II. Human Capital, Returns to Education and Experience. D. Estimating the Returns to Seniority and Specific Human Capital

Size: px
Start display at page:

Download "II. Human Capital, Returns to Education and Experience. D. Estimating the Returns to Seniority and Specific Human Capital"

Transcription

1 II. Human Capital, Returns to Education and Experience D. Estimating the Returns to Seniority and Specific Human Capital Plan 1. Issues with On-the-Job Training a. General vs. Specific Training b. Job Match 2. Estimation of the Returns Seniority a. Abraham and Farber b. Topel s model and extensions

2 1. Issues with On-the-Job Training For Mincer (1962), schooling is only the first stage in the process of the acquisition of occupational skill. The second stage, on-the-job (OJT) training, is thought to be as important and raises a different class of questions. In the case of formal training, Becker (1964) distinguishes general training from specific training, where general training is training that can be used in various firms, not just in firms that provide the training, whereas specific training is specific to the firm. This raises the question of who should pay for training? Is a government subsidy warranted? How should the training be evaluated? We distinguish general training and firm-specific training. o General: training that is useful at all firms once it is acquired o Specific: training that is useful only at the firm where it is acquired

3 a. General Training vs. Specific Training time=1 time=1.5 time=2 Workers produce output y and receive training τ at cost c(τ ). Workers face option to leave firm. Worker produces output y + f (τ ) at any firm (general training) Assuming increasing productivity and cost of training f'(τ )>0, c(0) = c'(0) = 0 and c'(τ ) > 0, c''(τ ) > 0, c'(τ ) τ. FOC social optimum for training? c'(τ ) = f'(τ ). o How do we know that τ > 0? Can firm pay worker w 1 = y, w 2 = y + f (τ ) c(τ )? Since the worker s opportunity (outside) wage rises by f(τ ) with training, firm must pay w 2 = y +f (τ ) or lose employee to a competitor

4 Basic insight of Becker model: Hold-up o In a competitive labor market, employers will not pay for general human capital. o The problem is holdup : If both parties need the other in order to succeed, each party is in a position to threaten a hold up of operations that may be extremely costly to the other party. A hold-up occurs when one party decides to renegotiate after the investment has been made. 1. Direct investments by workers School, training programs, etc. 2. Indirect payment to employer: Training wage, that is: w 1 = y c(τ ), w 2 = y + f (τ ). Since employers unwilling to fund socially productive training, is this a market failure? o Worker is full residual claimant to returns from training investments Efficient incentive for worker to make the investments. How do high rates of labor turnover affect training investment incentives? Not at all. How do worker credit constraints impact this model? Quite a bit.

5 o Cannot typically borrow against future stock of human capital (except from government, e.g., student loans). o Wages may not go low enough (e.g., negative) to cover efficient training expenditures. How relevant is this model? o Clearly, covers considerable territory: College education is paid for by families, society; Professional schools do not generally subsidize tuition: Law, medicine, business. o What about the case of military education? Military flight education costs multiple millions to provide. The skills are completely portable (to commercial aviation).

6

7 As with choices of educational investments, an employee s choice of occupation (and of the type of required training) may depend on the length of foreseen worklife.

8 On the other hand, if on-the-job training is firm specific, the costs of firmspecific training is shared by employers and employees Because the specific training is useful only with the firm that provides the training, other firms would have no incentive to pay higher earnings for such training and the trainee would not bear the cost because of an inability to reap the benefits in the form of higher earnings at other firms (if he/she is laid off) Firms thus have to sponsor specific training by offering the alternative wage and incurring the costs the training. * w VMP during training and the benefits VMP wa after a t More likely, when there is some uncertainty about the continuation of the employment relationship, specific human capital may be a shared investment, * where the firms pays wt wa and w wa. w a

9 Specific human capital investments provide an incentive for firms and workers to maintain their employment relationship in the face of external shocks to demand and supply. In specific capital is important, Mincer and Jovanovic (1981) show that increases in job tenure, holding labour market experience constant, should have a positive effect on earnings.

