Firm Specific Human Capital vs. Job Matching: A New Test

Firm Specific Human Capital vs. Job Matching: A New Test. Erwan Quintin John J. Stevens (cid:3) Federal Reserve Bank of Dallas Federal Reserve Board May 22, 2003 Abstract We use a unique data set on employee turnover by industry in Arizona to test competing theories of turnover. We (cid:12)nd that industries with lower establishment survival rates have more employee turnover, even after controlling for di(cid:11)erences in the distribution of employee tenure. This result is consistent with a model of turnover where employees choose how much (cid:12)rm speci(cid:12)c human capital to accumulate, but it is inconsistent with job matching models. Keywords: Firm survival; Employee turnover; Firm speci(cid:12)c human capital. JEL classification: J24; J31; J63. ∗E-mail: erwan.quintin@dal.frb.org and john.j.stevens@frb.gov. We wish to thank Patricia Anderson for pointing us to the data we used and Elias Brandt for his assistance. We also thank seminar participants at the Federal Reserve Bank of Atlanta, Southern Methodist University, and the Federal Reserve Bank of Minneapolis for their comments. The views expressed herein are those of the authors and may not reflect the views of the Federal Reserve Bank of Dallas, the Federal Reserve Board or the Federal Reserve System.

1 Introduction The negative correlation between a worker’s tenure and the likelihood that the worker quits or is laid o(cid:11) (see Anderson and Meyer, 1995, for a review) motivates two competing theories of turnover. In job matching models (e.g. Jovanovic, 1979a, or Mortensen and Pissarides, 1994), workers well-matched with their employers are less likely to quit or lose their job, and tend to have more seniority. In a second class of models (e.g. Oi, 1962, Becker, 1962, or Jovanovic, 1979b), workers endogenously decide how much (cid:12)rm speci(cid:12)c human capital to accumulate. Separation rates fall with tenure because workers with more tenure tend to have more speci(cid:12)c human capital.1 A largeliterature is devoted to separating these competing views. Most tests arecentered on two predictions of the endogenous, human capital view: 1) earnings should rise with tenure,2)involuntaryseparationsshouldbeassociatedwithhigherearningslossesforworkers with more tenure, even after controlling for initial employer-employee match quality. Those tests yield mixed results. For example, Altonji and Shakotko (1987) and Abraham and Farber (1987) (cid:12)nd that earnings rise little with job tenure, while Topel (1991) (cid:12)nds strong support for both predictions of the (cid:12)rm speci(cid:12)c human capital model. In this paper, we adopt a new strategy. If (cid:12)rm speci(cid:12)c capital is in part endogenous, then workers employed in (cid:12)rms that are less likely to survive should choose to accumulate less (cid:12)rm speci(cid:12)c capital than their counterparts with more secure jobs. In turn, those workers should be more likely to quit their jobs or be laid o(cid:11) than other workers, even at equal tenure. We begin by establishing those claims formally in the context of a dynamic general equilibrium model. Then we implement the corresponding test with a unique data set on turnover by industry in Arizona. We (cid:12)nd strong evidence that industries where survival rates are higher have less turnover in Arizona, even after controlling for di(cid:11)erences in the distribution of tenure and the size distribution of establishments across industries. This evidence, we conclude, provides new support for the view that endogenously accumulated (cid:12)rm speci(cid:12)c human capital accounts for part of the variance of earnings and turnover across 1The endogenousview includes models with contractualfrictions (Lazear,1981)whereworkerschooseto forego a fraction of the value of their current marginal product in return for an actuarially fair repayment promise. Employersusethispromisetoinducemoree(cid:11)ortoverthedurationoftheemploymentrelationship. Over time the value of the promise rises, and so separations become less likely. 1

