Does Employer Learning Vary by Occupation?

[Job Market Paper]

Hani Mansour

University of California at Santa Barbara�

November, 2008

Abstract

In this paper I provide evidence that initial occupational assignments a¤ectthe ability revelation of young workers in the labor market. Di¤erences in em-ployer learning about worker�s ability carry two implications: First, the growthin wage dispersion with experience will di¤er across initial occupations; Second,the growth in the coe¢ cient of a variable that employers initially do not observe,such as the Armed Forces Quali�cation Test (AFQT) score, will also vary acrossoccupations. Empirically, I �nd that occupations with high growth in the vari-ance of residual wages over the �rst ten years of the worker�s career are also theoccupations with high growth in the AFQT coe¢ cient. In contrast, the resultsfor occupations with low growth in the variance of residual wages suggest thatemployers have learned about the worker�s AFQT score at the time of hire. Ialso investigate the informational role of race across occupations. The coe¢ cienton the Black variable in occupations with low growth in the residual variance ispositive and decreases over time, but the racial gap widens over time in occupa-tions with high growth in the residual variance. These results have implicationsfor di¤erent topics in labor economics, such as occupational mobility, and thesignaling value of education.

JEL Codes: J24, J31, J71, J62

Keywords: Wage Dynamics, Occupational Choice, Statistical Discrimination.

�I am grateful to Peter Kuhn, Kelly Bedard, Olivier Deschênes, Peter Rupert, Javier Birchenall, andCatherine Weinberger for helpful comments and continuous support.

1 Introduction

Employers often hire new inexperienced workers without observing their full productivity.

Instead, they assess a worker�s value to the �rm based on information they receive from

job interviews, resumes, and recommendations (Spence, 1973). Altonji and Pierret (2001)

[hereinafter AP] show that if employers learn about the worker�s productivity, the coe¢ cient

of an ability correlate which is initially unobserved by employers, such as the Armed Forces

Quali�cation Test (AFQT) score, increases with experience.

AP�s analysis, and most other models that use a learning framework, assume that learning

is independent of job assignment. It is likely, however, that the amount of initial information

workers bring to the market varies across occupations and that employers learn about the

productivity of their workers at di¤erent rates.1

The main hypothesis in this paper is that initial occupational assignments a¤ect the

ability revelation of young workers in the labor market. I test this hypothesis by combining

AP�s insight with another implication of employer learning, namely that the distribution of

wages becomes more dispersed as a cohort of workers gains experience (Neal and Rosen,

1998). Combining these two observations, we would expect the AFQT coe¢ cient to grow

the most in occupations with high growth in wage dispersion.2

The main contribution of the paper is to provide evidence that initial occupational as-

1An exception is Altonji (2005) who presents a framework in which the rate of employer learning dependson the skill level of the job to show how, in this environment, statistical discrimination at the time ofhire a¤ects employment rates and wage growth, but he does not test the model empirically. In addition,Antonovics and Golan (2007), motivate the idea of "job shopping" within �rms using a model in whichlearning di¤ers across jobs.

2Matching models, such as Jovanovic�s (1979a) model, also generate increased wage dispersion withexperience. Although I adopt a learning framework, the main point in this paper is to show that the �owof information (whether about worker�s ability or quality of match) di¤ers across initial occupations. Miller(1984) discusses an environment where workers learn about the quality of their match at di¤erent ratesacross occupations.

1

signments are associated with di¤erent learning parameters. The results have important

implications for various models in labor economics that use a learning framework, but as-

sume that learning is independent of job assignment. These include models of statistical

discrimination, occupational mobility, wage dynamics within �rms, occupational wage dif-

ferences, earnings inequality, and labor market signaling.3

The model in the paper follows that of AP.4 Identical employers form expectations about

the worker�s productivity, and in each period update their initial belief based on a noisy signal

of output produced by the worker. Di¤erently from AP, the technology I specify assumes

that the precision of the signal of output di¤ers across occupations. If these di¤erences are

empirically signi�cant, the growth in the AFQT coe¢ cient and the growth in wage dispersion

will vary across initial occupations.5 ;6

Empirically, I use data from the Current Population Survey (CPS) Outgoing Rotation

Group (ORG) for the 1984-2000 period to calculate the two-digit occupation growth in

wage dispersion for workers with 0-5 and 6-10 years of experience.7 I then, merge the

estimated growth rates with the �rst occupation that workers report in the 1979 National

3Some of the relevant references are Topel and Ward (1992), Gibbons et. al (2005), Gibbons and Waldman(1999; 2006), Lemieux (2006), Lange (2007), among others.

4AP�s learning model closely follows that of Farber and Gibbons (1996) [hereinafter FB]. One importantdi¤erence between the two is that FB estimate a wage level regression while AP�s dependent variable is thelogarithm of wages.

5The �ow of information about the worker is public and observed by all labor market participants. Thedebate in the literature wether information �ows symmetrically across employers has not been resolved yet.Recent papers such as Kahn (2007) and Schönberg (2007) use a learning model to test whether informationbetween employers is symmetric. Although they use the same data set they reach opposite conclusions.

6Throughout the analysis, I do not model initial occupational choice and assume that it is associated witha �xed learning parameter for the worker�s entire career. In section 5, I show that the model�s predictionsand empirical results are consistent with an assignment rule in which higher-expected-ability workers chooseto work in occupations where output is sensitive to individual ability.

7Ideally, I would want to follow the growth in the variance of residual wages for a cohort of workers whostarted their career in the same occupation over time, rather than comparing the variance of two cohortswithin an occupation. This is not feasible in the NLSY79 because of sample size limitations. In section 6, Iperform robustness checks to address this concern.

2

Longitudinal Survey of Youth (NLSY79) and compare the growth in the AFQT coe¢ cient

across occupations with di¤erent growth rates in wage dispersion.8

My main �nding is that occupations with high growth in the variance of residual wages

over the �rst 10 years are also the occupations with high growth in the AFQT coe¢ cient over

the same period. This provides further support for the notion that the market learns some of

the worker�s characteristics, which are correlated with AFQT score, over the �rst part of the

career. It also suggests that initial occupational sorting may have important implications for

learning. I call occupations with high levels of growth in wage dispersion and in the AFQT

coe¢ cient "high learning" occupations. As for "low learning" occupations, I �nd that the

AFQT coe¢ cient is large and does not grow over time, and that they are associated with an

initial high level of residual variance. This indicates that, in these occupations, the market

has learned the worker�s AFQT score at the start of the career. I apply my results to test for

statistical discrimination, and show that the evolution of the Black coe¢ cient di¤ers across

high and low learning occupations.

The rest of the paper proceeds as follows. Section 2 presents a learning model which

incorporates the idea of di¤erential learning rates across jobs. Section 3 describes the data,

and section 4 includes the empirical analysis and results. Section 5 speci�es an initial occu-

pational assignment rule consistent with the model, and section 6 includes robustness checks.

Finally, section 7 concludes and discusses future work.

8Focusing on occupation-speci�c growth rates in the variance of residual wages compared to looking atlevels of residual wages is important since it accounts for the fact that employers in some occupations mighthave richer information about worker�s ability, that is not related to learning. It also accounts for the factthat levels of residual wages might be a mechanical consequence of how broadly or narrowly the occupationis de�ned and not of employer learning.

3

2 Model and Predictions

2.1 A Model of Employer Learning

The model builds on that of AP. Worker i enters the labor market with skill level, h, which

is decomposed into four components:

hi = asi + bqi + �zi + �i (1)

In (1), s represents variables observed by the employer and the econometrician (such as

education); q is a productivity component observed by the employer but is not in the data

(such as information obtained during a recruitment interview); z is a variable observed and

used only by the econometrician (such as AFQT score); and � is a productivity component

that neither the employer nor the econometrician observe. The focus in this paper is only

on time-invariant productivity components, such as innate ability, and abstracts from time-

variant components acquired, for example, from on-the-job training, although the two are

likely to be correlated.

Because employers do not observe z or �, they form expectations conditional on s and q.

In what follows, I suppress the index i. Assuming that these expectations are linear in both

s and q, de�ne:

z = E(zjs; q) + � = �1q + �2s+ � (2)

� = E(�js; q) + " = �1s+ �2q + "

4

By de�nition of an expectation, both � and " are uncorrelated with s and q and have mean

zero. Calculating the expected value of the skill level in (1) conditional on the information

available to employers upon entry to the labor market, and substituting (2) in (1) yields:

he = E(hjs; q) = (a+ ��2 + �1)s+ (b+ ��1 + �2)q + (��+ ") (3)

The term (��+ ") in (3) represents the initial error in the employer�s expectation about the

worker�s ability and is uncorrelated with s and q.

