Safe Port in a Storm: The Impact of Labor Market Conditions on ...

Download PDF
11 downloads 78657 Views 78KB Size Report
Comment
the unemployment rates of recent high school graduates and of all adults are ... For technical reasons this postscript file does not contain the two figures .... 2. The links between two-year public colleges and the business cycle are direct and.
93-12R

UNIVERSITY OF CALIFORNIA, SAN DIEGO DEPARTMENT OF ECONOMICS SAFE PORT IN A STORM: THE IMPACT OF LABOR MARKET CONDITIONS ON COMMUNITY COLLEGE ENROLLMENTS BY JULIAN R. BETTS AND LAUREL L. MCFARLAND

DISCUSSION PAPER 93-12R JANUARY 1995

Safe Port in a Storm: The Impact of Labor Market Conditions on Community College Enrollments

by Julian R. Betts, Department of Economics, University of California, San Diego and Laurel L. McFarland, The Brookings Institution, Washington, D.C.

This article is forthcoming in the Journal of Human Resources.

This article is based on a paper prepared for the "Is There a National Policy Interest in Subbaccalaureate Education?" session at the Association for Public Policy Analysis and Management research conference, October 31, 1992. We would like to thank Charles Hornbrook and Marc Rysman at the Brookings Institution and Shinichi Sakata, Randy Silvers, and Klaus Zauner at UCSD for research assistance. The following people at the U.S. Dept. of Education and the National Science Foundation have provided valuable assistance: Kristin Keogh and Fabrizio Galino. We would also like to thank Henry Aaron, George Borjas, Enid Jones, Joe Altonji, Jane Sjogren, two anonymous referees and participants in seminars at UCSD, RAND and the 1993 meetings of the Western Economic Association.

ABSTRACT The paper examines the impact of the business cycle on enrollments and finances at individual community colleges between the late 1960’s and the mid 1980’s. We find that 1 percent increases in the unemployment rates of recent high school graduates and of all adults are associated with rises in full-time attendance of about 0.5 percent and 4 percent respectively. Part-time enrollment exhibits similar anticyclical patterns. This link carries over in large part to degrees obtained. In contrast, state and local appropriations per student are procyclical. We interpret this funding pattern as a failure to integrate education policy sufficiently closely with labor-market policy.

Note

For technical reasons this postscript file does not contain the two figures which accompany this paper. If you would like to receive a copy of these, please send e-mail to [email protected] or phone Julian Betts at (619) 534-3369. Mail address: Dept. of Economics, UCSD, La Jolla, CA 92093-0508. Our apologies for any inconvenience this may have caused you.

I. Introduction Recessions drive people into community colleges, as community college administrators have long observed. Newly unemployed people enter community college to retrain for occupations less buffeted by unemployment, while some workers who are still in jobs may see further education as a vaccine against unemployment. Furthermore, loss of income may force individuals to enroll in lower cost community colleges, when in better times they might have attended private or public universities. Faced with the shifts in demand that follow economic change, community colleges have generally been admired for their quick response and flexibility. But for all of the awareness on the ground level that community colleges are affected by the business cycle and by shifting demands for skills, public policies towards higher education have not been shaped by labor market concerns. Instead, policy towards community colleges, particularly at the Federal level, has been part of an overall higher education policy that has emphasized equity and access. And for their part, community colleges have become the principal institution of access to higher education in this country. They provide entry at low cost for all kinds of people who might not otherwise have such opportunities. They also enroll large numbers of people: more than half of all first-time freshmen are at two-year colleges, and of all undergraduates, more than 4 in 10 attend community colleges.1 The attention to access, however socially vital and important, may obscure the vital economic dimension to community colleges. In a strict economic sense community colleges contribute middle level students to the middle level labor market and affect the productivity of our workforce. To be effective labor market intermediaries, community colleges must respond to economic change and labor market fluctuations. Therefore, a key question is: Are

community college enrollments closely linked to the unemployment rate? Is there a weak or strong link? How quickly does adjustment occur? Do community colleges act as "automatic stabilizers" to absorb and then retrain those workers displaced by recession and restructuring? Previous studies of enrollments and economic conditions have focused on determinants of enrollment demand and state-level enrollment forecasting. Curiously, for all the anecdotal evidence to the contrary, several empirical studies of demand for higher education suggest a rather weak relationship between unemployment and enrollments. Looking at four-year colleges, Manski and Wise, for example, find that "...the local unemployment rate is not significantly related to [four-year] college application. Thus, our findings weakly support the presumption that there is some interaction between local labor market opportunities and the continuation of schooling..." (Manski and Wise 1983, p.69). They find similar results for attendance (pp. 139-41). With respect to two-year colleges, little is known empirically about the role of unemployment and the business cycle in determining enrollments. Some academic work has been done on the general determinants of demand for two-year college, but most of it has concentrated on the effect of student aid on attendance.2 Grubb (1988) finds that enrollments at two-year public colleges are not especially sensitive to labor market conditions. In contrast, Lehr and Newman (1978) find that both two-year and four-year enrollments in Oregon are sensitive to economic conditions. But their cross-sectional analysis does not include unemployment rates, while their time series equation should be treated as no more than suggestive due to the study’s low number of observations.

2

A few state governments have also contributed to the literature on enrollments and economic conditions. States need reasonable forecasts of student numbers for planning and budgeting purposes and a few state higher education authorities have included cyclical economic variables in their models. For example, in research done by the state of California, it was found that community college enrollment has an "unemployment elasticity" of 0.2 to 0.4, in the sense that a 10 percent rise in unemployment leads to a 2 to 4 percent increase in enrollment by individuals seeking retraining (California Community Colleges Chancellor’s Office 1987). Unfortunately, most states do not consider the effects of unemployment explicitly in their education planning and forecasting. Seldom, if ever, do they analyze the long-term influence of the business cycle on enrollments and funding. And, to the detriment of national policy discussions, the few existing state analyses have not been collected and compared. Though most of the literature on enrollment demand emphasizes four-year enrollments, it does offer some useful models and approaches for examining two-year enrollments.3 When modeling enrollment at two-year colleges, though, the relative importance of variables may not be the same, and new variables may also be required. Traditional human capital models posit that three main factors influence postsecondary enrollment demand: rate of return to postsecondary education, cost of the education, and income of the student’s family. Much of the economic literature on higher education has portrayed the adjustment process of education-labor market imbalances as a "cobweb"4; would-be students observe the labor market conditions and enroll accordingly. Four years or more after enrollment, the supply of graduates in high-demand areas surges and drives down wages in the high demand field.

