The Role of Early Career Experiences in Determining Later Career Success: An International Comparison

First version : January 1999 This version : February 1999

David N. Margolis*, Erik Plug†, Véronique Simonnet‡ and Lars Vilhuber§

*

CNRS, TEAM-Université de Paris 1 Panthéon-Sorbonne and CREST University of Amsterdam ‡ TEAM-Université de Paris 1 Panthéon-Sorbonne § York University †

1. Introduction This paper represents an attempt to untangle the theoretically complicated, and empirically uncertain, links between the early career experiences of young people in the labor market and their labor market success or failure later in life. We approach this topic from 2 different perspectives. First, we consider many different measures of early-career experiences and later career results, and we look at outcomes measured between 5 and 20 years after the time of school leaving. Second, we use long longitudinal data sets from four countries (the United States, France, Germany and the Netherlands) with vastly differing institutions to see if workers in different labor markets are treated in the same manner when they have similar early career experiences. Although the subject of early-career experiences, such as “excessive” job mobility or taking a long time to find a first job, have already been treated in the literature,1 very little attention has been given to the effects at a much longer term (i.e. at least 5 years after school leaving) of these experiences.2 In most of these studies, the authors focus on a single measure of early career experiences and consider the effects from a short-term perspective, such as over the first 2 years after school leaving or in the first job found after finishing school. 1

See, for example, Topel and Ward (1991) for the United States and Balsan, Hanchane and Werquin (1989) for France. 2 Gardecki and Neumark (1998) represents a (rare) recent example for the United States that uses the same data as us. There are a number of important differences between their approach and ours. First, whereas we use information on all jobs to construct our measures of labor market insertion, they restrict their attention (in their main analysis) to the most recent job at interview time. Given the short average tenure on jobs in the initial period, this is likely to lose a lot of information, especially on industry and occupational mobility, but also on overall time spent working. Second, whereas Gardecki and Neumark concentrate on a fixed calendar date (1990-1992), we have fixed the time since labor market insertion to five years and more, each year generating a new observation on each individual. Third, Gardecki and Neumark consider the transition to occur at the latest when individuals leave two-year colleges, considering any education occurring after this transition to be training, while we focus less on the specific (high-)school-to-work transition and more on the actual entry into the labor market, irrespective of the educational attainment at the time of that insertion. Both approaches will tend to take the education-work decision as exogenous, but ours will tend to capture the actual process of insertion better, whereas theirs will describe the exit from (high-)school better. Finally, and related to the previous point,

1

2 Furthermore, these studies tend to consider only one “output” measure, typically log hourly wages. Our study expands upon this literature in several respects. First, we use a large set of measures of early career experiences in an attempt to control for the omitted variable bias that has made interpretation of many previous results risky. In particular, we control not only for the number of different employers an individual has had during the first two years after leaving school (by 6-month intervals) and the time it took to find the first job (be it any job or a job lasting at least 6 months), but many other measures as well. For example, we control for the number of different occupations and industries that and individual has tried during the first 6/12/18/24 months after school leaving, the percentage of time spent employed during the first 6/12/18/24 months, the average job duration during the first year and the first two years, and earnings growth between the first and second year after school leaving. Second, we consider a variety of different measures of later career success.

In

particular, we study the percentage of time in a given year spent in employment and (where possible) the following earnings measures: log hourly wage, log monthly earnings and log full year equivalent earnings. This set of measures allows us to control not only for earnings variation, but also hours variation while employed and differences in employment rates. Although using many different controls for early and later career experiences can render interpretation of the results more complex, the advantages of such an approach are numerous. First, the diversity of outcome measures provides us with a more complete view of the role of early career experiences in subsequent labor market success than was previously available. Second, the variety of early career measures allows us to study directly phenomena that have been proposed by the theoretical literature but that have not previously been testable.

Gardecki and Neumark consider “churning” measures over the five years following the exit from school, whereas we consider only a two-year period.

3 Finally, by including many different measures of early career experiences, we reduce the omitted variable bias in the parameter estimates that is found in many other studies. To answer these questions, we use long (at least 10 year) panel data sets available in the United States (the National Longitudinal Survey of Youth), France (the DADS Annual Social Data Reports), Germany (the GSOEP German Socio-Economic Panel) and the Netherlands (the OSA Panel Survey). As is often the case with international comparisons, not all variables can be calculated on all data sets. Therefore, for each country we report those results that can be derived from the relevant data, although we have made a considerable effort to harmonize the definitions of the different variables that do appear. The rest of the paper is organized as follows. Section 2 provides some very brief theoretical foundations for the analyses undertaken here. Section 3 describes the different data sets that are exploited in the econometrics, and section 4 discusses the descriptive statistics from the different countries to provide a basis for interpreting the results. Section 5 presents the results from the various estimations. Section 6 uses the results from section 5 to conduct several thought experiments concerning the role of early career experiences on later career outcomes. Section 7 discusses avenues for future research and concludes.

2. Theory The literature in labor economics has much to say about why early career experiences should be observable as affecting later career success. The literature can be broken down into 3 main strands: information-based learning models, sorting models and human capital models.

4 2.1 Information-Based Learning Models The information-based learning models3 maintain that, since information is not symmetric and complete in the labor market, agents will learn about unknown characteristics over time. This learning is typically modeled in a bayesian manner, with either employers or workers starting from a general prior distribution of beliefs over the unknown characteristics and using information that is revealed over time as a means of refining their beliefs. Since the fastest revisions of priors occur with the earliest observations, early career experiences could play an important role in determining later posterior distributions of beliefs over unknown characteristics, which should be correlated with observable “output” measures. One implication of these theories is that, as time goes by, the early-career experiences play a less and less important role in the determination of the posterior distribution of beliefs. Econometrically, this implies that the significance of early career experiences in determining later career outcomes should be decreasing over time. Although the literature has not typically posed the question in these terms, related work by Farber and Gibbons (1996), Altonji (1998) and Simonnet (1997) all show that firms use initially observable characteristics less and less in determining remuneration as time goes on, which the various authors interpret as employers relying increasingly on the (unobservable by the econometrician) posterior beliefs induced by years of repeated observation.

2.2 Sorting Models The fundamental hypothesis of sorting models is the same as that of learning models, namely that initially information is not symmetric and complete. However, sorting models differ in that, as information is revealed about the worker, the market assigns him or her to the

5 appropriate job. The founding paper in the signaling branch of this literature (Spence (1973)) was concerned with educational attainment signaling unobserved worker productive ability so that workers could be paid in relation to their productivity. The seminal paper in the dual labor markets branch (Roy (1951)) suggested that their were two sorts of jobs, and that queues for the good jobs allowed good employers to select the good workers, relegating the bad workers to the bad sector or to unemployment. In more complicated multiattribute versions, one can imagine this sorting mechanism assigning workers to the sectors which most highly value their characteristics. Econometrically, these models imply that the early career experiences will have an impact on the type of job the person has in the future, and thus on certain observable output measures. Furthermore, this model suggests that the importance of these initial experiences in determining later career outcomes does not diminish over time. This is because the assignment of a worker to a sector happens early, and since nothing changes after the initial assignment, the correlation between early career experiences and later career outcomes is constant. The good jobs/bad jobs model is at the base of most econometric studies of union non-union wage differentials (Lewis (1986)) as well as many other papers, whereas the pure signaling model has proved notoriously hard to test empirically, as its observable implications are often identical to those of other models, human capital models in particular.

2.3 Human Capital Models These models, originating with Becker (1993), are the workhorses of the labor economics literature, suggesting that workers’ characteristics change over time as they learn new things or as their old knowledge becomes outdated. The Ben-Porath (1967) life-cycle 3

Some of the earliest examples are Jovanovic (1979) and Miller (1984). More recent work includes Farber and

6 version of this model implies that the greatest investments in learning will be made at the beginning of the career, where there returns can be reaped over a longer time in the future. Other refinements of these models have typically be based on decomposing human capital into different types. Becker’s initial decomposition into general and specific human capital has been criticized as being too rigid, and recent work has proposed that human capital may be semispecific to the firm (of lower, yet positive value, to at least some firms other than the current employer (Stevens (1994))), occupation-specific (of equal value to all employers provided that the worker performs the same occupation as with the current employer) or sector specific (of equal value to all firms in the same sector as the current employer).4 The empirical implications of this type of model depend on the variant retained. In the sector specific approach, fewer different sectors for a given number of days worked implies the accumulation of more human capital, and an equivalent reasoning holds for occupational capital models.

In models where the human capital does not depreciate over time, the

significance of these effects should remain constant, as the skills learned early in the career can still be used later. In models where human capital depreciates, the capital garnered at the beginning of a career becomes less and less relevant over time, and thus measures of it should become less and less significant.

3. Data In this section, we discuss briefly, country by country, the data upon which our estimations are performed.

Gibbons (1997). 4 See Neal (1995, 1997), Parent (1996) and Vilhuber (1997, 1999) for tests of the different specifications.

7 3.1 The United States The data used here were extracted from the 1979-1993 NLSY data files. We exclude all oversamples, as well as the military subsample, leaving us with 4043 individuals. To be included in the regressions, an individual needs to have been in school at the time of the 1979 interview; this allows us to observe the transition into the labor force. The individual's “permanent” exit from school is computed by requiring an exit from school to be followed immediately by at least two full years during which the individual never returned to school.5 Given the widespread availability of part-time courses offered by high schools, community colleges, and universities, this “clean” transition is likely to be an idealization of the true transition. However, since we try to capture a “permanent” exit from school, this seems to be a reasonable definition. By construction, the chosen definition of school exit directly determines the definition of initial post-schooling labor market experience.6 In order to gather sufficient post-schooling data, we require the exit from school to occur before 1987, which leaves a minimum of 6 post-schooling years from which to compute outcome measures. The date of school exit is then taken to be the first interview for which these conditions are met. This criterion is satisfied by 78.28 percent of the above 4 043 persons. A number of individuals have missing data, leaving us with a maximal sample size (depending on the regression specification) of 2974 individuals. Table 1a provides sample statistics, while table 2a describes the evolution of the sample in terms of entry cohorts and observation years.

5

The date of school exit is taken to be the first interview for which these conditions are met. This date is then refined if the individual reports the exact month of school exit. If not, the interview date is taken as the exit date. 6 An alternative to the above “clear-cut” solution would be to include jobs held before the permanent exit from school. This, however, runs the risk of confusing some individuals' post-school jobs, which happen to overlap with post-school ongoing education, with other individuals' high-school jobs (Ruhm (1995)). In the present paper, we stick with the “clear-cut ” distinction, though further work will refine this argument.

8 3.2 France The French data are drawn from the Déclarations Annuelles des Données Sociales (DADS), or Annual Social Data Reports, from 1976 through 1996. These data are a 1/25th random sample of the French population, selected by the individual’s birth date. The data contain earnings, occupation, sector, region and days worked information for each year and each enterprise-individual match. The data also contain individual and enterprise identifiers that allow the researcher to follow both individuals and firms over time, and to know some basic characteristics of each. It should be noted that data from 1981, 1983 and 1990 were not made available to us, and as such we restrict our attention to school leaving cohorts 19761978, 1984-1987 and 1991. Furthermore, data for observation years 1981, 1983 and 1990 are unavailable. For one-tenth of the DADS data, we observe the level of education (in 8 degree categories, including “no known educational certification”) and the age at school leaving with a supplementary data set to the DADS called the Echantillon Démographique Permanent (EDP), or Permanent Demographic Sample. We restricted our attention to this subset of the DADS. In the EDP, age at the end of schooling is measured in completed years. Therefore, we use 2 different measures of the date at which an individual leaves school. The first assumes all individuals leave school on July 1 of the year in which they have the declared age at the end of schooling. The other exploits a third data set called the Enquête Emploi, or Labor Force Survey, which contains a monthly calendar allowing us to observe the month of school exit. We use this information to estimate a multinomial logit model of month of school leaving, conditional on sex, department of birth and month of birth (the only variables that are observable before school leaving, exogenous and common to all data sets), and then use the

9 coefficients of this model to predict a month of exit for each individual in the DADS.7 As a check on the validity of the imputed school leaving date, we compare this to the start date of the first job observed in the data. When the job started after the imputed school leaving date, we maintained this date. When the job began less than a year prior to the imputed exit date, we used the imputed exit date minus 1 year, to account for the possibility that the age at school leaving was mismeasured. When the first job began more than a year prior to the imputed school leaving date, we replaced the imputed date with the start date of the first job, reflecting the idea that the worker introduction to the labor market actually began prior to the end of formal schooling. In what follows, we report results using the second measure (where the month of school exit varies); results using the first measure are in all cases similar. Table 1b provides sample statistics, while table 2b describes the evolution of the sample in terms of entry cohorts and observation years.