There are J jobs (or occupations) in the economy.9 The output in each job is a function

of the worker�s skill level, h, and the experience pro�le of productivity, Gj(t). The worker�s

output if assigned to job j at time t is:

yjt = exp(h+Gj(t) + �jt) , j = 1; :::; J (4)

Gj(t) is assumed to be independent from s; q; z;or �. The error term �jt is job-speci�c and

represents a productivity shock drawn from a normal distribution with mean 0 and variance

�2j .10 Denote Yjt = log(yjt) and de�ne xjt = [Yjt�(as+bq)�Gj(t)] = �z+�+�jt to be a noisy

signal of productivity that employers can compute at the end of each period. This signal is

public and is observed by all employers. Di¤erently from previous studies the precision of the

signal di¤ers across occupations.11 Since employers form expectations about z and � based

9Throughout the paper I use the terms job type and occupation interchangeably. The empirical analysiswill use occupations to test for di¤erential employer learning.10The paper�s objective is to provide empirical evidence that learning rates di¤er across occupations.

Assuming that the variance of the productivity shock di¤ers across occupations is one possible mechanismthat generates such patterns. Other mechanisms, such as the one I present in section 5, also generate suchdi¤erences.11In the empirical analysis I use the two-digit occupation codes to look at learning across jobs. This

implicitly assume that the i.i.d shock, �, is the same across all the individual occupations under each code

5

on s and q, observing xjt is equivalent to observing bjt = xjt�E(hjs; q) = ��+"+�jt, which

is the sum of the initial error upon entry to the labor market and the productivity shock. The

worker�s output history in occupation j is summarized by the vector Bjt = fbj1; bj2; :::; bjtg.

Wages are set by spot-market contracting, and at the end of each period workers are paid

their expected output. That is, Wjt = E[yjtjs; q; Bjt] exp(� jt) where exp(� jt) is a classical

measurement error unrelated to other variables in the model. Using equations (1), (2), and

(4) we can derive the wage process:

Wjt = eGj(t) � es(a+��2+�1) � eq(b+��1+�2) � eE[��+"jBjt] � E(e�jt) � E(evjt) � e�jt (5)

where �jt is the di¤erence between [�� + "] and E[�� + "jBjt], and is assumed to be inde-

pendent of s,q, or Bjt. It follows that the log wage process in occupation j is:

wjt = G�j(t) + (a+ ��2 + �1)s+ (b+ ��1 + �2)q + E[��+ "jBjt] + � jt (6)

where wjt = log(Wjt) and G�j(t) = Gj(t)+log(E(e�jt))+log(E(evjt)). The term E[��+"jBjt]

re�ects the idea that the wage in occupation j changes because new information about

worker�s ability is extracted. For the purpose of this paper, I de�ne an occupation as the

entire expected career track associated with choosing a given initial occupation, so that

a worker�s entire career is associated with a �xed learning parameter. Initially, I do not

model initial occupational sorting, but in section 5 I propose an assignment rule to initial

occupations which is consistent with the �ndings of the paper.

and does not take into account the fact that within-occupation job heterogeneity might be substantial.

6

2.2 Estimation Framework

Recall that only variables in s and z are observed in the data and can be used for the

empirical analysis. Assuming that s and z are scalars, the log wage process from equation

(6) can be summarized in the following regression:

wjt = �sjts+ �zjtz +G�j(t), t = 0; :::; T (7)

The main insight of AP is that if employers learn and statistically discriminate, the coe¢ cient

on a correlate of productivity which is not observed by the employer upon hiring, such as

AFQT score, should increase over time, while, using the same logic, the coe¢ cient on an

easily observed variable, such as education, should decrease over time. The hypothesis in

regard to whether employers learn can be tested within each occupation. Proposition 1

summarizes this result and the proof can be found in AP.12

Proposition 1 Under the assumptions of the model, the regression coe¢ cient �zjt is non-

decreasing in t, and the regression coe¢ cient �sjt is nonincreasing in t.

If learning varies by occupation, the evolution of �zjt will di¤er across occupations. Specif-

ically, �zjt will grow more with experience in occupations where employers extract more

information about the worker�s ability.

We can estimate equation (7) by occupation and compare the evolution of the hard-

to-observe variable over time in order to test for di¤erences in employer learning. However,

because of sample size limitations, I use another insight from the model to group occupations12The wage regressions in AP control for initial occupations �xed e¤ects, but do not analyze the wage

dynamics within these occupations. The proof of proposition 1 is not a¤ected by introducing di¤erentinformational structure across occupations.

7

based on the growth in the variance of residual wages, and then test for di¤erences in the

evolution of hard-to-observe variables over time across these groups.

2.3 Variance of Residual Wages

A straightforward implication of employer learning is that the distribution of wages of a

cohort of workers becomes more dispersed as they gain experience (Neal and Rosen, 1998).

If employer learning di¤ers across occupations then we should expect di¤erent growth rates

in wage dispersion with experience.13

The use of the growth in the variance of residual wages as a measure of employer learning

is important since level di¤erences in the variance of residual wages across occupations might

re�ect, among other things, di¤erences in the amount of information employers have on their

employees at the time of hire. It also accounts for the fact that levels of residual wages might

be a mechanical consequence of how broadly or narrowly the occupation is de�ned and not

of employer learning.

To clarify the idea consider the following wage equation:

wijt = xijtrjt + �ijt (8)

where wijt is the natural logarithm of the hourly wage rate for person i in occupation j with

t years of labor market experience; xijt is a vector of observable skills such as education; rjt13Theoretically employer learning generates growth in wage dispersion. However, there are other reasons

which can explain growth in wage dispersion within and across occupations. For example, the growth inthe wage dispersion in traditionally unionized occupations might be small even if employer learning ratesare high (Lemieux, 2006). Ranking occupations based on growth in wage dispersion o¤ers a simple way toanalyze proposition 1 where the learning hypothesis is tested directly.

8

is the price of observable skills in occupation j; and �ijt is an occupation-speci�c regression

residual. The regression residual can be decomposed into two parts (Lemieux, 2006):

�ijt = pjtaijt + �ijt (9)

where pjt is the price of unobserved skill in occupation j, aijt is unobserved skill, and �ijt is

measurement error. Calculating the variance of the residual yields:

V ar(�ijt) = p2jtV ar(aijt) + V ar(�ijt) (10)

An increase in V ar(�ijt) with experience can be due to an increase in pjt, an increase in

V ar(aijt), and an increase in V ar(�ijt). Focusing on low levels of experience, we can safely

assume that the price of unobserved ability in occupation j does not change with experience

and that the variance of the measurement error does not grow systematically with experience,

so that pjt = pjr and V ar(�ijt) = V ar(�ijr) for t 6= r. Under these assumptions, the growth

in residual variance between t and r years of experience in occupation j where r > t is:

V ar(�ijr)� V ar(�ijt) = p2j [V ar(aijr)� V ar(aijt)] (11)

Equation (11) implies that even if the price of unobserved skill varies across occupations (for

example, pj > pk for j 6= k), observing di¤erent growth rates in the variance of residual wage

across occupations can be attributed, under reasonable set of assumptions, to di¤erences in

the growth of the variance of unobserved ability across occupations. Thus, a faster growth in

the variance of residual wages is evidence for faster updating of the employer�s beliefs about

9

the worker�s productive ability. This insight o¤ers a simple way to group occupations based

on a consistent measure of learning, which I will use to test the results of proposition 1.

3 Data

Most of the analysis in this paper uses data drawn from the NLSY79 covering the 1979-

2000 period. The NLSY79 is a nationally representative sample of 12,686 men and women

who were between the ages of 14 and 22 when they were �rst interviewed in 1979. The

survey was conducted annually until 1994 and since then the participants are interviewed

on a biennial basis. The NLSY79 has been used in many of the studies of employer learning

because of two main features: detailed information on work experience allows the calculation

of actual rather than potential labor market experience, and most importantly, the sample

includes some variables that are correlated with ability but are not observed (or used) by

the employer, such as AFQT scores and father�s education.

The work history �le in the NLSY79 provides information on the hours worked in each

week of the year. In order to calculate actual labor market experience, I follow AP�s method-

ology and accumulate the number of weeks in which the worker reports to have worked more

than 30 hours and divide it by 50. I focus only on jobs after an individual has left school

for the �rst time and drop all observations before that. The month and year when a person

left school for the �rst time is directly reported by the respondents and is used as the entry

date to the labor market. An individual can go back to school and remain in the sample.

To construct my sample, I include only white or black men which leaves me with 5,403

individuals. I drop 2,256 respondents who left school before 1979 because information on

10

their �rst occupation and actual labor market experience is hard to determine. I drop 362

respondents who never report to have left school and 8 individuals who do not report the

month they left school for the �rst time. 5 respondents who never completed more than 8

years of schooling are excluded from the sample, and 101 individuals who do not have a valid

AFQT score are dropped. I �nally drop 709 additional individuals who do not report their

�rst occupation.

