018
0.004
0.007
0.008
0.004
0.000
0.002
0.000
-0.001
-0.001
0.000
-0.001
-0.004
0.001
0.002
0.002
0.001
Physical demanding
0.002
0.008
-0.001
-0.003
-0.003
-0.002
Stress
0.006
0.025
-0.005
-0.010
-0.011
-0.006
Job satisfaction**
-0.006
-0.029
0.006
0.011
0.012
0.006
Wage satisfaction
0.005
0.023
-0.004
-0.009
-0.010
-0.005
Working hour satisfaction*
-0.010
-0.044
0.009
0.017
0.019
0.010
Hard to satisfy job demands***
-0.007
-0.035
0.006
0.013
0.015
0.008
No usual retirement age***
-1.000
1.000
0.000
0.000
0.000
0.000
Usual retirement age
-0.007
-0.034
0.006
0.013
0.014
0.007
38
362
82
86
66
41
Tenure Hours of work(*)
Number of observations
Note: The marginal effects are calculated on the basis of estimations from model 4 for men, cf. Table 3. The significance levels noted in the first column are obtained from this model as well. (*) significant at a 10% level, * significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
63
Table 6. Marginal effects of a unit change in explanatory variables from ordered probit estimation of the probability of planning to retire at different ages, women, Denmark. Less than 60
60
61-62
63
64-66
67 or above
Age***
-0.030
-0.182
0.075
0.075
0.043
0.018
Vocational training(*)
-0.013
-0.073
0.032
0.030
0.017
0.006
Higher education
-0.005
-0.027
0.012
0.011
0.006
0.002
Poor health
0.009
0.040
-0.020
-0.017
-0.009
-0.003
Fair health
0.004
0.019
-0.009
-0.008
-0.004
-0.002
Single***
-0.026
-0.199
0.073
0.080
0.050
0.023
Partner: same age/ younger
-0.006
-0.030
0.014
0.013
0.007
0.003
Log earnings
-0.008
-0.044
0.021
0.019
0.010
0.004
0.000
0.000
0.000
0.000
0.000
0.000
Compensation rate
-0.013
-0.068
0.032
0.029
0.015
0.006
Missing compensation rate
-1.000
0.000
0.000
0.000
0.000
1.000
Early retirement benefit***
0.090
0.514
-0.145
-0.197
-0.157
-0.105
Labour market pensions*
0.166
0.153
-0.152
-0.105
-0.047
-0.016
Private schemes**
0.340
-0.019
-0.167
-0.101
-0.042
-0.013
Other income**
0.269
0.045
-0.160
-0.099
-0.041
-0.013
Experience
0.000
0.002
-0.001
-0.001
-0.001
0.000
-0.002
-0.011
0.005
0.005
0.002
0.001
Private sector
0.011
0.053
-0.026
-0.023
-0.012
-0.004
Tenure*
0.001
0.003
-0.001
-0.001
-0.001
0.000
-0.001
-0.004
0.002
0.002
0.001
0.000
Physical demanding
0.009
0.052
-0.024
-0.022
-0.012
-0.004
Stress
0.008
0.044
-0.020
-0.019
-0.010
-0.004
-0.007
-0.038
0.018
0.016
0.009
0.003
Wealth*
Salaried worker/public servant
Hours of work
Job satisfaction*** Wage satisfaction
0.004
0.022
-0.010
-0.009
-0.005
-0.002
Working hour satisfaction***
-0.011
-0.059
0.028
0.025
0.013
0.005
Hard to satisfy job demands**
0.019
0.083
-0.041
-0.036
-0.018
-0.007
No usual retirement age(*)
-0.952
0.952
0.000
0.000
0.000
0.000
Usual retirement age*
-0.002
-0.011
0.005
0.005
0.002
0.001
Fixed effect***
-0.002
-0.011
0.005
0.005
0.002
0.001
28
363
84
55
28
16
Number of observations
Note: The marginal effects are calculated on the basis of estimations from model 5 for women, cf. Table 4. The significance levels noted in the first column are obtained from this model as well. (*) significant at a 10% level, * significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
64
The calculated marginal effects for the subjectively assessed job factors are in general fairly small, cf. Table 5 and 6 above. However, the general picture is that bad job conditions increase the probability that older Danish workers plan to retire at or before age 60, while good conditions decrease this probability and vice versa. In addition, the effects are typically larger for women than for men. Finally, the largest marginal effects for subjectively assessed job factors are found for exit at the age 60 for men and for age 60 and 61-62 for women, while the effects for the other retirement ages are tiny. However, since the majority of older Danish workers plan to retire at age 60, an effort directed towards postponing the retirement age a year or two from the age of 60 might still be a good investment.
The largest marginal effects are found for women for problems associated with meeting job demands. In fact, if women face such problems, the probability of planning to retire at the early retirement age is increased by more than 8 percentage points compared to women without these problems. Conversely, the probability that women plan to retire at age 61-62 and at age 63 respectively is increased by around 4 percentage points for women without these problems. In other words, it seems to be of particular importance to ensure that the skills of older female employees are concordant with the job demands.
A way to persuade both older men and women to prolong their working life is to increase working hour satisfaction. In fact, this kind of satisfaction is found to be the most important subjectively assessed job measure for men. However, the approach is more effective for women since for instance the effect of increasing working hour satisfaction one point decreases the prob65
ability to plan to retire at age 60 by almost 6 percentage points for women but only a little bit more than 4 percentage points for men. Similarly, such an increase in working hour satisfaction increases the probability of planning to retire at age 61-62 by almost 3 percentage points for women but only by around one point for men. Another way to retain older workers is to increase job satisfaction. The effect of a one-point increase in job satisfaction is a decrease in the probability of planning to retire at age 60 by almost 4 percentage points for women and 3 percentage points for men. The results suggest that compared to job satisfaction, working hour satisfaction is more important for both men and women when planning the retirement age.
The results for the usual retirement age were found to be significant for both men and women, cf. above. However, the effect is very small for women, in particular. A one-year increase in this age only decreases women’s probability of planning to retire at the early retirement age by one point. The effect for men is more than 3 percentage points. This difference might suggest that men are more affected by the retirement behaviour among their colleagues than women.
Looking at the remaining variables, the marginal effect for retirement income variables appears to be fairly large. Early retirement benefits in particular have large and lasting effects. For instance, if early retirement benefits are expected to be the most important income source when retired, the probability of planning to retire at age 60 is increased by more than 50 percentage points for men as well as women, which suggest that retirement income have a strong effect on early retirement in Denmark. The effect of 66
age and cohabitation status for women in particular and the effect of earnings and poor health for men also appear to be larger than the effect of subjectively assessed job factors.
All in all, results suggest that differences in planned retirement ages are only to a small extent explained by subjectively assessed job measures. Instead, they are to a much larger extent driven by retirement income variables in particular but also age, cohabitation status, earnings and health.
6. Data and results for the U.S. To put the Danish results into perspective, a similar analysis is carried out on data for the U.S. The rationale for choosing the U.S. is that there are some similarities between the two countries. For instance, the population in each country is well-educated and the labour force participation rate for older women is similar.19 At the same time, however, the pension systems differ to a large extent. While the Danish system is universal, pension is tied to labour market history in the U.S. and typically, the wage is at its highest level at the end of the career. The strong financial incentives builtin the American pension system imply that subjectively assessed job measures are expected to be less important in the U.S. than in Denmark and therefore, it is interesting to test whether or not this is indeed the case.
19
In 1998, the labour force participation rate was 51.2 for 55-64-year-old women in the U.S. and 45.3 for similar Danish women, cf. U.S. Bureau of Labor Statistics (www.bls.gov) and Statbank Denmark (www.statistikbanken.dk).
67
6.1. Data for the U.S. In the analysis for the U.S., data from the Health and Retirement study (HRS) and, in particular, the user-friendly RAND HRS version is used. The RAND version is supplemented with "raw" HRS data about current employment and private pension at age 62. The original HRS, which form the basis of the RAND HRS version, is a national biannually panel survey of individuals born in 1931-1941 and their spouses.20 At the moment, five HRS waves are available for study. That is, every second year from 1992 to 2000. In this study, information about wage earners aged 55-59 in 1996 is used. That is, people temporarily out of wage employment, self-employed and people outside the labour market are not included.
Table 7. Planned retirement age, the U.S. Men
Women
Total
< 62
15
15
15
62
23
23
23
2
1
2
20
25
22
7
5
6
Never
33
31
32
Total
100
100
100
Number of observations
718
752
1470
63-64 65 > 65
The early and the normal retirement age in the U.S. is 62 and 65, respectively. Therefore, I distinguish between the following six categories of
20
For further information about the HRS data, see e.g. Juster & Suzman (1995).
68
planned retirement age21: below 62, 62, 63-64, 65, above 65 and never.22 In this case, there is almost no gender difference in the distribution of the planned retirement in the applied sample, cf. Table 7 above.23 It is no matter of surprise that 62 and 65 are the salient planned retirement ages. However, planned retirement at these ages is not as prevalent as planned retirement at age 60 in Denmark. In fact, the shares that plan to retire at these ages are a little more than 20 per cent in each case. However, it is quite remarkable that one out of three plan never to retire. In fact, the big differences in the distribution of planned retirement in Denmark and the U.S. seem to reflect the difference in average retirement age in the two countries. In the period 1994-1999, the retirement age in Denmark averaged around 62 compared to 65 in the U.S.; see Scherer (2001).24 This difference also reflects that withdrawal at the normal retirement age is much more common in the U.S.
21
In the case of the U.S., the planned retirement age questions are used. The respondent is asked when he/she will stop working or retire. Some respondents mentioned some age or answered "never". They were given these values for planned retirement age. If the respond is that he/she has not given it much thought or has no plan, he/she was asked: "When do you think you will stop working?" Some respondents mentioned some age and were given these values for planned retirement age. Other answered "never". These answers are considered as missing information.
22
Having a separate category for “never” corresponds to what have been done in Panis et al. (2002). One might, however, argue that “never” should not be taken literally. Therefore, as a check of the robustness of the results, the responses “never” and “above 65” are combined. For men these results are qualitatively similar, while for women, a larger difference appears. Namely, model 3, which includes subjective job measures, is found to be the most appropriate model. Looking at the estimates, however, the only differences are that job satisfaction is found to be significant, while age is no longer significant. Still, for purposes of comparability to previous U.S. studies, we retain a separate category for “never”. 23
The results for the U.S. are weighted by person-specific sampling weights.
24
However, the Danish average is relatively high compared to most other EU countries, see Scherer (2001). In this context, the calculation of the average retirement age is based on the OECD labour force database.
69
The potential for sample selection bias is even bigger in the case of the U.S.25 Only 36 per cent of the original sample of 55-59-year-olds forms part of the analyses, cf. Table 8 below.26
Table 8. Individuals sorted out of the sample, the U.S. Break down by causes. Number of individuals and per cent of total sample. Number
Per cent
4035
100
527
13
1319
33
Variables with missing values
598
15
Planned retirement age: "never", but "have not given it much thought"
121
3
2565
64
Total sample Self-employed People not currently working for pay
Total of individuals sorted out of the sample
As in the case of Denmark, self-employed, people not working for pay and individuals with missing values for one or more variables are excluded from the analysis. In particular, the share of people that were not working for pay at the time of the interview is comparative large. Part of the difference might be explained by more housewives in the U.S. not working away
25
A minor part of this difference between the two countries is due to the inclusion of individuals that are temporarily out of employment in the analyses for Denmark. It is not possible to include these groups in the analyses for the U.S. 26
However, a correction for potential selection bias has also been conducted in this case using a Heckman two-step procedure. The included explanatory variables were age, race, education, health, working or not, single, partner on the labour market, number of children, experience, industry and occupation for job with longest tenure, total household income and the instruments unemployed and frequency of unemployment in the period 1982-1992. Several of these variables were significant. The Mills ratio was insignificant for men and significant for women but neither for men nor for women did the inclusion of this ratio change the results of the ordered probit analyses. Therefore, these results are not reported in tables but are available on request.
70
from home.27 In this case, people that "have not given it much thought" but answer "never" are also excluded from the analysis. In fact, "never" seem to be a rather uncertain statement in this context. This is confirmed by comparing this group with the "never" group included in the analysis. In fact, these two groups are heterogeneous.
To compare the Danish results to similar results for the U.S., it is important to succeed in finding variables for the U.S., in particular subjectively assessed job variables, similar to the Danish ones. For some variables such as “physical demanding” and “usual retirement age”, I succeed in finding variables for which the content seems to be quite similar, while for others such as “job satisfaction” and “working hour satisfaction”, larger differences exist, cf. Table A.1 in the Appendix. Only the wage satisfaction variable is not available in the U.S. data. These differences will be kept in mind when comparing the results for the two countries.
Variables related to expected income when retired differ between the two countries because of the very different pension systems. Definitions and descriptive statistics of the chosen variables for the U.S. are presented in Table A.4 in the Appendix.
27
This potential explanation might seem to be in contrast to the fact that the labour force participation rate for 55-64-year-old women are similar in the two countries, cf. above. However, while the labour force participation rate is much higher for 55-59year-olds than for 60-64-year-old Danish women (63.5 compared 24.2), this rate is assumed not to deviate as much for the corresponding groups in the U.S. due to higher average retirement age.
71
An ordered probit model is also estimated in this case.28 This permits me to include the 32 per cent that answered "never", cf. Table 7. Survey information from 1992, 1994 and 1996 is used to model wage determination in a fixed effects model. The results of this analysis are shown in Table A.5 in the Appendix.
6.2. Results for the U.S. The analyses are conducted in four stepwise-extended models, cf. Table 9 and 10 below.29 The nested models consist of:
(1) Individual factors, financial, social security and pension wealth variables (2) Labour market characteristics added (3) Subjectively assessed job measures added. (4) Fixed effects added.
28
A linear regression in which “never” is set equal to age 75 produces similar results. Similar probit analyses are conducted in which I distinguish between planned retirement at or before age 62 and after this age. For men these results are qualitatively similar and only differ from the results of the ordered probit analysis by the fact that age is not significant. In the case of women a larger difference is found. Namely, model 3, which includes subjectively assessed job measures, is found to be the most appropriate model. Compared to the results of the ordered probit analysis age, blue collar workers and tenure is not significant, while private pension age, job satisfaction and the usual retirement age in ones position is found to have a significant positive effect. 29
The U.S. data are weighted data, cf. above. The “likelihood” does not fully account for the “randomness” of the weighted sampling. Therefore, the log likelihood is not displayed in Table 9 and 10.
72
Table 9. Ordered probit estimates of determinants of planned retirement age, men, the U.S. (robust standard errors in parentheses). Age
Race
College
Poor health
Fair health
Single
Partner: same age/ older
Individual earnings
Wealth
Retiree health insurance
Private pension age
No private pension
Social security wealth
Private pension wealth
Experience
Services or sales
Blue collar workers
Public sector
(1)
(2)
(3)
(4)
0.065*
0.063*
0.063*
0.061(*)
(0.031)
(0.031)
(0.031)
(0.031)
0.072
0.073
0.030
0.036
(0.136)
(0.139)
(0.144)
(0.143)
0.257**
0.229*
0.226*
0.234*
(0.096)
(0.106)
(0.108)
(0.108)
-0.048
-0.046
-0.029
-0.027
(0.108)
(0.107)
(0.108)
(0.108)
-0.094
-0.080
-0.090
-0.088
(0.102)
(0.102)
(0.102)
(0.102)
0.063
0.077
0.100
0.101
(0.144)
(0.145)
(0.146)
(0.147)
0.043
0.050
0.049
0.049
(0.092)
(0.093)
(0.095)
(0.095)
-0.011
-0.005
-0.006
-0.007
(0.025)
(0.026)
(0.026)
(0.026)
-0.000
-0.000
-0.000
-0.000
(0.000)
(0.000)
(0.000)
(0.000)
-0.119
-0.110
-0.112
-0.112
(0.086)
(0.086)
(0.087)
(0.087)
0.041(*)
0.034
0.021
0.021
(0.021)
(0.021)
(0.021)
(0.021)
10.747(*)
8.884
5.654
5.438
(5.517)
(5.474)
(5.561)
(5.562)
0.069
0.071
0.068
0.075
(0.152)
(0.162)
(0.168)
(0.168)
-0.025*
-0.021(*)
-0.021(*)
-0.019
(0.010)
(0.011)
(0.011)
(0.011)
-
0.007
0.005
0.008
(0.009)
(0.009)
(0.010)
0.129
0.093
0.095
(0.128)
(0.130
(0.130)
-0.024
-0.054
-0.074
(0.117)
(0.119)
(0.125)
-0.029
0.041
0.075
(0.178)
(0.186)
(0.196)
-
-
-
73
(1)
(2)
(3)
(4)
-
-0.008(*)
-0.008
-0.007
(0.005)
(0.005)
(0.005)
0.003
0.003
0.002
(0.005)
(0.005)
(0.005)
-
0.097
0.102
(0.110)
(0.111)
-0.125
-0.117
(0.151)
(0.151)
-0.083
-0.078
(0.100)
(0.100)
0.025
0.027
(0.067)
(0.067)
0.075
0.075
(0.104)
(0.104)
0.151
0.143
(0.094)
(0.095)
0.042(*)
0.041(*)
(0.022)
(0.022)
6.733(*)
6.489(*)
(3.588)
(3.595)
-
-0.103
Tenure
Hours of work
-
Physical demanding
-
Stress - strongly agree
-
Stress – agree
-
Job satisfaction
-
Can reduce hours
-
Harder to satisfy job demands
-
Usual retirement age
-
No usual retirement age
-
Fixed effect
-
-
-
-
-
-
-
-
-
(0.153) Cut 1
5.294
5.079
6.989
6.676
Cut 2
6.082
5.872
7.793
7.487
Cut 3
6.136
5.926
7.847
7.535
Cut 4
6.654
6.447
8.373
8.061
Cut 5
6.840
6.634
8.560
8.249
Number of observations
718
(*) Significant at a 10% level, * significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level. Adjusted Wald test: F (14, 704) = 2.88; Prob > F = 0.000 F (6, 712) = 0.86; Prob > F = 0.525 F (8, 710) = 1.07; Prob > F = 0.384 F (1, 717) = 0.46; Prob > F = 0.499
74
Table 10. Ordered probit estimates of determinants of planned retirement age, women, the U.S. (robust standard errors in parentheses). Age
Race
College
Poor health
Fair health
Single
Partner: same age/ older
Individual earnings
Wealth
Retiree health insurance
Private pension age
No private pension
Social security wealth
Private pension wealth
Experience
Services or sales
Blue collar workers
Public sector
(1)
(2)
(3)
(4)
0.054(*)
0.051(*)
0.056(*)
0.056(*)
(0.029)
(0.029)
(0.029)
(0.029)
0.158
0.165
0.173
0.178
(0.110)
(0.114)
(0.117)
(0.117)
0.134
0.087
0.055
0.034
(0.088)
(0.093)
(0.093)
(0.095)
0.033
0.012
0.032
0.034
(0.106)
(0.106)
(0.107)
(0.107)
0.034
0.017
0.018
0.014
(0.098)
(0.099)
(0.099)
(0.099)
0.446***
0.432***
0.442***
0.445***
(0.104)
(0.104)
(0.104)
(0.104)
0.082
0.055
0.077
0.084
(0.106)
(0.105)
(0.106)
(0.107)
0.005
0.011
0.018
0.016
(0.021)
(0.023)
(0.023)
(0.024)
-0.000*
-0.000*
-0.000*
-0.000*
(0.000)
(0.000)
(0.000)
(0.000)
-0.119
-0.086
-0.081
-0.079
(0.096)
(0.096)
(0.097)
(0.097)
0.031
0.023
0.021
0.021
(0.023)
(0.023)
(0.023)
(0.023)
8.189
6.118
5.587
5.432
(5.954)
(6.026)
(6.105)
(6.114)
0.169*
0.102
0.104
0.122
(0.077)
(0.090)
(0.090)
(0.092)
-0.010
-0.009
-0.008
-0.007
(0.009)
(0.010)
(0.010)
(0.010)
-
0.013**
0.013**
0.020*
(0.005)
(0.005)
(0.008)
-0.071
-0.073
-0.044
(0.107)
(0.109)
(0.148)
-0.261(*)
-0.234
-0.070
(0.149)
(0.154)
(0.218)
0.107
0.149
0.101
(0.197)
(0.202)
(0.206)
-
-
-
75
(1)
(2)
(3)
(4)
-
-0.013*
-0.013*
-0.009
(0.005)
(0.005)
(0.006)
0.005
0.004
0.004
(0.004)
(0.005)
(0.005)
-
0.042
0.038
(0.089)
(0.090)
0.158
0.152
(0.123)
(0.124)
0.082
0.081
(0.103)
(0.103)
0.175*
0.174*
(0.072)
(0.072)
-0.058
-0.060
(0.096)
(0.096)
-0.008
-0.006
(0.094)
(0.094)
0.025
0.025
(0.020)
(0.020)
4.340
4.326
(3.279)
(3.294)
-
-0.222
Tenure
Hours of work
-
Physical demanding
-
Stress - strongly agree
-
Stress – agree
-
Job satisfaction
-
Can reduce hours
-
Harder to satisfy job demands
-
Usual retirement age
-
No usual retirement age
-
Fixed effect
-
-
-
-
-
-
-
-
-
(0.198) Cut 1
5.148
4.386
6.895
6.818
Cut 2
5.938
5.183
7.704
7.628
Cut 3
5.978
5.224
7.746
7.670
Cut 4
6.636
5.893
8.426
8.350
Cut 5
6.768
6.027
8.561
8.485
Number of observations
752
(*) significant at a 10% level, * significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level. Adjusted Wald test: F (14, 738) = 4.300; Prob > F = 0.000 F (6, 746) = 2.680; Prob > F = 0.014 F (8, 744) = 1.51; Prob > F = 0.149 F (1, 751) = 0.26; Prob > F = 0.262
76
As mentioned above, the intention has been to set up similar models for Denmark and the U.S. The results show that the models are much less appropriate for the U.S. than for Denmark. In fact, rather few variables are significant in this case. Adjusted Wald tests show that model 1 and 2 are the most appropriate models for men and women, respectively. Therefore, according to this study subjectively assessed job measures seem to be less important for older wage earners in the U.S. than for similar Danes as expected. It can, however, not be ruled out that different content of the job measures partly explain the difference.
The results for men suggest that age, education and age for access to private pension increase the planned retirement age, while private pension wealth decreases this age. The signs are as expected. The most significant result is the effect of education i.e. having a college degree, which increases the planned retirement age among men.
The most significant result that is found for U.S.-women is being single. Single women plan to retire later than other women. Further investigation of this result indicates that single women need to work due to financial reasons. Experience also increases women's planned retirement age. The positive sign is the opposite of what was found in the analysis for Denmark. The difference might reflect that Americans, who have worked for many years, have to continue to work for financial reasons while comparable Danes to a greater extent can afford to retire at the early retirement age. In fact, the social security system is more favourable for low wage earners in Denmark than in the U.S. Access to early retirement through Social and
77
Disability Pension30 in Denmark could also explain the difference in retirement behaviour of experienced workers in the two countries.
Wealth and tenure affect the planned retirement age negatively. I find that tenure is positively correlated with access to health insurance and negatively correlated with age for access to private pension. In fact, the significant result for tenure also seems to reflect financial circumstances. As was found for men, age also affects the retirement age positively among women. Finally, blue-collar workers plan to retire earlier than professional workers.
All in all, the results for men as well as women indicate that financial circumstances are important in the U.S. when the timing of the retirement age is planned. While early retirement benefits were found to hasten retirement in Denmark, the results for the U.S. suggest that the strong financial incentives built-in the actuarially fair American pension system contribute to a higher retirement age.
The result that the importance of subjectively assessed job measures is limited among wage earners in the U.S. might be in accordance with the results in Hurd & McGarry (1993). Namely, in that paper, the importance of these job characteristics compared to other factors is not assessed. In fact, in this paper, the usual retirement age is found to be significant in the case of U.S. men, while job satisfaction is found to be significant in the case of
30
See Section 9.1 in the Appendix for further description of this scheme.
78
women. However, the models, which include these variables, are not found to be the most appropriate in any of these cases.
7. Concluding remarks The changing labour market for aging workers and the demographic changes expected increase the need for convincing older workers to prolong their working life. This paper looks at the role of subjectively assessed job characteristics in retaining older workers in the Danish labour market. The focus is on wage earners in his or her fifties. To put the results for Denmark into perspective, a parallel analysis is also carried out on data for the U.S., as the two countries are similar in terms of background characteristics and labour force attachment of their older worker populations, yet significantly different in terms of the structure and incentives of their pension systems.
In Denmark, withdrawal from the labour market around the early retirement age, that is age 60, has more or less become the norm. Retirement at age 62 and 65 is typical in the U.S. However, retirement at these ages is not as prevalent as age norms as retirement at age 60 in Denmark. The different distributions of planned retirement age in the two countries reflect that the average retirement age is lower in Denmark. While Danish women plan to retire earlier than men, there are almost no gender differences in the U.S.
Knowledge about the subjectively assessed job factors that affect planned retirement age yields input to the ongoing debate about how to convince
79
older people to prolong their working life. The findings for Denmark suggest that job measures do play a role when the retirement age is planned. In fact, bad job conditions increase the probability that older Danish workers plan to retire at or before age 60, while good conditions decrease this probability and vice versa. In addition, the effects of job measures are typically found to be larger for women than for men. However, to get an appreciable effect on the retirement age, quite large changes in these factors are required. Focusing on the most important job measures for women and men respectively, females’ probability of planning to retire at the early retirement age is increased only by 8 percentage points if they find it hard to satisfy job demands compared to females without such problems. Correspondingly, a one point increase in males’ working hour satisfaction only increases their probability by 4 percentage points.
The largest marginal effects for subjectively assessed job factors are found for exit at the age for 60 for men and for age 60 and 61-62 for women, while the effects for the other retirement ages are tiny. However, since the majority of older Danish workers plan to retire at age 60, an effort directed towards postponing the retirement age a year or two from the age of 60 might still be a good investment.
While workplace conditions are found to have fairly small effects on the planned retirement age, the effect of retirement income variables in particular early retirement benefits appears to be much stronger and have more lasting effects. In fact, the probability of planning to retire at age 60 is increased by more than 50 percentage points for men as well as women if early retirement benefits are expected to be the most important income 80
source when retired. This result suggests that early retirement benefits have a strong effect on early retirement in Denmark. However, since previous studies suggest that rather large changes in economic policy variables are required in order to elicit substantial changes in retirement ages, and since cuts in benefits are politically speaking an unpopular option, an effort directed toward improving workplace conditions might still be a worthy investment.
Compared to Denmark, subjectively assessed job measures seem to be less important for older wage earners in the U.S. than for similar Danes as expected. It can, however, not be ruled out that different content of the job measures partly explain the difference. The results for men as well as women indicate that financial circumstances are important in the U.S. when the timing of the retirement age is planned. In fact, while pension program features were found to hasten retirement in Denmark, the strong financial incentives built-in the American pension system seem to contribute to a higher retirement age in the U.S. In general, the different results for the two countries might reflect that older American wage earners are more financially constrained when planning the timing of their retirement.
This study fills an existing gap in the literature by looking at the importance of subjectively assessed job characteristics on the retirement decision, factors other than those conventionally looked at such as individual traits and the financial net gains of retiring. The dearth of earlier research on this topic has been mainly due to the lack of suitable data. I have access to two sources of data on job measures, from Denmark and the U.S., allowing me to analyze this question both within and across different pension systems. 81
However, some important qualifications that need mentioning are the use of subjectively assessed job measures and that the outcome variable is planned, not actual retirement age, since sample information on both working conditions and actual retirement age was not available at the time the analyses were conducted. These assumptions should be kept in mind when interpreting the results for policy purposes. In fact, the relative importance of objective vs. subjective job measures could be one topic for further research, while another topic for future work could be a comparison of the determinants of actual versus planed retirement.
8. References Berkovec, J. & Stern, S. (1991) Job Exit Behavior of Older Men, Econometrica, Vol. 59, No. 1. Bernheim, B.D. (1989) The timing of retirement: A comparison of expectations and realizations, in The Economics of Aging, (ed) D.A. Wise, NBER, 1989. Bertrand, M. & Mullainathan, S. (2001) Do People Mean What They Say ? Implications for Subjective Survey Data, American Economic Review, Vol. 91, No. 2, pp. 6772. Bingley, P., Datta Gupta, N. & Pedersen, P.J. (2003) The Impact of Incentives on Retirement in Denmark, in Social Security and Retirement Around the World: Microestimation, (eds.) J. Wise and D. Gruber, NBER, 2003. Blöndal, S. and S. Scarpetta (1998) The Retirement Decision in OECD Countries, Aging Working Papers 1.4, Paris. Burtless, G. & Morfitt, R. A. (1985) The Joint Choice of Retirement Age and Postretirement Hours of Work, Journal of Labor Economics, Vol. 3, No. 2, pp. 209236. Christensen, B.J. & Datta Gupta, N. (2000) The Effect of a Pension Reform on the Retirement of Danish Married Couples (in Danish). Nationaløkonomisk Tidsskrift 138, pp. 222-242. Clark, A.E. (2001). What really matters in a job? Hedonic measurement using quit data, Labour Economics, Vol. 8, Issue 2. Danish Insurance Information Service (1997) Social Benefits 1997- who, what & when? (in Danish), Copenhagen.
82
Danø, A.M., Ejrnæs, M. & Husted, L. (2000) How is the Retirement Age Affected by the Reform of the Post Employment Wage programme? (in Danish), Nationaløkonomisk Tidsskrift 138, pp. 205-221. Danø, A.M., Ejrnæs, M. & Husted, L. (1998) Gender Differences in Retirement Behaviour, Institute of Local Government Studies - Denmark, Copenhagen. Department of Unemployment Insurance (2001) Statistics of Post Employment Wage and Transitional Benefit Programme First Half of 2001 (in Danish), digital document: http://www.adir.dk/ Disney, R. & Tanner, S. (1999) What can we learn from retirement expectation data? The Institute for Fiscal Studies, Working Paper Series No. W99/17. Duflo, E. & Saez, E. (2002) The Role of Information and Social Interactions in Retirement Plan Decisions: Evidence from a Randomized Experiment, National Bureau of Economic Research, Working Paper No. 8885. Friedberg, L. (2001) The Impact of Technological Change on Older Workers: Evidence from Data on Computer Use, National Bureau of Economic Research, Working Paper No. 8297. Gruber, J. & Wise, D.A. (1999) Social Security and Retirement around the World, The National Bureau of Economic Research, Chicago. Gustman, A. and Steinmeier, T.L. (2000) Retirement in Dual-Career Families: A Structural Model, Journal of Labor Economics, val 18, no. 3, p. 503-545. Gustman, A.L. & Steinmeier, T.L. (1986): A Structural Retirement Model, Econometrica, Vol. 54, No. 3. Hansen, H. (1999) Elements of Social Security. A comparison covering: Denmark, Sweden, Finland, Austria, Germany, The Netherlands, Great Britain and Canada, The Danish National Institute of Social Research 99:14, Copenhagen. Hurd, M.D. & Boskin, M.J. (1984) The Effect of Social Security on Retirement in the Early 1970s, The Quarterly Journal of Economics, November, pp. 767-790. Hurd, M. & McGarry, K. (1993) The Relationship Between Job Characteristics and Retirement, National Bureau of Economic Research, Working Paper, No. 4558. Juster, T. & Suzman, R. (1995) An Overview of the Health and Retirement Study, The Journal of Human Resources, 30 (supplement), pp. s7-s56. Jørgensen, K. (1997) The Elderly and the Working Life (in Danish), Danish Research and Development Centre for Adult Education, Copenhagen. Krueger, A.B. & Pischke, J.-S. (1992) The Effect of Social Security on Labor Supply: A Cohort Analysis of the Notch Generation, Journal of Labor Economics, Vol. 10, No. 4, pp. 412-437. Larsen, M. and Datta Gupta, N. (2004) The Impact of Health on Individual Retirement Plans: a Planel Analysis comparing Self-reported versus Diagnostic Measures, Unpublished. Lazear, E.P. (1979) Why is There Mandatory Retirement? The Journal of Political Economy, Vol. 87, Issue 6 (Dec., 1979), pp. 1261-1284.
83
Leonesio, M.V. (1990) Effects of the Social Security Earnings Test on the LaborMarket Activity of Older Americans: A Review of the Evidence, Social Security Bulletin Vol. 53, No. 5 (May), pp. 2-21. Michaud, P.-C. (2003) Joint Labour Supply Dynamics of Older Couples, IZA Discussion Paper No. 832, July 2003. Ministry of Finance (2001) The Financial Statement and Budget Report 2001 (in Danish), Copenhagen. Mitchell, O.S. & Fields, G.S. (1984) The Economics of Retirement Behaviour, Journal of Labor Economics, vol. 2, no. 1, p. 84-105. OECD (2000) Reforms for an Ageing Society, Social Issues, Paris. Panis, C., Hurd, M., Loughran, D., Zissimopoulos, J., Haider, S. & StClair, P. (2002) The Effects of Changing Social Security Administration's Early Entitlement Age and the Normal Retirement Age, Santa Monica. Pedersen, P.J. (1998) The Elderly and the Labour Market, in Smith, N. (ed), Work, Work Incentives and Unemployment (in Danish), The Rockwool Foundation Research Unit, Aarhus. Pedersen, P.J. & Smith, N. (1996) A Duration Analysis of the Decision to Retire Early, Chapter 5 in Wadensjö, E. (ed.) The Nordic Labour Markets in the 1990's, Amsterdam. Pedersen, P.J. & Smith, N. (1995) The Retirement Decision, in Mogensen, G.V. (ed), Work Incentives in the Danish Welfare State, New Empirical Evidence, The Rockwool Foundation Research Unit, Aarhus. Pozzebon, S. & Mitchell, O.S. (1989) Married Women's Retirement Behavior, Journal of Population Economics, No. 2, pp. 39-53. Quaade, T. (2002). Reform of the Post Employment Wage Programme with Limited Effect, in Danish National Institute of Social Research (2002) Social Research 2002:3 (in Danish), Copenhagen. Rust, J. (1989): A Dynamic Programming Model of Retirement Behavior, in Wise, D.A. (ed.): The Economics of Aging, Chicago. Schaumann, A. (2001) The Aging Society. Demography - Expenditure Pressure - What can be done? The Danish Board of Technology, digital document: http://www.tekno.dk/ Scherer, P. (2001) Withdrawal from the Labour Force in OECD Countries, OECD Occasional Paper, No. 49. Statistics Denmark (1999) Register based Labour Force Statistics 1 January 1998 (in Danish), Copenhagen. Stock, J.H. & Wise, D.A. (1990): Pensions, the Option Value of Work, and Retirement, Econometrica, vol. 58, no. 5.
84
9. Appendix
9.1. Measurement of expected income when retired Two measures of expected income when retired are included in the ordered probit model for Denmark. The first is a measure of the most important expected income source when retired, while the second is an estimate of the expected compensation rate in the first year as a pensioner.
The most important source when retired31 The information about the expected most important income source when retired is based on the survey question: "What do you think is your most important income source at the time when you stop working?" The possible answers are post employment wage, social and disability pension, old age pension, labour market pension, private pension, private savings, income of the spouse, own earnings and other sorts of income.
Post Employment Wage (PEW) is an early retirement scheme. To be entitled to PEW, a number of conditions must be fulfilled. These rules have been tightened during the 1990's. In 1997, the year in which the applied survey data were collected, entitlement presupposed that an individual was 60 to 66 years of age, had been a member of a unemployment insurance (UI) fund in 20 of the last 25 years, had been working in at least 52 weeks
31
For further information about the Danish pension schemes, see e.g. Bingley et al. (2003) and Hansen (1999).
85
in the last three years and lived in Denmark. After retirement, work was restricted up to maximum 200 hours per year. The PEW benefits did not depend on the age at entry. Instead, it was reduced as a function of time spent in the scheme. PEW benefits were equivalent to UI benefits the first two and a half years.32 UI benefits amount to 90 per cent of previous earnings subject to a ceiling.33 In 1997, PEW benefits were reduced to maximum 82 per cent of UI benefit after two and a half year. However, individuals that delayed retirement until age 63 received benefits equivalent to UI benefits in the whole PEW period. The main source funding of the PEW was general tax revenues. At the normal retirement age, PEW benefit recipients are transferred to old age pension, cf. the description of this scheme below. The PEW programme is generous in the sense that the amount of PEW benefits does not influence the level of old age pension.
In 1999, the requirement for entry into the PEW was tightened up. For instance, to be entitled to PEW, one had to have been a member of an unemployment insurance fund in 25 of the last 30 years and have paid a PEW contribution during this period.34 Benefits depend now on the age at entry instead of being reduced as a function of time spent in the scheme and benefits are means tested against income from labour market pension and lump-sum retirement income payment35 for people aged 60 and 61. Instead
32
However, particular rules applied to people that were unemployed during the two and a half years before early retirement through PEW.
33
This ceiling is reached quite fast. In fact in 1997, the gross compensation percentage was 53 for an average production worker, as defined by OECD, see Hansen (1999).
34
In 2001, almost 80 per cent of the 50-59-year-olds paid this contribution, see Quaade (2002) and presumably, most of them are entitled to this scheme at age 60. 35
I return to these schemes below.
86
of being restricted to work up to a maximum of 200 hours, a flexible scheme was introduced where the benefits are reduced gradually as the number of working hours increases. A tax premium is introduced for people that are entitled to early retirement benefits at age 60 but who continue working until at least the age of 62. From 2004, the official age of retirement will be lowered from 67 to 65 years of age and entitlement to PEW presupposed that the individual is between 60 to 64 years of age.
The Social and Disability Pension (SDP) system is quite complex because tax treatment and means testing differ for different components and amounts. SDP includes different programmes. These are disability pension, a public financed programme for widows’ pension and a programme for early OAP also for individuals whose (older) spouse was receiving the OAP already. Among these, disability pension is the predominant programme. SDP is residence based. Former working or contribution record is not required. To be awarded SDP, an individual must be 18 to 66 years of age (18 to 64 years from 2004). However, SDP is not an individual option like eligibility for PEW, cf. above. The pension amount is graduated according to loss of working capability and varies according to the age of the first time recipient. For instance, the highest level cannot be received for the first time in the age bracket 60-66 years. The basic pension is differentiated according to being single or married and is means tested against a wider range of other income sources. Conversely, the pension is completely independent of former work and income.
87
Respondents that expected that PEW and SDP respectively would be the most important source when retired are merged in the category "early retirement benefits".
National Old Age Pension (OAP) is available when people reach the official pension age, which is 67 in Denmark (65 from 2004, cf. above). Eligibility is only dependent on age and duration of stay. The expenditures are financed from general tax revenues. The OAP depends on marital status and is partly means tested. In fact, a base amount is means tested against earnings from work, a supplementary amount is means tested against other income in the household, while a special supplementary amount for singles is not means tested. In the variable "expected most important income source when retired", respondents, who answered OAP, constitute the baseline.
Labour Market Pension Programmes (LMP) are funded pension programmes. These sorts of programmes are under gradual implementation in Denmark. They cover both public and private employees and different groups of wage earners such as academics and a major part of the labour market for blue-collar workers. The typical structure is defined contributions of either 15 per cent (high wage groups) or 9 per cent (industrial workers) of the annual earnings. Typically, the programmes have an early retirement option before the normal retirement age, normally from the age of 60 and in most programmes with an actuarially reduction depending on the specific age. In this context, the category "labour market pensions" also includes Public Employee Pension (PEP) that is available for public servants. This pension is not funded. It is considered to be part of a lifetime 88
wage contract. The pension amount is calculated as a function of the wage, depending on seniority and position.