10 c. Job Match One major implication of the concept of firm-specific human capital is to emphasize the importance of the duration of a worker-firm match in determining the total pay-off to the investment. How to estimate returns to tenure, experience, and job match, when the job match component is endogenous? o Job match increases over the working life o Better job matches lead to longer tenure This raises the questions: o Do earnings rise with seniority (in a causal way)? o Or is it a selection issue? 1. High wage jobs tend to survive would be a selection of jobs (e.g. 100 best firms to work for ). 2. More productive or able people changing jobs less often would be a selection of workers.

11 2. Estimation of the Returns to Seniority a. Abraham and Farber Abraham and Farber [AF] (1987) argue that the measured positive cross-sectional return to seniority is a statistical artifact due to the correlation of seniority with an omitted variable representing the quality of the worker-firm match. Suppose that the earnings of a particular worker i on job j in year t can be written lnwijt 1Sijt 2EXPij ij ijt (1) where Wijt hourly earnings, Sijt current seniority (tenure), 1 return to seniority, EXP ij pre- job experience, 2 return to pre-job experience, ij is a person/jobspecific error term representing the excess of earnings enjoyed by this person on this job over and above the earnings that could be expected by a randomly selected person/job combination.

12 Letting that person/job-specific component depend on pre-job experience ij EXPij ij (2) where captures the growth in ij with experience and ij is the component of ij that is uncorrelated with EXP ij. Substitution (2) into (1), they get lnwijt 1 Sijt ( 2 ) EXP ij ij ijt So that the net return to seniority is 1 ( 2 ), estimating an equation of return to seniority, while ignoring the person/job specific component will biased the estimates of both seniority and pre-job experience. Because good workers and workers in good jobs or good matches are likely to stay on their jobs longer, the completed duration of jobs is positively related to ij. Thus, the idea is to control for good matches by introducing an estimate of the duration of a job.

13 Source: Abraham and Farber (1987) 288 THE AMERICAN ECONOMIC REVIEW JUNE 1987 TABLE 4a-SELECTED COEFFICIENTS FROM In (AVERAGE HOURLY EARNINGS) MODELS MANAGERIAL AND PROFESSIONAL NONUNION SAMPLE a Mean OLS IV OLS OLS [s.d.] (1) (2) (3) (4) Years of Experience [10.08] (.0027) (.0058) (.0027) (.0031) (Years of Experience) 2 [407.84] (.00006) (.00014) (.00007) (.00007) Years of Current Seniority [8.34] (.0011) (.00128) (.00178) (.00256) E(Completed Job Duration) [12.18] (.0024) (.0050) { E(Completed Job Duration)}2 [505.56] (.00006) (.00016) E(Job Duration) X [ = 1 if 3 < Seniority ~ 10] [10.46] (.00432) { E(Job Duration)} X [ = 1 if 3 < Seniority ~ 10] [325.3] (.00015) E(Job Duration) X [ = 1 if Seniority> 10] [15.78] (.00455) { E(Job Duration)} X [ = 1 if Seniority> 10] [572.9] (.00015) R <LAll models also include controls for education, race, marital status, disability, occupation, industry, region, and year. E (Completed Job Duration) is computed using the estimates in col. 1 of Table 2. The numbers shown in parentheses are standard errors. Sample size = TABLE 4b-SELECTED COEFFICIENTS FROM In (AVERAGE HOURI~Y EARNINGS) MODELS BLUE-COLLAR NONUNION SAMPLE a Mean OLS IV OLS OLS [s.d.] (1) (2) (3) (4) Years of Experience [11.14] (.0024) (.0040) (.0026) (.0026) (Years of Experience) 2 [470.81] (.00006) (.00009) (.00006) (.00006) Years of Current Seniority [7.46] (.0011) (.00172) (.00213) (.00302) E(Comple-ted Job Duration) [11.75] (.0021) (.0057) { E(Completed Job Duration)}2 [444.45] (.00006) (.00024) E(Job Duration) X [ = 1 if 3 < Seniority ~ 10] [8.27] (.00538) { E(Job Duration)} X [ = 1 if 3 < Seniority ~ 10] [211.7] (.00024) E(Job Duration) X [ = 1 if Seniority> 10] [12.90] (.0055) { E(Job Duration)} X [ = 1 if Seniority> 10] [461.9] (.00024) R <LAll models also include the controls listed in Table 4a, fn. a. E (Completed Job Duration) is computed using col. 2, Table 2. Standard errors are shown in parentheses. Sample size = 3554.