3

Conversely, the supply of graduates falls in occupations experiencing high unemployment. (More recent work has argued that students form rational expectations of future wages, and has shown that such models explain wage volatility about as well as cobweb models (Zarkin 1983; Siow 1984).) We will argue that four-year college demand models are of limited use for analyzing the link between unemployment and community college attendance. Not only are different variables more important to outcomes, but there is a fundamental difference in the nature of would-be community college students’ enrollment decisions. They are typically not just out of high school, are already in the labor force, have higher discount rates5, have lower savings, and little or no access to their parents’ assets (U.S. Department of Education 1992b, Table II.5). A recession hits this group harder than "traditional" college students. We begin our analysis by stating these working assumptions: i) student enrollments respond to labor-market returns to education and to the costs of that education. ii) Recession influences enrollments through business cycle effects (discussed above), and by changing the returns to education. That leads us to the principal hypotheses: 1. Community college attendance is countercyclical-- enrollments rise when unemployment rises, and fall when unemployment falls. 2. The links between two-year public colleges and the business cycle are direct and immediate.

4

II. The Model We consider a simple income maximization model of school choice. Suppose that a cohort of eighteen year old workers indexed by i has just graduated from high school in period 0. Each worker has three educational paths available: to begin work without undertaking any postsecondary education (ED=12), to attend a community college for two years and then work (ED=14), or to attend a four-year college for four-years and then start work (ED=16).6

7

The worker has T periods in which to acquire further education and

work, and discounts future net income at the rate r. Each worker chooses the educational path with the highest expected present value PV: (1)

maxPV = E(Return)

where E denotes expectations in period 0, and the Return is given by:  T(W ) if ED = 12  Φ t=1 it,12   T(W ) - Φ 2(Cost ) if ED = 14 (2)  Φ t=3 it,14 it,14 t=1   T(W ) - Φ 4(Cost ) if ED = 16  Φ t=5 it,16 it,16 t=1  where Φ is an operator which converts income t periods ahead, xt, into current dollars: (3)

Φxt = xt/(1+r)t,

the Wit,ED terms refer to earnings t periods into the future for person i with an education level given by ED and Costit,ED, the cost of education for such a person, equals fees, Feeit,ED, less scholarships, Aidit,ED.

5

Workers i differ in ability ai, which has continuous probability density. More able students have a higher economic return to education. Denoting population averages for variables by a horizontal bar, this model leads to the following comparative statics results: Result One Enrollment in community colleges falls with a rise in the expected value of any of the following: i) Wt,12 for any t ∈ (1,2,...,T) ii) W t,16 for any t ∈ (5,6,...,T) iii) Aidt,16 for any t ∈ (1,2,...,4) iv) Feet,14 for any t ∈ (1,2) Result Two Enrollment in community colleges rises with a rise in the expected value of any of the following: i) Wt,14 for any t ∈ (3,4,...,T) ii) Aidt,14 for any t ∈ (1,2) iii) Feet,16 for any t ∈ (1,2,...,4) In words, enrollment responds to changes in the expected returns to and costs of education in each of the three educational paths.8 An important extension to the model arises if some students enroll in community college with the intention of transferring to a four-year college at a later date. In such a case, a rise in the returns to four-year education will change enrollment in community colleges in an ambiguous manner, since workers previously indifferent between ED=14 and ED=16 will now choose ED=16, while workers who previously had been indifferent between ED=12 and ED=14 may now opt for community college, with the intention of transferring to a four-year

6

college at a later date. The overall impact of a change in the wages of graduates of four-year colleges, then, is ambiguous in theory. Similar arguments apply to the other four-year variables, and are noted below.9

Empirical Implementation Note that the EΦWit,ED terms refer to expected discounted annual earnings for a worker with the stated level of education at age 18 + t. This will depend positively on expected wages conditional upon employment and negatively upon the expected unemployment rate for workers with the given level of education. Therefore, we proxy these terms using the present discounted value of earnings of employed workers of the given educational level and the unemployment rate of workers of the given education level upon graduation. (We avoid using average unemployment rates for workers of all ages with the given level of education because demographic shifts are likely to have altered this average rate over time.) We use cross-sectional data from the Current Population Survey to estimate the present value of wages. But the earnings of older workers today may not reflect what young workers will earn in the future. So we use both the cross-sectional present value and the starting wages for each educational path to proxy for expected present value of earnings of a young worker. The starting wage serves as a shift variable which may capture technological and other trends which affect mainly younger generations of workers.10 A second issue concerns the age distribution of students at community colleges. The above model assumes that all enrollees are eighteen year old high school graduates. For older enrollees, the relevant proxies for the returns to each path of education will differ. While it is

7

impractical to include estimates of these returns for enrollees of every age between 18 and 65, the inclusion of the present discounted values (PDV’s) of earnings for each path should proxy reasonably well for the expected returns for an older student, since the PDV for a worker of age x contains (65 - x) terms in common with the PDV for an eighteen-year old. To control for the impact of unemployment rates faced by older workers, we add the overall unemployment rate for workers aged 26-65 in several regressions. The age of entry to college is particularly important for the distinction between fulltime and part-time enrollment. In 1987, 60.6 percent of full-time students at community colleges were 21 or under, compared to only 21.3 percent for part-time enrollees.11 We would thus expect part-time enrollment to respond more to the adult unemployment rate than to, for instance, the unemployment rate for recent high school graduates. A third thorny issue involves the treatment of expectations. We will use current earnings as a proxy for expected earnings, on the grounds that the labor market for broadly defined groups of workers (e.g. workers with education equal to 14 years) should adjust quite quickly to shocks. Such an assumption is less appealing for students’ expectations of college fees and aid. Since these prices are usually set by administrators rather than directly by market forces, it is likely that students can do much better than using just the current average fee and aid level to predict future levels. We therefore proxy workers’ expectations of oneperiod-ahead fees per full-time equivalent student (FTE) at each community college. We use the predicted value from a regression of the future value of fees on the college’s deficit per FTE and the most recent one period change in fees. This variable, in addition to current fees, which already appears in the enrollment regression (but not in the forecast equation for future

8

fees), proxies for the workers’ rational expectations of future fees. We use analogous proxies for expected aid per FTE at each community college, and for average cost at all regional four-year colleges. The above model ignores changes in demography, which can affect enrollment if demographic groups differ in their probability of attending community college. We therefore add YOUNG, the share of people aged 18 to 25 in the population aged 18 to 65, as well as the percent of the adult population which is black. We also include income per capita to test whether the income elasticity of enrollment is significantly different from zero. The other reason for including income per capita is to provide a proxy for ability to pay for college. This allows us to separate the effects of the business cycle into two components: the impact of incomes on ability to finance college, and the impact of unemployment rates on the opportunity costs of education. Finally, we add measures of the population aged 18 to 65 divided by the number of two-year and four-year colleges in the census region as a proxy for population density, on the presumption that a larger potential student body will increase enrollment at a given college. This leaves us with the empirical model shown below. Signs underneath each variable indicate predicted signs of coefficients. The dependent variable, ENRirt, is enrollment at community college i, which is in census region r, at time t. Fees, aid, and expected fees and aid at the individual community college i, are denoted by FEE/FTEirt, AID/FTEirt, E(FEE/FTE)ir,t+1, and E(AID/FTE)ir,t+1 respectively. Note that we measure all other variables at the level of the census region. The UR terms denote unemployment rates for people of the stated age and education, e.g. UR1812 is the rate for 18 year-olds with 12 years

