OCCUPATIONAL HUMAN CAPITAL: ITS ROLE AND IMPLICATION FOR EARNINGS
By
Kritkorn Nawakitphaitoon

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
Human Resources and Labor Relations
2012

ABSTRACT
OCCUPATIONAL HUMAN CAPITAL: ITS ROLE AND IMPLICATION FOR EARNINGS
By
Kritkorn Nawakitphaitoon

This dissertation research examines the role of occupational human capital in the
determination of workers’ wages and earnings losses following job displacement. In it, I explain
the important role of occupational skills transferability in the labor market. The first chapter
develops and estimates the measure of skills transferability applying two approaches. The first
approach is based on Shaw’s (1984) method, and the second one is based on Ormiston’s (2006)
method. The main difference between Shaw’s (1984) and Ormiston’s (2006) approaches is that
Shaw’s (1984) transferability matrix is a market-based approach, which reflects market as well
as technical conditions. In particular, Shaw (1984) estimates skills transferability by examining
an actual occupational change, arguing that there will be greater occupational mobility between
jobs that have greater rates of skills transferability. On the other hand, Ormiston’s (2006) skills
transferability is estimated based on the knowledge, skills, and abilities (KSAs) shared across
occupations. The accuracy of this estimate will be important for the subsequent two chapters, in
which I examine the determinants of workers’ wages over the lifecycle and earnings losses due
to job displacement.
The second chapter studies the effect of occupational human capital on the workers’
wages. Unlike previous studies that apply occupational tenure as a proxy for occupational
human capital, this chapter applies the concept of Shaw’s (1984) occupational human capital to
capture the transferability of occupational skills and estimates a new measure of occupational
human capital, so-called occupational investment. Using the National Longitudinal Survey of

Youth (NLSY) from 1979 to 2000, the key findings of this chapter suggest that occupational
skills from previous jobs can also affect workers’ wages at the current job and that occupational
investment is one of the important sources of wages supporting Shaw’s original work on wage
determination. Specifically, five years of (3-digit) occupational investment relative to the current
occupational tenure could lead to a wage increase of 7 to 16 percent. I also find that the general
labor market experience accounts for a large share of workers’ wages.
The third chapter investigates the role of occupational human capital in explaining
variations in earnings losses following job displacement. Unlike previous studies on job
displacement, this chapter uses a continuous measure of occupational skills transferability,
developed in Chapter 1, to measure the similarity between the pre- and post-displacement
occupations of reemployed displaced workers. Using the 2004, 2006, 2008, and 2010 Displaced
Worker Survey (DWS), the main finding is that post-displacement earnings losses are highly
correlated with the degree of similarity between pre- and post-displacement occupations.
Displaced workers who find jobs in occupations similar to their previous jobs suffer smaller
earnings losses than those who find less similar occupations. This relationship is non-linear in
that higher skills transferability reduces the earnings losses at a decreasing rate.
Overall, this dissertation examines the measurement and outcomes of occupational
human capital to support the idea that occupational skills are transferable across occupations and
that skills transferability can help to explain workers’ wages over the lifecycle and earnings
losses associated with job displacement.

ACKNOWLEDGMENTS

I would like to contribute my appreciation to my dissertation committee members, Dr.
Todd Elder, Dr. Peter Berg, and Dr. Mary Hamman, for reading the drafts and offering helpful
suggestions on my dissertation. Special thanks are especially due to Dr. Dale Belman, my
advisor and a chairperson of my dissertation committee, who mentors and helps me develop as a
scholar. I am also grateful to my colleagues both at the School of Human Resources and Labor
Relations and Department of Economics for their collaboration and support during my time at
Michigan State University. Finally, I am especially grateful to my parents and family in Thailand
for always giving me support and encouragement since I have come to study in the U.S.

iv

TABLE OF CONTENTS

LIST OF TABLES ....................................................................................................................... vii
LIST OF FIGURES ...................................................................................................................... ix
CHAPTER 1
THE ESTIMATION OF OCCUPATIONAL SKILLS TRANSFERABILITY .......................................... 1

Introduction ..................................................................................................................................1
The Estimation of Occupational Skills Transferability ...............................................................4
Shaw’s Skills Transferability Method .................................................................................4
Ormiston’s Skills Transferability Method ...........................................................................7
Comparison of Shaw’s and Ormistion’s Estimates .....................................................................9
Estimation Results .....................................................................................................................11
Conclusion .................................................................................................................................13
CHAPTER 2
OCCUPATIONAL HUMAN CAPITAL AND WAGES: THE ROLE OF SKILLS TRANSFERABILITY
ACROSS OCCUPATIONS ....................................................................................................................... 19

Introduction ................................................................................................................................19
Literature Review .......................................................................................................................21
Methodology .............................................................................................................................30
The Estimation of Occupational Human Capital Measure (Occupational Investment) ...30
Wage Determination ..........................................................................................................33
Data ............................................................................................................................................36
Estimation Results ......................................................................................................................38
Descriptive Statistics .........................................................................................................38
OLS Regression Results ...................................................................................................40
FE Regression Results ......................................................................................................43
Comparison of Shaw’s and Ormiston’s Skills Transferability .........................................46
Conclusion .................................................................................................................................48
CHAPTER 3
OCCUPATIONAL HUMAN CAPITAL AND EARNINGS LOSSES OF DISPLACED WORKERS:
DOES THE DEGREE OF SIMILARITY BETWEEN PRE- AND POST-DISPLACEMENT
OCCUPATIONS MATTER? ..................................................................................................................... 60

Introduction ................................................................................................................................60
Literature Review .......................................................................................................................62
Methodology ..............................................................................................................................67
Data ............................................................................................................................................70
Descriptive Statistics .................................................................................................................72
Estimation Results ....................................................................................................................74
OLS Regression Results ...................................................................................................74
Robustness Tests ...............................................................................................................78
v

Comparison of Shaw’s and Ormiston’s Skills Transferability .........................................81
Conclusion .................................................................................................................................83
APPENDICES ........................................................................................................................................... 93