3.3 Germany In Germany, we use the German Socio-Economic Panel (GSOEP, 1984-1996), a 13year household panel survey. We restrained our attention to 16 to 45 years old German nationals or foreigners residing in western Germany (the former Federal Republic of Germany) because data for eastern Germany only became available as of 1990. The GSOEP contains a total (over all years) of 11081 potentially eligible individuals, of which 5550 are men. Unfortunately, for reasons of attrition and sample replenishment, only 6300 individuals (approximately) are present in any given year..8 This problem with the structure of the GSOEP reduces our sample sizes dramatically. 7

We suppose that the individual left school on the last day of the month preceding the month of school exit predicted by the multinomial logit. 8 We lose many observations in part due to missing data, but most observations are lost because of the high rate of attrition in the GSOEP. Moreover, because new observations are added to the database each year (when new

10 In order to observe the impact of early career experiences on later career experiences starting at least 4 years after the school leaving date, we kept only those people who left school between 1984 and 1991, which left us with 2551 individuals, of which 1396 were men. As the year and the month of school leaving were not directly available as survey variables, we used the survey’s monthly activity calendar and the same school ending definition as in the United States. When an individual stops declaring that he is still undertaking his “initial education” and does not resume “initial education” for at least 2 years, we define school as having ended (at the first date following the end of “initial education”), and we begin to measure his early career experiences. As our study requires observations on individuals over a long period of time (we need to observe individuals each of the three years after their school leaving date to reconstruct their early career events, and individuals are required to provide some information regarding their employment status 4, 5, 6, 7 or 8 years after the end of school for use as outcome measures), our analysis sample represents a relatively small subsample of the initially available data. Of the 1396 men in our sample, only 624 were present 4 years after the end of schooling and provided information regarding their early career events, while 356 are present 8 years after. These numbers are reduced to 474 and 240, respectively, when we consider only people who are working at the time of the interview 4 to 8 years after the end of school.9 For the women (1155 in the sample), 487 were present 4 years after the end of school and provided information on their early career experiences and 265 of them are present 8 years after school leaving, but only 355 and 147, respectively, were employed at the interview date.

individuals join a household which has already been interviewed or if a child of this household reaches 15 years old, the individual becomes a new respondent in the survey), the total number of individuals available is a gross overestimate of the number of usable observations in the German data. 9 We are only able to observe earnings and hours information for people who were employed at the interview date, and thus this represents a large loss in the number of observations that will be available for the earnings regressions relative to the employment models.

11

3.4 The Netherlands For the Netherlands, we employ the OSA panel Survey.

These data include

information on individual schooling duration, labor market status and, earnings for the following years: 1985, 1986, 1988, 1990, 1992, 1994 and 1996. Each wave contains about 4500 observations and is intended to be representative for the Dutch labor force. Because we are interested in the role of early career experiences on later career outcomes and we define these later career outcomes to refer to at least 5 years since school leaving, we exploit the OSA panel survey in two ways. We construct our insertion variables based on information from the 1985, 1986, 1988 and 1990 waves; from these waves we are able to identify 199, 523, 70 and 90 school leavers, respectively, whom we follow for at least two years. Differences in the number of school leavers are caused by differences in the consecutive surveys. The 1985 wave refers to the period 1980 to 1985. The 1986 wave refers to the period 1980 to 1986 and can easily be linked to the 1988 wave. The 1988 and the 1990 wave cover only two years, 1986 to 1988 and 1988 to 1990. Both waves can be linked to the 1990 and 1992 waves. If we use the first five waves, we end up with a sample 884 observable school leavers. We construct the variables that refer to the later career outcomes from all seven waves of the OSA data. Linking the 884 school leavers to their later working career, we are left with 662 observations, of which 318 are female. Table 1d provides sample statistics for our Dutch data, while table 2d describes the evolution of the sample in terms of entry cohorts and observation years.

12

4. Descriptive Statistics Tables 1a-1d provide descriptive statistics for the various data sets we analyze below. In this section, we characterize the early career experiences in our four countries and describe the samples on which we estimate our models.

4.1 “Job Search” Variables In each of our countries, we calculate two measures of job search experiences: time to first job and time to first job of at least 6 months. Job search theory suggests that as reservation wages increase, offer arrival rates decrease, or the wage offer distribution declines, the time to first job should get longer. If the difference in time to first job and time to first job of at least 6 months is large, this suggests a high separation rate that could be either exogenous to the search process (layoffs) or endogenous (high offer arrival rates for on-the-job search). From a definitional perspective, there are several differences across the countries. In the United States, France and Germany, we excluded workers who did not find their first job of at least 6 months duration from the analyses. Descriptive statistics on the excluded workers suggest that there are indeed differences between the retained and excluded samples in the regressors and dependent variables, and thus in future work we intend to correct for the likely induced selection bias.10 In the Netherlands, the time to first job and time to first job of at least 6 months variables were top-coded at the maximum duration between school leaving and 5 years after.11 This was done in order to maintain reasonable sample sizes, although it can bias our estimates. Future work will attempt to impute an expected duration for the top-coded observations. 10

Another source of selection bias, coming from our sample construction criteria (such as requiring nonmissing values for the early career and later career variables), could also be considered at a later stage.

13 For all of our countries, it is clearly the United States that has the shortest observed time to first job, at 0.304 years (around 16 weeks) on average for men and 0.362 (around 19 weeks) for women. This is consistent with the differing labor market institutions, since school leavers are ineligible for most sorts of government transfers in the United States; this implies a low reservation wage and a subsequently faster first-job finding rate. The Netherlands is close behind, with an average time to first job of 0.671 years (35 weeks) for men and 0.514 years (27 weeks) for women, while in France and Germany it takes, on average, over a year to find the first job. Much of the difference between the European and U.S. data can be attributed to mandatory military service. In the Dutch data, military service is observed and is not counted as time spent in the labor market, i.e. if a person had not found a job in the 3 months preceding the start of his military service, but found one 1 month after, the time to the first job would be 4 months. For France, military service (of 10-12 months) is not observed at all. Since it is only mandatory for men (and many men avoid it though various means12) this could explain why French women find their first jobs earlier than French men. In Germany, men have the option of choosing civil service instead of military service. The duration of military service was reduced from 15 to 12 months on January 1, 1990, but men who chose civil service (roughly half) had a 18 month obligation (which was reduced to 15 months in 1990).13 In our analyses, civil service is counted as employment, whereas military service is not. Given these differences in the treatment of military service, a striking pattern across countries becomes evident. Men typically seem to find their first jobs faster than women, and

11

In other words, for those people who do not participate in military service (see below), the values are topcoded at 5. For all others, they are top coded at 5 minus the duration of the military service. 12 Aside from those men exempted or disqualified from military service, the most recent statistics (1997) from the French Ministry of Defense suggest that 16 percent of those who undertake their “national service” do not serve in the military (http://www.defense.gouv.fr/sn/dformes/dformes.html). 13 Roughly 11 percent of the German sample left school in 1989 or earlier. See table 2c for details.

14 the time to first job (net of the expected maximum time spent in the military) varies for men from 0.30 in the U.S. to 0.36 in France to 0.44 in Germany to 0.67 in the Netherlands.14 On the other hand, the military service issues typically do not affect women, and thus the differences there are more striking. Women in the United States and the Netherlands clearly find their first jobs faster than in France or Germany.15 Although some of this effect may be due to differences in female labor force participation rates, non-participation is unlikely to be able to explain the large differences between the two groups of countries. Not only are the (uncorrected) means for the time to first job very different, but the standard errors are as well. There seems to be very little variability in the time to first job in the United States relative to the European countries, and in particular relative to France and Germany. As with the means, part of this difference in variability may be due to the treatment of military service, as some men in France and Germany have an extra year to 15 months added to their real time to first job according to our measures, and some do not. Still, the differences persist for the women, and there is little difference within France and Germany across sexes in the variance of time to first job. All of this evidence taken together suggests that there may be substantially less variability in the United States with respect to the underlying reservation wages or offer arrival rates that determine this duration. In considering the time to the first job of at least 6 months, most of the aforementioned conclusions apply.

The United States stands out for the speed with which workers are

integrated into “long” jobs; on average, men only take an additional 8.4 weeks, and women an additional 7.6 weeks to find a stable job, while the figures for France (23.2 and 23.5 extra weeks) and Dutch men (25.6 extra weeks) suggest that the earliest jobs are less stable in these

14

The French calculation is 1.203 years - (84 percent * 1 year) = 0.363 years. The German calculation is 1.063 years -(50 percent * 1.25 years) = 0.438 years. 15 In fact, women should find their first jobs even sooner in countries with military service that in those without, ceteris paribus, because labor supply of their entering cohort is temporarily reduced while the men are in the military. This means less competition for jobs and, all else equal, faster job finding.

15 countries. As noted above, this could be due to a higher exogenous separation rate or faster job-to-job mobility early in the careers for these people. Given the results on the time to the first job, it seems likely that the first explanation (exogenous separation) is more important for France and the second (job-to-job mobility) is more important for Dutch men.16

4.2 Earnings Growth In all of our countries, we calculated earnings growth as follows : Earnings Growth =

( wt +1− wt ) wt +1

where wt refers to the relevant earnings measure in period t. We chose to normalize by period t+1, as, for all countries, the share of workers employed increased more or less monotonically across the six-month periods during the first two years after school leaving. As such, we reduce the number of observations lost due to missing data on the earnings growth variable. One disadvantage is that, whereas this measure is bounded above by 1, it is not bounded below. Overall, we find that the fastest earnings growth occurred in the Netherlands, where first year earnings were on average 27 percent lower than second year earnings for men, and 12 percent lower for women. Average earnings growth was lower for men in Germany with respect to the Netherlands (14 percent), but marginally higher for women (also 14 percent). Our earnings growth measures for the United States suggest slower average earnings growth than in Germany and the Netherlands, with women having earnings that are sufficiently

16

This interpretation is consistent with research found elsewhere in the literature. Although fixed-term contracts and special youth employment programs were introduced in France in the middle of our sample period, these made up an important part of initial hiring in France (Gelot and Osbert (1995)). On the other hand, recent models estimated by van de Berg and Ridder in the Netherlands suggest that the offer arrival rate for on-the-job search is high than that for unemployed search (van de Berg and Ridder (1993)).

16 decreasing over time so as to induce, on average across individuals, a decline in our measure of earnings growth. At the other end of the spectrum, our measure of earnings growth for France is largely negative on average for both men and women, but this is clearly driven by a few outliers. The median values of the earnings growth measure (across individuals, not observations) in France were 57 percent for men and 51 percent for women, and only 21 percent of men and 20 percent of women had first year earnings that were higher than second year earnings (thus inducing a negative value for our earnings growth measure).