9

of education, the PDV terms measure the present discounted value, from the point of view of a worker of age 18, of earnings of workers with the stated level of education, the WW terms equal weekly wages for workers of the stated age and education. Finally, letting COST4rt refer to average net cost of enrollment per full-time equivalent enrollee at four-year colleges in region r at time t, we have:

(4) ENRirt = f{UR1812rt, UR2014rt, UR2216rt, PDV12rt, PDV14rt, PDV16rt, WW1812rt, +

-

+/-

-

+

+/-

-

WW2014rt, WW2216rt, FEE/FTEirt, E(FEE/FTE)ir,t+1, AID/FTEirt, +

+/-

-

-

+

E(AID/FTE)irt, COST4rt, E(COST4)r,t+1, POP/(# 2-YR COLLEGES)rt, +

+/-

+/-

+

POP/(# 4-YR COLLEGES)rt, YOUNGrt, %BLACKrt, INCOME PER CAPITArt, YEAR} +/-

+

+/-

+/-

+/-

In order to reduce simultaneity bias between enrollment and current year fees and aid, all current-year college variables used as regressors are instrumented by regressing on a constant and the previous year’s value. Similarly, the one-step ahead forecasts are based on regressions using instrumented financial costs.12 We reiterate that the dependent variable,

10

enrollment, and the fee and aid variables refer to the individual community college. All other variables, including the cost of attending a four-year college, are calculated at the level of the nine census regions. In the analysis below, all variables except the unemployment rates and the proportion of the population which is young are in logs.

III. Data: Sources and Choice of Variables

The Educational Data Set: We drew institutional enrollment, degree attainment, and financial data from the U.S. Department of Education’s Higher Education General Information Survey/ Integrated Postsecondary Education Data System ("HEGIS"/"IPEDS") data system. This data set is an annual census of postsecondary institutions. It is based on the survey year’s October 15 enrollment figures, the previous school year’s record of degrees awarded, and the institution’s most recent fiscal year financial statistics. The two-year public college dataset contained 884 institutions, though our final dataset was slightly smaller because we removed military institutions, colleges in American territories, Federally-funded tribal colleges, and extremely small institutions. We also edited the data, in a procedure outlined in an appendix available from the authors, because of errors in the government’s data system. From our edited HEGIS/IPEDS data base, we drew full-time and part-time enrollment and degrees-awarded. We also used the fee and aid variables in the IPEDS data set, though we recognize there are limitations to their use.13

11

The Labor Market Data Set All labor-market data were calculated from the March Supplements of the Current Population Survey. Because community college administrators frequently list responsiveness to local labor-market conditions as a primary goal of the community college system, we calculated all labor-market variables separately for each of the nine census regions used by the Bureau of the Census. (A lack of degrees of freedom prevents reliable estimates at less aggregated levels, such as by state.) Unless explicitly mentioned, all labor-market variables were calculated based on the adult population aged 18 to 65 in the year preceding the March survey. Data for each individual are weighted by the March Supplement weight to obtain estimates which reflect the population average. The estimates of present discounted value of earnings (PDV) for a given level of education were calculated from the point of view of an 18-year old using estimated earnings for each age up to 65. OLS estimation was used in order to allow interpolation for the few age-education-geographical region cells for which there were no observations in a given year. Weekly earnings at each age were thus estimated by fitting them to a quartic function of potential experience (age - years of schooling completed - 6) for a given education group. This was used instead of the standard quadratic specification in light of the findings of Murphy and Welch (1990). A discount rate of 10 percent was used in the calculations, based on the work of Lawrance (1991). While she estimates discount rates which are often slightly higher, using 15 percent for the discount rate had little effect on the regressions.

12

The regressions below typically are based on data for the years 1969-85, although certain regressions involve longer time spans. The variation arises because of differences in the time span for which various HEGIS/IPEDS variables are available.

IV. Results Figure One depicts the relationship between the unemployment rate and attendance rate at community colleges for the aggregate national data, while Figure Two presents similar graphs for the New England region.14

15

(We postpone discussion of the second graph in

the Figures, which examines cycles in public subsidies of community colleges, to the penultimate section of the paper, which deals with policy implications.) The relationship is clear. Community college enrollments rise and fall remarkably in phase with the ups and downs of unemployment. There are some regional variations, but overall there is a notable consistency in the unemployment/attendance relationship across the country. Taking New England as an example (Figure Two), the unemployment spikes in 1971, 1975, and 1982 are matched by surges in community college attendance. Similarly, both attendance and unemployment declined in 1972, 1976, and 1984. The graph of the national data shows a similar relationship. (The correlation coefficient between the adult employment rate and the attendance rate at the national level is 0.820.) We now estimate the model developed earlier, to determine whether the unemployment rate:enrollment relation continues to hold once we control for other determinants of enrollment.

13

Econometric Results Given that we use panel data to model the absolute level of enrollment at each community college, ordinary least squares is likely to give biased standard errors and perhaps biased coefficients due to college-specific heterogeneity. Thus we would expect a fixed effects formulation, which allows for college-specific intercepts, or a random effects formulation, which treats college-specific effects as part of the error term, to outperform OLS. We therefore initially estimated all equations in this section using OLS, fixed effects and random effects methods. However, the OLS formulation was always rejected by F tests for a common intercept between colleges, with probability values < 0.00005. Similarly, the random effects formulation was always rejected when we performed Hausman tests, again with p-values of < 0.00005.16 Thus all regressions in levels presented below incorporate fixed effects for each college. In addition, since we measure the key regressor, the unemployment rate, at the regional level, we risk overstating the size of the t-statistics. We therefore specify an error term consisting of two parts: a white noise component and a time-varying error component specific to each combination of region and year. The estimation technique thus involves estimating a random effects model (with an error component related to the region and year) by GLS on data which has been mean differenced at the college level to control for collegespecific intercepts.17 Since our main goal is to study the link between enrollment and unemployment rates, we start by examining simple regressions before estimating the more complex model we developed above.