Appendix A: Knowledge, Skill, and Ability Categories Across Occupations as Defined by
O*NET .................................................................................................................94
Appendix B: Description of Main Variables in the NLSY Dataset ...........................................96
Appendix C: The Estimations of Occupational Skills Intensity ...............................................104
Appendix D1: Descriptive Statistics of Displaced Workers (2004–2010). ..............................110
Appendix D2: Descriptive Statistics for CPS and DWS (2004–2010) .....................................112
REFERENCES ...........................................................................................................................113

vi

LIST OF TABLES

TABLE 1. Score Categories Using to Calculate Occupational Skills Transferability .................15
TABLE 2. The Probability Transition Matrix (1983, 1985, 1989, and 1991 March CPS) ...........15
TABLE 3. The Skills Transferability Matrix for 10 Selected Occupations (Shaw’s Approach). .16
TABLE 4. Summary of Occupational Skills Transferability ........................................................17
TABLE 5. The Skills Transferability Matrix for 10 Selected Occupations (Ormiston’s Approach)
........................................................................................................................................................18
TABLE 6. Descriptive Statistics of Key Variables (1979-2000) ..................................................53
TABLE 7. Correlations among Key Human Capital Measures ....................................................53
TABLE 8. OLS Estimates of the Wage Equation .........................................................................54
TABLE 9. Cumulative Returns to Human Capital Measures .......................................................55
TABLE 10. FE Estimates of the Wage Equation .........................................................................56
TABLE 11. Comparison of Shaw’s and Ormiston’s Skills Transferability (OLS) .......................57
TABLE 12. Comparison of Shaw’s and Ormiston’s Skills Transferability (FE) ..........................58
TABLE 13. Comparison of additional information from Shaw’s and Ormiston’s occupational
investment .....................................................................................................................................59
TABLE 14. Comparison of Cumulative Returns to Shaw’s and Ormiston’s Occupational
Investment .....................................................................................................................................59
TABLE 15. Percentage of Reemployed Workers Who Change Occupations ..............................87
TABLE 16. The Earnings Losses Regression................................................................................89

vii

TABLE 17. The Earnings Losses Regression Excluding Workers Who Were Displaced at Least
One Year before the Survey Date .................................................................................................90
TABLE 18. Heckman’s Two-Step Estimation of the Earnings Losses ........................................91
TABLE 19. Comparison of the Return to Shaw’s and Ormiston’s Skills Transferability ...........92

viii

LIST OF FIGURES

FIGURE 1. Occupational Changes among Employed Workers at the 1-, 2-, 3-digit Occupations
from 1979-2000. ............................................................................................................................52
FIGURE 2. Fractions of Displaced Workers by Level of Skills Transferability ..........................87
FIGURE 3. The Average Change in ln (Real Weekly Earnings) by Level of Occupational Skills
Transferability between Pre- and Post-Displacement Occupations. ..............................................88

ix

Chapter 1
THE ESTIMATION OF OCCUPATIONAL SKILLS TRANSFERABILITY

1. Introduction
The idea of occupational human capital has become particularly important in recent
literature to understanding wage growth over the lifecycle and the distribution of earnings across
the labor force (for example, Sullivan, 2006; Kambourov and Manovskii, 2009). With few
exceptions, important contributions from most previous research have focused on the role of
occupation-specific skills (i.e., skills that would be lost or not be applicable whenever workers
change occupations) and neglected the idea that skills could be transferred across occupations.
Ignoring this aspect of occupational human capital might introduce measurement error in
estimation.
Most recent studies on the relationship between occupation-specific skills and wages use
occupational tenure, the numbers of year spent in one particular occupation, as a proxy for
occupational human capital (for example, Sullivan, 2006; Zangelidis, 2007; Kambourov and
Manovskii, 2009). Presumably, the amount of occupational skills increases with tenure in the
occupation and would not applicable whenever workers change occupations. These studies
generally found the positive and significant effect of occupational tenure on individuals’ wages.
For example, using data from the PSID Retrospective Occupation-Industry Supplemental Data
Files (Retrospective files) from 1968-1993 period, Kambourov and Manovskii (2009) find 5
additional years of occupational tenure increase wages by 12 to 20 percent, all else being equal.
However, occupational tenure measure may be an inadequate proxy for occupational
human capital because whenever workers change occupations, the tenure measure is reset to zero,
which implies that all the skills from the previous occupations would be lost. Yet it is likely that
1

at least some portion of skills is transferred across occupations. For example, an engineer should
be able to transfer some engineering skills to a clerical job. Arguably, including actual years of
work experience in the wage equation will capture some of this transferability, but this measure
does not vary by occupation and therefore cannot capture variation in the transferability of skills
across occupations.
From a purely econometric standpoint, the occupational tenure measure assumes that all
the skills from the previous occupations fall into the error term of wage equation. The
occupational tenure variable will be endogenous if it is correlated with the omitted occupational
skills transferability from the previous job that also affects wages. For example, if workers who
have higher skills transferability from the previous job tend to stay in the current occupation
longer to accumulate more specialized occupational skills and receive higher wage returns than
their counterparts who have less skills transferability, then the estimate of return to occupational
tenure would be (upwardly) biased and inconsistent.
In this chapter, I address the concerns raised here by developing estimates of
occupational skills transferability, and in subsequent chapters I test the importance of this
measure for estimating wage determination and earnings losses following job displacement. To
measure occupational skills transferability, I apply two different measurement approaches. The
first approach is based on Shaw’s (1984) method, and the second one is based on Ormiston’s
(2006) method. Shaw’s (1984) skills transferability is a market-based approach, based on an
actual occupational change while Ormiston’s (2006) skills transferability is a pure technical
measure based on the knowledge, skills, and abilities (KSAs) shared across occupations.
To estimate Shaw’s (1984) skills transferability, I use the March Current Population
Survey (CPS), which is a large cross-sectional dataset of over 60,000 households. For

2

Ormiston’s (2006) skills transferability, I use the O*NET dataset to compute the ratio of shared
120 standardized knowledge, skill, and ability categories (KSAs) across occupations. The
accuracy of skills transferability will be important for subsequent chapters (Chapters 2 and 3), in
which I examine the determinants of workers’ wages over the lifecycle and the earnings losses of
displaced workers. Despite the differences in the estimation methods, the estimated correlation
between Shaw’s and Ormiston’s skills transferability is around 0.45, indicating a positive and
modestly strong correlation.
Better estimates of skills transferability could lead to a further understanding of wage
inequality (e.g., gender wage gap, or racial wage disparities). Specifically, in the gender wage gap
literature, the male wage premium might exist because of the lower relative occupational mobility
or higher skills transferability across occupations more commonly held by men, which in turn will
contribute to higher wages. The skills transferability measure could also lead to a better estimate
of earnings losses following job displacement. Displaced workers who find more similar
occupations with higher skills transferability might suffer smaller earnings losses than those who
find less similar occupations. From a policy perspective, the skills transferability measure can be
used to estimate the costs of human capital destruction involved in structural change in economy.
This chapter is structured as follows. In Section 2, I explain the estimations of occupational
skills transferability applying Shaw’s (1984) and Ormiston’s (2006) approaches. Section 3
discusses the strengths and weaknesses of each approach. Section 4 presents the estimation results,
and the last section concludes.

