As we will demonstrate, demand for low-skilled hours is more elastic .... Iowa firms must comply with the minimum wage if their sales exceed. $3(K),0()0 per year.
Minimum Wage Effects on Hours, Employment, and Number of Firms: The Iowa Case* PETER F. ORAZEM and J, PETER MATTILA Iowa State University, Ames, lA 500] 1 \. Introduction Until recently, economists were uniquely united in the opinion that increases in the minimum wage reduced employment. Frey et al. (1984, p. 991) reported that almost nine often U.S. economists agreed with the statement "A minimum wage increases unemployment among young and unskilled workers." The consensus empirical result, as summarized by Brown et al. (1982). was that a 10 percent increase in the minimum wage reduced teenage employment by 1-3 percent. The least skilled segments of the population (especially school dropouts) appear to have been most adversely affected. This consensus was challenged in highly publicized studies by Card and Krueger (1995) and Katz (1992). Their findings, as summarized in Card and Krueger (1995, p. 1), constitute "a new body of evidence showing that recent minimum-wage increases have not had the negative employment effects predicted by the textbook model." In fact, "some of the new evidence points toward a positive effect of the tninimum wage on employment: most shows no effect at all." Their research was used to buttress arguments for increasing the federal minimum wage in 1996. Although several critiques of Card-Krueger ("Review Symposium," 1995; Bellante and Picone, 1999; Burkhauser et al., forthcoming) and several studies using new data sets (Partridge and Partridge. 1999; "Symposium," 1999) have subsequently been published, the minimum wage debate is yet to be resolved. No consensus will likely prevail until the weight of scientific evidence tends to dominate one side or other of the debate. Given this situation, it is important to investigate new data sets and make improvements in methodology if economists and policy makers are ever to reach a consensus. We contribute to the dialogue using a new data set and making, in our view, innovations and improvements in the methodology of minimum wage analysis. Our study is in the spirit of the longitudinal methodology used by Card-Krueger and others but has several important advantages. In order to minimize problems associated with aggregation bias, we use both firm-level and county-level data sets within a state (Iowa) rather than the more aggregate state-level data sets used in some studies (e.g., Partridge-Partridge). Like Card-Krueger, we collect our own primary data JOURNAL OF LABOR RESEARCH Volume XMII. Numlwr 1 Winlir 20112
JOURNAL OF LABOR RESEARCH set (a survey of retail and service firms), but we go beyond this and supplemetit it witb government data from payroll and tax records. These data provide information on whether firms are covered by the law as well as demographic characteristics, wages, and hours of their workers. One of the major advantages of our data set is tbat it allows us to disaggregate workers and impacts. That is. we separate workers paid less than the new minimum wage, and hence potentially directly impacted by an increase in the minimum, from workers who already earn more than tbe new minimum, wbo are less likely to be alfccied. Few other studies have been able to incorporate this dichotomy into their analysis (Linneman, 1982, is an exception). Such a dichotomy is important if one is to more precisely measure the magnitude of impact and distributional effects of minimum wage increases. These two groups of workers are likely to be substitutes such that declining employment of low-wage workers may be offset, in part., by rising employment of high-wage workers. Studies that aggregate these two groups of workers may underestimate minimum wage impacts on low-wage workers. Our data set also has the advantage of allowing us to estitiiate tbe impact on hours as well a.s numbers of workers. As we will demonstrate, demand for low-skilled hours is more elastic to wage increases than is demand for numbers of workers. Previous literature has been content to focus on the impact of rising minimum wages on employment. In contrast, we stress that the most meaningful approach is to analyze minimum wage impacts in terms of more conventional elasticities of demand for labor hours. We also have information on the number and size of firms which is rarely available. In recent years, economists have paid more attention to the creation and destruction of firms as part of labor market dynamics. An additional advantage is that our data span tbe years 1989-1992 wben the newly legislated Iowa minimum wage increased rapidly and exceeded most other minimum wage rates. Hence, we observe a large shock which should help to identify employment effects. Briefly, Iowa was subject only to the federal $3.35 minimum wage, having no state minimum wage law prior to 1990, Despite being a relatively low-wage state, Iowa established its first state minimum wage on January I, 1990, wbich at $3.85, exceeded the federal rate.' At tbe same time, Iowa expanded coverage to small retail and service firms having annual sales as low as 60 percent of the federal threshold.- Iowa's minimum rose to $4.25 on January 1, 1991 and to $4.65 on January 1, 1992. Throughout this period, Iowa's rates exceeded both the federal minimum and the minimum wage rates of surrounding states. One advantage of our data set is that we can study the impact of highlegislated rates in a relatively low-wage state. Many other minimum wage studies have focused on high-wage states such as New Jersey, Pennsylvania, and California. In Section II, we analyze Iowa county-level data for retail and nonprofessional service industries, based primarily on Unemployment Insurance records and Census data. This analysis bas tbe advantage of completeness, covering all firms and workers, and provides a benchmark with which to compare the firm-level analysis that follows. However, the county-level data suffer the limitations inherent in aggregation. In
PETER F. ORAZEM and J. PETER MATTILA particular, they don't allow us to separate low-wage workers who may be directly impacted from higher wage workers who already earn more than the minimum wage. In Sections III and IV we analyze firm-level data generated from our own survey. We measure employment and wage rates by worker so that they can be dichotomized into groups below and above the new minimum wage. Although small sample si7.es and other data limitations necessitate caution in interpreting some of our results, our estimates suggest that dichotomizing worker effects is important. That is. minimum wage impacts tend to be larger when workers are dichotomized by level of eamings than when they are not. Equally important, our methodology can fruitfully he applied as other micro-level data sets become available.