14

The first regression in Table One shows that the log of full time enrollment is significantly related to the overall unemployment rate of workers aged 18 to 65. The coefficient suggests that a 1 percent rise in the adult unemployment rate is associated with a rise in full-time enrollment of 0.0846 log points, or 8.8 percent. In the second regression, we check whether the significance of the unemployment variable in part proxies for ability to pay for college, by adding income per capita to the regression, and find that the coefficient on unemployment remains highly significant. Column Three repeats this regression using statelevel data on unemployment rates and income per capita, obtained from the Dept. of Labor and the Bureau of Economic Analysis respectively. The results are quite similar. The coefficient on unemployment is smaller than in the regional regression, which may reflect a greater degree of measurement error at the state level.18 Regressions Four through Six introduce the three unemployment rates we will use in the main model. UR1812, the unemployment rate among workers of age 18 with high school diplomas, is far and away the most significant explanatory factor of the three. Its significance remains high after including both income per capita and a time trend. This pattern continues to hold in the full model to be presented below. Table Two presents estimates of the model developed in the last section. The specification in regression #1 shows that unemployment rates, especially among high school graduates, remain highly significant factors in modeling full time enrollment even when controlling for many other possible influences. Higher starting wages of high school graduates reduce enrollment as expected. The fact that the present discounted earnings of high school graduates are positively related to

15

enrollment may reflect the fact that this variable in part proxies for family income, which is likely to be a positive determinant of postsecondary enrollment. We encountered some difficulty in obtaining a well identified model of enrollment at two-year colleges in a model which features both two-year and four-year college influences. The coefficients on the college financial variables are thus less robust than those on the labormarket variables. Higher fees do appear to deter enrollment, although the effect of aid is imprecisely measured and in general has an unexpected sign. Measurement error may explain this result as we found that the scholarships variable required substantially more cleaning than other variables. In most of the regressions, the expected cost of four-year college (as well as the sum of current costs and expected cost) is a positive determinant of two-year college enrollment, suggesting that community colleges act as substitutes for four-year colleges. Signs on the wage and discounted earnings variables for workers with 16 years of education support this inference. Note that a rise in the returns to four-year college, whether measured by unemployment rates, starting wage, or present discounted value, is not associated with a rise in enrollment at community colleges. Thus the data reveal no evidence that the average student is attracted into community college in order to take courses which he or she intends to later apply toward a four-year college degree. This finding squares with Grubb’s aforementioned finding that only about 20 percent of students in the High School and Beyond survey who had entered community college transferred to four-year colleges within four years. Regressions 2-4 test restrictions on regression 1. In general, the restrictions do not change the results on the remaining variables much. The unemployment rate coefficients

16

prove particularly robust. The bottom of the table presents F tests of the restrictions with regression 1 as the alternative. The hypothesis that the expected fee and aid variables do not belong in the model is strongly rejected. (See regression 2.) In regression 4, exclusion restrictions on the market shares are also strongly rejected.

19

As mentioned earlier, we wish to proxy the labor-market conditions faced by older workers, since many college students are more than 25 years old. While the PDV terms should proxy fairly well for the prospects faced by older workers, this is not the case for the measures of the unemployment rate currently in the model. We therefore add the unemployment rate for adults aged 26 to 65 to regression 1. The results, in regression 5, show that the adjusted R2 rises by over 10 percent. Given the collinearity between this and the other three unemployment rates, it is not surprising that the coefficients on these latter rates fall. Still, if unemployment were to rise by 1 percent for all measures of unemployment, regression 5 suggests a rise in full-time enrollment of 0.045 log points, or 4.6 percent, significantly above the 1.4 elasticity suggested by regression 1.20 In a similar vein, note that YOUNG, the share of young workers in the population, is in almost all cases statistically insignificant, suggesting that propensities to enroll do not drop off precipitously with age. (In 1987, students aged 30 and over accounted for 36.1 percent of full-time enrollment at public two-year institutions. (U.S. Dept. of Education, 1991, p. 171))21 In general, the labor-market variables for people of age 20 with 14 years of education are only marginally significant. Indeed, F-tests in regression #5 Table Two for the exclusion of the three variables related to those with 14 years of education show no evidence against

17

the null. Measurement error could explain the low t-statistics, because this group of workers consists not only of recent community college graduates, but also dropouts from four-year colleges.

22 23

Additional restrictions on regression #5, Table 2, that were tested and

rejected with p-values less than 0.01 include: a) the coefficients on the weekly wage (ww) terms are jointly zero. b) the coefficients on the present discounted value (pdv) terms are jointly zero. c) the coefficients on the four-year college variables are jointly zero. Thus far in the econometric analysis we have focused exclusively on full-time enrollment. Table 3 displays the results when the log of part-time enrollment was regressed on the variables used in regressions 1 and 5 in Table 2. For comparative purposes, the first column replicates the results when log full-time enrollment is the dependent variable. The table makes clear the reasons for our reluctance to model ’full-time equivalent’ enrollment, i.e. a weighted average of full-time and part-time enrollment. Among other differences, part-time enrollment reacts quite differently to the unemployment rate terms, although the sum of the unemployment elasticities is still positive and high. Many of the differences may reflect the fact that part-time students tend to be older than full-time students, as was documented in the previous section of the paper. For instance, enrollment of part-time students appears less strongly linked to job opportunities for young high school graduates, and more strongly linked to labor-market conditions for young graduates of four-year colleges. In addition, the coefficient on the unemployment rate for older workers is higher in the regression for part-time enrollment. Part-time enrollment also appears to be much more sensitive to costs, and appears to be a normal good, in that a rise in

18

regional income per head is associated with a rise in part-time enrollment, while the opposite obtains for full-time enrollment. One possible explanation for both of these differences is that part-time students are not in general eligible for Pell grants, so that they may be more sensitive to costs, and to variations in income per head, which is a measure of ability to pay for college. There are some similarities between the regressions for full-time and part-time enrollment, though. In particular, there is no evidence of a ’transfer function’ for part-time students, based on the logic presented earlier for full-time enrollment. Given that enrollment rises during a recession, we would like to know whether those students whom recession induces to enroll eventually go on to receive degrees. If students are merely ’parking’ themselves in community college temporarily, the policy implications would be quite different. Table Four considers this issue in two ways. First, it shows that the elasticities of current enrollment and future certificates awarded with respect to the current unemployment rate are quite similar, implying that there is a fairly high completion rate among students. Second, equations #4 and #6 add interactions between enrollment at time t and changes in the unemployment rate between t and t + 1, and between t + 1 and t + 2. If students enrolled at time t intend to drop out as soon as labor-market conditions improve, then the coefficients on these interaction terms should be large and positive. In fact, they are small relative to the coefficient on initial enrollment. For instance, suppose that in both of the years after the year in which students enroll the unemployment rate drops by 1%. Then, in regression #4, the elasticity of certificates with respect to full-time enrollment two years earlier would drop by