3

2. The Estimation of Occupational Skills Transferability
This section describes two estimation methods of occupational skills transferability. The
first method is based on Shaw’s (1984) approach, and the second one is based on Ormiston’s
(2006) approach. The main difference between Shaw’s and Ormiston’s approach is that Shaw’s
occupational skills transferability is a market-based approach. In particular, Shaw (1984)
estimates skills transferability by examining actual occupational mobility patterns and arguing
that there will be greater occupational mobility between jobs that have a higher rate of skills
transferability. On the other hand, Ormiston’s occupational skills transferability is a technical
approach based on a measure of commonality of knowledge skills and abilities (KSAs) across
occupations. Each approach has strengths and weaknesses which I will discuss below. Therefore,
these two methods of estimating measures of occupational skills transferability will be used and
tested in the following chapters.

2.1 Shaw’s Skills Transferability Method (Occupational Distance)
Shaw (1984) is the first one who recognizes the importance of transferability of
occupational human capital in establishing particular labor market outcomes. Particularly, Shaw
notes that the return to the stock of human capital from the former occupation is not completely
foregone upon an occupational switch. Instead, the return to one’s human capital stock depends
on the degree of transferability of skills from one occupation to the next. She hypothesizes skills
acquired in a particular occupation are portable across firms making them “occupation specific”,
rather than “firm specific”. Unlike the case of general human capital, individuals would be able
to transfer only a certain amount of their occupation-specific human capital stock in the event of

4

an occupational change, and that amount inherently depends on the degree of “transferability”
between the two occupations (Shaw, 1984).
Shaw (1984) estimates the transferability of skills using observed occupational changes.
She proposes the probability of a change from one occupation to another is increasing in the
transferability of skills between the two occupations. The greater the transferability is, the greater
the incentive to change occupation and vice versa. This implies that the measure of the
ij

ij

probability of occupational switch, P , is closely correlated with the transferability of skills, γ .
ij

However, the difficulty of directly using P as the approximation of skills transferability is that it
is likely to pick up the short-run causes and the determinants of mobility that are unrelated to
skills transferability. Instead, she proposes an alternative measurement of transferability which
ij

assesses occupational movement more closely than P . She calls this measure "occupational
ij

distance”, or D , and defines it as follows:

D ij 

K

| Pik  P jk |

i 1

K
ii  P jj )  ( P ij  P ji )  | P ik  P jk |
 (P

i 1
ik

jk

ij

ji

P is the probability of moving from occupation i to occupation k, and, P , P , and P
ii

jj

are defined equivalently. K is the set of all occupations. P and P are the probability of staying
within the same occupation i and j, respectively. This formula implies that the smaller the
ij

distance D , the closer the two occupations or the greater the skills transferability. Hence, for
occupations i and j to be close, or to have a high transferability of skills, they must have a close

5

equality of movement between occupation i and k, and between occupation j and k. In other
words, the main advantage of using the distance function rather than the probability of moving,
ij

P is that it is not only a function of the probability of moving between occupation i and j, but
also compares the probability of movement between occupation i and j and all other possible
occupations as well. For example, in order for a technician and engineer to be similar, the
probability of moving from a technician to any other occupations should be very similar to the
ij

probability of moving from an engineer to other occupations. Using D thus adds a greater
theoretical content to the measure of occupational skills transferability rather than the use of the
ij

probability of occupational switches, P (Shaw, 1981).
As we can see, if occupation i and j are completely different (unrelated) in terms of skills
ij

transferability, then D will be equal to 2 (i.e., all individuals stay in the same occupation such
ii

jj

that P + P is equal to 2, and the rest of the equation is equal to 0), and if they are exactly the
ij

same in terms of skills transferability, then D will be equal to 0 (i.e., all individuals change
their occupation such that all terms in the equation are equal to 0). Thus, we could scale the value
ij

of D to be bounded between 0 and 1 as follows:
ij

ij

d = 1- ( D / 2)
ij

Therefore, d is proposed to be the measure of proportion of skills that are transferable across
occupations ranging from 0 to 1.

6

ij

ij

ji

d is a symmetric matrix, d = d ; the direction of the occupational movement does not
matter. The skills transferability from occupation i to j is equal to that from occupation j to i, and
ij

d measures the average transferability between these two directions of movement.
In this chapter, Shaw’s occupational skills transferability is estimated at the 2-digit level
1

of the 1980 occupational classification system. To estimate it, I use the 1983, 1985, 1988,
and1991 March CPS. This dataset provides the current and “one year ago” occupations of
respondents, 18 years or older and not in school. The advantage of this dataset is that it is a very
large dataset drawn on a random sample of over 60,000 households, which is weighted to be
representative of the U.S. population. However, the short interval length of occupational
switches could also produce the very high retention rates, and it provides only the change in the
very short run (Shaw, 1984). Aggregating the 4 March CPS survey years, I have cross sectional
data for 638,380 individuals. I delete individuals who do not have occupations both at the current
year and one year ago. Thus, the final sample size is 303,022. Around 44.8 percent of the sample
is female, and the average age is approximately 37 years old.

2.2 Ormiston’s Skills Transferability Method
Ormiston (2006) develops an alternative method of estimating the occupational skills
transferability that directly measures the skill content of jobs. In particular, he estimates the skills
transferability based on the knowledge, skills, and abilities (KSAs) shared across occupations.
The dataset that is used to estimate the skills transferability is the O*NET, the authoritative
national database on vocational information coordinated by the United States Department of
1

The more desirable category would be based on the 3-digit level. However, estimating the skills
transferability based on Shaw’s approach using the detailed 3-digit occupational categories is tedious. It
results in almost 160,000 cell probability matrix with many cells with zero.

7

2

Labor. O*NET data characterizes 900 distinct professions along 120 standardized knowledge,
3

skill, and ability categories, as shown in Appendix Table A1. The skills transferability measure
is an estimated proportion of occupation-specific human capital that would be transferred from
one’s origin occupation to one’s destination occupation, and it is derived by the computation of a
ratio of shared KSAs between two occupations to the KSAs required within the origin
occupation. Thus, two occupations, based on Ormiston’s approach, are similar if they utilize or
share all 120 dimensions of KSAs in the similar proportions.
The O*NET dataset scores the KSAs utilized in each profession along two separate
indices: (1) the level of each KSA required of one to perform adequately, scored on a 1-7 scale,
and (2) the importance of each KSA to the responsibilities of the occupation, scored on a 1-5
scale. This two-dimensional approach is imperative in capturing both the breadth and the depth
of KSAs required in each profession (Ormiston, 2006). The product of the importance and level
scores of 120 KSA dimensions within each occupation will generate a “point” within an
occupation.
Using the KSAs for each occupation, I estimated the skills transferability by calculating
the proportion of KSAs that would be transferred from one’s origin occupation to one’s
destination occupation. I did so by computing the ratio of shared KSAs to total KSAs required
within the origin occupation. Table 1 provides the knowledge score of an industrial engineer and
human resource manager in “Administration and Management,” “Engineering and Technology,”

2

O*NET dataset is available from http:// www.onetcenter.org/database.html#archive.