(cid:12)rms and industries. 2 Firm survivor and employee turnover We consider a model with an in(cid:12)nite number of periods. Each period, constant measures of (cid:12)rms andworkers areborn. Firms belong either toindustry H orto industry L. The fraction of (cid:12)rms born in each industry does not vary over time, and (cid:12)rms remain in the industry to which they are born until they die. In industry H, (cid:12)rms survive from one period to the next with likelihood p while in industry L they survive with likelihood p < p .2 Industries H L H are otherwise identical. Firms in both industries can transform a quantity n of labor into f(n) units of the consumption good, where f is strictly concave, strictly increasing, and lim f(cid:48)(n) = +1. n(cid:55)→0 Workers are risk neutral and endowed with the same ownership share of all (cid:12)rms. They die with likelihood 1 − β 2 (0,1) at the end of each period. In each period, workers are employed by exactly one employer. The maximum quantity x of labor a worker can deliver to their employer is drawn from the set f0,1,2g. We refer to x as a worker’s productivity. In the (cid:12)rst period of employment with a given (cid:12)rm, x = 1 with probability 1. The evolution of a worker’s productivity in subsequent periods depends on the time s she devotes to training. With probability h(s), x moves up to 2 while with probability δ(s), x falls to 0, where h and −δ are increasing, h is strictly concave, h(0) > 0 and h(cid:48)(1) = 0. States x = 0 and x = 2 are absorbing. At the beginning of any period, workers can choose to quit and begin providing labor to a new employer. The process governing the evolution of x may be correlated across workers in a given subset of (cid:12)rms, but we assume no aggregate uncertainty. Speci(cid:12)cally, in each industry, a fraction h(s) of the workers who devote time s to training see their productivity rise, while a fraction δ(s) of these workers see their productivity fall. We only consider steady state equilibria, i.e. equilibria in which the price w of labor (the i 2Dunne etal.(1987)showthatexitratesdi(cid:11)ermarkedlyacrossindustriesandthatthosedi(cid:11)erencestend to persist over time. The data we present below for Arizona illustrates these empirical regularities. Our assumption that exit rates are exogenously (cid:12)xed is strong, but can be relaxed without altering our main result. 2

wage rate) in industry i 2 fH,Lg is constant. A steady state equilibrium is a pair of wage rates such that labor markets clear in both industries in all periods. To simplify the proof of existence we assume that βh(1) < δ(1). Under this assumption, all (cid:12)rms must hire new workers in every period in steady state.3 Therefore, wages adjust so that new workers are indi(cid:11)erent between the two industries. As a result, incumbent workers at productivity level x = 1 are indi(cid:11)erent between quitting and staying. We assume, for concreteness, that they do not quit. With this convention, workers quit if, and only if, their productivity level falls to x = 0. Let V (x) denote the expected income of a worker of productivity level x in industry i i 2 fH,Lg. From the previous discussion we know that in steady state new workers are indi(cid:11)erent between working in the two industries, so V (1) = V (1). Therefore, the expected H L income of a worker of productivity x in industry i 2 fH,Lg can be written as V (x) = max(1−s)xw +βfV (1)+p [(1−h(s))V (x)+h(s)V (2)]g (2.1) i i i i i i s∈[0,1] where (1 − s)xw is current period earnings if a fraction s of time is devoted to training. i Because the worker can change employers if necessary, her productivity is at least 1 in the next period. If her employer survives, which occurs with likelihood p , and if her training i e(cid:11)orts pays o(cid:11), which occurs with likelihood h(s), she moves up to productivity x = 2. Workers with x = 2 devote no time to training. Denote by si be the fraction of time workers with x = 1 devote to training in industry i 2 fH,Lg. Let Qi(t) denote the fraction of workers of tenure t who quit in industry i in any given period. By tenure we mean the number of periods a worker has worked for their current employer at the beginning of a given period. In particular, all workers have tenure 0 in their (cid:12)rst period of employment, and workers of tenure 0 do not quit by construction. We can now state: Proposition 1. A unique steady state equilibrium exists. Furthermore, in steady state, 1. w < w , H L 2. sL (cid:20) sH with a strict inequality if and only if sH > 0, 3SeeQuintinandStevens,2003,foracompleteargument. Thatpaperalsorelaxesmanyofthesimplifying assumptions made in this section and introduces unemployment. 3