Private pension includes both lump-sum retirement income payment and savings in e.g. a bank or an insurance company with current payments. Private pensions are merged with private savings in the category "private schemes".
Finally, income of the spouse, own earnings and other sorts of income are merged in the category "other income".
Compensation rate The expected compensation rate is estimated as if withdrawal took place in 1997. This rate is estimated as the ratio of the potential disposable income as a pensioner to the disposable income as a participant in 1996:
Expected compensation rate97
=
potential disposable income as a pensioner97 disposable income as a participant96
Estimation of the potential disposable income as a pensioner relates to the first year as pensioner and is calculated as the sum of two sources: The estimated amount of (potential) private pension and income from the expected most important income source when retired, cf. the description above of this variable.
89
To estimate the potential disposable income as a pensioner, the amount of private pension, which is a lump-sum retirement income payment, is equally distributed over the years from the expected retirement age until the mean life expectancy for the age and gender group in question. Unfortunately, a lot of people are not aware of the payable amount. Namely, 74 per cent of the sample reported that they could receive a private pension, when they stop working. However, only 45 per cent of this group reported the expected payable amount (although a ball park figure was accepted).
For people in the category "early retirement benefit", the estimation of the income from this source is based on the assumption that all of them expect to receive PEW. This assumption is made primarily because SDP is not an individual option, cf. above. Furthermore, only 2 per cent of the respondents in the category "early retirement benefit" answered SDP. Finally, it is difficult to make a reasonable estimate of (potential) income from SDP because of the complex SDP rules. The income from PEW is calculated as 90 per cent of earnings in 1997 subject to a ceiling, which differ depending on whether the UI is on a full-time or a part-time basis.
When estimating the (potential) income from OAP, marital status and the labour market status of the potential spouse is taken into account. For respondents, who where single in 1997, the (potential) income from OAP is calculated as the sum of the basic amount and the full36 and the special
36
The estimated amount of private pension is quite small for this group and therefore, it is assumed that the supplementary amount is not reduced due to other income in the household.
90
supplementary amount. For respondents with a spouse at least two years younger than themselves, who were in the labour force in 1997, the (potential) income is assumed to correspond to the basic amount. The residual group of respondents with spouses are assumed to receive the basic amount and the full supplementary amount.
It is not possible to estimate the amount of income from the expected most important income source when retired for all the categories of this variable. In fact, for respondents who answered "labour market pension" or "private schemes", the potential retirement income as a pensioner is estimated solely on the basis of the amount of private pension, if this amount is reported, cf. above. If this information is missing, information about the (potential) income as a pensioner is also missing.
In the category "other income", the (potential) income from this source is estimated as half of the spouse's earnings in 1997, if the respondent answered “income of the spouse”. If the answer was “own earnings”, the income was set equal to the respondent's earnings in 1997. If the answer was “other income”, information about the income is missing.
The estimate of the denominator, disposable income as a participant, which is based on earnings in 1996, is described in Pedersen & Smith (1996). A variable for missing compensation rate is added. The compensation rate is missing for people for whom, it is not possible to estimate this rate, e.g. respondents in the category "labour market pension", cf. above. Further, this variable is set to missing, if the estimated rate is in excess of 5. The 91
missing compensation rate variable is set equal to 1, if the compensation rate is missing and 0 otherwise.
9.2. Measurement of social security wealth In the HRS, Social Security Wealth information is only available in the restricted data, which I cannot use. Instead, the calculation of this variable is based on the approach applied in Michaud (2003) in which the estimate for Average Monthly Earnings (AME) is based on the average of observed wages. That is, wages in 1992, 1994 and 1996 in this paper. On the basis of this measure of AME, the Primary Insurance Amount (PIA) is computed. PIA is a piece-wise linear function of the AME. The function for computation of PIA for 1996 was
PIA1996 = c1 min[ AME1996 , p1 ] + c 2 min[max[ AME1996 − p1 ,0], p 2 − p1 .] + c3 max[ AME1996 − p1 − p 2 ,0]
where p1 = $426 and p2 = $2567, referred to as bendpoints, and (c1, c2 , c3) = (0.9, 0.32, 0.15) in 1996. In the case of Denmark, the pronounced feature is the distinction between retirement at or before the early retirement age or after this age. For comparison, the focus is also on the early retirement age in the analysis for the U.S. Therefore, social security wealth at age 62 is calculated. To obtain this measure, PIAt is multiplied by 0.8. The use of social security wealth at this age is based on the assumption that people only respond to this one value when they look at the future.
92
9.3. Tables Table A.1. Subjectively assessed job characteristics, question phrases and creation of variables, DK and U.S. DK, Ældredatabasen
The U.S., HRS
Physical demanding
Is the job physical demanding? 1. To a large extent, 2. To some extent, 3. No. Dummy is set equal to 1, if the answer is 1 or 2.
Statement: My job requires lots of physical effort. 1. All or almost all of the time, 2. Most of the time, 3. Some of the time, 4. None or almost none of the time. Dummy is set equal to 1, if the answer is 1, 2 or 3.
Stress
Is your job satisfaction disturbed by a) high speed or b) stress and a tight timetable? 1. Yes, 2. No. Dummy is set equal to 1, if the answer to a) or b) is 1.
Statement: My job involves a lot of stress. 1. Strongly agree, 2. Agree, 3. Disagree, 4. Strongly disagree. Strongly agree: Dummy is set equal to one, if the answer is 1. Agree: Dummy is set equal to one, if the answer is 2. Disagree: Dummy is set equal to one, if the answer is 3 or 4 (baseline).
Job satisfaction
If you had to decide today, would you then choose your current job again? 1. Yes, quite sure, 2. Yes, fairly certain, 3. Hard to say, 4. No, probably not, 5. No, definitely not. Continuous variable, inverted scale.
Statement: I really enjoy going to work. 1. Strongly agree, 2. Agree, 3. Disagree, 4. Strongly disagree. Continuous variable, inverted scale.
Wage satisfaction
How satisfied are you with your current wage? 1. Very satisfied, 2. Satisfied, 3. Neither nor, 4. Unsatisfied, 5. Very unsatisfied. Continuous variable, inverted scale.
Working hour satisfaction/ can reduce working hours
How satisfied are you with the length of your working week? 1. Very satisfied, 2. Satisfied, 3. Neither nor, 4. Unsatisfied, 5. Very unsatisfied. Continuous variable, inverted scale.
Could you reduce the number of paid hours in your regular work schedule? 1. Yes, 2. No. Dummy is set equal to 1, if the answer is 1.
Hard/harder to satisfy job demands
Do you find it hard to satisfy the job demands a) further training, b) use of new technology or c) readjustment to new tasks? 1. To a large extent, 2. to some extent,3. no. Dummy is set equal to 1, if the answer to a), b) or c) is 1 or 2.
Statement: My job requires me to do more difficult things than it used to. 1. Strongly agree, 2. Agree, 3. Disagree, 4. Strongly disagree. Dummy is set equal to 1, if the answer is 1 or 2.
Usual retirement age
Is there a fixed rule or customs or unwritten rules for the age at which withdrawal takes place in your position at your working place? If yes, at what age? Continuous variable for usual retirement age. If no such age, dummy for no usual retirement age is set equal to 1.
On your main job, what is the usual retirement age for people who work with you or have the same kind of job? Continuous variable for usual retirement age. If no such age, dummy for no usual retirement age is set equal to 1.
93
Table A.2. Definitionsa) and means of variables, men and women, Denmark (standard deviation in parentheses). Variable Age Education Health Partnership
Individual earnings Wealth Compensation rate
Missing compensation rate Early retirement benefits Labour market pension Private schemes Other income Experience Occupation for job with longest tenure Sector Tenure Hours of work Physical demanding Stress Job satisfaction Wage satisfaction Working hour satisfaction Hard to satisfy job demands Usual retirement age No usual retirement age Fixed effects Number of observations
Definition =1 if age equal to 57 Vocational training Higher education Poor Fair Single Partner of the same ageb) or older Partner of the same ageb) or younger Log average earnings 1995-1997, continuous variable Wealth in 1996 in 1.000 d.kr., continuous variable Expected compensation rate if withdrawal takes place in 1997, continuous variable =1 if compensation rate is missing =1 if most important expected income when retired =1 if most important expected income when retired =1 if most important expected income when retired =1 if most important expected income when retired Total years worked, continuous variable =1 if salaried worker or public servant =1 if employed in the private sector Tenure on current job, continuous variable Hours per week at current main job, continuous variable =1 if working in a job that is physical demanding =1 if working in a job that is stressful Continuous variable Continuous variable Continuous variable =1 if hard to satisfy job demands Continuous variable =1 if the usual retirement age is missing The fixed error component
0.40 0.49 0.23 0.17 0.16 0.09 0.49 -
Men (0.49) (0.50) (0.42) (0.37) (0.36) (0.29) (0.50) -
0.38 0.39 0.28 0.22 0.20 0.24 0.45
Women (0.48) (0.49) (0.45) (0.41) (0.40) (0.43) (0.50)
12.4
(0.32)
12.1
(0.32)
314
(549)
123
(304)
0.39c) 0.22
(0.19) (0.41)
0.52d) 0.23
(0.22) (0.42)
0.55
(0.50)
0.63
(0.48)
0.19
(0.39)
0.18
(0.38)
0.15
(0.36)
0.08
(0.27)
0.08
(0.28)
0.09
(0.29)
36.6
(5.14)
32.6
(5.85)
0.58 0.68
(0.49) (0.47)
0.81 0.35
(0.39) (0.48)
15.9
(10.5)
15.5
(9.40)
39.8
(6.99)
34.1
(7.03)
0.49 0.55 3.49 3.96 4.12 0.32 65.7e)
(0.50) (0.50) (1.47) (0.81) (0.87) (0.47) (3.49)
0.60 0.58 3.62 3.71 4.21 0.35 66.1f)
(0.49) (0.49) (1.41) (1.01) (0.90) (0.48) (3.80)
0.54 4.92
(0.50) (0.12)
0.58 4.69
(0.49) (0.24)
675
574
Notes: a) For a more precise definition of the subjectively assessed job characteristics, see Table A.1; b) “Same age”: +/- two years; c) Number of observations: 528; d) Number of observations: 440; e) Number of observations: 312; f) Number of observations: 242.
94
Table A.3. Fixed effects wage determination, 1993 to 1997, men and women, Denmark (standard errors in parentheses). Men Constant
Women
5.054
(0.045)
4.539
(0.060)
-0.156
(0.091)
-0.105
(0.047)
Vocational education No education Vocational/apprentice
(dropped)
(dropped)
Short higher
-0.199
(0.097)
0.026
(0.077)
Medium higher
-0.146
(0.069)
0.158
(0.138)
Long higher
-0.152
(0.184)
0.163
(0.174)
0.004
(0.019)
Sector Private
(dropped)
Public
-0.056
(0.023)
(dropped)
Not employed/missing
0.047
(0.097)
0.065
(0.095)
0.022
(0.034)
Industry Agriculture, fishing, raw material, extraction, manufacturing
(dropped)
Energy and water supply, construction
0.038
(0.042)
-0.074
(0.059)
Wholesale, retail, hotels, restaurants
0.131
(0.030)
0.028
(0.032)
Transportation, postal and telegraph services
-0.043
(0.034)
-0.114
(0.071)
Financing and business services
-0.002
(0.035)
-0.042
(0.030)
Public and private services
-0.018
(0.033)
(dropped)
Self-employed, assisting spouse
-0.009
(0.071)
0.035
(0.068)
Manager, high level wage earner
-0.077
(0.013)
-0.066
(0.011)
Medium level wage earner
-0.035
(0.013)
(dropped)
Occupation
Skilled worker
(dropped)
0.073
(0.007)
Unskilled worker
-0.025
(0.011)
-0.015
(0.012)
Not employed/missing
-0.055
(0.026)
0.058
(0.020)
-0.029
(0.045)
0.475
(0.130)
Place of residence Living in-/outside the Copenhagen Metropolitan area
R2 (within) Number of observations
3.3
9.3
3375
2870
95
Table A.4. Definitionsa) and means of variables, men and women, the U.S. (standard errors in parentheses). Variable Age Race College Health Partnership
Individual earnings Total wealth (excl. pension assets) Retiree health insurance Private pension age
No private pension Social security wealthe) Private pension wealth
Experience Occupation for job with longest tenure Sector Tenure Hours of work Physical demanding Stress Job satisfaction Can reduce hours Harder to satisfy job demands Usual retirement age No usual retirement age Fixed effects Number of observations
Definition Current age = 1 if black = 1 if (some) college or above Poor Fair Single Partner of the same ageb) or older Partner of the same ageb) or younger Log total earnings, continuous variable Total household non-pension assets in 1.000$, continuous variable =1 if have access to retiree health insurance Age for access to private pension; current age, if private pension age lower than current age, continuous variable =1 if private pension is missing Log total SS wealth at age 62, continuous variable Log total pension wealth age 62; mean log pension wealth if missing and has pension; 0 if no pension reported, continuous variable Total years worked, continuous variable Service or sales Blue collar workers =1 if public sector Tenure on current job, continuous variable Hours per week at current main job =1 if working in a job that is physical demanding = 1, if strongly agree = 1, if agree Continuous variable =1 if can reduce hours =1 if job requires more difficult things than it used to Continuous variable =1 if the usual retirement age is missing The fixed error component
56.9 0.07 0.48 0.34 0.32 0.12 0.40 10.3
Men (0.05) (0.01) (0.02) (0.02) (0.02) (0.01) (0.02) (0.07)
56.9 0.13 0.41 0.33 0.36 0.33 0.37 9.6
Women (0.05) (0.01) (0.02) (0.02) (0.02) (0.02) (0.02) (0.07)
195
(12.9)
173
(11.0)
0.47
(0.02)
0.32
(0.02)
58.7c) 0.52
(0.16) (0.02)
60.0d) 0.65
(0.21) (0.02)
6.74
(0.01)
6.31
(0.03)
8.75 37.9
(0.19) (0.18)
6.99 29.7
(0.21) (0.37)
0.19 0.45 0.08
(0.02) (0.02) (0.01)
0.55 0.11 0.05
(0.02) (0.01) (0.01)
15.21 43.6
(0.46) (0.38)
12.6 37.7
(0.37) (0.43)
0.66 0.15 0.44 2.99 0.26
(0.02) (0.01) (0.02) (0.03) (0.02)
0.64 0.25 0.40 3.08 0.30
(0.02) (0.02) (0.02) (0.03) (0.02)
0.53 62.7f) 0.25 2.67
(0.02) (0.14) (0.02) (0.12)
0.50 63.2g) 0.36 2.23
(0.02) (0.13) (0.02) (0.02)
718
752
Notes: a) For a more precise definition of the subjectively assessed job characteristics, see Table A.1; b) “Same age”: +/- two years; c) Number of observations: 341; d) Number of observations: 259; e) For measurement of this variable, see Section 9.2 in the Appendix; f) Number of observations: 538; g) Number of observations: 479.
96
Table A.5. Fixed effects wage determination, 1992, 1994 and 1996, men and women, the U.S. (standard errors in parentheses). Men Constant
Women
0.306
(0.561)
0.359
(0.217)
0.654
(0.120)
-0.171
(0.124)
0.496
(0.085)
0.366
(0.161)
0.200
(0.086)
Sector Public sector Industry Agriculture, forest, fishing, mining, construction Manufacturing
(dropped)
Transportation
0.311
(0.099)
0.129
(0.126)
Wholesale, retail
0.345
(0.078)
-0.032
(0.074)
Private services
0.383
(0.069)
(dropped)
Professionals
0.356
(0.069)
(dropped)
Sales
0.371
(0.097)
0.755
(0.091)
Clerical/ administrative support
0.074
(0.087)
0.692
(0.063)
Services
0.368
(0.091)
0.651
(0.068)
Farming, forestry, fishing
-0.340
(0.149)
1.417
(0.451)
Mechanics, operators etc.
(dropped)
0.885
(0.089)
Occupation
Union Union member
0.529
(0.067)
0.130
(0.072)
0.016
(0.002)
0.020
(0.003)
0.056
(0.031)
0.056
(0.015)
-0.000
(0.000)
-0.001
(0.000)
Tenure Tenure on current job Experience Total years worked Total years worked square 2
R (within)
25.7
27.4
Number of observations
2153
2253
97
98
Chapter 3 The Impact of Health on Individual Retirement Plans: a Panel Analysis comparing Self-reported versus Diagnostic Measures† Mona Larsen* and Nabanita Datta Gupta**
JEL Codes: I18, J14, J26. Abstract Earlier studies have concluded that the use of self-reported health in retirement models is likely to yield an unreliable impact of health on retirement due to “justification bias”. A few recent studies based on younger cohorts approaching retirement age have found little support for this hypothesis. This paper adds fresh evidence to this debate by considering the effect of health on retirement plans in samples of older workers and retirees drawn from a Danish panel survey from 1997-2002 merged to longitudinal register data. Using a wide array of alternative health measures, we compare the role of subjectively versus objectively measured health as a determinant of retirement planning. We control for unobserved heterogeneity as well as account for endogeneity and measurement error of health in retirement, and estimate separate models for women as well as men. As in the more studies, justification bias turns out not to be important. Self-rated physical and mental health are important predictors of retirement planning, in fact more important than economic factors, both among men as well as women. At a disaggregated level, back problems and myalgia significantly hasten male retirement, while back problems, osteoporosis and depression are conditions that significantly affect retirement among women. Retirement planning is in general unaffected by being hospitalised for a serious condition. Looking at health changes strengthens the conclusion that health is an important factor in retirement planning. In fact, health shocks seem to increase the propensity to retire earlier. However, health seems to be less important for retirement planning in Denmark compared to the US.
†
This work is part of the research of the Graduate School for Integration, Production and Welfare. Financial support from the Danish Social Science Research Council is gratefully acknowledged. This work benefited from comments by Richard Disney, Michael Hurd and Martin Browning and other participants at the Second Workshop of the RTN Project on the Economic of Ageing 2003 in Naples. *
Danish National Institute of Social Research; Aarhus School of Business and Graduate School for Integration, Production and Welfare, Herluf Trolles Gade 11, DK-1052 Copenhagen K,
[email protected].
**
Aarhus School of Business, Silkeborgvej 2, DK-8000 Aarhus C,
[email protected].
99
1. Introduction The future demographic changes that many countries around the world expect to face imply that it becomes increasingly important to add to our knowledge about the factors that affect the retirement decision. A number of previous studies have shown that health is an important determinant of preferences for retirement. Much of the available empirical evidence in this area suggests that poor health causes workers to retire earlier (Bound, 1991; Anderson and Burkhauser, 1985, Bazzoli, 1985; Dwyer and Mitchell, 1999; McGarry, 2002). Yet, from a theoretical perspective, poor health actually has an ambiguous effect on retirement (Sammartino, 1987). In part, developing a health problem can reduce productivity at work by affecting work capacity. In addition, workers may have to reduce hours or take sick days to attend to their health condition, which increases the value of leisure time relative to consumption. On the other hand, poor health can increase the value of consumption relative to leisure at the margin if developing an acute condition means that the expected work life is shortened, so that the worker now has fewer years to allocate between retirement and work (Grossman, 1972). While most previous studies have found that the former effects dominate the latter so that poor health brings on retirement, there is much disagreement as to the precise effects of health on retirement. This is because various econometric issues arise when health is proxied by way of survey-based self-reports or even by more objectively measured indicators. These issues have been detailed in a number of previous studies and therefore they are mentioned only briefly below (see for example, Benitez-Silva et al., 2000, Bound, 1991, Dwyer and Mitchell, 1999, and McGarry, 2002).
100
Subjective reports of health made by those already retired may lead to potential “justification bias”. That is, failing health is used as a socially acceptable excuse for retirement, rather than an accurate description of the reason why individuals leave the labour market. This likely biases the estimated impact of health on labour market outcomes and introduces a bias in the coefficients of any variables correlated with health too. Another consideration is that health may be endogenous to labour market outcomes. This might lead to mis-estimation of the effect of health if, for instance, withdrawal from the labour market improves or worsens health. This is a direct effect leading to overestimation of the effect of health. An indirect effect is generated if unobserved differences across individuals correlate with both health and retirement behaviour, for example, differences in workers’ time rate of preference. The inability to control for these variables could lead to omitted variable bias. There may also be correlation between health status and financial aspects. Since disability pension is only available for those who suffer from loss of working capability, some people will have a financial incentive to identify themselves as disabled, leading to a correlation between subjective health measures and financial variables. Thus, although the impact of health itself is correctly measured, the estimates of the impact of economic variables may be biased. Use of objective health measures might correct for this problem but objective health measures themselves may suffer from other problems, namely, that such health indicators need not necessarily measure the individual’s reduced ability to perform work, the relevant measure in any labour study. In fact, in most cases, objective measures are proxies for general health or presence of health conditions rather than work capacity. For example, a measure such as subsequent mortality typically suffers from this measurement error problem, particularly as mortality often occurs abruptly or following a short-lived serious illness 101
and is therefore not a good measure of current work capacity. Conversely, many chronic conditions such as arthritis may severely limit one’s ability to work, but have less of an effect on life expectancy, which may be important if workers form expectations taking into account their expected remaining time horizon. Finally, there need not be any relationship between the level of current health and retirement probabilities, if individuals and jobs are generally well matched so that employers tend to accommodate worker’s health problems at the workplace, or workers themselves undergo neutralizing treatment or lifestyle changes. Rather, it might be the onset of diseases or conditions, or sudden or gradual changes in health that likely influence retirement decisions. Therefore, in addition to analysing the effect of the level of health, the role of health changes is also examined.
These considerations pose a considerable challenge to empirical economists interested in quantifying the impact of health on retirement and other labour market outcomes. Our main purpose is to assess the importance of precisely measured health relative to economic factors in particular, on planned retirement behaviour in Denmark taking careful account of the problems inherent in obtaining a precise estimate of health impacts outlined above. By combining a repeated survey of elderly individuals to longitudinal register data on labour market information, we are able to create a panel sample of older individuals for whom we have data on planned retirement age, health, income, job characteristics, labour market and background measures. This gives us a unique opportunity to quantify for the first time the effect of subjective as well as objective (including diagnostic-based) measures of health on retirement in Denmark. In doing so, we attempt to build upon the work in previous studies by taking into account various is-
102
sues that arise in the measurement of health that were discussed above. For instance, similar to McGarry (2002), by using repeated observations on planned retirement age we are able to control for unobserved heterogeneity across individuals that may be correlated with both health and retirement and thereby cause omitted variable bias. However, going beyond McGarry, we also account for the endogeneity of health and measurement error as in Dwyer and Mitchell (1999), but within a framework of a panel model instead of the static model used in that study, allowing us to simultaneously account for both unobserved effects and endogeneity. Further, we employ a wide array of health measures in this study, both subjective and objective, including information of actual diagnoses made at the time of hospitalisation extracted from the Danish National Patient Registry (LPR, Landspatientregister). By employing a wide array of health measures, including self-rated physical and mental health, health compared to others, work capacity, work and functional limitations, presence of diseases conditions and episodes of hospitalisation caused by a serious condition, we are able to provide outer bounds on the “true” effect of health on retirement spanned by subjective and objective measures. This method has been suggested by Bound (1991) and has been implemented by Dwyer and Mitchell (1999) and Kreider and Pepper (2001, 2002). In addition, one methodological improvement on previous studies is that in some specifications we experiment with a ordered probit panel model which allows for discontinuities in planned retirement age that may arise if the data are bunched around the early and normal retirement ages (see for example Figure 2 below). Thus, we do not necessarily assume like previous studies (Dwyer (2001), Dwyer & Mitchell (1999) and Dwyer & Hu (1998)) that planned retirement age is a smoothly distributed variable. Finally, and importantly, while most previous studies have concentrated on the impact of health on retirement be103
haviour of men, we conduct separate but symmetric analyses on samples of men as well as women, providing for the first time, precise estimates of the effects of health on female retirement.
The focus in this study is to estimate the effect of health status on planned retirement age in Denmark, cf. above. Denmark is interesting in this context because unlike for instance health care arrangements in the US, under the Danish welfare system, health insurance is universal and access to most health services is free for all regardless of economic situation. In addition, health-related exit from the labour market is possible in Denmark through Social Disability Pension (SDP, førtidspension).37 These differences between the two countries imply that we expect that health is a less important factor in retirement planning in Denmark compared to the US. First, the universal system in Denmark might imply that older Danes in the labour force are healthier than similar American due to easier access to preventive and neutralizing health care services. Second, as a consequence of the existence of SDP, older Danes in the labour force probably constitute a more selected group than comparable Americans since older Danes with very poor health have already withdrawn themselves from the labour market.
37
SDP is residence based. Former work or contribution record is not required. To be awarded SDP, recipients must be 18 to 66 years of age (18 to 64 from 2004). The basic pension is differentiated according to marital status and is means tested against a wider range of other income sources. Conversely, the pension is complete independent of former work and income. In 2003, a reform of SDP was conducted with big simplification of the benefit structure. Before 2003, the pension was graduated according to loss of working capability and varied according to the age of the first time recipient.
104
While earlier U.S. studies based on Retirement History Longitudinal Survey (RHS) data have concluded that the use of self-reported health in retirement models is likely to yield an unreliable impact of health on retirement due to the presence of “justification bias”, we find little support for the justification hypothesis in these data. Our findings are therefore more in line with the recent U.S. studies based on younger cohorts approaching retirement age drawn from the U.S. Health and Retirement Study (HRS) data (Benitez-Silva et al., 2000, Dwyer and Mitchell, 1999, McGarry 2002). We also find that self-rated physical and mental health are important predictors of retirement planning, in fact as important as economic factors, both among men as well as women. Again, these results confirm recent studies from the U.S. that find that health problems influence retirement plans more strongly than economic variables. For instance, Dwyer (2001) finds that economic factors such as net worth, pensions, and social security do not play a big role in retirement outcomes conditional on plans and McGarry (2002) finds that self-rated health in particular is a powerful predictor of retirement behaviour. Finally, as expected health seems to be a less important factor in retirement planning in Denmark than in the US.
The rest of the paper is organized as follows: Section 2 gives some descriptive evidence and institutional background on health and retirement in Denmark. Section 3 presents our empirical model, Section 4 the data and Section 5 the results of the estimation. Section 6 offers some concluding remarks.
105
2. Health and retirement in Denmark In Denmark, as in most other OECD countries, the ageing of the population implies that the elderly will constitute an increasing share of the population in the coming decades. The projection for Denmark for the next 35 years is that there will be a decline in the number of people in the active population relative to the number aged 65 and above, from a ratio of 4.5 to l to a ratio of 3 to 1 (see Schaumann (2001)). In a scenario of smaller entering birth cohorts and larger cohorts exiting to retirement, precise estimates of the effect of health on retirement become increasingly necessary in order for governments and policy planners to plan the right amount and types of investments in healthcare that would support an aging workforce.
One important health-care systemic difference compared to for instance the U.S. is that health care services in Denmark are publicly financed through general income taxes and for most of the services, offered directly by the public authorities. Thus, unlike in the U.S. where availability of insurance is an important predictor of retirement, health insurance is as mentioned above universal and access to most health services is free for all regardless of economic situation.
Health-related exit from the labour market is possible through SDP, cf. Section 1. At the end of 2002, 7 per cent of people aged 18-66 years received SDP. Among the 50-59-year-olds, the share was 12 per cent. This share was about the same in the middle of the 1980’s, cf. Figure 1 below.
106
Figure 1. 50-59-year-olds on SDP, men and women, 1984-2002. Per cent of 50-59-year-olds in the population. Per cent 16 14 12 10 8 6 4 2 0 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 Men
Women
Early retirement though SDP is most prevalent among women in this age group. As SDP in Denmark in the decades of the 1980s incorporated as well retirement for individuals who fulfilled certain social criteria, the share of men and the share of women increased in the eighties. For women, the cancellation of the widow’s pension in 1984 contributed to a corresponding take-up of SDP, particularly among widowed women. However, in the nineties, the share of women in disability decreased much more than the share of men. That is, the share of women decreased 8 per cent from 1984 to 2002, while the share of men increased 12 per cent. Part of the explanation of this gender difference might be the prevalence of the Transitional Benefit Program (TBP, overgangsydelse) in the mid-nineties that was available for people aged 55-59 years (from 1994 also 50-54-year-olds) who were members of an unemployment insurance fund and who had been unemployed for at least 12 out of the last 15 months. The existence of this scheme probably contributed to the reduction in the inflow of people in to SDP. In particular, the inflow of women was reduced since two out of three individuals who received TBP benefits were women. During the 1990s, 107
changes made to improve work-disabled peoples’ attachment to work evolved around the idea of ‘activation’. Thus, by emphasizing active measures such as vocational rehabilitation, the dependency on social security benefits was reduced, cf. Høgelund (2003). Furthermore, subsidized jobs on special terms were introduced in the late nineties as a means of reducing the inflow to SDP of people of working age (fleksjob) and to increase the labour force by enabling SDP pensioners to re-enter the labour force at the same time as they maintain their SDP benefits (skånejob). These policy changes also contributed to the decrease until 2001 in the share of men and women that received SDP benefits. However, these shares increased from 2001-2002. This trend, which was seen across age groups, seems to be due to an effort by the municipalities directed towards closing as many cases as possible before a reform of the SDP came into force 1 January 2003. Comparable trends are seen in connection with an earlier reform within this area, cf. National Social Appeals Board (2003). SDP is relatively expensive for the government. Thus, the cost of this program amounted to 2.2 per thousand relative to GNP in 2002. SDP benefits to people aged 50 or above amounted to 1.4 per thousand. By way of comparison, it is worth mentioning that the cost of Old Age Pension (OAP, folkepension) and early retirement schemes38 amounted to 4.2 and 1.6 per thousand, respectively.
A few previous Danish studies have also estimated the effect of health on retirement, employing either duration or option-value models of retirement. The general finding seems to be that bad health affects early retirement positively. However, while Christensen & Datta Gupta (2000) and Peder38
In this context, early retirement schemes include TBP and Post Employment Wage (PEW, efterløn), cf. the description of the latter in Section 4.
108
sen & Smith (1996) show that bad health has a positive impact on early retirement for both men and women, Danø et al. (1998, 2000) find that bad health seems to hasten retirement for men, while the effect is insignificant for women. The first two studies define their health measure on a single index, which is based on whether or not the worker qualifies for sickness insurance, typically granted after a sickness spell of a minimum of 13 weeks. The study by Danø et al. base health on the number of physician visits in a year, not accounting for the reason for visit. None of these studies however takes into account the issues of endogeneity and measurement error of health.
3. Empirical model Following Bound (1991), Dywer and Mitchell (1999) and other previous studies, we consider a model of the retirement decision of individuals approaching retirement age, in which the (continuous) planned retirement age of individual i at time t, Rit, depends on economic factors, health status and demographic variables: R it = β w it + γ H it* + λ Z it + ε it .
H* is unobserved health status, w are economic incentive variables, Z are demographic factors and ε is a random disturbance term. Although H* is unknown, we observe alternative proxies of underlying health, a subjective measure, Hs, and an objective measure, Ho. The subjective measure depends both on underlying health status H* and on economic incentive measures w such that:
109
H its = δwit + θ s H it* + µ its
The latent objective measure on the other hand, does not depend on w. H ito = θ o H it* + µ ito
Assume that H* is orthogonal to ε, µs and µo. If “justification bias” is present, then ε and µs are correlated. For example, an individual planning to retire earlier than average may try to justify this by reporting themselves to be in worse than average health, leading to simultaneity bias. Another problem is that as Hs is not a perfect predictor of H*, measurement error will bias the coefficient to H*, γ. Bound (1991) shows that as long as the correlation between ε and µs is positive, it will tend to bias the coefficient to H*, γ, upwards, while variance in µs leads to the classic errors-in-variables problem and will tend to bias it downwards. The net effect on γ depends on the relative magnitudes of these two biases, but even if on balance they cancel out, there will in general still exist a downward bias on β arising from the dependence of Hs on w i.e. δ. This means, that while the coefficient to the subjective proxy despite the rationalization problem may end up in practice close to the true effect of health on retirement, the coefficients to the economic variables will tend to be smaller than they should be.
There are problems inherent in using the objective measures as well. Here, justification bias is not present, µo is uncorrelated with ε, and further, Ho is not dependent on w, but as long as it is not a perfect predictor of underlying health, the use of it will tend to underestimate the effect of health and overestimate the effects of economic variables. Solutions to identification may 110
lie in finding instruments for the subjective and objective measure (Dwyer and Mitchell, 1999). Or, in limiting the analysis to a sample of workers only (McGarry (2002)). Other solutions could be to find exogenous sources of variation in w or changes in the structure of retirement benefits that could identify the underlying parameters, or even in using information on the reliability of self-reports. In this paper, we employ the instrumental variables (IV) method to solve the identification issues discussed above.
More recently, as subsequent waves of the HRS have become available, some authors have applied longitudinal data to the question of the effect of health on retirement plans.
McGarry (2002) estimates a fixed effects
model of the subjective probability of continued work on workers only and finds strong effects of subjective measures of health even on a sample for which justification bias is purportedly low. While the use of longitudinal data in itself does not solve the identification issues that arise in the cross section, it does allow purging the data of unobserved individual effects, which may be correlated with health and retirement. Examples of such effects mentioned by McGarry (2002) may be individual’s time-rate of discount, tastes for work or even time-constant measurement error in health. The fixed effects model allows for individual effects to be correlated with the included variable. However, the effects are estimated as strict parametric shifts of the outcome function. Thus, the model is more appropriately applied to settings in which the sample constitutes the population under study, for example, an inter-country comparison which includes the full set of countries, cf. Greene (2003, p. 293). As most labour studies are based on smaller samples drawn from a large population, a more appropriate econometric specification models the error term as randomly distributed across
111
individuals i.e. the random effects model, and therefore this model is the chosen statistical framework.39 Pooled OLS and random effects models are tested against each other by way of Breusch-Pagan LM and likelihood ratio tests, which are based on OLS and random effects residuals respectively, as tests of the random effects.
The econometric model is specified as follows. In the case of subjective health measures, we have the following system of equations:
R it = β s w it + γ s H its + λ s Z it + η is + ε its
H its = π s Yit + µ its
In the case of the objective measures, the model is given by the system
R it = β o w it + γ o H ito + λ o Z it + η io + ε ito
39
However, the random effects model is restrictive in that the unobserved individual effect is assumed to be uncorrelated with the included regressors, and in fact, a Hausman specification test shows that this assumption is only fulfilled in the case of men. However, the fixed effects approach which allows correlation between the unobserved component and the observed variables, is costly in terms of degrees of freedom lost which is particular problematic in our case since our sample is small and since T is equal to 2. Consequently, using fixed effects in this case corresponds to throwing away half of the observations. The problem is illustrated by the fact that although Hausman specification test suggests that both the fixed and the random effects approach are consistent in the case of men, results based on these two approaches differ considerably. In fact, very few of the estimates based on the fixed effects approach are significant, while several of the estimates from the random effects specification are significant and have the expected signs. The results for women based on the fixed effects approach suffer from similar problems (see footnote 47).
112
H ito = π oYit + µ ito .
The error components each are assumed to be mean-zero with covariance matrix
σ η2 I N η E ε (η ′, ε ′, µ ′) = 0 µ 0
0
σ ε2 I NT 0
0 0
σ µ2 I NT
.
Note that as true health status H* is unobserved, the subjective and objective health measures are used alternatively as proxies for it. However, to account for the endogeneity and measurement error, we employ an instrumental variables estimation approach where Y are a set of exogenous instruments spanning the relevant health measure. The parameters of this random effects IV model are derived by way of the full-information G2SLS estimator developed by Balestra and Varadharajan-Krishnakumar (1987).
If justification bias is present, then subjective health measures should have a big impact and economic variables a small impact on planned retirement age. In that case, instrumentation should reduce the impact of the subjective variables and increase the relative importance of the economic variables. On the other hand, if measurement error is a problem with respect to the objective measures, then the impact of health will be small and economic effects big, and instrumentation should increase health effects and reduce effects of economic variables on planned retirement. 113
4. Data and descriptives The primary data used in this study are obtained from a longitudinal database of elderly people (Ældredatabasen), a survey which was fielded and collected by the Danish National Institute of Social Research. The database consists of two waves of survey data from 1997 and 2002. Thus, repeated observations over time enable us to obtain knowledge about how individuals update their retirement plans when new information arrives, particularly with respect to the role of health on planned retirement age. The survey data are merged with longitudinal register data from 1993 to 2001, in order to supplement the database with information on individual’s labour market characteristics (the economic variables).
The first survey was conducted in 1997 face-to-face in the homes of a representative sample of individuals born every fifth year from 1920 to 1945. 5,864 individuals from the six cohorts were interviewed. The response rate was 70 per cent. The second survey was conducted primarily by phone. The same respondents were contacted for a second interview.40 79 per cent of the first wave respondents participated in the second wave. Thus, 4,634 individuals form part of both waves.
40
In addition, a representative sample of individuals born in 1950 and new respondents born every fifth year from 1920 to 1945 to replace the attrition in the first wave were contacted for an interview.
114
In order to minimize sample selection due to retirement, the sample used in this paper is limited to individuals born in 1940 and 1945. That is, people aged approximately 52 and 57 years in 1997, which corresponds to 2,259 individuals, who are observed again in 2002. The sample is restricted to individuals that were in the labour force in the first wave. Thus, one source of potential sample selection is that we omit individuals who were already retired in 1997. This exclusion of individuals outside the labour force in Wave 1 can be problematic if transition to SDP and early retirement schemes are self-chosen. However, we lack key information on some health measures and many relevant economic characteristics (see Appendix A3, Table A1), which therefore does not allow the possibility of including these individuals in the analysis. Instead, a full comparison of means is made of those in the labour force in Wave 1 to those out of the labour force in Wave 1. Note that we do include individuals who retired between the two waves in our sample, for purposes of maximizing sample size, as we have full information on all covariates on these individuals.41 These restrictions leave us with a sample of 1,834 individuals. Finally, since we compare retirement plans in 1997 and 2002 we are only able to include individuals that report a given age in both years when asked about planned retirement age, see below. Unfortunately, a large share of individuals does not meet this demand42 and therefore, we end up with a sample that consists of 1,103 individuals. That is, 49 per cent of the original sample.
41
Planned retirement age is set equal to the actual retirement age (reported in the survey) for these individuals.
42
Instead, their answer was “don’t know” or “as long as possible”. It is of course feasible to set “as long as possible” to a maximum of e.g. 75 or 85 years, as in some previous studies (Dwyer and Mitchell, 1999). However, we hesitate to do so, as this latter group of individuals turns out to be a very heterogeneous group whose characteristics do not in general resemble those who cite a high retirement age when queried. Nevertheless, as
115
Planned retirement age is the dependent variable in our analyses. We treat this generally as a continuously distributed variable, but we do sensitivity analysis in which this variable is grouped to account for the concentrations of responses at age 60, 62 and 65, cf. Figure 2 below, which in turn reflect pension policy rules and eligibility criteria. Thus, we distinguish between six categories of planned retirement age: below 60, 60-61, 62, 63-64, 65-66 and 67 and above. Ages 60 and 67 were the early and the normal retirement ages in 1997, respectively. Furthermore, if people waited until the age of 63 to take up PEW, benefits were increased. In 2002, the incentives to retire at age 62 were strengthened. Thus, changes in PEW, which is the most popular early retirement scheme in Denmark, were put in effect in 1999. Among other changes, means testing of benefits against income from labour market pension and lump sum retirement income payment for people aged 60 and 61 were introduced. A tax premium was introduced as well for people that were entitled to PEW benefits at age 60 but who continued working until at least the age of 62. At the same time, it was decided that the official age of retirement would be lowered from 67 to 65 years of age from 2004. Thus, in 2002 the normal retirement age was 65 for the individuals in our survey.