4.3 Employment Variables We have constructed several different measures of early career employment, each measured over 6 month periods beginning at the time of school leaving. These variables were the number of different employers, the time spent in employment and the average job duration (measured as the share of time employed divided by the number of employers times the length of the interval). As average job duration is a function of the other two variables, we will only discuss the percentage of time employed and number of employers statistics at this point. For all of the countries in our sample and for both sexes (with the exception of Dutch women and, to a much lesser extent, Dutch men), young people spent more and more time in employment as time elapsed following the end of schooling. Part of this effect is mechanically related to the time to first long job measures, since by construction the six months following the first long job will have 100 percent employment. Nevertheless, this measure suggests that unemployment is not evenly distributed throughout the insertion period, and that either workers tend to spend less and less time unemployed, or that fewer and fewer workers spend any time unemployed, as the labor market insertion progresses.

17 There does not appear to be any common pattern across countries in terms of the number of different employers visited during the first 2 years of one’s career. In the United States, the trend is essentially decreasing, which suggests that jobs are becoming more and more stable over time (as would be implied by standard job search theory). On the other hand, the number of different employers is monotonically increasing on average in France, while in the Netherlands there is no clear pattern.17 These differences suggest that care should be taken when trying to transpose results, such as those found in Gardecki and Neumark (1998), from American data based studies to the European context, as the base descriptive statistics are not even the same.

4.4 Occupational and Industry Mobility Variables Certain recent modifications of human capital theory suggest that, for a given number of employers, those individuals who change occupation or industry more often will acquire less human capital in any given industry that may be of use later in life. In addition, Neal’s (1996) analysis of the complexity of early career mobility suggests that workers who often switch occupation or sector are having problems finding an adequate match, which delays the start of their real careers. With these theories in mind, we constructed measures of the number of different occupations and industries experienced during the first two years after the end of schooling, measured in six month intervals.18 The patterns in the means of the data do not suggest that workers settle in to an industry or occupation within their first two years on the labor market. Only in the case of American women is there a trend in industry mobility that looks somewhat monotonically declining. For French men and women, the tendency is toward an increase in the number of 17

The GSOEP does not allow us to construct this measure for our German data.

18 different occupations and industries experienced. This effect may be due, once again, to military service delaying the true start of job search (since the number of occupations or industries experienced while an individual is in military service is set equal top zero). However, as there does not seem to be a major difference in the time paths of the means between French men and women, this is unlikely to be the main explanation.

4.5 Evolution of the Samples Over Time Tables 2a to 2d describe the evolution of the analysis samples over time. As these tables make clear, we are working with unbalanced panels in all of our analyses. Sample attrition is an important issue for all of the European panels, with over 15 percent of each wave, on average, disappearing before the end of the observation period. The problem is particularly serious for the Dutch data, where (on average) 48 percent of each male wave and 45 percent of each female wave disappears before the end of the observation period. For this reason, we decided to analyze our data as pooled samples, rather than following cohorts over time or looking at mixtures of cohorts at a fixed number of years after school leaving. The earlier cohorts tend to be the most heavily represented in our samples, which is encouraging in that it implies that there is some chance of having a reasonable number of observations observed far after school leaving. In fact, in even though we start our analysis at five years after school leaving, 14 percent of American men and women and 64 percent of French men and women are observable 10 or more years after they left school. As tables 2a and 2c show, for both the United States and Germany, we constrained ourselves to look at a maximum of 10 and 8 years of post schooling labor market experience, respectively. In the French and Dutch cases we included workers observed through the end of

18

Once again, the structure of the GSOEP data does not allow us to construct these measures for Germany.

19 the sample windows, which implies a maximum observable post-schooling experience of 20 years for France and 16 years for the Netherlands. It should be noted, however, that the highexperience cohorts in the Dutch data are rather sparsely populated, and as such we preferred to include control variables in our regressions for observation years and not post-schooling experience, since these will be better identified.19

5. Results In this section, we present the results of estimating models of the effects of early career experiences on later career outcomes. For each country, we estimate linear models of the form y it = x i β 1 + zit β 2 + ε it where y it is the later-career earnings outcome of interest (log hourly wages, log monthly earnings or log full-year equivalent earnings), x i is a vector of measures of early-career experiences, and z it is a vector of other covariates measured at the same date as the dependent variable y it that control for other characteristics likely to be important, such as age, seniority, education, region, industry and occupation. For the percentage of time spent in employment, we use a standard tobit specification based on the latent model specified in (1), bounded above by 1 and below by 0, with the same control variables. Since not all of the y it , x i and z it variables are available for all of the data sets, the absence of a country from a results table implies that either the dependent variable or the explanatory variables were not available for that country. Below we present results for our early career variables in three sections: time to first job and earnings growth variables; number

19

We do, however, control for the year of cohort entry, age through age4, and education, so this restriction is less important.

(1)

20 of employers, share of time employed and average duration variables; and number of occupations and number of industries variables.20

5.1 Time to First Job and Earnings Growth Measures Table 3 shows the results of our estimations for the explanatory variables Time to First Job, Time to First Job of at Least 6 Months and Earnings Growth Between First and Second Post Exit Years. For the United States, this last variable is replaced by 3 semester-on-semester earnings growth measures. These variables are available for all of our countries, but not all models were estimated for all countries. The first thing to note is the lack of significance of the time to first job for the determination of the hourly wage in the United States, Germany and the Netherlands. However, this variable becomes larger, significant and of essentially a constant value (for men) for Germany in the Log Monthly Income and Log Full-Year Equivalent Earnings measures. This suggests that, in Germany at least, those who took longer to find their first job, although not significantly different for the others in terms of hourly wage, do tend to work more hours, and thus have higher monthly and annual earnings. The positive coefficients on monthly earnings for men and women in Germany, however, are not found in the Netherlands for women, nor are they found in France for either men or women. The French results (which are significantly different from zero) imply that men and women who find their first jobs faster in France tend to do better in earnings terms later in their careers. The fact that the effect of time to first job on women’s earnings becomes more positive both in France and in Germany when going from monthly to full-year equivalent earnings is reasonable in light of the results on employment probabilities. If a women spends part of the

20

Statistics concerning model performance and number of observations appear in table 6.

21 year out of the labor market, the full-year equivalent earnings measure increases her earnings proportionally more than a woman who spends more time in the labor market. Long times to first jobs in Germany are associated with higher subsequent employment probabilities (as they are for American women), and since time to first job is positively associated with monthly earnings, the relation is strengthened through the covariance of days worked with time to first job. In France, the opposite relation between time to first job and percentage of time employed holds (as in the Netherlands and for American men), namely that those who take longer to find a first job are less likely to be employed later in their careers. Thus the relation observed between monthly income and time to first job should be weakened, via the negative covariance of time to first job with the employment probability. The French (and Dutch women’s) results are consistent with a straightforward search theoretic model. A person with a higher discount rate will have a lower reservation wage. This leads to faster job finding at the beginning of the career and quicker job acceptances later in the career as well. Hence the negative correlation in France between time to first job and employment probability. On the other hand, since the reservation wage is lower, earnings for high discount rate people should be lower, and table 3 suggests that they are. The German results can be interpreted in the context of a sorting model with heterogeneous workers and heterogeneous jobs.21 In such a model, the jobs that pay higher initial wages offer flatter seniority profiles, and these jobs are relatively abundant and attract less stable workers.22 In this case, unstable workers will take jobs quicker but be less likely to be employed at any given time. In addition, those who took longer to find their jobs will be, on average, more stable and thus have more seniority than those who found jobs faster. Which

21

See Margolis (1996) for a model that generates an equilibrium non-degenerate distribution of jobs with different returns to seniority and different separation rates. 22 Many efficiency wage models also predict the existence of such an “ocean of small firms”, where employment is readily available at a fixed wage as compared to other firms which offer seniority returns profiles.

22 group has the highest expected wages at any point is a function of the model parameters, but given that we are observing our individuals 5-20 years after they started looking for jobs, it is very possible that the return to seniority effect dominates, thus giving the observed positive sign on wages in Germany and for American women.23 The results concerning the first stable job tend to go in the opposite direction as those concerning the first job. Note that these coefficients are identified off the individuals for whom the first job did not last more than 6 months, and as such refer to individuals whom either accepted first jobs at low wages (and hence moved as their on-the-job search brought out better offers) or accepted unstable first jobs (signifying a preference for short term earnings over long term employment and earnings growth). In the United States, Germany and the Netherlands, taking a long time to find one’s first job of at least 6 months is associated with lower wages and (generally) lower values of the other earnings measures. This suggests that the first story, namely a low reservation wage, is what is driving these results. That said, the negative coefficients in employment probabilities suggest that long periods between time to first job and time to first job of at least 6 months may be due to heterogeneity in worker stability. The French results, unlike the other three countries, suggest a positive link between a long time to first job of at least six months and subsequent earnings. There exist several models that could explain such a result, such as insider-outsider models, dual labor market models and queuing/screening models. In general, these models suggest (more or less explicitly) that workers would try to keep the desirable jobs, once obtained, and thus are also consistent with the positive coefficient on the employment probability.

23

Margolis (1996) has shown that including simple seniority terms in a regression is not sufficient to capture the non-linearities in earnings induced by the combination of the explicit compensation policy and the evolution of the quality of the job entry cohort.

23 Finally, the early career earnings growth results, in the vast majority of cases, suggest that workers who managed faster earnings growth at the beginning or their careers tend to earn more later in their working lives.

This could be a sign of efficient on-the-job search,

heterogeneous career ladders, good quality matches or a simple autocorrelation in the earnings series,24 but not all of these explanations have straightforward implications for the link between early career earnings growth and later career employment rates.

The Jovanovic (1979)

matching model, for one, does have such implications. In this model, faster early career earnings growth is consistent with a high quality match. High quality matches are less likely to end, and as such the French and American results can be interpreted as being generally consistent with such a model. The German and Dutch results, on the other hand, might be considered consistent with a model of internal labor markets (tournaments), where there is stiffer competition higher up the hierarchy, and an up-or-out promotion rule. Thus faster early career wage growth indicates quicker movement up the career ladder, but as the person approaches the higher rungs, the layoff risk increases. This could lead to the observed pattern of a positive correlation between early career earnings growth and later career earnings, coupled with the negative correlation between earnings growth and employment probabilities.

24

If such an autocorrelation were present, it would have to be a very long memory process, as our right hand side variables are measured at a minimum 3 years prior to the left hand side variable, and in some cases as long as 18 years earlier. In addition, there would have to be an absence of an offsetting reduction in initial earnings for those with faster wage growth, whereas the literature has tended to find that faster wage growth is correlated with lower initial earnings (Abowd, Kramarz and Margolis (1999), Margolis (1996)). Nevertheless, if autocorrelation is a serious problem, then our right hand side earnings growth variables would be endogenous, and we would need to instrument them. Unfortunately, there are few variables that readily suggest themselves as instruments for early career earnings growth that do not otherwise appear in the models on their own, and as such the necessary exclusion restrictions for applying instrumental variables are not easily satisfied.

24 5.2 Number of Employers, Share of Time Employed and Average Duration Measures

Table 4 presents the coefficients on the number of employers, share of time employed and average job duration measures from our various models. Each of our measures is broken into six month intervals, which allows us to see whether it is the earliest experiences alone that count, or whether the entire early career period (the two years following school leaving by our definition) plays a role in later career outcomes. First of all, our measures of the number of employers visited during the first 2 years, in general, suggests that for men in the United States and the Netherlands, the more different employers are tried early in the career, the higher the earnings will be later on.

These

coefficients are significant only for men in the Netherlands for the 6-18 month interval, but they are consistent with on the job search early in the career being a route to higher earnings later on.25 That said, we have controlled for early career earnings growth, so the early career mobility may be sorting workers into better suited careers, as Neal’s (1996) model would suggest.

Unfortunately for Neal’s model, however, are the results on employment

probabilities, in which the early career instability is negatively related to future employment. For American women and the French, on the other hand, these measures are very often significant, and although the 12-18 month interval is significant and positive, all of the other intervals are significant and negative, for both monthly and full year equivalent earnings in France. The size of the positive coefficient is not enough to counterbalance the net effect of many different employers in each period, and thus France seems different from American men and the Netherlands. Part of this negative effect may be the result of fixed-term and youth employment promotion contracts, which some authors have suggested serve to push young

25 people into a secondary labor market of short term, low pay, unstable jobs.