Aside from initial occupation, individuals in the NLSY79 report information on all the

jobs they held between two interviews. I only use information from the job they hold at the

time of the interview (CPS item). I exclude jobs without pay, jobs at home, and military

jobs. The wage measure in the sample is the hourly wage rate of pay at the most recent

job from the CPS section of the NLSY79. I use CPS de�ators to calculate real wages in

1984 prices and I drop wages below $1 or above $100. My �nal sample includes 22,409

observations on 1,962 distinct individuals.

Since NLSY79 participants took the AFQT at di¤erent ages, I standardize the AFQT

scores to have mean zero and a standard deviation of 1. I do this by subtracting the mean

score for a person of that age group and divide it by the standard deviation for that age

group. Table 1 provides summary statistics for the main variables I use in the empirical

analysis.

In order to calculate growth in residual variance by occupation, I use information from

the ORG CPS. The optimal way to calculate growth in the residual variance and attribute

it to employer learning would be to track an individual at di¤erent points in the experience

pro�le. Since a large panel that allows analysis within occupations is not available, I use

a cohort analysis in which the residual variance of workers with 0-5 years of experience in

11

an occupation is compared to workers with 6-10 years of experience in the same occupation

and year. The ORG data set is adequate for the purposes of this paper because it contains

information about the hourly wage rate, the same measure I use in the NLSY79 sample.

Moreover, as Lemieux (2006) shows, the wage information in the ORG is less noisy compared

to the March CPS because it measures directly the hourly wage of workers paid by the hour.

I pool data from 1984-2000 and keep in the sample only white or black men who are

18-45 years old and have 0-10 years of potential experience.14 I exclude from the sample

self employed and full or part time students and individuals with below 8 years of schooling.

I use the hourly wage rate for workers paid by the hour (about 60 percent of the sample)

and calculate an hourly wage rate for the other workers by dividing their weekly earnings by

their usual weekly hours. Real wages are calculated in 1984 prices using the CPS de�ator.

I drop wages that are below $1 and above $100, and adjust top-coded earnings by a factor

of 1.4 (Lemieux, 2006). Since actual experience is not reported in the ORG, I calculate a

measure of potential experience (age-education-6). As it is well known, in 1992 the U. S.

Bureau of Labor Statistics changed the educational attainment question in the CPS from

one that was based on years spent in school to one that focuses on the highest degree

received. I use the method proposed by Jaeger (1997; 2003) to create a consistent measure

of years of schooling (to calculate potential experience) and to group individuals in consistent

educational categories.

Table 2 presents some summary statistics from the ORG sample. Panel A reports the

sample share of four educational groups in the 1984-2000 period. As can be seen, the

14I analyzed other shorter periods of time such as 1984-1988 and 1992-2000. The results regarding thegrowth in the variance of wage residuals within occupations are not sensitive to the time period. In order toincrease my sample size I choose to pool the entire period of 1984-2000.

12

distribution across the sample period remains relatively constant with a steady increase

throughout the period in the share of college graduates and those with some college. Panel

B reports the sample share of four experience groups: workers with less than �ve years of

experience and workers with 6-10 years of experience. The distribution across experience

groups is similar across the sample years with a small increase in the share of workers with

6-10 years of experience. This suggests that compositional e¤ects over the sample period

are small. In a separate analysis which I do not report here, I compute the distribution of

occupations over the sample period and �nd no evidence of substantial changes. Moreover, I

analyzed the distribution of education by occupation-experience cells and found no evidence

for signi�cant changes in the education level within an occupation over the experience pro�le.

4 Empirical Speci�cations and Results

4.1 Ranking Occupations

As mentioned earlier, the growth in the wage dispersion of workers over time can be at-

tributed to employer learning. Ideally, in order to rank occupations based on employer

learning we would want to follow the growth in wage dispersion of a cohort of workers who

start their career in the same occupation, over time. Instead, I calculate the growth in the

variance of residual wages for two experience-cohorts within occupation in a given period of

time.

Although this growth measure does not account for the possibility that occupational

mobility might di¤er across initial occupations, it has two main advantages. First, this

13

growth measure does not confound the increase in wage dispersion due to learning with the

increase in wage dispersion due to an increase in the price of unobserved skill because of

skill-biased technological change (Katz and Murphy, 1992). Second, the growth measure will

not be contaminated by potential composition e¤ects due to large changes in the education

and experience of the U.S. population (Lemieux, 2006; Card and DiNardo, 2002). In section

6, I address the concern regarding occupational mobility, and show that it is not what derives

my results.

I start the empirical analysis by computing the variance of the residual wage by occupation-

experience cells using ORG data. The residual wage is computed from a �exible regression

of log hourly wage on an unrestricted set of dummies for age and years of schooling. I also

include interaction terms between six schooling dummies and a quartic in age.15 To capture

any year-speci�c di¤erences I include a full set of year dummies. I also include a set of oc-

cupation dummies to capture di¤erences in the level of wage residuals between occupations.

I weight the regression using weights provided by the CPS.

To calculate the variance in the residual wages, I use the 1980 2-digit occupation codes

to create 45 occupation dummies and interact them with 2 experience cells, one for work-

ers with 0-5 years of experience, and the other for workers with 6-10 years of experience.

The coe¢ cients of a regression of the squared residuals from the wage regression on the

occupation-experience cells (90 coe¢ cients) are the coe¢ cients of interest. I calculate the

growth in the variance between these two experience categories and rank occupations based

on this measure. I focus on the �rst 10 years of the worker�s career because employer learning

15The six schooling dummies I use to interact with a quartic in age are 9-10, 11, 12, 13-15, 16, and 17+. Ido not include an unrestricted set of age-education dummies because many of the cells are empty. I do notinclude other typical demographic variables like marital status and race in order to focus on direct measuresof skill.

14

is likely to happen at the beginning of the career, so that the growth in the variance of wage

residuals at later stages in the experience pro�le might re�ect other factors rather than em-

ployer learning (Lange, 2007).16 I test whether the growth in the variances are statistically

di¤erent from zero (reported in Table 3). I also apply an F-test and reject the hypothesis

that the occupation speci�c residual variances are all equal to each other.

I want to emphasize, before presenting the results, that ranking occupations using growth

in the variance of residual wages provides an indirect way to group occupations based on

employer learning, but one can suggest alternative explanations for these di¤erences (such as

di¤erent types of contracts across occupations, and di¤erential on-the-job training). Even-

tually, however, these occupation-speci�c values are going to be matched with NLSY data

where I test directly the learning hypothesis.

Table 3 provides a list of the occupations ranked by the growth in the variance of residual

wages, and lists the average years of schooling completed by occupation. A large growth rate

indicates that the amount of learning that takes place in the �rst 10 years in these occupa-

tions is large, while employers in occupations with low growth in the residual variance do not

learn much about the worker in the same time period. Occupations with high growth in the

residual variance include health diagnosing occupations, sales workers, and administrators

while occupations with low growth in the residual variance include engineers, teachers, and

farm operators. Notice that the ranking of occupations based on the growth in residual vari-

ance is di¤erent from ranking occupations based on mean education. For example, physicians

16Calculating the growth in the variance of wage residuals between workers with 0-5 years of experienceand workers with 6-10 years of experience allows more precision in the estimation because of larger samplesize. I also experimented comparing the variances of workers with 0-2 years of experience and 3-5 years ofexperience. The results are very similar to the ones I report here (but are estimated less precisely) and theranking of occupations is robust to this choice.

15

have the highest growth in the variance while college professors experience no growth, but

both have high mean education. Nonetheless, some occupations with low or no growth in

the residual variance appear to have high mean education, perhaps re�ecting the fact that

employers in these occupations have more information about their workers�abilities upon

their hire.

4.2 Evidence for Employer Learning

In this section, I reproduce AP�s results using the updated sample. As discussed in section

2, the idea of testing for employer learning hinges on the availability of a variable which is

(positively) correlated with ability, is unobserved by the employer at the time of hiring but

is available to the econometrician. The empirical speci�cation that stems from equation (7)

when learning is assumed to independent of occupation is:

logWit = �s0si + �z0zi + Ti + �s10(si �Ti10) + �z10(zi �

Ti10) + it (12)

Throughout the empirical analysis education (s) and AFQT (z) are interacted with expe-

rience divided by 10, so that the coe¢ cients on the interaction terms represent the change in

the wage slope between T = 0 and T = 10. Table 4 reproduces AP�s results using the original

sample period of 1979-1992, and using the extended sample for the 1979-2000 period. All the

reported standard errors are based on White-Huber standard errors. In the regression, ex-

perience is modeled with a cubic polynomial, and I control for the 1980 two-digit occupation

codes at the �rst job, urban residence, race, year �xed e¤ects, and a cubic trend interacted

with education and race. The base year for the time trend is 1992. These interactions with

16

the time trend control for economywide changes in the wage structure during the sample

period (Katz and Murphy, 1992).