3. Qi(t) < Qi(t+1) for all t (cid:21) 0 and i 2 fH,Lg, 4. QH(t) (cid:20) QL(t) for all t > 0, with a strict inequality if and only if sH > 0. Proof. Existence and uniqueness can be established with standard arguments (see Quintin and Stevens, 2003). Assume by way of contradiction that w > w in steady state. Then, H L since the return to any training policy is higher in industry H, we have V (1) > V (1), H L which can’t be in steady state. Now consider the second item in the proposition. Given (2.1), workers in industry i 2 f1,2g choose an si that satis(cid:12)es w (cid:20) h(cid:48) (si )βp [V (2)−V (1)] i i i i with a strict equality if and only if si > 0. (Recall that h(cid:48)(1) = 0.) To establish the second item of the proposition, we only need to argue that p [V (2)−V (1)] > p [V (2)−V (1)]. H H H L L L Assume that this inequality does not hold. Then, because w > w , inspection of (2.1) L H shows that V (1) > V (1), an inequality which cannot hold in equilibrium as no worker H L would join (cid:12)rms of industry H. This result yields the second item of the proposition. In both industries the evolution of a worker’s productivity with tenure follows a Markov chain with states f1,2g. Because h(0) > 0 the chain’s expected value rises strictly with tenure, which is item 3 of the proposition. In addition, sH (cid:21) sL implies that the Markov process governing a worker’s productivity in industry H (cid:12)rst order stochastically dominates the process governing the evolution of a worker’s productivity in industry L. This yields the last item of the proposition, and completes the proof. On average, workers in industries with high survival rates have more tenure, quit less often, and are more productive. This result is true even if there is no training (sH = sL = 0) because h(0) > 0. The no-training outcome prevails, for instance, when h(s) = h(0) for all s, in which case the model has the same predictions for turnover as exogenous, job matching models. When sH > 0, which occurs for instance when lim h(cid:48)(s) = +1 , workers in the high s(cid:55)→0 survival industrychoose toaccumulate more(cid:12)rmspeci(cid:12)c capitalthanworkers inlowsurvival industries. Therefore, if we interpret s as the time devoted to training, we can, in principle, 4

separate the exogenous view (sH = 0) from the endogenous view (sH > 0) by comparing training intensity across industries. However, as Barron et al. (1997) point out, the state of existing evidence on training, particularly informal training, does not allow for a convincing test along those lines. The last item of proposition 1 suggests another approach. If the endogenous view of (cid:12)rm speci(cid:12)c capital is empirically relevant, then turnover rates should be lower in high survival industries, even after controlling for differences in the distribution of tenure across industries. If, on the other hand, the evolution of worker productivity is mostly exogenous, then controlling for tenure should su(cid:14)ce to account for di(cid:11)erences in turnover rates across industries. We implement this test after describing our dataset. 3 The data Our test requires industry level information on turnover rates, (cid:12)rm or establishment death rates, and the distribution of tenure. We use separation rates computed from state unemployment records by Arizona’s Department of Economic Security (DES) as our measure of turnover for 1994, 1996, and 1998. A separation is said to have occurred when an individual received earnings from a speci(cid:12)c employer in a given quarter, but did not receive any earnings from that employer in the subsequent quarter. For 1994 and 1996, the reported separation rates equal annual separations divided by annual employment; for 1998, the ratio of annual separations to average quarterly employment was reported.4 In each of the three years, the separation rates were published for 2-digit Standard Industry Classi(cid:12)cation (SIC) industries. DES also reports average quarterly employment by industry in Arizona. Annualexitratesforestablishments byindustryarecalculatedbytheU.S.CensusBureau using consecutive years of the data underlying the U.S. County Business Patterns (CBP) data. The Census exit rate tabulations are for the nation as a whole, and while Arizonaspeci(cid:12)c rates would be preferable, we do not expect that it would have a substantively 4If the quarterly separation rates were constant over the year, then the two measures would be equal. If the separation rate are seasonal, then the measures are not equivalent. Likewise, if total employment is seasonal then the measures will not be equal even if the separation rates just exhibit random noise (due to weighting di(cid:11)erences between the two measures). 5