A large shift in planned retirement age took place from 1997 to 2002 among the individuals in our sample, cf. Figure 2. In 1997, retirement
a check of the robustness of the results, “as long as possible” is tentatively set equal to 75, on the assumption that that people enjoy working and want to maintain a relationship to the labour market as long as possible, and in an alternative experiment, to 67, which might apply if people are financially constrained and would therefore need to continue working until the normal retirement age. By including individuals that answer “as long as possible”, the sample size increases from 1103 to 1331 individuals. However, including these individuals does not change the estimates of any of the health coefficients significantly using either definition.
116
around the early retirement age was more or less the norm among women in particular but also among men born in 1945. However, from 1997 to 2002 the shares of individuals that planned to retire at age 60 and to a minor degree at age 67 decreased, while the shares that planned to retire at age 62 and 65 respectively increased. The shift to planned retirement at age 65 was most pronounced among people born in 1940, while the shift to planned retirement at age 62 first took place among people born in 1945. In general, the changes were most pronounced for men. Part of the changes in planned retirement is probably due to the upward adjustment that normally takes place, when people approach retirement age, cf. e.g. Dwyer & Hu (1999). However, it also seems reasonable to believe that these changes at least to some extent reflect the policy changes described above.
It is potentially difficult to isolate the effect of health on retirement plans in an environment in which changes in pension policy are taking place simultaneously. However, as the pension wealth accrual variable is updated to reflect the expected value of the pension policy change and given that the correlation between changes in pension policy and health is small, we are able to identify the effects of health in a period that spans the changes in PEW eligibility and generosity described above.43
43
One source of correlation between health and pension policy reform could be if individuals are more likely to report themselves in poor health following a pension reform that increases eligibility age in an attempt to justify early exit. However, the PEW reform of 1999 kept the first available age of early retirement at 60 but simply increased (the already present) financial incentives of later retirement at age 62 or later. Thus, retirement at 60 is still a legally available and much used option and there are no social stigma attached to retiring at 60. The reform merely makes it more financially attractive for individual to defer retirement age to 62 or later.
117
Figure 2. Planned retirement age 1997 and actual and planned retirement age 1998-2002, per cent, (a) men born in 1940; (b) women born in 1940; (c) men born in 1945; (d) women born in 1945. Figure 2a)
Figure 2b)
Per cent 70 60 50 40 30 20 10 0
Per cent 70 60 50 40 30 20 10 0
57 58 59 60 61 62 63 64 65 66 67 68 69 70 (Planned) retirement age 1997
1998-2000
1997
Figure 2c)
1998-2002
Figure 2d) Per cent 70 60 50 40 30 20 10 0
Per cent 70 60 50 40 30 20 10 0 52
57 58 59 60 61 62 63 64 65 66 67 (Planned) retirement age
54
56
58 60 62 64 66 (Planned) retirement age 1997
1998-2000
68
70
51
53
55
57 59 61 63 65 (Planned) retirement age 1997
67
69
1998-2002
Definitions of the explanatory variables included in the analyses are provided in Table A1 in Appendix A3. Poor health is proxied alternatively by 8 separate measures, 4 subjective and 4 objective. The subjective variables are self-reported general health, self-reported mental health, self-rated health compared to others and self-rated work capacity. Among the objective measures, we have the presence of work limitations, the presence of functional limitations, the presence of a disease condition and the most objective measure of all, presence of a (serious) condition diagnosed at the time of hospitalisation, based on information obtained from LPR records
118
and merged to the panel sample.44 A detailed description of health measures can be found in Appendix A3, Table A1. Other key explanatory variables include individual labour market earnings and other income, wealth and a variable capturing pension wealth accrual measured as a replacement rate, the compensation rate (see Appendix A2 for a description of how the compensation rate is calculated). Additional controls include birth cohort (1940 or 1945), vocational education, cohabitation status, age difference between partners, experience, occupation, sector, tenure, hours of work, whether the job is physical demanding, level of job satisfaction and a dummy variable for missing compensation rate.
Descriptive statistics for these variables are also included in Table A1. Looking at gender differences among the individuals in the labour force in Wave 1 (the estimation sample), planned retirement age is significantly higher for men (61.68) than women (60.86) in our sample, but for both groups, on average, individuals in the labour force plan to retire through some form of an early retirement program. The mean values for general health are similar by gender with 17% of women and 18% of men reporting themselves to be in poor health and 5% of each gender group rating their health to be worse than others. However, women report having significantly worse mental health than men, and a higher share of women report having functional limitations that bother them in their daily life. Also, more women than men are bothered by at least one disease condition, particularly, osteoarthritis, myalgia, osteoporosis and depression.
Conversely,
men to a larger extent are hospitalised with a serious condition and are 44
Instead of 0-1 dummies, counts of disease conditions, diagnoses and work limitations have been tried and results remain robust to these alternative specifications.
119
more likely to be diagnosed with heart diseases, diabetes and high blood pressure than women. As regards the remainder variables in our model, more men than women are vocationally trained, and more women than men live alone, reflecting in part gender mortality differences. Turning to financial and job characteristics, men have higher earnings and other income and higher average wealth, although the dispersion in the wealth variable within the group of men is much larger. Conversely, the compensation rate from pensions is on average expected to be higher for women than for men when withdrawal takes place, indicating that women’s labour market earnings and other income is typically lower. Unsurprisingly, men have higher experience, tenure and a longer working week than women on average. In terms of occupation of the job with the longest tenure, the average male in our sample to a greater extent has been self-employed/assisting spouse or a skilled/unskilled worker, while the typical women has been a salaried worker or a public servant.
Turning next to in-out of-labour force differences, we see that both men and women outside the labour force have significantly worse health as measured by all the self-reported subjective proxies, as well as in terms of having any functional limitation, disease or diagnosed condition (including heart conditions, strokes, lung diseases for men, and heart conditions, lung diseases, diabetes and arthritis for women). The finding of poorer selfreported health among the individuals outside the labour force supports the justification bias hypothesis. On the other hand, that fact that individuals outside the labour force tend to suffer more from underlying diseases and diagnosed conditions, imply that they are to some extent probably in worse health than individuals in the labour force. Mean age is also higher in this
120
group than corresponding men and women in the labour force. In addition, men outside the labour force have considerably less higher education, are more likely to live alone, have lower income on average, considerably less wealth and less labour market experience than men in the labour force. Similarly, women outside the labour force have less vocational training, and a significantly lower fraction of them have higher education. They are also more likely to be living alone or have a partner who is the same age or younger (baseline is having an older partner), have lower average income and wealth and less experience than their counterparts in the labour force. In sum, there are sufficiently many differences between the two groups, indicating that our findings should be appropriately applied to individuals approaching retirement age rather than those who have already withdrawn themselves from the labour market. On the other hand, the descriptive evidence also indicates that by excluding the already retired in Wave 1, the estimated impact of health on retirement is less likely to reflect justification bias.
Correlations between health measures are shown in Appendix A3, Table A2 for men, and A3 for women. In general, correlations between measures are not high except for a few notable exceptions. The correlations between self-reported measures appear higher (upper left quadrant), for instance for men (women), the correlation between general health and mental health is 0.32 (0.24) and between general health and health compared to others is 0.38 (0.41) which suggests that these should be considered as alternative ways of measuring individuals’ underlying health. On the other hand, correlations between the (self-reported) diseases conditions appear small except for conditions that affect the same part of body or system or share the
121
same pathological causes for example, between back problems and myalgia or back problems and osteoarthritis. There appears to be no correlation between individual diagnosed conditions. However, both for diseases conditions as well as diagnosed conditions, the aggregated variable that is defined to be the presence or absence of any condition, is correlated with the separate components. Overall, these considerations point to treating the various available measures as alternative proxies for underlying health.
5. Results of estimation Results of the analysis of the factors that affect retirement planning are presented in Tables 1-20. In particular, we focus on the role of health compared to economic factors, and therefore report only the relevant coefficients.45
5.1. Pooled OLS analysis To examine the role of health in retirement planning, pooled OLS models are estimated for men and women separately, as a benchmark case against a more general model that accounts for unobserved heterogeneity in Section 5.2. The findings from this simple specification in which the various health measures (subjective and objective) are treated as alternative proxies for underlying health H*, show that being in poor health in general reduces planned retirement age for men and women as almost all measures of poor
45
Estimated coefficients on the other explanatory variables are available and can be provided on request.
122
health are estimated to have negative impacts on retirement age, cf. Table 1 and 2 below. In contrast, as may be expected, higher income increases planned retirement age for both men and women while greater wealth and a higher compensation rate of pensions reduce it for men.46 That is, individuals that were wealthy in the year prior to the survey year are more inclined to precipitate retirement than others. However, this result only applies to men. For women, while wealth in most cases has the usual negative effect, the compensation rate has unexpectedly positive effects on planned retirement age, but in general, neither of these effects is significant. Interestingly, for both men and women, economic factors are highly stable and are estimated to have the same impacts irrespective of the health measure being considered, indicating an absence of correlation between health and these factors.
The subjective health measures are in general highly significant, while the objective measures not, and this holds for both men and women. The only objective measures that are significant are having a disease condition for
46
Other factors also affect planned retirement age. In this case, the results are very similar for men and women. As expected, adjustment in an upward direction takes place, when people approach retirement age. Unsurprisingly, people with higher education are more inclined to increase their planned retirement age than people without any qualifying education. Further, an increase in job satisfaction increases the planned retirement age as expected while an increase in the physical job demands decreases this age. Women living alone are more inclined to increase their planned retirement age than partnered women. This result probably reflects that single women have a greater preference for work than married women, see also Larsen (2004). Finally, planned retirement age is lower for men with a partner about the same age or younger. A potential explanation is that the desire to retire is stronger when one’s partner is expected to retire at the same time or before oneself.
123
men, and the presence of work limitations for women. On the face of it, this result could match the predictions by Bound (1991) that subjective measures tend to be inflated in the presence of justification bias while objective measures are weakened by measurement error. A formal test of this hypothesis is carried out in Section 5.3 where endogeneity and/or measurement error in health is corrected for by way of IV analysis.
Table 1. Pooled OLS estimates of the effect of health and economic factors on men’s retirement age. General health Poor health Individual income Wealth Compensation rate Adjusted R2 Number of observations
-1.384*** (0.189) 0.603*** (0.143) -0.129* (0.063) -0.372 (0.386) 16.1
Subjective health Objective health Mental Health Working Work Functional Disease Diagnoses health compared capacity limitations limitations conditions to others -1.260*** -1.318*** -0.125* 0.198 -0.409 -0.715*** -0.410 (0.378) (0.335) (0.050) (0.188) (0.421) (0.173) (0.281) 0.625*** 0.645*** 0.605*** 0.635*** 0.637*** 0.640*** 0.634*** (0.145) (0.145) (0.146) (0.146) (0.146) (0.145) (0.146) -0.130* -0.116 -0.133* -0.140* -0.137* -0.124 -0.130* (0.064) (0.064) (0.064) (0.064) (0.064) (0.064) (0.064) -0.549 -0.543 -0.466 -0.600 -0.563 -0.568 -0.581 (0.392) (0.391) (0.395) (0.394) (0.394) (0.391) (0.393) 13.1 13.4 12.8 12.4 12.4 13.6 12.5 1224
Note: Additional controls include birth cohort, education, cohabitation status, age differences between partners, experience, occupation, sector, tenure, hours of work, whether job is physical demanding, job satisfaction level and dummy variable for missing compensation rate. * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
How does health compare to economic factors? To be able to compare the predicted estimates, we standardize changes in the continuous explanatory variables by computing the estimated impacts on planned retirement age of a 1 standard deviation change in each explanatory variable. We find for men, focusing only on the significant variables, a one standard deviation increase in (log) income increases planned retirement age by 0.33-0.35 years, while a one standard deviation increase in wealth reduces it by 0.140.16 years. Looking next at the estimated coefficients on the health meas124
ures that are significant, being in poor general health reduces men’s planned retirement age by 1.4 years, while a one-unit worsening in mental health or being in worse health than others reduces planned retirement age by 1.3 years. A one-point decrease in working capacity (10-point scale) reduces planned retirement age by 0.13 years while being hospitalised for a serious condition reduces planned retirement age by 0.72 years. Thus, except for the work capacity, health seems to have a larger impact than wealth or income. Dwyer and Mitchell (1999) also find that health effects are larger than economic incentive effects in a similar analysis based on cross-sectional HRS data. In their study, those in poor health plan to retire about 2 years earlier than those in better health. Thus, (general) health effects are about half as strong in Denmark. In other words, our expectation that health is a less important factor in Denmark compared to the US seems to be confirmed.
Table 2. Pooled OLS estimates of the effect of health and economic factors on women’s retirement age. General health Poor health Individual income Wealth Compensation rate Adjusted R2 Number of observations
-0.720*** (0.178) 0.518** (0.196) -0.003 (0.166) 0.572 (0.401) 16.0
Subjective health Objective health Mental Health Working Work Functional Disease Diagnoses health compared capacity limitalimitations condito others tions tions -1.073*** -0.642* -0.180*** -0.364* -0.510 -0.112 -0.175 (0.330) (0.319) (0.043) (0.167) (0.305) (0.144) (0.374) 0.564** 0.550** 0.486* 0.573** 0.540** 0.558** 0.561** (0.196) (0.197) (0.196) (0.197) (0.197) (0.197) (0.197) -0.004 0.005 0.015 -0.015 0.000 -0.016 -0.008 (0.166) (0.167) (0.166) (0.167) (0.167) (0.167) (0.167) 0.561 0.628 0.616 0.638 0.597 0.637 0.633 (0.402) (0.403) (0.400) (0.403) (0.404) (0.404) (0.404) 15.5 14.9 16.1 15.0 14.8 14.6 14.6 982
See Table 1 for notes. * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
125
For women, a one standard deviation increase in income increases planned retirement age by between 0.20-0.24 years, while wealth and compensation effects are not significant. Being in poor general health reduces female retirement age by about 8 months, a one point increase in the (poor) mental health index reduces it by 1.07 years, while being in worse health than others reduces planned retirement age by two-thirds of a year. A one-point increase in working incapacity reduces planned retirement age by 0.18 years while having at least one limitation that affects work reduces retirement age by little more than 1/3 of a year. Having a disease condition or being hospitalised for a serious condition does not significantly impact women’s planned retirement age. Here again, the significant health effects are stronger than income effects on retirement. However, health effects are not as strong as those for men.
5.2 Unobserved heterogeneity In this section, results from the random effects specification are reported and compared to the simple pooled OLS model in Section 5.1 by way of Breusch-Pagan LM and likelihood ratio tests. In all models considered, for both men and women, the Breusch-Pagan LM and the likelihood ratio test statistics reported in Tables 3 and 4 below clearly indicate that the null hypothesis of no unobserved heterogeneity is strongly rejected. Results from the random effects specification on the health variables are qualitatively similar to those derived from a pooled OLS model, indicating that unobserved heterogeneity, although present, is not large in the case of the health variables. For men, one difference seems to be that working incapacity is no longer a significant determinant of retirement planning. Accounting for
126
unobserved heterogeneity seems to be more important in the case of the economic variables. For example, for men, wealth and the compensation rate are now estimated to have smaller impacts than before, roughly half the magnitude of the OLS effects. The relative importance of health to income is unchanged, while health becomes even more important compared to wealth and the compensation rate. The precise effects of health can be summarized as follows: For men, being in poor general health reduces planned retirement age by 1.3 years, while a one-point increase in the (poor) mental health index reduces planned retirement age by 1.2 years. Having health worse than others decreases planned age of retirement by 1.2 years while having a diseases condition means that males adjust their planned retirement age down by about ½ a year. In comparison, a one standard deviation increase in (log) income increases planned retirement age by 0.33-0.35 years.
Table 3. Random effects estimates of the effect of health and economic factors on men's retirement age. General health Poor health Individual income Wealth Compensation rate R2 (overall) BreuschPagan χ2 Likelihood ratio test Number of observations
-1.266*** (0.198) 0.604*** (0.146) -0.092 (0.068) -0.144 (0.358) 16.8
Subjective health Mental Health health compared to others -1.219** -1.221*** (0.406) (0.348) 0.617*** 0.638*** (0.148) (0.148) -0.091 -0.084 (0.069) (0.069) -0.238 -0.254 (0.361) (0.361) 13.9 14.2
Working capacity -0.067 (0.053) 0.615*** (0.149) -0.092 (0.069) -0.207 (0.364) 13.4
Work limitations 0.172 (0.186) 0.627*** (0.149) -0.095 (0.069) -0.274 (0.363) 13.2
Objective health FunctioDisease Diagnoses nal limi- conditions tations -0.428 -0.546** -0.463 (0.431) (0.177) (0.257) 0.631*** 0.643*** 0.624*** (0.149) (0.148) (0.148) -0.094 -0.087 -0.088 (0.069) (0.069) (0.069) -0.247 -0.258 -0.256 (0.362) (0.361) (0.361) 13.2 14.3 13.2
101.6***
109.8***
109.0***
106.5***
110.8***
111.2***
104.6***
111.9***
116.7***
126.8***
125.8***
124.3***
128.6***
128.9***
121.4***
130.0***
1224
The likelihood ratio tests are obtained from the maximum-likelihood random-effects estimator, which produce estimates that are very nearly the same as those produced by the full-information G2SLS estimator. See also Table 1 for notes. * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
127
Table 4. Random effects estimates of the effect of health and economic factors on women's retirement age. General health Poor health Individual income Wealth Compensation rate R2 (overall) BreuschPagan χ2 Likelihood ratio test Number of observations
-0.659*** (0.180) 0.424* (0.201) -0.066 (0.171) 0.692 (0.379) 17.4
Subjective health Mental Health health compared to others -0.961** -0.573 (0.356) (0.324) 0.457* 0.444* (0.202) (0.202) -0.060 -0.055 (0.172) (0.172) 0.667 0.735 (0.381) (0.381) 16.9 16.4
Working capacity -0.155*** (0.044) 0.414* (0.201) -0.043 (0.171) 0.701 (0.379) 17.5
Objective health Work Functional Diseases Diagnoses limita- limitations conditions tions -0.275 -0.419 -0.124 -0.214 (0.170) (0.328) (0.150) (0.346) 0.460* 0.437* 0.447* 0.451* (0.202) (0.203) (0.202) (0.202) -0.069 -0.061 -0.075 -0.068 (0.172) (0.172) (0.173) (0.173) 0.733 0.707 0.729 0.732 (0.381) (0.381) (0.381) (0.381) 16.4 16.3 16.1 16.1
66.0***
65.6***
67.6***
63.3***
66.4***
67.4***
68.5***
68.5***
71.8***
71.4***
73.7***
68.9***
72.5***
73.5***
74.7***
74.8***
982
See Table 3 for notes. * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
For women, purging the effects of unobserved heterogeneity, make wealth effects considerably larger and of the right sign (negative). Income effects are slightly smaller and the compensation rate is about the same magnitude as before and remains positive but insignificant. Health effects remain roughly the same as before, except that having health worse than others and work limitations are no longer significant at the 5% level, so that only subjective health measures impact retirement planning. The precise impacts are as follows: Being in bad health leads to a decrease in the planned retirement age by about 8 months. A one-point increase in the (poor) mental health index reduces planned retirement age by little short of a year while a one-point increase in working incapacity reduces the planned age by about 2 months. On the other hand, a one standard deviation increase in (log) income increases planned retirement age by between 0.17-0.19 years. As be-
128
fore, poor health affects men’s retirement planning more strongly than it does women’s.47 Table 5. Random effects estimates of the effect of health and economic factors on men’s retirement age. Working sample only. Subjective health Objective health Mental Health Working Work Functional Diseases Diagnoses health compared capacity limitations limitations conditions to others -1.261*** -1.211** -1.221** -0.052 -0.025 -0.033 -0.511** -0.476 (0.223) (0.450) (0.396) (0.059) (0.196) (0.492) (0.191) (0.295) 0.570*** 0.568*** 0.587*** 0.572*** 0.584*** 0.583*** 0.592*** 0.576*** (0.144) (0.146) (0.146) (0.147) (0.147) (0.147) (0.146) (0.147) -0.127 -0.122 -0.119 -0.124 -0.124 -0.125 -0.117 -0.117 (0.066) (0.068) (0.068) (0.068) (0.068) (0.068) (0.068) (0.068) -0.533 -0.623 -0.623 -0.598 -0.628 -0.632 -0.635 -0.641 (0.368) (0.372) (0.371) (0.375) (0.374) (0.373) (0.372) (0.372) 23.3 20.5 20.7 20.0 19.7 19.7 20.8 19.8
General health Poor health Individual income Wealth Compensation rate R2 (overall) BreuschPagan χ2 Likelihood ratio test Number of observations
48***
54***
54***
51***
54***
54***
51***
55***
61***
68***
68***
66***
68***
69***
64***
70***
884
See Table 3 for notes. * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
47
A fixed effects estimation is also tried as explained in footnote 39, and a Hausman specification test fails to reject the hypothesis that the individual effects are uncorrelated with the other regressors for women, but not for men. Based on the fixed effects IV results for women, we find that neither endogeneity nor measurement error is of obvious importance. Results derived from simple fixed effects estimation suggest that only poor general health has a significant effect on retirement plans for women but this result is only marginally significant (10% level). Poor general health is found to lower women’s planned retirement age by 8 months. In addition, diseases conditions such as osteoporosis and being hospitalised for heart conditions are both found to hasten retirement by more than 2 years. However, the effect of heart conditions is only marginally significant.
129
Table 6. Random effects estimates of the effect of health and economic factors on women's retirement age. Working sample only. General health Poor health Individual income Wealth Compensation rate R2 (overall) BreuschPagan χ2 Likelihood ratio test Number of observations
-0.573** (0.210) 0.480* (0.223) 0.146 (0.202) 0.081 (0.435) 23.1
Subjective health Objective health Mental Health Working Work Functional Diseases Diagnoses health compared capacity limitations limitations conditions to others -0.741 -0.543 -0.119* -0.280 -0.137 0.094 -0.647 (0.434) (0.370) (0.055) (0.194) (0.459) (0.176) (0.468) 0.492* 0.492* 0.464* 0.507* 0.500* 0.507* 0.504* (0.223) (0.224) (0.224) (0.223) (0.224) (0.224) (0.223) 0.151 0.164 0.183 0.154 0.152 0.159 0.153 (0.203) (0.203) (0.203) (0.203) (0.203) (0.204) (0.203) 0.074 0.108 0.049 0.097 0.096 0.099 0.114 (0.437) (0.437) (0.437) (0.437) (0.437) (0.437) (0.437) 22.6 22.4 22.9 22.6 22.2 22.2 22.3
27***
27***
27***
26***
26***
28***
27***
28***
30***
30***
30***
29***
28***
30***
30***
31***
652
See Table 3 for notes. * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
5.3 Workers only Although the sample from the start is restricted to those individuals who were in the labour force in the first survey year (1997), the preceding analysis also includes those individuals who retired between the two survey years. One way to minimize justification bias would be to restrict the analysis to individuals who remain in the workforce in both years, as retired individuals are more likely to cite failing health as a reason for quitting work. McGarry (2002) uses this strategy and argues that the estimates of the impact of subjective health measures on the probability of continued work cannot be attributed to justification bias when the focus is exclusively on workers. To test this hypothesis, in Tables 5 and 6 above we re-estimate the random effects model on workers only and compare it to the previous findings in Tables 3 and 4. Focusing only on the subjective health measures in these tables, as expected, coefficients are slightly smaller in magnitude
130
when non-workers are excluded, particularly for females, indicating that there may exist some potential justification bias in the self-reported health measures. However, not all economic variables are estimated to be larger here, only wealth and compensation rate for men, and income and wealth for women.
Table 7. Random effects IV estimates of the effect of health and economic factors on men’s retirement age. General health Poor health Individual income Wealth Compensation rate R2 (overall) Hausman test of exogeneity Number of observations
-1.892* (0.857) 0.590*** (0.148) -0.091 (0.068) -0.088 (0.368) 16.6 0.63
Subjective health Objective health Mental Health Working Work Functional Diseases Diagnoses health compared capacity limitations limitations conditions to others -5.186 -3.049* -0.754 -2.656 -3.896 -1.847* -1.344 (2.708) (1.553) (0.498) (1.660) (2.650) (0.735) (0.811) 0.572*** 0.647*** 0.454* 0.680*** 0.631*** 0.669*** 0.612*** (0.157) (0.150) (0.197) (0.165) (0.152) (0.152) (0.150) -0.080 -0.068 -0.075 -0.068 -0.093 -0.068 -0.077 (0.072) (0.071) (0.075) (0.077) (0.071) (0.071) (0.070) -0.181 -0.253 0.261 0.121 -0.225 -0.263 -0.269 (0.374) (0.364) (0.511) (0.459) (0.372) (0.369) (0.363) 10.2 13.0 8.7 5.3 9.9 12.5 12.6 2.51
1.57
1.94
2.94
1.79
3.32
1.32
1224
See Table 1 for notes. * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
131
Table 8. Random effects IV estimates of the effect of health and economic factors on women's retirement age. General health Poor health Individual income Wealth Compensation rate R2 (overall) Hausman test of exogeneity Number of observations
-1.866 (1.090) 0.362 (0.211) -0.074 (0.176) 0.643 (0.389) 15.3 1.48
Subjective health Objective health Mental Health Working Work Functional Diseases Diagnoses health compared capacity limitations limitations Conditions to others -2.760 -1.216 -0.249 -1.348 -3.937 -0.854 -0.193 (1.655) (1.538) (0.219) (1.253) (2.489) (0.585) (0.829) 0.468* 0.471* 0.383 0.501* 0.336 0.473* 0.451* (0.204) (0.201) (0.209) (0.210) (0.229) (0.203) (0.202) -0.047 -0.021 -0.035 -0.076 0.002 -0.096 -0.068 (0.174) (0.174) (0.175) (0.176) (0.185) (0.176) (0.173) 0.552 0.705 0.691 0.752 0.500 0.705 0.731 (0.399) (0.388) (0.381) (0.390) (0.429) (0.393) (0.381) 15.3 16.2 17.4 14.2 10.5 14.2 16.1 1.25
n.a.
0.59
0.79
2.28
n.a.
n.a.
982
See Table 1 for notes. * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
5.4 Endogeneity and measurement error A more fundamental statistical method of correction for justification bias is to account for the endogeneity of health in retirement. Tables 7 and 8 above report panel IV estimates of the effect of health and economic variables on retirement. The advantage of the IV approach is that problems of endogeneity (affecting the subjective measures) and measurement error in health (affecting mainly the objective measures) are rectified once health is instrumented for.48 In terms of appropriate instruments that could be correlated with health but uncorrelated with the disturbance term in the retirement equation, we use parental mortality status, number of discharges, number of doctor visits per year, assistance with home work, assistance with transportation and unique information on accidents as well as a vari-
48
This approach is used by Dwyer and Mitchell (1999) but in a static model.
132
able denoting the amount of exercise the individual gets.49 A detailed description of the instruments is provided in Appendix A1. First-stage results can be seen in Appendix A3, Tables A4 (men) and A5 (women). For each of the health measures considered, we perform chi-squared tests of the power of the instruments and in every case we are able to reject the null hypothesis of joint insignificance of the instruments at the 0.1-5% level (see Tables A4 and A5). Further, the instruments affect health in the expected directions in the majority of cases: having a parent alive is associated with better health, while the greater the number of discharges and doctors’ visits, the poorer is health. Individuals requiring assistance with homework or transportation in general have poorer health as do individuals who experience accidents. The result on the no exercise variable is consistent with poor health for men, although yields mixed results for women.
The results of the panel IV estimation are shown in Tables 7 and 8 and can be compared to the panel model without endogeneity in Tables 3 and 4. If justification bias is present, then IV estimates of the subjective health measures that are corrected for this type of endogeneity should be smaller than those from the model which treats health as exogenous. This is apparently not the case as in each gender group, the panel IV estimates on the subjective measures are larger in magnitude than those obtained from sim-
49
One might argue that the number of physician visits is not exogenous. First, frequent visits at a doctor might be one way of justifying early retirement. Second, if the process of applying for SDP involves one or more doctor visits, planned retirement and doctor visits might be correlated. Therefore, as a check of robustness of the results the number of doctor visits is eliminated as an instrument. However, the only change is that the chisquared test of the power of the instruments fails to reject the null hypothesis of joint insignificance of the instruments in 1 out of the 8 health outcome equations (the work limitation regression) for women.
133
ple random effects estimations. On the other hand, if measurement error is present, then the random effects IV estimates of the objective measures that are corrected for this type of error should be larger than those from the simple random effects. This is evidenced in Tables 7 and 8, where for almost all objective measures, for both men and women, random effects IV estimates are indeed larger. Fewer health measures are significant in the random effects IV compared to the simple random effects estimates. However, and more fundamentally, Hausman tests of exogeneity fail to reject the null hypothesis in the case of each health measure considered (see bottom rows in Tables 7 and 8), that both the simple random effects and the random effects IV estimators are consistent.
In order to check the robustness of the IV estimates, tests of overidentifying restrictions are conducted, see Table A6 for men and Table A7 for women in Appendix A3. For each of the health measures considered, we estimate specifications in which one instrument or group of instruments is eliminated in turn and compare these results to the estimates obtained when all instruments are included. In case of men as well as women, the parameter estimates are remarkably stable. The estimate for each of the health measures is found to be negative and similar in magnitudes in almost all specifications. Further, in all cases Hausman test of exogeneity fail to reject the null hypothesis that both the simple random effects and the random effects IV estimates are consistent.
The results indicate that issues of correlation between health and the disturbance term in the retirement equation or measurement error in health are not of obvious importance here and that therefore the ordinary random ef134
fects estimates are preferable on grounds of efficiency.50 This model (Tables 3 and 4) is retained as the preferred specification in the following analysis.
5.5 Cohort differences So far, we have pooled together two age cohorts in the analysis, primarily to obtain a larger sample and sufficient variation in program rules. However, health effects may become increasingly important as individuals age and the horizon approaches. To test this hypothesis, in Tables 9-12 below individuals born in 1940 and 1945 are analysed separately. Treating the sub-samples as independent, t-tests of equality of health coefficients across cohorts are performed. The results reveal that in no case are health coefficients different across cohorts, either for men or for women. However, the impact of economic variables could differ across cohorts but all these variables, annual income, wealth and the compensation rate, are estimated to be of the same magnitudes across cohorts for both men and women, except in the case of wealth for men. Still, as health effects are of equal importance to retirement planning to the two cohorts, we do not lose relevant information by pooling across cohorts for the purposes of this study.
50
The same result is obtained in a static context by Dwyer and Mitchell (1999) who are unable to reject the consistency of the least square estimator.
135
Table 9. Random effects estimates of the effect of health and economic factors on retirement age. Men born in 1940. General health Poor health Individual income Wealth Compensation rate R2 (overall) BreuschPagan χ2 Likelihood ratio test Number of observations
-0.883** (0.284) 0.874** (0.340) 0.224(*) (0.115) -0.899(*) (0.533) 14.9
Subjective health Mental Health health compared to others -1.160(*) -1.158(*) (0.594) (0.603) 0.866* 0.928** (0.344) (0.342) 0.218(*) 0.237* (0.116) (0.116) -0.961(*) -0.961(*) (0.534) (0.537) 13.2 14.4
Working capacity
Work limitations
-0.079 (0.075) 0.882* (0.345) 0.222(*) (0.116) -0.904(*) (0.540) 13.8
0.665** (0.260) 0.913** (0.341) 0.199(*) (0.115) -1.082* (0.537) 14.5
Objective health Functional Diseases limitations
Diagnoses
-0.473 -0.464(*) (0.751) (0.273) 0.928** 0.936** (0.344) (0.343) 0.221(*) 0.217(*) (0.116) (0.116) -0.937(*) -1.003(*) (0.538) (0.537) 13.3 13.7
-0.172 (0.332) 0.913** (0.344) 0.217(*) (0.116) -0.951(*) (0.537) 13.1
79***
81***
78***
76***
79***
79***
79***
79***
101***
104***
98***
97***
100***
101***
101***
102***
512
Note: Additional controls include education, cohabitation including age differences between partners, experience, occupation, sector, tenure, hours of work, physical demanding job, job satisfaction and a dummy variable for missing values for compensation rate. The likelihood ratio tests are obtained from the maximum-likelihood random-effects estimator, which produce estimates that are very nearly the same as those produced by the full-information G2SLS estimator. See also Table 1 for notes. (*) Significant at a 10% level, * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
Table 10. Random effects estimates of the effect of health and economic factors on retirement age. Men born in 1945. General health Poor health Individual income Wealth Compensation rate R2 (overall) BreuschPagan χ2 Likelihood ratio test Number of observations
-1.746*** (0.272) 0.604*** (0.168) -0.208* (0.082) 0.146 (0.478) 22.7
Subjective health Objective health Mental Health Working Work Functional Diseases Diagnoses health compared capacity limitations limitations to others -1.503** -1.439*** -0.068 -0.299 -0.498 -0.746** -0.710(*) (0.556) (0.446) (0.077) (0.262) (0.552) (0.237) (0.367) 0.618*** 0.643*** 0.622*** 0.641*** 0.630*** 0.641*** 0.629*** (0.173) (0.173) (0.174) (0.174) (0.174) (0.173) (0.174) -0.200* -0.190* -0.201* -0.202* -0.205* -0.184* -0.190* (0.085) (0.085) (0.086) (0.086) (0.086) (0.085) (0.086) 0.013 0.002 0.037 0.021 -0.015 0.029 -0.020 (0.485) (0.483) (0.490) (0.487) (0.486) (0.485) (0.485) 17.8 17.9 16.7 16.7 16.5 18.4 16.9
27***
36***
38***
38***
39***
39***
33***
39***
30***
39***
42***
42***
43***
44***
37***
43***
712
See Table 9 for notes. Significant at a 10% level, * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
136
Table 11. Random effects estimates of the effect of health and economic factors on retirement age. Women born in 1940. General health Poor health Individual income Wealth Compensation rate R2 (overall) BreuschPagan χ2 Likelihood ratio test Number of observations
-0.602* (0.283) 0.273 (0.225) 0.047 (0.220) -0.641 (0.587) 22.7
Subjective health Objective health Mental Health Working Work Functional Diseases Diagnoses health compared capacity limitations limitations to others -1.692*** -0.335 -0.116(*) -0.323 -0.435 -0.208 0.016 (0.506) (0.527) (0.061) (0.266) (0.392) (0.218) (0.407) 0.306 0.275 0.265 0.296 0.271 0.272 0.279 (0.224) (0.227) (0.226) (0.227) (0.226) (0.226) (0.227) 0.099 0.077 0.071 0.053 0.079 0.063 0.076 (0.218) (0.221) (0.220) (0.221) (0.220) (0.220) (0.221) -0.760 -0.583 -0.574 -0.568 -0.626 -0.566 -0.580 (0.584) (0.590) (0.588) (0.589) (0.591) (0.589) (0.590) 25.0 22.0 23.4 22.2 22.2 21.8 21.9
25***
21***
25***
21***
25***
25***
25***
24***
28***
24***
28***
24***
28***
28***
29***
28***
406
See Table 9 for notes. (*) Significant at a 10% level, * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
Table 12. Random effects estimates of the effect of health and economic factors on retirement age. Women born in 1945. General health Poor health Individual income Wealth Compensation rate R2 (overall) BreuschPagan χ2 Likelihood ratio test Number of observations
-0.610** (0.237) 0.773* (0.332) -0.261 (0.252) 1.165* (0.503) 17.1
Subjective health Mental Health health compared to others -0.455 -0.569 (0.491) (0.418) 0.811* 0.815* (0.333) (0.333) -0.271 -0.257 (0.254) (0.254) 1.168* 1.201* (0.506) (0.505) 16.1 16.2
Objective health Work Functional Diseases limitations limitations
Working capacity
Diagnoses
-0.159** (0.062) 0.745* (0.333) -0.211 (0.254) 1.123* (0.504) 17.0
-0.234 (0.223) 0.829* (0.333) -0.267 (0.254) 1.189* (0.506) 16.2
-0.361 (0.505) 0.808* (0.334) -0.271 (0.254) 1.181* (0.506) 16.1
0.013 (0.204) 0.825* (0.333) -0.273 (0.255) 1.188* (0.506) 15.9
-0.356 (0.535) 0.826* (0.333) -0.269 (0.254) 1.192* (0.506) 16.0
33***
34***
34***
34***
33***
34***
34***
34***
36***
37***
37***
36***
36***
37***
37***
37***
576
See Table 9 for notes. (*) Significant at a 10% level, * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
137
5.6 Discontinuities in planned retirement age In previous studies, planned retirement age has been treated as a continuously distributed variable in econometric analyses, where researchers typically employ ordinary least squares methods in the analysis of its determinants. However, the descriptive evidence in most countries indicates that the distribution of planned retirement age peaks around the early and normal ages of retirement, or in the European context, is bunched around the ages of first eligibility of various labour market related exit programs. In Denmark for example, several different regimes could be identified, consistent with the clustering at ages 60, 62 and 65, cf. Figure 2 above, which in turn reflect pension policy rules and eligibility criteria. Ages 60 and 67 are the early and the normal retirement ages in 1997, respectively, while in 2002 the normal retirement age is 65, following the 1999 reform. Several waiting incentives in the PEW program also make retirement at 60-61, 62 and 63-64 typical options. We distinguish therefore between six categories of planned retirement age: below 60, 60-61, 62, 63-64, 65-66 and 67 and above. This allows us to adopt a random effects ordered probit model specification of the dependent variable, planned retirement age.
Results of this estimation are shown in Tables 13 and 14 below. Although we cannot formally test this model versus the ordinary random effects model, all estimated cut parameters are significant in these tables confirming the usefulness of this approach in the context of modelling retirement planning. A further advantage, other than exploiting the clustering in the observed distribution of planned retirement age, is that adopting such an approach allows for impacts of health to vary by regime, reflecting pension policy reform that affects program age of eligibility or age-based incentives 138
cf. Section 4. Therefore, we report marginal effects of health coefficients on retirement in these tables. For purposes of compactness, we report aggregated marginal effects of the category 62 and above for each significant health coefficient.51 As is evident, results from the random effects ordered probit specification on both the health and the financial variables are qualitatively similar to those derived from the random effects specification (Tables 3 and 4). However, worse relative health for women and being hospitalised for a serious condition for men are only significant in the random effects ordered probit specification.
Focusing on the significant health variables for men, quite large marginal effects are found for the subjective health measures. In fact, the probability of planning to retire at age 62 or later is reduced by 26-28 percentage points for poor general health, worse mental health and worse relative health, while having a disease condition or being hospitalised for a serious condition only reduces this probability by 13 percentage points. For comparison, a one-point increase in (log) individual income increases the probability by 20-22 percentage points. Since the impact of the broad subjective health measures again is found to be larger than the impact of income, the assessment of the relative importance of health is unchanged compared to what was derived from the random effects specification.
51
This is possible because marginal effects sum to zero as the probabilities sum to one. The dividing line is set between the categories 60-61 and 62 because for all variables in the tables the marginal effects change signs between these two categories. Separate marginal effects for each age category are available on request.
139
Table 13. Random effects ordered probit estimates of the effect of health and economic factors on men’s retirement age. General health
Poor health Individual income Wealth Compensation rate Cut 1 Cut 2 Cut 3 Cut 4 Cut 5 Rho Log likelihood Number of obs.