It is important to mention that the NLSY79 sample used in the analysis is very similar

to the one constructed by AP with two exceptions. First, the sample covers a longer period,

from 1979-2000, while in AP�s sample 1992 is the last year used. Second, individuals who left

school before 1979 are excluded from the sample while AP includes them. These workers are

excluded because their work history before 1979 is not complete and it is hard to determine

their �rst occupation. Although this restriction weakens the results when compared to the

ones reported by AP, the main picture regarding employer learning is not a¤ected.17

Column 1 in Table 4 reports the results of estimating equation (12) when the sample

covers the 1979-1992 period and the interaction of AFQT with experience is excluded. The

coe¢ cient on education is 0.06 and statistically signi�cant and does not change over time. As

in other studies that have used AFQT scores in wage regressions, it enters the equation with

a positive (0.03) and statistically signi�cant coe¢ cient (Neal and Johnson, 1996). Column

2 uses the same restricted sample but includes an interaction term between AFQT and

experience. As predicted by the model, the coe¢ cient on AFQT when T = 0 becomes

zero and statistically insigni�cant while the coe¢ cient on the interaction term enters the

regression with a large coe¢ cient. At T = 10 the coe¢ cient on AFQT is 0.07 and statistically

signi�cant. This indicates that employers do not observe the worker�s AFQT score upon

entry to the labor market but do learn about it over time. Extending the sample to cover

the 1979-2000 period does not change the main results. As reported in column 3 and 4 in

17In a separate analysis using only the 1979-1992 sample I include workers who left school before 1979.The results con�rm quite closely the ones reported by AP. These results can be provided upon request.

17

Table 4, the magnitude of the AFQT-experience coe¢ cient becomes smaller (0.05) but the

notion of employer learning remains unchanged.

4.3 Occupation-Speci�c Employer Learning

To test whether employer learning di¤ers across occupations, I match the �rst occupation

of each worker in the NLSY sample with the appropriate growth in the variance of residual

wages of that occupation.18 To start the analysis, I group occupations in terciles based on

the growth measure: The �rst group includes occupations where the growth in the variance is

greater or equal to 0.036, the second includes occupations where the growth in the variance

is between 0.010 and 0.036, and the third includes occupations where the growth in the

variance is below 0.010 (and is mostly not statistically signi�cant). If the growth measure

re�ects (at least partially) employer learning, the growth in the AFQT coe¢ cient should be

increasing with the variance of residual wages. After 10 years of experience, we expect the

highest growth in the AFQT coe¢ cient in occupations with the highest growth rates. In

contrast, we expect no growth in the AFQT coe¢ cient in occupations with zero growth rates

in the variance of residual wages. I will report results from running separate regressions for

each of these groups, and then proceed to report results using di¤erent speci�cations. Notice

that all models include occupation �xed e¤ects, which among other things, also capture level

di¤erences in the residual variance upon the worker�s hire.

Table 5 reports the results from estimating equation (12) for the three separate groups

18About 80 percent of the workers in my sample change their initial 2-digit occupation after 10 years inthe workforce. However, it may be inappropriate to consider only workers who never switch occupations.For example, an entry level sales position in a �rm might be the starting point for many possible careerpaths in the company, and thus we would not want to restrict the sample to workers who stayed in sales.The real interest of the paper is in the entire menu of career paths that are associated with di¤erent initialoccupational assignments.

18

I describe above. The results on the AFQT variable (at T = 0) suggest that while it

is very small and statistically insigni�cant in occupations where the growth in variance is

high, it increases monotonically as the growth in the variance decreases. The coe¢ cient in

occupations with intermediate growth rates is 0.023 (0.009) and grows to 0.050 (0.024) in

occupations with no growth in variance (Columns 2 and 3, respectively). On the other hand,

the coe¢ cient on the AFQT-experience interaction is monotonically increasing with the

growth in the residual variance. It is as high as 0.071 (0.012) in occupations with the highest

growth rates and is not signi�cantly di¤erent from zero in occupations with low growth in the

residual variance. The coe¢ cient on the interaction term for occupations with intermediate

growth rates is 0.038 (0.01). I call occupations with high growth in the variance of residual

wages and high growth in the AFQT coe¢ cient "high learning" occupations. Occupations

with low growth rates in both the variance and the AFQT coe¢ cient are "low learning"

occupations, and those in between are "intermediate learning" occupations.

Of course, occupations can be low learning for two di¤erent reasons: First, it might be

the case that the market has already learned the worker�s AFQT at the start of the career;

Second, it might mean that the market has not learned the worker�s AFQT even after 10

years into the career. The large coe¢ cient on AFQT at T = 0 in column 3 and the large level

of residual variance for workers with 0-5 years of experience in these occupations (Column

1, Table 3) support the �rst scenario. It indicates that the worker�s AFQT in low learning

occupations is revealed immediately, or soon after their hire. This is plausible if employers

in these occupations have access to richer information about workers�skills upon their hire

(e.g. more information in q) which enables them to correctly estimate the worker�s AFQT

score.

19

It is also interesting to look at the coe¢ cient on the black dummy in the regressions

across the di¤erent groups. Table 5 shows that the coe¢ cient on the black dummy in high

learning occupations is negative and large, but is positive for workers in low learning occu-

pations (Column 3).19 A �nal point that deserves attention is the behavior of the education

variable over time. According to proposition 1, if education is used by the employer to as-

sess the average ability of the worker upon hiring and employers learn, then the coe¢ cient

on education should be nonincreasing over time. In fact, the coe¢ cient on the education-

experience interaction in high and intermediate learning occupations is negative (but small

and statistically insigni�cant) while the coe¢ cient on the education-experience interaction

in low learning occupations is positive (but small and statistically insigni�cant).

Table 6 reports the results from a regression where the variables are directly interacted

with dummy variables based on the growth in variance (with similar classi�cations as before).

The results in Column 1 are based on estimating equation (12) with the addition of two

interaction terms of the AFQT-experience variable with high and intermediate growth in

variance dummies (the omitted group is occupations where the growth is below 0.010).

Column 2 adds interactions of the AFQT variable (at T = 0) with the variance dummy

variables and Column 3 is based on a fully interacted model, where both education and AFQT

are interacted with the growth in the variance dummy variables. The results are consistent

with di¤erential learning. Notice that in all models, the coe¢ cient on the interaction of

AFQT-experience with high variance growth is large and statistically signi�cant, while the

coe¢ cient on the interaction with intermediate growth in variance is small (0.038), in column

19Arcidiacono, Bayer, and Hizmo (2008) report similar results on the black dummy for high school grad-uates and college graduates and conclude that black college graduates do not face statistical discriminationin the labor market.

20

1, and decreases and becomes statistically insigni�cant in column 3. The coe¢ cients on the

interaction terms with high and intermediate growth in variance are statistically di¤erent

from each other in all models. Finally, although the patterns on the AFQT at T = 0 are not

estimated precisely, the coe¢ cients on AFQT when interacted with the high-variance growth

dummy are smaller that the coe¢ cients on the AFQT when interacted with the intermediate

variance growth dummy.

In Table 7, instead of classifying occupations into groups based on the growth in the

residual variance, I interact the variables in the regression with the raw growth measure.

To simplify the comparison, column 1 repeats the results of estimating equation (12) when

learning is assumed to be similar across occupations. Column 2 adds an interaction term

between the AFQT-experience interaction and the growth in variance, and column 3 adds

an interaction of AFQT (at T = 0) with the growth in variance. Finally, column 4 adds

interactions of education (both at T = 0 and at T = 10) with the growth in variance. In

all models (column 2-4) the coe¢ cient on the AFQT-experience-variance growth variable is

large, positive, and statistically signi�cant, suggesting that there is a monotonic relationship

between the growth in variance and the growth of the AFQT coe¢ cient over time.

4.4 Do Employers Discriminate on the Basis of Race?

An important insight of AP is that we can use learning models to test for statistical discrim-

ination based on race. Although the focus of this paper is di¤erent, it is interesting to show

how the coe¢ cient on the race variable evolves over time in occupations with di¤erent pat-

terns of employer learning. The race variable can be treated as an s variable or a z variable.

21

AP show that if there is a negative correlation between race and skill, and employers use it

as a source of information about workers�skills, the coe¢ cient on the black variable should

be negative at T = 0. However, if employers initially statistically discriminate against black

workers but learn over time, we would expect the racial gap to shrink, that is the coe¢ cient

on race to rise over time (decrease in absolute value). If employers, however, do not discrim-

inate on the basis of race, the race variable can be treated as a z variable. In this case, if

race is negatively correlated with productivity and employers learn, the race gap will widen

as workers accumulate experience.

In the �rst and second column of Table 8 I use the full sample to reproduce AP�s results.