3

Ormiston’s occupational skills transferability is first based on the Standard Occupational Classification
(SOC) and then is matched to the 2- and 3-digit occupational categories of the 1980 occupational
classification system provided in the CPS. While there are some differences between these two systems, I
carefully matched these categories with guidance from the CPS codebook supplement in order to prevent
the recoding problems.

8

and “Sales and Marketing” categories. For example, an industrial engineer uses 16.01 points of
the administration and management knowledge area, while the HR manager uses 19.68 points.
Within this knowledge category, the two occupations share 16.01 points. While the industrial
engineer can apply all 16.01 points of their “Administration and Management” knowledge in
his/her role as an HR manager, the HR manager, on the other hand, can apply only a portion of
his/her 19.68 points in a role as the industrial engineer, given the smaller requirement of
“Administration and Management” knowledge in the engineering occupation. Using the
transferability estimate outlined above, in this example, the transferability of the “Administration
and Management” knowledge category from the industrial engineer position to the HR manager
position is equal to 1 (16.01/16.01), while the transferability of the “Administration and
Management” knowledge category from the HR manager position to the industrial engineer
position is equal to 0.814 (16.01/19.68).
By summing this score within all 120 categories of KSAs for each pair of occupations,
and then averaging the totals to provide equal weighting, I then can derive the 467 X 467
occupational skills transferability matrix based on the 3-digit U.S. census code as well as 45 X
45 matrix at the 2-digit level, representing the estimated proportion of occupational skills
transferability.

3. Comparison of Shaw’s and Ormistion’s Estimates
Both Shaw’s and Ormiston’s skills transferability approaches have strengths and
weaknesses. Shaw’s skills transferability is the market-based approach. It combines skill factors
with demand factors as well as social factors. High demand for a given set of occupational skills
will cause more movement into that occupation from individuals who are more distant because

9

the demand curve and wages will be moving up. On the other hand, Ormiston’s skills
transferability is a technical-based approach, which is a pure measure of commonality of
knowledge skills and abilities (KSAs) among occupations.
Compared with Ormiston’s approach, the main pitfall of Shaw’s estimation is it assumes
skills transferability to be symmetric, in which the skills transferability from occupation i to j is
equal to the skills transferability from occupation j to i. In other words, the direction of the skills
ij

transfer is irrelevant, and the d , therefore, measures the average transferability values between
these two directions of movement. This assumption might not be entirely reasonable, especially
when moving from a more to less skilled occupation. It is less likely that the same proportions of
skills are transferred as in the opposite direction (Shaw, 1984). For example, this approach
implies skills transferability from nursing to being a doctor is the same as from being a doctor to
a nurse. In reality, however, we would anticipate a doctor that later became a nurse could transfer
more of the medical skill to a nurse but not vice versa. The same story holds for a university
professor who later becomes a teacher as well. Also, a symmetric relation of the distance
measure could obscure the fact that there are also non-negligible asymmetries in the skills
transferability when comparing a move from occupation i to j to a move from j to i. Ormiston’s
skills transferability, on the other hand, is a non-symmetric method, in which the skills
transferability from occupation i to j does not have to be equal to that from occupation j to i.
Another limitation of a market-based approach is that it must be driven by demand. If
there is little demand for a given occupation, there will be little movement into it (i.e., low
transferability), even from very similar occupations. As a consequence, it will mislead the
estimates if we are interested in measuring the effect of the occupational switches on wages.

10

On the other hand, the main disadvantage of Ormiston’s approach is that it is calculated
purely based on the commonality of KSAs, and may not reflect actual labor market movement
between occupations. In particular, this measure can yield a high value of skills transferability
between two occupations although the movement between those two particular occupations
rarely occurs in the market, such as the movement from a biologist to physicist, while Shaw’s
approach is based on the actual movement of individuals in the labor market reflecting both
market and technical conditions. If the actual movement between two occupations is low, the
skills transferability based on Shaw’s method is more likely to approach zero. Another problem
with Ormiston's measure could arise because the KSAs cannot possibly capture every single skill
attribute of each occupation. Thus, there are other skill factors that are not captured in the KSAs
that would explain the lower occupational mobility such as between a doctor and janitor. The
subsequent section presents the estimation results of these two approaches.

4. Estimation Results
ij

Table 2 presents the probability matrix of occupational movement (P ) for 10 selected
4

occupations at the 2-digit level. It suggests most of individuals stay in the same occupation
between “one year ago” and current year. For example, 97 percent of individuals who work as
engineers one year ago remain in the same occupation in the current year while only 0.02 percent
becomes mechanics and repairers, and 0.1 percent becomes college teachers. The very high
ij

ii

jj

retention rate would result in the small value of skills transferability, d (i.e., P + P is very
close to two). Table 3 demonstrates the estimated Shaw’s skills transferability matrix for 10

4

These 10 occupations are chosen from the different 1-digit occupational categories.

11

selected occupations. Not surprisingly, the estimated skills transferability across 2-digit
occupations is relatively low and most of them are less than 10 percent. For example, the skills
transferability from engineers to college teachers is approximately 0.027, suggesting that
approximately 2.7 percent of occupational skills used by engineers could be able to transfer to
the academics profession while around 2.3 percent would be able to transfer to mechanics and
repairers. As mentioned above, the direction of the occupational movement is irrelevant for the
skills transferability based on Shaw’s method. Therefore, in this case the skills transferability
from engineers to college teachers is equal to that from college teachers to engineers.
The average skills transferability estimate based on Shaw's approach in this sample is
approximately 0.065 with a standard deviation of 0.148 (see Table 4). The highest occupational
skills transferability is between “farm workers” and “freight, stock, and material handlers”,
which is around 0.567 while the lowest occupational skills transferability is between “lawyers
and judges” and “health diagnosing occupations”, which is 0.002.
Table 5 presents the Ormiston’s skills transferability matrix for 10 selected occupations at
the 2-digit level. For example, the skills transferability from engineers to college teachers is
approximately 0.5, suggesting that 50.0 percent of KSAs used by engineers would transfer to the
academic profession. On the other hand, only 21.7 percent of KSAs could be transferred from
engineers to construction laborers. As opposed to Shaw’s approach, the direction of occupational
movement does matter for Ormiston’s approach. While 71.9 percent of skills from executive
managers could be transferred to engineers, 53.7 percent of skills from engineers could be
transferred to executive managers.
The average skills transferability based on Ormiston’s approach in this sample is
approximately 0.564 with a standard deviation of 0.147 (see Table 4). The highest 2-digit

12

occupational skills transferability is from “supervisors, proprietors, and sales” to “public
administrators”, which is around 0.93 while the lowest occupational skills transferability is from
“public administrators” to “construction laborers”, which is 0.186. The bottom of Table 4 also
shows the summary of Ormiston’s skills transferability at the more detailed 3-digit level. The
highest skills transferability is from “ushers” to “police and detectives”, which is around 0.998,
and the lowest skills transferability is from “architects” to “garbage collectors”, which is around
0.051.
As we can see from Table 3 and Table 5, two approaches yield very different magnitudes
of skills transferability. For example, Shaw’s estimate suggests that the skills transferability from
engineers to college teachers is around 0.0271 while Ormiston’s estimate indicates the
transferability is around 0.5. In spite of this difference, the estimated correlation between Shaw’s
and Ormiston’s approach is approximately 0.45, indicating a positive and modestly strong
correlation.