IL County-Level Analysis One advantage of using eounly-level aggregate data is that one can study the impact of minimum wages on the number and size of firms. In other contexts, economists have studied job creation and destruction associated with the birth and death of firms (Hamermesh, 1993), but relatively little is known about the impact of minimum wages on firms and tirni size. In addition, county-level data can provide an aggregate henchmiirk of employment and earnings effects against which to compare estimates using firm-level data. Our strategy is to analyze changes in county-industry cell employment (or earnings or number of firms) as a function of changes in Ihe minimum wage relative to county-industry wage levels, while controlling for national changes in industry employment and wage levels and for county changes in income levels. In addition, we allow for minimum wage interactions with county-industry measures of coverage under the law and county measures of the rural composition of the population as shown in equation 1: In Yij,/Yij,.i = oto + (a, + otjCy + a.,/?, + c^Qj * Ri) In [MW,/U^j,_|]
(1 >
+ a5 In (E;i/Ef,.O + Oft In (Wfj/Wfl.O + a^ In (/y,//,,^) + H^,, where Y is alternatively number of firms, number of employees, or quarterly earnings in the county/industry cell; C is the proportion of firms covered by the Fair Labor Standards Act in the county-industry cell; /? is a dummy variable if the county is rural; MW, is the Towa minimum wage;Wf/,_i is the predicted hourly wage rate by industry and county as described below; Ef^is national employment in the industry at time /; VVf'is the national average hourly wage in the industry at time /; and I, is per capita income in the county in period /. Changes in national industry employment and wages control for exogenous shifts in industry demand, while changes in county income control for localized demand shifts. The national data came from Employment and Earnings. By law. Iowa firms must comply with the minimum wage if their sales exceed $3(K),0()0 per year. Data on the proportion of firms by county and industry with sales above $300.0(X) were obtained from the Iowa Department of Revenue and Finance.-* County per capita income was obtained from tapes provided by the Bureau of Eco-
JOURNAL OF LABOR RESEARCH nomic Analysis. Counties were classified as rural if county population in the 1990 Census of Population was greater than 38 percent rural. The specification allows for the minimum wage effect to differ between rural and urban markets and between covered and uncovered markets. The dependent variables are taken from legally mandated quarterly unemployment insurance (UI) forms collected by the Iowa Department of Workforce Development (IDWD) and aggregated to the county and industry level. We focus only on those low-wage retail and service industries likely to be impacted by minimum wage changes. These include each two-digit retail trade (SIC 52-.i7.59) and nonprofessional service (SIC 70-79) industry but exclude eating and drinking establishments {SIC 58} due to the complicatiotis associated with tips. Unfortunately, these UI data do not report hours or hourly wage rates, which we need to estimate equation 1. However, the IDWD. in cooperation with the U.S. Bureau of Labor Statistics, also collects "establishment" data on hours and wage rates in Its "shuttle survey." We use the latter to estimate hourly wage rates for each of our countyindustry cells. Although more detail can be obtained from the authors, we obtained 15 quarters of this "shuttle survey" data 1989:2 through 1992:4 for a cumulative total of 171,947 workers In Iowa. We regressed log hourly wage rates by worker on individual attributes (gender, production worker status, proportion of overtime hours), countylevel variables (proportiotis rural, female, with high school diploma, and with college degree), national variables (industry employment and wage rates), and the current minimum wage rate alone and interacted with county and industry dummy variables.** Predicted wage rates by county /, industry j . and quarter; (H^;^,) were computed such that cross-sectional variation is attributable to differences in mean wage rates across industries and differences in county attributes. Temporal variation occurs because of changes in the national variables and changes in the minimum wage. Our estimates of equation I are reported in Table I. One-quarter changes in the data refer to changes from the quarter before (e.g.. 1989:4) to the quarter in which the minimum increased on January 1 (e.g., 1990:1). Four-quarter changes refer to fourthquarter to fourth-quarter changes in the data (e.g., 1989:4 to 1990:4). Elasticities measuring the response of each of the dependent variables to changes in the minimum wage rate are shown at the bottom of Table 1. The estimates imply that a ten percent increase in the minimum wage relative to the previous wage causes a decrease in the number of firms of 1.7 percent in one quarter and 2.5 percent over four quarters. The effect on firm numbers is only marginally larger in the eovered sector and does not differ much between urban and rural counties. Quarterly employment also falls in response to the minimum wage increase. The effect is larger in the covered than in the uncovered sector and is larger after four quarters than after one quarter. The employment elasticity of around -. I is consistent with the lower end of the estimates in Brown et al. (1982). Taken together, the more elastic response of firm numbers than of employees to the minimum wage increase implies tbat average firm size rises with the minimum wage.
PETER F. ORAZEM and J. PETER MATTILA
d.
Ic •S Op
a
oo
fO OS
^
^
o • "-o^ .a U.
•-' — in
r
f
00 O
— oo
•S
I .—.•q'.—. m -—^00.—O — — . O — Q r o v • " i - m t " - — • O ^ f O O ^ •
00
o
— r o < n p ^
— o
r-
'^
f->
I
u
I I-3 ^s
JOURNAL OF LABOR RESEARCH On the other hand, we have already pointed out that the employment response may be underestimated. The county-level data combine workers whose wages in (-1 were below the new minimum wage {hereafter, the subminimum group) and those whose wages in f-I exceeded MW, (hereafter, the superminimum group). If these groups are substitutes in production, aggregate data will capture the net change but not the impact on the subminimum group. We attempt to address the latter issue in the next section. Quarterly earnings fall in all sectors in response to the minimum wage increase. The covered sector eaming's elasticity is approximately -. 12 over one quarter and -. i 5 over four quarters. The earnings elasticities are greater in magnitude than are the employment elasticities. In fact, negative earnings elasticities require that the aggregate hours elasticity with respect to the minimum wage must be in the elastic range.^ The results from Table I are similar to the pre-Card/Krueger consensus estimates for employment responses to the minimum wage. Hours elasticities are infrequently reported in the literature, although our tindlngs of elastic hours responses are consistent with those reported by Linneman (1982). To our knowledge, there are no recent studies of the response of Hrm numbers to the minimum wage increases, but neoclassical theory predicts that an increase in input prices will reduce tinii numbers, other things equal. To the extent that larger (irms have higher wages and more capital-intensive production processes, their costs will rise by a smaller proportion than will the costs of small firms. This accords well with our finding that firm size rises in response to the minimum wage.