Such an

explanation would seem consistent with our data, except that the very large, positive coefficient in the employment probability model on the 18-24 month interval suggests that the jobs found later in life by these individuals may be, in fact, more stable than those found by others who had less turbulent beginnings. The question of early career stability, as measured by the number of employers, is closely linked to that of early career employability, as measured by share of time spent in employment (controlling for the time to the first job). For American, French and German women, spending more time at the beginning of one’s career in employment clearly leads to higher earnings later in life. This may be reflecting a labor force attachment effect, in which women who spend less time employed at the beginning of their careers signal to future employers a possible weak labor force attachment, and thus inciting employers to invest less in these women. The lower investments bring lower returns, and thus lower earnings. For men, the effects of spending more time employed early in the career are less obvious. In the Netherlands, it once again seems that the start of the first full year after school leaving (months 12-18) is the most important period, bringing with it a strongly negative relation between time spent employed and future earnings. In the United States the first six months seem the most crucial, with a significant negative coefficient, although the sum of the subsequent 18 months erases this negative effect. In France, the coefficients alternate in sign, while in Germany similar instability in the estimation coefficients is observed. Such results are difficult to interpret coherently. It is interesting to note, however, that “employability” does not seem to be an individual-specific fixed characteristic. If workers were more or less employable, and this was constant over time, we would expect to see positive coefficients on share of time spent 25

Gardecki and Neumark (1998) found similar results for the United States.

26 employed everywhere for the employment probability models. This is far from being the case, as there are more negative and significant coefficients than positive and significant ones. Furthermore, it does not seem that “employability” is learned as the labor market insertion period progresses. In fact, the United States is the only country in which the share of time spent employed during the last 6 months of our 2-year insertion period is positively correlated with time spent employed later in the career, and this effect is only significant for men. Finally, we also included a measure for average job duration, which (as mentioned above) is an interaction between time spent employed and the inverse of the number of employers. Our results suggest that, in general, visiting more employers for a given time spent in employment (thereby reducing the average time worked at each employer) improves ones earnings later in the career by more than would be expected by the simple increase in the number of employers. Coefficients on these variables are typically negative in the United States and France, although staying with the same employer early in the career seems to have positive earnings implications for the Netherlands. Although the Dutch result seems consistent with specific human capital theory, the American and French results require a more subtle analysis. The average duration variable in these countries may be proxying for the size of the local labor market. If the local labor market is large, an individual may test several employers for a given amount of time spent employed. Demand-side competition in the labor market may then bid up wages. If our later career data correspond to the same labor market as the early career data, or if individuals move between similar labor markets, we would expect to see a negative relation between average job duration and earnings. In addition, since workers are required to (and can) exercise the threat of leaving on occasion to remain credible, the generally positive coefficients on employment probability suggest that the more competitive labor markets experience more turnover. Superficially, it does not seem implausible that Dutch employers have more monopsony power over wages than American or French employers.

27

5.3 Number of Occupations and Number of Industries Measures Although there has been some recent work on occupation and industry specific human capital (Neal (1996), Parent (1995), Stevens (1994), Vilhuber (1997,1999)), most of it suggesting that at least some of the human capital that workers acquire may be transferable to other firms in which the worker does the same job or other firms in the same sector, there is little information on whether frequent mobility across occupations or industries early in the career is detrimental to workers. From a matching perspective, frequent mobility across occupations or sectors may be a good thing, as it is a sign that the worker is finding the most adequately suited job, rather than settling for a suboptimal career path.

With these

considerations in mind, table 5 presents the coefficients from our various estimations on the number of occupations and number of industries measures, evaluated at 6-month intervals.26 Unfortunately, these measures were not computable from the GSOEP data for Germany. Our results suggest that women who change occupations often during the first six months of their labor market time in the Netherlands and France tend to earn more later in life, in terms of hourly wages and monthly earnings.

The negative coefficients on full year

equivalent earnings for women in France, once again, are due to an employment effect. Although the point estimates are not precise, French women who change occupations less at the beginning of their careers tend to spend less time employed later on. Since these are the women who earn less in monthly earnings, and since full year equivalent earnings inflates these women’s earnings more than those of women who tried more occupations, the sign on full year equivalent earnings can switch. For Dutch men, changing occupations often at the beginning of

26

It should be noted that including these measures, along with the number of different employers in 6 month intervals measures, is likely to introduce significant multicollinearity in our results. The model performance statistics in table 6 indicate that this is likely to be a problem, especially for France.

28 the career tends to be penalized later on, although the opposite effect is observed for France (when taking into account the sizes of the coefficients). The results for the United States suggest a somewhat negative link between early career occupational mobility and later career earnings, although little is significant. Insofar as industry mobility is concerned, the effects seem somewhat more clearly in favor of the idea that changing industries often at the beginning of one’s career is a signal of searching for a better job that is better paid later in life, rather than supporting the idea that such mobility dilutes one’s human capital. Although the role of such mobility on employment probabilities is also generally positive in the United States (albeit insignificant), early career inter-industry mobility seems detrimental to later career employability in France. Although the American results can be rationalized by allowing heterogeneity in worker “adaptability”, which improves one’s chances of finding a job and would be rewarded by employers by higher wages, interpretations of the French results seem less straightforward.

6. A Few Thought Experiments Now that we have a large set of results concerning four different measures of later career success and 23-25 early career explanatory variables, all of which are intimately related, we present several thought experiments to help clarify the roles of different early career experiences on later career outcomes. Table 6 presents these experiments in tabular form. Each of these experiments corresponds to changing one (or a set of) explanatory variable(s) and holding all other controls fixed at the sample means for the relevant population. For the tobit results, we apply the same censoring rule, namely that shares of time spent employed must lie between 0 and 1, inclusive.

29 First of all, we consider how things would be different if all workers found their first job immediately upon leaving school. In terms of hourly wages, American, German and Dutch workers would be worse off in this scenario than if they took the average time in each country for each sex to find their first job. However, such a change would be associated with 1 percent higher monthly earnings for French men and 7.5 % higher monthly earnings for French women. The average German man, although losing only slightly in terms of hourly wages (1.2 percent), would see his monthly earnings drop by much more, 3.6 percent. In this alternative scenario, German women fare much worse in terms of subsequent employment probabilities, with an average women going from spending 74 percent of the time employed to spending only 22 percent of the time employed. The effects for finding the first long job right away (which implies finding the first job immediately by construction) are quite different. The hourly wage declines become wage gains for German and Dutch men, and elsewhere the size of the wage decline is smaller. Monthly earnings go up for all observable groups except French men (stable) German women (a slight decline).

However, the biggest differences are in terms of employment probabilities.

It

appears that “locking oneself in to a job” too early in France can have very strong negative effects on subsequent employment probabilities, with the average man spending barely more than half as much time employed later in his career than if he were given that average amount of time to find his first stable job. For French women there is also a drop, as there is for German and American women (albeit much smaller). On the other hand, if the average German man were to find his first long job right after leaving school, his chances of being employed at a given point in time later in life would essentially be 100 percent. Perhaps surprisingly, the impacts of our two experiments on earnings growth (no earnings growth during the first two years or double the observed earnings growth during the first two years) seem less important than our thought experiments concerning time to first

30 job.27 The only major differences with respect to the baseline case lie with Dutch women, for whom the probability of future employment for a woman with average characteristics increases by 3.3 percentage points when her earnings growth between years 1 and 2 is doubled. Experiments with the number of employers seem to have little effect on earnings, but a major effect on employment probabilities in France. Constraining French workers to stay with the same employer, but not changing their share of time spent employed (thereby forcing them back to the old employer after a period of unemployment) basically excludes these workers from the labor market. A more realistic experiment, where we allow the person to remain continuously employed throughout the entire insertion period with the same employer, has more mitigated effects for men, causing the future employment probability to drop only to 54 percent, while that of women leaps up to essentially 100 percent. The experiment of eliminating early career unemployment, while holding the number of employers constant, has much more important effects, particularly for women’s hourly wages in the United States and Germany, and for men’s hourly wages in the Netherlands. The effects on monthly earnings are relatively important for Dutch men and women, as they are for French and German women, and to a lesser extent French men. Eliminating unemployment early in the career has very big impacts on subsequent employment probabilities in all countries, with French women, German men and Dutch men and women all being pushed to the truncation bounds of the tobit model. Forcing the worker to spend the additional time with the same employer seems to have additional effects in the Netherlands, and for American women in terms of hourly earnings. As noted above, the French are affected in terms of employment probabilities in this thought

27

It should be noted that, since in the American case earnings growth was measured as a difference between 6 month periods rather than years, the though experiment performed here is to double each semester-onsemester’s observed earnings growth. In general, this will not be equivalent to doubling annual earnings growth.

31 experiment, and the effects are also seen in terms of lower monthly earnings than if they were allowed to change employers. Given the unstable patterns of the coefficients on the number of occupations and number of industries, it is not surprising to note that pursuing the thought experiments in which workers spend their entire insertion period in the same industry or the same occupation provide little insight.

None of our outcome variables is seriously affected by such a change in

occupational or industry mobility. This may suggest that further thought needs to be given to those models which predict an impact for such variables later in life, as we find no evidence of such an impact in our data.

7. Conclusion In this paper, we have considered the effects of a large variety of measures of early career experiences on later career outcomes using data drawn from four different countries. As suggested in the introduction and in section 2, there are many different theories that link our early career variables to our later career outcomes, and estimation of very simplified models is likely to lead to misleading interpretations, given the complexity of the subject. Our results have shown that the subject is, indeed, very complex. There seem to be very few “general principles” that permeate all four of our economies in the same manner, and this diversity of relations should be a fruitful ground for further research into the institutional origins of these differences. Nevertheless, most theories of labor market behavior that link some aspect of the start of the career to success later on in life can find at least lukewarm support in at least one of our countries. This suggests that one needs to be very careful not to predispose oneself to a certain type of model for all settings, as the appropriate model seems quite context-dependent.

32 Finally, we conducted several thought experiments in section 6 that could help inform a policy maker’s decision process on potentially fruitful areas for intervention in the job market at the beginning of the career. Clearly, as mentioned above, there is no one-size-fits-all solution to increasing later career wages, earnings or employability.

Often a possible

intervention might have positive effects for one outcome variable but negative effects for another. A useful direction for further research would be to examine the persistence of these effects over time, to see if certain effects diminish in intensity as the career proceeds, leaving others to dominate the rest of an individual’s working life.

33 References Abowd, John M., Francis Kramarz and David N. Margolis (1999). “High Wage Workers and High Wage Firms,” Econometrica, March. Altonji, Joseph (1998). “Employer Learning and Statistical Discrimination,” Inaugural Address given at the 15émes Journées de Microéconomie Appliquée, Pointe-à-Pitre, Guadeloupe, June 4. Balsan, Didier, Saïd Hanchane and Patrick Werquin (1996). “Mobilité professionnelle initiale : éducation et expérience sur le marché du travail” Economie et statistique. Becker, Gary S. (1993). Human Capital, Third Edition (Chicago: University of Chicago Press). Ben Porath, (1967). “The Production of Human Capital and the Life Cycle of Earnings,” Journal of Political Economy, August. van den Berg, Gerard and Geert Ridder (1993). “An Empirical Equilibrium Search Model of the Labour Market,” University of Amsterdam working paper, July 19. Farber, Henry S. and Robert Gibbons (1996). “Learning and Wage Dynamics,” Quarterly Journal of Economics, November. Gardecki, Rosella and David Neumark (1998). “Order from Chaos ? The Effects of Early Labor Market Experiences on Adult Labor Market Outcomes,” Industrial and Labor Relations Review, January. Gelot, Didier and Gerard Osbert (1995). “Policies for Youth Employment in France Over the Past 20 Years,” presented at the NBER Summer Institute, Franco-American Seminar, July 28. Jovanovic, Boyan (1979). “Job Matching and the Theory of Turnover,” Journal of Political Economy. Lewis, H. Gregg (1986) Union Relative Wage Effects: A Survey (Chicago: University of Chicago Press). Margolis, David N. (1996). “Firm heterogeneity and Worker Self-Selection Bias Estimated Returns to Seniority,” CIRANO working paper, January. Miller, Robert A. (1984). “Job Matching and Occupational Choice,” Journal of Political Economy, December. Neal, Derek (1995). “Industry-Specific Human Capital: Evidence from Displaced Workers,” Journal of Labor Economics, October. Neal, Derek (1996). “The Complexity of Job Mobility Among Young Men,” presented at the NBER Summer Institute Labor Studies Group, July 25.