5. Conclusion
This chapter estimates the skills transferability applying Shaw’s and Ormiston’s
approaches as well as discusses the strengths and weaknesses of each approach. These two
approaches produce very different results of skills transferability although they have a fairly high
correlation; therefore, the choice of measure could significantly impact the estimates of study.
On the one hand, since Ormiston’s skills transferability might not be able to capture all
aspects of occupation skills, and there might be other skill factors that are not captured in the
KSAs that would explain the lower movement between occupations, Shaw’s skills transferability
might be preferred over Ormiston’s measure. On the other hand, if we believe that the lack of

13

mobility is not due to skills transferability but rather to discrimination or other factors, then we
want to use Ormiston's measure. For example, the reason workers do not change occupations
such as from a janitor to doctor often has more to do with limited access to medical school or
social class. In that case, it is factors other than skill that explain the lack of mobility, and we
should prefer Ormiston's skills measure over Shaw's because Shaw's is contaminated with these
social factors.
In the subsequent chapters, the estimates of these two approaches will be used to develop
a new measure of occupational human capital based on Shaw’s (1984) model to examine the
determinants of workers’ wages in Chapter 2 and used to determine the earnings losses following
job displacement in Chapter 3.

14

Table 1. Score Categories Using to Calculate Occupational Skills Transferability.
Score of 3 selected knowledge categories
Administration
Engineering and Sales and
and Management Technology
Marketing

Occupation
HR manager

19.68

1.03

5.99

Industrial engineer

16.01

23.50

9.76

Table 2. The Probability Transition Matrix (1983, 1985, 1989, and 1991 March CPS)
Initial Occupation
(1980 code)
Executives, administrators
(2)
Engineers (4)
Teachers, college and
university (9)
Lawyers and judges (11)
Sales reps, finance,
business (17)
Computer equipment
operators (22)
Mail and message
distributing (25)
Mechanics and repairers
(34)
Construction laborers (41)
Farm workers (44)

Destination Occupation (1980 code)
(11)
(17)
(22)
(25)

(2)

(4)

(9)

(34)

(41)

(44)

0.941
0.006

0.001
0.970

0.001
0.001

0.000
0.000

0.003
0.001

0.000
0.000

0.000
0.000

0.003
0.000

0.000
0.000

0.000
0.000

0.007
0.004

0.001
0.001

0.916
0.001

0.001
0.983

0.001
0.001

0.001
0.000

0.000
0.000

0.001
0.000

0.000
0.000

0.001
0.000

0.012

0.000

0.000

0.001

0.924

0.000

0.001

0.002

0.000

0.000

0.003

0.000

0.001

0.000

0.002

0.916

0.002

0.001

0.002

0.001

0.003

0.000

0.001

0.000

0.002

0.000

0.925

0.002

0.002

0.001

0.005
0.003
0.004

0.001
0.000
0.000

0.000
0.000
0.000

0.000
0.000
0.000

0.001
0.001
0.002

0.000
0.002
0.000

0.000
0.001
0.001

0.919
0.006
0.014

0.003
0.856
0.008

0.004
0.006
0.829

15

Table 3. The Skills Transferability Matrix for 10 Selected Occupations (Shaw’s Approach)
Initial Occupation
(1980 code)
Other executives, administrators
(2)
Engineers (4)
Teachers, college and university
(9)
Lawyers and judges (11)
Sales reps, finance, business
(17)
Computer equipment operators
(22)
Mail and message distributing
(25)
Mechanics and repairers (34)
Construction laborers (41)
Farm workers (44)

Destination Occupation (1980 code)
(11)
(17)
(22)
(25)

(2)

(4)

(9)

(34)

(41)

(44)

1.000
0.025

0.025
1.000

0.043
0.027

0.014
0.012

0.061
0.023

0.044
0.022

0.046
0.020

0.039
0.023

0.045
0.021

0.047
0.024

0.043
0.014

0.027
0.012

1.000
0.017

0.017
1.000

0.042
0.014

0.040
0.014

0.034
0.014

0.028
0.013

0.033
0.012

0.037
0.014

0.061

0.023

0.042

0.014

1.000

0.045

0.048

0.035

0.048

0.050

0.044

0.022

0.040

0.014

0.045

1.000

0.052

0.038

0.054

0.051

0.046
0.039
0.045
0.047

0.020
0.023
0.021
0.024

0.034
0.028
0.033
0.037

0.014
0.013
0.012
0.014

0.048
0.035
0.048
0.050

0.052
0.038
0.054
0.051

1.000
0.047
0.065
0.065

0.047
1.000
0.076
0.092

0.065
0.076
1.000
0.558

0.065
0.092
0.558
1.000

16

Table 4. Summary of Occupational Skills Transferability
Shaw's approach
(2-digit)
Mean
Standard Deviation
Most similar

From

To

Farm workers

Freight, stock, and material
handlers

Other handlers, equipment
cleaners, and laborers
Forestry and fishing occupations

Skills
transferability
0.065
0.148

0.567

Construction laborers
Freight, stock, and material
handlers

0.558

Lawyers and judges
Sales related occupations
Forestry and fishing occupations

Health diagnosing occupations
Health diagnosing occupations
Health diagnosing occupations

0.002
0.003
0.004

From

To

0.548

Least similar

Ormiston's
approach (2-digit)
Mean
Standard Deviation
Most similar

Skills
transferability

0.564
0.147
Supervisors, proprietors, and
sales
Supervisors, proprietors, and
sales
Sales representatives, finance
and business

Public administrators
Supervisors-administrative
support

0.929

Public administrators

0.856

Construction laborers
Construction laborers
Other transportation and
material moving

0.186
0.188

0.884

Least similar
Public administrators
Lawyers and judges
Lawyers and judges
Ormiston's
approach (3-digit)
Most similar

