Wages and the composition of experience.

1. Introduction

The human capital model (Becker 1962; Mincer 1974) typically categorizes skills as being "firm specific," or those that increase productivity in only one firm, and "general," or those that increase productivity in more than one firm. Recent research by Loewenstein and Spletzer (1998) finds that employer-financed forms of training have positive impacts on wages at subsequent jobs, suggesting that these forms of training are portable across employers, or are general. Related studies by Neal (1995) and Parent (2000) challenge the traditional notion of skills being either firm specific or general, however. They find that the wage returns to tenure at the current job are similar to the wage returns to work experience at prior jobs in the same industry as currently employed and conclude that skills are "industry specific" rather than firm specific. Also, earlier research by Shaw (1984) indicates that skills acquired in a particular occupation are portable across employers, or that skills are "occupation specific" rather than firm-specific.

Taken together, these studies provide substantial evidence that skills are not necessarily firm specific but are likely to be specific to either an industry or an occupation. These findings suggest that reexaminations of theoretical models of human capital and job mobility are required, as these models generally focus on the match between a worker and firm rather than between a worker and a particular line of work, or a "career." For example, research by Neal (1999) indicates that workers typically search initially for a career and subsequently search between firms once a career match is found.

Yet while these prior studies indicate that certain skills appear to be more valued by subsequent employers than others, there remains considerable uncertainty as to what types of skills are the most transferable and what sets of skills constitute a career. In particular, the concept of industry-specific skills can be very different than that of occupation-specific skills. For instance, suppose an individual currently works as a residential real estate agent. Given that there are skills that are idiosyncratic to the real estate industry, the real estate agent might be able to easily move from residential sales into working as an accountant with a real estate firm or doing administrative work for a mortgage insurance company. This type of transition involves a change in occupation but not in industry, and the wage returns to prior experience as real estate agent should be equivalent to that of current job tenure in another job in the real estate field if prior experience truly is industry specific.

Conversely, the individual might find that the sales skills acquired as a real estate agent are very transferable and may move into another area of sales, such as the sale of pharmaceuticals, books, or steel. This type of transition involves a change in industry but not in occupation. The wage returns to prior experience as a realtor will be the same as that of current tenure in another sales job if skills actually are occupation specific.

While industry and occupation categories are not always readily discernable and many job changes will likely involve a change in both industry and occupation, making a distinction between industry-specific versus occupation-specific skills is important in trying to understand the skill acquisition process and in trying to get a handle on what is implied by the term "career." In this paper, an attempt is made to disentangle the contribution of industry experience and occupation experience to wages. In particular, prior labor market experience is segmented into mutually exclusive categories based on industry and occupation. This exercise sheds light on whether experience in the same industry or in the same occupation is rewarded to a greater extent than experience in other industries or in other occupations.

The results indicate that the wage returns to prior experience in either the same industry or the same occupation are equivalent to the wage returns to current job tenure. The return to prior experience acquired outside both the person's current industry and occupation is lower than the returns to job tenure and to the other forms of experience, however. This finding suggests that, other than time spent working outside the current industry and occupation, experience is homogeneous. It is important to point out, however, that a large proportion of workers' employment histories, particularly early in their careers, is spent outside the current industry and occupation. The estimates of the wage returns to alternative types of experience provide evidence on the benefits, and hence opportunity costs, of different employment paths.

The remainder of the paper is organized as follows. Section 2 describes the data and the empirical strategy adopted. The procedure used to segment prior workplace experience by both industry and occupation is discussed in this section, and evidence on the distribution of prior experience is offered. Section 3 presents wage equation estimates under alternative assumptions about the degree of homogeneity of prior workplace experience. Some extensions of the findings related to employer learning and worker ability are explored in section 4. Section 5 provides a robustness test of the results in regard to the issue of unobserved heterogeneity, while section 6 presents the conclusions.

2. Data and Empirical Strategy

What is the monetary return to alternative forms of skill acquisition, including different types of prior workplace experience? To answer this question, we estimate different models of wage determination. These equations are characterized by varying degrees of aggregation to determine the perceived relation between alternative workplace experiences and productivity.

Data

The analysis is based on the 1996 release of the National Longitudinal Survey of Youth (NLSY). The NLSY is a panel study of men and women who were between the ages of 14 and 22 in 1979 and who have been interviewed annually from 1979 to 1994. After 1994, the survey moved to a biennial cycle. A key feature of the NLSY is that it garners information in an event history format, in which dates are collected for the beginning and ending of important life events. In particular, starting dates and ending dates for all jobs are recorded, so it is possible to create fairly precise measures of current job tenure and prior work experience.

The sample used here is restricted to white males who were working for pay at the 1996 interview with nonmissing information on the variables used in the analysis (details on sample creation are provided in Appendix A). It should be mentioned that over the 1979-1996 waves of the NLSY, information on industry and occupation is consistently available only for those jobs that involved working at least nine weeks and 20 hours per week. (1) Consequently, the sample of workers is limited to those who held a job that met these restrictions at some time over the interview year. Also, the tenure and prior experience variables represent the number of weeks in which a person worked more than 20 hours per week (divided by 50). For episodes in which individuals held multiple full-time jobs simultaneously, the job that involved the most hours worked per week was used when generating the experience variables.