III. Firm-Level Analysis: Data and Stylized Facts As discussed previously, the county-level data have a serious limitation in that they aggregate sub- and superminimum workers. This should bias toward zero tbe estimated employment and earnings responses lo the minimum wage increase. To examine this possibility, we collected data on hourly wage rates (or average work hours) for a sample of firms in these retail and service industries. Multi-establishment firms were excluded to avoid confusion of respondent location with establishment location. Tbe exclusion of multi-establisbment firms effectively excluded general merchandise firnis (SIC 53), although the exclusion was not by design. In general, we focus on smaller firms than exist in the population. Because the minimum wage was raised in the first quarter of 1990, after remaining constant at $3.35 since 1982. the universe of retail and service sector firms was taken to be those firms in business in fourth quarter 1989. Sampled firms were followed longitudinally from 1989:4 through 1992:1, a period which contained the successive minimum wage increases in 1990:1. 1991:1, and 1992:1. Selection of Finns and Collection of Data. IDWD records of firms paying into Unemployment Insurance contained 17,362 single-e.stablishment retail and nonprofessiona! service firms in 1989:4. Recall that neither hours nor hourly wages are provided, necessitating our survey. A random sample of 1,201 firms (roughly 7 percent
PETER F. ORAZEM and J. PETER MATTILA of the universe) was selected for inclusion in the study. Of these, 329 had changed owners, merged, eliminated all employees (which halted reporting to IDWD). or closed by 1992:3. These firms were excluded. Thirty-seven firms did not have telephone numbers. This left 835 firms still in existence in 1992 which had the same owner, slill had employees, and were sending quarterly reports to IDWD. Because of confidentiality rules, firms were first contacted by IDWD for permission to participate in the survey and for the release of their unemployment insurance records. These initial phone contacts were made in March 1993. Of the 835 existing firms with phone numbers. 55 percent agreed to release their records; 25 percent refused to release their records; and 20 percent could not be contacted for various reasons (disconnected phones, unavailable owners, or no answer). The majority of those refusing said they had no records or records were difficult to locate. Others stated that they did not have the time to participate or were reluctant to disclose wage rates. Unemployment insurance records included information on quarterly employment and earnings for individual employees. Driver's license records were merged by social security number to get the gender and age of each worker. A survey was sent to the 460 firms which agreed to participate. The 460 cooperating firms were distributed across industries and urban and rural counties in roughly equal proportions lo their distribution in the universe of firms. In addition to other questions, firms were asked to list hourly wage rates for each of their workers in 1989:4, 1990:1, 1990:4. 1991:1. 1991:4 and 1992:1. Ultimately, 212 firms returned the survey, 139 of which supplied useable hourly wage data on an average of 772 workers per quarter. All 460 firms are incorporated into the analysis below. One innovation of our study is that we measured the coverage status of each firm. For the 460 firms that agreed to release their records, the Iowa Department of Revenue and Finance (IDRF) released sates tax records for 1990. I99I. and 1992. Firms with reported sales of S300,0(X) or more were considered covered (as stated by Iowa statute) and those below $300,000 were considered uncovered in the analysis below. About one-quarter of retail employees and one-third of service sector workers in the 460 firms fell into the uncovered sector.^ Although pioneering, our minimum wage coverage variable has some inherent ambiguities and must be interpreted with caution. First, the $300,000 annual minimum sales criterion is a moving test, the results of which may change each quarter. A firm with $320,000 sales prior to the first quarter would legally be required to pay minimum wages but would later be exempt if its annual sales fell to $295,000 prior to the fourth quarter. Since we only have calendar-year sales data, our coverage variable will not capture switching of this nature. On the other hand, it is questionable whether such a firm would actually change its pay practices in such circumstances. More likely, the firm would either continue to pay the minimum wage in both quarters or to ignore the law in both quarters. Since the law is enforced on a complaint basis, it is plausible that some firms may claim (believe) and inform their employees that they are exempt even though they are not exempt. Other firms may believe that they are covered even when they are not.
JOURNAL OF LABOR RESEARCH
Second, the law provides exemptions and special cases which our data aren't sufficiently delailed to handle. For instance, individual workers engaged in interstate transactions such as credit card sales or shipping/receiving are .suhject (as individuals) to the federal minimum wage even though their firm is exempt from both the state and federal minimum wage laws. Seasonal amusement and recreation firms may be exempt from both the state and federal rates, even though their total revenues exceed $300,000. Full-time students, learners, teenagers (during their first 90 days of employment), and the handicapped may also he paid less than the minimum wage under certain conditions. Given these complications, it is appropriate to regard our coverage variable with some caution. Our coverage variable might alteniatively be interpreted as measuring firm sales rather than legal ohiigations concerning wage rates. Wage Estimation. While direct information on hourly wages was available for an average of 772 individuals each quarter, relying on observed wages could cause significant biases in the estimation. Minimum wage changes could alter the skill composition ofthe iahor force. That is, firms might want to switch to higher skilled workers. Consequently, changes in average wages would partially retieet changes in average skills and not the desired change in the wage per unit of skill. Ideally, we would like to hold skills constant at their prior levels. In order to do this, we estimate an earnings function of the form In W;., = Yi/^- + yiAi: + y3.AJ,+yMi,*Fi
^
+ y^Al^Fi + YfeQ + y^NEMPj
(2)
y5Cy, + S n,Ai,y, + v,,,