19

0.01 times the sum of the coefficients on the two interaction terms. In other words, the elasticity would drop from 0.536 to 0.533. The effect is even smaller in equation #6 which uses full-time equivalent enrollment. Thus it appears that a subsequent drop in the unemployment rate lowers only slightly the probability of the student obtaining a certificate.24 Earlier, we hypothesized that adjustment of enrollment to changes in the unemployment rate is very rapid. We therefore studied the dynamics of log full-time freshman (LFTF) enrollment. We re-estimated regression 5 from Table Two, with LFTF as the dependent variable. Again, the unemployment rate for adults aged 26-65 was the most significant of the unemployment rate terms, with a coefficient of 4.31 (t=4.78). A test of this model against an identical model with the addition of the four unemployment rates lagged once led to retention of the first model (p-value=0.64). This result indicates that adjustment of freshman enrollment to the unemployment is very rapid.25 In summary, we conclude that the link between community college enrollment and the unemployment rate is significant both statistically and in terms of the absolute elasticity. This link appears in large part to carry over to degrees obtained. Freshman enrollment responds strongly to changes in the unemployment rate, adjusting within a year at the most.

V. Policy Discussion Community colleges are affected even more by the business cycle than our results on enrollments imply: while recessions lead to enrollment increases, they also lead to revenue decreases. Community colleges in most states are heavily dependent on state appropriations

20

for their revenues--and as state revenue falls in recession, many states have cut funding to community colleges. This pattern of public funding may limit the benefits available to those who enroll. Policymakers should consider whether labor-market policy, in particular unemployment benefits, could be linked to education policy. Such a link would provide resources to meet the strong counter-cyclical demand for community college education that our results have documented. The graphs of state and local appropriations in Figures One and Two demonstrate how attendance patterns and appropriations are at odds with each other. Appropriations fall just as enrollment demand surges.26

27

At the national level, appropriations per FTE dipped

in 1971, 1975, and 1980-82. The latest figures for 1989 through 1991 (not shown on Figures One and two) show precipitous declines in appropriations per FTE.28 This has undone some of the dramatic increase in appropriations per FTE in the mid-1980s. Appropriations per FTE rise in 1972, 1976, and 1983-85. In the recent recession of the early 1990s, public higher education has suffered even more significant cuts: in 17 states, state appropriations were lower in 1992-93 than two years previously, and a total of 36 states provided lower aid in real terms (Jaschik, 1992, A21). It is also clear that when recession reduces tax revenues, states and localities do not extend any special budgetary protection to community colleges. In addition, the recessionary effects are meted out to them fairly rapidly: for example, mid-year state budget cuts to community colleges were made in 23 states in Fiscal 1991 (Higher Education and National Affairs, 1991).

21

Recession puts fiscal stress upon community colleges just when citizens demand their services the most, and the quality of education available may suffer.29 In most states, community colleges must embrace open admissions policies--all applicants must be accommodated--which means that, at the height of recession, the college cannot cap surging enrollments despite the drop in appropriations. Community colleges must either dilute the quality of education, raise tuition at the very institution designed to provide "access", or frustrate student progress by rationing classes. Many institutions in the short term react by reducing the number of course sections, which intensifies the ration-by-queue effects. The resulting increased class size and compression of curriculum can dilute the quality of education. It could also lengthen the time to graduation and increase the drop-out rate. A rationing effect is also apparent: in 1991 about 45,000 students in California community colleges withdrew between the second and fourth week of term, because they "didn’t obtain the courses they sought" (California Community Colleges Chancellor’s Office, 1992, Appendix B). So, while enrollment may initially surge because of the economy, the effects of the recession may include inferior educational services, poorer provision of training, and some students rationed from the system entirely. Indeed, our estimates of the sensitivity of enrollment to the unemployment rate may be understated, due to this rationing of potential students during recessions. Given that community college enrollment acts as a stabilizing market adjustment mechanism, these findings suggest that a more anticyclical pattern of public funding of community colleges is in order. Countercyclical funding sources such as unemployment compensation could be linked to community college funding. Income support for unemployed

22

individuals could be coordinated with institutional aid to expand training provision during high unemployment periods. These countercyclical funds would then fall automatically in boom times, when both states and individuals had more money to spend on higher education. Such investment in human capital at the bottom of the business cycle would also be made when the unemployed workers’ opportunity costs were lowest.

VI. Conclusion This paper investigates how community college enrollments respond to changing labor market conditions, in particular unemployment. Combining panel data on colleges from HEGIS/ IPEDS with labor market data from the CPS, we estimate equations for enrollment based on a simple income maximization model of school choice. Our results show that there is a powerful link between the education "market" and the labor market at the middle level. We find that 1 percent increases in the unemployment rates of recent high school graduates and of all adults are associated with rises in full-time attendance of about 0.5 percent and 4 percent respectively. This link appears to carry over in large part to degrees obtained. Our findings suggest that community college enrollments are very sensitive to unemployment and economic conditions. Moreover, the enrollment response is almost coincident with movements in unemployment. The high elasticity of enrollment with respect to unemployment implies that the government does not need to create programs to encourage people to attend community college in recessions for retraining. Hundreds of thousands of individuals are already reading labor market signals quickly, and signing up for courses.

23

The results indicate the size of the recession-induced demand for education and training that community colleges must struggle to meet. This pressure on community college resources is exacerbated by the tendency for state funding to stagnate or fall in recent recessions. These patterns of community college enrollments and funding suggest that education and labor market policy should be linked by providing countercyclical funding for community college training.