All variables that refer to a job, including industry and occupation, are measured as of the start of the job. It should be mentioned that some workers report changes in industry and occupation without changing jobs. While these changes may be more likely to occur for occupation changes for a given employer, it is impossible to distinguish between a "real" occupation change and a reporting or coding error. These types of misclassification errors are discussed by Neal (1999). (2)

Table 1 provides a description of the key work experience variables used in this analysis. Means of the prior experience variables are provided at the one- and three-digit industry/occupation level in Table 1. Means for all the variables used in the empirical analysis are provided in Appendix B. NLSY respondents were ages 31 to 39 in 1996 and on average had more than 14 years of total experience, spending close to six years on the current job and about eight years with prior employers. At the one-digit level, about three out of the eight years of prior work experience, or about one-third of the total, are spent in the same occupation. Similarly, about one-third of prior work experience is spent in the same industry. Using more disaggregated measures of prior experience based on both industry and occupation (same industry/same occupation, same industry/different occupation, different industry/same occupation, and different industry/different occupation) indicates that individuals spend only about a year and a half in the same industry and occupation as their current job, while about four years, or over half of all prior work experience, is spent in a completely different occupation and industry as the current job. The means at the three-digit level are even more dramatic, as they indicate that close to three-quarters of all prior experience is spent in jobs that are of a different industry and occupation than the current job.

These percentages are consistent with other findings that indicate that there is a great deal of workplace mobility among young workers. Other figures using the NLSY indicate that individuals hold more than nine jobs between the age of 18 and their mid-30s (Bureau of Labor Statistics 2000). The industry/occupation work experience averages reported in Table 1 suggest that young workers not, only "job shop" but also "career shop" in the sense that changes in industry, occupation, and both industry and occupation are very common. The average tenure figure of approximately six years, however, suggests that these workers do settle into a particular job and career and stay there during their late 20s and early 30s.

Model Specification

The conventional assumption is that experience gained prior to the current job is homogeneous, or that all forms of prior experience contribute equally to subsequent productivity. This model, referred to here as the homogeneous experience model, is specified as

In w = [[alpha].sub.1] + [[lambda].sub.1](Tenure) + [[phi].sub.1](Exp) + [[theta].sub.1](X) + [[epsilon].sub.1], (1)

where ln w is the natural log of the 1996 wage a worker receives on their current job (individual subscripts are suppressed), Tenure is years of work experience with the current employer, Exp is years of work experience with prior employers, the vector X consists of variables measuring personal characteristics other than tenure and prior experience, and [[gamma].sub.1] is a standard error term. (3) The X vector includes education, urban residence, marital status, current industry, and current occupation. In order to account for skill investments that occur during the early stages of a job, a dummy variable representing tenure less than one year is also included in the X vector. (4) In addition, an individual's score on the Armed Forces Qualifying Test (AFQT) is included and taken to be a measure of aptitude. (5)

Equation 2 allows the value of prior experience to a worker's current employer to depend on both the industry and the occupation of the earlier job and is referred to here as the heterogeneous experience model:

ln w = [[alpha].sub.2] + [[lambda].sub.2](Tenure) + [[psi].sub.A](ExpSOSI) + [[psi].sub.B](ExpSODI) + [[psi].sub.C](ExpDOSI) + [[psi].sub.D](ExpDODI) + [[theta].sub.2](X) + [[epsilon].sub.2] (2)

where ExpSOSI and ExpSODI are weeks of prior experience in the same occupation/same industry and in the same occupation/different industry, respectively. (6) Weeks of earlier experience in a different occupation/same industry and in a different occupation/different industry are represented by ExpDOSI and ExpDODI, respectively.

Estimation of this unrestricted model fosters comparison of the rate of return to a wider range of alternative prior workplace experiences given that experience is classified by industry and by occupation. This specification allows inferences to be drawn regarding the importance of within-industry experience relative to within-occupation experience. The contribution to a person's wage of prior same occupation experience is [[psi].sub.A] + [[psi].sub.B], while the contribution of prior same industry experience is [[psi].sub.A] + [[psi].sub.C]. Therefore, the importance of earlier occupation experience to an individual's wage relative to earlier industry experience can be determined by comparing [[psi].sub.B] and [[psi].sub.C]. If an additional week of experience in the same occupation--even if the job is in a different industry--enhances productivity more than an additional week of experience in the same industry but in a different occupation (i.e., [[psi].sub.B] > [[psi].sub.C]), then prior occupation experie nce will have a greater wage return than does prior industry experience.

The coefficient estimates from Equation 2 also permit a comparison of the return to current experience or tenure, [[lambda].sub.2], with the return to the various forms of earlier experience. Of particular interest is whether prior experience in the same industry and same occupation, ExpSOSI, contributes as much to current productivity as a comparable increase in time on the current job. If so, we expect [[psi].sub.A] = [[lambda].sub.2].

Equations 1 and 2 are specified essentially as variants to the standard earnings equation developed by Mincer (1974) and are estimated by ordinary least squares. It is important to mention that the returns to both tenure and experience may be impacted by the presence of various forms of unobserved heterogeneity. The current employment relationship is clearly the result of particular human capital and job search processes, and consequently these job match effects likely impact the estimated returns to tenure. Similarly, the returns to experience in a particular industry/ occupation may reflect unobserved differences between workers that lead to the choice of a particular industry/occupation. In addition, there are unobserved differences across individuals in general ability and other factors that may bias the estimated returns to experience and tenure. Unfortunately, there is no "clean" way to eliminate all these potential biases, particularly given the likely correlation between prior experience in a given in dustry/occupation and the unobserved quality of the match at the current job. Section 5 provides some robustness checks to test whether the results from the standard wage equations estimated here are influenced by different forms of unobserved heterogeneity.