di(cid:11)erent e(cid:11)ect, as industry survival rates in Arizona will be correlated with their nationwide counterparts. A weak correlation would only weaken the signi(cid:12)cance of our exit rate variables. If anything therefore, using nationwide data for exit rates biases our test against the endogenous view. Our data on the distribution of tenure come from the February 1996 and 1998 Job Mobility and Tenure Supplements of the Current Population Survey.5 We also make use of these CPS data to compute the average age of employees as well as the fraction of men and white workers in all industries. In all our calculations, we include all employees between the ages of 16 and 65, and rely on CPS weights. Finally, we added to our set of variables various statistics that describe the organization of production in Arizona using CBP data. Large establishments have less turnover, a phenomenonfor which avariety ofreasons have beenadvanced (see Even andMacpherson, 1996, for a review). We want to make sure that it is not simply because high survival industries have larger (cid:12)rms (due for instance to learning by doing) that they have less turnover. Table 1 provides summary statistics for the major industry group with the largest employment in each of the 9 industry division for which we have data. Our entire data set is available upon request. Table 1 reveals that exit rates and employee turnover rates vary greatly across industries, and that those di(cid:11)erences are very persistent over time. 4 The test The test is simple. Do industry exit rates for establishments help account for the variation in industry separation rates? The literature has ignored the role of exit, instead focusing on the somewhat mechanical relationship between tenure and separation rates. Given this relationship, we should expect industries where average tenure is high to have low separation rates.6 A regression of separation rates on tenure and other demographic and industry variables, shown in Table 2 under the Model I column, con(cid:12)rms that separation rates decline 5The amount of time spent working for one’s current employer is not available in 1994. 6Note however that even though employees with more tenure are less likely to quit or be laid o(cid:11), it is conceivable for industries with high average tenure to also have high separation rates. This would be the case for instance in an industry where 50 percent of employees quit after 1 year and the other 50 percent have 20 years of tenure on average. 6

as tenure rises. For example, in 1994, an industry whose employees have one more year of tenure than the average industry will, everything else equal, have a quit rate that is 3 percentage points lower. The results for1994 and1996 arequite similar in allthree models; the results for1998 are similar in sign but often vary in magnitude and degree of statistical signi(cid:12)cance due to the di(cid:11)erent methodforcalculatingseparationrates. Nonetheless, somegeneralpatternsemerge. Industries with higher shares of women and whites exhibit lower separation rates (although the signi(cid:12)cance of these results varies across years and speci(cid:12)cations).7 The dummy for agricultural industries is always signi(cid:12)cant and may reflect fundamental di(cid:11)erences in the nature of the employer-employee relationship in this sector. In contrast, the dummy for manufacturing is never signi(cid:12)cant, despite what seems to be a common belief that separation rates for manufacturing di(cid:11)er systematically from those in the trade and service industries. In Model II we extend the mechanical, tenure-only explanation by including industry exit rates and an interaction term between tenure and exit rates. Our test hinges upon the sign and signi(cid:12)cance of these variables. We (cid:12)nd that exit rates help account for separation rates: An industry with an exit rate that is one percentage point higher than average will, everything else equal, have a separation rate that is nearly 3-1/2 percentage points higher. The interaction term enters negatively, which suggests that as average tenure rises, the exit rate plays a diminishing role. We were particularly concerned that the behavior of tenure was not adequately captured by a simple average tenure measure, so we also included the level of tenure for the employee at the 25th percentile. This measure, when statistically signi(cid:12)cant, tends to lower separation rates. We were also concerned that our results could be a(cid:11)ected by the share of employees with less than one year of tenure, as quits tend to be concentrated in the (cid:12)rst year, but including this share did not alter our results.8 While this test favors the endogenous view of (cid:12)rm speci(cid:12)c capital, exit rates could be proxying for other industry characteristics such as the distribution of establishment size 7It is important to note that our unit of analysis is industries and not individuals. If we had individuallevel data, we would expect to (cid:12)nd that women have higher separation rates than men after controlling for industry fixed effects. In our industry-level regression,the share of men is likely proxying for other industry characteristics, such as poor working conditions, that tend to reduce the share of women and increase separation rates. 8Results from this speci(cid:12)cation, and from speci(cid:12)cations that use a variety of tenure statistics in each industry are available upon request. 7