where F is a female dummy variable; A is age; C is firm-coverage status; NEMP is the number of employees in the firm; the M are county labor market variables (including per capita income, proportion rural, proportion of women in the labor force, and the proportion with either high school or college degrees); and the SIC are industry dummy variahles. Decomposition into Subminimum and Superminimum Groups. Estimates of equation (2) are reported in Appendix Table AI for all (six) first and fourth quarters in the sample period. The model explains 41 percent to 51 percent ofthe variation in wage rates, which is quite good for micro wage data. Age, as a proxy for experience, has the expected quadratic effect. Females earn less than males (evaluated at the mean age of the interaction effects). Covered firms with higher sales pay higher wages, although the small negative impact of employment size is a puzzle in the two regressions for which it is statistically significant. Wage rates tend to he lower in rural counties and higher in counties having a higher percentage of females. We suspect that the latter reflects the endogenous tendency of high-wage areas (such as Des Moines) to attract more young females. By holding fixed the earnings function coefficients in a given quarter (hereinafter, the quarter's earnings structure), we can predict period t hourly wage rates for work-
PETER F. ORAZEM and J. PETER MATTILA ers employed in al! other periods. This generates a wage rate distribution for each period, while holding the time /earnings structure fixed, but allowing the worker attributes to reflect those of the workers actually employed in eacb period. This predicted wage distribution (W',) allows those employed in period / to be decomposed into two groups. The subminimum group is defined as those for whom \n{MW,+ i/W',) > 0 and the superminimum group is those for wbom \iUMW,+ ]/W',) < 0.^ where MW,^\ is the minimutn wage implemented one quarter after period /. The estimated parameters in (2) also enable us to predict what workers in a different period, r', would have been paid at time /. This procedure has the advantage that we can include newly hired employees at /'. even thougb they weren't employed at time /. We derive the subminimum population at time f' as all employees for whom \niMW,^\/W}') > 0. holding the time / earnings structure constant. In Table 2. we employed this strategy using tbree different base earnings structures. The first two columns present the predicted subminimum group, using the 1989:4 wage structure and the 1990:1 implemented minimum wage of $3.85. The two middle columns use the 1990:4 earnings structure and the 1991:1 implemented minimum wage of $4.25. The final two columns use the 1991:4 wage structure and the 1992:1 implemented minimum wage of $4.65. This strategy holds returns to firm and individual attributes (the earnings structure) fixed over time, so the simulated changes in etnployment shares are due to changes in the distribution of attributes of the 460 firms' employees. Whatever the possible biases in the assignment into super- and subminimutn groups, those biases are fixed over time. The simulated subminimum employment shares were generated for all employees in the 460 firms and aggregated separately for urban, rural, covered, and noncovered employees. The major purpose of this exercise is to see if employment patterns differ between niral and urban areas and between covered and noncovered firms. If all firms were subject lo general economic trends, we might expect to observe similar patterns for all of these groups. That is. if there were a general increase iti the demand for more experienced workers, we should observe increasing superminimum employment shares and decreasing subminimum shares In Table 2. The interesting result is that we only observe ihis pattern for urban tu-eas and covered firms (and all employed). That is, comparing the earliest and the final corresponding quarters (so as to avoid seasonal variation), [he superminimum shares increaseJ by 0.7 to 3.0 percentage points in urban ftrms and by 1.3 to 2.6 percentage points in covered firms. In contrast, the superminimum employment shares declined in all but one case for noncovered firms (-0.1 to -1.9 declines except for one +0.2 increase) and for three of six cases for rural firms (the six changes being -1.0, -0.3, -0.2, 0. +1.2, and +2.4).^ Further examination of the data indicated that at least part of tbe differential patterns for noncovered and rural firms can be traced to teenage employtnent. Whereas the proportion of teenagers declined in other sectors by 5 percent to 22 percent, the proportion increased by at least 17 percent in the rural and noticovered sectors.^ One interpretation is that less skilled workers (e.g., teenagers) spilled over to the noncov-
12
JOURNAL OF LABOR RESEARCH
Table 2
Simtilated Suhminimum Employment Shares, Based on Currently Employed Workers in 460 Firms, Holding Wage Structures Fixed at the 1989:4, 1990:4, and 1991:4 Levels Year and Quarter
1989:4 Cijefticients Subminimum" Superminimum
1990:4 Coefficients Subminimun 1 Supcrminimuni
199l:4Coefficiems Subminimum*^^ Superminimum
All Employed 89:4
27.4'^
72.6%
18.6%
81.4%
29.0%
71.0%
90:1
22.5
77.5
14.2
85.8
26.0
74.0
90:4
27.8
72.2
18.1
81.9
28.9
71.1
91:1
22.9
77.1
14.5
85.5
25.5
74.5
91:4
27.0
73.0
17.3
82.7
27.7
72.3
92:1
21.5
78.5
13.4
86.6
23.8
76.2
Rural Counties Only 89:4
29.77r
70.3%
21.5%
78.5%
36.1%
63.9%
90:1
28.8
71.8
19.7
80.3
35.7
64.3 62.9
90:4
31.8
68.2
22.9
77.1
37.1
91:1
30.2
69.8
21.4
78.6
36.3
63.7
91:4
29.9
70.1
22.5
77.5
36.4
63.6
92:1
27.0
73.0
20.0
80.0
33.3
66.7
74.7%
Urban Counties Only 89:4
26.2%
73.8%
17.1%
82.7%
25.3%
9():|
18.8
81.2
II.O
89.0
20.4
79.6
90:4
25.5
74.5
15.4
84.6
24.3
75.7
91:1
[8..1
81.7
10.2
89.8
18.8
81.3
91:4
25.3
74.7
14.3
85.7
22.7
77.3
92:1
18.1
8L9
9.3
90.7
18.0
82.0
Not Covered Firms Only 89:4
28.1%
71.9%
21.1%
78.9%
36.3%
63.7%
911:1
26.9
73.1
19.7
80.3
35.2
64.8
90:4
28.9
71.1
2i.9
78.1
36.3
63.7
91:1
28.8
71.2
21.3
78.7
35.3
64.7
91:4
30.0
70.0
21.3
78.7
36.4
63.6
92:1
28.8
71.2
20.8
79.2
35.0
65.0
89:4
27.8%
72.8%
17.7%
82.3%
26.4%
73.6%
90:1
20.8
79.2
i2.i
87.9
22.5
77.5
90:4
27.4
72.6
16.7
83.3
26.2
73.8
91:1
20.6
79.4
11.9
'88.1
21.8
78.2
91:4
25.9
74. i
15.8
84.2
24.5
75.5
92:1
18.9
81.1
10.8
89.2
19.9
80.1
Covered Firms Only
Noles:
^Proponion of wnrkers employed in the year and quaner wilh predicied pay below $3.85 in 1989:4. Proportiun of workers employed inihe year and qiiiirtcr wiih predicted piiy below $4.25 in 1990:4. ••Pixiportionof workers employed inihe year anil quarter with predicted puy below $4.65 in IWI:4.
PETER F. ORAZEM and J. PETER
M
A
T
T
I
L
A
1
3
ered sectors as would be predicted in Mincer's (1976) model.'" On average, rural areas have a higher proportion of subminimum workers than urban areas, especially in noncovered firms (footnote 10). In the next section, we present evidence that subminimum employment in rural firms is less negatively attected by changes in minimum wages than in urban firms.