Equations 1 and 2 are estimated using both one- and three-digit codes to identify occupation and industry of current employment and to categorize the prior experience measures. As the definition of occupation and industry becomes finer, moving from one-digit definitions to three-digit codes should provide more precise estimates of the contribution to wages of prior experience in the same occupation or the same industry as the current job.

3. Results

Table 2 presents estimates of the relation between human capital investment and wages for the homogeneous experience model and the heterogeneous experience model. Every form of traditional human capital investment, including formal schooling, cognitive talents accumulated, current workplace experience, and earlier time spent on the job, contribute significantly to a person's wage rate in each of these models of wage determination.

An additional year of prior experience boosts wages by 1.9% when all forms of prior experience are treated as if they have an equivalent impact on worker productivity. An individual's wage will rise by 2.8% if they stay with their current employer an additional year. An F-test reveals that the monetary reward to an additional year of tenure significantly exceeds the financial gain of an additional year of prior experience.

Estimates of the wage returns to all four types of prior experience indicate that there is a positive and significant relation between a person's wage and all forms of workplace experience. However, the return to prior experience gained outside both the person's current industry and occupation (ExpDODI) is significantly lower than the return to other forms of workplace experience at the one-and three-digit levels. There is no statistically significant difference between the returns to tenure and the returns to prior experience in the same occupation/same industry (ExpSOSI), same occupation/different industry (ExpSODI), and different occupation/same industry (ExpDOSI). An additional year of prior experience in both a different occupation and a different industry (ExpDODI) raises an individual's current wage only by 1.2%, which is significantly less than the 2.8% rise for tenure.

Whether there is a difference in the impact on wages of prior same occupation experience [[psi].sub.A] + [[psi].sub.B] and prior same industry experience [[psi].sub.A] + [[psi].sub.C] depends on the relative values of [[psi].sub.B] and [[psi].sub.C]. Therefore, the wage gain associated with an additional year of experience in the same occupation but in a different industry ([[psi].sub.B]) is compared with the wage gain arising from an additional year of experience in the same industry but in a different occupation ([[psi].sub.C]). The F-test reported in Table 2 indicates that the null hypothesis of [[psi].sub.B] = [[psi].sub.C] cannot be rejected. Thus, the estimates indicate that greater within-occupation and greater within-industry experience enlarge wages by equivalent amounts.

Table 3 presents estimates from wage equations where the forms of experience that are shown to provide similar returns are aggregated into single measures. In particular, one specification is estimated where all forms of prior experience other than ExpDODI are grouped into a single measure (ExpSOSI + ExpSODI + ExpDOSI). Another specification is estimated where tenure and prior experience other than ExpDODI are aggregated (Tenure + ExpSOSI + ExpSODI + ExpDOSI). Estimates at both the one- and the three-digit level again indicate that current job tenure and prior experience other than ExpDODI have very similar impacts on wages, whereas ExpDODI provides significantly lower returns than other forms of experience. In particular, the results indicate that the returns to additional year of ExpDODI are between 1.3 and 1.5%, while the returns to all other forms of experience including current job tenure are between 2.6% and 2.8%.

The estimates of the wage returns to alternative forms of human capital, including various types of prior experience, suggest that experience is largely homogeneous. For the most part, the composition of a person's workplace experience does not influence his current wage. Only a job change leading to employment in both a different occupation and a different industry results in an opportunity cost in terms of lost wages and might be considered a "career change." So, prior experience is not necessarily industry specific or occupation specific as sometimes considered but appears to be largely portable other than for experience that is completely outside either industry or occupation. Still, given that nearly half of prior experience is spent in a different occupation and a different industry, this form of experience plays a large role in earnings determination and should not be dismissed or ignored.

4. Related Findings on the Returns to Experience

Employer Learning and the Returns to Prior Experience

Employers have imperfect information about the contribution of prior skills to an individual's current productivity. If their judgment about the portability of a worker's prior skills is inaccurate, then the wage increases that they link to various forms of prior experience will be set inappropriately. Altonji and Pierret (2001) demonstrate that determinants of productivity that are difficult to observe at the time of hire will have a larger impact on wages as the firm learns more about the worker. Employers may be less able to accurately evaluate the likely contribution to current productivity of prior experience the further the proximity of this experience from the industry/occupation setting of a person's present job. For instance, the employer may be less familiar with firms in a different industry/occupation and will not be as knowledgeable about the nature of the work and be less able to contact others in the field to find out about the worker's productivity. Hence, one explanation for the relatively l ow return to remote experience--experience acquired in both a different industry and a different occupation (ExpDODI)--is that employers are less able to value the influence of this experience on a worker's current productivity.

As employers become more familiar with a worker, they should be able to more accurately assess the portability of their skills acquired on previous jobs. Thus, as a worker's tenure with a firm rises, leading the employer to learn more about the employee's skills acquired at previous jobs, the returns to experience may adjust upward for undervalued prior work experience and gravitate downward for overvalued earlier work experience.