β -0.679*** (0.123) 0.502*** (0.134) -0.047 (0.043) -0.116 (0.218) 5.462*** (1.679) 7.755*** (1.693) 8.586*** (1.700) 9.151*** (1.704) 10.341*** (1.711) 0.500*** (0.036) -1724
ME -0.263 0.200 -0.019 -0.046
Subjective health Mental health Health compared to others β ME β ME -0.803*** -0.256 -0.748*** -0.282 (0.250) (0.213) 0.509*** 0.202 0.544*** 0.217 (0.135) (0.135) -0.046 -0.018 -0.041 -0.016 (0.043) (0.043) -0.160 -0.064 -0.170 -0.068 (0.218) (0.218) 4.680** 6.065*** (1.745) (1.693) 6.966*** 8.352*** (1.757) (1.708) 7.801*** 9.184*** (1.763) (1.715) 8.365*** 9.749*** (1.766) (1.719) 9.549*** 10.936*** (1.773) (1.727) 0.512*** 0.510*** (0.035) (0.036) -1734 -1733
Objective health Working Work Functional Diseases capacity limitations limitations conditions β β β β ME -0.057 0.120 -0.263 -0.334** -0.133 (0.032) (0.112) (0.261) (0.107) 0.517*** 0.536*** 0.540*** 0.532*** 0.212 (0.136) (0.136) (0.136) (0.135) -0.046 -0.050 -0.048 -0.044 -0.017 (0.043) (0.043) (0.043) (0.043) -0.127 -0.179 -0.161 -0.168 -0.067 (0.219) (0.219) (0.219) (0.218) 5.727*** 6.123*** 6.130*** 6.007*** (1.706) (1.700) (1.702) (1.689) 7.999*** 8.398*** 8.409*** 8.285*** (1.721) (1.715) (1.717) (1.704) 8.827*** 9.229*** 9.240*** 9.113*** (1.727) (1.722) (1.724) (1.711) 9.388*** 9.792*** 9.804*** 9.674*** (1.731) (1.726) (1.728) (1.715) 10.569*** 10.980*** 10.992*** 10.859*** (1.739) (1.733) (1.735) (1.722) 0.509*** 0.515*** 0.516*** 0.507*** (0.036) (0.035) (0.035) (0.036) -1738 -1739 -1739 -1734 1224 the β estimate of the health measure is significant.
Diagnoses β -0.321* (0.158) 0.532*** (0.136) -0.045 (0.043) -0.169 (0.219) 5.996*** (1.703) 8.278*** (1.718) 9.111*** (1.725) 9.676*** (1.728) 10.867*** (1.736) 0.517*** (0.035) -1737
ME -0.127 0.212 -0.018 -0.067
Note: Marginal effects (ME) are calculated only for models in which In this case, ME refers to the change in the probability of having reported a retirement age at or above 62. ME are based on the linear prediction from the estimated coefficients and are calculated at the mean values of the explanatory variables, while the two components of the error term εit (υit and ui) are set equal to zero. See also Table 1 for notes. * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
Table 14. Random effects ordered probit estimates of the effect of health and economic factors on women's retirement age. General health
Poor health Individual income Wealth Compensation rate Cut 1 Cut 2 Cut 3 Cut 4 Cut 5 Rho Log likelihood Number of obs.
β -0.416*** (0.131) 0.313* (0.139) -0.003 (0.118) 0.289 (0.269) 3.340* (1.644) 5.941*** (1.654) 6.913*** (1.659) 7.501*** (1.662) 8.566*** (1.669) 0.405*** (0.047) -1141
ME -0.138 0.112 -0.001 0.103
Subjective health Objective health Mental health Health compared to Working capacity Work Functional Disease Diagnoses others limitations limitations conditions β ME β ME β ME β β β β -0.649** -0.254 -0.495* -0.155 -0.118*** -0.046 -0.203 -0.448 -0.047 -0.122 (0.248) (0.234) (0.032) (0.121) (0.233) (0.105) (0.248) 0.333* 0.120 0.323* 0.116 0.301* 0.108 0.335* 0.315* 0.327* 0.329* (0.138) (0.139) (0.137) (0.139) (0.139) (0.139) (0.139) 0.001 0.000 0.007 0.003 0.015 0.005 -0.004 0.003 -0.005 -0.003 (0.117) (0.118) (0.116) (0.117) (0.118) (0.118) (0.118) 0.264 0.095 0.316 0.113 0.281 0.101 0.305 0.282 0.307 0.309 (0.269) (0.270) (0.267) (0.269) (0.269) (0.270) (0.270) 2.761 3.496* 2.864 3.630* 3.427* 3.576* 3.615* (1.664) (1.643) (1.634) (1.637) (1.642) (1.647) (1.646) 5.340*** 6.087*** 5.442*** 6.208*** 6.012*** 6.162*** 6.201*** (1.673) (1.653) (1.645) (1.647) (1.652) (1.657) (1.656) 6.306*** 7.058*** 6.407*** 7.176*** 6.982*** 7.135*** 7.173*** (1.678) (1.658) (1.650) (1.652) (1.657) (1.662) (1.661) 6.893*** 7.644*** 6.990*** 7.762*** 7.568*** 7.722*** 7.761*** (1.681) (1.661) (1.653) (1.655) (1.660) (1.665) (1.664) 7.956*** 8.705*** 8.042*** 8.821*** 8.625*** 8.786*** 8.825*** (1.687) (1.668) (1.660) (1.662) (1.667) (1.672) (1.671) 0.399*** 0.407*** 0.388*** 0.402*** 0.404*** 0.411*** 0.410*** (0.048) (0.047) (0.048) (0.048) (0.047) (0.047) (0.047) -1142 -1144 -1139 -1144 -1144 -1146 -1146 982
See Table 13 for notes. * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
141
For women, the income effect is only half the size of the effect for men. In fact, a one-point increase in (log) individual income only increases the probability to plan to retire at age 62 or later by 11-12 percentage points. The effect of the significant health measures differs considerably ranging from a 5 percentage points decrease if working incapacity is increased one point to a 25 percentage points reduction for worse mental health. The figures for poor general health and worse relative health are 14 and 16 percentage points respectively. Here again, the relative importance of health to income is unchanged. Further, this analysis also suggests that health effects are not as strong as for men. The only exception is the effect of mental health.
5.7 Disaggregated disease conditions and diagnoses The random effects ordered probit specification is useful in capturing discontinuities in planned retirement age, particularly in an environment of changing pension program eligibility criteria. However, in order to compare our findings to previous studies, we return to the random effects models in Tables 3 and 4 and disaggregate to the level of specific conditions and diagnosed diseases in order to obtain information about which particular illnesses can affect retirement plans. The results of these analyses are given in Tables 15-18 below. For men, myalgia and back problems significantly lowers planned retirement age by respectively 7 months and little more than a year, while for women, diseases conditions such as back problems, osteoporosis, and depression significantly hasten retirement, particularly the two latter conditions, by 6 months, 2 years and nearly a year respectively. Being hospitalised for a serious condition does not appear to
exert any appreciable effect, except in the case of heart diseases, which is marginally significant (10% level) for both men and women and lowers planned retirement age by 8 months and a year and 8 months respectively.52 The limited effect of diagnoses might be due to the fact that individuals who have been hospitalised receive neutralizing treatment and therefore are able to continue to work. Another potential explanation is that people who suffer from a serious condition may in fact be advised by their doctors (or choose themselves) to continue working because of the potential therapeutic effects of work on health through engagement in a challenging and rewarding activity, the social contact with colleagues etc. Finally, a third explanation might be that these individuals’ retirement plans are affected but we do not capture this effect because of small sample size as the incidence of diagnoses is in general quite low.
52
In comparison, McGarry (2002) does not find that specific conditions are significantly related to the probability of working beyond the age of 62, while Dywer and Mitchell (1999) find that limitations of daily living, back problems, head injuries and circulatory problems lower planned retirement age. Mental health (which is broader than depression) does not have a significant effect in their study and conditions such as osteoporosis and myalgia are not defined in their data. Thus, in part differences in findings are due to differences in data definitions and availability but back problems appear important in both their study and ours.
143
Table 15. Random effects estimates of the effect of disaggregated diseases conditions and economic factors on men's retirement age. High Diabetes Bronchiblood tis/asthma pressure Poor health -0.828 1.287 -0.268 (0.545) (0.917) (0.515) Individual 0.646*** 0.635*** 0.632*** income (0.149) (0.148) (0.149) Wealth -0.091 -0.094 -0.094 (0.069) (0.069) (0.069) Compensa-0.263 -0.236 -0.250 tion rate (0.362) (0.362) (0.362) R2 13.4 13.3 13.1 (overall) Breusch109*** 110*** 111*** Pagan χ2 Likelihood ratio test 126*** 128*** 129*** Number of obs.
Osteoarthritis
Diseases Myalgia
Osteoporosis/ Back Depression decalcification problems of bones -2.058 -1.056*** -0.285 (2.175) (0.255) (0.602) 0.631*** 0.638*** 0.633*** (0.149) (0.147) (0.149) -0.093 -0.072 -0.096 (0.069) (0.069) (0.069) -0.262 -0.248 -0.248 (0.362) (0.360) (0.362)
-0.328 (0.248) 0.631*** (0.148) -0.091 (0.069) -0.253 (0.362)
-0.592* (0.267) 0.628*** (0.148) -0.097 (0.069) -0.243 (0.362)
13.4
13.9
13.2
14.7
13.1
108***
105***
110***
107***
110***
126***
122***
128***
124***
128***
1224
See Table 3 for notes. * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
Table 16. Random effects estimates of the effect of disaggregated diseases conditions and economic factors on women's retirement age.
Poor health Individual income Wealth Compensation rate R2 (overall) BreuschPagan χ2 Likelihood ratio test Number of obs.
High Diabetes Bronchiblood tis/asthma pressure -0.199 0.488 -0.194 (0.442) (1.977) (0.433) 0.448* 0.454* 0.451* (0.202) (0.203) (0.202) -0.066 -0.067 -0.068 (0.173) (0.173) (0.173) 0.730 0.727 0.729 (0.381) (0.381) (0.381)
Osteoarthritis
Diseases Myalgia
-0.040 (0.197) 0.449* (0.202) -0.067 (0.173) 0.727 (0.381)
-0.021 (0.186) 0.450* (0.202) -0.068 (0.173) 0.729 (0.381)
Osteoporosis/ Back Depression decalcification problems of bones -2.211* -0.506* -0.973* (0.552) (0.223) (0.414) 0.448* 0.470* 0.436* (0.201) (0.202) (0.202) -0.091 -0.074 -0.061 (0.171) (0.172) (0.172) 0.685 0.671 0.747* (0.378) (0.381) (0.380)
16.1
16.0
16.0
16.0
16.0
17.4
16.7
16.7
67***
68***
68***
68*
68***
68***
66***
66***
74***
75***
75***
75***
75***
74***
72***
72***
982
See Table 3 for notes. * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
144
Table 17. Random effects estimates of the effect of disaggregated diagnoses and economic factors on men's retirement age.
Poor health Individual income Wealth Compensation rate R2 (overall) BreuschPagan χ2 Likelihood ratio test Number of observations
0.188 (1.124) 0.631*** (0.149) -0.094 (0.069) -0.248 (0.362) 13.1
-0.163 (0.679) 0.630*** (0.149) -0.094 (0.069) -0.248 (0.362) 13.1
Diagnoses Lung diseases 0.477 (0.795) 0.631*** (0.149) -0.094 (0.069) -0.246 (0.362) 13.1
111***
111***
111***
129***
129***
129***
Heart conditions -0.700 (0.377) 0.632*** (0.148) -0.086 (0.069) -0.246 (0.362) 13.3
Strokes
Cancers
Diabetes -0.862 (1.008) 0.628*** (0.149) -0.095 (0.069) -0.253 (0.362) 13.1
High blood pressure -0.855 (0.749) 0.628*** (0.149) -0.094 (0.069) -0.256 (0.362) 13.1
Arthritis
111***
111***
111***
111***
129***
130***
129***
129***
-0.513 (0.625) 0.627*** (0.149) -0.093 (0.069) -0.249 (0.362) 13.1
1224
See Table 3 for notes. * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
Table 18. Random effects estimates of the effect of disaggregated diagnoses and economic factors on women's retirement age.
Poor health Individual income Wealth Compensation rate R2 (overall) BreuschPagan χ2 Likelihood ratio test Number of observations
Heart conditions -1.723 (0.954) 0.442* (0.202) -0.078 (0.172) 0.750* (0.380) 16.1
Strokes
Diagnoses Cancers
-0.768 (1.354) 0.453* (0.202) -0.069 (0.173) 0.728 (0.381) 16.0
-0.610 (0.553) 0.458* (0.202) -0.065 (0.173) 0.731 (0.381) 16.1
Lung diseases -0.323 (0.722) 0.449* (0.202) -0.070 (0.173) 0.732 (0.381) 16.1
Arthritis
70***
69***
69***
67***
68***
76***
75***
75***
74***
74***
0.944 (0.667) 0.453* (0.202) -0.070 (0.172) 0.737 (0.381) 16.2
892
See Table 3 for notes. * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
5.8 Health changes So far, we have focused on the effect of the level of health on retirement planning. However, it might be that the effect of sudden or gradual changes 145
in health is even larger than the effect of health status. Means for poor health and worsened health by changes in retirement plans is reported for men and women respectively in Table A8 and A9 in Appendix A3. For men as well as women, the share of individuals that plan to postpone retirement between the two waves is larger than the share that plan to retire earlier than originally planned and in general, the expected correlation between worsened health and hastened retirement is not found. Only the share of individuals that has been hospitalised for a serious condition is larger among individuals that bring retirement forward. This result indicates that diagnoses to a greater extent than the other health measures reflect health shocks.
To examine these effects more closely, health changes occurring between the 1st and 2nd waves are included in the random effects estimations while still controlling for health status in wave 1, cf. Table 19 and 20 below. The results obtained from these analyses are not directly comparable to those derived from the analyses of the effect of health status (Tables 3 and 4) since these models are not nested. However, health changes seem to be important in particular in the case of men. Both worsened general and mental health hasten men’s planned retirement, while development of functional limitations lower women’s planned retirement age. Looking directly at the estimated coefficients for health changes, worsened general and mental health reduces men’s planned retirement age by 9 months and more than 2 years respectively, while development of functional limitations reduce this age by 1 year and 3 months for women. These results suggest that health shocks increase the propensity to hasten retirement.
146
Table 19. Random effects estimates of the effect of health status, health changes and economic factors on men’s retirement age. Subjective health General health
Mental health
Poor health -1.610*** -1.509** wave 1 (0.256) (0.500) Health -0.760** -2.132*** change (0.287) (0.660) Individual 0.599*** 0.614*** income (0.146) (0.148) Wealth -0.090 -0.084 (0.068) (0.069) Compensa-0.130 -0.244 tion rate (0.359) (0.361) R2 (overall) 16.9 14.4 Breusch101.1*** 107.8*** Pagan χ2 Likelihood ratio test 115.6*** 124.5*** Number of obs.
Health compared to others -1.567*** (0.460) -0.435 (0.489) 0.639*** (0.148) -0.077 (0.069) -0.216 (0.361)
Objective health Working capacity
Work Functional Disease Diagnoses limitations limitations conditions
-0.218*** (0.067) -0.114 (0.076) 0.587*** (0.149) -0.090 (0.069) -0.168 (0.362)
0.139 (0.271) 0.182 (0.250) 0.631*** (0.149) -0.098 (0.069) -0.264 (0.363)
-0.694 (0.602) -0.917 (0.676) 0.629*** (0.149) -0.091 (0.069) -0.237 (0.362)
-1.003*** (0.239) -0.166 (0.241) 0.635*** (0.148) -0.082 (0.068) -0.247 (0.360)
-0.377 (0.463) -0.348 (0.345) 0.630*** (0.149) -0.087 (0.069) -0.255 (0.362)
14.3
14.1
13.1
13.2
15.0
13.2
108.2***
109.4***
110.5***
111.1***
106.1***
110.9***
124.9***
126.2***
128.3***
128.8***
122.2***
128.7***
1224
Note: Health changes are defined as the differences between the values for health status in wave 1 and 2. See also Table 3 for notes. * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
Table 20. Random effects estimates of the effect of health status, health changes and economic factors on women's retirement age. General health Poor health -0.838*** wave 1 (0.249) Health -0.119 change (0.229) Individual 0.429* income (0.201) Wealth -0.055 (0.171) Compensa0.685 tion rate (0.380) R2 17.6 (overall) Breusch64.1*** Pagan χ2 Likelihood ratio test 69.6*** Number of obs.
Subjective health Mental Health health compared to others -1.153** -0.816 (0.424) (0.446) -0.682 -0.117 (0.527) (0.399) 0.459* 0.436* (0.202) (0.202) -0.062 -0.060 (0.172) (0.172) 0.697 0.710 (0.380) (0.381)
Working capacity
Objective health Work Functional Diseases Diagnoses limitations limitations conditions
-0.231*** (0.057) -0.043 (0.059) 0.376 (0.202) -0.055 (0.171) 0.736 (0.379)
-0.496* (0.232) -0.290 (0.221) 0.465* (0.202) -0.077 (0.172) 0.725 (0.381)
-0.786 (0.402) -1.214* (0.535) 0.414* (0.202) -0.050 (0.172) 0.692 (0.381)
-0.110 (0.192) -0.241 (0.207) 0.450* (0.202) -0.082 (0.173) 0.751* (0.381)
-0.193 (0.631) 0.170 (0.411) 0.443* (0.203) -0.063 0.173 0.730 (0.381)
17.0
16.5
18.0
16.6
16.7
16.2
16.1
65.6***
67.3***
63.6***
66.9***
66.7***
68.1***
68.2***
71.4***
73.4***
68.8***
72.9***
72.6***
74.3***
74.4***
982
See Table 19 for notes. * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
147
As is evident, results on both health status and financial variables are qualitatively similar to those derived from the analyses of the effect of health status. However, working capacity for men and work limitations for women are only significant when health changes are included. Similarly, inclusion of health changes in general increases the effect of health status while the effect of income remains unchanged. That is, including health changes does not change the assessment of the relative importance of health compared to income. Further, this analysis also suggests that health effects are stronger for men than for women. In general, these results strengthen the conclusion that health is important when retirement age is planned.
6. Conclusions Using a wide array of alternative health measures including both selfreported and diagnostic measures extracted from LPR records, we compare the role of subjectively versus objectively measured health as a determinant of retirement planning, after controlling for income, labour market, job and background characteristics. The sample consists of older workers and retirees drawn from a Danish panel survey from 1997-2002 that is merged to longitudinal register data. Extending the existing literature, we estimate a panel model of retirement planning that controls both for unobserved heterogeneity as well as accounts for endogeneity and measurement error of health in retirement, and estimate separate models for men as well as women. We find that self-rated health is both an important predictor of retirement as well as a valid measure of health. Our analysis therefore brings fresh evidence to the debate initiated by earlier U.S. studies based on
148
the older Retirement History Longitudinal Survey (RHS) data. These studies concluded that the use of self-reported health in retirement models was likely to yield an unreliable impact of health on retirement due to “justification bias”. However, our study, like more recent U.S. studies based on the newer HRS data find little support for this hypothesis and neither endogeneity nor measurement error turn out to be important sources of concern in the Danish data.
Unobserved heterogeneity however, turns out to be important and estimates from random effects models show that self-rated physical and mental health are important predictors of retirement planning, in fact at least as important as economic factors, both among men as well as among women. However, health seems to be relatively more influential in men’s retirement planning. Being in poor general health or poor mental health significantly reduces planned retirement age for both men and women. Other health measures, in particular having health worse than others or having a disease for men and a reduction in working capacity for women, also lower planned retirement age significantly. At a disaggregated level, back problems and myalgia significantly hasten male retirement, while back problems and particularly osteoporosis and depression are significant factors triggering retirement among women. Retirement planning is in general unaffected by being hospitalised for a serious condition, except in the case of hospitalisation for heart diseases, which reduces planned retirement age for both men and women, but only marginally so. Looking at health changes strengthens the conclusion that health is an important factor in retirement planning. In fact, health shocks seem to increase the propensity to retire earlier. However, as expected, health seems to be less important for retire-
149
ment planning in Denmark compared to the US due to the subsidized and fully-covered health care system and the easier access to health-related exit.
As stated above, results suggest that health has the strongest effect on men’s retirement planning. According to Danø et al. (1999) one possible explanation of this gender difference is that men to a greater extent than women are employed in jobs that are inconsistent with poor health.53 Another explanation might be that the gender difference is due to sample selection. That is, women outside the labour force in wave 1 suffered from poor health to a greater extent than similar men. However, this hypothesis is not confirmed when looking at the means for the health measures for these two groups, cf. Table A1 in Appendix A3. In fact, women outside the labour seem to be at least as healthy as similar men. However, this might be due to the fact that a relatively large share of women retired from the labour market through TBP in the mid-nineties, cf. Section 2. Since access to this program did not depend on health criteria, retirement through this program probably implied that the average level of health among women outside the labour force was higher than it would otherwise have been. In other words, the sample selection problem might be hidden due to the high take-up rate of a particular early retirement program in the mid-nineties among relatively healthy women.
53
Tentatively, we have tested this hypothesis by interacting each of the eight health measures with dummy variables for private sector and physical demanding job respectively. However, no systematic difference between men and women appeared.
150
If the identified effect of health on retirement planning in Denmark reflected an exploitation of the Danish welfare system by individuals in the labour force using failing health as an excuse for early retirement, our conclusions would lead to the policy recommendation of a less generous welfare system as a way of postponing retirement among older workers in the labour force. However, the identified effect of health seems to be real since justification bias does not appear to be a problem. Further, the fact that only selected, and not all health measures are found to have a significant effects confirms this interpretation. Therefore, based on our findings, the policy recommendation would be to expand preventive and neutralizing health care services. A particular effort should be directed toward preventing diseases such as osteoporosis, back problems, myalgia and depression.
All in all, our results confirm that health is an important determinant of preferences for retirement and that poor health causes workers to retire earlier. Finally, by estimating separate models for women as well as men, we add to the existing knowledge about the impact of health on retirement behaviour of older workers in the population.
7. References Anderson, K.H. and Burkhauser, R.V. (1985): The Retirement-Health Nexus: A New Measure of on Old Puzzle, The Journal of Human Resources, 20 (3), pp. 315330. Balestra, P. and Varadharajan-Krishnakumar, J. (1987): Full Information Estimations for a System of Simultaneous Equations with Error Component Structure. Econometric Theory, 3, pp. 223-246. Bazzoli, G.J. (1985): The Early Retirement Decision: New Empirical Evidence on the Influence of Health, The Journal of Human Resources, 20 (2), pp. 214-234.
151
Benítez-Silva, H., Buchinsky, M., Chan, H.M., Cheidvasser, S. and Rust, J. (2000): How Large is the Bias in Self-Reported Disability?, NBER Working Paper 7526. Bound, J. (1991): Self-Reported versus Objective Measures of Health in Retirement Models. The Journal of Human Resources, 26, 1, 106-138. Christensen, B.J. and Datta Gupta, N. (2000): The Effect of a Pension Reform on the Retirement of Danish Married Couples (in Danish). Nationaløkonomisk Tidsskrift, 138, pp. 222-242. Danish Insurance Information Service (2001, 2002) Social Benefits – who, what & when? (in Danish), Copenhagen. Danø, A.M., Ejrnæs, M. and Husted, L. (2000): How is the Retirement Age Affected by the Reform of the Post Employment Wage Programme? Nationaløkonomisk Tidsskrift, 138, pp. 205-221. Danø, A.M., Ejrnæs, M. and Husted, L. (1998): Gender Differences in Retirement Behaviour, Institute of Local Government Studies - Denmark, Copenhagen. Dwyer, D.S. (2001): Planning for Retirement: The Accuracy of Expected Retirement Dates and the Role of Health Shocks. CRR WP 2001-08. Dwyer, D.S. and Hu, J. (1998): Retirement Expectations and Realizations: The Role of The Health Shocks and Economic Factors. PRC WP 98-18. Dwyer, D.S. and Mitchell, O.S. (1999): Health problems as determinants of retirement: Are self-rated measures endogenous? Journal of Health Economics 18 (1999), p. 173-193. Greene, W.H. (2003): Econometric Analysis, Prentice Hall, Upper Saddle River, New Jersey, pp. 293-303. Grossman, M. (1972): On the Concept of Health Capital and the Demand for Health, Journal Political Economy, 80 (2), pp. 223-255. Høgelund, J. (2003): In Search of Effective Disability Policy. Comparing the Developments and Outcomes of Dutch and Danish Disability Policies, Amsterdam University Press. Kreider, B. and Pepper, J. (2001): Inferring Disability Status from Corrupt Data, mimeo, Iowa State University. Kreider, B. and Pepper, J. (2002): Disability and Employment: Reevaluating the Evidence in Light of Reporting Errors, mimeo, Iowa State University. Larsen, M. (2004) Retaining Older Workers in the Danish Labour Market: The Effect of Subjective Job Characteristics on Early Retirement. Unpublished. Lausten, M. (2001) Transfer Incomes and Income Distribution (in Danish), Copenhagen. Mitchell, O.S. and Fields, G.S. (1984): The Economics of Retirement Behavior. Journal of Labour Economics, 1984, vol. 2, nr. 1, p. 84-105. McGarry, K. (2002): Health and Retirement: Do Changes in Health Affect Retirement Expectations? NBER Working Paper 9317.
152
National Social Appeals Board (2003): Press statement, 27 June 2003, digital document: http://www.dsa.dk/. Pedersen, P.J. and Smith, N. (1996): A Duration Analysis of the Decision to Retire Early, Chapter 5 in Wadensjö, E. (ed.). The Nordic Labour Markets in the 1990's, Amsterdam. Sammartino, F.J. (1987): The Effects of Health on Retirement, Social Security Bulletin, 50, pp. 31-47. Schaumann, A. (2001) The Aging Society. Demography - Expenditure Pressure - What can be done? The Danish Board of Technology, digital document: http://www.tekno.dk/.
Appendix A1. Health measures Both survey and register based health measures are included in our data. Subjective health measures are obtained from the survey while more objective health measures are obtained from both surveys and registers.
The subjective health measures from the survey are self-rated. These are: • General health: “All in all, how would you assess you current health? This variable is equal to one, if the answer is “very poor”, “poor” or “somewhat poor” and zero otherwise (“good” or “very good”). • Mental health: An index for health is created by counting the number of mental problems/conditions present such as memory loss, fear, anxiety, depression and loneliness. • Health compared to others: “How do you think your health is relative to others at your age?” This variable is equal to one, if the answer is “worse than others” and zero otherwise (“better than others” or “like others”).
153
• Working capacity: “If we say that your working capacity was given 10 points when it was at its highest level, how many points would you give your working capacity today?” This scale has been inverted. Consequently, one point is the maximum, while ten points is the minimum.
The objective health measures obtained from survey and registers are split into four categories: work limitations, functional limitations, doctor diagnosed diseases and hospitalisation due to specific diseases: • Work limitations: “Do you find it difficult to do your job due to impairment of memory or concentration, reduced sight or hearing, tiredness, low spirits or lack of sleep?” A dummy variable for work limitations is set equal to one, if the individual suffers from at least one of these problems. • Functional limitations: A dummy variable is set equal to one if the individual normally has difficulties cutting toe nails, climbing stairs, walking around outdoors or inside the home, getting washed or putting on clothes or shoes. • Diseases: “Has a doctor told you that you have – or within the last year have had – high blood pressure, diabetes, bronchitis/asthma, osteoarthritis, myalgia, osteoporosis/ decalcification of bones, back problems or depression?” If yes: “Does this disease bother you in normal everyday life?” A dummy for diseases is set equal to one if at least one of these diseases bothers the individual in normal everyday life.
154
• Diagnoses: Information about diagnoses is obtained from registers. A dummy variable for diagnoses is set equal to one if the individual has been hospitalised for either heart conditions, stroke, cancer, lung disease, diabetes, high blood pressure or arthritis.54
The instruments for health are: • Discharges: Number of discharges. • Physician visits: Number of physician visits. • Accident etc.: Hospitalisation due to accident, attack or self-inflicted injury. • Parent alive: At least one of the parents alive • Assistance with work at home: Have received assistance from children, family or friends (not cohabitants) with cleaning, washing, shopping, cooking or keeping up the house or the garden. • Assistance with transportation etc.: Have received assistance from children, family or friends (not cohabitants) with money affairs, contact to authorities, getting to examinations, treatment etc. or going outside, going on a visit or getting to leisure activities. • No exercise: Exercise includes walking or cycling at least half an hour, doing gymnastics, doing any sport or taking dancing lessons. On a scale from 1 to 5 (1 = daily, 5 = never).
54
For the first survey year, which is 1997, diagnoses are recorded on the basis of information for the period 1993 to 1996, while in the case of the second survey year, which is 2002, diagnoses are recorded for the years 1998-2001. If retirement takes place between to two survey years, diagnoses are recorded based on information for the years between the first survey year and the retirement year.
155
The instruments discharges, physician visits, accidents etc. are based on information from registers, while parent alive, assistance with work at home, assistance with transportation etc. and exercise are survey-based. 55
Appendix A2. Compensation rate The compensation rate is estimated as the ratio of (potential) disposable income as a pensioner in year t to disposable income as a participant in year t-1:
Compensation ratet
=
( potential ) disposable income as a pensionert disposable income as a participant t −1
Year t is the survey year except when retirement takes place between two survey years. In these cases, year t is the actual retirement year (reported in the survey). The calculation is based on the first year as pensioner and is calculated as the sum of the two sources: The estimated amount of (potential) private pension and income from what respondents state as their expected most important income source when retired.
55
The number of discharges and physician visits are measured in year t-1, where year t is the survey year, or alternatively the retirement year if the respondent retires between the two survey years. Recording of accidents is carried out as diagnoses, cf. above. The survey-based instruments are measured in the two survey years. However, we do not know whether the answer given in 2002 by individuals that retire between the two survey years is related to the situation before or after the actual retirement year. We assume that this information is related to the before situation.
156
To estimate potential disposable income as a pensioner, the amount of private pension, which is a lump-sum retirement income payment, is equally distributed over the years from the planned or actual retirement age until the mean life expectancy for the age and gender group in question. Unfortunately, some people are not aware of the payable amount. Namely, 3 out of 4 in the sample reported that they could receive a private pension when they stop working. However, only 66 per cent of this group reported the expected payable amount although a ball park figure was acceptable.
The information about the expected most important income source when retired is based on the survey question: "What do you think is your most important income source at the time, when you stop working?" The possible answers are56: • Post Employment Wage (PEW, efterløn), • Social Disability Pension (SDP, førtidspension), • National Old Age Pension (OAP, folkepension), • Labour market pension, • Private pension, • Private savings, • Income of the spouse, • Own earnings and • Other sorts of income.
56
For more information about these schemes, see Larsen (2004).
157
In case of planned retirement, these answers are applied directly in the calculation, while in case of actual retirement these answers only are applied directly if they are consistent with the retirement age. An example of inconsistent information is retirement at age 58 and PEW as the most important income source when retired. In cases of planned or actual retirement where the answers are labour market pension, private pension, private savings or other sorts of income, no information about income is available.57 If information is inconsistent or no information is available, we assume that SDP is the retirement scheme for individuals that plan to or actually retire before the age of 60, PEW is assumed to be the retirement scheme for individuals that plan to or actually retire at age 60-66, while OAP is assumed to be the retirement scheme for individuals that plan to retire at age 67 or later. This corresponds to the approach applied in Pedersen & Smith (1996).
The income for potential recipients of PEW benefits is calculated as 90 per cent of earnings in year t-1 subject to a ceiling, which differs depending on whether unemployment insurance is on a full-time or a part-time basis. For some of the self-employed, the registered income in year t-1 is negative. In these cases, benefits are set equal to the ceiling. For retirement after 1999,
57
An exception is private pension, cf. above. However, as mentioned above, information about the payable amount is sometimes missing. Further, in many cases the payable amount is too small to be the only source of income. Therefore, if seems reasonable to add a second income source to individuals in this group.
158
income is calculated as 91 per cent of full benefits for individuals that plan to or actual retire at age 60 or 61.58
Estimation of income for recipients of SDP benefits is more complicated. First of all, different levels of SDP are taken into consideration: highest level, intermediate level, increased ordinary level and ordinary level. In addition, we distinguish between men and women and between singles and partnered individuals. For recipients of SDP in our sample, information about the level of SDP is unknown. Therefore, the distribution on the four levels of SDP for each year for men and women respectively is obtained from Statistics Denmark (1997-2002), which includes yearly information about award of SDP. Since the distribution on the four levels of SDP found in Statistics Denmark (1997-2002) is not broken down by family type, an assumption has to be made. We assume that the distribution on these four levels is the same for singles and partnered individuals. These distributions are used as weights to calculate yearly average SDP benefit for each of the four groups.
To estimate an average measure for income as a pensioner for people in the category OAP, three types of OAP benefits are taken into account: Base amount, supplementary amount and special supplementary amount. In addition, distinctions are made between singles and married individuals and between men and women. For each of these four groups, the composition
58
These changes were part of the PEW reform that was conducted in the middle of 1999, see Larsen (2004). Due to the age composition of our sample, these changes are not taken into account until 2000.
159
of OAP benefits for each year for an average recipient is identified on the basis of yearly information about receipts of OAP obtained from Statbank Denmark. Some recipients receive a reduced supplementary amount. The rate of reduction is unknown but we assume that it is 50 per cent on average.
Yearly information about benefit levels and ceilings for PEW, SDP and OAP is found in Lausten (2001) and Danish Insurance Information Service (2001, 2002).
If the expected most important income source is income of the spouse, the (potential) income is estimated as half of the spouse's earnings in year t. If own earnings is the expected most income source, the income is set equal to the respondent's earnings in year t.
The estimate of the denominator, disposable income as a participant, which is based on earnings in year t-1, is described in Pedersen & Smith (1996).
A variable for missing compensation rate is added. That is, compensation rate is set equal to missing, if the estimated rate is equal to or below 0 because a negative compensation rate is assumed to be invalid. In addition, the compensation rate is set equal to missing if this rate is in excess of 1.5. Since all individuals in the sample are in the labour force before retirement, it seems reasonable to assume that the disposable income as a pensioner in year t at its highest is 50 per cent higher than the disposable income as a
160
participant in year t-1. The missing compensation rate variable is set equal to 1, if the compensation rate is missing and 0 otherwise.
161
Appendix A3. Tables Table A1. Means of variables by labour force (LF) status, men and women (standard deviations in parentheses).
Planned Retirement Age
In LF Wave 1 Men Women 61.68 (2.67) 60.86 (2.22)
Out of LF Wave 1 Men Women -
-
Poor health General health Mental health Health compared to others Working capacity Work limitations Functional limitations Diseases Conditions - High blood pressure - Diabetes - Bronchitis/asthma - Osteoarthritis - Myalgia - Osteoporosis/ decalcification of bones - Back problems - Depression Diagnoses - Heart conditions - Strokes - Cancers - Lung diseases - Diabetes - High blood pressure - Arthritis
0.18e 1.21ge 0.05e 2.35 0.19 0.03ge 0.22ge 0.02e 0.01e 0.02e 0.11ge 0.09ge 0.00ge 0.10e 0.02ge 0.07ge 0.03ge 0.00e 0.01 0.01e 0.00g 0.01g 0.01
(0.38) (0.20) (0.22) (1.51) (0.39) (0.17) (0.42) (0.14) (0.08) (0.14) (0.31) (0.28) (0.04) (0.30) (0.12) (0.26) (0.18) (0.06) (0.09) (0.08) (0.06) (0.09) (0.10)
0.17e 1.24e 0.05e 2.28 0.20 0.05e 0.32e 0.02e 0.00e 0.03e 0.15e 0.17e 0.02e 0.12e 0.03e 0.03e 0.00e 0.00 0.01 0.01e 0.00e 0.00 0.01e
(0.37) (0.20) (0.21) (1.59) (0.40) (0.22) (0.47) (0.15) (0.03) (0.16) (0.36) (0.37) (0.12) (0.32) (0.17) (0.18) (0.06) (0.05) (0.11) (0.08) (0.00) (0.00) (0.10)
0.64 1.43 0.47 0.35 0.50 0.10 0.03 0.06 0.29 0.21 0.01 0.30 0.14 0.17 0.09 0.03 0.01 0.03 0.00 0.01 0.01
(0.48) (0.39) (0.50) (0.48) (0.50) (0.29) (0.18) (0.24) (0.45) (0.41) (0.11) (0.46) (0.35) (0.38) (0.29) (0.17) (0.11) (0.18) (0.07) (0.11) (0.11)
0.55 1.37 0.35 0.32 0.55 0.06 0.02 0.11 0.35 0.26 0.07 0.29 0.10 0.11 0.03 0.01 0.02 0.02 0.01 0.00 0.03
(0.50) (0.33) (0.48) (0.47) (0.50) (0.23) (0.14) (0.32) (0.48) (0.44) (0.26) (0.45) (0.30) (0.31) (0.16) (0.07) (0.13) (0.14) (0.11) (0.04) (0.16)
0.42e 0.48g 0.23e 0.10ge 0.48e -
(0.49) (0.50) (0.42) (0.31) (0.50) -
0.41e 0.39e 0.26e 0.20e 0.46e
(0.49) (0.49) (0.44) (0.40) (0.50)
0.57 0.41 0.10 0.38 0.31 -
(0.50) (0.49) (0.31) (0.49) (0.46) -
0.52 0.30 0.07 0.25 0.40
(0.50) (0.46) (0.25) (0.43) (0.49)
12.1e 0.20e 0.57d) 0.18 34.5e 0.10 0.15 0.32 16.5 34.3 0.57 3.60 982
(0.41) (0.41) (0.19) (0.38) (6.38) (0.29) (0.36) (0.47) (10.5) (8.44) (0.49) (1.44)
12.0 (0.61) 0.22 (0.56) 33.2 (8.10) 230f)
Demographic characteristics Born in 1940 Vocational training Higher education Living alone Partner at the same agea) or older Partner at the same age a) or younger
Financial and job characteristics Log income, average of year t-1 and t-2b) Wealth in year t-1 in 1,000,000 d.kr. Compensation rate in year t Compensation rate is missing Experience Self-employed, assisting spouses etc. Skilled, unskilled workers Private sector Tenure
Hours of work Physical demanding job Job satisfactione) Number of observations
12.4ge (0.54) 0.51ge (1.17) 0.46c)g (0.20) 0.11g (0.31) 38.6ge (5.55) 0.13g (0.34) 0.34g (0.47) 0.64g (0.48) 17.5g (11.7) 40.7g (8.58) 0.49g (0.50) 3.59 (1.46) 1224
11.1 (2.88) 0.13 (0.43) 29.3 (8.25) 548 f)
Notes: a) “Same age”: +/- two years; baseline in each case is the most usual pattern b) In cases where earnings is less than zero, this information is replaced by information about surplus of own firm; c) 1093 observations; d) 808 observations; e) “Would choose the current job again”: 1= yes, quite sure; 5 = no, certainly not; f) Number of observations differ for some variables due to missing values. gSignificant gender difference, p0.05.
162
General health 1.00 Mental health 0.32 1.00 Health compared to others 0.38 0.17 Working capacity 0.28 0.32 Work limitations 0.19 0.28 Functional limitations 0.26 0.11 Diseases Conditions 0.30 0.20 High blood pressure 0.06 0.08 Diabetes 0.02 -0.02 Bronchitis/asthma 0.05 0.07 Osteoarthritis 0.29 0.19 Myalgia 0.22 0.20 Osteoporosis etc. -0.02 0.00 Back problems 0.29 0.16 Depression 0.06 0.27 Diagnoses 0.14 0.06 Heart conditions 0.12 0.04 Strokes 0.01 -0.01 Cancers 0.00 0.01 Lung diseases 0.04 0.00 Diabetes 0.00 0.04 High blood pressure -0.01 -0.04 Arthritis 0.12 0.05
1.00 0.24 1.00 0.16 0.24 1.00 0.29 0.12 0.07 0.29 0.24 0.15 0.13 0.05 0.02 0.08 0.03 0.05 0.02 0.06 0.07 0.27 0.21 0.12 0.23 0.22 0.19 -0.01 -0.04 -0.02 0.28 0.22 0.12 0.00 0.09 0.04 0.14 0.08 0.02 0.13 0.03 0.05 0.05 0.02 0.01 -0.02 -0.02 0.00 0.03 0.09 0.01 -0.01 -0.02 -0.03 -0.02 0.02 -0.04 0.12 0.09 0.01
Arthritis
High blood pressure
Diabetes
Lung diseases
Cancers
Strokes
Heart conditions
Diagnoses
Depression
Back problems
Osteoporosis etc.