34

Parent, Daniel (1995). “Industry-Specific Capital and the Wage Profile: Evidence from the NLSY and the PSID,” CRDE working paper 0895, February. Roy, A. D. (1951). “Some Thoughts on the Distribution of Earnings,” Oxford Economic Papers. Ruhm, Christopher J. (1995). “The Extent and Consequences of High School Employment,” Journal of Labor Research, Summer. Simonnet, Véronique (1997). “Déterminants et rentabilité de la mobilité sur le marché du travail : Analyse théorique et empirique (Allemagne, Etats-Unis, France)”, Ph.D. Thesis, Université de Paris 1 Panthéon-Sorbonne. Spence, Michael (1973). “Job Market Signalling,” Quarterly Journal of Economics, August. Stevens, Margareth (1994). “A Theoretical Model of On-the-Job Training with Imperfect Competition,” Oxford Economic Papers. Topel, Robert H. and Michael P. Ward (1992). “Job Mobility and the Careers of Young Men,” Quarterly Journal of Economics, May. Vilhuber, Lars (1997). “Sector-Specific On-the-Job Training: Evidence from U.S. Data,” CIRANO working paper 97s-42, December. Vilhuber, Lars (1999). “Sector-Specific Training and Mobility in Germany,” CIRANO working paper 99s-03, February.

35 Table 1a: Descriptive Statistics, United States Data Men Number of Standard Variable Observations Mean Deviation Date of School Exit 1471 81.923 2.272 Time to First Job 630 0.304 0.007 Time to First Job of at Least 6 Months 669 0.466 0.009 Number of Employers in First 6 Months 1471 1.456 0.999 Number of Employers in First Months 6-12 1471 1.250 0.884 Number of Employers in First Months 12-18 1467 1.213 0.822 Number of Employers in First Months 18-24 1468 1.236 0.893 Share of Time Spent in Employment in First 6 Months 1471 0.327 0.185 Share of Time Spent in Employment in First 12 Months 1471 0.361 0.194 Share of Time Spent in Employment in First 18 Months 1467 0.375 0.188 Share of Time Spent in Employment in First 24 Months 1468 0.376 0.190 Average Job Duration During First 6 Months 1471 0.239 0.170 Average Job Duration During First Months 6-12 1471 0.300 0.195 Average Job Duration During First Months 12-18 1467 0.322 0.192 Average Job Duration During First Months 18-24 1468 0.319 0.195 Number of Different Occupations in First 6 Months 1394 1.090 0.608 Number of Different Occupations in Months 6-12 1394 1.140 0.617 Number of Different Occupations in Months 12-18 1394 1.039 0.534 Number of Different Occupations in Months 18-24 1394 1.154 0.624 Number of Different Industries in First 6 Months 1394 1.072 0.579 Number of Different Industries in Months 6-12 1394 1.085 0.587 Number of Different Industries in Months 12-18 1394 1.017 0.505 Number of Different Industries in Months 18-24 1394 1.098 0.588 Wage Growth Between First and Second Post-Exit Semesters 1138 0.003 0.982 Wage Growth Between Second and Third Post-Exit Semesters 1150 -0.006 0.737 Wage Growth Between Third and Fourth Post-Exit Semesters 1224 0.064 0.562 Age 7193 27.167 2.879 Share of Time Spent in Employment 7007 0.878 0.266 Log Real Hourly Wage (1985 USD) 5481 -2.570 0.035 Source: Authors' calculations from NLSY data. Notes: All employment figures refer to civilian (non-military) employment.

Women Number of Observations 1503 606 586 1502 1503 1503 1503 1502 1503 1503 1503 1502 1503 1503 1503 1400 1400 1400 1400 1400 1400 1400 1400 1099 1127 1213 7367 7191 5044

Mean 81.811 0.362 0.508 1.409 1.226 1.143 1.156 0.316 0.346 0.353 0.364 0.236 0.287 0.305 0.312 1.036 1.075 0.973 1.071 1.029 1.021 0.966 1.007 -0.040 -0.018 -0.018 26.981 0.748 -2.830

Standard Deviation 2.194 0.008 0.009 0.979 0.930 0.807 0.879 0.186 0.199 0.199 0.198 0.169 0.196 0.200 0.198 0.554 0.590 0.511 0.570 0.555 0.547 0.496 0.511 1.638 0.613 1.658 2.786 0.373 0.035

36 Table 1b: Descriptive Statistics, French Data Men Number of Standard Variable Observations Mean Deviation Date of School Exit 57135 78.519 3.243 Time to First Job 57135 1.203 1.469 Time to First Job of at Least 6 Months 57135 1.649 1.570 Number of Employers in First 6 Months 57135 2.126 4.209 Number of Employers in First Months 6-12 57135 2.914 4.851 Number of Employers in First Months 12-18 57135 3.454 5.193 Number of Employers in First Months 18-24 57135 3.866 5.536 Share of Time Spent in Employment in First 6 Months 57135 0.348 0.428 Share of Time Spent in Employment in First 12 Months 57135 0.474 0.468 Share of Time Spent in Employment in First 18 Months 57135 0.530 0.487 Share of Time Spent in Employment in First 24 Months 57135 0.524 0.508 Average Job Duration During First 6 Months 57135 0.078 0.122 Average Job Duration During First Months 6-12 57135 0.079 0.121 Average Job Duration During First Months 12-18 57135 0.077 0.113 Average Job Duration During First Months 18-24 57135 0.065 0.110 Number of Different Occupations in First 6 Months 57135 0.685 0.864 Number of Different Occupations in Months 6-12 57135 0.850 0.940 Number of Different Occupations in Months 12-18 57135 0.947 0.967 Number of Different Occupations in Months 18-24 57135 0.994 1.002 Number of Different Industries in First 6 Months 57135 0.530 0.606 Number of Different Industries in Months 6-12 57135 0.634 0.636 Number of Different Industries in Months 12-18 57135 0.687 0.628 Number of Different Industries in Months 18-24 57135 0.695 0.623 Earnings Growth Between First and Second Post-Exit Years 38512 -0.866 14.657 Age 57135 29.323 5.675 Tenure 46592 4.043 4.440 Share of Time Spent in Employment 57135 0.707 0.460 Log Real Monthly Gross Earnings (1980 KF) 46592 1.240 1.016 Log Real Full-Year Equivalent Gross Earnings (1980 KF) 46592 3.992 0.666 Year of Observation of the Dependant Variable 57135 89.801 4.230

Women Number of Observations 46160 46160 46160 46160 46160 46160 46160 46160 46160 46160 46160 46160 46160 46160 46160 46160 46160 46160 46160 46160 46160 46160 46160 31854 46160 30834 46160 30834 30834 46160

Mean 78.311 1.166 1.617 2.073 2.924 3.529 4.013 0.341 0.483 0.548 0.551 0.082 0.084 0.079 0.067 0.646 0.803 0.906 0.948 0.530 0.644 0.708 0.728 -0.682 29.562 4.113 0.574 0.842 3.641 89.532

Standard Deviation 2.966 1.459 1.540 4.133 4.851 5.162 5.463 0.420 0.461 0.485 0.499 0.129 0.127 0.117 0.108 0.797 0.856 0.881 0.906 0.598 0.625 0.623 0.625 12.726 5.437 4.493 0.499 1.138 0.770 4.216

Source: Authors' calculations from DADS data. Notes: All employment figures refer to civilian (non-military) employment.

Table 1c: Descriptive Statistics, German Data Men Number of Standard Variable Observations Mean Deviation Date of School Exit 2497 86.640 4.267 Time to First Job 2497 1.063 1.340 Time to First Job of at Least 6 Months 2497 1.305 1.665 Share of Time Spent in Employment in First 6 Months 2497 3.029 2.828 Share of Time Spent in Employment in Months 6-12 2497 3.233 2.901 Share of Time Spent in Employment in Months 12-18 2497 3.298 2.908 Share of Time Spent in Employment in Months 18-24 2497 3.593 2.824 Earnings Growth Between First and Second Post-Exit Years 2497 0.136 0.480 Age 2497 27.714 5.972 Tenure 2205 50.410 44.681 Share of Time Spent in Employment 2275 0.807 0.360 Log Nominal Hourly Wage 1594 2.606 0.819 Log Nominal Monthly Gross Earnings 2134 7.795 0.424 Log Nominal Full-Year Equivalent Gross Earnings 1781 10.669 0.521 Year of Observation of the Dependant Variable 2497 92.362 4.930 Source: Authors' calculations from GSOEP data. Notes: All employment figures refer to civilian (non-military) employment.

Women Number of Observations 1936 1936 1936 1936 1936 1936 1936 1936 1936 1586 1739 1197 1517 1240 1936

Mean 86.824 1.067 1.273 2.752 2.959 3.068 3.379 0.142 26.584 41.390 0.736 2.320 7.415 10.306 92.534

Standard Deviation 4.408 1.227 1.413 2.838 2.864 2.903 2.858 0.373 5.242 33.074 0.405 0.778 0.456 0.581 4.970

37 Table 1d: Descriptive Statistics, Dutch Data Men Number of Standard Variable Observations Mean Deviation Date of School Exit 344 83.878 2.885 Time to First Job 344 0.671 0.970 Time to First Job of at Least 6 Months 344 1.163 1.198 Number of Employers in First 6 Months 344 0.669 0.648 Number of Employers in First Months 6-12 344 0.660 0.686 Number of Employers in First Months 12-18 344 0.802 0.791 Number of Employers in First Months 18-24 344 0.689 0.740 Share of Time Spent in Employment in First 6 Months 344 0.482 0.463 Share of Time Spent in Employment in First 12 Months 344 0.489 0.474 Share of Time Spent in Employment in First 18 Months 344 0.490 0.454 Share of Time Spent in Employment in First 24 Months 344 0.484 0.468 Average Job Duration During First 6 Months 344 0.221 0.222 Average Job Duration During First Months 6-12 344 0.220 0.228 Average Job Duration During First Months 12-18 344 0.212 0.216 Average Job Duration During First Months 18-24 344 0.223 0.228 Number of Different Occupations in First 6 Months 344 0.648 0.617 Number of Different Occupations in Months 6-12 344 0.631 0.648 Number of Different Occupations in Months 12-18 344 0.741 0.712 Number of Different Occupations in Months 18-24 344 0.692 0.690 Earnings Growth Between First and Second Post-Exit Years 271 0.266 0.450 Age 344 30.645 5.077 Tenure 344 6.120 4.921 Share of Time Spent in Employment 344 0.895 0.299 Real Hourly Wage (1990 HFL) 344 14.190 7.262 Real Monthly Gross Earnings (1990 HFL) 344 2375.032 1220.355

Women Number of Observations 318 318 318 318 318 318 318 318 318 318 318 318 318 318 318 318 318 318 318 240 318 318 318 318 318

Mean 84.031 0.514 0.743 0.786 0.843 0.918 0.654 0.654 0.655 0.625 0.561 0.308 0.302 0.275 0.273 0.761 0.799 0.827 0.635 0.121 28.783 5.555 0.838 9.656 1297.190

Standard Deviation 3.010 0.877 0.998 0.588 0.674 0.753 0.556 0.454 0.457 0.448 0.477 0.223 0.227 0.221 0.237 0.550 0.603 0.639 0.526 0.317 4.877 4.524 0.348 6.335 789.697

Source: Authors' calculations from OSA data. Notes: All employment figures refer to civilian (non-military) employment.