An employer's familiarity with a worker grows with the worker's seniority or length of time they have been on the present job. Therefore, to test the hypothesis that the returns to prior experience are influenced by worker seniority, we respecify Equation 2, the heterogeneous experience model, by interacting tenure with prior experience:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (3)

Given this specification, the gain in wages due to an additional year of experience in both a different industry and occupation is [partial](In w)/[partial](ExpDOSI) = [[psi].sub.D] + [[pi].sub.D]Tenure. The change in the monetary return to this form of prior experience as Tenure advances by one year is [partial][[partial](ln w)/[partial](ExpDODI) = [[psi].sub.D] + [[pi].sub.D]Tenure]/[partial](Tenure) = [[pi].sub.D]. Therefore, the estimated coefficient on the ExpDODI* Tenure interaction term reveals how the return to this type of experience is altered by spending an additional year with the current employer.

Estimates of Equation 3 are reported in Table 4. The wage return to an additional year of prior experience is independent of a worker's seniority for three of the four types of prior experience. However, as tenure increases by a year, leading the employer to become more familiar with a worker, the wage gain to an additional year of remote experience increases significantly. This result suggests that while skills acquired in a different industry and occupation may be of less apparent value to employers, the usefulness of these more general skills is likely to be revealed as job tenure increases. The estimates from Table 2 indicate that the returns to tenure are about 2.3 times as large as the returns to ExpDODI, while the results in Table 4, where tenure and ExpDODI are interacted, indicate that the returns to tenure (evaluated at the mean tenure and experience levels) are about 1.8 times as great as ExpDODI, suggesting that the "penalty" for experience outside the current industry/occupation is about 20% les s than that indicated by the results in Table 2. Thus, the opportunity cost facing a career change diminishes with advances in seniority.

Employee Ability and the Returns to Experience

The influence of prior experience on subsequent productivity may depend on the rate at which a person learns on the job or acquires and refines skills through work. Persons of greater ability should acquire more skills during a work period than their less capable coworkers. Neal (1998) argues that more able workers have a comparative advantage in accruing highly specialized skills. If true, then the influence of an additional year of the more specific forms of experience on a person's wage should rise with ability. (7)

To determine if the returns to more specific forms of prior experience change with ability, the heterogeneous experience model is altered by interacting all work experience variables with a measure of ability:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (4)

Following Neal (1998), we use education level and AFQT score as proxies for ability. The relationship between prior experience and ability is probably not as apparent at the one-digit level because the one-digit categories are less likely to be indicative of an individual's type of work for those jobs that require higher talent levels. For instance, at the one-digit occupation level, lawyers and dentists are placed into the same broad category of "professional workers," whereas at the three-digit level, "lawyers" and "dentists" are distinct categories. Therefore, we focus primarily on the three-digit-level findings.

The results from estimating Equation 4, presented in Table 5, provide some limited evidence that the more able are more likely to gain from specialization in more specific types of job skills. The wage return to additional occupational experience, in the same industry (ExpSOSI) or in a different industry (ExpSODI), is larger for persons with greater education at the three-digit level. These estimates suggest that the highly educated earn a higher return to occupation-specific skills. Alternatively, these estimates may indicate that education assists in providing a more refined definition of a "job" than does the occupational classification. Certainly for some jobs, such as a dentist, occupational category implicitly defines education as well. Yet for other jobs, such as office managers, salesmen, or technicians, education may serve as an additional descriptor for job type. Hence, the results may imply that for particular occupations, those with more education are really in a different job type and receive gre ater wage returns than those with less education.

The estimate for the interaction between education and tenure is positive and significant, which suggests that those with more education earn greater returns to firm-specific skills. This finding is consistent with the notion that more educated workers acquire more training and develop more specialized skills and reap the returns from these investments. The estimated interaction between AFQT score and both occupation- and firm-specific skills is positive, but these estimates are imprecise.

5. Is It Just Heterogeneity?

It could be true that the lower estimated wage returns to prior experience in a different occupation and industry are due to unobserved differences between individuals or to the unobserved quality of the match between an individual and a particular sector of employment. Specifically, if individuals with lower unobserved ability are more likely to obtain experience in a different occupation and industry, the estimated return to this form of experience will be biased downward. Similarly, if those individuals with greater experience in a different occupation and industry are poorly matched in that specific sector, this also will lead to the estimated return to prior experience that is biased downward.

To provide a test as to whether these forms of heterogeneity are driving the results, the following error components model similar to that of Parent (2000) is specified:

ln [W.sub.ijkt] = [[beta].sub.1]([EXP.sub.it]) + [[beta].sub.2] ([SPECEXP.sub.ikt]) + [[mu].sub.i] + [[omega].sub.ij] + [[gamma].sub.ij] + [[epsilon].sub.ijkt], (5)

where [W.sub.ijkt] is the wage for individual i on job j in sector k at time t. The term EXP is total labor market experience, SPECEXP is prior experience specific to sector k, [[mu].sub.i] is an individual fixed effect, [[omega].sub.ij] is an individual job match fixed effect, [[gamma].sub.ik] is an individual sector match fixed effect, and [[epsilon].sub.ijkt] is a transitory mean zero error team. Unfortunately, it is difficult to eliminate all the potential biases introduced by these unobserved heterogeneity components. For instance, in order to generate estimates that eliminate the individual fixed effect ([[mu].sub.i]), at least two observations on each person are required. Similarly, in order to generate estimates that eliminate the individual job match effect ([[omega].sub.ij]), at least two observations on the same person at the same job are required.