Myalgia
Osteoarthritis
Bronchitis/ asthma
Diabetes
High blood pressure
Diseases Conditions
Functional limitations
Work limitations
Working capacity
Health compared to others
Mental health
General health
Table A2. Correlation between health measures, men.
1.00 0.09 1.00 -0.03 0.26 1.00 0.05 0.14 0.15 1.00 0.01 0.26 -0.02 -0.01 1.00 0.11 0.64 0.05 -0.03 0.01 1.00 0.08 0.57 0.04 0.02 -0.02 0.31 1.00 -0.01 0.08 -0.01 0.00 -0.01 0.12 0.13 1.00 0.12 0.61 0.01 -0.02 0.04 0.46 0.31 0.12 1.00 -0.02 0.23 0.13 -0.01 0.08 0.06 0.15 -0.01 0.09 1.00 0.08 0.07 0.01 0.02 -0.01 0.07 0.05 -0.01 0.06 0.07 1.00 0.02 0.06 0.04 -0.01 0.01 0.02 0.06 -0.01 0.06 0.12 0.68 1.00 0.07 0.00 -0.01 0.00 -0.01 -0.02 0.03 0.00 -0.02 -0.01 0.21 -0.01 1.00 0.03 -0.03 -0.01 -0.01 -0.01 0.00 0.00 0.00 0.00 -0.01 0.35 -0.02 -0.01 1.00 0.04 0.01 -0.01 -0.01 -0.01 0.01 0.01 0.00 0.01 -0.01 0.30 0.04 0.00 -0.01 1.00 -0.01 0.00 -0.01 0.17 -0.01 -0.02 -0.02 0.00 -0.02 -0.01 0.23 0.06 0.00 -0.01 -0.01 1.00 0.04 -0.02 0.06 -0.01 -0.01 -0.03 -0.03 0.00 -0.03 -0.01 0.31 0.09 0.16 -0.01 -0.01 -0.01 1.00 0.03 0.10 -0.01 -0.01 -0.01 0.17 0.02 0.00 0.07 -0.01 0.38 -0.02 -0.01 -0.01 -0.01 -0.01 -0.01 1.00
Number of observations: 1224.
163
1.00 -0.01 0.08 -0.01 0.00 -0.01 -0.01 -0.01 0.00 0.00 0.00 0.00 . . 0.00
1.00 0.05 1.00 0.01 0.34 1.00 0.03 0.04 0.08 1.00 -0.06 0.39 0.36 0.16 1.00 -0.03 0.06 0.15 0.08 0.15 1.00 0.14 0.07 0.03 0.02 0.06 0.04 1.00 0.09 -0.03 0.01 0.12 -0.02 -0.01 0.35 1.00 -0.01 -0.02 -0.02 -0.01 -0.02 -0.01 0.25 0.00 1.00 -0.02 0.03 0.05 -0.01 0.02 0.04 0.61 -0.01 -0.01 1.00 0.21 -0.04 -0.01 0.09 -0.03 -0.01 0.46 0.18 0.00 0.10 1.00 . . . . . . . . . . . . . . . . . . . . . . 0.11 0.14 0.01 -0.01 0.13 0.05 0.52 -0.01 0.00 -0.01 -0.01
. . .
Arthritis
High blood pressure
Diabetes
Lung diseases
Cancers
Strokes
Heart conditions
Diagnoses
Depression
Back problems
Osteoporosis etc.
Myalgia
Osteoarthritis
Bronchitis/ asthma
High blood pressure
1.00 0.23 1.00 0.17 0.16 1.00 0.22 0.21 0.14 1.00 0.07 0.05 0.09 0.22 1.00 0.07 0.06 0.14 0.05 0.00 0.00 0.06 0.10 0.25 -0.03 0.17 0.15 0.19 0.63 0.05 0.14 0.13 0.04 0.66 0.06 0.12 0.00 0.01 0.18 -0.02 0.19 0.14 0.15 0.53 0.05 0.18 0.19 0.05 0.25 0.02 0.12 0.08 0.25 0.10 0.01 0.10 0.01 0.06 -0.01 -0.01 0.05 0.03 0.09 0.02 0.15 0.03 -0.01 0.06 0.04 -0.02 0.05 0.02 0.04 0.02 -0.01 . . . . . . . . . . 0.08 0.11 0.27 0.12 -0.01
Diseases Conditions
1.00 0.34 0.17 0.31 0.24 0.13 -0.01 0.20 0.20 0.09 0.25 0.18 0.14 0.32 0.14 0.10 0.11 0.10 . . 0.29
Diabetes
Functional limitations
Work limitations
Working capacity
1.00 0.24 1.00 0.41 0.16 0.32 0.19 0.19 0.32 0.26 0.18 0.36 0.21 0.12 0.05 0.07 -0.04 0.09 0.00 0.29 0.16 0.26 0.15 0.12 0.10 0.34 0.19 0.22 0.30 0.18 0.02 0.10 -0.03 0.04 0.02 0.05 0.02 0.06 -0.05 . . . . 0.16 0.06
Health compared to others
General health Mental health Health compared to others Working capacity Work limitations Functional limitations Diseases Conditions High blood pressure Diabetes Bronchitis/asthma Osteoarthritis Myalgia Osteoporosis etc. Back problems Depression Diagnoses Heart conditions Strokes Cancers Lung diseases Diabetes High blood pressure Arthritis
Mental health
General health
Table A3. Correlation between health measures, women.
. . 1.00
Number of observations: 982. Note: All the values for the diagnoses diabetes and high blood pressure are missing because none of the women in the sample have been hospitalised due to these diseases.
164
Table A4. First-stage random effects IV estimates of health equations, men. General health
Parent alive Number of discharges Physician visits 4-10 Physician visits 11-20 Physician visits > 20 Home work assistance Transportation assistance etc. Accident etc. No exercise χ2 test for power of instruments Number of observations
-0.018 (0.022) 0.004 (0.003) 0.011 (0.021) 0.126*** (0.027) 0.172*** (0.038) 0.034 (0.024) 0.079 (0.078) 0.056 (0.066) 0.037*** (0.009) 125***
Subjective health Objective health Mental Health Working Work Functional Diseases Diagnoses health compared capacity limitations limitations Conditions to others 0.007 -0.007 -0.004 -0.008 -0.005 0.004 -0.014 (0.011) (0.013) (0.087) (0.025) (0.011) (0.025) (0.017) 0.000 0.007*** -0.011 -0.005 0.005*** 0.005 0.026*** (0.002) (0.002) (0.012) (0.003) (0.001) (0.004) (0.002) 0.015 0.019 -0.010 0.034 -0.018 0.072** 0.015 (0.011) (0.012) (0.085) (0.024) (0.010) (0.025) (0.017) 0.049*** 0.056*** 0.196 0.073* 0.002 0.211*** 0.046* (0.014) (0.016) (0.108) (0.031) (0.013) (0.032) (0.021) 0.065*** 0.095*** 0.462** 0.136** 0.013 0.268*** 0.053 (0.019) (0.022) (0.151) (0.043) (0.019) (0.044) (0.030) -0.001 0.033* -0.004 0.025 0.019 -0.006 -0.013 (0.012) (0.014) (0.095) (0.027) (0.012) (0.028) (0.019) 0.046 -0.005 0.249 -0.067 0.013 -0.048 -0.017 (0.041) (0.045) (0.318) (0.091) (0.039) (0.093) (0.063) -0.060 0.031 0.154 -0.139 -0.053 -0.147 0.004 (0.034) (0.038) (0.266) (0.076) (0.033) (0.078) (0.053) 0.013** 0.014** 0.061 0.010 0.012** 0.018 0.005 (0.005) (0.005) (0.037) (0.011) (0.005) (0.011) (0.007) 129***
93***
138***
70***
48**
103***
145***
1224
Note: Additional controls include individual earnings, wealth, compensation rate and a dummy variable for missing values for compensation rate, birth cohort, education, cohabitation including age differences between partners, experience, occupation, sector, tenure, hours of work, physical demanding job and job satisfaction. * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
Table A5. First-stage random effects IV estimates of health equations, women. General health
Parent alive Number of discharges Physician visits 4-10 Physician visits 11-20 Physician visits > 20 Home work assistance Transportation assistance etc. Accident etc. No exercise χ2 test for power of instruments Number of observations
-0.038 (0.025) 0.001 (0.006) 0.000 (0.028) 0.063 (0.034) 0.132*** (0.040) 0.070* (0.029) -0.013 (0.073) 0.029 (0.096) -0.020 (0.013) 63***
Subjective health Objective health Mental Health Working Work Functional Diseases Diagnoses health compared capacity limitations limitations Conditions to others -0.025 -0.019 -0.303** -0.026 -0.005 -0.019 -0.023 (0.013) (0.014) (0.103) (0.027) (0.014) (0.030) (0.012) -0.003 0.013*** 0.030 -0.002 0.010** 0.003 0.041*** (0.003) (0.003) (0.025) (0.007) (0.004) (0.008) (0.003) -0.013 -0.006 0.032 0.018 0.015 0.087* -0.026 (0.014) (0.016) (0.115) (0.031) (0.016) (0.035) (0.014) 0.044* -0.007 0.055 0.046 0.024 0.179*** -0.009 (0.017) (0.019) (0.139) (0.037) (0.020) (0.042) (0.017) 0.079*** 0.036 0.500** 0.151*** 0.052** 0.347*** -0.043* (0.020) (0.022) (0.161) (0.043) (0.023) (0.049) (0.019) 0.005 -0.004 -0.081 0.006 0.001 0.022 -0.005 (0.015) (0.016) (0.119) (0.032) (0.017) (0.037) (0.014) 0.066 0.126** 0.843** 0.112 0.037 0.199* 0.084* (0.037) (0.041) (0.297) (0.080) (0.042) (0.092) (0.036) 0.018 0.018 0.763 -0.017 0.040 -0.009 0.005 (0.049) (0.054) (0.391) (0.105) (0.056) (0.121) (0.047) 0.018** -0.014* -0.123* -0.013 0.006 -0.016 -0.005 (0.006) (0.007) (0.052) (0.014) (0.007) (0.016) (0.006) 103***
81***
115***
42*
48**
100**
224***
982
See Table A4 for notes. * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
166
Table A6. Test of overidentifying restrictions, men (standard error in parentheses). General health RE bpoor health REIV All instruments bpoor health Exogeneity test
Subjective health Mental Health Working Work health compared capacity limitations to others
Objective health Functional Diseases Diagnoses limitations conditions
-1.266*** (0.198)
-1.219** (0.406)
-1.221*** (0.348)
-0.067 (0.053)
0.172 (0.186)
-0.428 (0.431)
-0.546** (0.177)
-0.463 (0.257)
-1.892* (0.857)
-5.186 (2.708)
-3.049* (1.553)
-0.754 (0.498)
-2.656 (1.660)
-3.896 (2.650)
-1.847* (0.735)
-1.344 (0.811)
0.63
2.51
1.57
1.94
2.94
1.79
3.32
1.32
-5.242 (2.750)
-3.066* (1.555)
-0.737 (0.505)
-2.663 (1.661)
-3.944 (2.665)
-1.847* (0.733)
-1.353 (0.814)
2.89
1.53
2.21
2.95
1.79
3.40
1.34
-3.118 (1.781)
-0.879 (0.528)
-3.346 (1.801)
-4.047 (3.390)
-1.791* (0.745)
-4.368* (2.147)
1.61
2.54
3.93
1.16
2.97
5.80
-1.698 (2.006)
0.454 (0.962)
-0.667 (2.037)
-2.736 (2.847)
-2.423 (1.770)
-1.021 (0.828)
0.22
10.71
16.47
0.71
4.44
0.55
-3.471* (1.634)
-0.749 (0.500)
-2.868 (1.720)
-4.420 (2.781)
-1.844* (0.735)
-1.336 (0.813)
2.01
1.95
3.16
2.13
3.31
1.29
-3.009 (1.553)
-0.841 (0.522)
-2.543 (1.674)
-4.020 (2.660)
-1.823* (0.735)
-1.331 (0.812)
1.49
2.40
2.67
1.89
3.20
1.30
-5.104 (2.788)
-3.174* (1.561)
-0.823 (0.510)
-2.655 (1.835)
-3.582 (2.713)
-1.811* (0.747)
-1.348 (0.812)
2.01
1.74
2.23
2.40
2.47
3.17
1.32
-6.386* (3.216)
-3.330* (1.665)
-0.868 (0.558)
-2.684 (1.693)
-4.654 (3.026)
-1.890* (0.746)
-1.339 (0.813)
3.30
2.33
2.08
2.90
2.00
3.55
1.30
Minus parent alive bpoor health -1.902* (0.860) Exogeneity test 0.58
Minus number of discharges bpoor health -1.817* -5.099 (0.867) (2.710) Exogeneity 0.55 2.49 test Minus number of physician visits bpoor health -0.814 -2.241 (1.416) (4.177) Exogeneity test 0.25 0.12 Minus home work assistance bpoor health -2.036* -5.174 (0.880) (2.712) Exogeneity 0.81 2.55 test Minus transportation assistance etc. bpoor health -1.963* -5.798* (0.858) (2.743) Exogeneity test 0.97 2.86 Minus accident etc. bpoor health -1.843* (0.860) Exogeneity test 0.86 Minus no exercise bpoor health -2.380* (0.981) Exogeneity test 1.90
(*) Significant at a 10% level, * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
167
Table A7. Test of overidentifying restrictions, women (standard error in parentheses). Subjective health Mental Health health compared to others
Working capacity
-0.659*** (0.180)
-0.961** (0.356)
-0.573 (0.324)
-0.155*** (0.044)
-0.275 (0.170)
-0.419 (0.328)
-0.124 (0.150)
-0.214 (0.346)
-1.866 (1.090)
-2.760 (1.655)
-1.216 (1.538)
-0.249 (0.219)
-1.348 (1.253)
-3.937 (2.489)
-0.854 (0.585)
-0.193 (0.829)
1.48
1.25
n.a.
0.59
0.79
2.28
n.a.
n.a.
-2.313 (1.707)
-0.775 (1.558)
-0.164 (0.244)
-1.059 (1.259)
-3.662 (2.381)
-0.827 (0.555)
-0.048 (0.837)
0.74
n.a.
0.09
0.65
2.30
n.a.
n.a.
-1.804 (2.044)
-0.251 (0.223)
-1.366 (1.259)
-6.820 (3.895)
-0.854 (0.583)
0.858 (2.680)
n.a.
0.57
0.80
2.75
n.a.
3.30
-0.735 (1.678)
-0.190 (0.255)
0.975 (2.444)
-3.052 (2.871)
-0.740 (1.344)
-0.360 (0.843)
n.a.
n.a.
0.90
1.44
n.a.
0.06
-1.308 (1.544)
-0.281 (0.220)
-1.359 (1.232)
-3.905 (2.494)
-0.824 (0.569)
-0.219 (0.829)
n.a.
0.62
0.76
2.22
n.a.
0.00
-2.032 (1.795)
-0.379 (0.249)
-1.773 (1.419)
-4.514 (2.691)
-0.996 (0.646)
-0.298 (0.840)
3.69
1.54
1.24
2.38
2.37
0.03
-2.657 (1.534)
-0.939 (1.555)
-0.156 (0.228)
-1.410 (1.264)
-3.421 (2.666)
-0.842 (0.607)
-0.181 (0.829)
5.65
n.a.
0.11
0.84
1.30
1.80
0.00
-2.453 (1.820)
-1.647 (1.590)
-0.383 (0.239)
-1.665 (1.327)
-3.583 (2.146)
-0.920 (0.612)
-0.246 (0.831)
0.72
n.a.
1.29
1.12
7.72
2.25
0.01
General health RE bpoor health REIV All instruments bpoor health Exogeneity test
Minus parent alive bpoor health -1.606*** (1.142) Exogeneity test 0.87
Minus number of discharges bpoor health -1.861 -2.875 (1.091) (1.679) Exogeneity 1.47 1.37 test Minus number of physician visits bpoor health -1.530 -3.943 (1.489) (2.872) Exogeneity test 4.62 1.10 Minus home work assistance bpoor health -1.615 -2.670 (1.191) (1.643) Exogeneity 0.91 1.22 test Minus transportation assistance etc. bpoor health -1.852 -3.218 (1.089) (1.685) Exogeneity test 1.39 2.10 Minus accident etc. bpoor health -1.777 (1.087) Exogeneity test 1.29 Minus no exercise bpoor health -2.341* (1.151) Exogeneity test 2.28
Objective health Work Functional Diseases Diagnoses limitations limitations conditions
(*) Significant at a 10% level, * Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level.
168
Table A8. Means for health variables by changes in (planned) retirement age, men (standard deviations in parentheses). All Poor general health Both waves Wave 2 Only Worsened mental health Wave 1 to wave 2 Worse health than others Both waves Wave 2 Only Reduced working capacity Wave 1 to wave 2 Work limitations Both waves Wave 2 Only Functional limitations Both waves Wave 2 Only Diseases conditions Both waves Wave 2 Only Diagnoses Both waves Wave 2 Only Number of observations
Changes in (planned) retirement age Wave 1R > Wave 1R = Wave 1R < Wave 2R Wave 2R Wave 2R
0.12 0.05
(0.32) (0.22)
0.18 0.03
(0.39) (0.17)
0.15 (0.35) 0.08 (0.27)
0.06 0.05
(0.25) (0.21)
0.24
(0.43)
0.16
(0.37)
0.23 (0.42)
0.30
(0.46)
0.03 0.02
(0.17) (0.15)
0.06 0.02
(0.24) (0.15)
0.04 (0.19) 0.03 (0.17)
0.01 0.02
(0.09) (0.14)
0.24
(0.42)
0.19
(0.39)
0.19 (0.40)
0.29
(0.45)
0.10 0.09
(0.30) (0.28)
0.12 0.07
(0.32) (0.25)
0.13 (0.34) 0.05 (0.22)
0.06 0.13
(0.25) (0.34)
0.02 0.01
(0.13) (0.09)
0.02 0.01
(0.15) (0.09)
0.02 (0.15) 0.00 (0.07)
0.01 0.01
(0.09) (0.11)
0.14 0.09
(0.35) (0.29)
0.24 0.07
(0.43) (0.26)
0.14 (0.35) 0.08 (0.27)
0.09 0.11
(0.29) (0.32)
0.00 (0.00) 0.09 (0.28) 136
0.02 (0.14) 0.07 (0.26) 211
0.01 (0.11) 0.03 (0.18) 265
0.01 (0.11) 0.06 (0.24) 612
169
Table A9. Means for health variables by changes in (planned) retirement age, women (standard deviations in parentheses). All Poor general health Both waves Wave 2 Only Worsened mental health Wave 1 to wave 2 Worse health than others Both waves Wave 2 Only Reduced working capacity Wave 1 to wave 2 Work limitations Both waves Wave 2 Only Functional limitations Both waves Wave 2 Only Diseases conditions Both waves Wave 2 Only Diagnoses Both waves Wave 2 Only Number of observations
170
Changes in (planned) retirement age Wave 1R > Wave 1R = Wave 1R < Wave 2R Wave 2R Wave 2R
0.09 0.09
(0.29) (0.29)
0.18 0.05
(0.38) (0.21)
0.06 (0.23) 0.11 (0.31)
0.07 0.10
(0.26) (0.30)
0.27
(0.44)
0.19
(0.39)
0.27 (0.45)
0.31
(0.46)
0.02 0.03
(0.15) (0.17)
0.06 0.02
(0.23) (0.14)
0.01 (0.12) 0.04 (0.19)
0.01 0.03
(0.11) (0.17)
0.27
(0.45)
0.15
(0.36)
0.31 (0.46)
0.30
(0.46)
0.11 0.09
(0.32) (0.29)
0.19 0.01
(0.39) (0.10)
0.11 (0.31) 0.10 (0.30)
0.07 0.13
(0.26) (0.33)
0.03 0.01
(0.18) (0.09)
0.08 0.00
(0.28) (0.00)
0.03 (0.18) 0.01 (0.10)
0.01 0.01
(0.08) (0.11)
0.22 0.10
(0.42) (0.31)
0.33 0.06
(0.47) (0.23)
0.22 (0.41) 0.12 (0.33)
0.17 0.11
(0.38) (0.32)
0.00 (0.05) 0.04 (0.19) 491
0.01 0.06
(0.10) (0.23) 107
0.00 (0.00) 0.04 (0.20) 208
0.00 0.02
(0.00) (0.13) 176
Chapter 4 An Experimental Analysis of the Effect of an Increase in Delaying Incentives in the Post Employment Wage Program on Retirement Age†
Mona Larsen∗
JEL Codes: J14, J26.
Abstract This paper looks at the effect of a change of the popular Danish early retirement program, known as the post employment wage (PEW) or efterløn, on retirement age. In order to reduce early retirement through PEW, the scheme was changed in 1992 by increasing the incentives to delay retirement until age 63. To analyse the effect of this change on retirement outcomes, an experimental analysis is carried out using longitudinal register data covering two per cent of the population for the period 1980-1998. Results suggest that the response of retirement age to the change in the PEW program was relatively small.
†
This work is part of the research of the Graduate School for Integration, Production and Welfare. Financial support from the Danish Social Science Research Council is gratefully acknowledged. This work benefited from comments by Jeffrey Smith, Paul Bingley, Peder Pedersen, Nabanita Datta Gupta and Paul Schou on earlier drafts of this paper. Thanks also to the participants at the IZA Summer School 2003, the CEBR Conference 2003 in Copenhagen and the First Meeting with the Strategic Advisory Board of the Danish National Institute of Social Research 2003 in Korsør for helpful comments and suggestions. All remaining errors are my own. ∗
The Danish National Institute of Social Research, Herluf Trolles Gade 11, DK-1052 Copenhagen K,
[email protected].
171
1. Introduction Early retirement schemes are regarded as some of the crucial pull-factors that make senior workers retire early, see e.g. Börsch-Supan (2000); Gruber and Wise (1999) and Blöndal and Scarpetta (1998). If fact, these schemes are considered to be too generous. That is, they contain large disincentives to continue to work after the early retirement age. Conversely, previous international studies indicate that the response of retirement to changes in social security benefits is either rather limited or non-existent. According to Krueger and Pischke (1992) there is no evidence that reducing Social Security benefits slow down the trend to earlier retirement. Mitchell and Fields (1984) show that government practices, which alter the rewards for retirement, will influence older workers’ labour market behaviour in predictable ways but their results indicate a relatively small response. Namely, in general they find that rather large changes in policy variables such as taxes or benefits would be required in order to elicit substantial changes in retirement ages. Similarly, Börsch-Supan (2000) finds that an introduction of an actuarially fair benefit formula induces a shift of the cumulative distribution of retirement age to the right. The effects are most powerful for very early retirement. However, calculations on the basis of Börsch-Supan (2000, Figure 8) show that the marginal effects of this introduction are quite limited. In fact, the average retirement age only rises by 0.45 years59.
The response to changes in social security benefits has also been examined in the Danish context. Danø et al. (2000) have entered the main elements of
59
I owe this point to Nelissen (2002).
172
the Post Employment Wage (PEW, efterløn) reform in 1999 in a quantitative study to simulate the effects of the policy changes. They show that the reform will increase the average retirement age. The effect differs considerably between different levels of education and income and the effect is larger for women than for men. Christensen and Datta Gupta (2000) use micro data on couples and simulate the effect of two different types of pension reforms. They find that delaying PEW eligibility from 60 to 62 will lead to a greater increase in the average retirement age (0.6 years for husbands and 0.75 years for wives) than a reduction in benefits (0.22 years for husbands and 0.65 years for wives). Finally, Bingley et al. (2003) show that the provisions of the social security system play an important role in determining retirement behaviour in Denmark. Among other things, they simulate the effect of a raise in the age of first eligibility of all early retirement programs and normal retirement by three years. They find quite a large delaying impact of this three-year reform on the retirement of both men and women.
The focus in this paper is on the effect of changes in the PEW program on the average retirement age. The PEW program is the most popular early retirement scheme in Denmark60. When the PEW scheme was introduced in 1979, the intention was to offer some form of early retirement for workers who were worn-out as a result of a physically or psychologically demanding work-life. However, today the use of the scheme is widespread even among relatively healthy older workers. In order to reduce early retirement through PEW, this scheme was changed in 1992 and 1999. The effect of 60
In 2001, almost 160,000 received PEW benefits, see Department of Unemployment Insurance (2001). That is, more than 40 per cent of the 60-66-year-olds.
173
the changes in 1999 is of current interest in Denmark. However, the focus in this paper is on the effect of the changes in 1992 because information from many more after-treatment years is available. In 1992, the incentives to delay retirement until at least the age of 63 were increased. Before 1 March 1992, PEW benefits were equivalent to unemployment benefits the first two and a half years. After two and a half years, PEW benefits were reduced to a maximum of 80 per cent of unemployment benefit. After 1 March 1992, the reduction in PEW benefits after two and a half years was removed for individuals that delayed retirement until at least age 63, who instead received a benefit equivalent to the unemployment benefit over the whole PEW period. For further details, see Section 2.
While previous Danish studies of the response to changes in social security are based on simulations, an experimental analysis is carried out in this paper. Experimental analysis may be useful in this regard because causal responses of retirement to policy changes can be inferred. Namely, by using variation in explanatory variables generated by policy changes this approach allows me to obtain variation that is plausible exogenous. Further, this variation is readily examined without requiring strict structural modelling assumptions. Difference-in-differences (DD) and difference-indifference-in-differences (DDD) estimators are applied. The use of this framework to examine the response to changes in social security benefits is limited61 because social security most often are national programs, in which system parameters apply uniformly to all individuals and reforms are universal. It is possible to create a comparison group and apply the approach 61
An example is Baker and Benjamin (1999), who apply the DD approach to examine the introduction of early retirement provisions in Canada’s two public pension plans.
174
in this context because not all individuals are eligible for the PEW benefits62. The comparison group is people eligible for Public Employee Pension (PEP, tjenestemandspension). PEP is available for public servants that are state-employed or employed in the public school system, the national church or the Danish Parliament. To receive PEP, at least 10 years of service is required. People eligible for PEP make a suitable comparison group to people on PEW because this group is not affected by the change in the PEW policy in 1992 and because it seems reasonable to assume that the composition of this group is exogenous. For further information about this group, see Section 3.
The analysis is based on longitudinal register data created for administrative purposes. The dataset covers a representative two per cent sample of persons and their spouses respectively for the period 1980-1998. The results control for the announcement of the policy change. Employing two different treatment groups, changing the time period and conducting separate analyses for each of the ages in focus constitute further control.
The remainder of the paper is organized as follows. The PEW program and the changes in 1992 are introduced in Section 2 while the PEP scheme is described in Section 3. Data and empirical models are described in Section 4 and 5 respectively. The results are presented in Section 6 whereas Section 7 concludes and discusses the results.
62
The rules for entitlement to PEW benefits are described in Section 2 below.
175
2. Post Employment Wage (PEW, efterløn) and the changes in 1992 Not all individuals are eligible for the PEW benefits. To be entitled to these benefits, a number of conditions must be fulfilled. The conditions have been changed several times since the introduction of the program in 1979. To be entitled to PEW on 1 March 1992, at which time the changes in the PEW program came into force, the worker must: 1) be 60-66 years of age. 2) have been a member of an unemployment insurance fund in 20 of the last 25 years. However, for people born before 1 March 1952, it was 10 of the last 15 years provided that the membership of the unemployment insurance fund was continuous from 31 March 1992 until transition to PEW63. 3) have been working full-time in at least 26 weeks in the last three years64, be entitled to unemployment insurance benefit or have received transitional benefit. The Transitional Benefit Program (TBP, overgangsydelse) was introduced 1 March 1992 and was in force in the period 1992 to 199665. After 1 January 1995, people that have received part-time PEW were also entitled. 4) live in Denmark.
63
Before 1 March 1992, it was 10 out of 15 years for everyone.
64
After 1 January 1997, the condition was changed to at least 52 weeks in the last three years.
65
For more information about this scheme, see Section 4.
176
In principle, it is possible for people aged 60-66 years to leave the PEW scheme once and re-enter in the scheme provided that the third condition mentioned above is fulfilled once again. In practice, however, very few make use of this possibility and therefore, these transitions are ignored in the analyses.
In 1992, the normal retirement age was 67 (now it is 65 for people born after 1 July 1939). At this age, PEW benefit recipients are transferred to old age pension. The PEW program is generous in the sense that the amount of benefits does not influence the level of old age pension, as it is independent of previous earnings. The replacement rate, when PEW benefits are received, is on average 70 per cent of labour market income for low-income workers and 40 per cent for high-income workers, cf. Bingley et al. (2003).
In the period studied, the initial PEW benefits did not depend on the age at entry. Instead, it was reduced as a function of time spent in the scheme. Before 1 March 1992, PEW benefits were equivalent to unemployment benefit the first two and a half years. Unemployment benefits amount to 90 per cent of previous earnings subject to a ceiling66.
After two and a half years, PEW benefits were reduced to maximum 80 per cent of unemployment benefits (after 1 January 1994, it was changed to maximum 82 per cent of unemployment benefits). After 1 March 1992, the reduction in PEW benefits after two and a half years was removed for some 66
This ceiling is reached quite fast. In fact, the gross compensation percentage for an average production worker (as defined by OECD) was 53 in 1997, see Hansen (1999).
177
people. That is, persons that delayed retirement until at least age 63 received a benefit equivalent to the unemployment benefit in the whole PEW period67. Thereby, annual PEW benefits for people retiring at age 63 and 64 were increased on average, while the situation was unchanged for people that retired at age 60 to 62 and age 65 to 66, cf. Figure 1 below.
Figure 1. The average annual rate of PEW benefit before and after the changes of the PEW program in 1992. Per cent of unemployment benefit. By retirement age. Average UIB rate pr. year 100 95 90 85 80 60
61
62
63
64
65
66
Retirement age Before ´92
After ´92
Note: Reduced PEW benefits correspond to maximum 80 per cent of unemployment benefit.
67
The rules described applied until the reform of the PEW in 1999. In 1999, the requirements of entry into the scheme were tightened up. The worker must have been a member of an unemployment insurance fund in 25 of the last 30 years. Entitlement also presupposes that a contribution to PEW is paid for a certain number of years. That is, PEW is moving from a pay-as-you-go system toward a funded system. In 2001, 77 per cent of the 50-59-year-olds paid PEW contribution; see Quaade (2002), and presumably, most of them are entitled to this scheme at age 60. Benefits depend on the age at entry instead of being reduced as a function of time spent in the scheme and benefits are means tested against income from all other pension schemes for people aged 60-61. Furthermore, a flexible scheme has been introduced where the benefits are reduced gradually as the number of working hours increases. Finally, a tax premium has been introduced for people who are entitled to early retirement benefits at age 60 but who continue to work at least to the age of 62.
178
The changes in 1992 implied that the financial incentives to delay early retirement from age 60 to 62 until the age of 63 were increased. The effect on the incentives to retire at age 64 is uncertain. On the one hand, PEW benefits when retirement occurs at age 64 was increased compared to the situation before the change, and the benefit amount was the same when retiring at age 65 to 66, which increases the incentives to retire at age 64. However, the benefit amount was also the same when retirement occurs at age 63, which decreases the incentives to retire at age 64. Namely, pension benefits are actuarially adjusted so that they are typically higher per year, when retirement is delayed; see e.g. Mitchell and Fields (1984). But people that delayed retirement from age 63 to age 64 were not compensated for receiving benefits for a shorter period. Similarly, people that delayed retirement until age 65 or 66 did not get any compensation either, and therefore, the incentives to delay retirement until these ages were decreased. All in all, the aggregate effect of the changes in the PEW policy on the average retirement age is uncertain. Provided that the changes of the financial incentives have the intended effect, the effect on the average retirement age depends on the distribution of the retirement age before 1992.
3. Public Employee Pension (PEP, tjenestemandspension) PEP is available for public servants that are state-employed or employed in the public school system, the national church or the Danish Parliament. A public servant is entitled to PEP from the age of 60 and retirement is mandatory at the age of 70 unless lower age limits for the job in question are specified. Furthermore, to receive PEP, at least 10 years of service is re-
179
quired68. PEP is the only Danish public sector program that ties benefits to past record, see e.g. Bingley et al. (2003). Namely, the programs are actuarially adjusted, as a reduction in benefit amount is present for retirement either before age 6769 or before the mandatory retirement age, if this is lower than 67. The reduction depends on the retirement age and the number of “pension years” that have been accumulated up to the point of retirement70. The number of “pension years” is the number of years that a public servant has been full-time employed in public service after attaining the age of twenty-five. The number of pension years cannot exceed 37. PEP recipients are not transferred to old age pension, when they reach sixty-seven years of age. Instead, they continue to receive PEP the rest of their lives.
People entitled to PEP are chosen as the comparison group to people entitled to PEW. Firstly, this is because these individuals are not affected by the change in the PEW policy in 1992. In addition, it seems reasonable to assume that the composition of the group of individuals entitled to PEP is exogenous. One problem could have been that people were jumping between PEW and PEP but it is unlikely that the composition of the two groups change due to the policy changes in 1992 because rules dictate that eligibility to PEW and PEP respectively is a function of a long labour market history way back in time. Another objection could be that the comparison group is not pure in the sense that in addition to people eligible for PEP
68
However, particular rules apply to public servants with at least three years of service that leave their job before the early retirement age.
69
After 1 July 1999, this was changed to age 65 for people born after 1 July 1939.
70
After 1 January 1994, the maximum benefit was 57 per cent of the wage when retiring.
180
other public employees are also included in this group. However, the composition of the comparison group is fairly stable over time, cf. Table A.1 in the Appendix. Consequently, this group in its entirety seems to be a suitable comparison group.
4. Data The data used are obtained from a Danish longitudinal register dataset created for administrative purposes. The dataset contains a representative two per cent sample of individuals and their spouses respectively for the period 1980-1998. In this paper, the focus is on 60-66-year-olds that were in the labour market when aged 59. Therefore, information about people born in the period from 1921 to 1938 is applied, resulting in a sample in which individuals in the oldest cohort were 59 years old in 1980, while individuals in the youngest cohort were 60 years old in 1998. This selection implies that a representative sample of 60-66-year-olds that were in the labour market when aged 59 is available for the period 1987-1998. This sample includes 10,922 individuals.
To analyse the effect of the change of the PEW policy in 1992, DD and DDD analyses are conducted. The treated are people entitled to PEW, while people eligible for PEP constitute the control group. Entitlement to PEW is determined on the basis of information about receipt of post employment wage benefits and retrospective information on membership in an unemployment insurance fund. More precisely, individuals are considered as entitled to PEW if they received post employment wage benefits during the sample period or had been a member of an unemployment insurance 181
fund for at least 10 years until age 60 or, if left-truncated, every year from 1980 until age 60. In counting the number of membership years, the condition of continuous membership on and after 1992 is taken into account. Two treatment groups are constructed. A broader group (PEW1) consists of individuals that fulfil the conditions just mentioned, while a smaller group (PEW2) consists of a subsample of individuals from the PEW1 group, who have been publicly employed for at least 10 years until age 60 or, if lefttruncated, every year from 1980 until age 60. The PEW1 group makes up 78 per cent of the selected sample, while the PEW2 group accounts for 26 per cent. In the residual group, eligibility to PEP is determined on the basis of information about receipt of public employment pension and retrospective information on public employment. In fact, if individuals have received public employment pension during the sample period71 or had been publicly employed according to the same criteria as individuals in the PEW2 group, they are included in the PEP group (PEP). The PEP group accounts for 12 per cent of the selected sample.
The idea behind constructing two treatment groups is to be able to control the robustness of the results to selections made to increase comparability between the treatment and control group. However, two control groups are not constructed because while the number of individuals in the selected sample that is entitled to PEW is relatively large, the number of individuals included in the PEP group is too small to impose further restrictions on this group. Another potential control group could have been people entitled to neither PEW nor PEP. However, this group is too heterogeneous and there71
Unfortunately, information about receipt of public employment pension is only available for the years 1997 and 1998.
182
fore not suitable for this purpose. The point of restricting PEW2 to public employees with long tenures is to make the treatment group more comparable with respect to education, income, previous unemployment and sector - all factors that affect retirement behaviour. The larger comparability as regards previous unemployment and sector is obtained through the selection criteria. These criteria have also resulted in larger comparability with respect to education and income as intended, although a relative large difference still remains, cf. Table A.1 in the Appendix. Conversely, however, PEW2 differs more from PEP with respect to gender than PEW1. In fact, PEW2 is dominated by women, while a minor predominance of men is found in PEP and PEW1. Gender has also a pronounced effect on retirement behaviour and larger comparability would be obtained if separate analyses for men and women were conducted and if the treatment group was restricted further with respect to education and/or income. Due to small sample size, however, this is not possible. Instead, differences with respect to gender, education and income are at least partly controlled for since they are included as explanatory variables in the DD and DDD analyses.
The sample is restricted to individuals who were wage earners or temporarily out of employment due to unemployment or leaves when aged 59. Individuals, who were outside the labour force when aged 59, are excluded because they were not entitled to either PEW or PEP when aged 60. Selfemployed and assisting spouses are excluded to make the individuals in the two PEW groups as comparable as possible to the group of people entitled to PEP, which consists exclusively of wage earners. Labour market status at the age of 59 and the retirement year are determined on the basis of
183
yearly information about the primary labour market status at the end of November.
5. Empirical models The focus is on the average treatment effect. That is, the average effect of the change in the program on people with access to PEW rather than only on people, who choose to participate in the program. To examine the effect on the treated, two cross-section samples of 60-66-year-olds entitled to PEW before and after 1992 respectively are compared to the control group on the basis of two cross-section samples of 60-66-year-olds eligible for PEP.
Both DD and DDD estimators are applied. The DDD framework is the most appropriate approach in this case. However, to be able to control the results and due to the small size of the comparison group, DD analyses are also conducted.
The legitimacy of the DD approach is grounded on three assumptions: The change in the PEW policy provided exogenous variation in the outcome of retirees; secular trends in labour market behaviour are common to individuals in both groups and there was no scheme specific shock to behaviour co-incident with the change in the PEW policy in 1992.
As for the first assumption, the Minister of Labour brought in a bill about changing the PEW program in October 1991 and the law was passed in De184
cember 1991. According to the Minister, the idea about changing the program arose from an article in a newspaper half a year earlier, cf. “Sortbog” (1991). The timing of these events makes it likely that my results are biased by the actions of individuals in 1991 who anticipated the changes of the PEW program in 1992. I control for the announcement of the policy change in 1992 by comparing results of two analyses in which the before observation is measured in 1990 and 1991 respectively.
Since the reform in 1992 implied that nobody entitled to PEW in 1991 was financially worse off if they delayed retirement until after the policy changes in 1992, while in particular the 60-62-year-olds were better off, the potential effect of the announcement on the retirement age in 1991 would be positive. This indicates that this potential effect would decrease the estimated effect of the changes in the PEW policy in 1992 when 1991 is used as the before observation. However, in 1991 it was also suggested that the age at which PEW benefits were first available should be raised from 60 to 62 years. For a while, the Minister of Labour received many letters from angry citizens that believed that this change would be effected the 1st January 199272, cf. “Sortbog” (1991). These expectations in 1991 might have induced some people to retire earlier than planned. Therefore, it is not straightforward to predict the potential effect of using 1990 and 1991 respectively as the before observation. In fact, the two contrary effects might have cancelled out.