24

TABLE ONE Simple Fixed Effect Regressions of Log Full Time Enrollment on Unemployment Rates, by Census Region and State Regression #3 includes observations for 1970-89. All other regressions in the table cover the period 1967-89. All regressions in all tables include an error component for region/year or, in the case of regression #3 below, state/year. T-statistics appear in parentheses. Regression # Variable Adult unemp. rate

#1 8.4614 (20.19)

#2

#3

#4

#5

#6

7.6664 (22.63)

As above, state

3.4779 (15.23)

UR1812

2.9304 (10.57)

2.7798 (11.36)

1.4842 (6.51)

UR2014

1.0196 (3.43)

0.9894 (3.79)

0.8228 (3.79)

UR2216

1.4200 (5.16)

0.6332 (2.55)

0.6021 (2.92)

1.4236 (12.96)

0.0413 (0.28)

Income per capita

1.3511 (15.34)

As above, state

0.1699 (18.25)

YEAR

0.0181 (11.97)

R-Squared

0.0235

0.0444

0.0433

0.0150

0.0261

0.0393

Adjusted R-Squared

0.0115

0.0325

-0.0151

0.0027

0.0139

0.0272

Number of Colleges

818

818

817

818

818

818

Number of Observations

16988

16988

14651

16988

16988

16988

25

TABLE TWO Fixed Effect Estimates of Log Full Time Enrollment Model Parentheses enclose t-statistics. Regressions include observations for 1969 through 1985. Regression # Variable

#1

#2

#3

#4

#5

UR1812

1.0648 (5.62)

1.1092 (5.87)

1.1034 (5.74)

1.1526 (6.23)

0.5042 (2.66)

UR2014

0.1369 (0.84)

0.0400 (0.25)

0.1739 (1.05)

0.1679 (1.05)

-0.0545 (-0.36)

UR2216

0.1669 (0.95)

0.2200 (1.26)

0.1948 (1.09)

0.1386 (0.80)

-0.0223 (-0.14)

UR, Ages 26-65

4.0906 (6.74)

WW1812

-0.1815 (-3.39)

-0.1813 (-3.39)

-0.1616 (-2.99)

-0.1884 (-3.56)

-0.1512 (-3.05)

WW2014

-0.0018 (-0.05)

0.0002 (0.01)

-0.0186 (-0.56)

0.0080 (0.24)

0.0046 (0.15)

WW2216

-0.0661 (-1.76)

-0.0621 (-1.65)

-0.0817 (-2.15)

-0.0430 (-1.16)

-0.0137 (-0.39)

PDV12

1.7496 (6.96)

1.7783 (7.08)

1.2961 (6.10)

2.0248 (8.32)

1.4615 (6.18)

PDV14

0.1231 (0.89)

0.1193 (0.86)

0.0209 (0.15)

0.2195 (1.61)

0.0759 (0.59)

PDV16

-0.4996 (-3.70)

-0.5722 (-4.28)

-0.6542 (-5.02)

-0.4572 (-3.42)

-0.4941 (-3.92)

FEE/FTE

-0.1179 (-11.53)

-0.1188 (-12.45)

-0.1176 (-11.49)

-0.1173 (-11.45)

-0.1138 (-11.12)

E(FEE/FTE(+1))

0.0100 (0.34)

0.0188 (0.64)

-0.0153 (-0.54)

-0.0052 (-0.18)

AID/FTE

-0.0107 (-2.21)

-0.0115 (-2.37)

-0.0102 (-2.11)

-0.0109 (-2.25)

E(AID/FTE(+1))

-0.0094 (-0.66)

-0.0081 (-0.57)

-0.0092 (-0.65)

-0.0083 (-0.59)

0.0759 (0.91)

-0.0976 (-1.14)

-0.0025 (-0.03)

0.3092 (4.07)

0.2693 (3.64)

0.1971 (2.76)

-0.1291 (-0.20)

-1.5854 (-2.56)

-0.7309 (-1.21)

COST4R (4-YEAR COLLEGES) -0.0273 (-0.31) E(COST4R(+1))

0.3136 (4.18)

YOUNG

-0.6885 (-1.06)

-0.0121 (-2.72)

0.2164 (3.30)

-0.7228 (-1.12)

26

POP/(# 2-YR COLL)

0.1494 (1.95)

0.1631 (2.18)

0.1939 (2.55)

0.1804 (2.44)

POP/(# 4-YR COLL)

0.3744 (4.27)

0.3269 (3.91)

0.5105 (6.42)

0.1664 (1.87)

YEAR

0.0187 (3.45)

0.0180 (3.34)

%BLACK

0.0384 (1.14)

0.0312 (0.93)

INCOME PER CAPITA

-1.0893 (-4.87)

R-Squared

0.0369 (8.37)

0.0120 (2.32)

0.0459 (1.35)

0.0585 (1.77)

0.0965 (2.87)

-1.2339 (-5.58)

-0.6812 (-3.56)

-1.1440 (-5.42)

-0.4294 (-1.87)

0.0814

0.0795

0.0796

0.0779

0.0897

Adjusted R-Squared

0.0612

0.0596

0.0594

0.0578

0.0695

Number of Colleges

760

760

760

760

760

Number of Observations

8027

8027

8027

8027

8027

F Test of Model vs. Regression 1 above ) p-value:

0.00104

27

0.00000

TABLE THREE Model of Log of Part-Time Enrollment (LPT) Column #1 shows regression #1 from Table 2, with log full-time enrollment as the dependent variable, for comparative purposes. Regressions cover the period 1969-85. DEPENDENT VAR.:

LFT

LPT

LPT

UR1812

1.0648 (5.62)

-0.2867 (-0.96)

-0.8485 (-2.76)

UR2014

0.1369 (0.84)

0.4094 (1.61)

0.2016 (0.82)

UR2216

0.1669 (0.95)

0.8116 (2.95)

0.6127 (2.30)

Regressors:

UR, Ages 26-65

4.6408 (4.69)

WW1812

-0.1815 (-3.39)

-0.1905 (-2.28)

-0.1662 (-2.09)

WW2014

-0.0018 (-0.05)

-0.0529 (-1.02)

-0.0425 (-0.86)

WW2216

-0.0661 (-1.76)

-0.2211 (-3.76)

-0.1624 (-2.84)

PDV12

1.7496 (6.96)

1.3788 (3.44)

1.1293 (2.92)

PDV14

0.1231 (0.89)

0.5244 (2.42)

0.4779 (2.30)

PDV16

-0.4996 (-3.70)

-0.9049 (-4.22)

-0.9492 (-4.61)

FEE/FTE

-0.1179 (-11.53)

-0.2789 (-15.08)

-0.2735 (-14.78)

E(FEE/FTE(+1))

0.0100 (0.34)

0.1167 (2.20)

0.0971 (1.83)

AID/FTE

-0.0107 (-2.21)

-0.0296 (-3.37)

-0.0296 (-3.38)

E(AID/FTE(+1))

-0.0094 (-0.66)

0.0186 (0.72)

0.0209 (0.81)

COST4R

-0.0273 (-0.31)

-0.5558 (-3.90)

-0.5113 (-3.72)

E(COST4R(+1))

0.3136 (4.18)

-0.0226 (-0.19)

-0.1361 (-1.16)

YOUNG

-0.6885 (-1.06)

1.9698 (1.93)

1.8356 (1.87)

28

POP/(# 2-YR COLL)

0.1494 (1.95)

-0.0976 (-0.76)

-0.0392 (-0.31)

POP/(# 4-YR COLL)

0.3744 (4.27)

0.3675 (2.46)

0.1354 (0.88)

YEAR

0.0187 (3.45)

0.0891 (10.21)

0.0809 (9.46)

%BLACK

0.0384 (1.14)

-0.0855 (-1.50)

-0.0177 (-0.31)

INCOME PER CAPITA

-1.0893 (-4.87)

0.7576 (2.11)

1.5144 (4.00)

R-Squared

0.0814

0.3153

0.3299

Adjusted R-Squared

0.0612

0.3002

0.3150

Number of Colleges

760

760

760

Number of Observations

8027

8015

8015

29

TABLE FOUR Certificates Awarded Two Years into the Future as a Function of Current Enrollment and Changes in the Unemployment Rate Note: The dependent variable in the first regression is log of full-time enrollment (LFT), which is shown for comparative purposes. CERT = log of certificates granted by the college, and LFTE is the log of full-time equivalent enrollment. Regression #1 covers the period 1967-89. All other regressions cover the period 1967-86. Regression # #1 Dependent Var.