From

To

0.193
Skills
transferability

Ushers
Insulation workers
Shoe machine operators

Police and detectives
Electricians
Machinists

0.998
0.996
0.995

Architects
Chief executives and general
administrators
Psychologists

Garbage collectors

0.051

Garbage collectors
Garbage collectors

0.082
0.084

Least similar

17

Table 5. The Skills Transferability Matrix for 10 Selected Occupations (Ormiston’s Approach)
Initial Occupation
(1980 code)
Other executives, managers
(2)
Engineers (4)
Teachers, college and
university (9)
Lawyers and judges (11)
Sales reps, finance, business
(17)
Computer equipment
operators (22)
Mail and message
distributing (25)
Mechanics and repairers
(34)
Construction laborers (41)
Farm workers (44)

Destination Occupation (1980 code)
(11)
(17)
(22)
(25)

(2)

(4)

(9)

(34)

(41)

(44)

1.000
0.537

0.719
1.000

0.650
0.500

0.626
0.469

0.570
0.413

0.538
0.464

0.352
0.285

0.298
0.313

0.210
0.217

0.442
0.378

0.684
0.692

0.704
0.695

1.000
0.707

0.678
1.000

0.560
0.563

0.557
0.536

0.339
0.336

0.289
0.265

0.199
0.188

0.433
0.409

0.795

0.776

0.745

0.717

1.000

0.659

0.456

0.364

0.258

0.507

0.667

0.767

0.662

0.599

0.576

1.000

0.454

0.410

0.304

0.522

0.754

0.806

0.707

0.664

0.697

0.795

1.000

0.573

0.485

0.669

0.542
0.571
0.651

0.767
0.773
0.760

0.513
0.524
0.621

0.438
0.471
0.550

0.476
0.499
0.537

0.619
0.682
0.646

0.501
0.629
0.496

1.000
0.707
0.513

0.499
1.000
0.416

0.596
0.680
1.000

18

Chapter 2
OCCUPATIONAL HUMAN CAPITAL AND WAGES: THE ROLE OF SKILLS
TRANSFERABILITY ACROSS OCCUPATIONS

1. Introduction
The concept of occupational human capital has become particularly important in recent
literature for understanding the wage growth over the lifecycle, wage inequality, and the
distribution of earnings across the labor force (Sullivan, 2006; Zangelidis, 2007; Kambourov and
Manovskii, 2009; etc.). For example, Kambourov and Manovskii (2009) find that five years of
tenure in the current occupation on average increase wages by 12 to 20 percent. With few
exceptions, the important contributions from most previous studies have focused on the role of
occupation-specific skills (i.e., skills that would be lost or not be applicable whenever workers
change occupations) and neglect the idea that skills could be transferred across occupations.
Ignoring skills transferability might introduce a measurement error into the estimation and result
in a misleading estimate of the effect of occupational human capital on wages. Therefore, in this
chapter, I intend to examine the role of occupational human capital as well as skills
transferability in the determination of workers’ wages, applying Shaw’s (1984) model. To my
knowledge, this is the first study that applies Shaw’s model to the study of wage determination.
I use data from 1979 through 2000 of the National Longitudinal Survey of Youth (NLSY)
to estimate the share of wages that can be explained by variations in occupational human capital.
This dataset provides detailed information about the employment of a nationally representative
sample of young men and women (12,826 individuals) who were between the ages of 14 and 21
when first interviewed in 1979. In particular, the 2- and 3-digit occupation codes for each job in

19

this dataset are used to create a series of occupation tenure variables for each person, and later to
create a new measure of occupational human capital.
Building on Shaw’s (1984) model, the first contribution of this chapter is to examine
whether the workers’ wages increased due to the skills transferred from the previous occupations.
The findings show that controlling for the tenure at the current occupation, two additional years
of skills accumulation transferred from the previous occupation on average increased wages by
1.2 to 2.6 percent. In addition, the new measure of occupational human capital incorporating
skills transferability, or so-called occupational investment, is one of the important sources of
wages, supporting Shaw’s original work on wage determination. Specifically, the return to (3digit) occupational investment relative to five years of occupational tenure could lead to a wage
increase of 7 to 16 percent. This implies that the skills from the previous occupation are
successfully transferred to the current occupational skills to increase wages. I also find that the
general labor market experience accounts for a large share of workers’ wages, which supports
the importance of general skill accumulation. The second contribution is to analyze and compare
the return to occupational human capital with other components of human capital. Overall, the
results of this chapter demonstrate the positive and statistically significant effect of general, firm-,
industry-, and occupation-specific human capital on wages. This chapter also presents the
separate estimates for Shaw’s and Ormiston’s skills transferability at the 2-digit level. Overall,
both approaches produce the fairly similar results. In particular, the return to 2-digit occupational
investment relative to five years of occupational tenure based on Shaw’s method increases wages
by 6.2 to 13.3 percent, while the return to occupational investment relative to five years of
occupational tenure based on Ormiston’s method increases wages by 6.1 to 15.8 percent.

20

The analysis of this chapter serves as a next step toward understanding how occupations
are related and how occupational human capital contributes to individuals’ wages. In particular,
the specification of this model allows us to quantify the importance of occupational skills from
the previous occupations for wage determination, relative to other dimensions of human capital
(e.g., general, firm-, and industry-specific human capital). In addition, it is particularly important
for understanding wage growth over the lifecycle and the distribution of earnings across the labor
force. For example, the wage-occupational specificity profile is a key source of the extent to
which the wage power of individuals is tied to specific occupations, and it is important for
assessment of the losses suffered by “displaced” workers (Kambourov and Manovskii, 2009).
Little is still known about the losses experienced by workers who find re-employment in
occupations that are not the same but are similar to their pre-displacement occupation, in which
the occupational skills transferability is relatively high. Chapter 3 will explore this relationship.
This chapter is organized as follows. In Section 2, I summarize the literature on the return
to several components of human capital, particularly occupational tenure. Section 3 describes the
estimation of occupational human capital applying Shaw’s (1984) model as well as the empirical
methodology. Then, Section 4 discusses the data used in the main analysis. Section 5 presents the
descriptive statistics and empirical results. Section 6 concludes and discusses some implications
for future research.