14 o They introduce an expected duration of a job match as regressor in the earnings equation. It is estimated as the half of the average observed job seniority, under the assumption that observed seniority in a cross-section is, on average, half of completed job duration. Alternatively, he uses the residuals from a regression of seniority on predicted completed job duration as instrument for seniority. They show that returns to seniority fall from 1.1 (1.4)%/year to 0.6 (0.3)%/year when seniority in white (blue) collar. b. Topel s model and extensions Topel (1991), Altonji and Shakoto [AS] (1987), Altonji and Williams [AW] (1997, 2005) develop alternative methodologies to deal with the endogeneity of the unobserved job match with experience and tenure. Topel finds that 10 years of tenure increases wages by over 25 percent, but AS and AW find have average wage gains from 4 to 7 percent.

15 Source: Topel (1991)

16 Their assumptions are: Individuals switch jobs when they are offered something better (improved job match) As experience increases, you receive more offers so are more likely to be in a job with a better match Since job match is unobservable, this leads to a bias in the estimates of returns to experience and tenure As [AF], they base their estimations on the PSID, a panel of individuals, with multiple employment spells. Topel s first stylized fact supporting the existence of sizeable returns to seniority comes from the experience of displaced workers: o The average displaced worker suffers a 14 percent reduction in earnings

17 If we have panel data on individuals, we can write W ijt the log real wage of person i in job j in period t, X ijt the total labour market experience, and Tijt tenure with the employer, as Wijt 0 x X ijt TTijt ijt (1) where the error term is decomposed into ijt i ij jt uit (2) a fixed individual specific component, i ij a fixed job match specific error, jt a time varying job match specific component u a measurement error that is time varying and person-specific it But ideally, we would think of the tenure effect sought as the interaction term: T I * F *, where I is the individual, F is the firm and is time.

18 W The proper identification of that effect X I F I F F u I I F ijt 0 x ijt i F would require the inclusion of I (done with i ), F (need more than one worker by firm, done with employer-employee linked data base [Bronars and Famulari (1997), Abowd et al (1999), Bingley and Westergaad-Nielsen, (2003)], (Topel argues against the use of a time trend!) and I * F (done with ij ), F * (done with jt ), I * (done with u it, but ignored in Topel and AS.) But data on F is not available in the PSID, so letting the quality of the job match depends on experience and tenure, Topel considers the auxiliary regression ij b1 X ijt b2tijt ijt (3) o The likely sign of b1 is positive as suggested by matching models and conventional search models. ij jt it T

19 o The likely sign of b2 is ambiguous. Selection induced by voluntary job changes will lead low tenure to be associated with large values of ij, so that b2 could be negative. But if good matches are associated with lower probability of layoffs, b2 will be positive. Topel uses a two-step approach (2SFD). In the first step, he applies OLS to a within-job growth equation for job stayers to identify ˆ ˆ X ˆ T. o Writing equation (1) in differences: Wij x X ijt T Tijt ijt for job stayers ( X ij Tij 1), we get W ij In the second step, he writes X X 0 T where X 0 is the experience at the start of a job, so that Wijt 1 X ij0 Tijt ij0 Wijt ˆ ( T ) 1X ij0 eijt, where e T ( ˆ ijt ijt ) and ˆ is the OLS estimate from the first step. The linear tenure coefficient is the difference: ˆ T 1. x T ij

20 But there could be some heterogeneity in the individual component: i c1 X ijt c2tijt ijt (4) Using equation (1)-(4), Altonji and Williams (1997) show that the OLS estimates OLS OLS in (1) are likely biased: X X b 1 c1 and T T b 2 c2, with job match heterogeneity (the b s) and individual heterogeneity (the c s). AS uses an instrumental variables strategy to address the problems of individual and job match heterogeneity in the wage equation: they use deviations from mean tenure in job match ij ( DT ijt Tijt T.. t ), which is orthogonal to i and ij, and Since there is a likely positive correlation between ijt coefficient of tenure is downward biased. X ijt and t as instruments in (IV1). X and ij, AS s estimated Because both i and ij are included in eijt and may be correlated with X ij0 estimator may be biased., Topel