LFT

#2

#3

#4

#5

#6

CERT(+2) CERT(+2) CERT(+2) CERT(+2) CERT(+2)

Regressors: UR1812

3.7190 (14.76)

3.7258 (15.40)

LFT

2.3906 (11.05) 0.5511 (67.28)

1.1414 (6.09)

0.5361 (64.67)

(LFT)X∆UR1812

0.2166 (7.29)

(LFT)X∆UR1812(+1)

0.1299 (4.96)

LFTE

0.6038 (72.80)

0.5919 (68.98)

(LFTE)X∆UR1812

0.1416 (5.86)

(LFTE)X∆UR1812(+1)

0.0879 (4.19)

R-Squared

0.0127

0.0161

0.2404

0.2487

0.2708

0.2766

Adjusted R-Squared

0.0005

0.0038

0.2307

0.2390

0.2615

0.2672

Number of Colleges

818

817

816

816

816

816

Number of Observations

16988

14580

14273

14273

14273

14273

30

References

Altonji, Joseph. 1993. "The Demand for and Return to Education when Education Outcomes are Uncertain", Journal of Labor Economics, 11(1), Part 1: 48-83.

Berne, Robert. 1980. "Net Price Effects on Two-Year College Attendance Decisions", Journal of Education Finance, 5(4):391-414.

Betts, Julian R and Laurel L. McFarland. 1993. "Safe Port in a Storm: The Impact of Labor Market Conditions on Community College Enrollments", Discussion Paper 93-12, UCSD Department of Economics.

Breneman, David, and Susan Nelson. 1981. Financing Community Colleges: An Economic Perspective, Washington, D.C.: The Brookings Institution.

California Community Colleges, Chancellor’s Office, 1992. "Estimate of Fall 1991 Enrollment", Research and Analysis Memo 91-28". unpublished.

California Community Colleges, Chancellor’s Office, 1987. "Study of Fee Impact". unpublished.

31

Clotfelter, Charles, Ronald Ehrenberg, Malcolm Getz, and John Siegfried. 1991. Economic Challenges in Higher Education. Chicago, Ill.: The University of Chicago Press for the National Bureau of Economic Research.

Freeman, Richard B. 1971. The Market for College-Trained Manpower. Cambridge, Mass.: Harvard University Press.

______. 1976a. "A Cobweb Model of the Supply and Starting Salary of New Engineers", Industrial and Labor Relations Review, 29(1):236-248.

______. 1976b. The Overeducated American. New York: Academic Press.

Grubb, W. Norton. 1988. "Vocationalizing Higher Education: The Causes of Enrollment and Completion in Public Two-Year Colleges, 1970-1980", Economics of Education Review, 7(3):301-319.

Grubb, W. Norton. 1991. "The Decline of Community College Transfer Rates", Journal of Higher Education, 62 (2): 194-222.

Grubb, W. Norton. 1993. "The Varied Economic Returns to Postsecondary Education: New Evidence from the Class of 1972", Journal of Human Resources, 28(2): 365-382.

32

Hausman, J.A. 1978. "Specification Tests in Econometrics", Econometrica. 46(6): 1251-1271.

"Colleges Facing Severe Financial Problems". 1991. Higher Education and National Affairs, American Council on Education, 40(15): 1-3.

Hoenack, Stephen, and Eileen Collins. 1990. The Economics of American Universities: Management, Operations, and Fiscal Environment. Albany, New York: State University of New York Press.

Jaschik, Scott. 1992. "1% Decline in State Support for Colleges Thought to be First 2-Year Drop Ever", The Chronicle of Higher Education, 39(9):A21-A26.

Kane, Thomas J. and Cecilia E. Rouse. 1993. "Labor Market Returns to Two- and Four-Year Colleges: Is a Credit a Credit and Do Degrees Matter?", Working Paper #4268, National Bureau of Economic Research.

Lawrance, Emily C. 1991. "Poverty and the Rate of Time Preference: Evidence from Panel Data", Journal of Political Economy, 99(1): 54-77.

Lehr, Dona, and Jan M. Newton. 1978. "Time Series and Cross-Sectional Investigations of the Demand for Higher Education", Economic Inquiry, 16(2): 411-422.

33

Levin, Andrew, and Chien-Fu Lin. 1992. "Unit Root Tests in Panel Data: Asymptotic and Finite-Sample Properties", Discussion Paper 92-23, UCSD Department of Economics.

Loury, Linda Datcher. 1990. "Effects of Cohort Size on Postsecondary Training." In A Future of Lousy Jobs? The Changing Structure of U.S. Wages, ed. Gary Burtless, 165-193. Washington, D.C.: The Brookings Institution.

Manski, Charles, and David A. Wise, 1983. College Choice in America. Cambridge, Mass.: Harvard University Press.

McPherson, Michael, and Morton Owen Schapiro. 1991. Keeping College Affordable: Government and Educational Opportunity. Washington, D.C.: The Brookings Institution.

Murphy Kevin M., and Finis Welch. 1990. "Empirical Age-Earnings Profiles", Journal of Labor Economics, 8(2): 202-229.

Siow, Aloysius. 1984. "Occupational Choice under Uncertainty", Econometrica, 52(3):631-645.

Sulock, Joseph. 1982. "The Demand for Community College Education", Economics of Education Review, 2(4): 351-61.

34

U.S. Department of Education, National Center for Education Statistics. 1991. Digest of Education Statistics. Washington, DC: GPO.

________. 1992a. National Postsecondary Statistics, Collegiate and Noncollegiate: Fall 1991. Early Estimates. Washington, D.C.: GPO.

________. 1992b. National Postsecondary Student Aid Study: Parental Support for Undergraduate Education. NCES 92-390. Washington, D.C.: GPO

Willis, R. and Sherwin Rosen. 1979. "Education and Self-Selection", Journal of Political Economy, 87(5), part 2: S7-S36.