The �rst column includes interaction terms of education and the Black dummy with expe-

rience. The coe¢ cient on the Black dummy at T = 0 is -0.022 (0.011) and the coe¢ cient

on the black dummy when interacted with experience is -0.099 (0.012). Adding the AFQT-

experience interaction to the regression changes the coe¢ cients to -0.051 (0.012) at T = 0

and the coe¢ cient on the interaction term to -0.055 (0.014) (columns 2). The results are

similar to AP�s, and suggest that employers use race as a source of information upon the

worker�s hire but that they do not statistically discriminate based on race because the race

gap widens over time, rather than shrinking.

Does this conclusion hold across occupations with di¤erent learning rates? Columns (3)-

(5) report the results of a similar regression that is used in column (2) for high, intermediate

and low learning occupations. The results in column (3) and (4) seem to be in line with

the results of the full sample. For high learning occupations, including the Black-experience

interaction does not reduce the coe¢ cient on the Black dummy at T = 0. It is -0.097 (0.018),

and the racial gap widens slightly after 10 years. For intermediate learning occupations, the

22

coe¢ cient on the Black dummy is small -0.032 (0.016) at T = 0, and the race gap widens

over time to about -0.110. According to AP�s interpretation, although the Black coe¢ cient

at T = 0 in both groups of occupations is di¤erent, the fact that the racial gap widens over

time indicate that employers treat race as a z variable and that they do not statistically

discriminate Black workers.

The results in column (5) portray a di¤erent scenario. The coe¢ cient on the Black

dummy at T = 0 in low learning occupations is 0.175 (0.043) and declines over time to

0.041 (0.175-0.134).20 This result is puzzling because, on the one hand, it suggests that

employers in these occupations use race as an "s" variable (similar to education) and update

their beliefs over time. On the other hand, the low growth in the residual variance and in

the AFQT coe¢ cient indicate that employers in these occupations learn about the worker�s

AFQT at the beginning of the career. This suggests that the change in the race coe¢ cient

over time is driven by something other than employer learning about the worker�s AFQT,

and casts doubts about the interpretation of the change in the race coe¢ cient in high learning

occupations.

5 Initial Occupational Sorting

In the previous analysis I did not model initial occupational sorting. In reality, of course,

there are a number of factors that might in�uence a worker�s initial occupational assignment,

including risk aversion, credit constraints, private information about ability, and others.

Generally speaking, any occupational assignment rule that would disproportionately place

20These results are consistent with the results of Arcidiacono et. al (2008) who �nd that black workersare discriminated against in the high-school labor market but not in the college graduates labor market.

23

workers with high growth in the AFQT coe¢ cient into occupations with high growth in the

variance of residual wages for reasons that are unrelated to general learning would constitute

a selection bias that could be an alternative explanation of the paper�s main results.

It is hard to think of a selection-into-occupations mechanism that could be a plausible

alternative explanation of the results. Consider, however, an example of such an assignment

rule: If occupations with high growth in the variance of residual wages o¤er dispropor-

tionately more on-the-job training, and if training is correlated with AFQT score, then an

assignment rule that places the least trained workers into occupations with the most on-the-

job training would be consistent with the results of the paper. Although this is a possible

mechanism there is no evidence that indicate large di¤erences in training across occupations.

Certainly a model in which workers with di¤erent expected mean ability select into di¤erent

starting occupations, such as the one I present in this section, does not pose a problem, since

it just a¤ects the level of wages in each occupation, not the growth in the residual, which is

the variable of interest.

In the remainder of this section, I examine the e¤ects of one factor that might in�uence a

worker�s initial occupational assignment, namely the e¤ects of the market�s initial, publicly-

held priors about the worker�s ability. Consider the following production function:

yjt = exp(dj + cjh+Gj(t) + �jt) , j = 1; :::; J (13)

where dj and cj are constants, common knowledge among all labor market participants, and

where cJ > cJ�1 > ::: > c1 > 0 and d1 > d2 > ::: > dJ . Gj(t), as before, is the experience

pro�le of productivity, which is assumed to be independent from s; q; z;or �. The error

24

term �jt represents a productivity shock drawn from a normal distribution with mean 0 and

variance �2.

The signal that employers extract now is: bjt = xjt � E(hjs; q) = �� + " +vjtcj. In this

case the precision of the signal of productivity depends on the sensitivity of the output to

the worker�s productive ability. In particular, the higher cj is the higher is the signal-to-

noise ratio. Thus, jobs characterized with higher cj are more informative about the worker�s

ability compared to jobs with low cj.

As before, de�ne an occupation as the entire expected career track associated with choos-

ing a given initial occupation, and assume that each of the chosen career tracks are associ-

ated with a �xed learning parameter for the worker�s entire career.21 Workers�objective is

to choose an initial occupation that maximizes expected present value of lifetime earnings,

given the information they have when they enter the labor force.

Denote with hj the skill level that solves dj + cjhj = dj+1 + cj+1hj. That is, hj is the

skill level at which a worker is equally productive at job j and j + 1, and assume that

hj+1 > hj, 8 j. Employers (and workers) form initial priors about the worker�s skill based on

the information they have from s and q. Given these priors, a worker chooses (or is assigned

to) an initial job that maximizes his expected lifetime output. Proposition 2 formalizes the

assignment rule, with its proof provided in the appendix.

Proposition 2 Suppose that at the beginning of a worker�s career, employers in J occupa-

tions form expectations about the worker�s skill level based on available information. Under

the assumptions of the model, initial occupational assignment and the logarithm of wages are

21Since workers� objective is to maximize lifetime income, di¤erential learning might well involve re-assignment of workers to occupations other that the initial occupation.

25

given by (i)-(iii):

(i) If he < h1;then worker i is assigned to occupation 1 and is paid w1t.

(ii) If hj � he < hj+1, then worker i is assigned to occupation j + 1 and is paid wj+1t.

(iii) If he > hJ�1, then worker i is assigned to occupation J and is paid wJt.

This assignment rule implies that workers sort into initial occupations monotonically,

based on the market�s prior on their productivity, with higher-expected-ability workers choos-

ing jobs where output depends more on individual ability. Thus, in relation to the learning

model, workers with higher priors enter occupations in which output depends more on ability

and are more informative about the worker�s ability at the start of the career. One other

implication of proposition 2 is that the starting wage level is higher for workers who enter the

market with higher initial priors. In fact, the mean starting wage level for workers who enter

low learning occupations is $9.74/hour, while the mean starting wages for the other groups

of occupations are $7.25/hour and $6.6/hour for occupations with intermediate and high

learning, respectively. This is consistent with the idea that workers with low expected abil-

ity sort into occupations where employers have a lot to learn about the workers, while high

expected ability workers sort into occupations where ability is revealed (and compensated)

soon after their hire.

26

6 Robustness Check

6.1 Wage Dispersion Using NLSY

Because of data limitations in the NLSY, the preceding analysis calculated the growth in

wage dispersion using CPS data for workers with 0-5 years of experience and workers with 6-

10 years of experience within an occupation. The theoretical model, however, implies that we

need to track the growth in wage dispersion for workers who start their career in occupation

j over their experience pro�le. For some occupations, where occupational mobility is low,

these two measures are expected to be similar (e.g. nurses, or engineers), and for other

occupations the two measures might be very di¤erent.

To check the sensitivity of my results to the inclusion of occupations for which these

measures are di¤erent, I calculate two measures of growth-in-dispersion using NLSY data.

The �rst compares the wage dispersion of di¤erent experience cohorts within an occupation,

and the other compares the wage dispersion of workers with 0-5 years of experience who

start their career at the same occupation, with their own wage dispersion when they have

6-10 years of experience (regardless of the current occupation in which they are in). I, then,

restrict my sample to occupations for which the two measure are not di¤erent from each

other. There are 19 such occupations (out of 45), covering 49 percent of my original sample

size. Examples of such occupations are engineers, administrative occupations, construction

trades, health technicians, and farm workers.

Table 9 repeats the results of Table 5 with the restricted sample. I group occupations into

three categories, similar to the ones I describe earlier using the growth measure calculated

from the CPS. I do not use the growth measures which I calculated from the NLSY because

27

they are estimated with less precision. However, the ranking of occupations based on high,

medium, or low growth in wage dispersion remains unchanged, regardless which data set is

used to calculate them.

As can be seen, the results of this sub-sample are stronger, indicating that the results I

report for the whole sample are not driven by occupations for which the wage dispersion of

workers over time di¤ers from the dispersion of di¤erent cohorts within an occupation. The

positive monotonic relationship between the AFQT-experience coe¢ cient with the growth-

in-variance measure appears to be stronger than what I initially report in Table 5, and so is

the negative relationship between the AFQT coe¢ cient at T = 0 and the growth measure.

The other coe¢ cients in Table 8 appear to be similar to the previous results.

6.2 Employer Learning Across Educational Groups

The model in section 2 links the skill required by a job to the learning rate of employers.