IV. Firm-Level Analysis: Comparative Statics In this section, we analyze these firm-level data using an employment shares demand model, an approach similar in spirit to that used in Section 11 but with the advantage of distinguishing between subminimum and superminimum workers. As emphasized previously, the effect of minimum wages on changes in employment and earnings of these two groups of workers should differ. We first briefly summarize our model, then discuss the data before presenting our results. An appendix which presents our model in more detail is available from the authors on request. Methodology. Designate total subminimum hours for firm / as //f (for below), the wage rate paid to this group as W,^, and this group's share of total earnings at time t as Sll= {WlJHfjC,,). where C,, is total cost of subminimum and superminimum workers. Assuming output constant demand for factor inputs at time /, neutral technological change, ignoring non-labor inputs, and assuming that the superminimum group's wage rate can be characterized by a random error term," the first difference ofthe subminimum share equations can be written as; Sft^^ -Sf,= a ln{MW,,i/W,^) + e,,,
(3)
where e,, is an expression involving the error terms from eaeh share equation and the random walk process. It can be shown that the parameter a in (3) provides an estimate of the elasticity of subminimum earnings share with respect to the wage change. Dropping subscripts for simplicity: d5«/d \niW^) ^S^ + S^ r]BB - ( 5 ^ = a,
(4)
where Tl/j^ is the ela.sticity of demand for hours of the subminimum group with respect to the wage rate. Given estimates of a and S, an estimate of r\BB can easily be derived from (4). Alternatively, the subminimum group share can be measured as the employment share, L'/= Ef/Ej, where E^ is number of subminimum employees and £ ' is the sum of subminimum and superminimum employment. Equation (3) could be estimated substituting the employment share as the dependent variable. This would allow us to estimate a corresponding coefficient (a^) on the wage ratio term. The parameter a ' provides an estimate of the elasticity of the subminimum employment share with respect to the wage change: ^
+
%B L^)
= a^
(5)
JOURNAL OF LABOR RESEARCH where Ogg is the elasticity of detnand for subminimum employment with respect to its own wage rate; Q^n is the cross elasticity of demand for superminimum employment with respect to the wage of subminimum workers: and L^ is the employment share of superminimum workers. It can be shown that the subminimum employment elasticity can be measured as:
).
.
(6)
If hours are less costly to adjust than are numbers of employees, demand for labor hours should be more elastic than demand for employees, implying that ria^ < 9^^. However, the opposite may be true if hiring and training costs are very low, as may be the case with unskilled labor. Constructing the Variables. The firm's earnings and employment shares were estimated using the earnings structures reported in Appendix Table Al. The 1989:4 eamings structure was used to estimate hourly wage rates for all those employed in the 460 firms in 1989:4, 1990:1. and 1990:4. All workers in these periods whose predicted wage in 1989:4 is less than $3.85. the minimum wage implemented in 1990, are considered subminimutn workers. In quatter / (where / = 89:4, 90:1, or 90:4) the subminimum group are those for whom \nO.?,5IWi^y) > 0. This methodology allows us to assign workers to sub- or superminimum groups, even if they were not employed in 1989:4. The employment share for subminimum workers is the ratio of predicted subminimum workers to all workers. The eamings share of subminimum workers is the quarterly earnings for those identified as subminimum workers relative to total quarterly earnings. One-quarter changes (89:4 to 90:1) and four-quarter changes (89:4 to 90:4) in these shares provide the dependent variables for equation (3). The method is repeated for the 1991 minimum wage of $4.25. using the 1990:4 earnings structure. In this context, we define the subminimum group as those for whom In(4.25/Wj()) > 0 where / = 90:4. 91:1, or 91:4. Likewise for the 1992 minimum wage increase to $4.65, the 1991:4 earnings structure identifies subminimum groups using ln(4.65/H'^) for / = 91 4 or 92:1. It is very important to emphasize that this method does not force tbe subminimum employment share to decline as workers* wages are raised to the new higher rate. Actual wage rates are not used in constructing these data. Rather, tbe employment sbares change solely due to changes in the demographic mix of employees in each firm. That is. changes in the proportions of teenagers and women change the subminimum employment shares. By holding constant the wage structure that existed prior to tbe minimum wage increase, we prevent (1) automatic decreases in the subminimum population and (2) bias due to changes in the returns to skill that might occur as a result of the legislated rate increase. The key independent variable in (3) is tbe ratio of the minimum wage relative to the average predicted wage of the subminimum group in the period before the minimum wage increase.'- This ratio was set to one if (he firm employed no subminimum workers. In log form, the variable takes a minimum value of zero for titms employing only superminimum workers in both periods (and hence face zero effective cbange in
PETER F. ORAZEM and J. PETER MATTILA
15
the subminimum wage) and takes on positive values for employers having subminitnum workers. The log ratio of minimum wage to average previous subminimum wage is interacted with a coverage dummy variable (C). a rural dummy variable (R), and the product {CR) as previously in (I). The specification allows estimation of several potential differences in responses to the minimum wage: impacts on subminimum versus superminimum workers, on rural versus urban workers, and on covered versus noncovered workers. Results. The regressions of changes in subminimum employment shares and of changes in subminimum quarterly earnings shares are reported in Table 3.'