For the purposes of estimating the returns to sector of prior experience ([[beta].sub.2]), it is of particular interest to eliminate the impact of the individual sector fixed effect ([[gamma].sub.ik]). In order to conduct this estimation, at least two observations for the same person employed in the same sector are required. Yet the prior experience variable SPECEXP changes over time only when an individual changes jobs. Hence, to estimate the returns t prior experience where the influence of the individual sector fixed effects are eliminated, at least two observations are required for individuals who changed jobs but remained in the same experience sector. Estimating a differenced equation for such individuals eliminates the influence of both the individual fixed effect and the sector-specific effect on the estimated returns to experience in a given sector. It does not eliminate however, the job match fixed effect. Since it is often thought that workers improve their match quality by moving from one job to another (even when changing sector of employment), the returns to prior experience in a given sector will be biased upward. (8) Hence, the returns to experience in a different occupation and industry, which are anticipated to be less than that of other forms of experience, will likely be biased upward using approach.

Table 6 presents results form estimating first-estimating a first-differenced wage equation between 1996 and 1993. The difference in log wages is regressed on the difference in total experience and the difference in experience in a given sector for those who changed jobs but remained in the particular experience sector as categorized by ExpSOSI, ExpSODI, ExpDOSI, and ExpDODI. It is important in mention that the EXP term in Equation 5 captures all total experience, including current job tenure. If experience is homogeneous, the estimates on the returns to experience in a particular sector will not be significantly different from zero, and the returns to EXP should not change. If experience in a given sector has greater returns than other forms of experience, the estimated coefficient for experience n that sector should be positive, and the estimated coefficient on the EXP term will fall. Conversely, if experience in a given sector offers lower returns, as thought to be the case for ExpDODI, the estimated coeff icient for this variable should be negative, and the estimated coefficient for EXP will rise.

It should also be noted that although the choice of 1993 as a comparison year to 1996 for the differenced wage equation is somewhat is somewhat arbitrary, it is used here because it allows for a comparison of experience accumulated by 1996 with experience that is not dated not too far in the remote past while also providing for an adequate sample of job changers. Hence, all individuals included in this sample have changed jobs in the past three years, and any change to their stock of prior experience is relatively recent. Since the change in prior experience is "aged" similarly for all individuals, the estimated impact of prior experience should not be influenced by any life cycle differences in the acquisition of the particular forms of experience.

In Table 6, it is shown that 505 individuals changed jobs between 1993 and 1996, and the estimates indicate that the change in total experience increased wages by about 5% over this time. Although the sample sizes for some of the specifications that are stratified by the number of job changers who remained in a particular sector are somewhat small, the estimates suggest that ExpDODI is the only form of experience that provides lower returns than the other forms of experience. That is, the estimated coefficient for this variable is negative and significant at both the one-and the three-digit level, while the other forms of experience are not significantly different from total experience. Also, the return to EXP increases in the specifications where the ExpDODI variable is included. These results for the ExpDODI term are consistent with the prior findings that this form of experience has a lower return than the other types of experience and suggest that the lower return to ExpDODI is not due simply to individua l heterogeneity or the match quality of sector of employment.

6. Conclusions

This paper provides evidence on the relation between alternative forms of experience and wages. These findings are used to draw inferences about whether experience can be treated as heterogeneous in models of wage determination. Our primary finding is that for the typical person, most forms of experience, including tenure at the current job, provide a comparable return. While there is some evidence that skills are occupation specific at higher levels of education, for the most part only skills acquired in a different industry and different occupation are of less value than other forms of experience. Consequently, a movement in both the industry and the occupation of employment appears to constitute a career change. Yet it is important to mention that career changes are very common among younger workers, as over half of all prior work experience is in a different industry/occupation among workers in their mid-30s.

Referring to the example mentioned in section 1, experience acquired while a real estate agent is valued similarly as tenure at other occupations, such as accounting, within the real estate industry. In addition, the experience as a real estate agent is valued similarly to tenure at other industries, such as the pharmaceutical industry, if continuing in the occupation of sales. If the real estate agent becomes an accountant in the pharmaceutical industry, however, the experience as a real estate agent is of less value than that within accounting or the pharmaceutical industry.

The results also indicate that prior experience outside the current industry/occupation increases in value as current job tenure rises. This suggests that certain "remote" skills reveal themselves to employers and become more valuable over time. Hence, for the real estate agent who becomes an accountant in the pharmaceutical industry, while the sales skills or knowledge of the real estate industry may seem of little immediate use to an employer, some of this knowledge becomes useful as the employer becomes more acquainted with the worker.

Appendix A

Sample Appendix

Simple creation
  NLSY total sample in 1996           9964
  Respondents in 1996                 8636
Deletions                                   Remaining Sample
  Female                              4361  4275
  Nonwhite                            2122  2153
  Missing/invalid hours for 1996 job   430  1723
  Missing/invalid wage                  47  1676
  Missing/invalid tenure                92  1584
  Missing industry/occupation codes     81  1503
  Missing AFQT score                    67  1436
  Missing Urban variable                10  1426
Appendix B

Sample Means

Variable                              Mean

Tenure                                5.68
Experience                            8.28
Education                            13.54
Standardized AFQT score               0
Urban locality                        0.74
Married                               0.68
Tenure > 1 year                       0.80
Professional and technical            0.19
Manager                               0.19
Sales                                 0.05
Clerical                              0.05
Operative                             0.14
Laborers and farmers (omitted)        0.09
Craft worker                          0.23
Service and private household         0.06
Agriculture and mining                0.04
Construction                          0.14
Manufacturing (omitted)               0.23
Transportation                        0.08
Wholesale and retail trade            0.16
Finance                               0.05
Business                              0.09
Personal services and entertainment   0.03
Professional services                 0.12
Pubic administration                  0.06
Lnwage                                2.68
Table 1