21 Source: Altonji and Williams (2005) 378 INDUSTRIAL AND LABOR RELATIONS REVIEW Table 2. OLS, IV1, and 2SFD on PSID Topel Replication Sample, (Dependent Variable: EARN_MW68) Replicated Topel Sample Topel (1991) OLS IV1 2SFD OLS IV1 2SFD Variable (1) (2) (3) (4) (5) (6) Linear Tenure Coefficient (0.0081) (0.0063) (0.0082) (0.006) (0.0079) Linear Experience Coefficient (0.0123) (0.0128) (0.0178) (0.0181) 2 Years of Tenure (0.0127) (0.0103) (0.0132) 5 Years of Tenure (0.0219) (0.0194) (0.0237) (0.0098) (0.017) (0.0235) 10 Years of Tenure (0.0250) (0.0277) (0.0287) (0.0105) (0.024) (0.0341) 15 Years of Tenure (0.0262) (0.0367) (0.0300) (0.0110) (0.028) (0.0411) 20 Years of Tenure (0.0287) (0.0485) (0.0317) (0.0116) (0.035) (0.0438) 5 Years of Experience (0.0406) (0.0423) (0.0601) 10 Years of Experience (0.0530) (0.0553) (0.0808) 30 Years of Experience (0.0524) (0.0633) (0.0783) Notes: White standard errors in parentheses for the OLS and IV1 estimators. Standard errors for the 2SFD estimator account for the fact that it is a two-step estimator and for person-specific heteroskedasticity and serial correlation in the error terms. Columns (1) through (3) use the replicated Topel sample, while columns (4) through (6) contain estimates reported by Topel (1991). The specifications in columns (1) and (3) do not contain a time trend. The specification in column (2) contains an exogenous time trend. Columns (4) and (6) are taken from Table 3 of Topel. Column (5) is taken from Table 6, column (2) of Topel. All specifications include a quartic in tenure, a quartic in experience, years of education, and dummies for marital status, union membership, current disability, residence in an SMSA, residence in a city with a population of more than 500,000, and eight Census regions.

22 Both the AS, AW and the Topel estimators are biased down by job match heterogeneity, and the Topel estimator is biased up by individual heterogeneity. Another difficult issue with the PSID is whether to treat time as exogeneous, AW (2005) explore this issue in detail. Taking the potential biases into account, AW conclude that reasonable estimates of the return to ten years of tenure should range between 0.06 and 0.13, which are between a quarter to a half of the OLS estimates. Thus, job seniority plays only a modest role in the determination of wages. How does one reconcile the large wage decreases experience by displaced workers with substantial seniority (e.g. as reported in Topel)? Parent (2000) argues that industry-specific human capital may be more important than firm-specific human capital. The more recent literature has explored differences in return to seniority by education groups (e.g. Connolly and Gottschalk, 2001)

23

24 Source: Parent (2001)

25 Using employer-employee dataset, a number of papers (e.g. Beffy, Buchinsky, Fougère, and Kamionka, 2006) for France and (Buchinsky, Fougere, Kramarz, and Tchenis, 2008) for the US have estimated structural models of inter-firm mobility. The large wage losses, as well as the unemployment histories, of older displaced workers have led governments to sponsor training. In the US, it started as early as 1958 and was followed by the Manpower Development and Training Act of The evaluation of training programs and other active labour market policies (ALMP) in the US began in the 1970s and now include a few meta-analysis of policies implemented in many countries. Card, Kluve and Weber (2010) analyze the results of 97 studies of ALMP between 1995 and o They conclude that job search assistance (JSA) and related programs have generally positive impacts. o Classroom and on-the-job training programs are not particularly likely to yield positive impacts in the short-run, but yield more positive impacts after two years.

26 o Importantly, they find that the mean differences between the experimental and non-experimental impact estimates are small and statistically insignificant (t<0.5), and they find no evidence of publication bias.

27 Basic readings: J. Mincer (1962) "On-the-Job Training: Costs, Returns, and Some Implications" Journal of Political Economy, 70.5 (October), K. Abraham and H. Farber. Job Duration, Seniority, and Earnings, American Economic Review, 77.3 (June 1987), Altonji, J.G. and N. Williams, Do Wages Rise with Job Seniority? Review of Economic Studies, Vol. 54, no. 179 (1997): ) Altonji, J.G. and N. Williams, Do Wages Rise with Job Seniority? A Reassessment Industrial and Labor Relations Review, Vol. 58, no. 3 (April 2005):