^ Estimates for one-quarter (3 changes per firm) and four-quarter (2 per firm) changes are shown separately to allow for possible lagged adjustments by firms. Joint F-tests of significance of all coefficients involving iMW,/W,_[) variables were computed to determine whether minimum wages had neutral effects on superminimum and subminimum employment. The neutrality hypothesis is rejected in every specification. While the Rh are small, the coefficients are quite stable across dependent variables, specifications and time intervals. We have generated employment and hours elasticities (Table 4) from the regression estimates (Table 3). Estimates of the share elasticities (a and a^) are always negative and are statistically significant at the 5 percent level in 15 out of 16 of the estimates in which industry dummy variables were entered. Likewise, estimates of the hours elasticity r\BB and the employment elasticity QBB are always negative. The employment elasticities are in the range -.22 to -.85. varying with the speciticiition. The magnitudes are larger than those obtained using the county-level data. As expected, disaggregation makes a difference. Subminimum employment using firmlevel data is more strongly affected than is total employment at the county level. Although there is no clear pattern between covered and noncovered firms, the employment elasticities are uniformly less negative in rural areas than urban areas. Although there is no evidence of a Mincer-type spillover effect that raises subminimum employment, we do find evidence that the reduction is smaller in rural firms. As discussed in the previous section, rural firms have more subminimum workers and may be less subject to enforcement. The hours responses are more elastic, ranging from -1.0 to -1.5 with the specification. This is consistent with the hypothesis that hours are easier (lower cost) to adjust ihan is employment. When we control for industry, a 10 percent increase in the minimum wage will cause about a 6 percent reduction of subminimum employment in urban firms one year later, but a 13-15 percent reduction in hours worked. As with employment, the hours elasticities typically are less negative in rural areas than urban areas. We note that these estimates are short-run elasticities. Over longer periods of lime, employment may adjust more fully, implicitly, capital is con.stant in our analysis. Over longer periods, capital may be substituted for labor. This possibility is somewhat allayed by the fact that small retail and service sector firms are not capital intensive and may have limited capital-labor substitution possibilities.''^
JOURNAL OF LABOR RESEARCH Table 3 Iowa Firm-Level Estimation of Changes in Subminimum Shares As a Function of the Minimum Wage Employment Shares
Eamings Shares
1
2
3
4
-.072" (6.13)
-.088" (7.06)
-.028" (2.49)
-.036" (3.03)
.004 (.26)
-.019 (1.07)
-.007 (.43)
-.021 (1.25)
R * \n(\fW,/W,_i)
.04\" (2.99)
.041" (3.03)
.012 (.91)
.011 (.86)
C*R* ln(MW,/W,_i)
-.022 (1.09)
.008
(38)
-.002 (-13)
.014 (.71)
,062"
.088"
.016"
.029"
1274
1274
1274
1274
\n(MW,/W,_t)
-.065" (3.63)
-.079" (4.22)
-.029* (1.91)
-.036" (2.30)
C*\n(.MW,/W,_i)
.029 (1.18)
,010 (.40)
-.004 (.17)
-.020 (.94)
R*]n(MW,/W,_i)
.007 (.35)
.006 (,30)
-.010 (1.28)
-.015 (.85)
C ^ R * \n (MW,/W,_O
.001 (.02)
,024 (.75)
.008 (1,78)
.052" (1.98)
One-Ouarter Chanee
SIC dummies included*" R^
•
Four-Ouaner Change
SIC dummies included''
•
•
R-
.039"
.055^
.022'
,044=
N
845
845
845
845
Notes: l-stalisiics in parenlheses. • ( " ) indicates signiticance al the .10 (.05) level. "F-tesl oflhe joini hypolhesis ihal all cwnkienls except the ranstani icmi are equal lo zero is rejecled at the .01 level. ^A constant term was also included in the spetilifaiions thai excluded frulusiry dummy variables.
17
PETER F. ORAZEM and J. PETER MATTILA
Table 4
Demand Elasticities Implied by the Firm-level Demand Regressions No Industry Controls Share Ela.sticity
Industry Controls
Demand Elasticity
Share Ela.>iticity
Demand Elasticity
Fmplovnienl
fii
Emplovment
Rmplovmenl
oL
Urhan Covered
1 quarter changes 4 quarter changes
-.068" -.035
-.54 -.28
-.106" -.069"
-.85 -.55
Uncovered
1 quarter changes 4 quarter changes
-.072" -.065"
-.58 -.52
-.088" -.079"
-.70 -.63
Co\ered
I quarter changes 4 quarter changes
-.049" -.028*
-.39 -.22
-.057" -.039"
-.46 -.31
Uncovered
1 quaner changes 4 quarter changes
-.032" -.058"
-.26 -.46
-.046" -.073"
-.37 -.58
a
Hours (r\Bn)
OS
Hours (n«ft)
Covered
I quarter changes 4 quarter changes
-.035" -.032*
-1.27 -1.24
-.057" -.056"
-1.50 -1.49
Uncovered
1 quarter changes 4 quaner changes
-.028" -.029*
-1.20 -1.21
-.036" -.036"
-1.28 -1.28
Covered
1 quarter changes 4 quarter changes
-.026" -.010
-1.17 -1.01
-.031" -.019
-1.23
Uncovered
1 quaner changes 4 quaner changes
-.016* -,034"
-1.07 -1.26
-.025" -.051"
-1.16 -1.44
Rural
Urban
Rural
Noiey. The cmploymenl based demand elasticities are compuicd using equation (6), and the eaming.s based demand elaslicilics are computed using equation (4). Parameter values are taken from Table 3. The top number is the one-quaner elasticity. Thi; bottom number is the foor-qiiiirier elasticity. * (**( represents a value of a.significantly differeni from zero at the .HK.O.'il level. Over the period, those predicted to be in the subminitnum group have average employnienl sbare of .1.10 and average earning share oi" .096.
18
JOURNAL OF LABOR RESEARCH
V. Conclusions Our results show that minimum wages reduce employment opportunities for workers. Like the early studies, our county-level (more aggregate) estimates imply fairly mtxlest impacts with elasticities of approximately -0.1. However, our firm-level estimates for suhminimum workers imply much more elastic responses. In particular, hours elasticities are in the elastic range. One may question why we find much more elastic demand responses compared to the earlier studies. The real mystery, however, is why minimum wage studies have yielded such small elasticities when studies of the demand for teenage and unskilled labor routinely find more elastic demand.'"^ Our interpretation is that the level of disaggregalion makes a difference. The more closely a minimum wage study focuses on young and low-wage (what we call subminimum) workers, the larger the measured impact is likely to be. The corollary is that youth, women, and minority workers are most likely to be adversely affected by high minimum wage rate increases. Our results imply declining aggregate eamings for such workers.'^ If economists and policy makers are to reach consensus on minimum wage effects, it is important to obtain and study new data sets and to improve the methodology. We make contributions in both regards. First, we analyze a new firm-level data set, augmented with information on individual workers. Relatively few such data sets with the necessary variables exist. Second, we have used a methodology which we believe provides improvements. In particular, our technique for separating subminimum from superminimum workers and isolating the minimum wage impacts on the former may be useful to others in tbis field of research. Our data set. like most, is not perfect. Admittedly, our firm-level sample is small and limited to a three-year period in one state. On the other hand, our data have several advantages over otbers. in addition to the level of disaggregation already emphasized. For example. Iowa is a relatively low-wage state, so the sizeable minimum wage increases caused larger shocks than found in many data sets. Our study is one of the few that can explore tbe effect on worker hours as well as numbers. We also have one of the few data sets that allows us to distinguish between rural and urban areas, finding smaller employment impacts in rural areas. And altbough it did not have much impact on our results, we were able to distinguisb wbetber individual firms are covered by the minimum wage laws. Finally, we are among the few lo study the impact of minimum wages on the number and size of firms, fmding thai minimum wages reduce the number but increase average firm size.