Types of Experience

                                                  One-Digit
                                                       % Prior
Variable                                    Mean      experience

Years of tenure (Tenure)                    5.78          -
Years of experience (Exp)                   8.28        100%
Years of experience: Same occupation        2.86         35%
Years of experience: Different occupation   5.42         65%
Years of experience: Same industry          3.10         37%
Years of experience: Different industry     5.18         63%
Years of experience: Same occupation/same   1.54         17%
 industry (ExpSOSI)
Years of experience: Same occupation/       1.30         14%
 different industry (ExpSODI)
Years of experience: Different occupation/  1.54         18%
 same industry (ExpDOSI)
Years of experience: Different occupation/  3.90         51%
 different industry (ExpDODI)
Number of observations                      1426

                                                 Three-Digit
                                                       % Prior
Variable                                    Mean      experience

Years of tenure (Tenure)                    5.78          -
Years of experience (Exp)                   8.28        100%
Years of experience: Same occupation        1.36         16%
Years of experience: Different occupation   6.92         84%
Years of experience: Same industry          1.54         19%
Years of experience: Different industry     6.74         81%
Years of experience: Same occupation/same   0.66          8%
 industry (ExpSOSI)
Years of experience: Same occupation/       0.70          8%
 different industry (ExpSODI)
Years of experience: Different occupation/  0.88         11%
 same industry (ExpDOSI)
Years of experience: Different occupation/  6.04         73%
 different industry (ExpDODI)
Number of observations
Table 2

Earnings Functions Estimates: Heterogeneous Experience Model

                                                  Heterogeneous
                                                Experience Model
                                Homogeneous
Variable                      Experience Model     One-Digit

Education                           .055 **          .055 **
                                   (.008)           (.008)
AFQT score                          .071 **          .069 **
                                   (0.18)           (.018)
Tenure                              .028 **          .028 **
                                   (.004)           (.004)
Exp                                 .019 **
                                   (.004)
ExpSOSI                                              .025 **
                                                    (.006)
ExpSODI                                              .028 **
                                                    (.006)
ExpDOSI                                              .020 **
                                                    (.006)
ExpDODI                                              .012 **
                                                    (.005)
[H.sub.0]: Tenure = Exp             6.32 **
                                   [.012]
[H.sub.0]: Tenure = ExpSOSI                          0.32
                                                    [5.70]
[H.sub.0]: Tenure = ExpSODI                          0.00
                                                    [.981]
[H.sub.0]: Tenure = ExpDOSI                          2.14
                                                    [.144]
[H.sub.0]: Tenure = ExpDODI                          13.33 **
                                                    [.001]
[H.sub.0]: ExpSODI = ExpDOSI                         1.32
                                                    [.251]
Adjusted [R.sup.2]                                    .32
N                                                    1426

                                 Heterogeneous
                               Experience Model

Variable                         Three-Digit

Education                           .055 **
                                   (.008)
AFQT score                          .069 **
                                   (.018)
Tenure                              .028 **
                                   (.004)
Exp

ExpSOSI                             .024 **
                                   (.008)
ExpSODI                             .031 **
                                   (.008)
ExpDOSI                             .031 **
                                   (.008)
ExpDODI                             .015 **
                                   (.004)
[H.sub.0]: Tenure = Exp

[H.sub.0]: Tenure = ExpSOSI         0.60
                                   [.418]
[H.sub.0]: Tenure = ExpSODI         0.88
                                   [.348]
[H.sub.0]: Tenure = ExpDOSI         0.23
                                   [.629]
[H.sub.0]: Tenure = ExpDODI         13.40 **
                                   [.001]
[H.sub.0]: ExpSODI = ExpDOSI        0.18
                                   [.719]
Adjusted [R.sup.2]                   .33
N                                   1426

Other covariates include marital status, occupation, industry, first
year at current job, and urban location dummies. Standard errors are
shown in parentheses (rounded to .001). F-statistics and their
associated p-values shown in square brackets are reported for tests of
differences in the wage return of greater tenure or experience. Wages
are measured in 1996 dollars.

* Statistically significant at the 90% level.

** Statistically significant at the 95% level.
Table 3

Earnings Functions Estimates: Industry Experience versus Tenure Effect
and Occupation Experience versus Tenure Effect

                                Exp = ExpSOSI + ExpSODI + ExpDOSI
Variable                          One-Digit         Three-Digit

Education                           .055 **             .055 **
                                   (.008)              (.008)
AFQT score                          .069 **             .069 **
                                   (.018)              (.018)
Tenure                              .028 **             .028 **
                                   (.004)              (.004)
Exp                                 .024 **             .029 **
                                   (.004)              (.005)
ExpDODI                             .013 **             .015 **
                                   (.005)              (.014)
[H.sub.0]: Tenure = Exp            1.15                 .080
                                   [.284]              [.784]
[H.sub.0]: Tenure = ExpDODI       13.13 **            12.64 **
                                   [.000]              [.000]
[H.sub.0]: Exp = ExpDODI           7.01 **             9.51 **
                                   [.008]              [.002]
Adjusted [R.sup.2]                  .32                 .33
N                                   1426                1426

                                Exp = Tenure + ExpSOSI
                                  +ExpSODI + ExpDOSI
Variable                             One-Digit