School Leaving Year 1979 Men 1980 1981 1982 1983 1984 1985 1986 1987 Total 1979 Women 1980 1981 1982 1983 1984 1985 1986 1987 Total

1984 192 0 0 0 0 0 0 0 0 192 214 0 0 0 0 0 0 0 0 214

Table 2a: Evolution of the Sample, United States Data Observation year 1985 1986 1987 1988 1989 1990 194 192 193 195 195 0 180 181 181 178 181 180 0 218 216 215 219 216 0 0 212 213 212 214 0 0 0 151 149 146 0 0 0 0 108 106 0 0 0 0 0 77 0 0 0 0 0 0 0 0 0 0 0 0 374 591 802 952 1064 939 214 214 213 213 212 0 214 212 213 213 212 212 0 217 217 215 217 217 0 0 233 234 234 233 0 0 0 137 137 137 0 0 0 0 115 116 0 0 0 0 0 82 0 0 0 0 0 0 0 0 0 0 0 0 428 643 876 1012 1127 997

Source: Authors' calculations from NLSY Data. Notes: Tables show number of observations available for percentage of time worked models.

1991 0 0 217 215 147 107 77 75 0 838 0 0 217 232 137 116 82 81 0 865

1992 0 0 0 213 146 105 76 77 60 677 0 0 0 235 138 116 82 81 42 694

1993 0 0 0 0 148 106 78 75 60 467 0 0 0 0 139 116 82 81 42 460

Total 1161 1081 1301 1279 887 532 308 227 120 6896 1280 1276 1300 1401 825 579 328 243 84 7316

38

School Leaving Year 1976 Men 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 Total 1976 Women 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 Total

1981 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

1982 1852 1384 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3236 1550 1208 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2758

1983 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

1984 1805 1351 1094 0 0 0 0 0 0 0 0 0 0 0 0 0 4250 1504 1163 1053 0 0 0 0 0 0 0 0 0 0 0 0 0 3720

Table 2b: Evolution of the Sample, French Data Observation year 1985 1986 1987 1988 1989 1990 1784 1764 1737 1722 1693 0 1329 1310 1290 1275 1260 0 1078 1064 1048 1032 1020 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 34 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 4191 4138 4075 4029 4007 0 1474 1442 1420 1388 1360 0 1146 1122 1086 1052 1027 0 1040 1011 988 967 947 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 46 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 3660 3575 3494 3407 3380 0

1991 1647 1222 986 0 0 0 0 0 33 576 35 0 0 0 0 0 4499 1309 975 900 0 0 0 0 0 43 315 43 0 0 0 0 0 3585

1992 1619 1211 969 0 0 0 0 0 33 562 34 800 0 0 0 0 5228 1279 941 876 0 0 0 0 0 43 304 41 576 0 0 0 0 4060

1993 1583 1179 948 0 0 0 0 0 33 550 34 781 0 0 0 0 5108 1235 913 849 0 0 0 0 0 42 292 37 549 0 0 0 0 3917

1994 1542 1133 931 0 0 0 0 0 33 524 33 753 0 0 0 0 4949 1189 879 803 0 0 0 0 0 37 277 35 504 0 0 0 0 3724

1995 1512 1110 915 0 0 0 0 0 31 511 33 721 0 0 0 0 4833 1156 841 772 0 0 0 0 0 34 261 31 482 0 0 0 0 3577

1996 1429 1042 852 0 0 0 0 0 29 481 26 647 0 0 0 86 4592 1073 772 695 0 0 0 0 0 28 234 26 401 0 0 0 74 3303

Source: Authors' calculations from DADS Data. Notes: Tables show number of observations available for percentage of time worked models.

School Leaving Year 1984 Men 1985 1986 1987 1988 1989 1990 1991 Total 1984 Women 1985 1986 1987 1988 1989 1990 1991 Total

1988 101 0 0 0 0 0 0 0 101 69 0 0 0 0 0 0 0 69

Table 2c: Evolution of the Sample, German Data Observation year 1989 1990 1991 1992 1993 1994 98 90 86 82 0 0 100 89 85 86 85 0 0 89 89 80 83 80 0 0 77 69 68 56 0 0 0 71 64 58 0 0 0 0 72 69 0 0 0 0 0 55 0 0 0 0 0 0 198 268 337 388 372 318 63 61 57 55 0 0 74 71 67 64 62 0 0 72 64 66 62 58 0 0 49 48 48 48 0 0 0 55 48 49 0 0 0 0 66 62 0 0 0 0 0 46 0 0 0 0 0 0 137 204 237 288 286 263

Source: Authors' calculations from GSOEP Data. Notes: Tables show number of observations available for percentage of time worked models.

1995 0 0 0 57 56 67 54 59 293 0 0 0 47 46 61 45 56 255

1996 0 0 0 0 52 63 51 56 222 0 0 0 0 43 59 42 53 197

Total 457 445 421 327 301 271 160 115 2497 305 338 322 240 241 248 133 109 1936

39

School Leaving Year 1980 Men 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 Total 1980 Women 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 Total

1985 24 0 0 0 0 0 0 0 0 0 0 24 26 0 0 0 0 0 0 0 0 0 0 26

1986 18 18 0 0 0 0 0 0 0 0 0 36 14 23 0 0 0 0 0 0 0 0 0 37

1987 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

Table 2d: Evolution of the Sample, Dutch Data Observation year 1988 1989 1990 1991 1992 7 0 2 0 2 11 0 3 0 3 14 0 5 0 5 14 0 4 0 4 0 0 0 0 0 0 0 11 0 11 0 0 0 0 24 0 0 0 0 4 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 46 0 25 0 53 6 0 0 0 0 12 0 4 0 4 9 0 4 0 5 8 0 2 0 2 0 0 0 0 0 0 0 4 0 3 0 0 0 0 30 0 0 0 0 8 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 35 0 14 0 52

Source: Authors' calculations from OSA Data. Notes: Tables show number of observations available for percentage of time worked models.

1993 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

1994 3 4 6 8 0 13 35 3 11 7 0 90 3 6 5 2 0 5 32 12 8 15 0 88

1995 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

1996 2 4 5 4 0 10 19 3 10 5 8 70 1 5 7 2 0 3 28 6 5 6 3 66

Total 58 43 35 34 0 45 78 10 21 12 8 344 50 54 30 16 0 15 90 26 13 21 3 318

40 Table 3 : Time to First Job and Early-Career Earnings Growth Measures Early-Career United States France Variable Men Women Men Women

Dependent Variable Log Hourly Wage

Time to First Job Time to First Job of at Least 6 Months

0.1012 (0.1800) -0.0599 (0.0711)

0.0920 (0.2138) -0.0271 (0.0697)

Wage Growth Between Second and Third Post-Exit Semesters Wage Growth Between Third and Fourth Post-Exit Semesters

Germany Women

Netherlands Men Women

0.0117 0.0131 0.0230 (0.0157) (0.0140) (0.0441) -0.0313 * -0.0047 -0.0191 (0.0125) (0.0139) (0.0304) 0.0035 -0.0492 * 0.0873 (0.0118) (0.0220) (0.0746)

0.0162 (0.1648) -0.0070 (0.1029) 0.0303 (0.3233)

-0.0101 * -0.0647 * 0.0343 * 0.0057 0.0129 (0.0046) (0.0085) (0.0131) (0.0143) (0.0515) 0.0072 * 0.0445 * -0.0320 * 0.0133 -0.0292 (0.0016) (0.0032) (0.0105) (0.0133) (0.0355) 0.0007 * -0.0001 0.0035 0.0534 * 0.0066 (0.0001) (0.0002) (0.0105) (0.0210) (0.0870)

-0.0276 (0.2047) -0.0492 (0.1279) 0.4854 (0.4017)

Earnings Growth Between First and Second Post-Exit Years Wage Growth Between First and Second Post-Exit Semesters

Men

0.1135 * 0.0502 (0.0495) (0.0444) 0.0893 * 0.0191 (0.0309) (0.0124) 0.0119 0.0435 (0.0181) (0.0326)

Log Monthly Income Time to First Job Time to First Job of at Least 6 Months Earnings Growth Between First and Second Post-Exit Years Wage Growth Between First and Second Post-Exit Semesters Wage Growth Between Second and Third Post-Exit Semesters Wage Growth Between Third and Fourth Post-Exit Semesters Log Full-Year Equivalent Earnings Time to First Job

-0.0219 * -0.0485 * 0.0345 * 0.0108 (0.0026) (0.0054) (0.0170) (0.0180) 0.0050 * 0.0289 * -0.0350 * 0.0128 (0.0010) (0.0021) (0.0135) (0.0179) 0.0006 * -0.0009 * 0.0025 0.0311 (0.0001) (0.0001) (0.0130) (0.0240)

Time to First Job of at Least 6 Months Earnings Growth Between First and Second Post-Exit Years Wage Growth Between First and Second Post-Exit Semesters Wage Growth Between Second and Third Post-Exit Semesters Wage Growth Between Third and Fourth Post-Exit Semesters Employment Probability Time to First Job Time to First Job of at Least 6 Months

-0.0180 0.0518 (0.0304) (0.0659) -0.0320 * -0.0097 (0.0093) (0.0194)

Earnings Growth Between First and Second Post-Exit Years Wage Growth Between First and Second Post-Exit Semesters Wage Growth Between Second and Third Post-Exit Semesters Wage Growth Between Third and Fourth Post-Exit Semesters

-0.0093 (0.0086) 0.0046 (0.0051) 0.0000 (0.0031)

-0.0048 -0.1369 0.1462 0.4816 * -0.0030 (0.0600) (0.0774) (0.0877) (0.1086) (0.0358) 0.2007 * 0.2525 * -0.4554 * -0.3194 * -0.0168 (0.0214) (0.0283) (0.0632) (0.0870) (0.0236) 0.0029 * 0.0014 -0.0210 -0.1282 -0.0342 (0.0009) (0.0013) (0.1081) (0.1588) (0.0632)

-0.0010 (0.0110) 0.0052 (0.0027) 0.0055 (0.0088)

Sources: United States - NLSY and authors' calculations; France - DADS and authors' calculations; Germany - GSOEP and authors' calculations; Netherlands - OSA and authors' calculations. Notes for Log Hourly Wage , Log Monthly Income and Log Full-Year Equivalent Earnings . Least squares standard errors in all regressions are corrected for arbitrary heteroskedasticity. NL: Hourly wages and Monthly Income measured in levels, not logs. US: Regressions control for log hours, years of education, age, age2, seniority, seniorty 2, rural residency, nonwhite, married, 6 entry cohorts, 9 years, 4 regions, 8 sectors and 6 occupations. FR: Regressions control for 8 educational categories, age, age2 , age3, age4, seniority, seniorty 2, Paris region, 8 entry cohorts, 13 years, 8 sectors and 6 occupations. DE: Regressions control for 9 educational categories plus years of education, age, age2 , age3, age4, seniority, seniorty 2, 8 entry cohorts, 9 years, marital status, 4 firm sizes and 3 occupations. NL: Regressions control for years of education, age, age2, age3 , age4, seniority, seniorty2 , firm size, firm size2, 10 entry cohorts, 7 years, cohabitation, 6 sectors and 6 occupations. Notes for Employment Probability . US: linear probability model of percentage of time employed with heteroskedsticity-consistent standard errors. FR, DE and NL: Tobit of percentage of time employed bounded above by 1 and below by 0. US: Regressions control for log hours, years of education, age, age2, seniority, seniorty 2, rural residency, nonwhite, married, 6 entry cohorts, 9 years, 4 regions, 8 sectors and 6 occupations. FR: Regressions control for 8 educational categories, age, age2 , age3, age4, 8 entry cohorts and 13 years. DE: Regressions control for 9 educational categories plus years of education, age, age2 , age3, age4, 8 entry cohorts, 9 years and marital status. 2

3

4

2

2

NL: Regressions control for years of education, age, age , age , age , seniority, seniorty , firm size, firm size , 10 entry cohorts, 7 years, cohabitation, 6 sectors and 6 occupations.