72
This part of the bill was not passed. Today, the age at which PEW benefits are first available is still 60 years.
185
Figure 2. Average retirement age for 60-66-year-olds, PEW1, PEW2 and PEP, 1987-1994. Average retirement age 64
63
62
61 1987
1988
1989
1990
1991
1992
1993
1994
Year PEW1
PEW2
PEP
The second assumption is that secular trends in labour market behaviour are common to individuals in both the treatment and the control group. Comparing the development of the average retirement age for the control group and the two treatment groups respectively can test this assumption, cf. Figure 2 above. It appears that the average retirement age did not tend to move in parallel up to 1992. The largest difference is found when PEP is compared to PEW1. Part of this difference might be explained by the fact that contrary to people entitled to PEW, public servants are in general characterized by few layoffs and a high degree of job protection. The curves for PEW2 and PEP are more similar although not completely parallel either. Part of this difference might be explained by different gender composition of the two groups. Therefore, the overall picture is that individuals eligible for PEP do not seem to experience all of the other influences that affect people entitled to PEW.
186
One way to get rid of the different trend is to control for this difference by including a third dummy capturing the trend adjustment in the analysis. In this way the DD estimator is extended to a DDD estimator. I return to this approach in the end of this Section.
The third assumption that there was no scheme specific shock to behaviour co-incident with the changes in the PEW policy in 1992 that affected the two groups differently might also be problematic. In fact, at the same time as the PEW program was changed in 1992, the TBP was introduced. This scheme, which was a sort of extension of the PEW program, was available in the period 1992 to 1996 for people 55-59 years old (from 1994 also the 50-54-year-olds) who were members of an unemployment insurance fund and had been unemployed for at least 12 out of 15 months. That is, people entitled to this scheme constitute a subset of people that are entitled to PEW when aged 60. Therefore, the existence of the TBP program might cause a potential sample selection problem since people that take up this scheme are not selected for the PEW in the analysis because they are outside the labour force at the age of 59. However, the entry of people to the TBP scheme was relative limited until 1995. In fact, only up to 2 per cent of the 55-59-year-olds received TBP benefits in period 1992-1994, while the share rose to 5 and 9 in 1995 and 1996, respectively. Therefore, to minimize the potential sample selection problem and due to small sample size for the control group, information after 1994 is not utilized in the analyses.
Furthermore, using different time periods controls for the assumption that there was no scheme specific shock to behaviour co-incident with the 187
change in the PEW policy in 1992. In fact, the situation in 1990 and 1991 is compared to the situation after 1992 based on both pooled information on 1993 and 1994 and information on 1993 and 1994, respectively. Descriptive statistics for 1990 and 1991 for 60-66-year-olds that retired the year in question is shown in Table A.2 in the Appendix.
Finally, separate analyses are also conducted for each of the retirement ages. In fact, the effect of the changes of the PEW program might be that retirement in fact was delayed from age 60 to 62 until the age of 63, but the effect on average was zero because retirement at age 65 to 66 at the same time were hastened. Therefore, in addition to the analysis on the whole sample of 60-66-year-olds that retire the year in question, separate analyses for each of the ages 60 to 66 years are conducted. In this case, the focus is on individuals that were in the labour market at age x-1.
The DD estimator measures the excess outcome growth for the treated compared to the non-treated. If only focusing on the treatment indicator, formally the DD estimator can be written as:
T
T
C
C
1
0
1
0
α~DD = (Y t − Y t ) − (Y t − Y t )
(1)
T
C
where α is the treatment impact and Y and Y are the mean outcomes for the treatment and the control group respectively.
188
The DD estimate of the effect of the change in the PEW policy in 1992 is computed in an OLS regression on repeated cross sections consisting of pooled micro data for groups and years; see Angrist and Krueger (1999). The regressors consist of dummies for years and groups and an interaction term equal to the product of the dummies for years and groups. A vector of individual characteristics is also added:
Yit =β1X’it + β2 AFTERt + β3 PEWi + β4 (AFTERt*PEWi) + εit
(2)
The included individual characteristics (Xit) are gender, education, cohabitation status and lagged income. The dummy for year (AFTERt) is equal to one when the after observation is 1993/1994, 1993 and 1994, respectively and the dummy for group (PEWi) is equal to one for individuals in the PEW1 and the PEW2 group respectively. The coefficient in focus is the one on the interaction term (β4). A significant positive estimate of this coefficient indicates that the treatment had the intended effect. In the analysis of the whole sample of 60-66-year-olds, the dependent variable (Yit) is the retirement age for individuals that retired the year in question, while in the separate analyses for each of the ages 60 to 66 years, this variable is a dummy that equals one if retirement takes place at the age in question.
It is problematic to assume that secular trends in labour market behaviour are common to individuals entitled to PEW and PEP respectively. For that reason, the use of the DD estimator is debatable. As stated in Blundell and Costa Dias (2000), formally the problem can be written as:
189
E (α~DD ) = α + (k T − k C )(θ t1 − θ t0 )
(3)
where θt is the macroeconomic effect and kT and kC indicate the differential macro effect across the treatment (T) and the control (C) groups. The true effect of the treatment is only recovered when kT = kC. A possible solution to this problem is according to Blundell and Costa Dias (2000) that another time interval t* to t** is taken over which a similar macro trend has occurred. More precisely, a period for which the macro trend matches the term (kT − k C )(θ t − θ t ) in (3) is chosen. The differential adjusted DD or the 1
0
DDD estimator, which will consistently estimates α, takes the following form:
T
T
C
C
1
0
1
0
T
T
C
C
α~DDD = [(Y t − Y t ) − (Y t − Y t )] − [(Y t − Y t ) − (Y t − Y t )] **
*
**
*
(4)
In order to identify a period in which a macro trend similar to the trend around 1992 has occurred, a figure corresponding to Figure 2 but excluding observations for 1992 is constructed, cf. Figure A.1 in the Appendix. If observations for 1990 and/or 1994 are included, it is not straightforward to identify a period in which the macro trend is similar to the trend around 1992. Therefore, 1991-1993 is the chosen period in focus. Comparing the development in this period to the development in the preceding years, the period 1988-1989 seems to be the most appropriate choice. In other words, t* is set equal to 1988, while t** is set equal to 1989. A similar approach has been applied in Blundell and Costa Dias (2000). By using observations for 1991, it is assumed that the announcement of the policy change in 1992 did
190
not affect the estimated effect of this change, while the use of observations for 1993 implies that only short run effects are examined.
By extending (2) with a dummy to adjust for the different macroeconomic effect, the DDD estimate of the effect of the change in the PEW policy in 1992 is obtained:
Yijt = β1X’ijt + β2 AFTERt + β3 PEWi + β4ADJUSTj + β5 (AFTERt*PEWi)+ β6 (AFTERt*ADJUSTj) + β7 (PEWi* ADJUSTj) + β8 (AFTERt*PEWi* ADJUSTj) + εijt
(5)
This estimator is calculated on the basis of information for 1988, 1989, 1991 and 1993. The dummy for adjusting for different trend (ADJUSTj) equals one for observations for 1991 and 1993, while the dummy for years (AFTERt) equals one for observations for 1989 and 1993. In this model, the estimate of the coefficient on the interaction term equal to the product of the dummies for years, groups and for adjustment for different trends (β8) indicates whether the treatment had the intended effect.
6. Results To examine the effect of the increase in delaying incentives in the PEW program on the retirement age, DD and DDD analyses are conducted. However, before turning to the results of these analyses, the distribution of retirement age before and after 1992 and retirement age hazards are shown.
191
6.1. Distribution of retirement age and retirement age hazards The changes of the PEW program in 1992 caused the financial incentives to retire at age 60-62 and age 65-66 to decrease, the incentive to retire at age 63 to increase, while the effect on retirement at age 64 was uncertain. Therefore, provided that the changes of the financial incentives have the intended effect, the effect on the average retirement age depends on the distribution of retirement age before the changes of the PEW policy.
Before 1992, the majority of individuals entitled to PEW retired before the age of 63, cf. Figure 373 below. Therefore, the average retirement age increased if the changes in the financial incentives in 1992 had the intended effect. For PEW1, however, the distribution of the retirement age after 1992 was very similar to the situation before 1992. For PEW2, the share of people that retired at age 62 certainly decreased, but at the same time retirement at age 60 increased, retirement at age 65 decreased, while the tendency to retire at age 63 was almost unchanged. These results indicate that the changes of the financial incentives did not work as intended.
73
The curves for the average retirement age before 1992 are based on pooled information for the period 1987-1990. Information about 1991 is excluded due to the potential announcement effect, cf. above. The “after 1992” curves are based on pooled information for the years 1993 to 1994. Curves for PEW based on pooled information for the period 1993-1998 are very similar to the “PEW1 after 1992” and the “PEW2 after 1992” curves shown in Figure 3. Due to lack of observations for PEP, it is not possible to construct an “after 1992” curve based on pooled information for the period 19931998.
192
Figure 3. Distribution of the retirement age among 60-66-year-olds, a) PEW1 and PEP, b) PEW2 and PEP, before and after 1992. Figure 3a)
b) Figure 3b)
Per cent
Per cent
45 40 35 30 25 20 15 10 5 0
40 35 30 25 20 15 10 5 0
60
61
62
63
64
65
66
Retirement age
60
61
62
63
64
65
66
Retirement age
PEW1 before 1992
PEW1 after 1992
PEW2 before 1992
PEW2 after 1992
PEP before 1992
PEP after 1992
PEP before 1992
PEP after 1992
Another way to look at the situation before and after 1992 is to show retirement age hazards for PEW1 and PEW2 74 75. Here, we get a slightly different picture, cf. Figure 4 below. For PEW1 as well as PEW2, the hazard rate for retirement at age 63 increased after 1992 as expected, while the hazard rate for retirement at age 62 decreased slightly. In addition, the hazard rate for retirement at age 64 increased. This increase might be due to increased PEW benefits after 1992 when retirement occurs at this age. That is, the financial incentives seem to work as intended. However, at the same time the hazard rate for retirement at age 60 also increased suggesting that changes of the financial incentives in 1992 were too small to persuade people to delay retirement from age 60 to age 63. In addition, the hazard rate 74
The curve ”after 1992” does not include information about retirement at age 66 because the oldest individuals in the sample (individuals that were 60 years old in 1993) were only 65 years old in 1998, which is the last year in the dataset.
75
Unfortunately, it is not possible to calculate similar retirement age hazards for people entitled to PEP due to small sample size.
193
for retirement at age 65 decreased. Therefore, the retirement age hazard functions also indicate that the positive effect on the average retirement age is doubtful.
Figure 4. Retirement age hazards of transition to retirement, a) PEW1, b) PEW2, before and after 1992. Figure 3a)
Figure 3b)
Hazard rate
Hazard rate
0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.1
0.0
0.0 60
61
62
63
64
65
Retirement age Before 1992
66
60
61
62
63
64
65
66
Retirement age After 1992
Before 1992
After 1992
6.2. DD and DDD analyses The fact that Figures 3-4 indicate that the changes in 1992 did not work as intended might be due to other influences that affect not only the treatment group but also other individuals. Therefore, DD and DDD analyses are conducted in which the comparison group is used to control for these other influences. Of these two approaches, the DDD framework is the most appropriate one. However, to be able to control the results and due to small size for the control group, DD analyses are also conducted.
194
First of all, the DD analysis is conducted for a sample of 60-66-year-olds in which PEW1 and PEW2 respectively are included as the treatment group, cf. Table 1 below. For each sample, before observations for 1990 and 1991 are compared to after observations for the period 1993-1994 and for 1993 and 1994, respectively. The rationale for conducting the analyses on the basis of different years is to ensure that the final conclusion does not rely solely on particular circumstances in one of these years. Information about gender, education, cohabitation status and lagged income is included in these and the subsequent analyses. The estimated coefficients on these variables are not reported in the tables but are available on request.
The DD estimates on the sample of 60-66-year-olds also indicate that the changes in the financial incentives did not work as intended. In general, the estimated coefficients for the interaction terms between PEW and the “after” observation are smaller for PEW2 than for PEW1 which might be explained by either larger comparability or a smaller sample size for PEW2. When 1994 is included as the “after” observation year, the interaction terms are even negative in most cases, but regardless of the choice of treatment group and of the before and after observations, none of these terms are significant. This indicates that the change of the PEW policy in 1992 did not increase the average retirement age for people entitled to PEW. In other words, the changes in financial incentives in 1992 seem to have been insufficient to delay retirement significantly.
195
Table 1. Difference-in-differences estimates of the effect of the changes of the PEW policy in 1992 on the retirement age, PEW vs. PEP. Pooled OLS results for 60-66-year-olds. PEW1 AFTER PEW AFTER*PEW Number of obs. PEW2 AFTER PEW AFTER*PEW Number of obs.
1990-1993/1994
1990-1993
1990-1994
1991-1993/1994
1991-1993
1991-1994
-0.663** (0.328) -1.495*** (0.285) 0.473 (0.345) 1456
-0.767* (0.392) -1.537*** (0.292) 0.626 (0.410) 1013
-0.558 (0.371) -1.426*** (0.284) 0.286 (0.391) 898
-0.073 (0.306) -0.907*** (0.256) -0.083 (0.322) 1542
-0.158 (0.375) -0.916*** (0.262) 0.060 (0.392) 1099
0.011 (0.355) -0.855*** (0.257) -0.248 (0.375) 984
-0.696** (0.335) -1.028*** (0.323) 0.365 (0.390) 526
-0.765* (0.400) -1.042*** (0.329) 0.501 (0.458) 357
-0.586 (0.371) -0.989*** (0.315) 0.114 (0.437) 337
-0.139 (0.325) -0.572* (0.304) -0.121 (0.381) 555
-0.222 (0.402) -0.565* (0.314) 0.016 (0.462) 386
-0.029 (0.373) -0.557* (0.303) -0.376 (0.444) 366
*: Significant at a 10 per cent level; **: significant at a 5 per cent level; ***: significant at a 1 per cent level. Note: Additional controls include gender, education, cohabitation status and lagged income.
As expected, the results show that individuals entitled to PEW retire earlier than individuals entitled to PEP in general. The coefficient on the “after” dummy is found to be significant negative in the analyses in which observations from 1990 is compared to 1993-1994 and 1993, respectively reflecting that the average retirement age was relatively high in general in 1990.
Nevertheless, the DD estimates might not be reliable because the average retirement age for the control group and each of the two treatment groups did not tend to move in parallel up to the policy change in 1992. Therefore, a DDD analysis is also conducted to adjust for the different trend.
The DDD estimates on the sample of 60-66-year-olds also suggest that the average retirement age did not increase as intended. Certainly, a positive 196
coefficient for the interaction term equal to the product of the dummies for year, group and adjustment for different trend is found for PEW1, while the coefficient for PEW2 in fact turns out to be negative, cf. Table 2 below. However, despite the inclusion of the adjustment variable, the effect on the average retirement age is insignificant for PEW1 as well as PEW2. Again, the retirement age is found to be lower in for individuals entitled to PEW as expected.
Table 2. Difference-in-difference-in-differences estimates of the effect of the changes of the PEW policy in 1992 on the retirement age, PEW1 vs. PEP, PEW2 vs. PEP. Pooled OLS results for 60-66-year-olds based on 1988, 1989, 1991 and 1993. PEW AFTER PEW*AFTER ADJUST PEW*ADJUST AFTER*ADJUST PEW*AFTER*ADJUST Number of observations
PEW1 -1.433*** (0.282) -0.126 (0.357) 0.065 (0.381) -0.580 (0.360) 0.481 (0.380) -0.052 (0.518) 0.014 (0.548) 1956
PEW2 -1.098*** (0.327) -0.258 (0.372) 0.097 (0.450) -0.573 (0.373) 0.521 (0.441) -0.028 (0.539) -0.018 (0.634) 716
*: Significant at a 10 per cent level; **: significant at a 5 per cent level; ***: significant at a 1 per cent level. For notes, see Table 2.
In general, the results suggest that the change of the PEW policy in 1992 did not increase the average retirement age significantly as intended. However, the effect of the change might have been that retirement in fact was delayed from age 60 to 62 until the age of 63, while retirement at age 65 to 66 at the same time were hastened. This might explain that the effect was zero on average. To control for this, separate analyses for each of the ages
197
60 to 66 years are conducted. The focus is on individuals that were in the labour market at age x-1.
Contrary to the previous results, these DD analyses suggest that the change of the PEW program in 1992 did significantly affect the retirement behaviour of the treated. In fact, an increase in retirement at age 64 among individuals entitled to PEW that were in the labour market at the age of 63 is indicated for PEW1 as well as PEW2, cf. Table 3 below. This result suggests that the fact that the PEW benefits increased on average when retirement occurred at age 64 was more important than the fact that people that delayed retirement from age 63 to age 64 were not compensated for receiving benefits for a shorter period. Further, results for PEW2 suggest that the policy change also had unexpectedly positive effects on retirement at age 66. One possible explanation however might be that this effect is driven by the different trend for PEP and PEW2. Unsurprisingly, people entitled to PEW are found to retire to a greater extent at age 60 to 63 than people eligible for PEP.
198
Table 3. Difference-in-differences estimates of the effect of the changes of the PEW policy in 1992 on retirement at age t conditioned on being in the labour market at age t-1, PEW1 vs. PEP, PEW2 vs. PEP. Pooled OLS results. PEW1 Retirement at age 60 AFTER PEW AFTER*PEW Number of obs. Retirement at age 61 AFTER PEW AFTER*PEW Number of obs. Retirement at age 62 AFTER PEW AFTER*PEW Number of obs. Retirement at age 63 AFTER PEW AFTER*PEW Number of obs. Retirement at age 64 AFTER PEW AFTER*PEW Number of obs. Retirement at age 65 AFTER PEW AFTER*PEW Number of obs. Retirement at age 66 AFTER PEW AFTER*PEW Number of obs. (to be continued)
1990-1993/1994
1990-1993
1990-1994
1991-1993/1994
1991-1993
1991-1994
0.112** (0.066) 0.270*** (0.057) -0.083 (0.071) 1677
0.102 (0.075) 0.264*** (0.058) -0.032 (0.080) 1111
0.121 (0.079) 0.279*** (0.057) -0.132 (0.084) 1106
-0.067 (0.072) 0.1021 (0.065) 0.069 (0.077) 1687
-0.080 (0.081) 0.082 (0.065) 0.127 (0.086) 1121
-0.053 (0.085) 0.109* (0.065) 0.013 (0.090) 1116
0.045 (0.067) 0.127*** (0.062) 0.033 (0.074) 1054
0.027 (0.078) 0.104* (0.063) 0.110 (0.086) 705
0.051 (0.074) 0.142** (0.061) -0.033 (0.082) 690
0.030 (0.066) 0.187** (0.059) -0.033 (0.073) 1087
0.019 (0.079) 0.169** (0.060) 0.038 (0.087) 738
0.037 (0.075) 0.193*** (0.059) -0.098 (0.083) 723
0.002 (0.061) 0.118** (0.054) 0.002 (0.069) 785
0.008 (0.075) 0.119** (0.054) -0.012 (0.084) 546
0.001 (0.072) 0.125** (0.054) 0.014 (0.082) 534
-0.032 (0.068) 0.178** (0.062) -0.063 (0.076) 753
-0.017 (0.081) 0.183** (0.064) -0.083 (0.091) 514
-0.042 (0.080) 0.176** (0.065) -0.043 (0.091) 502
0.027 (0.072) 0.173** (0.065) 0.021 (0.082) 667
0.004 (0.082) 0.161** (0.065) 0.085 (0.094) 467
0.048 (0.087) 0.195** (0.064) -0.047 (0.099) 457
-0.012 (0.072) 0.150** (0.065) 0.056 (0.084) 643
-0.027 (0.083) 0.149** (0.066) 0.115 (0.096) 443
0.000 (0.088) 0.163** (0.065) -0.001 (0.100) 433
-0.073 (0.083) 0.012 (0.079) 0.177* (0.098) 446
-0.109 (0.096) 0.003 (0.080) 0.257** (0.115) 303
-0.042 (0.094) 0.022 (0.080) 0.106 (0.113) 300
-0.016 (0.076) 0.043 (0.071) 0.147 (0.090) 482
-0.043 (0.089) 0.043 (0.071) 0.211** (0.105) 339
0.006 (0.087) 0.048 (0.071) 0.084 (0.104) 336
-0.044 (0.081) 0.033 (0.079) 0.064 (0.096) 374
-0.075 (0.094) 0.029 (0.080) 0.111 (0.111) 273
-0.013 (0.096) 0.031 (0.081) 0.006 (0.116) 247
-0.001 (0.088) 0.134 (0.087) -0.041 (0.105) 343
-0.029 (0.099) 0.131 (0.088) 0.007 (0.119) 242
0.035 (0.104) 0.131 (0.089) -0.103 (0.128) 216
-0.066 (0.088) 0.066 (0.089) 0.091 (0.108) 301
-0.087 (0.102) 0.032 (0.090) 0.181 (0.126) 205
-0.057 (0.100) 0.082 (0.087) 0.002 (0.122) 200
-0.132 (0.095) 0.067 (0.095) 0.111 (0.114) 306
-0.157 (0.112) 0.056 (0.100) 0.212 (0.135) 210
-0.091 (0.108) 0.101 (0.095) -0.011 (0.130) 205
199
(continued) PEW2 Retirement at age 60 AFTER PEW AFTER*PEW Number of obs. Retirement at age 61 AFTER PEW AFTER*PEW Number of obs. Retirement at age 62 AFTER PEW AFTER*PEW Number of obs. Retirement at age 63 AFTER PEW AFTER*PEW Number of obs. Retirement at age 64 AFTER PEW AFTER*PEW Number of obs. Retirement at age 65 AFTER PEW AFTER*PEW Number of obs. Retirement at age 66 AFTER PEW AFTER*PEW Number of obs.
1990-1993/1994
1990-1993
1990-1994
1991-1993/1994
1991-1993
1991-1994
0.081 (0.059) 0.178*** (0.059) -0.012 (0.073) 639
0.068 (0.065) 0.177*** (0.058) 0.005 (0.081) 431
0.094 (0.070) 0.179*** (0.058) -0.030 (0.085) 421
-0.075 (0.068) 0.044 (0.068) 0.119 (0.080) 627
-0.092 (0.076) 0.033 (0.068) 0.144 (0.091) 419
-0.059 (0.080) 0.048 (0.069) 0.101 (0.095) 409
0.043 (0.062) 0.127* (0.067) 0.021 (0.079) 491
0.028 (0.071) 0.110* (0.067) 0.068 (0.090) 333
0.051 (0.069) 0.138** (0.066) -0.026 (0.089) 312
0.013 (0.061) 0.172** (0.064) -0.026 (0.079) 509
-0.005 (0.072) 0.168*** (0.065) 0.019 (0.091) 351
0.032 (0.070) 0.170*** (0.064) -0.068 (0.091) 330
-0.004 (0.055) 0.071 (0.057) 0.009 (0.072) 408
-0.020 (0.070) 0.056 (0.060) 0.076 (0.090) 276
-0.000 (0.062) 0.080 (0.055) -0.042 (0.080) 294
-0.023 (0.061) 0.120* (0.066) -0.049 (0.081) 373
-0.017 (0.076) 0.114 (0.071) 0.003 (0.100) 241
-0.026 (0.069) 0.119* (0.067) -0.092 (0.091) 259
0.022 (0.067) 0.156** (0.070) 0.017 (0.089) 347
-0.003 (0.074) 0.161** (0.068) -0.025 (0.101) 254
0.048 (0.085) 0.167** (0.072) 0.053 (0.112) 240
-0.024 (0.068) 0.123 (0.071) 0.068 (0.091) 338
-0.041 (0.075) 0.135* (0.070) 0.022 (0.103) 245
-0.013 (0.085) 0.118 (0.073) 0.128 (0.113) 231
-0.088 (0.078) -0.020 (0.086) 0.186* (0.109) 262
-0.137 (0.090) -0.027 (0.085) 0.261** (0.128) 176
-0.056 (0.092) -0.012 (0.089) 0.133 (0.130) 182
-0.014 (0.071) 0.033 (0.079) 0.103 (0.100) 281
-0.038 (0.084) 0.037 (0.080) 0.134 (0.120) 195
0.005 (0.084) 0.020 (0.081) 0.070 (0.120) 201
-0.058 (0.075) 0.099 (0.087) -0.117 (0.106) 225
-0.106 (0.089) 0.083 (0.090) -0.064 (0.124) 162
-0.006 (0.094) 0.097 (0.094) -0.184 (0.138) 146
-0.008 (0.079) 0.127 (0.089) -0.131 (0.111) 215
-0.046 (0.091) 0.120 (0.091) -0.080 (0.127) 152
0.029 (0.099) 0.126 (0.095) -0.180 (0.144) 136
-0.057 (0.081) -0.107 (0.111) 0.282** (0.129) 187
-0.090 (0.095) -0.128 (0.111) 0.450*** (0.149) 121
-0.044 (0.086) -0.084 (0.101) 0.130 (0.132) 120
-0.130 (0.091) -0.005 (0.109) 0.175 (0.129) 193
-0.165 (0.108) -0.016 (0.115) 0.326** (0.155) 127
-0.082 (0.101) 0.031 (0.107) 0.005 (0.145) 126
*: Significant at a 10 per cent level; **: significant at a 5 per cent level; ***: significant at a 1 per cent level. For notes, see Table 2.
200
Table 4. Difference-in-difference-in-differences estimates of the effect of the changes of the PEW policy in 1992 on retirement at age t conditioned on being in the labour market at age t-1, PEW1 vs. PEP, PEW2 vs. PEP. Pooled OLS results based on 1988, 1989, 1991 and 1993. PEW1 PEW AFTER PEW*AFTER ADJUST PEW*ADJUST AFTER*ADJUST PEW*AFTER*ADJUST Number of observations PEW2 PEW AFTER PEW*AFTER ADJUST PEW*ADJUST AFTER*ADJUST PEW*AFTER*ADJUST Number of observations
60
61
62
63
64
65
66
0.157*** (0.053) 0.004 (0.072) 0.011 (0.079) 0.113 (0.076) -0.055 (0.081) -0.083 (0.107) 0.114 (0.115) 2121
0.137** (0.054) 0.012 (0.068) -0.023 (0.076) 0.059 (0.069) 0.032 (0.077) 0.006 (0.100) 0.060 (0.111) 1512
0.204*** (0.061) 0.046 (0.072) -0.186** (0.081) 0.048 (0.076) -0.033 (0.086) -0.073 (0.109) 0.109 (0.122) 1097
0.135** (0.064) 0.063 (0.080) -0.148 (0.093) 0.020 (0.077) 0.011 (0.089) -0.086 (0.113) 0.260** (0.131) 880
0.095 (0.066) 0.094 (0.080) -0.115 (0.095) 0.065 (0.079) -0.055 (0.093) -0.134 (0.117) 0.316** (0.138) 708
0.122 (0.077) 0.130 (0.089) -0.155 (0.107) -0.004 (0.099) -0.008 (0.117) -0.169 (0.134) 0.178 (0.161) 545
-0.061 (0.085) -0.161* (0.095) 0.185 (0.119) 0.021 (0.101) 0.097 (0.124) 0.006 (0.142) 0.019 (0.174) 428
0.071 (0.056) -0.005 (0.066) 0.066 (0.082) 0.103 (0.068) -0.019 (0.083) -0.082 (0.096) 0.072 (0.118) 839
0.070 (0.055) 0.021 (0.061) 0.029 (0.077) 0.074 (0.061) 0.096 (0.079) -0.020 (0.089) -0.014 (0.113) 745
0.175*** (0.066) 0.045 (0.068) -0.138 (0.088) 0.047 (0.071) -0.066 (0.095) -0.071 (0.103) 0.153 (0.134) 562
0.189*** (0.069) 0.068 (0.074) -0.243** (0.100) 0.022 (0.071) -0.040 (0.095) -0.091 (0.104) 0.270* (0.141) 491
0.126 (0.081) 0.094 (0.076) -0.183 (0.112) 0.063 (0.075) -0.077 (0.109) -0.129 (0.111) 0.307* (0.159) 394
0.035 (0.085) 0.124 (0.083) 0.016 (0.128) -0.011 (0.092) 0.078 (0.128) -0.179 (0.125) -0.073 (0.184) 323
-0.080 (0.104) -0.167* (0.091) 0.180 (0.140) 0.015 (0.097) 0.050 (0.148) 0.001 (0.136) 0.157 (0.201) 267
*: Significant at a 10 per cent level; **: significant at a 5 per cent level; ***: significant at a 1 per cent level. For notes, see Table 2.
The DDD analysis is also conducted for each of the age groups, cf. Table 4 above. First of all, these results also suggest that the change of the PEW policy increased the incentives to retire at age 64 as intended. Correspondingly, the incentives to retire at age 63 seem to have increased as well. The results for PEW2 for retirement at age 63 and 64 are less significant than for PEW1 (10% compared to 5% level). Again, the difference might be explained by either larger comparability or a smaller sample size for PEW2. The results indicate that the changes of the financial incentives to retire at 201
age 63 and 64 have been sufficient to change the retirement behaviour in the intended direction for individuals in the treatment group around these ages.
In contrast, transition to retirement at age 60-62 and age 65-66 seems not to have been affected significantly. That is, the significant effect that was found in the DD analysis for retirement at age 66 is not confirmed in this analysis. The results of the DDD analysis moderate the conclusion that the change of the PEW policy did not have any effect. However, the results suggest that the aggregate effect on the retirement age was rather limited.
7. Concluding remarks The focus in this paper is on the PEW program, which is the most popular early retirement scheme in Denmark. The purpose is to examine the effect of a change of the PEW policy in 1992 on the retirement age. To reduce early retirement through PEW, the scheme was changed to increase the incentives to delay retirement until the age of 63.
All in all, results suggest that the response of retirement age to the change in the PEW policy was relatively small at least in the short run. Distributions of retirement age and retirement age hazard functions indicate that the changes did not have the intended effect. However, DDD analyses suggest that the change of the PEW policy in fact increased the incentives to retire at age 63 and 64 as intended. But the analyses also suggest that the aggregate effect was small in the sense that the effect on transition to retirement
202
at age 60-62 and at age 65-66 was not affected significantly. Consequently, the retirement age for people entitled to PEW did not increase on average. However, it cannot be ruled out that in the longer run, a larger effect might have arisen.
The results are similar to what has been found in previous international studies, which confirms that rather large policy changes are required to elicit substantial changes in retirement ages. The changes of financial incentives in 1992 seem to have been too small for the purposes of significantly delaying retirement at least in the short run. Part of the problem might be that there was no stick, only carrots. People entitled to PEW got a reward if they delayed retirement until age 63. But if they retired at age 60 to 62, the PEW benefits were exactly the same as before 1992. Moreover, the carrots were rather small. If retirement was delayed until the age of 63, the change first set in 2½ years after the transition to retirement. And, in fact, PEW benefits did not increase in absolute terms after 2½ years. Instead, the reduction in benefits after 2½ years, which was in force for everyone receiving PEW benefits before 1992, was removed for this group. Therefore, the changes were hardly noticeable in the individual case.
The results found in this paper deviate from what has been found in previous Danish studies. Presumably, the most important explanation of this difference is that the changes of the PEW program in 1992 were too small to affect the average retirement age. In fact, these changes were much less farreaching than the changes in focus in the previous studies. In part, the deviation could be due to the fact that unlike previous Danish studies, this paper employs an experimental approach to study retirement behaviour and 203
thereby the resulting estimates can be thought of as causal responses of retirement to policy changes.
Contrary to the policy aim, the changes of the PEW policy in 1992 might have caused the public expenditures to increase as a result of the “no stick, only carrots” construction of the policy change. Namely, if the effect on retirement behaviour as the results of this paper suggest was limited. If so, the change in 1992 increased the burden on the providers instead of diminishing it.
8. References Angrist, J. D. and A. B. Krueger (1999), Empirical Strategies in Labor Economics, in: Ashenfelter, O. and D. Card (eds), Handbook of Labor Economics, vol. 3A, Amsterdam, 1277-1366. Baker, M. and D. Benjamin (1999), Early Retirement Provisions and the Labor Force Behavior of Older Men: Evidence from Canada, Journal of Labor Economics, vol. 17, no 4, pt. 1, 724-756. Bingley, P., N. Datta Gupta and P.J. Pedersen (2003). The Impact of Incentives on Retirement in Denmark, in Social Security and Retirement Around the World: Microestimation, (eds.) Wise, J. and Gruber, D., NBER, 2003. Blöndal, S. and S. Scarpetta (1998), The Retirement Decision in OECD Countries, Aging Working Papers 1.4, Paris. Blundell, R. and M. Costa Dias (2000) Evaluation Methods for Non-Experimental Data, Fiscal Studies, vol. 21, no. 4, pp. 427-468. Börsch-Supan, A. (2000), Incentive Effects of Social Security on Labour Force Participation: Evidence in Germany and Across Europe, Journal of Public Economics 78, 25-49. Christensen, B.J. and N. Datta Gupta (2000), The Effect of a Pension Reform on the Retirement of Danish Married Couples (in Danish), Nationaløkonomisk Tidskrift 138, 222-242. Danø, A.M., M. Ejrnæs,and L. Husted (2000), How is the Retirement Age Affected by the Reform of the Post Employment Wage programme? (in Danish), Nationaløkonomisk Tidskrift 138, 205-221.
204
Department of Unemployment Insurance (2001), Statistics of Post Employment Wage and Transitional Benefit Programme first half of 2001 (in Danish), digital document: www.adir.dk Gruber, J. (1994), The Incidence of Mandated Maternity Benefits, The American Economic Review, vol. 84, issue 3 (June), 622-641. Gruber, J. and D. Wise, eds. (1999), Social Security and Retirement around the World, Chicago. Hansen, H. (1999), Elements of Social Security, A comparison covering: Denmark, Sweden, Finland, Austria, Germany, The Netherlands, Great Britain and Canada, The Danish National Institute of Social Research 99:14, Copenhagen. Krueger, A.B. and J.-S. Pischke (1992), The Effect of Social Security on Labor Supply: A Cohort Analysis of the Notch Generation, Journal of Labor Economics, vol. 10, no 4, 412-437. Mitchell, O.S. and G.S. Fields (1984), The Economics of Retirement Behavior, Journal of Labor Economics, vol. 2, no 1, 84-105. Nelissen, J.H.M. (2002), Early Retirement: The Impact of Changes in the Benefit Level, Tinbergen Institute Discussion Paper TI 2002-031/ 3, Rotterdam. OECD (2003), OECD Economic Surveys, Denmark, Vol. 2003/10 – July, Paris. Quaade, T. (2002), Reform of the Post Employment Wage Programme with Limited Effect (in Danish), in: Socialforskningsinstituttet, Social forskning 2002:3, Copenhagen, 10-11. Schaumann, A. (2001), The Aging Society. Demography – Expenditure Pressure – What can be done? (in Danish), digital document: www.tekno.dk Scherer, P. (2001), Withdrawal from the Labour Force in OECD Countries, OECD Occasional Paper, no 49, Paris. “Sortbog” (1991), (Law No 29 of 27 December 1991) Law on Change of the Law on Employment Service and Unemployment Insurance etc. (Post Employment Wage). Bill no L29 (in Danish), Copenhagen.
205
9. Appendix
Table A.1. Descriptive statistics for employed and retired 60-66-year-olds that were in the labour market at the age of 59, PEW1, PEW2 and PEP, 1987-1998. Means and standard errors (in parentheses). PEW1 Women Vocational training Higher education Single Lagged income (log) Number of observations
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
0.43 (0.49) 0.31 (0.46) 0.08 (0.28) 0.26 (0.44) 7.05 (5.49)
0.43 (0.50) 0.32 (0.47) 0.08 (0.28) 0.25 (0.44) 6.89 (5.57)
0.44 (0.50) 0.33 (0.47) 0.08 (0.27) 0.26 (0.44) 6.73 (5.66)
0.45 (0.50) 0.34 (0.47) 0.09 (0.28) 0.26 (0.44) 6.84 (5.66)
0.45 (0.50) 0.36 (0.48) 0.09 (0.29) 0.26 (0.44) 7.01 (5.64)
0.46 (0.50) 0.36 (0.48) 0.10 (0.30) 0.25 (0.43) 6.78 (5.69)
0.46 (0.50) 0.38 (0.48) 0.10 (0.30) 0.25 (0.43) 6.72 (5.73)
0.46 (0.50) 0.38 (0.49) 0.11 (0.31) 0.25 (0.44) 6.59 (5.75)
0.47 (0.50) 0.39 (0.49) 0.12 (0.32) 0.26 (0.44) 6.37 (5.77)
0.48 (0.50) 0.39 (0.49) 0.13 (0.34) 0.26 (0.44) 6.29 (5.78)
0.48 (0.50) 0.39 (0.49) 0.14 (0.34) 0.26 (0.44) 6.42 (5.79)
0.48 (0.50) 0.40 (0.49) 0.15 (0.35) 0.25 (0.44) 6.55 (5.85)
3068
3054
3063
3062
3077
3059
3132
3179
3270
3321
3353
3385
0.62 (0.48) 0.26 (0.44) 0.13 (0.34) 0.31 (0.46) 7.64 (5.34)
0.63 (0.48) 0.27 (0.45) 0.14 (0.35) 0.30 (0.46) 7.60 (5.42)
0.64 (0.48) 0.28 (0.45) 0.14 (0.35) 0.31 (0.46) 7.42 (5.57)
0.65 (0.48) 0.28 (0.45) 0.16 (0.37) 0.31 (0.46) 7.54 (5.58)
0.64 (0.48) 0.29 (0.45) 0.17 (0.37) 0.30 (0.46) 7.97 (5.46)
0.64 (0.48) 0.30 (0.46) 0.18 (0.38) 0.29 (0.45) 7.92 (5.46)
0.65 (0.48) 0.32 (0.47) 0.19 (0.39) 0.29 (0.45) 7.89 (5.49)
0.66 (0.48) 0.32 (0.47) 0.21 (0.41) 0.29 (0.45) 7.77 (5.55)
0.67 (0.47) 0.32 (0.47) 0.22 (0.47) 0.32 (0.47) 7.87 (5.51)
0.68 (0.47) 0.33 (0.47) 0.22 (0.42) 0.31 (0.46) 7.74 (5.55)
0.68 (0.47) 0.33 (0.47) 0.23 (0.42) 0.30 (0.46) 7.63 (5.60)
0.70 (0.46) 0.34 (0.47) 0.24 (0.43) 0.30 (0.46) 7.69 (5.64)
1071
1047
1034
1007
994
979
973
961
1015
1067
1118
1165
0.46 (0.50) 0.20 (0.40) 0.33 (0.47) 0.23 (0.42) 9.50 (4.51)
0.43 (0.49) 0.22 (0.41) 0.33 (0.47) 0.22 (0.41) 9.70 (4.49)
0.42 (0.49) 0.22 (0.41) 0.35 (0.48) 0.22 (0.41) 9.70 (4.54)
0.40 (0.49) 0.23 (0.42) 0.35 (0.48) 0.19 (0.39) 9.46 (4.74)
0.40 (0.49) 0.23 (0.42) 0.38 (0.49) 0.20 (0.40) 9.55 (4.72)
0.41 (0.49) 0.24 (0.43) 0.39 (0.49) 0.22 (0.41) 9.36 (4.89)
0.41 (0.49) 0.22 (0.42) 0.41 (0.49) 0.21 (0.41) 9.19 (5.00)
0.41 (0.49) 0.20 (0.40) 0.45 (0.50) 0.22 (0.42) 9.14 (4.99)
0.43 (0.50) 0.19 (0.39) 0.47 (0.50) 0.21 (0.41) 9.16 (5.01)
0.44 (0.50) 0.20 (0.40) 0.49 (0.50) 0.22 (0.41) 9.28 (4.88)
0.46 (0.50) 0.20 (0.40) 0.51 (0.50) 0.24 (0.42) 9.46 (4.78)
0.46 (0.50) 0.20 (0.40) 0.52 (0.50) 0.25 (0.43) 9.29 (4.94)
567
555
530
524
500
493
499
471
439
435
417
418
PEW2 Women Vocational training Higher education Single Lagged income (log) Number of observations
PEP Women Vocational training Higher education Single Lagged income (log) Number of observations
206
Table A.2. Descriptive statistics for 60-66-year-olds that were in the labour market at the age of 59 and retired the year in question, PEW1, PEW2 and PEP, 1990 and 1991. Means and standard errors (in parentheses).