Can these di¤erences be captured simply by estimating equation (12) by education group?

If so, we should expect a monotonic relationship between the AFQT coe¢ cient at T = 0

across the educational groups, and a consistent relationship between the coe¢ cient on the

AFQT-experience interaction and educational attainment. However, if occupations with

similar educational requirements have di¤erent learning rates, the evidence from estimating

equation (12) by educational group will not be conclusive.

Table 10 reports the results of estimating the wage equation for four separate education

groups: high school dropouts (8-11 years of schooling), high school graduates (12 years of

schooling), some college (13-15 years of schooling), and college graduates (16-20 years of

28

schooling). Interesting patterns emerge: the coe¢ cient on AFQT at T = 0 is essentially

zero for dropouts and workers with some college, but is positive and statistically signi�cant

for high school and college graduates. The coe¢ cient on the AFQT-experience interaction

term, on the other hand, is similar for workers with high school education or above (around

0.040), but is bigger for dropouts (0.070). As can be seen, there is no consistent monotonic

relationship between education and the coe¢ cient on the AFQT variable at T = 0 or at

T = 10. Therefore, it is hard to draw a clear prediction about di¤erential employer learning

based on educational attainment.

However, there is some evidence that the average years of schooling increases as the

growth in the variance of wage residuals decreases. The mean educational attainment within

occupations grouped based on the growth in wage dispersion is 13.31 in occupations with

the highest growth in wage dispersion, 13.77 in occupations with intermediate growth, and

16.33 years of schooling in occupations with low growth in wage dispersion. The fact that the

sorting of workers across jobs based on educational attainment does not capture di¤erences

in employer learning highlights the heterogeneity in the informational content of jobs which

require similar educational credentials.

The remaining results in table 10 are also of interest. The model in section 2 predicts

that if education is used as signal, its weight in a wage equation should be nonincreasing

over time. The coe¢ cients on the education-experience interaction for dropouts and college

graduates are consistent with this prediction but high school graduates and those with some

college have large positive coe¢ cients on the education-experience interactions (about 0.090).

According to AP�s interpretation, this suggests that employers do not use education to infer

these workers�abilities upon their hire. Instead, the large and positive coe¢ cient on the

29

education-experience interaction might re�ect the fact that workers in these educational

groups receive signi�cant amount of on-the-job training.

7 Conclusion

In this paper I provide evidence that employer learning about the worker�s time-invariant

productivity varies across occupations. In the model, employers initially do not observe the

worker�s ability, but over time they use output produced by the worker as a signal to update

their initial beliefs. Di¤erently from other studies, I assume that the precision of the output

signal di¤ers across occupations.

Employer learning has two implications which I utilize in the analysis. First, employer

learning generates the well known fact that the wage dispersion of a cohort of workers

increases as they gain experience. In particular, employers in occupations with large growth

in the variance of residual wages learn more about the worker�s productivity than employers

in occupations with low growth in the variance of residual wages. Second, if employers

initially do not observe a correlate of ability that is available in the data and employers

learn, then the coe¢ cient of this ability measure in a wage regression will be nondecreasing

with experience. As in previous studies of employer learning, the ability correlate I use is

the AFQT score. If employers learn at di¤erent rates across jobs, the growth in the AFQT

coe¢ cient will be larger in occupations with higher learning rates.

Empirically, I estimate an occupation-speci�c growth rate in the variance of residual

wages for workers with 0-5 and 6-10 years of experience using ORG CPS data for the 1984-

2000 period, and match these values with data from the NLSY79. Consistent with the model,

30

the results show that occupations with high growth in the variance of residual wages over

the �rst 10 years are also the occupations with high growth in the AFQT coe¢ cient. This

supports the notion that initial occupational assignments may have important implications

for learning. On the other hand, I �nd that the AFQT coe¢ cient in occupations with

low growth in the variance of residual wages is large and does not grow over time. This,

combined with the observation that these occupations have a high initial level of residual

variance, supports the idea that the market in these occupations has already learned the

worker�s AFQT at the start of the career.

Moreover, my results in regard to statistical discrimination on the basis of race suggest

that employers in occupations with high and intermediate learning rates do not statistically

discriminate at the time of hire. Interestingly, the coe¢ cient on the Black dummy in low

learning occupations is positive and large at the time of hire and decreases over time. These

di¤erent patterns suggest that the informational role of race di¤ers across occupations.

There are several important implications of the results reported that need to be explored

in future research. The most notable, perhaps, is to model how di¤erential employer learning

a¤ects occupational mobility during the worker�s career. Moreover, it raises the importance

of private information that workers might have about their own ability, and how it might

a¤ect their educational choices and occupational sorting. Another important issue relates

to the spot market contracting assumed throughout this paper (and in all other learning

models of which I am aware). What are the optimal contracts �rms should o¤er in the

presence of di¤erential learning? And what are the implications for career mobility in such

environments? These are some of the research veins to pursue in future research.

31

Appendix

Proposition 2: Suppose that at the beginning of a worker�s career, employers in Joccupations form expectations about the worker�s skill level based on available information.Under the assumptions of the model, initial occupational assignment and the logarithm ofwages are given by (i)-(iii):(i) If he < h1;then worker i is assigned to occupation 1 and is paid w1t.(ii) If hj � he < hj+1, then worker i is assigned to occupation j + 1 and is paid wj+1t.(iii) If he > hJ�1, then worker i is assigned to occupation J and is paid wJt.

Proof: hj is the skill level that solves dj + cjhj = dj+1 + cj+1hj. Assume also thathj+1 > hj, 8 j. A Worker objective at t = 0 is to maximize expected lifetime income Vjt

=TPt=0

�twjt, given the initial expected skill he. Consider a worker who enters the labor market

with hj � he < hj+1. The expected lifetime income if the worker chooses job j + 1 is biggeror equal than the expected lifetime income if job j is chosen. To see this, rewrite V for bothtypes of jobs as the summation of the expected output in period t = 0 and the expectedoutput in all subsequent periods:

wj +TXt=1

�twjt � wj+1 +TXt=1

�twj+1t (14)

Notice that if hj = he the initial wage will be equal in occupation j and j + 1, andso will be the stream of income in all subsequent periods (from the worker�s stand pointat t = 0). This choice is also strictly better than choosing any other job with j

0< j.

For hj < he < hj+1, choosing job j + 1 will generate a higher initial wage for the worker.Since there are no reassignments, this choice will also maximize the stream of income in allsubsequent periods, given the initial prior he. For this worker, the choice of job j+1 is strictlybetter than choosing any other job j

0> j + 1. In other words, in this simple framework,

because reallocation is not possible and the production function is linear, the expected valueof lifetime output is ex-ante maximized when the �rst period output is maximized. That is,although there might be ex post regret for some workers (because they update their initialprior), choosing an initial occupation that maximizes output is the optimal policy from theworker�s stand point at t = 0.

32

References[1] Altonji, J.G. (2005): "Employer Learning, Statistical Discrimination and Occupational

Attainment," American Economics Review, Papers and Proceedings, 96:112-117.

[2] Altonji, J.G. and C.R. Pierret (2001): "Employer Learning and Statistical Discrimina-tion," Quarterly Journal of Economics, 113:79-119.

[3] Altonji, J.G. and C.R. Pierret (1997): "Employer Learning and Statistical Discrimina-tion," National Bureau of Economic Research Working Paper No. 6279.

[4] Antonovics, K. and L. Golan (2007): "Experimentation and Job Choice," WorkingPaper.

[5] Arcidiacono P., Bayer P., and A. Hizmo (2008): "Beyond Signaling and Human Capi-tal: Education and the Revelation of Ability," National Bureau of Economic ResearchWorking Paper No. 13951.

[6] Autor, D., Katz, L.F., and A.B. Krueger (1998): "Computing Inequality: Have Com-puters Changed the Labor Market?," Quarterly Journal of Economics, 113:1169-1213.

[7] Card, D. and J.E. DiNardo (2002): "Skill-Biased Technological Change and RisingWageInequality: Some Problems and Puzzles," Journal of Labor Economics, 20: 733-783.

[8] Farber, H.S. and R. Gibbons (1996): "Learning and Wage Dynamics," Quarterly Jour-nal of Economics, 111:1007-1047.

[9] Gibbons R. and L.F. Katz (1991): "Layo¤s and Lemons," Journal of Labor Economics,9:351-380.

[10] Gibbons R., Katz L.F., Lemieux T., and D. Parent (2005): "Comparative Advantage,Learning, and Sectoral Wage Determination," Journal of Labor Economics, 23:681-723.

[11] Gibbons R. and M. Waldman (1999): "A Theory of Wage and Promotion Dynamicsinside Firms," Quarterly Journal of Economics, 114:1321-1358.

[12] Gibbons R. and M. Waldman (2006): "Enriching a Theory of Wage and PromotionDynamics inside Firms," Journal of Labor Economics, 24:59-107.