PETER F. ORAZEM and J. PETER MATTILA
19
NOTES * We thank the USDA Cooperative State Research Service Grant j('92-37401 -8284 for financial support. Steve Smith and Dick Sampson at the Iowa Department of Employmt-ni Service. Rich Jacobs and John Godwin al Ihe luwa Departmeni of Revenue and Finance, and Mark Imerman, Iowa t-xlension Program S|xx-ialist. helped us obtain our data. We appreciate Ihe assistance of Darin Wohlgemulh. Jooseop Kim. and Alex Turk in coordinating, programming, and analyzing this data set. Kelly Cordaro and Norma Meadc did the telephone survey and key entry work. Donna Otto prepared the manuscript. 'in 1993. Iowa ranked 42 out of 50 states in average annual income for unemployment insurance covered wage and salary workers. In contrast. New Jersey ranked third {Statistical Ah.strart of the United States. 1995). -Apparently. Iowa legislators thought these would be the mandated minimum wage levels in federal legislation which was under consideration and expected to pass. After the Iowa legislation was passed. U.S. Congress passed an amended bill with a lower level. Therefore, the higher minimum wage in Iowa was probably an accident, but Iowa legislators are hesitant to admit thai on the record. -^To test for sensitivity, coverage was also measured by proportion of total sales in the county-industry cell made by covered firms. The re.sults were not changed appreciably. A better coverage measure mighl be the proportion of workers covered per county/industrj' cell, but that measure was not available. •'The results of this regression are available on request. ^Let quarterly earnings be W H. The elasticity of quarterly eamings with respect to wages is r\vfn = d In(W-7/)/d \n{W) = iW/dW) (/VdW + WdWW-H) = 1 + HH where r|,, is the aggregate hours elasticity of demand (or all workers. HivH < " implies that rj^ < - I . Note that r\,i = r\,r + !!«/,.. where r]/.- is the elasticity of employment numbers and TlH/f is the elasticity of hours per worker. The results in Table I suggest r|£ = - . I and r|H = -1.1. so TIH/^ is approximately -1.0. 'The 460 C(N)perating finns were distributed across industries and urban and rural counties in roughly equal proportions to the distribution In the universe of firms. ^Predicted subminimum and superminimum status was compared to the actual hourly wage rate for the subset of workers for whom data was supplied by our surveyed firms. These predictions were correctly classified 79 percent of the time, aggregating over all 6 quarters. The wage equation generated smaller subminimum groups (by 11 percent) and larger superminimum groups {by 4 percent) than did the reported wages. Appendix Table A2 summarizes the actual versus predicted sub- and superminimum group.s by quarter. ''These results are not due solely to ihe use of wage structures. As shown in Orazem and Mattila (1995). covered sector employment shares for teenagers fell sharply in Iowa over the period while increasing lor those aged 20-24. "^ See Orazem and Mattila (1995), Table 1.11. '"The proportions of workers paid below the current minimum based on reported wages (top number) or predicted wages (bottom number in parentheses) are reported lor the first quarters of 1990 and 1992. Generally, uncovered or rural hrms arc more likely than covered or urban firms tn pay wages below the minimum. However, there are large numbers of employees paid below the minimum wage in all sectors. Urban 1990:1 1990:2
Covered 3.5 (33.1) 5.8 (23.1)
Rural Uncovered 5.5 (24.7) 6.7 (48.8)
Covered 4.9 (39.9) 4.6 (30.4)
Uncovered 15.5 (53.5) 17.0 (61.6)
'' Ashenfelter and Card (1982) found that quarterly wage series were well represented by an AR( I) process wilh a coeflicient insignificantly different from I. Analysis which relaxed this restriction by incorporating
20
JOURNAL OF LABOR RESEARCH
predicted changes in superminimum wages in response to the minimum wage yielded results similar to those reported in this study. '-We considered the possibility that superminimum wages mighl have ripple effects when the minitnum wage rises. To allow for possible impacts on demand we experimented by including changes in both subminimum and superminimum wages in equation (3). The results yielded nearly identical demand elasticities as those in Table 3 (Orazem and Mattila, 1995). " Summary statistics for the ptwied one- and four-quarter underiying data are: £* = subminimum employed per firm E^W^ = subminimum aggregate eamings per firm MW = minimum wage W^ = predicted hourly wage rate In(AfW/H'*) = log of ratio R = rural C = covered
firm firm
AZ.* = change in subminimum employment share AS* = change in subminitnum earnings share
8.95 $24,567
15.9 $50,580
$4.03 $5.37
$.498 $ 1.34
-39* 514
.548 .500
.^2
.499
-.(X)5
.119
-.0004
,108
In addition to these variables, we also entered a sample selection control variable in the regression equations of Table .1. However, this correction, based on exogenous variables such as floodini; that ttwik place at time of the survey, had no significant effect. Full discussion and results are available in Orazem and Mattila (1995). '•"We do not have capital data to control directly for such adjustments. An alternative approach, used by Neumark. Schweitzer, and Wascher (2(HX)). is to attempt to use lagged changes in wage rates lo capture any lagged impact resulting from substitution of labor saving inputs, .Although this is a reasonable approach in principle, we are concerned that the lags will capture a variety of extraneous phenomenon, not just the impact of minimum wages. Sorting out the.se effects would require a longer time series than our two-year sample period. '*See Table 3.9 in Hamermesh (1993) for a summary of results of demand studies disaggregated by age. Elasticities for teenage workers are generally above .5 in absolute value with some estimates well into the elastic range. '^Linneman (1982) also found that eamings for subminimum workers fell in response to minimum wage increases.