Retirement age Women Vocational training Higher education Single Lagged income (log) Number of obs.
PEW1 61.69 (1.82) 0.45 (0.50) 0.33 (0.47) 0.08 (0.28) 0.22 (0.42) 10.66 (3.42) 411
1990 PEW2 62.02 (1.76) 0.69 (0.46) 0.23 (0.42) 0.18 (0.38) 0.28 (0.45) 11.58 (1.88) 123
PEP 63.39 (1.77) 0.41 (0.50) 0.20 (0.41) 0.43 (0.50) 0.25 (0.44) 11.88 (0.74) 44
PEW1 61.63 (1.80) 0.49 (0.50) 0.38 (0.49) 0.08 (0.27) 0.22 (0.42) 10.17 (4.00) 485
1991 PEW2 62.04 (1.85) 0.62 (0.49) 0.32 (0.47) 0.14 (0.34) 0.23 (0.42) 11.60 (2.05) 140
PEP 62.77 (2.07) 0.43 (0.50) 0.18 (0.39) 0.39 (0.49) 0.29 (0.46) 11.77 (1.09) 56
Figure A.1. Average retirement age for 60-66-year-olds, PEW1, PEW2 and PEP, 1987-1991, 1993-1994. Average retirement age 64 63 63 62 62 61 61 1987
1988
1989 PEW1
1990
1991
PEW2
1993
1994
PEP
207
208
Chapter 5 Pathways to Retirement in Denmark, 1984 – 2000†
Mona Larsen* and Peder J. Pedersen**
JEL Codes: J14, J26 Abstract Not much is known about the pathways leading to either normal Old Age Pension (OAP, folkepension) or to exit from the labour force into a program for early retirement. This paper describes and analyses the multitude of pathways to retirement in Denmark since the middle of the 1980’s. The analyses are based on a 10 per cent panel sample of the Danish population 45-67 years old followed from 1984 onwards. A multinomial logit approach is applied in order to analyse the characteristics of individuals that retire through each pathway compared to individuals that remain in the labour force. We find that the transition from work to retirement is complex and far from the conventional idea of exit typically occurring from a job at the official pension age. Eight pathways from work to normal OAP or to an early retirement program are identified, which seem to fall into three groups. One group is transitions directly from employment that covers 75 per cent of all transitions in the sample period. Another group is the pathways dominated by UIB (unemployment insurance benefits) covering 20 per cent. The remaining 5 per cent are pathways dominated by benefits reflecting a low attachment to the labour force in the period prior to transition to retirement. The relative magnitude of these aggregated pathways is affected by the cyclical profiles over time and the temporary opening of the Transitional Benefit Program (TBP, overgangsydelse) in the middle of 1990’s. Across all three pathways, the Post Employment Wage (PEW, efterløn) destination becomes increasingly important over time, while a very low incidence of entry directly to OAP is observed. Overall, availability and/or generosity of retirement programs seem to be very important for retirement through the employment and UIB dominated pathways, while individual background factors are of minor importance. For retirement through other pathways, however, personal characteristics seem to be at least as important as retirement programs.
†
This work is part of the research of the Graduate School for Integration, Production and Welfare. Financial support from the Danish Social Science Research Council is gratefully acknowledged. We are grateful for comments from Michael Rosholm, Peter Jensen and other participants at the Seminar on Welfare Research in december 2003 in Nyborg.
*
Danish National Institute of Social Research; Aarhus School of Business and Graduate School for Integration, Production and Welfare,
[email protected].
**
Department of Economics, University of Aarhus and Danish National Institute of Social Research,
[email protected].
209
1. Introduction Traditionally, work and retirement has been considered as distinct states, reflecting the idea of a life cycle where full-time employment in a job is followed directly at a specific and common age by entry into National Old Age Pension (OAP, folkepension). This view is still true for some people in the labour force. However, increasingly the transition from work to OAP can consist of many combinations of programs for early retirement, unemployment insurance benefits, sickness benefits and social welfare benefits. Further, the period between full-time work and retirement at the official pension age can be bridged by part-time work or by early activation of private pension programs at an actuarial discount. OECD (2003) presents evidence for the member countries showing that withdrawal from the labour market starts well before the official retirement age. The range of labour market participation rates across the countries is quite big for the age group 50 to 64 years old in 2000, i.e. with a range between 40.2 per cent in Belgium and 89.9 per cent in Iceland.
Kohli and Rein (1991) define a pathway as an institutional arrangement or a sequential combination of a number of arrangements making a bridge between the exit from work and entry into the normal OAP. In the present paper we use a broader concept and focus on pathways leading to either normal OAP or to exit from the labour force into a program for early retirement. We look at six different final destinations outside the labour force and identify eight different pathways to these destinations. That is, up to 48 combinations are possible. In addition, we have a long observation period of 17 years. Furthermore, the pathways differ between some, which are choice based and others for which people are only eligible after having 210
passed a visitation based on medical and/or social criteria. An optimisation model for the selection to pathways from work is thus a highly complex undertaking, which is one reason that this is a descriptive paper. Another reason is that not much is known about the relative magnitude of these pathways. Therefore, the purpose of this paper is to describe the multitude of pathways to exit from the labour force in an institutional setting with a very broad range of available programs. This type of knowledge is important in research and policy discussions about reforms of welfare and retirement programs in a setting with big demographic changes in the future age composition of the population in the rich OECD countries. Furthermore policies to promote higher participation and employment rates among the seniors in the labour force also depend on more solid evidence about the withdrawal process from the labour force.
In Section 2 we describe briefly the retirement programs and the pathways from work to these programs in Denmark. Next, Section 3 draws on register based data for the period 1984–2000 to describe the actual distribution on a number of pathways from work to normal OAP or to an early retirement program for people who retire during this period. Section 4 describes the data set, being a 10 per cent panel sample of the Danish population 18– 67 years old followed during the period from 1984 and on. Further, Section 4 introduces the empirical model, which is a multinomial logit estimation of the allocation of individuals in the sample to different pathways, including the outcome of remaining in the labour force the year in question as the group of reference. In Section 5 we present the results from the estimations and Section 6 concludes the paper.
211
2. Pathways and retirement programs In this section, the social security programs constituting the pathways and the retirement programs applied in the paper are described.
2.1. Pathways Three types of social security benefits are taken into consideration when identifying pathways, namely unemployment insurance benefits, social welfare benefits and sickness benefits.76 These types of benefits are described briefly in the following.
(i) Unemployment Insurance Benefits (UIB) Unemployment insurance is voluntary in Denmark. Both employees and self-employed can be insured. Since the middle of the 1980’s, the rules of eligibility have been tightened up, as it has become more difficult to regain eligibility for UIB once it has been lost and the maximum duration of the benefit period has been reduced.
Eligibility to UIB is conditional on at least 1 years membership of an unemployment insurance fund, on a minimum period of former work (52 weeks within 3 years in 2000), and on being available for the labour market.
76
For further information about these programs, see e.g. Lausten (2001), Hansen (2000) and Høgelund (2003).
212
UIB benefits corresponds to 90 per cent of former earnings subject to a ceiling, which differs depending on whether unemployment insurance is on a full-time or a part-time basis. In 2000, the maximum benefit period is 4 years in general. After having received UIB for 1 year, the unemployed have a claim as well as an obligation to enter an activation program such as job training or education. However, the maximum benefit period is longer if aged 55-59 years, but shorter if aged 60 to 66. In our analysis, periods with UIB include activation periods.
(ii) Social Welfare Benefits (SWB) Recipients of SWB consist of two groups: a) uninsured unemployed and insured unemployed who have not yet gained or have lost the right to UIB; b) individuals, who are not registered as unemployed but receive SWB due to social problems. In principle, the duration of SWB is unlimited. However, different arrangements and schemes such as activation and Social Disability Pension, cf. below, usually contribute to move the recipients out of the SWB system again. Benefits amounts to between 60 and 80 per cent of the ceiling for UIB, cf. above. Periods with SWB include activation periods in our analysis.
(iii) Sickness Benefits (SB) SB are paid in connection with being unable to work due to illness, injury, childbirth or adoption. The SB scheme was changed several times during the 1980’s. Both employees and self-employed are eligible for SB. Eligibility to SB is conditional on a certain employment record (in 2000, 74 hours of employment within the preceding 8 weeks). SB correspond to
213
earnings subject to the ceiling for UIB, cf. above. However, it is usually for the employer to supplement the public benefit. In fact, white collar workers receive wages, while almost all blue collar workers have received full wages in the first 2 weeks since 1994. The maximum benefit period is 1 year within 36 months.
We identify eight different pathways out of the labour force, either to an early retirement program or to OAP. One possibility is obviously to retire from a job as a wage earner or from being self-employed in one’s own business. In the econometric analyses in Section 5 these two pathways are aggregated to one, i.e. retirement from employment. Next, we identify two pathways used by people out of employment but with a close labour force attachment who are provided for by UIB and to a lesser extent by SB up to the time of retirement. These two pathways are aggregated to one, called ”UIB dominated pathways”, in the econometric analyses below.
Another pathway is defined by SWB being the dominant income up to the time of retirement. Further, two pathways are defined by combinations of SWB, SB and UIB as the main sources of income until retirement. Finally, we identify a residual pathway. These four individual pathways are aggregated to one, called ”other pathways”, in the econometric analyses below.
214
2.2. Retirement programs The structure in the Danish pension and retirement system consists of a number of public sector programs and private arrangements. In this section, the destinations outside the labour force applied in the paper are described.
(1) Old age pension (OAP) Eligibility to OAP is dependent on age and duration of residence in the country. The eligibility age is 67 for the period we analyse (65 from 2004). OAP consist of a base amount, which is means tested against earnings from work and a supplementary pension, which is means tested against all other income. As only few people work beyond the age of 67 this implies that the base amount is nearly universal. A recent study (Arendt et al., 2003) found that about 25 per cent of retired people had no income besides the OAP. Consequently, 75 per cent of retired people either get no or reduced supplementary pension from the OAP.
The other major exit routes from the labour market are the many programs for early retirement. It should be emphasized that some of these can be chosen freely if objective conditions for eligibility are fulfilled while others are open only after visitation on either medical or social criteria. The major public sector financed programs for early retirement consists of Social Disability Pension (SDP, Førtidspension) and pensions to certain groups of public employees in permanent positions, the Public Employees' Pension scheme (PEP, Tjenestemandspension), which has an early retirement option from age 60. Further, there is a labour market related program open for people 60 - 66(64) years old, called the Post Employment Wage (PEW, 215
Efterløn). Private sector arrangements cover a broad range from mature pension funds, over funds in the blue collar part of the labour market still in a build-up phase, and to fully individual arrangements of which most are tax subsidized. Common for these arrangements is an early retirement option beginning from age 60 based on actuarial principles. We describe the different programs briefly in the following77.
(2) Social Disability Pension (SDP, Førtidspension) The main principles behind the rules for SDP were enacted through a major reform in 1984 of the public sector programs regarding early retirement. The SDP was intended to replace a number of earlier programs. The biggest among these was disability pension, which could be granted on three levels according to health criteria. Other programs, which were included into the SDP, were a public financed program for widows’ pension, and a program for early OAP also for persons whose (older) spouses were receiving the OAP already. In 2003, a new reform with a big simplification of the benefit structure was enacted. For further description of this reform, see e.g. Høgelund (2003).
In principle, granting of SDP depends on an application being decided upon relative to a set of medical and social criteria. SDP is thus not an individual option like eligibility for e.g. labour market pension from a specific age. Until 2003 the SDP system was quite complex as the rules differed regarding tax treatment and regarding the means testing or not of the
77
A discussion of the ongoing policy reforms in this area can be found in OECD (2000).
216
different components and amounts that made up the program. In this period SDP could be granted on three different levels. The highest level was applicable to persons younger than 60 whose work capacity had become (or always had been) reduced to almost nothing. The intermediate level SDP was open for those younger than 60 with a work capacity reduced to onethird of the normal level and to people 60 to 66 years old with almost no remaining work capacity. Eligibility for the highest and the intermediate levels SDP was decided on medical criteria. Finally, eligibility for the lowest level, so-called, ordinary level SDP depended on work capacity having been reduced to below half the normal level based on health criteria or on a combination of health and social criteria. Recipients of the ordinary level SDP younger than 60 were entitled to a supplementary amount. From 2003 there are only two levels of SDP benefits for new entrants to the program with the level depending on marital status.
(3) Post Employment Wage (PEW, Efterløn) In contrast to the SDP the PEW scheme introduced in 1979 provides the possibility of early retirement from age 60 without having to fulfil any health criteria. It was intended to be a labour market policy instrument with the purpose of creating jobs for young people by advancing the retirement age for older workers with a number of years. In 1999 a fairly complicated reform of the PEW was enacted in which incentives to postpone entry until the age of 62 or later was introduced.
After the reform in 1999, the age of eligibility is still from 60. However, to become eligible for PEW an individual is required to have been in an un-
217
employment insurance fund for 25 out of the last 30 years. Further, the rules were changed to make it more attractive to continue work for a number of hours at the same time as collecting PEW. The reform has so far resulted in a decline in the take up of PEW for the 60-61 years old. (Economic Council, 2001). The initial impact on retirement is analysed using micro data by Jørgensen (2004), Bjørn and Larsen (2003), Quaade (2001, 2002) and Danø et al. (2000).
PEW can be entered both from employment and from unemployment. If a person enters directly from a job, benefits in the PEW system are equal to the amounts to which she or he would be entitled in case of unemployment. The benefits in the PEW program are not means tested against the income of other family members, but income from pension schemes from previous employers are deducted from the PEW benefits and SDP cannot be collected at the same time as PEW.
(4) Transitional Benefits Program (TBP, Overgangsydelse) This program was introduced in 1992 and entry to the program was extended in 1994. Eligible persons for entry were initially 55-59 years old members of unemployment insurance funds who had been unemployed for at least 12 out of the most recent 15 months. From the beginning of 1994 the program was extended to cover the age group 50-54 years with the same labour market criteria as for the 55-59 years old group. Benefits were set at 82 per cent of maximum unemployment insurance benefits and the maximum duration was until transition to PEW at the age of 60. Entry to the program was terminated at the beginning of 1996. As a consequence,
218
this program will no longer appear among the possible retirement states as from 2006.
Occupational pensions (5) Public Employees Pension Scheme (PEP, Tjenestemandspension) and (6) Labour Market Pensions (LMP) called occupational pensions, are old age pensions that are available with an actuarial discount at an earlier age than 67. PEP is a program covering part of the employees in the public sector. The pension is considered to be part of a lifetime wage contract. The pension is consequently not funded. The pension amount is calculated as a function of the wage depending on seniority and position and can on actuarial terms be taken up from age 60.
There is a wide and expanding coverage with LMP-programs. The building up of pension funds began some 40 years ago for fairly small highly educated groups. Coverage has since broadened and during the last decade a major part of the labour market for blue-collar workers has also been covered with pension plans. Like the PEP these programs have an early retirement option from age 60. Typically, the pension funds build on defined contributions of either 15 per cent (high wage groups) or 9 per cent (industrial workers) of the annual earnings.
(7) Private Pensions Saving Finally, there is a broad coverage with private pension plans, mostly into some broad categories of savings arrangement which until recently have
219
been treated quite favourably by the tax rules. These too, can mostly be taken up from age 60.
In principle, a big number of combinations are possible as pathways from an ordinary full-time job to normal OAP or to an early retirement program. It turns out however that a fairly small number of simple or composite pathways dominate the retirement process. In the next section we turn to a description of these pathways based on our panel data for the years since 1984. One further possibility to bridge an eventual gap between a full-time job and exit from the labour market is a shift to part-time work as a “bridge” job. Unfortunately, we are unable to identify a shift to a part-time job in our micro database78 and therefore, information about the extent of “bridge” jobs is solely obtained from earlier studies.
3. The distribution on pathways, 1984-2000 After the description of the separate pathways in Section 2 we turn to look into the actual flows through these different ways into retirement. First, as mentioned, shift to a part-time job could be one of the pathways. Ruhm (1990) shows that partial retirement is very widespread in the U.S. In the case of Denmark, many individuals, in particular self-employed, express a desire to retire partially at the end of their working lives, cf. Pedersen (1998), Nørregaard (1996) and Pedersen & Smith (1995). However, paradoxically, the extent of partial retirement through less working hours seems to be limited in Denmark. We are not able to identify shifts to part-time 78
Usable information is either not available or of bad quality.
220
work as bridge jobs to retirement in the present paper. The fairly modest use of bridge jobs is discussed in Chapter 1 based on panel survey data from 1997 and 2002. In earlier studies Nørregaard (1996) finds that only relatively few retire through part-time work, flexible working hours etc. Similarly, a survey conducted by the Danish National Institute of Social Research in 1999 shows that only 6 per cent of already retired individuals had reduced their working week gradually through part-time work. Furthermore, very few individuals have joined part-time retirement schemes, cf. Pedersen & Smith (1995). In 1998, only 4,000 retired on the part-time pension scheme while 1,600 retired on a part-time PEW scheme79, cf. Initiative Committee on Senior Policy, Ministry of Labour (1999).
Next, we go on to look more closely at the pathways from work to OAP or to an early retirement program that we can identify in our longitudinal dataset. A first summary overview is shown in Figure 1 below. The figure shows the relative distribution over the whole period 1985–2000 on the pathways through which entry can occur either to early retirement or to OAP. Entry from a job is found to be dominant covering about 65 per cent of all transitions. The two other big pathways from which entry occurs are UIB and from a working life as self-employed, which in this case includes assisting spouses. The remaining about 10 per cent of states from which entry occurs are combined states where people are being provided for by SWB alone or by UIB, SWB or SB in a composite sequence.80
79
The part-time PEW scheme was abolished 1 July 1999.
80
The abbreviation in brackets expresses the type of social security benefits from which the largest amount of benefits has been received in the period from work until retire-
221
Figure 1. Pathways to retirement, pooled observations, 1985-2000. Wage earner UIB Self-employed UIB+SB (UIB) SWB UIB+SWB (UIB) UIB+SB (SB) Others 0
10
20
30
40
50
60
70 Per cent
Figure 2. Share of people entering OAP or early retirement coming from employment, 1985-2000. Per cent 100 90 80 70 60 50 40 30 20 10 0
1985
1987
1989
1991
1993
1995
1997
1999
In Figure 2 above the focus is on the cyclical impact on the use of pathways by showing the share of people entering early retirement or OAP directly from employment, in a job or as self-employed. Figure 2 is a clear reflection of the cyclical profiles during the period with a long phase of increasing unemployment lasting from 1987 to 1993, followed by a strong cyclical upswing from 1994. On top of that, Figure 2 gives a clear impression of the
ment. For instance, the title “UIB + SB (UIB)” expresses a combination of UIB and SB where UIB is the dominant type of benefits.
222
impact from TBP to which long run unemployment was a condition for entry, with the broadest age criterion, 50 – 59 years, in the period 1994 to 1996. In fact, only half of individuals retiring in 1995 retired directly from employment compared to about 80 per cent in the last half of the 1980’s as well as around 2000.
In Figure 3 below the focus is on the profile over time in the states from which people enter early or normal retirement, conditional on entry not being from a job or from working in your own business. UIB alone and in combinations reflects to some extent the Danish business cycle especially for the period since 1994 with a big decline in unemployment. Finally, we find a jump up in the second half of the 1990s in the residual group “Others”, which is mainly due to the fact that individuals in leave schemes, which were introduced in 1994, are included in this category.
Figure 3. Relative distribution on pathways to retirement, non-employed people, 1985-2000. 100% 80% 60% 40% 20% 0% 1985 1987 1989 1991 1993 1995 1997 1999 UIB
UIB+SB (UIB)
UIB+SB (SB)
UIB+SWB (UIB)
SWB
Others
223
Next, Figure 4 below shows the average age at entry into early or normal retirement by the three aggregated pathways. For people entering from employment the average age appears to be very stable with a very small decline during the period. The average age for people with a closer labour market attachment, UIB (all), has a very special profile reflecting the natural experiment evident in the temporary opening up of an early exit possibility in form of the TBP, which implied a temporary decline in the average age of retirement of four years for this group. Finally, the average age for the aggregated pathway, SWB + others, consisting of people with a more loose attachment to the labour force is quite volatile, but without any clear trend.
Figure 4. Average age at retirement by different pathways, 1985-2000. Retirement age 62 60 58 56 54 52 50 1985
1987
1989
Employed
1991
1993
1995
UIB (all)
1997
1999
SWB+others
Figure 5 below shows gender differences in pathways to retirement pooled for the whole period. We find a more than 50 per cent share of women who retire through UIB and UIB in combination with SB. SWB alone or in combination with UIB are on the other hand very male dominated pathways. 224
Figure 5. Share of women in the pathways, pooled observations, 19852000. Others SWB UIB + SWB (UIB) UIB + SB (SB) UIB + SB (UIB) UIB Self-employed Wage earner 0
10
20
30
40
50
60
70 Per cent
In Figure 6 below we shift the focus to destination states and illustrate how people without a formal education are distributed on retirement programs during the period. The clearest trend is an increasing importance of PEW for this group and a corresponding decrease in the relative importance of the destination state Others+PEP. Further, Figure 6 illustrates very clearly the importance of TBP for people in this educational group. Finally, the “conventional” idea of direct entry to OAP is seen to be very unusual in this group completely dominated by early exit from the labour force.
225
Figure 6. Retirement destination state for people without formal education, per cent, 1985–2000. 100% 80% 60% 40% 20% 0% 1986
1988
1990
PEW
TBP
1992 SDP
1994
1996
Others+PEP
1998
2000
OAP
In Figure 7 below we link pathways and destinations by looking at how people who exit by each of the three aggregated pathways are distributed on destinations. In the top panel (a) the main trend is an increasing importance of PEW. In panel (b) PEW is even more important and we observe once again the importance of the temporary opening of the TBP which is aggregated with PEW in the figure. Finally, in panel (c) including the group with the loosest attachment to the labour force we find the same increasing importance of the aggregate PEW destination although as expected at a lower level and with dominance of SDP and Other exit destinations. We observe also the very low incidence of entry directly to OAP, even for the group who retire directly from employment.
226
Figure 7. Relative distribution on retirement destination states for people coming from a) Employment, b) UIB dominated Pathways and c) Other Pathways, per cent, 1985-2000. a) Employment
100% 80% 60% 40% 20% 0% 1985 1987 1989 1991 1993 1995 1997 1999 PEW
SDP
PEP+Others
OAP
b) UIB dominated Pathways 100% 80% 60% 40% 20% 0% 1985 1987 1989 1991 1993 1995 1997 1999 PEW
SDP
PEP+Others
OAP
c) Other Pathways 100% 80% 60% 40% 20% 0% 1985 1987 1989 1991 1993 1995 1997 1999 PEW
SDP
PEP+Others
OAP
227
Finally, we look in Table 1 below at another aspect of the connection between pathways and destinations focusing on the fact that most of the pathways are fully or dominantly financed by public expenditures. For each of the 5 years preceding retirement we show the average share of gross income coming from social security benefits in each of the pathways81. Using this indicator we find clear differences between the three groups of people for all five years up to retirement with the biggest difference between the employment pathway and the two other aggregates. Another interesting observation is that the benefit shares even five years back in time seem to be unrelated to the cyclical state of the macro economy which was much better in 2000 than in the two other years.
81
The sum of OAP, PEW, PEP, SDP, TBP, UIB, SWB, SB and benefits for rehabilitation and leave programs. A minor part of the financing comes from labour market contributions and the remaining, dominating part, is financed from general tax revenues.
228
Table 1. The share of gross income that comes from social security benefits 1-5 years before the retirement year by aggregate pathways, 1990, 1995 and 2000.82 1990
1995
2000
Employment t-1 t-2 t-3 t-4 t-5 Max. number of observationsa)
0.19 0.14 0.11 0.09 0.09 3069
0.19 0.16 0.14 0.13 0.12 3332
0.16 0.17 0.17 0.17 0.16 3555
UIB Dominated Pathways t-1 t-2 t-3 t-4 t-5 Max. number of observations a)
0.73 0.48 0.34 0.29 0.27 647
0.77 0.49 0.37 0.30 0.25 3081
0.69 0.49 0.35 0.30 0.26 561
Other Pathways t-1 0.83 0.87 0.84 t-2 0.67 0.65 0.75 t-3 0.51 0.49 0.66 t-4 0.42 0.48 0.59 t-5 0.38 0.43 0.49 135 308 359 Max. number of observations a) a) Notes: Number of observations differs slightly for some years due to missing values.
82
In some cases, the gross income is lower than the amount of social security benefits. In these cases, the share of social security benefits is set to be equal to one. If instead, these observations are set equal to missing, we get the following figures for year 2000: Employment: t-1: 0.13; t-2: 0.13; t-3: 0.13; t-4: 0.12; t-5: 0.11. Max. number of observations: 3433. UIB Dominated Pathways: t-1: 0.67; t-2: 0.47; t-3: 0.33; t-4: 0.29; t-5: 0.25. Max. number of observations: 548. Other Pathways: t-1: 0.73; t-2: 0.66; t-3: 0.56; t-4: 0.51; t-5: 0.41. Max. number of observations: 306.
229
4. Data and empirical model The analyses in this paper draw on longitudinal register data created for administrative purposes. The database consists of a 10 per cent panel sample of the Danish population aged 18-67 years and comprises register-based information on labour market status, receipts of social security, income and labour market and other background characteristics for the period 19842000. We restrict the sample to individuals aged 45-67 years. This sample, which consists of 248,000 individuals, is representative for the 45-67-yearolds through the whole sample period since new 45-year-olds are added every year. In addition, we restrict the sample to individuals that have been in the labour force at least one year during the period covered by the panel. This condition is fulfilled by 77 per cent of 45-67-year-olds in our sample, which corresponds to almost 190,000 individuals.
In this paper, the date of retirement is the first year an individual is out of the labour force full-time and remains so for the rest of the period covered by the sample.83 The criterion of being full-time out of the labour force is determined on the basis of yearly information about the dominant income source.
83
As stated in Palme & Svensson (2002) an obvious problem with this way of measuring date of retirement is that workers who are regarded as retired could in fact have returned to the labour market after the last year included in the data. Therefore, on average, the date of retirement will be underestimated. As retirement or early retirement is predominantly an absorbing state, we expect this problem to be of minor empirical relevance.
230
Several different pathways from labour market participation to OAP or to an early retirement program are identified, cf. Section 3. Yearly information about the primary labour market status by the end of November is also applied for the purpose of determining the labour market status previous to the retirement year. In particular, this information tells us whether individuals are wage earners, self-employed or temporary out of employment. In the latter case, information about social security benefits is applied to distinguish between recipients of UIB, SB and SWB in particular, cf. Section 2. If individuals are working in the year prior to the retirement year, this status as either wage earner or self-employed is classified as the pathway to retirement. However, if individuals are temporary out of employment the first year, information about labour market status two years prior to the retirement year is taken into account. Similarly, information three years prior to this year is taken into account if individuals are also temporary out of employment the second year and so on until five years prior to the retirement year. If information is left-censored, information for the available years is applied. Since the two most dominant types of social security benefits for each individual on average constitute almost 100 per cent of total benefits per year, only two types of benefits per year are included when the pathways are identified. In cases where pathways include two types of benefits, the dominant type is identified. The dominant type is defined as the type of benefits from which the largest amount has been received in the period from work until retirement.
The predominant pathways to retirement are made up of seven pathways that account for almost 98 per cent of the transitions to permanent exit from the labour force. The remaining pathways are included in a residual cate-
231
gory “Others”, cf. Section 3. In the analyses, these eight pathways are aggregated up to three pathways. Individuals who are not yet retired the year in question are chosen as the reference category.
Results of previous studies emphasize the importance of distinguishing between exit states in the analysis of the retirement decision, see e.g. Pedersen & Smith (1996) and Dahl et al. (2000). Therefore, a multinomial logit (MNL) model for the pathways from work to retirement is estimated for selected years. The MNL approach has also been applied in a number of previous studies of pathways to retirement; see e.g. Riphahn (1997), Hernæs et al. (2000) and Dahl et al. (2000, 2003).
The probabilities in a MNL model in which there are J+1 choices for a decision maker i with characteristics x can be written as
Prob(Yi = j ) =
e J
β ' j xi
∑ e β ' k xi
for j = 0,1,2,.....J .
k =0
The MNL estimator treats each category as distinct but of equal rank. Independence from irrelevant alternatives (IIA) is assumed. The MNL model is a choice model. However, the pathways in focus are a mixture of choices and risks. Therefore, in this case the MNL approach is applied simply to expand the two-way correlations in Section 3 to multivariate analyses of the characteristics of individuals that retire through each of the three pathways compared to individuals still remaining in the labour force.
232
Significant gender differences are found in the decision to retire early in Denmark as well as in other countries, see e.g. Pedersen & Smith (1996), Danø et al. (1998, 2000), Christensen & Datta Gupta (2000) and Dahl et al. (2000, 2003). Therefore, separate analyses for men and women are conducted. Denmark is particularly suitable for the study of gender differences in retirement behaviour because the participation rate for women is high compared to the rate in most other countries outside Scandinavia. A number of explanatory variables are included in the analysis, namely age, immigrant, cohabitation status, education, wealth and industry. 84 For descriptive statistics for these variables, see Table A.1 in the Appendix.
Age is entered as interval variables, i.e. 46-49 years, which is the reference category, 55–59 years, which is the relevant age interval for TBP during the whole period from 1992 to 1996, and 50–54 years, which is relevant relative to TBP in the period 1994 to 1996. The interval variable 60–66 years is relevant relative to PEW. In other words, the results for the age variables give an impression of the importance of availability and/or generosity of retirement programs.85 A dummy variable is set equal to one for
84
Initially, the compensation rate was also included as an explanatory variable. However, neither structural nor with respect to interpretation did this variable make sense. First, since we look at a mixture of pathways of which some are choices while others reflect exclusion, we focus on who and not why individuals enter different pathways. Second, the benefit levels are the same within the non-employment groups and therefore, the compensation rate is less interesting in these cases. Consequently, this variable was excluded from the analysis.
85
Finer age intervals would have made it possible to better see the age pattern. However, disaggregation in neither one-year nor two-year intervals is possible due to small sample size for “other pathways”.
233
people being immigrants.86 Cohabitation status is also entered as a dummy variable which is set equal to one for single people.87 Education is divided into three categories: no qualifying education, vocational education and higher education. Dummy variables for vocational and higher education respectively are entered while no qualifying education is the reference category. Wealth is entered as a continuous variable and converted into millions of Danish kroner in 1984-prices. The variable wealth captures only taxable/registered wealth. The wealth tax which was abolished by the end of 1996 had a fairly high tax-exempt base level of wealth. Furthermore, the accrual of pension wealth is not registered in the administrative data on which our sample is based. As a consequence, this variable is a somewhat imperfect measure of true wealth.
In order to get an idea about the importance of the business cycle, a continuous variable for the unemployment rate in the county in year t-1 is entered.88 Dummy variables for the most recent industry of occupation within the last 5 years are entered. We construct 5 categories of industries with employment in the public sector as the reference category. The remaining
86
It would have been optimal to distinguish between Western and non-Western immigrants since the labour market situation for these two groups differ to a large extent. Unfortunately, however, since for most years, our sample contains less than 20 immigrants in each of the two groups UIB dominated pathways and “other pathways”, further disaggregation of the immigrant variable is not possible. 87
It would have been informative to split the group of singles by divorced, widowed and other singles. However, information on divorced and widowed is not available in the dataset.
88
Counties are chosen rather than municipalities because typically Danish municipalities are fairly small. In fact, around 60 per cent of all individuals live in municipalities with less than 40,000 inhabitants. Therefore, the mobility across boundaries is considerable for which reason the unemployment rate at the municipality level is assumed to be less important.
234
categories are “primary or secondary industries”, “wholesale, retail, hotels and restaurants”, “transportation, postal and telegraph services, financing and private services” and finally, “non-employed or information about industry missing”. The latter category is supposed to give an idea about the importance of non-employment within the last 5 years. However, industrial information is in some cases missing for self-employed which might contribute to underestimate the importance of non-employment within the last 5 years since self-employed and non-employed are found in two different pathway categories. On the other hand, however, the construction of the industry variable is based on information for the end of November. That is, if individuals are employed during the year but not by the end of November, this information is not captured by the industry variable leading to overestimating the importance of non-employment within the last 5 years. However, to be employed in particular seasons (excluding November) every year during a five year period is assumed to have a low incidence.
5. Results In Tables 2–7 below, we show the results separately for men and women and for the years 1990, 1995 and 2000 of estimating the multinomial logit model and calculating marginal effects for the allocation to three aggregate pathways out of the labour force. Those in the sample who are not yet retired by the year in question are the reference group relative to the interpretation of the results reported in Tables 2-7. This group constitutes as much as 92-97 per cent of individuals included in each of the analyses.
235
Table 2. Multinomial logit estimates and marginal effects of determinants of retirement through different pathways, men, 1990 (standard errors in parentheses). Variable
Employment Coefficient
Aged 50-54 Aged 55-59 Aged 60-66 Immigrant Single Vocational training Higher education Wealth Local unemployment rate Primary and secondary industries Wholesale, retail, hotels, restaurants Transportation, postal, telegraph, financing, private services Non-employed/ missing Constant Log likelihood Number of observations
0.696*** (0.173) 1.470*** (0.158) 3.783*** (0.139) 0.252 (0.178) 0.322*** (0.072) -0.152* (0.064) -0.835*** (0.101) -0.140*** (0.022)
Marginal effect 0.0074 0.0213 0.1558 0.0025 0.0031 -0.0013 -0.0059 -0.0012
UIB dominated Other pathways pathways Coefficient Marginal Coefficient Marginal effect effect 1.071* 0.0017 0.332 0.0002 (0.459) (0.335) 1.607*** 0.0032 0.021 -0.0000 (0.438) (0.385) 4.397*** 0.0344 1.574*** 0.0015 (0.387) (0.304) 0.749* 0.0013 0.750 0.0007 (0.297) (0.386) 0.847*** 0.0013 1.435*** 0.0016 (0.133) (0.237) -0.231 -0.0003 -0.346 -0.0002 (0.134) (0.256) -1.554*** -0.0012 -0.793 -0.0004 (0.279) (0.447) -0.148*** -0.0002 -0.131* -0.0001 (0.041) (0.052)
-0.015 (0.014)
-0.0001
0.078** (0.030)
0.0001
0.133* (0.057)
0.0001
-0.177* (0.079)
-0.0015
-0.321* (0.156)
-0.0004
0.237 (0.452)
0.0002
-0.217* (0.103)
-0.0018
-0.530* (0.218)
-0.0005
0.646 (0.515)
0.0006
0.036 (0.090) -1.479*** (0.180) -5.313*** (0.202)
0.0003
-1.077*** -0.0009 (0.245) -0.825*** -0.0007 (0.256) -8.181*** (0.505) -6724 52066
0.143 (0.539) 1.936*** (0.426) -9.266*** (0.757)
0.0001
-0.0079
0.0034
* Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level. The group of comparison consists of natives aged 46-49 years that are cohabiting, have no vocational or higher education and for whom the last industry within the last five years was employment in the public sector.
236
Table 3. Multinomial logit estimates and marginal effects of determinants of retirement through different pathways, men, 1995 (standard errors in parentheses). Variable
Employment Coefficient
Aged 50-54 Aged 55-59 Aged 60-66 Immigrant Single Vocational training Higher education Wealth Local unemployment rate Primary and secondary industries Wholesale, retail, hotels, restaurants Transportation, postal, telegraph, financing, private services Non-employed/ missing Constant Log likelihood Number of observations
1.278*** (0.156) 1.835*** (0.153) 4.099*** (0.141) -0.086 (0.165) -0.031 (0.073) -0.083 (0.061) -0.867*** (0.090) -0.047** (0.016)
Marginal effect 0.0114 0.0215 0.1648 -0.0006 -0.0002 -0.0006 -0.0048 -0.0003
UIB dominated Other pathways pathways Coefficient Marginal Coefficient Marginal effect effect 3.214*** 0.0456 1.315*** 0.0017 (0.249) (0.246) 3.772*** 0.0891 1.147*** 0.0014 (0.248) (0.268) 3.794*** 0.0962 0.938** 0.0008 (0.252) (0.317) 0.358** 0.0022 0.850*** 0.0013 (0.130) (0.242) 0.276*** 0.0016 0.876*** 0.0012 (0.073) (0.173) -0.174** -0.0009 -0.399* -0.0004 (0.067) (0.173) -1.328*** -0.0052 -2.184*** -0.0014 (0.123) (0.431) -0.047** -0.0002 -0.054** -0.0001 (0.017) (0.020)
0.353*** (0.011)
0.0024
0.376*** (0.012)
0.0020
0.295*** (0.030)
0.0003
-0.359*** (0.073)
-0.0024
-0.223** (0.082)
-0.0011
-0.736** (0.258)
-0.0007
-0.458*** (0.096)
-0.0027
-0.301** (0.107)
-0.0014
-0.320 (0.300)
-0.0003
-0.200* (0.082) -2.603*** (0.204) -9.210*** (0.201)
-0.0013
-0.844*** -0.0035 (0.112) -0.0083 -0.687*** -0.0028 (0.114) -10.745*** (0.294) -10181 56902
-0.969** (0.328) 0.607** (0.222) -9.780*** (0.464)
-0.0007 0.0008
* Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level. The group of comparison consists of natives aged 46-49 years that are cohabiting, have no vocational or higher education and for whom the last industry within the last five years was employment in the public sector.
237
Table 4. Multinomial logit estimates and marginal effects of determinants of retirement through different pathways, men, 2000 (standard errors in parentheses). Variable
Employment Coefficient
Aged 50-54 Aged 55-59 Aged 60-66 Immigrant Single Vocational training Higher education Wealth Local unemployment rate Primary and secondary industries Wholesale, retail, hotels, restaurants Transportation, postal, telegraph, financing, private services Non-employed/ missing Constant Log likelihood Number of observations
0.411*** (0.124) 0.540*** (0.126) 3.303*** (0.106) 0.229 (0.135) 0.218*** (0.065) -0.026 (0.058) -0.552*** (0.077) -0.039*** (0.010)
Marginal effect 0.0059 0.0082 0.1722 0.0033 0.0031 -0.0003 -0.0065 -0.0005
UIB dominated Other pathways pathways Coefficient Marginal Coefficient Marginal effect effect 0.246 0.0005 0.443 0.0007 (0.318) (0.234) 0.528 0.0011 0.172 0.0002 (0.313) (0.255) 3.315*** 0.0247 1.037*** 0.0016 (0.262) (0.266) 0.554* 0.0014 1.070*** 0.0026 (0.266) (0.234) 0.505*** 0.0011 1.027*** 0.0021 (0.144) (0.170) 0.311* 0.0006 -0.204 -0.0003 (0.138) (0.177) -0.596** -0.0010 -0.878*** -0.0010 (0.208) (0.274) -0.040*** -0.0001 -0.041*** -0.0001 (0.011) (0.012)
0.252*** (0.021)
0.0034
0.259*** (0.051)
0.0005
0.345*** (0.068)
0.0005
-0.150* (0.069)
-0.0020
0.044 (0.181)
0.0001
-0.026 (0.279)
-0.0000
-0.098 (0.088)
-0.0013
-0.059 (0.233)
-0.0001
-0.109 (0.366)
-0.0001
0.008 (0.076) -1.701*** (0.219) -6.123*** (0.173)
0.0001
0.097 (0.202) 0.775*** (0.227) -8.502*** (0.432) -8290 55493
0.0002
-0.167 (0.327) 1.955*** (0.258) -8.685*** (0.517)
-0.0002
-0.0123
0.0022
0.0078
* Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level. The group of comparison consists of natives aged 46-49 years that are cohabiting, have no vocational or higher education and for whom the last industry within the last five years was employment in the public sector.