[13] Jaeger, D.A. (1997): "Reconciling the New Census Bureau Education Questions: Rec-ommendations for Researchers," Journal of Business and Economic Statistics, 15:300-309.

[14] Jaeger, D.A. (2003): "Estimating the Returns to Education Using the Newest CurrentPopulation Survey Education Questions," Economic Letters, 78:385-394.

[15] Jovanovic, B. (1979a): "Job Matching and the Theory of Turnover," Journal of PoliticalEconomy, 87:972-990.

[16] kahn, L.B. (2007): "Asymmetric Information between Employers," Harvard University,mimeo.

33

[17] katz, L.F. and K.M. Murphy (1992): "Changes in Relative Wages, 1963-1987: Supplyand Demand Factors," Quarterly Journal of Economics, 107:35-78.

[18] Lange, F. (2007): "The Speed of Employer Learning," Journal of Labor Economics,25:1-35.

[19] Lemieux, T. (2006): "Increasing Residual Wage Inequality: Composition E¤ects, NoisyData, or Rising Demand for Skill?," American Economic Review, 96: 461-498.

[20] Miller, A. R. (1984): "Job Matching and Occupational Choice," Journal of PoliticalEconomy, 92: 1086-1120.

[21] Mincer, J. (1958): "Investment in Humans Capital and Personal Income Distribution,"Journal of Political Economy, 66:281-301.

[22] Neal, D. and W.R. Johnson (1996): "The Role of Premarket Factors in Black-WhiteWage Di¤erences," Journal of Political Economy, 104:869-895.

[23] Neal, D. and S. Rosen (1998): "Theories of the Distribution of Labor Earnings," Na-tional Bureau of Economic Research Working Paper No. 6378.

[24] Topel, R. andM.Ward (1992): "Job Mobility and the Careers of Young Men,"QuarterlyJournal of Economics, 107:441-479.

[25] Schönberg, U. (2007): "Testing for Asymmetric Employer Learning," Journal of LaborEconomics, 25:651-692.

[26] Sicherman, N. and O. Galor (1990): "A Theory of Career Mobility," Journal of PoliticalEconomy, 98:169-192.

[27] Spence, A.M. (1973): "Job Market Signaling," Quarterly Journal of Economics, 87:355-379.

[28] Von Wachter, T. and S. Bender (2006): "In the Right Place at the Wrong Time -The Role of Firms and Luck in Young Workers�Careers," American Economic Review,96:1679-1705.

34

Table 1: Descriptive Statistics, NLSY79 sample, 1979-2000Variable Mean Standard Minimum Maximum

DeviationReal hourly wage 9.13 6.32 3 99.38Log of real hourly wage 2.06 0.53 1.09 4.60Education 13.77 2.43 8 20Black dummy 0.27 0.44 0 1Actual experience 6.96 5.04 0 23.70Standardized AFQT score 0.069 0.99 -2.99 2.41The sample includes 22,409 observations from 1962 individuals.

Table 2: Descriptive Statistics, ORG CPS sample for workers with 0-10 yearsof experience, 1984-2000

A. Education - % share B. Experience - % shareYear Dropouts High-school Some College 0-5 6-10 Sample

graduates college graduates years years size1984 9.07 39.90 24.35 26.68 47.35 52.65 27,5741985 9.23 39.34 24.95 26.48 47.14 52.86 27,7771986 8.77 39.43 24.58 27.22 45.92 54.08 26,8731987 8.66 38.76 24.89 27.68 46.22 53.78 25,6101988 8.73 38.82 24.42 28.03 45.68 54.32 23,8641989 8.84 38.17 25.00 27.99 46.02 53.98 23,7071990 8.87 37.83 25.22 28.07 46.66 53.34 23,8731991 8.41 36.73 26.04 28.82 45.83 54.17 22,1931992 6.81 37.68 27.46 28.05 46.13 53.87 21,1471993 6.61 36.61 28.45 28.33 45.69 54.31 20,3121994 6.49 35.47 28.80 29.24 45.92 54.08 18,4711995 6.97 34.53 28.53 29.97 45.72 54.28 18,1971996 7.05 34.55 28.40 30.00 46.27 53.73 16,0971997 7.66 34.96 27.78 29.60 45.60 54.40 16,3241998 7.22 33.22 29.50 30.06 48.19 51.81 16,4831999 7.40 34.12 28.54 29.94 27.42 52.58 16,4182000 8.21 33.68 28.03 30.08 47.95 52.05 16,6101984-2000 8.08 37.10 26.44 28.38 46.43 53.57 361,530

35

Table3:Rankingof19802-digitoccupationcodesbythegrowthinthevarianceofwageresiduals,ORGCPSsample-1984-2000

Occupation

ResidualVarianceResidualVarianceDi¤erence

MeanYears

0-5experience

6-10experience

ofEducation

Healthdiagnosingoccupations

0.359

0.477

0.118*

17.85

Personalserviceoccupations

0.150

0.260

0.110*

13.23

Salesworkers,retailandpersonalservices

0.133

0.213

0.080*

13.15

Salesrepresentatives,Finance,andBusinessService

0.242

0.307

0.065*

15.20

Foodserviceoccupations

0.131

0.191

0.060*

12.47

Secretaries,stenographers,andtypists

0.124

0.183

0.059

13.97

Forestryand�shingoccupations

0.227

0.278

0.051

12.17

Otherhandlers,equipmentcleanersandlaborers

0.102

0.149

0.047*

12.07

Constructionlaborer

0.129

0.170

0.041*

12.07

Freight,stockandmaterialhandlers

0.108

0.149

0.041*

12.25

SupervisorsandProprietors,salesoccupations

0.139

0.178

0.039*

13.91

Farm

workersandrelatedoccupations

0.139

0.177

0.038*

12.27

Otherexecutive,Administrators,andmanagers

0.183

0.217

0.034*

15.00

Healthtechnologistsandtechnicians

0.140

0.173

0.034

14.35

Constructiontrades

0.130

0.163

0.034*

12.32

Motorvehicleoperators

0.132

0.164

0.032*

12.47

Supervisors-Administrativesupport

0.108

0.139

0.031

14.39

Cleaningandbuildingserviceoccupations

0.114

0.144

0.030*

12.25

OtheradministrativeSupportOccupations

0.111

0.140

0.028*

13.54

Mathematicalandcomputerscientists

0.139

0.164

0.026**

16.01

Salesrepresentatives,commodities,exceptretail

0.179

0.204

0.025*

14.86

Protectiveserviceoccupations

0.149

0.172

0.023*

13.53

*,**indicatesthatthechangeinthevarianceissigni�cantlydi¤erentfrom

zeroatthe95-percentcon�dencelevel

andthe90-percentcon�dencelevel,respectively.

36

Table3continue:Rankingof19802-digitoccupationcodesbythegrowthinthevarianceofwageresiduals,ORGCPS

sample,1984-2000.

Occupation

Residualvariance

ResidualVarianceDi¤erence

Mean

0-5experience

6-10experience

education

Fabricators,assemblersandinspectors,andsamplers

0.113

0.136

0.023*

12.34

Mechanicsandrepairers

0.128

0.149

0.022*

12.65

Machineoperatorsandtenders,exceptprecision

0.104

0.125

0.021*

12.19

Technicians,excepthealthengineering,andscience

0.195

0.216

0.021**

15.34

Healthserviceoccupations

0.092

0.113

0.021

13.04

LawyersandJudges

0.208

0.227

0.019

17.81

transportationoccupationsandmaterialmoving

0.129

0.149

0.019**

12.21

Professionalspecialityoccupations

0.237

0.256

0.019*

15.64

Mailandmessagedistributing

0.134

0.153

0.019

13.09

Managementrelatedoccupations

0.140

0.156

0.016*

15.79

Computerequipmentoperators

0.135

0.150

0.015

13.91

Financialrecords,processingoccupations

0.131

0.145

0.015

14.03

Farm

operatorsandmanagers

0.194

0.208

0.014

13.54

Otherprecisionproductionoccupations

0.121

0.132

0.011*

12.71

Engineeringandsciencetechnicians

0.132

0.139

0.007

13.94

Administratorsando¢cials,publicadministration

0.148

0.147

-0.001

16.02

Teachers,exceptcollegeanduniversity

0.174

0.172

-0.002

16.22

Engineers

0.118

0.114

-0.004

16.13

Privatehouseholdserviceoccupations

0.337

0.336

-0.001

12.34

Healthassessmentandtreatingoccupations

0.183

0.156

-0.027

15.83

Naturalscientists

0.223

0.176

-0.047

16.72

Teachers,collegeanduniversity

0.330

0.256

-0.074*

17.17

Salesrelatedoccupations

0.262

0.172

-0.090

14.02

*,**indicatesthatthechangeinthevarianceissigni�cantlydi¤erentfrom

zeroatthe95-percentcon�dencelevel

andthe90-percentcon�dencelevel,respectively.