-0.2222 * (0.0920) 0.1252 * (0.0477) -0.2761 * (0.1166)

41

Dependent Variable Log Hourly Wage

Table 4 : Number of Employers, Share of time Employed and Average Duration Measures Early-Career United States France Variable Men Women Men Women Number of Employers in First 6 Months Number of Employers in Months 6-12 Number of Employers in Months 12-18 Number of Employers in Months 18-24 Share of Time Spent in Employment in First 6 Months Share of Time Spent in Employment in Months 6-12 Share of Time Spent in Employment in Months 12-18 Share of Time Spent in Employment in Months 18-24 Average Job Duration During First 6 Months Average Job Duration During Months 6-12 Average Job Duration During Months 12-18 Average Job Duration During Months 18-24

0.0581 (0.0304) -0.0005 (0.0356) 0.0259 (0.0361) 0.0001 (0.0353) -0.5020 * (0.2345) 0.0173 (0.2482) 0.0338 (0.2667) 0.5279 (0.2738) 0.5911 * (0.2228) -0.0477 (0.2211) 0.2118 (0.2292) -0.1131 (0.2292)

Men

-0.0741 * (0.0341) -0.0072 (0.0362) -0.0168 (0.0413) -0.0085 (0.0361) 0.7944 * (0.2686) 0.0477 (0.2648) 0.5292 (0.3071) -0.1026 (0.2790) -0.4973 * (0.2420) 0.1637 (0.2367) -0.3166 (0.2746) -0.0192 (0.2249)

Germany Women

-0.0024 (0.0059) -0.0072 (0.0072) 0.0099 (0.0068) -0.0032 (0.0061)

0.0070 (0.0056) 0.0006 (0.0067) -0.0082 (0.0062) 0.0153 * (0.0050)

Men

Netherlands Women

0.0670 (0.1196) 0.4167 (0.1267) 0.4050 (0.1210) 0.0400 (0.0695) 0.1136 (0.2925) -0.4509 (0.3305) -1.1211 (0.2171) -0.2754 (0.2463) 0.1919 (0.4849) 0.4901 (0.5657) 1.8210 (0.4378) 0.8380 (0.4300)

* *

*

*

-0.3398 (0.2657) -0.1626 (0.3000) 0.5023 (0.2614) 0.0238 (0.2673) -0.3970 (0.8334) 0.6710 (0.6892) -0.6052 (0.6288) 0.9631 (0.8278) 0.0060 (1.3200) -0.8035 (1.1789) 0.8643 (1.3400) -2.8883 (1.5180)

Log Monthly Income Number of Employers in First 6 Months Number of Employers in Months 6-12 Number of Employers in Months 12-18 Number of Employers in Months 18-24 Share of Time Spent in Employment in First 6 Months Share of Time Spent in Employment in Months 6-12 Share of Time Spent in Employment in Months 12-18 Share of Time Spent in Employment in Months 18-24 Average Job Duration During First 6 Months Average Job Duration During Months 6-12 Average Job Duration During Months 12-18 Average Job Duration During Months 18-24

-0.0061 (0.0004) -0.0028 (0.0004) 0.0074 (0.0005) -0.0030 (0.0004) -0.0333 (0.0055) 0.0816 (0.0046) -0.0128 (0.0047) 0.0455 (0.0039) 0.0430 (0.0137) -0.1013 (0.0127) -0.0042 (0.0136) -0.2426 (0.0143)

*

-0.0021 (0.0003) -0.0044 (0.0003) 0.0034 (0.0003) -0.0032 (0.0002) -0.0417 (0.0034) 0.0619 (0.0031) -0.0132 (0.0032) 0.0267 (0.0025) -0.0253 (0.0096) -0.0488 (0.0080) -0.0799 (0.0083) -0.1146 (0.0085)

*

-0.0204 (0.0102) 0.0090 (0.0112) -0.0042 (0.0119) 0.2094 (0.0095) -0.2135 (0.0798) 0.3145 (0.0670) -0.1368 (0.0698) -0.2483 (0.0551) 1.0183 (0.2160) -0.9072 (0.1805) 0.8860 (0.1905) 0.3342 (0.1710)

*

* * * * * * * * *

*

-0.0010 (0.0007) -0.0021 (0.0008) -0.0005 (0.0008) 0.0076 (0.0006) 0.0408 (0.0095) 0.0238 (0.0081) 0.0732 (0.0093) -0.0004 (0.0089) -0.0868 (0.0259) -0.1237 (0.0211) -0.1283 (0.0255) 0.0250 (0.0287)

*

* * * *

0.0499 (0.1396) 0.1842 (0.1478) 0.2998 * (0.1413) -0.0543 (0.0811) 0.3259 (0.3413) -0.2225 (0.3857) -0.7412 * (0.2534) -0.0665 (0.2874) -0.2871 (0.5660) -0.0518 (0.6602) 1.0568 * (0.5109) 0.3640 (0.5019)

0.0067 (0.0048) -0.0047 (0.0058) 0.0049 (0.0053) 0.0064 (0.0048)

0.0009 (0.0056) 0.0030 (0.0068) 0.0018 (0.0064) 0.0135 * (0.0051)

0.0080 (0.0061) -0.0094 (0.0076) 0.0002 (0.0069) 0.0085 (0.0061)

0.0028 (0.0071) 0.0141 (0.0086) 0.0024 (0.0077) 0.0038 (0.0063)

-0.0683 (0.0457) -0.0202 (0.0559) -0.0253 (0.0488) -0.0301 (0.0411)

0.0189 (0.1147) -0.0319 (0.1084) -0.0483 (0.1054) -0.0790 (0.0577) 0.1389 * 0.2880 (0.0428) (0.2819) -0.0077 0.1190 (0.0523) (0.2016) -0.0928 0.3119 (0.0495) (0.2087) 0.0515 -9.1522 (0.0408) (12.9962) -0.0124 (0.4455) -0.9721 * (0.4769) -0.3040 (0.4132) -0.5208 (0.3616)

* * *

-0.0624 (0.3301) 0.3021 (0.3727) 0.3942 (0.3248) 0.0617 (0.3321) -1.2871 (1.0356) 0.0803 (0.8563) -0.3143 (0.7814) 0.6828 (1.0285) 1.7682 (1.6401) 0.1431 (1.4648) 0.5526 (1.6650) -2.1290 (1.8862)

Log Full-Year Equivalent Earnings Number of Employers in First 6 Months Number of Employers in Months 6-12 Number of Employers in Months 12-18 Number of Employers in Months 18-24 Share of Time Spent in Employment in First 6 Months Share of Time Spent in Employment in Months 6-12 Share of Time Spent in Employment in Months 12-18 Share of Time Spent in Employment in Months 18-24 Average Job Duration During First 6 Months Average Job Duration During Months 6-12 Average Job Duration During Months 12-18 Average Job Duration During Months 18-24

* * * * * * * * * * *

-0.0010 (0.0005) -0.0024 (0.0005) 0.0014 (0.0006) 0.0017 (0.0005) -0.0002 (0.0061) -0.0028 (0.0052) 0.0258 (0.0061) 0.0247 (0.0055) 0.0565 (0.0167) -0.0944 (0.0144) -0.0119 (0.0173) -0.1379 (0.0149)

*

0.0288 (0.0111) 0.0113 (0.0120) 0.0427 (0.0128) 0.2398 (0.0101) -0.0959 (0.0971) -0.0776 (0.0798) -0.4607 (0.0840) -0.5916 (0.0738) 1.2698 (0.2587) 0.0640 (0.2083) 2.3489 (0.2300) 1.0545 (0.2260)

*

* * *

* * * *

*

Employment Probability Number of Employers in First 6 Months Number of Employers in Months 6-12 Number of Employers in Months 12-18 Number of Employers in Months 18-24 Share of Time Spent in Employment in First 6 Months Share of Time Spent in Employment in Months 6-12 Share of Time Spent in Employment in Months 12-18 Share of Time Spent in Employment in Months 18-24 Average Job Duration During First 6 Months Average Job Duration During Months 6-12 Average Job Duration During Months 12-18 Average Job Duration During Months 18-24

0.0009 (0.0051) -0.0109 (0.0054) 0.0077 (0.0061) -0.0055 (0.0055) -0.0145 (0.0396) 0.0610 (0.0398) -0.0749 (0.0436) 0.0974 (0.0435) 0.0341 (0.0373) -0.0834 (0.0355) 0.0900 (0.0391) 0.0036 (0.0375)

*

*

* *

0.0112 (0.0111) -0.0148 (0.0116) 0.0084 (0.0126) 0.0221 (0.0118) 0.0861 (0.0883) -0.0066 (0.0854) 0.0317 (0.0914) 0.0558 (0.0863) 0.0533 (0.0804) -0.0286 (0.0755) 0.1161 (0.0823) 0.1658 * (0.0725)

* * *

* * * *

* *

* * *

* *

Sources: United States - NLSY and authors' calculations; France - DADS and authors' calculations; Germany - GSOEP and authors' calculations; Netherlands - OSA and authors' calculations. Notes for Log Hourly Wage , Log Monthly Income and Log Full-Year Equivalent Earnings . Least squares standard errors in all regressions are corrected for arbitrary heteroskedasticity. NL: Hourly wages and Monthly Income measured in levels, not logs. US: Regressions control for log hours, years of education, age, age2, seniority, seniorty2, rural residency, nonwhite, married, 6 entry cohorts, 9 years, 4 regions, 8 sectors and 6 occupations. FR: Regressions control for 8 educational categories, age, age2, age3, age4, seniority, seniorty2, Paris region, 8 entry cohorts, 13 years, 8 sectors and 6 occupations. DE: Regressions control for 9 educational categories plus years of education, age, age2, age3, age4, seniority, seniorty2, 8 entry cohorts, 9 years, marital status, 4 firm sizes and 3 occupations. NL: Regressions control for years of education, age, age2, age3, age4, seniority, seniorty2, firm size, firm size2, 10 entry cohorts, 7 years, cohabitation, 6 sectors and 6 occupations. Notes for Employment Probability . US: linear probability model of percentage of time employed with heteroskedsticity-consistent standard errors. FR, DE and NL: Tobit of percentage of time employed bounded above by 1 and below by 0. US: Regressions control for log hours, years of education, age, age2, seniority, seniorty2, rural residency, nonwhite, married, 6 entry cohorts, 9 years, 4 regions, 8 sectors and 6 occupations. FR: Regressions control for 8 educational categories, age, age2, age3, age4, 8 entry cohorts and 13 years. DE: Regressions control for 9 educational categories plus years of education, age, age2, age3, age4, 8 entry cohorts, 9 years and marital status. NL: Regressions control for years of education, age, age2, age3, age4, seniority, seniorty2, firm size, firm size2, 10 entry cohorts, 7 years, cohabitation, 6 sectors and 6 occupations.