238
Table 5. Multinomial logit estimates and marginal effects of determinants of retirement through different pathways, women, 1990 (standard errors in parentheses). Variable
Employment Coefficient
Aged 50-54 Aged 55-59 Aged 60-66 Immigrant Single Vocational training Higher education Wealth Local unemployment rate Primary and secondary industries Wholesale, retail, hotels, restaurants Transportation, postal, telegraph, financing, private services Non-employed/ missing Constant Log likelihood Number of observations
0.907*** (0.126) 1.484*** (0.121) 3.522*** (0.108) 0.235 (0.149) 0.144* (0.064) -0.382*** (0.068) -0.619*** (0.092) -0.184** (0.064)
Marginal effect 0.0172 0.0366 0.2251 0.0039 0.0023 -0.0054 -0.0078 -0.0028
UIB dominated Other pathways pathways Coefficient Marginal Coefficient Marginal effect effect 0.673* 0.0016 -0.178 -0.0001 (0.335) (0.451) 1.568*** 0.0055 0.027 -0.0000 (0.303) (0.453) 4.070*** 0.0510 1.794*** 0.0017 (0.268) (0.350) 0.418 0.0011 0.312 0.0002 (0.293) (0.602) -0.334* -0.0007 0.864** 0.0007 (0.145) (0.282) -0.282* -0.0006 -0.547 -0.0003 (0.135) (0.338) -1.466*** -0.0021 -0.815 -0.0004 (0.291) (0.543) -0.285* -0.0006 -0.582*** -0.0004 (0.139) (0.172)
-0.011 (0.013)
-0.0002
0.052 (0.028)
0.0001
-0.124 (0.067)
-0.0001
-0.214** (0.076)
-0.0030
-0.135 (0.160)
-0.0003
-0.356 (0.570)
-0.0002
-0.309*** (0.093)
-0.0042
0.453** (0.156)
0.0012
0.670 (0.446)
0.0006
-0.222* (0.091) -1.398*** (0.135) -4.614*** (0.167)
-0.0031
-1.295*** -0.0018 (0.316) -0.124 -0.0002 (0.178) -7.126*** (0.386) -7028 43825
0.015 (0.568) 1.644*** (0.344) -6.395*** (0.733)
0.0000
-0.0137
0.0022
* Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level. The group of comparison consists of natives aged 46-49 years that are cohabiting, have no vocational or higher education and for whom the last industry within the last five years was employment in the public sector.
239
Table 6. Multinomial logit estimates and marginal effects of determinants of retirement through different pathways, women, 1995 (standard errors in parentheses). Variable
Employment Coefficient
Aged 50-54 Aged 55-59 Aged 60-66 Immigrant Single Vocational training Higher education Wealth Local unemployment rate Primary and secondary industries Wholesale, retail, hotels, restaurants Transportation, postal, telegraph, financing, private services Non-employed/ missing Constant Log likelihood Number of observations
1.233*** (0.120) 1.688*** (0.120) 3.807*** (0.110) -0.051 (0.153) -0.399*** (0.067) -0.330*** (0.064) -0.660*** (0.082) -0.117** (0.040)
Marginal effect 0.0157 0.0263 0.2003 -0.0006 -0.0040 -0.0035 -0.0061 -0.0013
UIB dominated Other pathways pathways Coefficient Marginal Coefficient Marginal effect effect 3.683*** 0.0989 1.749*** 0.0036 (0.204) (0.262) 4.072*** 0.1784 1.289*** 0.0022 (0.204) (0.293) 3.844*** 0.1673 1.084** 0.0011 (0.211) (0.353) 0.249* 0.0025 0.704* 0.0015 (0.122) (0.283) -0.785*** -0.0059 0.464** 0.0008 (0.069) (0.176) -0.368*** -0.0031 -0.516** -0.0007 (0.054) (0.189) -1.885*** -0.0115 -1.362*** -0.0015 (0.117) (0.342) -0.029 -0.0003 -0.216*** -0.0003 (0.041) (0.058)
0.333*** (0.011)
0.0037
0.384*** (0.010)
0.0034
0.274*** (0.031)
0.0004
-0.331*** (0.085)
-0.0034
0.296*** (0.069)
0.0030
0.853*** (0.250)
0.0018
-0.239** (0.091)
-0.0025
0.294*** (0.074)
0.0030
0.431 (0.305)
0.0008
-0.372*** (0.090) -2.218*** (0.157) -8.288*** (0.170)
-0.0036
-0.195* -0.0016 0.063 (0.082) (0.321) -0.0132 -0.838*** -0.0057 1.232*** (0.094) (0.215) -10.306*** -10.162*** (0.236) (0.466) -12250 49117
0.0001 0.0032
* Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level. The group of comparison consists of natives aged 46-49 years that are cohabiting, have no vocational or higher education and for whom the last industry within the last five years was employment in the public sector.
240
Table 7. Multinomial logit estimates and marginal effects of determinants of retirement through different pathways, women, 2000 (standard errors in parentheses). Variable
Employment Coefficient
Aged 50-54 Aged 55-59 Aged 60-66 Immigrant Single Vocational training Higher education Wealth Local unemployment rate Primary and secondary industries Wholesale, retail, hotels, restaurants Transportation, postal, telegraph, financing, private services Non-employed/ missing Constant Log likelihood Number of observations
0.087 (0.119) 0.455*** (0.118) 3.501*** (0.098) 0.242 (0.137) 0.022 (0.061) -0.231*** (0.063) -0.446*** (0.074) -0.120*** (0.024)
Marginal effect 0.0012 0.0071 0.2220 0.0038 0.0003 -0.0032 -0.0057 -0.0017
UIB dominated Other pathways pathways Coefficient Marginal Coefficient Marginal effect effect 0.679* 0.0017 0.897*** 0.0022 (0.313) (0.249) 0.849** 0.0024 0.354 0.0008 (0.316) (0.273) 3.811*** 0.0466 2.237*** 0.0093 (0.279) (0.253) 0.211 0.0005 0.263 0.0006 (0.285) (0.259) -0.126 -0.0003 0.257 0.0006 (0.143) (0.155) 0.083 0.0002 -0.548*** -0.0011 (0.130) (0.163) -1.044*** -0.0019 -1.248*** -0.0021 (0.226) (0.267) -0.153*** -0.0003 -0.155*** -0.0003 (0.034) (0.038)
0.232*** (0.023)
0.0033
0.207*** (0.050)
0.0005
0.262*** (0.059)
0.0005
-0.219* (0.086)
-0.0029
0.520** (0.175)
0.0014
0.131 (0.277)
0.0003
0.027 (0.085)
0.0004
0.187 (0.205)
0.0005
0.329 (0.276)
0.0008
-0.200* (0.085) -1.622*** (0.186) -5.601*** (0.166)
-0.0027
0.371* (0.182) 0.797*** (0.191) -7.934*** (0.412) -7768 47836
0.0010
0.417 (0.249) 2.175*** (0.182) -7.959*** (0.436)
0.0010
-0.0129
0.0026
0.0138
* Significant at a 5% level, ** significant at a 1% level, *** significant at a 0.1% level. The group of comparison consists of natives aged 46-49 years that are cohabiting, have no vocational or higher education and for whom the last industry within the last five years was employment in the public sector.
241
In general, the most pronounced effects are found for the age variables. In fact, relatively large marginal effects89 are found for age for men as well as women for the employment and the UIB dominated pathways across 1990, 1995 and 2000 suggesting that availability and/or generosity of retirement programs play a dominating role when individuals retire through these pathways. This result applies to women in particular. Retirement through PEW seems to be important independently of the cyclical situation. For the years 1990 and 2000 the impact from being aged 60 to 66 years on the retirement probability through the Employment pathway is increased by 20 – 23 percentage points for women and by 16 – 17 percentage points for men. In 1995 the TBP was open and this is seen as expected to have a strong impact on the retirement probability through the UIB dominated pathways, especially for women. For both women and men the marginal effect from being aged 55 – 59 increased in 1995 to the same level as for the 60 – 66 years old when we consider the UIB dominated pathways. Thus, the program impact is very strong.
The marginal effects from the age variables are small for the “other pathways” group, which is composed of persons in different programs before entry to retirement, are somewhat complicated to interpret. In general, the marginal effects are rather small due to the small volume of this group. Once again, there is an impact from the TBP which was open in 1995. In this year the marginal effect for the 55 – 59 years old group – for both women and men – is higher than for the 60 – 66 years old group. For the
89
The size of marginal effects is difficult to interpret across pathways due to the big difference regarding volume. Therefore, we only focus on the relative size of these effects within each of the three groups of pathways.
242
other two years, the marginal effect is highest for the 60 – 66 years old, reflecting presumably the impact from PEW in the mixed programs for this group. Comparing the marginal effects across years for women and men it seems reasonable to assume that TBP dominates SDP in 1995 and that part of the explanation of the relatively big marginal effect for 60 – 66 years old women in 2000 is transitions from leave schemes to retirement for this group.
As regards the remaining explanatory variables, marginal effects are either fairly small in general or only relatively large for individuals retiring through “other pathways”. The latter result is found for all three years for women and for 1990 and 2000 for men for one of the industry controls that are entered with public sector employment as the excluded category, namely the “non-employed/ information about industry missing” category. For individuals retiring through this type of pathways, it seems reasonable to assume that this variable typically indicates non-employment rather than self-employment, see also Section 4. Missing industrial information increases the probability of retiring through “other pathways” by 0.3-0.8 percentage points for men and 0.2-1.5 percentage points for women which is fairly high compared with the other non-age explanatory variables. The lack of information regarding employment for the preceding 5 years seems to become relatively more important through time, especially for women, probably reflecting that this becomes a more selected group – dominated by non – employment or employment in marginal jobs – after 5 years of cyclical upswing between 1995 and 2000.
243
With the reservation in mind that there are rather few immigrants in the sample, cf. Section 4, a relatively large positive marginal effect is found for being a male immigrant retiring through UIB dominated and “other pathways” in the second half of the 1990s. On the other hand the coefficient to the immigrant variable is insignificant for the Employment pathway. In other words, not surprisingly immigrants are in particular found to retire through the pathways reflecting the most loose attachment to the labour market. The share of immigrants aged 45-66 years increased during the sample period which probably explains that being an immigrant is only found to be important in 1995 and 2000. The coefficient to being a female immigrant are insignificant in 1990 and 2000, while relatively high marginal effects in 1995 relative to UIB dominated pathways and “other pathways” presumably reflects the impact from TBP for these groups of workers.
Being single has a significant impact for men increasing the probability for retirement through all pathways, except the employment related in 1995, which might reflect that employed single men to a larger extent than similar cohabiting men postponed retirement due to the economic upturn in the second half of the nineties. Looking at the marginal effects, they tend to be of the same magnitude as what we find for being an immigrant.
In accordance with prior expectations, having a higher education implies a significantly lower probability for exit from the labour force. The variable is however insignificant for the group of “other pathways” in 1990 which most probably reflects that very few highly educated people were leaving the labour force through those pathways at that time. 244
The coefficient to wealth is significantly negative regarding all pathways except the UIB dominated pathways for women in 1995 where the TBP once again tends to dominate the impact from other factors. It will however, based on the negative sign, be premature to conclude that the ongoing building up of pension assets will result in later retirement per se as pension wealth obviously correlates with other important retirement age determinants.
Looking at the local unemployment rate we find a significant positive impact in 1995 and 2000 in accordance with prior expectations. In 1990 two of the coefficients and positive for men but the marginal effects are extremely small. For women the local unemployment rate is insignificant in 1990 for all three groups of pathways. The different 1990 results probably reflects that this was a year of very high and fairly evenly distributed unemployment while 1995 was the initial phase of a cyclical upswing of which 2000 was a peak with a higher relative variation in regional unemployment rates.
6. Conclusions The purpose of this paper is to describe and analyse the multitude of pathways to exit from the labour force. We focus on the pathways leading to either normal OAP or to exit from the labour force into a program for early retirement.
245
Eight pathways from work to normal OAP or to an early retirement program are identified. Entry from a job is dominant covering almost 2/3 of all transitions. The two other big states from which entry occurs are UIB and from a working life as self-employed. The remaining about 10 per cent of the pathways from which entry occurs are states where people are being provided for by SWB alone or by UIB in combinations with SWB or SB. Gender differences are present. Retirement through UIB and UIB in combination with SB is dominated by women, while SWB alone or in combination with UIB are very male dominated pathways.
The share of people who have retired directly from employment, whether wage earners or self-employed, clearly reflects the cyclical profiles since the middle of the 1980’s. Further, this share is exceptional low in the middle of the 1990’s due to the temporary opening up of an early exit possibility in form of the TBP scheme, which was available for unemployment insured who had been long-term unemployed. Focusing on retirement from non-employment, the share of retired through pathways including UIB reflects also to some extent the Danish business cycle in the sample period.
The eight pathways seem to fall into three groups. One group is transitions directly from employment for which the main trend regarding destination is an increasing importance of PEW over time. Another group is the pathways dominated by UIB for whom PEW is even more important. Further, the TBP is an important destination for this group. Finally, the last group is dominated by benefits reflecting a low attachment to the labour force in the period prior to transition to retirement. For this group of pathways called “other pathways” the same increasing importance of the aggregate PEW 246
destination is observed although unsurprisingly at a lower level and with dominance of SDP and Other exit destinations. For all three groups of pathways, we observe the very low incidence of entry directly from OAP.
Looking at the average share of gross income coming from social security benefits for each of the 5 years preceding retirement we find a clear difference between the three groups of pathways for all five years with the biggest difference between the employment pathway and the non-employment pathways. Further, we find that the benefit shares even five years back in time seem to be unrelated to the cyclical state of the macro economy.
We estimate the interaction between a big number of background variables and use of the three aggregated pathways by estimation of a multinomial logit model for selected years with those in the sample who are not yet retired by the year in question as the reference group. Our approach is descriptive trying to judge the importance of different background factors relative to the use of pathways with a focus also on differences by gender. The pathways differ in a qualitative sense as the employment pathway is choice based while the non-employment pathways can be used only conditional on fulfilling the criteria for collecting benefits, be it UIB, SB or SWB. Due to this, an optimising approach is outside the scope of the present paper.
Turning to the estimation results, a number of background variables have the same impact across all or nearly all pathways and years. This is the case for having a higher education and for wealth which for higher values re-
247
duce the probability of using these pathways. The opposite, systematic positive impact on use of these pathways is found for people aged 60-66 years, for single men and for people with an indicator for missing industrial information.
Other background variables show more variation in their impact, mostly consistent with prior expectations. The local unemployment rate has as expected a systematic positive impact on the use of all three pathways but only in 1995 and 2000 which might reflects a high relative unemployment rate for older workers in these years. Similarly, being male immigrant makes retirement through the non-employment pathways more probable in 1995 and 2000. The results reflect strongly the fact that the TBP was open in 1995. Namely, we find a significant positive impact for people aged 5054 and 55-59 years on the use of all three pathways.
Overall, the results suggest that availability and/or generosity of retirement programs as reflected in the marginal effects from the age variables, are very important for retirement through the employment and UIB dominated pathways while individual background factors are of minor importance. This result applies in particular to women. For retirement through “other pathways”, however, personal characteristics seem to be at least as important as retirement programs when looking at the marginal effects from age variables relative to other individual factors.
The transition from work to retirement is complex and far from the conventional idea of exit typically occurring from a job at the official pension age.
248
An interesting approach in future work would be to have access to health data making it possible to build a competing risks model where some pathways are used due to health shocks and others are chosen based on economic optimisation comparing compensation rates with disutility from continued work.
7. References Arendt, J.N., Hansen, E.B., Olsen, H., Rasmussen, M., Bentzen, J. & Rimdal, B. (2003) Living conditions among old age pensioners without supplementary income (in Danish). Report 03: 12. The Danish National Institute of Social Research, Copenhagen. Bingley, P., Datta Gupta, N. & Pedersen, P.J. (2003) The Impact of Incentives on Retirement in Denmark, in Social Security and Retirement Around the World: Microestimation, (eds.) Jonathan Wise and David Gruber, NBER, 2003. Bjørn, N.H. & Larsen, M. (2003) Delay of Retirement (In Danish). Social forskning, 2003: 2. Copenhagen. Christensen, B.J. & Datta Gupta, N. (2000) The Effect of a Pension Reform on the Retirement of Danish Married Couples (in Danish). Nationaløkonomisk Tidskrift, 138, pp. 222-242. Dahl, S.Å., Nilsen, Ø.A. & Vaage, K. (2000) Work or retirement? Exit routes for Norwegian elderly. Applied Economics, 32, pp. 1865-1876. Dahl, S.Å., Nilsen, Ø.A. & Vaage, K. (2003) Gender differences in Early Retirement Behaviour, European Sociological Review, Vol. 19, No. 2, 179-198. Danø, A.M., Ejrnæs, M. & Husted, L. (1998) Gender Differences in Retirement Behaviour, Institute of Local Government Studies - Denmark, Copenhagen. Danø, A.M., Ejrnæs, M. & Husted, L. (2000) How is the Retirement Age Affected by the Reform of the Post Employment Wage program? (in Danish), Nationaløkonomisk Tidsskrift, 138, pp.205-221. Economic Council (2001) Danish Economy, Spring 2001. Copenhagen. Hansen, H. (2000) Elements of Social Security, A comparison covering: Denmark, Sweden, Finland, Austria, Germany, The Netherlands, Great Britain and Canada, The Danish National Institute of Social Research 00:7, Copenhagen. Hernæs, E., Sollie, M. & Strøm, S. (2000) Early retirement and Economic Incentives. Scandinavian Journal of Economics 102(3), 481-502.
249
Høgelund, J. (2003) In Search of Effective Disability Policy. Comparing the Developments and Outcomes of Dutch and Danish Disability Policies, Amsterdam University Press. Initiative Committee of Senior Policy, Ministry of Labour (1999) The Seniors and the Labour Market now and in the Future (In Danish), Copenhagen. Jørgensen, M.S. (2004) The Danes postpone retirement from the labour market a bit (in Danish). Social forskning, 2004:1. Copenhagen. Kohli, M. & Rein, M (1991) The changing balance of work and retirement, in Time for Retirement Comparative Studies of Early Exit from the Labour Force (Eds.) M. Kohli, M. Rein, A.M. Guillemard & H.v. Gunstern, Cambridge University, pp. 1-35. Lausten, M. (2001) Transfer Incomes and Income Distribution (in Danish), Copenhagen. Nørregaard, C. (1996) Work and Retirement in the Nineties – and Retired Persons in the Future (in Danish). Copenhagen: Socialforskningsinstituttet. OECD (2000) Employment Outlook. Paris. OECD (2003) Employment Outlook. Paris. Palme, M. & Svensson, I. (2003) Income Securiy and Retirement in Sweden, forthcoming in Social Security and Retirement Around the World: Microestimation, (eds.) Jonathan Wise and David Gruber, NBER, 2003. Pedersen, P.J. (1998) The Elderly and the Labour Market, in Smith, N. (ed) Work, Work Incentives and Unemployment (in Danish), Aarhus, pp. 151-178. Pedersen, P.J. & Smith, N. (1995) The Retirement Decision, in Mogensen, G.V. (ed) Work Incentives in the Danish Welfare State, New Empirical Evidence, The Rockwool Foundation Research Unit, Aarhus. Pedersen, P.J. & Smith, N. (1996) A Duration Analysis of the Decision to Retire Early, in Wadensjö, E. (ed.) The Nordic Labour Markets in the 1990's, Amsterdam, pp. 31-68. Quaade, T. (2001) Retirement from the Labour Market (In Danish). The Danish National Institute of Social Research. Report 01: 7. Copenhagen. Quaade, T. (2002) Reform of the Post Employment Wage Programme with Limited Effect (in Danish), in: Socialforskningsinstituttet, Social forskning 2002:3, Copenhagen, 10-11. Riphahn, R. T. (1997) Disability retirement and unemployment – substitute pathways for labour force exit? An empirical test for the case of Germany, Applied Economics, 29, pp. 551-61. Ruhm, C.J. (1990) Bridge Jobs and Partial Retirement. Journal of Labor Economics, vol. 8, no. 4.
250
8. Appendix
Table A.1. Descriptive statistics, men and women, 2000. Men Means Std. err.
Women Means Std. err.
Age Aged 50-54 Aged 55-59 Aged 60-66
0.32 0.28 0.15
(0.47) (0.45) (0.36)
0.35 0.27 0.11
(0.48) (0.44) (0.32)
Immigrant
0.04
(0.21)
0.04
(0.20)
Single
0.20
(0.40)
0.24
(0.43)
Education Vocational training Higher education
0.44 0.25
(0.50) (0.43)
0.37 0.27
(0.48) (0.44)
Wealth in 1,000,000 d.kr., 1984-prices
0.80
(3.99)
0.38
(2.45)
Unemployment rate in the county in year t-1
5.35
(1.18)
5.33
(1.18)
0.35 0.13
(0.48) (0.34)
0.12 0.11
(0.33) (0.31)
0.21
(0.41)
0.14
(0.35)
The most recent industry within 5 years Primary or secondary industries Wholesale, retail, hotels, restaurants Transportation, postal and telegraph services, financing and private services Non-employed/ information about industry missing Number of observations
0.06 0.23 55493
0.07 (0.26) 47836
251
252
Resume (Summary in Danish)
Mona Larsen
253
1. Introduktion Ældres tilbagetrækning fra arbejdsmarkedet er i fokus, fordi de demografiske ændringer i de kommende årtier betyder, at ældre vil udgøre en stigende andel af arbejdsstyrken. Desuden er der sket markante ændringer på arbejdsmarkedet for ældre. Ældres erhvervsfrekvens er således faldet de seneste årtier og deres arbejdsløshed har været relativt høj siden sidste halvdel af 1990’erne. Adgangen til tidligere tilbagetrækningsordninger har tillige været under forandring. Efterlønsordningen blev således ændret i 1992 og 1999 med det formål at få ældre til at udskyde deres tilbagetrækningstidspunkt. Omvendt var der midthalvfemserne midlertidigt åbnet mulighed for permanent tilbagetrækning for 50-59-årige via overgangsydelsesordningen. Disse ændringer giver anledning til bekymring for fremtiden især hvad angår potentiel fremtidig arbejdskraftmangel og stigende forsørgerbyrde. Problemerne kan bl.a. imødekommes ved at fastholde ældre længere tid på arbejdsmarkedet, fordi man ad den vej på samme tid øger arbejdsudbudet og skattegrundlaget, mens de offentlige udgifter til pension og tidlige tilbagetrækningsordninger som efterløn og førtidspension formindskes.
Formålet med denne ph.d.-afhandling er at forøge den eksisterende viden om de faktorer, der påvirker tilbagetrækningsbeslutningen specielt med henblik på at få forøget indsigt i, hvordan man kan fastholde ældre på arbejdsmarkedet efter 60 års alderen. Afhandlingen indeholder fire papirer, der vedrører forskellige aspekter af tilbagetrækningsadfærden. Både effekten af ikke-økonomiske aspekter, hvilket her omfatter helbred samt jobfaktorer og jobtilfredshed, og effekten af økonomiske incitamenter er undersøgt. Desuden er afgangsvejene til tilbagetrækning er kortlagt. Tidligere
254
danske og internationale undersøgelser viser, at der er signifikante forskelle mellem mænd og kvinder, når det drejer sig om beslutningen om at trække sig tilbage, se f.eks. Pedersen & Smith (1996), Danø et al. (1998, 2000), Christensen & Datta Gupta (2000) og Dahl et al. (2000, 2003). Med det formål at forøge vores viden om kønsforskelle, hvad angår tilbagetrækningsadfærden, er der derfor i alle de tilfælde, hvor det har været muligt, udarbejdet separate analyser for kvinder og mænd.
2. Jobforhold og jobtilfredshed påvirker tilbagetrækningstidspunktet Normalt inkluderer tilbagetrækningsmodeller ikke virksomhedsrelaterede forhold, hvilket bl.a. skyldes mangel på data, se f.eks. Hakola (2003).90 Jeg har imidlertid adgang til et unikt datasæt som indeholder en række informationer om lønmodtageres egen vurdering af jobforhold og jobtilfredshed. Disse data er anvendt i det første papir: “The Effect of Job Characteristics and Job Satisfaction on Planned Retirement Age”. Formålet med papiret er at se på, om lønmodtageres vurdering af jobfaktorer og jobtilfredshed spiller en rolle, når tilbagetrækningstidspunktet planlægges. Resultaterne af papiret giver dermed et væsentligt input til den igangværende seniorpolitiske debat om, hvordan man skal overbevise ældre (50 år eller derover) om, at de skal forblive længere tid på arbejdsmarkedet. Effekten af jobforhold og jobtilfredshed er analyseret i en ordered probit model, hvor der samtidig kontrolleres for effekten af helbred, uddannelse, samliv, finansielle forhold,
90
Som undtagelser kan nævnes papirer af Hurd & McGarry (1993) og Friedberg (2001), hvor der er anvendt Health and Retirement study data.
255
forventet tilbagetrækningsindkomst og objektive jobfaktorer. Desuden er der korrigeret for uobserverbar heterogenitet. Resultaterne for det danske arbejdsmarked er perspektiveret ved at gennemføre en tilsvarende amerikansk analyse baseret på Health and Retirement study data. USA er valgt, fordi de to lande på den ene side ligner hinanden, hvad angår baggrundskarakteristika og arbejdsmarkedstilknytning for den ældre del af befolkningen, mens strukturen af og de indbyggede incitamenter i de to landes pensionssystemer på den anden side er meget forskellig.
Danske lønmodtageres egen vurdering af jobforhold og jobtilfredhed synes at have betydning for, hvornår de forventer at trække sig tilbage. Disse faktorer har dog væsentlig mindre betydning end f.eks. den indkomstkilde, der forventes at være den vigtigste, når man trækker sig tilbage – dette gælder især, hvis denne indtægtskilde er efterløn. Når lønmodtagere overvejer, om de vil trække sig tilbage som 60-årige eller tidligere eller om de vil vente til de har passeret 60 års alderen synes især jobtilfredshed, tilfredshed med arbejdstiden, den normale tilbagetrækningsalder i ens stillingskategori på arbejdspladsen og vanskeligheder ved at leve op til arbejdspladsens krav at spille en rolle. Disse faktorer er generelt vigtigere for kvinder end for mænd. Fjernelse af barrierer i forhold til at opfylde jobkrav som f.eks. krav om brug af ny teknologi er tilsyneladende den mest velegnede fremgangsmåde, når kvinder skal overbevises om, at de skal forlænge deres arbejdsliv efter de 60 år, mens en forøgelse af tilfredsheden med arbejdstiden synes at være den mest effektive metode i forhold til mænd. Der er dog behov for relativt store ændringer i disse faktorer for at opnå en mærkbar effekt på tilbagetrækningsalderen. Sammenlignet med situationen blandt danske lønmodtagere er jobforhold og jobtilfredshed mindre vigtige i USA, hvilket
256
synes at hænge sammen med et mindre finansielt råderum for ældre amerikanske lønmodtagere.
3. Dårligt helbred fremskynder tilbagetrækningstidspunktet I det andet papir ”The Impact of Health on Individual Retirement Plans: a Panel Analysis comparing Self-reported versus Diagnostic Measures”, som er udarbejdet i samarbejde med Nabanita Datta Gupta, undersøges, om helbred eller økonomiske faktorer er vigtigst for, hvornår ældre forventer at trække sig tilbage fra arbejdsmarkedet, jf. kapitel 3. I papiret sammenlignes den rolle, som hhv. subjektive og objektive helbredsmål spiller som determinanter for tilbagetrækningsforventninger, når der samtidig kontrolleres for indkomst, arbejdsmarkeds-, job- og baggrundskarakteristika. En lang række forskellige helbredsmål anvendes, herunder både surveybaserede selvrapporterede mål og diagnoser hentet fra Landspatientregisteret. Papiret bidrager desuden med ny viden til debatten om ”justification bias”, dvs. den positive bias, som tilbagetrukne personer overfører til selvrapporterede helbredsmål ved at angive svigtende helbred som undskyldning for tidlig tilbagetrækning. Endelig udvides den eksisterende litteratur af Dwyer & Mitchell (1999) og McGarry (2002) ved både at kontrollere for uobserverbar heterogenitet og tage højde for endogenitet og målingsfejl samt ved at estimere separate modeller for kvinder og mænd. Vores analyse er baseret på en panelmodel for effekten af helbred på individuelle tilbagetrækningsplaner og er estimeret på basis af en stikprøve af ældre erhvervsaktive og tilbagetrukne udtrukket fra Ældredatabasen, der er en dansk panel survey fra 1997 og 2002, som er koblet med longitudinale registerdata fra perioden 1993-2001. 257
Hypotesen om ”justification bias” er tilsyneladende ikke er opfyldt, hvilket svarer til konklusionen i de fleste nyere amerikanske studier. Desuden synes hverken endogenitet eller målingsfejl at give anledning til bekymring i de danske data. Uobserverbar heterogenitet viser sig derimod at være vigtigt og estimater fra random effects modeller viser, at selvvurderet fysisk og mentalt helbred er vigtige for forventninger om tilbagetrækningsalderen, faktisk mere vigtige end økonomiske faktorer. Dette gælder både for kvinder og mænd. Helbred synes dog at være relativt vigtigst for mænds tilbagetrækningsforventninger. Dårligt generelt helbred reducerer således mænds og kvinder forventede tilbagetrækningsalder med henholdsvis 1,3 år og 8 måneder. Dårligt mentalt helbred sænker mænds forventede tilbagetrækningsalder med omkring 1,2 år og kvinders med omkring et år. Mænds tilbagetrækningsalder nedsættes også, hvis deres helbred er dårligere end andres, eller hvis de har en sygdom. Tilsvarende fremskynder nedsat arbejdsevne kvinders forventede tilbagetrækningstidspunkt. På det disaggregerede niveau fremskynder rygsygdomme og myoser mænds tilbagetrækning, mens rygsygdomme og især knogleskørhed/afkalkning af knogler og depression er tilstande, der udløser tidligere tilbagetrækning blandt kvinder. Tilbagetrækningsforventninger er generelt upåvirket af hospitalsindlæggelse for en alvorlig sygdom, undtagen i tilfælde af indlæggelse som følge af hjertesygdomme, hvilket reducerer den forventede tilbagetrækningsalder for både mænd og kvinder om end effekten kun er marginal. Ved at rette blikket mod helbredsændringer bestyrkes konklusionen om, at helbred er en vigtig faktor, når det gælder planlægning af tilbagetrækningstidspunktet. Helbredschok forøger således tilsyneladende tilbøjeligheden til at trække sig tidligt tilbage. Dog synes helbred at være mindre vigtigt for planlægningen af tilbagetrækningstidspunktet i Danmark sammenlignet med USA,
258
hvilket formentligt kan tilskrives det substituerede og universelle danske sundhedssystem og den lettere adgang til førtidspension.
4. Tilbagetrækningsordninger skal ændres markant for at ændre tilbagetrækningsadfærden Fokus i det tredje papir “An Experimental Analysis of the Effect of an Increase in Delaying Incentives in the Post Employment Wage Program on Retirement Age” er på de ændringer i efterlønsordningen, der blev gennemført i 1992, jf. kapitel 4. Formålet med dette papir er at belyse effekten af denne ændring på den gennemsnitlige tilbagetrækningsalder. Ændringen bestod i, at incitamenterne til at forblive på arbejdsmarkedet indtil 63 års alderen blev forøget. Det foregik i praksis ved, at reduktionen i efterlønsydelsen efter 2½ år, der tidligere berørte alle efterlønsmodtagere, blev fjernet for dem, der trak sig tilbage ved de 63 år eller senere.
Mens tidligere danske studier af effekten af ændringer i pensions- og tidlige tilbagetrækningsordninger er baseret på simuleringer, se Danø et al. (2000), Christensen & Datta Gupta (2000) og Bingley et al. (2003), er der gennemført en eksperimentel analyse i dette papir, hvilket omfatter difference-indifference (DD) og difference-in-difference-in-difference (DDD) analyse. Eksperimental analyse kan være nyttig i denne sammenhæng, fordi kausale effekter af politikændringer på tilbagetrækning kan udledes. ”Behandlingsgruppen” består af personer, der er berettiget til efterløn, mens personer, som kan få tjenestemandspension, udgør kontrolgruppen. Analysen er base-
259
ret på en repræsentativ 2% stikprøve af personer og deres partnere for perioden 1980-1998.
Alt i alt tyder resultaterne på, at effekten af ændringen i efterlønsordningen i 1992 på tilbagetrækningsalderen var relativt lille i det mindste på kort sigt. Fordelinger af og hazard-funktioner for tilbagetrækningsalderen indikerer, at ændringerne ikke havde den tilsigtede effekt. Imidlertid indikerer DDD analyser, at ændringen i efterlønsordningen formindskede incitamenterne til at trække sig tilbage i 63 og 64 års alderen, hvilket var en del af formålet. Men analyserne tyder også på, at den samlede effekt var lille i den forstand, at lovændringen ikke signifikant påvirkede overgang til tilbagetrækning ved hverken 60-62 år eller 65-66 år. Derfor steg den gennemsnitlige tilbagetrækningsalder for personer berettiget til efterløn ikke.
5. De fleste trækker sig tilbage fra beskæftigelse før pensionsalderen Formålet med det fjerde papir “Pathways to retirement in Denmark, 19842000”, som er udarbejdet i samarbejde med Peder J. Pedersen, er at beskrive mangfoldigheden af afgangsveje fra arbejde til tilbagetrækning inden for en institutionel ramme med en lang række tilgængelige ordninger, jf. kapitel 5. Dvs. tilgang enten til folkepension eller til tidlige tilbagetrækningsordninger som efterløn eller førtidspension fra et job som lønmodtager, fra en position som selvstændig eller fra forskellige typer af sociale eller arbejdsmarkedsrelaterede ydelser eller kombinationer heraf. Vi fokuserer på udviklingen siden midten af 1980’erne. Den eksisterende viden om den relative størrelse af disse afgangsveje til tilbagetrækning er begrænset. Viden 260
af denne type er vigtig i forskning og politikdiskussioner af reformer af velfærdsydelser og tilbagetrækningsordninger.
Multinomial logit (MNL) analyser af veje fra arbejde til tilbagetrækning er gennemført for udvalgte år. MNL modellen er valgt, fordi den muliggør analyse af flere afgangsveje til tilbagetrækning. Modellen er baseret på valg, men da afgangsvejene i fokus består af en blanding af valg og risiko er MNL tilgangen i denne sammenhæng anvendt til at gennemføre multivariate analyser af karakteristika ved personer, der trækker sig tilbage via forskellige afgangsveje til tilbagetrækning sammenlignet med personer, der forbliver i arbejdsstyrken. Datagrundlaget er en 10% panelstikprøve af den danske befolkning i alderen 45-67 år, som følges i perioden 1984-2000.
Analyserne viser, at overgange fra arbejde til tilbagetrækning er komplekse og langt fra den konventionelle ide om, at afgang fra arbejdsstyrken typisk forekommer fra et job til folkepension, når man når pensionsalderen. Vi finder frem til otte forskellige afgangsveje fra arbejde til folkepension eller til en tidlig tilbagetrækningsordning. Disse afgangsveje opdeler vi i tre grupper: En gruppe er overgang direkte fra beskæftigelse, som omfatter tre ud af fire af alle overgange i analyseperioden. En anden gruppe er afgangsveje, der er domineret af arbejdsløshedsdagpenge, hvilket omfatter 20 procent. De resterende 5 procent er afgangsveje domineret af ydelser, der afspejler en svag tilknytning til arbejdsstyrken i perioden forud for tilbagetrækning. Den relative størrelse af disse aggregerede veje er påvirket af konjunktursvingningerne i perioden og af eksistensen af overgangsydelsen i midten af 1990’erne. Efterlønsordningen benyttes i stigende omfang på tværs af alle tre grupper, mens tilgangen direkte til folkepension er meget 261
begrænset. Generelt synes det forhold, at tilbagetrækningsordninger er tilgængelige og/eller er økonomisk attraktive at være meget vigtigt for tilbagetrækning både direkte fra beskæftigelse og via de afgangsveje, der er domineret af arbejdsløshedsdagpenge, mens individuelle baggrundskarakteristika har mindre betydning. For tilbagetrækning via andre afgangsveje synes personlige karakteristika omvendt at være mindst lige som vigtige som tilbagetrækningsordninger.
6. Opsummering Resultaterne i denne ph.d.-afhandling tyder på, at faktorer som jobforhold og jobtilfredshed, helbred, ændringer i tilbagetrækningsydelser og afgangsveje til tilbagetrækning at vigtige er tage i betragtning i forbindelse med bestræbelserne på at fastholde ældre længere tid på arbejdsmarkedet. Betydningen af de enkelte faktorer er forskellig for mænd og kvinder. Dette taler for, at forskellige fremgangsmåder skal tages i brug for at tilskynde både mænd og kvinder til at forlænge deres arbejdsliv.
7. Referencer
Bingley, P., Datta Gupta, N. & Pedersen, P.J. (2003) The Impact of Incentives on Retirement in Denmark, in Social Security and Retirement Around the World: Microestimation, (eds.) Jonathan Wise and David Gruber, NBER, 2003. Christensen, B.J. & Datta Gupta, N. (2000) The Effect of a Pension Reform on the Retirement of Danish Married Couples (in Danish). Nationaløkonomisk Tidskrift, 138, pp. 222-242. Dahl, S.Å., Nilsen, Ø.A. & Vaage, K. (2000) Work or retirement? Exit routes for Norwegian elderly. Applied Economics, 32, pp. 1865-1876. Dahl, S.Å., Nilsen, Ø.A. & Vaage, K. (2003) Gender differences in Early Retirement Behaviour, European Sociological Review, Vol. 19, No. 2, 179-198.
262
Danø, A.M., Ejrnæs, M. & Husted, L. (1998) Gender Differences in Retirement Behaviour, Institute of Local Government Studies - Denmark, Copenhagen. Danø, A.M., Ejrnæs, M. & Husted, L. (2000) How is the Retirement Age Affected by the Reform of the Post Employment Wage program? (in Danish), Nationaløkonomisk Tidsskrift, 138, pp.205-221. Dwyer, D.S. & Mitchell, O.S. (1999): Health problems as determinants of retirement: Are self-rated measures endogenous? Journal of Health Economics 18 (1999), p. 173-193. Friedberg, L. (2001) The Impact of Technological Change on Older Workers: Evidence from Data on Computer use, National Bureau of Economic Research, Working Paper No. 8297. Hakola, T. (2003) Alternative Approaches to Model Withdrawals from the Labour Market - A Literature Review, Working Paper Series, Department of Economics, Uppsala University, No. 2003:4. Hurd, M. & McGarry, K. (1993) The Relationship Between Job Characteristics and Retirement, National Bureau of Economic Research, Working Paper, No. 4558. McGarry, K. (2002): Health and Retirement: Do Changes in Health Affect Retirement Expectations? NBER Working Paper 9317. Pedersen, P.J. & Smith, N. (1996) A Duration Analysis of the Decision to Retire Early, in Wadensjö, E. (ed.) The Nordic Labour Markets in the 1990's, Amsterdam, pp. 31-68.
263