and age (two characteristics that are empirically correlated). To test the robustness of our result to this possibility, we included in Model III the share of Arizona employment that is in establishments with 9 or fewer employees.9 Depending upon the year, this variable was marginally signi(cid:12)cant to insigni(cid:12)cant, and the exit rate results were not substantively changed either qualitatively or quantitatively. All of the results reported in Table 2 are from weighted regressions using industry employment inArizonaastheweights. Wealso ranunweighted regressions, but ourconclusions, with one exception, were unchanged. The exception is that in 1998 the exit rate is not significant in Models II and III, however, we think that these results give too much weight to very small industries. To check this possibility we reran the unweighted regressions excluding industries with fewer than 10,000 employees|for Arizona this roughly corresponds to the 25 percentile of industries. The results of these regressions were not substantively di(cid:11)erent from the weighted regressions with all industries included. Finally, we might expect that, everything else equal, unionized industries should have lower separation rates. However, the share of unionized employees by industry (from the CPS) did not have any explanatory power after controlling for tenure. 5 Conclusion Using a unique data set on employee turnover, this paper provides new evidence that employee turnover is higher in industries where (cid:12)rms or establishments are more likely to exit the industry. We show that tenure alone cannot explain the higher separation rates in low survival industries. The evidence we report favors endogenous, speci(cid:12)c human capital models of employee turnover rather than job matching models. This result also has implications for the literature on industry rents. Speci(cid:12)cally, our results suggest that industry rents may reflect di(cid:11)erences in job security across industries. Workers may receive higher wages in low survival industries, because they face a greater risk of unemployment, and the corresponding loss of (cid:12)rm-speci(cid:12)c human capital, than their counterparts in industries where jobs are more secure. 9This share is calculated from the CBP data as in Holmes and Stevens (2002). 8

References Abraham, Katharine, G., and, Farber, Henry S.,\Job Duration, Seniority, and Earnings." American Economic Review 77 (June 1987): 278-97. Altonji, Joseph, G., and, Shakotko, Robert, A., \Do Wages Rise with Job Seniority?" Review of Economic Studies 54 (July 1987): 437-59. Anderson,PatriciaM.,andMeyer,BruceD.\TheExtentandConsequencesofJobTurnover." Brookings Papers on Economic Activity: Microeconomics (1994): 177-236. Barron, John M., Berger, Mark C., and Black, Dan A. \How Well Do We Measure Training?" Journal of Labor Economics 15 (July 1997): 507-28. Becker, Gary, S., \Investment in Human Capital: A Theoretical Analysis." Journal of Political Economy 70 (1962): 9-49. Dunne, Timothy, Mark J. Roberts, andLarry Samuelson, \Patterns of FirmEntry andExit in U.S. Manufacturing Industries," The RAND Journal of Economics 19:4 (Winter, 1988) 495-515. Even, William, E., and Macpherson, David, A., \Employer Size and Labor Turnover: The Role of Pensions" Industrial and Labor Relation Review 49 (July 1996): 707-727. Holmes, Thomas J., and John J. Stevens. \Geographic Concentration and Establishment Scale." The Review of Economics and Statistics 84 (November 2002): 682-690. Jovanovic, Boyan. \Firm-speci(cid:12)cCapitalandTurnover." The Journal of Political Economy 87 (December 1979a): 1246-1260. Jovanovic, Boyan. \Job Matching and the Theory of Turnover." The Journal of Political Economy 87 (October 1979b): 972-990. Lazear, Edward P. \Agency, Earnings Pro(cid:12)les, Productivity, and Hours Restrictions," The American Economic Review, 71 (1981) 606-620. 9

Mortensen, Dale, T., and Pissarides, Christopher, A., \Job Creation and Job Destruction in the Theory of Unemployment" Review of Economic Studies 61 (1994) 397-415. Quintin, Erwan, andJohnJ.Stevens, \GrowingOldTogether: FirmSurvivalandEmployee Turnover" mimeo (2003). Oi, W., Y., \Labor as a Quasi-Fixed Factor" Journal of Political Economy 70 (1962): 538-55. Topel, Robert H., \Speci(cid:12)c Capital, Mobility, and Wages: Wages Rise with Job Seniority" Journal of Political Economy 99 (February 1991): 145-76. 10