Education                               .055 **
                                       (.008)
AFQT score                              .070 **
                                       (.018)
Tenure

Exp                                     .026 **
                                       (.004)
ExpDODI                                 .013 **
                                       (.005)
[H.sub.0]: Tenure = Exp

[H.sub.0]: Tenure = ExpDODI

[H.sub.0]: Exp = ExpDODI              12.21 **
                                       [.001]
Adjusted [R.sup.2]                      .33
N                                       1426

                               Exp = Tenure +
                             ExpSOSI +ExpSODI +
                                   ExpDOSI
Variable                        Three-Digit

Education                          .055 **
                                  (.008)
AFQT score                         .069 **
                                  (.018)
Tenure

Exp                                .028 **
                                  (.004)
ExpDODI                            .015 **
                                  (.004)
[H.sub.0]: Tenure = Exp

[H.sub.0]: Tenure = ExpDODI

[H.sub.0]: Exp = ExpDODI         15.81 **
                                  [.001]
Adjusted [R.sup.2]                 .33
N                                  1426

Other covariates include marital status, occupation, industry, first
year at current job, and urban location dummies. Standard errors are
shown in parentheses (rounded to .001). F-statistics and their
associated p-values shown in square brackets are reported for tests of
differences in the wage return of greater tenure or experience. Wages
are measured in 1996 dollars.

* Statistically significant at the 90% level.

** Statistically significant at the 95% level.
Table 4

Earnings Functions Estimates: Experience Variable Interacted with Tenure

                              Heterogeneous
                            Experience Model:
                                Exp-Tenure
                            Interaction Terms
Variable                 One-Digit       Three-Digit

Education                  .056 **          .056 **
                          (.008)           (.008)
AFQT score                 .069 **          .068 **
                          (.018)           (.018)
Tenure                     .023 **          .023 **
                          (.005)           (.005)
ExpSOSI                    .021 **          .017 **
                          (.007)           (.010)
ExpSODI                    .030 **          .035 **
                          (.008)           (.011)
ExpDOSI                    .018 **          .034 **
                          (.008)           (.011)
ExpDODI                    .004             .009 **
                          (.006)           (.005)
ExpSOSI * Tenure           .0011            .0015
                          (.0017)          (.0032)
ExpSODI * Tenure          -.0008           -.0002
                          (.0019)          (.0031)
ExpDOSI * Tenure           .0004           -.0009
                          (.0016)          (.0022)
ExpDODI * Tenure           .0027 **         .0019 **
                          (.0011)          (.0009)
Adjusted [R.sup.2]         .33               .33
N                          1426             1426

Other covariates include marital status, occupation, industry, first
year at current job, and urban location dummies. Standard errors are
shown in parentheses (rounded to .001). F-statistics and their
associated p-values shown in square brackets are reported for tests of
differences in the wage return of greater tenure or experience. Wages
are measured in 1996 dollars.

* Statistically significant at the 90% level.

** Statistically significant at the 95% level.
Table 5

Earnings Functions Estimates: Experience Variables Interacted with
"Ability" Measures

                       Heterogeneous Experience Model:
                        Exp-Apility Interaction Terms
                            "Ability" = Education
Variable               One-Digit        Three-Digit

Education                .036 *            .032
                        (.020)            (.020)
AFQT score               .067 **           .063 **
                        (.018)            (.018)
Tenure                  -.016             -.020
                        (.020)            (.020)
ExpSOSI                 -.017             -.048
                        (.029)            (.041)
ExpSODI                 -.001             -.051
                        (.035)            (.043)
ExpDOSI                  .034              .033
                        (.032)            (.039)
ExpDODI                  .030              .023
                        (.026)            (.022)
Tenure * Ability         .003 **           .004 **
                        (.001)            (.001)
ExpSOSI * Ability        .003              .005 *
                        (.002)            (.003)
ExpSODI * Ability        .002              .007 **
                        (.003)            (.003)
ExpDOSI * Ability       -.001             -.001
                        (.002)            (.003)
ExpDODI * Ability       -.001             -.001
                        (.002)            (.002)
Adjusted [R.sup.2]       .33               .33
N                        1426              1426

                        Heterogeneous Experience Model:
                         Exp-Apility Interaction Terms
                            "Ability" = AFQT score
Variable                   One-Digit          Three-Digit

Education                    .055 **             .055 **
                            (.008)              (.008)
AFQT score                   .056                .052
                            (.053)              (.053)
Tenure                       .028 **             .028 **
                            (.004)              (.004)
ExpSOSI                      .025 **             .025 **
                            (.006)              (.008)
ExpSODI                      .028 **             .037 **
                            (.006)              (.008)
ExpDOSI                      .021 **             .032 **
                            (.006)              (.008)
ExpDODI                      .013 **             .015 **
                            (.005)              (.004)
Tenure * Ability             .004                .004
                            (.004)              (.004)
ExpSOSI * Ability            .006                .012
                            (.006)              (.008)
ExpSODI * Ability           -.001                .005
                            (.006)              (.008)
ExpDOSI * Ability           -.006               -.004
                            (.006)              (.007)
ExpDODI * Ability           -.002               -.003
                            (.005)              (.004)
Adjusted [R.sup.2]           .33                 .33
N                            1426                1426

Other covariates include marital status, occupation, industry, first
year at current job, and urban location dummies. Standard errors are
shown in parentheses (rounded to .001). F-statistics and their
associated p-values shown in square brackets are reported for tests of
differences in the wage return of greater tenure or experience. Wages
are measured in 1996 dollars.