Zarkin, Gary. 1983. "Cobweb Versus Rational Expectations Models: Lessons from the Market for Public School Teachers’, Economics Letters, 13 (1): 87-95.

35

Endnotes

1. For first-time freshmen in Fall 1989 the percentage of first-time freshmen in two-year public colleges was 50.8 percent (U.S. Dept. of Education 1991, Table 169). For all undergraduates, 41.3 percent attended two-year public colleges in 1989, and among all students in public higher education, 51.2 percent were at two-year public colleges (ibid., Table 166). Department of Education estimates for 1991 show 44.4 percent of all undergraduates at two-year public colleges (U.S. Dept. of Education, 1992a). 2. See for example, Breneman and Nelson (1981), and McPherson and Schapiro (1991), pp. 44-54. For specific studies of community college demand and student aid, see Berne (1980) and Sulock (1982). 3. See for example, Clotfelter, Ehrenberg, Getz, and Siegfried (1991), Chapter 3. 4. Cobweb models were popularized by Richard Freeman in a series of articles on the labor market for college graduates. See Freeman (1971), (1976a), and (1976b). 5. The evidence suggests that individuals from lower socioeconomic groups have higher subjective rates of time preference. See, for example, Lawrance (1991). 6. A referee noted that in the model no worker is indifferent between ED=12 and ED=16, and suggested that this result had more appeal if the intermediate path was ’14 years of education’, including both a community college degree and 2 years of work at a four-year college. While we are sympathetic to this point of view, the empirical literature on the

36

distinction between these two paths in terms of economic returns is quite unsettled. Compare for instance the results of Grubb (1993) and Kane and Rouse (1993). 7.For a review of models of self-selection into higher education, as well as an extension to the case in which the outcome is uncertain, see Altonji (1993). 8.For a more detailed discussion of the underlying assumptions of this model, and the derivation of the results, see Betts and McFarland (1993). 9.The percentage of students who transfer from community colleges to four year colleges is small but significant. Grubb (1991) reports that 47 percent of high school graduates in the High School and Beyond study who entered community colleges stated that they desired a Bachelor’s degree or higher. Within four years, 20.2 percent of community college students had transferred to four year colleges. 10. This procedure is somewhat similar to the way in which Willis and Rosen (1979) model college choice as a function of the starting wage and wage growth. 11. These calculations were made by the authors based on data compiled by the Department of Education. 12. Due to a lack of degrees of freedom, we did not estimate instruments for the cost of attending a four-year college two to four years into the future. However, the current cost and the one-year ahead forecast should contain much of this information. 13. Most notably, the Department of Education has sometimes included and sometimes excluded Pell from the aid category for different years. This limits the ability to construct a time series for the entire period covered by this paper. 14. The nine regions all show similar trends. The full set of graphs are available upon request

37

from the authors. 15. We define the "attendance rate" as the percentage of 18-65 year olds in the Census region attending a community college, with attendance adjusted to a full-time equivalent basis. (Population and unemployment data were obtained from the Bureau of the Census.) As with the rest of the paper, enrollment data come from the NSF CASPAR database and the U.S. Department of Education’s HEGIS/IPEDS institutional database. Financial variables are expressed in 1987 values. In calculating Full Time Equivalents, the Federal government generally counts part time students as 1/3 of a full time student. We use this convention throughout the paper. 16. See Hausman (1978). 17.In practice, the addition of a time-varying region-specific error component to the standard fixed effect formulation causes the t-statistics on the regional labor-market variables to drop by half or more, while leaving the coefficients little changed. 18. In this regression, we estimate a random effects model with a time-varying error component specific to each state, rather than each region, as is the case in the other regressions in the paper. 19. Another source of competition for postsecondary enrollment comes from private vocational schools. Unfortunately, the IPEDS data set contains only very sketchy information on these institutions, so that it was not possible to create a ’market share’ estimate for these institutions. 20. To check the robustness of the enrollment:unemployment relation, we re-estimated equation #5 in Table 2 in first differences. The result could be quite different in the case in

38

which enrollment had a unit root. The sum of the four unemployment elasticities falls 24 percent to a smaller but still very substantial 3.4. The overall adult unemployment rate remains the most significant variable in the regression. Levin and Lin (1992) have recently developed a method of testing for unit roots in panel data. However, to the best of our knowledge nobody has yet developed critical values for testing for unit roots in unbalanced panels with missing data. 21.One referee suggested that rises in the unemployment rate and enrollment rates in the late 60’s and early 70’s could be related to the entry of baby boomers to the labor market. The evidence here suggests that if anything the proportion of young people in the population is negatively related to enrollment rates. See Loury (1990) for a review of the literature on demographics and enrollment, and further evidence of a negative linkage between the size of young cohorts and enrollment rates. 22. In the restricted version, the coefficients and significance of the remaining variables were very little changed. 23. On a similar note, a referee suggested to us that the measurement of the labor-market variables at the level of the Census region unavoidably creates measurement error, which in turn may explain why the coefficients on the proxies for earnings are not larger and more statistically significant. 24.When regressions #4 and #6 were repeated without the current unemployment rate as a regressor, the size and level of significance of the interaction terms were even smaller. 25. We should point out that this conclusion is not robust to the way in which the error term is modelled. In initial regressions in which we used fixed effects for colleges but specified the

39

error term without an error component for region and year, we found that this restriction was rejected (p-value = 0.01), but that a test of the model with unemployment rates lagged once against the same model with the addition of the unemployment rates lagged twice retained the null (p-value=0.05). It is thus safest to conclude that changes in the unemployment rate in year t are fully reflected in fall enrollment figures by year t + 1 at the latest. 26. The correlation between appropriations per student and the attendance rate is -0.278. The correlation between appropriations per student and the unemployment rate is -0.391. 27. Because some states link a part of appropriations to capitation, we divided the appropriations figures by FTE enrollment to observe the non enrollment-linked volatility in appropriations. The graphs depicting only appropriations levels show a similar result, though the relationship is somewhat more muted given that both enrollment and appropriations have trended up over time. 28. The newsletter Higher Education and National Affairs (1991), for example, noted that more than half of all states had made "significant" cuts in 1990-91 appropriations to community colleges. 29. It is possible that rather than observing fiscal stress, we may just be witnessing declining average cost at two-year colleges as enrollments rise, but field evidence collected as part of this project suggests that colleges are less interested in adding students during recession. College presidents we interviewed attribute this to the fact that most community colleges in normal economic times operate at or above capacity, and that the extra, recession-induced enrollment increases average cost. A more general discussion of higher education production functions and their links to quality can be found in Hoenack and Collins (1990).

40