37

Table 4: The e¤ect of standardized AFQT and schooling on wagesModel (1) (2) (3) (4)Education 0.0569� 0.0643� 0.0610� 0.0685�

(0.008) (0.008) (0.005) (0.005)Black -0.1474� -0.1201� -0.1265� -0.1197�

(0.020) (0.021) (0.010) (0.010)Stan AFQT 0.0341� 0.0023 0.0451� 0.0120

(0.004) (0.007) (0.004) (0.006)Education*exp/10 0.0072 -0.0069 0.0026 -0.0074

(0.007) (0.008) (0.005) (0.006)Stan AFQT*exp/10 0.0722� 0.0490�

(0.013) (0.007)Sample period 1979- 1979- 1979- 1979-

1992 1992 2000 2000R2 0.38 0.38 0.40 0.40N 15,714 15,714 22,409 22,409* indicates signi�cance at the 5% level. The dependent variable is log real hourlywage. The experience measure is actual experience and is modeled as a cubicpolynomial. All models control for the two-digit �rst occupations, urban residence,year e¤ects, and a cubic time trend, with base year 1992, interacted with educationand the black dummy. All standard errors are Huber-White standard errors.

38

Table 5: The e¤ect of standardized AFQT and schooling on wagesby growth rates in the variance of wage residuals

Model (1) (2) (3)High growth in Intermediate growth in Low growth inwage dispersion wage dispersion wage dispersion

Stan AFQT -0.0067 0.0231* 0.050*(0.010) (0.009) (0.024)

Stan AFQT*exp/10 0.0709* 0.0376* 0.0057(0.012) (0.010) (0.024)

Education 0.0692* 0.0743* 0.0270(0.008) (0.007) (0.018)

Education*exp/10 -0.0046 -0.0125 0.0247(0.010) (0.008) (0.020)

Black -0.1483* -0.1190* 0.080(0.016) (0.014) (0.036)

R2 0.39 0.41 0.39N 9,354 11,301 1,754* indicates signi�cance at the 5% level. The dependent variable is log real hourly wage.The experience measure is actual experience and is modeled as a cubic polynomialAll models control for the two-digit �rst occupations, urban residence, year e¤ects, and acubic time trend, with base year 1992, interacted with education and the black dummy.All standard errors are Huber-White standard errors.

39

Table 6: The e¤ect of standardized AFQT and schoolingon wages including interactions with residual variance.

Model (1) (2) (3)

Stan AFQT 0.0111 0.0025 0.0109(0.006) (0.018) (0.020)

Stan AFQT*�V arlow 0.0202 0.0096(0.018) (0.021)

Stan AFQT*�V arhigh -0.0038 -0.0107(0.019) (0.021)

Stan AFQT*exp/10 0.0092 0.0157 0.0050(0.014) (0.019) (0.020)

Stan AFQT*exp/10*�V arlow 0.0379* 0.0211 0.0333(0.014) (0.020) (0.022)

Stan AFQT*exp/10*�V arhigh 0.0521* 0.0573* 0.0686*(0.014) (0.020) (0.022)

R2 0.43 0.43 0.43N 22,409 22,409 22,409* indicates signi�cance at the 5% level. The dependent variable is log real hourly wage.The experience measure is actual experience and is modeled as a cubic polynomialAll models control for the two-digit �rst occupations, urban residence, year e¤ects, and acubic time trend, with base year 1992, interacted with education and the black dummy.All standard errors are Huber-White standard errors.

40

Table 7: The e¤ect of standardized AFQT and schoolingon wages including linear interactions with residual variance.

Model (1) (2) (3) (4)

Stan AFQT 0.0120 0.0109 0.0252* 0.0272*(0.006) (0.006) (0.010) (0.010)

Stan AFQT*�V ar -0.3958 -0.4580(0.226) (0.244)

Stan AFQT*exp/10 0.0490* 0.0253* 0.0134 0.0149(0.007) (0.009) (0.011) (0.011)

AFQT*exp/10*�V ar 0.7285* 1.0662* 1.0186*(0.168) (0.264) (0.284)

Education 0.0685* 0.0682* 0.0680* 0.0636*(0.005) (0.005) (0.005) (0.006)

Education*�V ar 0.1302*(0.067)

Education*exp/10 -0.0074 -0.0075 -0.0071 -0.0076(0.006) (0.006) (0.006) (0.006)

Education*exp/10*�V ar 0.0064(0.018)

Black -0.1197* -0.1195* -0.1192* -0.1195*(0.010) (0.010) (0.010) (0.010)

R2 0.43 0.43 0.43 0.43N 22,409 22,409 22,409 22,409* indicates signi�cance at the 5% level. The dependent variable is log real hourly wage.The experience measure is actual experience and is modeled as a cubic polynomialAll models control for the two-digit �rst occupations, urban residence, year e¤ects, and acubic time trend, with base year 1992, interacted with education and the black dummy.All standard errors are Huber-White standard errors.

41

Table 8: The e¤ect of standardized AFQT and schooling on wages including a black-experienceinteraction.

Model (1) (2) (3) (4) (5)Full sample Full sample High growth in Intermediate growth in Low growth in

wage dispersion wage dispersion wage dispersion

Stan AFQT 0.0450* 0.0124* -0.0095 0.0246* 0.0684*

(0.004) (0.006) (0.010) (0.016) (0.024)Stan AFQT*exp/10 0.0483* 0.0748* 0.0355* -0.0167

(0.008) (0.013) (0.010) (0.025)Education 0.0652* 0.0712* 0.0725* 0.0774* 0.0292

(0.005) (0.005) (0.008) (0.007) (0.018)Education*exp/10 -0.0010 -0.0093 -0.0071 -0.0147 0.0220

(0.005) (0.005) (0.009) (0.008) (0.020)Black -0.0215* -0.0512* -0.0965* -0.0318* 0.1747*

(0.011) (0.012) (0.018) (0.016) (0.043)Black*exp/10 -0.0994* -0.0551* -0.0312 -0.0774* -0.1336*

(0.012) (0.014) (0.021) (0.019) (0.045)R2 0.42 0.43 0.39 0.40 0.45N 22,409 22,409 9,354 11,301 1,754

* indicates signi�cance at the 5% level. The dependent variable is log real hourly wage. The experience measureis actual experience and is modeled as a cubic polynomial. All models control for the two-digit �rst occupations,urban residence, year e¤ects, and a cubic time trend, with base year 1992, interacted with education.All standard errors are Huber-White standard errors.

42

Table 9: The e¤ect of standardized AFQT and schooling on wagesby growth rates in the residual variance of limited occupations

Model (1) (2) (3)High growth in Intermediate growth in Low growth inwage dispersion wage dispersion wage dispersion

Stan AFQT -0.0265 0.0352* 0.1928*

(0.016) (0.010) (0.058)Stan AFQT*exp/10 0.0956* 0.0235 -0.0651

(0.019) (0.013) (0.057)Education 0.0938* 0.0762* 0.0019

(0.012) (0.009) (0.038)Education*exp/10 -0.0449* -0.0144 0.0914*

(0.015) (0.010) (0.038)Black -0.1989* -0.1054* 0.2431*

(0.026) (0.018) (0.078)R2 0.37 0.41 0.50N 3,392 7,177 397* indicates signi�cance at the 5% level. The dependent variable is log real hourly wage.The experience measure is actual experience and is modeled as a cubic polynomialAll models control for the two-digit �rst occupations, urban residence, year e¤ects, and acubic time trend, with base year 1992, interacted with education and the black dummy.All standard errors are Huber-White standard errors.

43

Table 10: The e¤ect of standardized AFQT and schooling on wages,by education group.

Model (1) (2) (3) (4)Dropout High School Some College College

Education -0.0068 - -0.0366 0.0857�

(0.048) - (0.032) (0.015)Black -0.0627� -0.1410� -0.2052� -0.0383

(0.030) (0.016) (0.023) (0.020)Stan AFQT 0.0126 0.0218� 0.0001 0.0361�

(0.020) (0.009) (0.013) (0.012)Education*exp/10 -0.0168 0.0883� 0.0970� -0.0622�

(0.046) (0.008) (0.035) (0.019)Stan AFQT*exp/10 0.0710� 0.0420� 0.0322� 0.0422�

(0.029) (0.011) (0.016) (0.013)R2 0.28 0.28 0.33 0.34N 2,157 8,147 4,790 7,315* indicates signi�cance at the 5% level. The dependent variable is log real hourlywage. The experience measure is actual experience and is modeled as a cubicpolynomial. All models control for the two-digit �rst occupations, urban residence,year e¤ects, and a cubic time trend, with base year 1992, interacted with educationand the black dummy. All standard errors are Huber-White standard errors.

44