-0.0355 (0.1460) -0.0710 (0.1526) 0.0935 (0.1362) -0.2044 (0.1420) 0.4336 (0.3460) -0.3822 (0.3210) 0.6190 (0.4289) -18.8888 * (9.2110) -0.4417 (0.7014) -0.4153 (0.6159) 0.8430 (0.6817) -1.0750 (0.7870)

42

Table 5 : Number of Occupations and Number of Industries Measures Early-Career United States France Variable Men Women Men Women

Dependent Variable Log Hourly Wage

Number of Different Occupations in First 6 Months Number of Different Occupations in Months 6-12 Number of Different Occupations in Months 12-18 Number of Different Occupations in Months 18-24 Number of Different Industries in First 6 Months Number of Different Industries in Months 6-12 Number of Different Industries in Months 12-18 Number of Different Industries in Months 18-24

-0.0126 (0.0358) -0.0387 (0.0288) 0.0183 (0.0377) -0.0474 (0.0287) 0.0055 (0.0407) 0.0042 (0.0310) 0.0169 (0.0422) 0.0096 (0.0315)

Men

Germany Women

-0.0986 * (0.0403) 0.0241 (0.0330) 0.0162 (0.0449) -0.0286 (0.0338) 0.0924 * (0.0414) -0.0027 (0.0403) -0.0267 (0.0505) -0.0216 (0.0420)

Netherlands Men Women -0.1577 0.6527 * (0.1123) (0.2571) -0.2954 * 0.0388 (0.1117) (0.2331) -0.2940 * -0.1668 (0.1070) (0.1521) -0.0709 -0.1602 (0.0621) (0.2762)

Log Monthly Income Number of Different Occupations in First 6 Months

0.0118 (0.0030) 0.0387 (0.0033) -0.0540 (0.0031) 0.0342 (0.0025) 0.0134 (0.0036) -0.0482 (0.0037) 0.0299 (0.0035) -0.0044 (0.0027)

Number of Different Occupations in Months 6-12 Number of Different Occupations in Months 12-18 Number of Different Occupations in Months 18-24 Number of Different Industries in First 6 Months Number of Different Industries in Months 6-12 Number of Different Industries in Months 12-18 Number of Different Industries in Months 18-24

* * * * * * *

0.0120 (0.0050) -0.0438 (0.0055) -0.0176 (0.0057) 0.0331 (0.0042) -0.0653 (0.0065) 0.0761 (0.0052) 0.0254 (0.0063) -0.0607 (0.0054)

* * * *

-0.1283 (0.1310) -0.0941 (0.1304) -0.1841 (0.1248) 0.0238 (0.0725)

0.6918 * (0.3195) -0.2099 (0.2896) -0.3015 (0.1890) -0.1395 (0.3432)

-0.0454 (0.0995) 0.0115 (0.0945) 0.0133 (0.0909) 0.0751 (0.0526)

-0.2341 (0.1285) -0.0490 (0.1252) 0.0486 (0.0791) 0.1973 (0.1293)

* * * *

Log Full-Year Equivalent Earnings Number of Different Occupations in First 6 Months

-0.0097 (0.0020) 0.0159 (0.0023) -0.0071 (0.0021) 0.0190 (0.0014) 0.0164 (0.0023) -0.0136 (0.0022) 0.0088 (0.0023) 0.0112 (0.0018)

Number of Different Occupations in Months 6-12 Number of Different Occupations in Months 12-18 Number of Different Occupations in Months 18-24 Number of Different Industries in First 6 Months Number of Different Industries in Months 6-12 Number of Different Industries in Months 12-18 Number of Different Industries in Months 18-24

* * * * * * * *

-0.0075 (0.0034) -0.0089 (0.0036) -0.0345 (0.0040) 0.0375 (0.0033) -0.0096 (0.0044) 0.0268 (0.0041) 0.0363 (0.0041) -0.0272 (0.0037)

* * * * * * * *

Employment Probability Number of Different Occupations in First 6 Months Number of Different Occupations in Months 6-12 Number of Different Occupations in Months 12-18 Number of Different Occupations in Months 18-24 Number of Different Industries in First 6 Months Number of Different Industries in Months 6-12 Number of Different Industries in Months 12-18 Number of Different Industries in Months 18-24

-0.0090 (0.0061) 0.0045 (0.0049) -0.0042 (0.0064) 0.0002 (0.0048) 0.0111 (0.0068) 0.0023 (0.0051) 0.0112 (0.0072) -0.0001 (0.0053)

0.0010 (0.0131) -0.0016 (0.0106) -0.0044 (0.0144) -0.0066 (0.0110) 0.0106 (0.0133) 0.0016 (0.0128) 0.0019 (0.0163) 0.0205 (0.0134)

0.0849 (0.0461) 0.1915 (0.0508) -0.1088 (0.0472) -0.1354 (0.0365) 0.0304 (0.0534) -0.2547 (0.0518) 0.1838 (0.0514) -0.0677 (0.0440)

* * *

* *

0.0181 (0.0569) 0.0256 (0.0624) -0.0830 (0.0590) 0.1445 (0.0460) -0.2395 (0.0690) 0.1776 (0.0653) 0.0316 (0.0625) -0.2105 (0.0533)

* * *

*

Sources: United States - NLSY and authors' calculations; France - DADS and authors' calculations; Germany - GSOEP and authors' calculations; Netherlands - OSA and authors' calculations. Notes for Log Hourly Wage , Log Monthly Income and Log Full-Year Equivalent Earnings . Least squares standard errors in all regressions are corrected for arbitrary heteroskedasticity. NL: Hourly wages and Monthly Income measured in levels, not logs. US: Regressions control for log hours, years of education, age, age2, seniority, seniorty2, rural residency, nonwhite, married, 6 entry cohorts, 9 years, 4 regions, 8 sectors and 6 occupations. FR: Regressions control for 8 educational categories, age, age2, age3, age4, seniority, seniorty2, Paris region, 8 entry cohorts, 13 years, 8 sectors and 6 occupations. DE: Regressions control for 9 educational categories plus years of education, age, age2, age3, age4, seniority, seniorty2, 8 entry cohorts, 9 years, marital status, 4 firm sizes and 3 occupations. 2

3

4

2

2

NL: Regressions control for years of education, age, age , age , age , seniority, seniorty , firm size, firm size , 10 entry cohorts, 7 years, cohabitation, 6 sectors and 6 occupations. Notes for Employment Probability . US: linear probability model of percentage of time employed with heteroskedsticity-consistent standard errors. FR, DE and NL: Tobit of percentage of time employed bounded above by 1 and below by 0. US: Regressions control for log hours, years of education, age, age2, seniority, seniorty2, rural residency, nonwhite, married, 6 entry cohorts, 9 years, 4 regions, 8 sectors and 6 occupations. FR: Regressions control for 8 educational categories, age, age2, age3, age4, 8 entry cohorts and 13 years. DE: Regressions control for 9 educational categories plus years of education, age, age2, age3, age4, 8 entry cohorts, 9 years and marital status. 2 3 4 2 2 NL: Regressions control for years of education, age, age , age , age , seniority, seniorty , firm size, firm size , 10 entry cohorts, 7 years, cohabitation, 6 sectors and 6 occupations.

43 Dependent Variable Log Hourly Wage

Table 6 : Simulated Effects of Changes in Early Career Experiences United States France Thought Experiment Men Women Men Women Baseline Adjusted R-Squared Number of Observations Find First Job Immediately After School Find First Long Job Immediately After School No Earnings Growth During First 2 Years Double the Observed Earnings Growth Between Years 1 and 2 Only 1 Employer Throughout First 2 Years No Unemployment In First 2 Years 1 Employer, No Unemployment During First 2 Years Only 1 Occupation Throughout First 2 Years Only 1 Industry Throughout First 2 Years

-2.570 0.295 4794 -2.601 -2.573 -2.571 -2.570 -2.602 -2.547 -2.416 -2.557 -2.572

-2.830 0.311 4323 -2.864 -2.850 -2.827 -2.833 -2.795 -1.978 -2.104 -2.826 -2.834

0.000

0.000

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

Baseline Adjusted R-Squared Number of Observations Find First Job Immediately After School Find First Long Job Immediately After School No Earnings Growth During First 2 Years Double the Observed Earnings Growth Between Years 1 and 2 Only 1 Employer Throughout First 2 Years No Unemployment In First 2 Years 1 Employer, No Unemployment During First 2 Years Only 1 Occupation Throughout First 2 Years Only 1 Industry Throughout First 2 Years

0.000

0.000

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

Baseline Adjusted R-Squared Number of Observations Find First Job Immediately After School Find First Long Job Immediately After School No Earnings Growth During First 2 Years Double the Observed Earnings Growth Between Years 1 and 2 Only 1 Employer Throughout First 2 Years No Unemployment In First 2 Years 1 Employer, No Unemployment During First 2 Years Only 1 Occupation Throughout First 2 Years Only 1 Industry Throughout First 2 Years

0.000

0.000

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

Baseline Adjusted R-Squared / Log Likelihood Number of Observations Find First Job Immediately After School Find First Long Job Immediately After School No Earnings Growth During First 2 Years Double the Observed Earnings Growth Between Years 1 and 2 Only 1 Employer Throughout First 2 Years No Unemployment In First 2 Years 1 Employer, No Unemployment During First 2 Years Only 1 Occupation Throughout First 2 Years Only 1 Industry Throughout First 2 Years

0.878 0.087 6141 0.883 0.898 0.878 0.878 0.880 0.921 0.932 0.878 0.877

Germany Men Women

Netherlands Men Women

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

2.606 0.845 1594 2.594 2.635 2.606 2.607 2.606 2.613 2.613 2.606 2.606

2.320 0.841 1197 2.306 2.312 2.327 2.313 2.320 2.287 2.287 2.320 2.320

14.190 0.642 250 14.175 14.197 14.167 14.213 14.447 13.305 14.508 13.927 14.190

9.656 0.525 212 9.648 9.653 9.652 9.660 9.607 9.946 9.278 9.732 9.581

1.240 0.877 31494 1.252 1.240 1.240 1.239 1.242 1.277 1.147 1.247 1.236

0.842 0.822 20822 0.917 0.845 0.842 0.842 0.826 0.914 0.767 0.838 0.829

7.795 0.677 2134 7.759 7.801 7.795 7.796 7.795 7.765 7.765 7.795 7.795

7.415 0.262 1517 7.408 7.392 7.407 7.422 7.415 7.371 7.371 7.415 7.415

2375.032 0.553 250 2375.023 2375.057 2375.030 2375.034 2375.154 2374.675 2375.107 2374.912 2375.032

1297.190 0.390 212 1297.204 1297.241 1297.131 1297.249 1297.278 1296.954 1297.051 1297.210 1297.165

3.641 0.815 20822 3.697 3.651 3.640 3.641 3.638 3.662 3.579 3.635 3.649

10.669 0.229 1781 10.633 10.678 10.669 10.670 10.669 10.652 10.652 10.669 10.669

10.306 0.271 1240 10.295 10.279 10.302 10.311 10.306 10.260 10.260 10.306 10.306

0.000

0.000

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

3.992 0.944 31494 4.018 4.010 3.993 3.991 4.004 4.004 3.901 3.991 4.001

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

0.748 0.076 6198 0.729 0.734 0.748 0.748 0.742 0.859 0.915 0.749 0.748

0.707 -31149 50370 0.713 0.382 0.709 0.704 0.123 0.551 0.535 0.756 0.665

0.574 -28675 31854 0.734 0.325 0.575 0.573 0.000 0.000 1.000 0.585 0.477

0.807 -1630 2275 0.652 1.000 0.810 0.804 0.807 1.000 1.000 0.807 0.807

0.736 -1437 1739 0.222 0.628 0.754 0.717 0.736 0.577 0.577 0.736 0.736

0.895 32.660 271 0.897 0.917 0.904 0.886 0.856 0.000 0.000 0.910 0.895

0.838 2.438 240 0.952 0.859 0.871 0.805 0.756 0.000 0.000 0.853 0.880

Log Monthly Income

Log Full-Year Equivalent Earnings

Employment Probability

Notes: Simulations are based on coefficients in tables 3 to 5 and are evaluated for a person with the mean values of the relevant explanatory variables as found in tables 1a to 1d. See the notes to tables 1a - 1d and 3 - 5 for further information.