2. Literature Review
The traditional human capital model (Becker, 1964; Mincer, 1974) decomposes skills
into two components. The first component is “general skills,” defined as skills that are
productive and transferable across different firms and usually measured as years of labor market

21

experience. The second component is “firm specific skills,” which are defined as skills that are
productive at only one firm, generally measured as years of work in the current firm, and
nontransferable across firms. Typically, workers acquire these specific skills through both formal
and informal (specific) on-the-job training, familiarity with employers’ processes, production
lines, or business culture, and as the amount of these skills increases with tenure at the firm, the
wages also rise.
Empirical studies generally find a positive relationship between years of labor market
experience and wages. This positive relationship is due at least in part to the significant amount
of (general) on-the-job training that occurs both formally and informally on the job (Becker,
1964). However, the Current Population Survey (CPS) and other cross-sectional datasets used in
past studies lack the information on actual work histories needed to precisely measure the market
experience. Instead, researchers use a proxy measure called “potential experience,” which is
5

calculated as an individual’s age minus years of education minus six. Although this measure is
commonly used in the previous literature, it is a poor proxy for market experience for at least two
reasons. First, it assumes that individuals immediately enter the labor market after they leave
school (Altonji and Blank, 1999). However, in reality, many individuals, particularly women,
postpone entering the labor market by one or more years. Second, potential experience assumes
continuous work histories. This assumption is also implausible in reality, especially for women
because many of them experience interruptions throughout their careers, such as leaving the
labor market during their child-bearing period (Regan and Oaxaca, 2009). Ignoring these
interruptions could introduce measurement error and thus bias estimation results. Longitudinal
datasets such as the Panel Study of Income Dynamics (PSID) and National Longitudinal Survey

5

I use six because, in general, people do not start school until age of six years old.

22

of Youth (NLSY), on the other hand, allow researchers to construct a more precise “actual
experience” measure of general human capital. Regan and Oaxaca (2009) compare both
measures to examine the gender wage gap. They show that the measure of market experience
used is important. In particular, when a potential experience variable is used, almost all of the
gender wage gap is unexplained. However, when an actual work experience measure is used,
around 30 percent of the gender wage gap is attributable to differences in the mean values of the
regressors for males and females. As a result, in this analysis, I will apply the precise “actual
experience” available from the NLSY to the estimation of workers’ wages.
There are also other studies that emphasize the impact of firm-specific human capital,
which is generally measured by employment duration with a current firm, on workers’ wages.
They have found a large and positive effect of job seniority (firm tenure) on wages; i.e., the
increase in wages with tenure, holding general experience constant, is attributed to the workers’
investments in firm-specific skills (for example, Bartel and Borjas, 1981; Borjas, 1981; Willis,
1986; Topel, 1991). Topel (1991) finds that, on average, ten years of job tenure with one’s
current employer increases one’s wages by over 25 percent. Willis (1986) finds that one
additional year of job tenure is correlated with a 2 to 3 percent increase in wages. This finding,
therefore, provides the evidence that human capital is highly firm specific, which implies that the
transferability of human capital is very limited.
However, the idea that workers’ wages rise as their employment duration with the same
employer increases has been controversial. Much of the debate has focused on the potential
endogeneity of the tenure variable (for example, Altonji and Shakotko, 1987; Abraham and
Farber, 1987; Topel, 1991; Altonji and Williams, 1997). The tenure variable will be endogenous
if it is correlated with unobserved individual and job characteristics that affect wages. For

23

example, if workers who have higher (unobserved) ability tend to stay in the firm longer to
accumulate more specialized skills and receive higher wage returns than less able counterparts,
then the estimate of firm tenure would be positively biased (Neal 1998). To address this
endogeneity problem, Altonji and Shakotko (1987) use the PSID dataset and apply an
instrumental variable (IV) method. Their IV is constructed as the deviation of the tenure measure
6

from its mean within a worker firm match. Their OLS estimates are slightly lower than Topel’s
(1995) study, in which ten years of firm tenure increases wages by 18 percent. However, when
applying the IV method, they find the partial effect of firm tenure on wages is small and
statistically insignificant. Specifically, on average, ten additional years of firm tenure will
increase wages by only 6.6 percent, with much of the increase occurring in the first year on the
job.
In addition to the endogeneity problem associated with the firm tenure measures, other
more recent studies have critiqued the appropriateness of decomposing a worker’s total labor
market experience into only two components: firm-specific and general experience, as in
traditional human capital theory. Instead, these studies posit skills and experience obtained may
also be specific to certain industries or occupations, and they are more likely to contribute to the
workers’ wages, rather than general or firm-specific skills. Neal (1994) and Parent (2000)
provide evidence to suggest that industry-specific human capital is much more important than
firm-specific human capital, and the correlation between the workers’ wages and firm tenure
might also be explained to some extent by the omitted years in the industry. Using the NLSY and
PSID, Parent (2000) finds a very strong effect of industry-specific human capital, measured by
6

For example, the instrumental variable for employer tenure Tenureijt is Tenureijt  T enureij , where

T enureij is an average tenure within individual i working for employer j across time period t. By
construction, the instruments are orthogonal to the unobserved worker ability and job match.

24

industry tenure, on wages. In particular, he finds that ten years in an industry increases wages
over 20 percent. Looking at industry-specific human capital and job displacement, Neal (1994)
finds that displaced workers who switch 1-digit industry following displacement, on average,
suffer greater wage losses than those who find new jobs in their pre-displacement industry. In
addition, the post-displacement wages among displaced workers who stay in their old industry
increase more sharply with the pre-displacement tenure and years of labor market experience
than among those who switch industries. However, in the concluding section of his paper, he
raises the possibility that his results may reflect the importance of skills that are not truly specific
to given industries, but rather specific to a set of skills associated with occupations (Neal, 1995).
Thus, aside from the role of industry in explaining the wage determination, many studies
have examined the effect of occupational skills, generally measured by tenure at the current
occupation, on wages. The common conclusion across studies using different datasets is that
skills specific to occupation are likely to be an important source of wage determination. The
magnitude of these estimates, however, varies by data and empirical specifications. Using data
from the PSID Retrospective Occupation-Industry Supplemental Data Files (Retrospective files)
from 1968–1993, Kambourov and Manovskii (2009) find that five additional years of
occupational tenure increases wages by 12 to 20 percent, all else being equal. In addition, they
also find that when an individual’s occupational tenure is included in the wage equation, firm
and industry tenures explain little of the variation in wages, implying that some portions of both
firm- and industry-specific human capital have been captured by the occupational dimension of
human capital. Zangelidis (2007) extends Kambourov and Manovskii’s (2009) study using the
British Household Panel Survey (BHPS) from 1991–2001. Although he finds a much smaller
estimated effect of occupational tenure than Kambourov and Manovskii’s (2009) study, the