Table 1: Selected statistics for the largest major industry group in each industry division Average quarterly Separation Establishment Establishments with 9 Average tenure employment rate (%) exit rate (%)† employees or fewer (%) in years Agricultural Services 1994 49,782 48.8 10.4 82.6 n.a. 1996 51,517 44.2 10.2 80.0 5.3 1998(cid:63) 45,952 74.0 12.7 77.2 5.4 Metal Mining 1994 11,260 6.2 18.0 50.8 n.a. 1996 13,355 8.2 19.7 54.5 12.2 1998 9,163 19.2 17.5 58.6 11.8 Special Trade Contractors 1994 128,985 31.9 12.5 75.8 n.a. 1996 147,321 31.4 12.1 72.4 n.a. 1998 153,491 59.8 12.1 72.6 n.a. Electronics and Other Electrical Equipment 1994 53,844 6.0 8.2 42.2 n.a. 1996 59,903 7.5 8.0 41.9 7.5 1998 48,232 59.4 8.5 43.9 8.4 Trucking and Warehousing 1994 28,943 23.4 13.0 74.3 n.a. 1996 28,708 23.9 13.6 75.1 6.2 1998 33,947 54.0 13.4 76.2 5.9 Wholesale Trade, Durables 1994 78,340 16.0 9.1 73.6 n.a. 1996 90,776 16.4 9.0 72.4 6.6 1998 93,154 40.1 9.3 72.6 7.1 Eating and Drinking Places 1994 206,411 34.3 12.1 49.9 n.a. 1996 227,014 35.0 11.9 48.6 3.0 1998 219,122 69.8 12.7 47.1 3.1 Depository Institutions 1994 42,567 12.8 7.0 52.3 n.a. 1996 44,680 15.9 5.9 41.4 7.0 1998 93,154 37.1 9.3 55.2 7.0 Health Services 1994 169,035 13.7 7.0 82.8 n.a. 1996 179,783 16.0 8.0 81.0 6.5 1998 192,484 50.4 7.9 80.0 6.8 Sources: Average quarterly employment and Separation rates (for Arizona): Arizona Departement of Economic Security. Establishment exit rates (for the U.S.): Census tabulations based on County Business Pattern (CBP) data. Establishments with 9 employees or fewer (for Arizona): Authors’ calculations based on CBP data. Average tenure (for the U.S.): February Current Population Survey. (cid:63) In 1998,separation rates were computed using a di(cid:11)erent formulae. † Exit rates for year t are establishment deaths between year t − 1 and t divided by initial number of establishments. 11

Table 2: Explaining Separation Rates Model I Model II Model III Variable 1994 1996 1998 1994 1996 1998 1994 1996 1998 Constant 47.98 47.70 118.16 5.27 7.71 42.81 24.98 17.70 40.64 (12.89) (12.25) (29.33) (14.62) (13.87) (25.25) (14.61) (13.44) (25.96) Fraction of men 13.55 9.77 14.65 8.20 5.30 8.76 9.12 5.92 8.54 (3.85) (3.84) (7.51) (3.35) (3.45) (6.81) (3.39) (3.58) (6.95) Fraction of whites -15.20 -13.17 -51.90 -8.58 -9.31 -39.99 -21.22 -16.24 -37.20 (14.33) (13.53) (34.75) (14.43) (13.22) (25.04) (13.96) (12.81) (25.42) Average tenure -3.17 -2.74 -6.20 2.17 2.70 4.66 0.68 2.00 4.60 (0.92) (0.85) (1.24) (1.67) (1.25) (3.98) (1.16) (1.19) (4.02) Tenure, 25th-percentile -1.79 -2.49 8.51 -2.42 -3.81 1.81 -1.20 -2.98 1.84 (2.14) (1.76) (5.87) (1.23) (1.11) (5.80) (1.25) (1.10) (5.83) Agriculture dummy 21.86 17.63 14.70 25.14 20.22 19.33 25.21 20.32 19.09 (3.52) (2.56) (5.32) (1.59) (1.60) (3.58) (1.64) (1.64) (3.65) Manufacturing dummy -1.40 -2.25 0.58 -2.41 -3.82 -1.98 0.24 -2.05 -2.35 (3.10) (2.81) (4.73) (2.72) (2.64) (4.72) (1.72) (1.99) (4.34) Exit rate 3.42 3.34 5.57 2.64 2.99 5.57 (0.73) (0.72) (1.90) (0.78) (0.73) (1.93) Exit rate (cid:2) avg. tenure -0.44 -0.43 -0.79 -0.32 -0.38 -0.79 (0.084) (0.095) (0.28) (0.093) (0.093) (0.29) Average size -0.063 -0.036 0.012 (0.030) (0.020) (0.042) R2 0.77 0.78 0.53 0.83 0.84 0.61 0.85 0.91 0.61 Number of obs. 56 56 65 56 56 65 56 56 65 Note. The dependent variable in all models is the separation rate. All regressions are weighted by industry employment in Arizona. Unweighted regression results (available from the authors) are similar for 1994 and 1996; the exit rates are not signi(cid:12)cant in 1998 for the unweighted regressions. Heteroskedasticity-consistent standard errors are in parentheses. 12