* Statistically significant at the 90% level.

** Statistically significant at the 95% level.
Table 6

First-Differenced Earnings Function Estimates

Variable
Difference in:                One-Digit   Three-Digit  One-Digit

Exp                  .051 **     .050 **     .023        .044 **
                    (.008)      (.006)      (.029)      (.020)
ExpSOSI                         -.012       -.005
                                (.012)      (.26)
ExpSODI                                                  .007
                                                        (.62)
ExpDOSI

ExpDODI

Adjusted [R.sup.2]   .08          .06        .02         .09
N                    505          137         69          91

Variable
Difference in:      Three-Digit  One-Digit  Three-Digit  One-Digit

Exp                    .056 **     .066 **     .044 **     .074 **
                      (.030)      (.023)      (.026)      (.017)
ExpSOSI

ExpSODI                .002
                      (.019)
ExpDOSI                           -.002        .001
                                  (.026)      (.021)
ExpDODI                                                   -.019 **
                                                          (.008)
Adjusted [R.sup.2]      .11        .10         .06         .09
N                        61         99          67          178

Variable
Difference in:      Three-Digit

Exp                    .099 **
                      (.014)
ExpSOSI

ExpSODI

ExpDOSI

ExpDODI              -0.25 **
                      (.006)
Adjusted [R.sup.2]     .13
N                       308

The dependent variable = Log Wage 1996 - Log Wage 1993. Both the
dependent and independent variables represent the change in values
between 1996 and 1993. Wages measured in 1996 dollars. Standard errors
are shown in parentheses (rounded to .001). Wages are measured in 1996
dollars.

* Statistically significant at the 90% level.

** Statistically significant at the 95% level.

Received October 2001; accepted March 2002.

(1.) From 1979 to 1985, industry and occupation codes are available only for those jobs that involved working at least nine weeks and 20 hours per week. Starting with the 1986 interview, the 20-hours-per-week restriction was reduced to 10 hours per week.

(2.) In his discussion of misclassification errors and data edits, Neal (1999) asserts that industry and occupation codes are available for all jobs that involved working at least nine weeks and 10 hours per week in the NLSY. However, as mentioned in the text, this is true only starting with the 1986 interview.

(3.) Squared and cubic terms of the tenure and experience variables are excluded for ease of presentation. The results reported here are qualitatively similar when the squared and cubic terms are included.

(4.) The less-than-one-year-tenure dummy variable is included in the specifications estimated by Parent (2000).

(5.) The AFQT was administered to all respondents in 1980. The scores used in the estimations are age adjusted and normalized to have a standard deviation of one.

(6.) The coefficient on a variable that appears in more than one equation will carry a subscript equivalent to the number of the estimating equation. Letter subscripts are used for coefficients on prior experience variables that appear in only one equation.

(7.) Neal (1998) finds that more able workers are less mobile and accumulate more specialized skills relative to less able counterparts because the return to specific training is likely greater for more able workers.

(8.) Workers are generally thought to improve their job match when changing jobs because the majority of job separations are voluntary (for a more detailed discussion of this issue, see Parent 2000). In the sample of job changers analyzed here, between 60% and 70% of those who changed jobs reported doing so voluntarily for each of the sectors of prior experience.

References

Altonji, Joseph, and Charles Pierret. 2001. Employer learning and statistical discrimination. Quarterly Journal of Economics 16:313-50.

Becker, Gary. 1962. Investment in human beings. Journal of Political Economy 70:9-49.

Bureau of Labor Statistics. 2000. Number of jobs held, labor market activity, and earnings growth over two decades: Results from a longitudinal survey. News Release 00-119. Washington, DC: Bureau of Labor Statistics.

Loewenstein, Mark A., and James R. Spletzer. 1998. Dividing the costs and returns to general training. Journal of Labor Economics 16:142-71.

Mincer, Jacob. 1974. Schooling, experience, and earnings. New York: Columbia University Press.

Neal, Derek. 1995. Industry-specific human capital: Evidence from displaced workers. Journal of Labor Economics 13:653-77.

Neal, Derek. 1998. The link between ability and specialization. Journal of Human Resources 33:173-200.

Neal, Derek. 1999. The complexity of job mobility among young men. Journal of Labor Economics 17:237-61.

Parent, Daniel. 2000. Industry-specific capital and the wage profile: Evidence from the national longitudinal survey of youth and the panel study of income dynamics. Journal of Labor Economics 18:306-23.

Shaw, Kathryn. 1984. A formulation of the earnings function using the concept of occupational investment. Journal of Human Resources 14:319-40.

Arthur H. Goldsmith * and Jonathan R. Veum +

* Department of Economics, Washington and Lee University, Lexington, VA 24450, USA; E-mail [e-mail]; corresponding author.

+ Freddie Mac, 8200 Jones Branch Drive, McLean, VA 22102, USA; E-mail [e-mail].

The authors are grateful to two anonymous referees for helpful suggestions. The authors also wish to thank Stuart Low for his contributions during the early stages of this paper. In addition, they wish to thank Noel Gaston who offered valuable insights and suggestions during numerous discussions of the ideas in this paper, especially when Goldsmith was a Visiting Professor of Economics at Bond University, Gold Coast, Australia. Summer Research Grants from Washington and Lee University provided financial support for this research. Finally, the views expressed in this paper are those of the authors and do not reflect the policies or views of Freddie Mac.