PETER K ORAZEM and J. PETER MATTILA
21
REFERENCES Ashenfelter. Orley and David Card. 'Time Series Representations of Economic Variables and Alternative Models of the Labour Market." Review of Economic Studies 59 (Supplemenl, 1982): 761-82. Bellanie. Don and Gabriel Picone. "Fast Food and Unnatural Experiments: Another Perspective on the New Jersey Minimum Wage." Journal of Labor Reseanrh 20 (Fall 1999): 463-77. Brown. Charles. Curtis Gilroy. and Andrew Kohen. "The Effect of the Minimum Wage on Employment and Unemployment." Journal of Economic Ulermure 20 (June 1982): 487-528. Btirkhauser. Richard V., Kenneth A. Couch, and David C. Wittenburg. "A Reassessment of the New Economics of the Minimum Wage Literature with Monthly Data from the CPS." Journal of Labor Economics 18 (October 2000): 653-80. Card, David and Alan B. Krueger. Myth and Measurement: The New Economics of the Minimum Wage. Princeton. N.J.: Princeton University Press, 1995. Frey, Bruno, Werner Pommerehne. FHedrich Schneider, and Guy Gilbert. "Consensus and Dissension Among Economists: An Empirical Inqairy." American Economic Review 74 (December 1984): 986-94. Hamermesh, Daniel S. "The Demand for Labor in the Long Run." In D. Ashenfeiter and R. Layard. eds. Handbook of Labor Economics. Vol. I. Amslerdam: Norlh Holland, 1986. . Labor Demand. Princeton, N.J.: Princeton University Press. 1993. Kat/. Lawrence F. and Alan B. Krueger. "The Effect of the Minimum Wage on the Fast Food Industry." Industrial and Uibor und Relations Review 46 (October 1992): 6-21. Linneman, Peter. "The Economic Impacts of Minimum Wage Laws." Journal off^litical Economy 90 (June 1982): 443-69. Mincer. Jacob. "Unemploymenl Effects of Minimum Wage Changes." Journal of Political Economy 84 (August 1976):S87-SI04. Ncumark. David. Mark Schweii/er. and William Wascher. "The Effects of Minimum Wages Throughoui ihc Wage Dhlnhutiim" NBER Workini- Paper Series 75\9 (Fcbniary 2000). Orazem. Peter F. and J. Peter Mattila. "The Impact of Minimum Wages on Rural Retail and Service Sector I..abor Markets." December. 1995. Iowa Stale University, mimeo. Partridge. Mark D. and Jamie S. Partridge. "Do Minimum Wage Hikes Reduce Employment? State-level Evidence from the Low-Wage Retail Sector." Journal of Labor Research 20 (Summer 1999): 393-413. "Review Symposium: Myth and Measuremenc Tlie New Economics of the Minimum Wage," by David Card and Alan B. Krueger. Industrial and Labor Relations Review 48 (July 1995): 827-49. "Symposium: Minimum Wages. Entry-level Employment and Employees and the Transition from Welfare to Work." Jotirnal of Labor Research 20 (Fall 1999): 443-537. U.S. Bureau of Labor Statistics. Employment and Ettmings. Washington, D.C.: U.S. Government Printing Office, various years. U.S. Dcpariincnl of Commerce. 5wrm/f«M/i.«ra(7n///»^ United States, /9P5. Washington. D.C.: U.S. Governmenl Printing Oflice, 1995.
22
JOURNAL OF LABOR RESEARCH
Appendix Table A1 Log Hourly Wage Rate Equations I989;4 Inlercept
-1.71 (1.61)
F (FEMALE)
.51 (3.20)"
A (AGE)
.07 (11.3)"
1990:1 -.24 (.23)
1990:4
1991:1
1991:4
i992:1
-1.43 (1.35)
-1.48 (1.40)
-.17 (.18)
-.10 (.11)
.27 (1.70)
.38 (2.62)"
.24 (1.57)
.23 (1.70)
.17 (1.15)
.05 (8.78)"
.06 (9.85)"
.06 (9.65)"
.05 (9.41)"
.05 (8.05)"
A^
-.001 (9.75)"
-.001 (7.38)"
-.001 (8.61)"
-.001 (8.54)"
-.001 (8.24)"
-.0005 (6.91)"
A* F
-.037 (4.33)"
-.022 (2.60)"
-.028 (3.60)"
-.022 (2.65)"
-.019 (2.62)"
-.016 (2.03)"
A^* F
.0004 (3.65)"
.0002 (2.05)*
.0003 (3.06J"
.0002 (2.30)'
.0002 (2.03)"
.(XX) 1 (1.49)
NEMP (SIZE)
-.002 (2.04)"
-.002 (2.03)'
-.001 (1.05)
-.0002 (.30)
-.0007 (1.11)
-.0002 (.32)
C (COVERED)
.125 (3.70)"
.142 (4.60)"
.098 (3.29)"
.119 (4.06)"
.107 (3.68)"
.161 (5.71)"
SlCs Included
•
•
•
•
•
•
% Rural
-.37 (3.38)"
-.45 (4.15)"
-.28 (2.53)"
-.36 (3.24)"
-.35 (3.28)"
-.35 (3.24)"
Per Capital Income
-.0001 (3.67)'*
-.0001 (4.53)"
-.0001 (1.88)
-.0001 (2.99)"
-.0001 (1.86)
-.0001 (3.84)"
% H.S. Grads
-.008 (.61)
-.011 (.86)
.005 (.40)
.010 (.76)
-.002 (.17)
-.004 (.29)
% College Grads
.004 (.50)
.009 (1.29)
-.001 (.21)
-.003 (.42)
.001 (.13)
.004 (.66)
8.80 (4.63)"
6.57 (3.61)"
5.28 (2.87)"
5.38 (2.97)"
3.87 (2.29)'
5.02 (3.06)"
% FEM in LF.
X
.05 (.46)
.13 (1.28)
-.01 (.08)
-.02 (.20)
.06 (.55)
.12 (1.22)
R^
.508
.492
.445
.460
.410
.449
N
733
713
788
776
867
756
Notes: f-ratios in parentheses; * (**) significant ai 5% (\%) level.
PETFR F ORAZEM and J. PETER MATTILA
23
Appendix Table A2 Comparison of Predicted and Actual Hourly Wage Rates Correctly Predicted Below MW
Incorrectly Predicted to be Below MW Above MW
Correctly Predicted Above MW
19«9:4 95 (60.5%)
77
62
505 (86.8%)
29 (64.4%)
118
16
476 (80.1%)
85 (42.7%)
54
114
519 (90.6%)
19 (31.1%)
46
42
573 (92.6%)
114 (57.6%)
87
106
534 (86.0%)
Number (% correct)
45 (75.0%)
178
15
417 (54.0%)
560
{% correct)
Number (% correct) 1990:1 Number (% correct) 1990:4 Number (% correct) 1991:1 Number (% correct) 1991:4 Number (% correct) 1992:1
ratal
438 (71.1%)
355
3,045 (84.5%)