25

occupational tenure effect is still statistically significant and larger than the estimated effect of
firm tenure. Specifically, five years of occupational tenure leads to an estimated increase in
wages of around 2.5 percent, all else equal, while five years of firm tenure increases wages by
1.8 percent. In addition, he also finds little support for the role of industry tenure in wage
determination. Schmieder (2007) uses a German administrative dataset, the IAB Employment
Sample IAS (Regional File) from 1975–2001, and finds that the return on five years of
occupational tenure is around 2 to 9 percent. While the return to firm tenure is very sensitive to
the inclusion of occupational and industry tenure variable, the result still shows the significant
effect of return to firm tenure, which is around 3 percent, all else equal. He explains that the
reason that the return to occupational tenure is less in U.S. datasets (e.g., PSID, NLSY) is
because there are large institutional differences between the German and the U.S. labor markets
(e.g., the importance of unions, the German apprenticeship system, etc.) that make it possible for
wage growth over the career of a worker to be significantly different. For example, in the
German labor market, workers have already acquired most occupational skills during the
apprenticeship period before they actually enter the full-time labor market (Schmieder, 2007). In
contrast to Kambourov and Manovskii’s (2009) and Zangelidis’s (2007) results, Sullivan (2006)
uses data from the NLSY and estimates that both occupation and industry specific human capital
are key determinants of wages. Specifically, five years of occupational tenure, on average,
increases wages by 6 to 13 percent, all else equal, and five years of industry tenure increases
wage by 4 to 6 percent, all else equal, while the estimated effect of firm tenure is very small and
statistically insignificant. Goldsmith and Veum (2002), on the other hand, use the NLSY and
find that all components of human capital, including firm, industry, and occupational tenures, are
estimated to have a positive and significant effect on individuals’ wages. For example, they find

26

that the workers’ wages rise by 2.6 to 2.8 percent if they stay at their current firm for an
additional year while an additional year of experience in the same occupation increases wages by
2.5 percent, and an additional year of experience in the same industry raises wages by 2 percent.
In summary, these studies of occupational tenure on wage determination find that
including occupation tenure has large effects on the estimated return for firm tenure in wage
regression. In particular, with few exceptions, they show that when the occupational tenure is
included in the wage equation, firm tenure has little impact on explaining the variation in wages.
However, the magnitude of return to occupational tenure varies considerably across studies
depending on the datasets used. Taken together, the above studies indicate that the traditional
view of human capital, where skills are either firm specific or general, might not be appropriate.
Instead, it is likely that a significant component of worker skills is specific to either industry or
occupation, and these industry- or occupation-specific skills are an important determinant of
workers’ wages. This finding has important labor market implications. For example, it suggests
that wage losses associated with losing a job with a particular firm might be lower if a worker is
able to stay in the same occupation (industry) to preserve and continue accumulating
occupational (industry) specific skills.
All of the above studies, however, use occupational tenure as a proxy for occupational
human capital in wage determination analysis. Using occupational tenure does not account for
differences in the skills transferability across occupations. In other words, the occupational
tenure variable will be endogenous if it is correlated with the omitted occupational skills
transferability from the previous job that also affects wages. Therefore, the use of occupational
tenure in lieu of more precise measures of occupational human capital may introduce a
measurement error into the estimation and thus bias the results.

27

Shaw (1984) is the first one who recognized the importance of transferability of
occupational human capital in establishing particular labor market outcomes. More specifically,
Shaw notes that the return to the stock of human capital from the former occupation is not
completely foregone in the midst of an occupational switch. Instead, the return to one’s human
capital stock depends on the degree of transferability of skills from one occupation to the next.
She hypothesizes that skills acquired in a particular occupation are portable across firms, making
them “occupation specific” rather than “firm specific.” Unlike the case of general human capital,
individuals would be able to transfer only a certain amount of their occupation-specific human
capital stock in the event of an occupational change, and that amount inherently depends on the
degree of “transferability” between the two occupations (Shaw, 1984). For example, most of the
skills acquired as an engineer may not be transferable to a career as a doctor, while the skills
acquired as an accountant may have a high degree of transferability to a job as an actuary. Thus,
occupational human capital, based on Shaw’s definition, is neither completely general nor purely
specific, but partially transferable across occupations.
In spite of Shaw’s findings and contributions, only a few recent studies have started to
focus on the importance of skills transferability. For instance, Lazear (2009) introduces a novel
theoretical model called the “skill-weight” approach, suggesting that all skills are general across
firms, but each firm might require a different combination of skills. Thus, the amount of human
capital transferability across firms depends on the difference in skill requirements between the
old job and the new job. Gathmann and Schönberg (2010) use the repeated cross-section German
Qualification and Career Survey to estimate the transferability of three aggregate groups of skills
(analytical, manual, and interactive) across 64 occupations. Their results show that individuals
are more likely to move to the occupations with similar skill requirements, and more than 40

28

percent of their wage growth can be attributed to the proportion of these portable skills. Building
on Shaw’s (1984) idea, Ormiston (2006), uses the O*NET dataset to develop a transferability
matrix of knowledge, skills, and abilities (KSAs) across occupations, and then he applies this
transferability matrix to examine occupational movement associated with displacement in the
manufacturing sector. With the role and application of skills transferability, there have been only
few studies that incorporate the idea of skills transferability across occupations into the
framework of wage determination process. Thus, in this chapter, I intend to develop a new
measure of occupational human capital that directly attaches the value of occupational skills
transferability, estimated in Chapter 1, to the occupational tenure to examine wage equation.
Given the growing acceptance of occupational human capital role but the remaining
inadequacy of the occupational human capital measure commonly used (i.e., occupational
tenure), it is necessary to revisit this topic. In this chapter, I will add to the findings in the
previous literature by developing a conceptually better proxy for occupational specific human
capital, based on Shaw’s model, that incorporates skills transferability rather than occupational
tenure and applies it in wage determination. My proposed analysis will also use a rich sample of
individuals, the NLSY, which helps to more precisely estimate the impact of occupational human
capital on the workers’ wages as mentioned above. The following section describes the
estimation of proposed occupational human capital measure as well as the methodology used in
this analysis.

29

3. Empirical Methodology
3.1 The Estimation of Occupational Human Capital Measure (Occupational Investment)
This section describes the estimation of a new occupational human capital measure—
occupational investment, based on Shaw’s (1984) model. Consider an individual changes
ij

occupations from occupation i to occupation j, and γ percent of his or her skills in occupation i
are transferred to occupation j. During work in occupation i, the individual accumulates
~
occupational specific investment that equals K i . Therefore, combining the definitions of skills

transferability and occupational specific investment, the individual’s accumulated quantity of
“occupational investment,” which is the accumulation of skills an individual acquires to perform
work within an “occupation” at time t is defined as:
~
~g
j ~j
Kt  Kt   ij K i
 ...   gj K
t j 1
t g 1

(2.1)

~
~
~
where K j , K i ,…, K g represent stocks of occupational specific investment for the current

~j
occupation j, i,…,g, respectively. Define K t 

t

 kh

h t j

j

j
for kt = the percent of full income

7

ij

invested in occupation j in year t, or the intensity of investment. The skills transferability, γ
gj

and γ , are the percent of skills transferred from occupation i to j and g to j, respectively. tj and
tg are the years of entry into occupations j and g. In other words, the occupational investment in
occupation j is a weighted sum of the individual’s accumulated quantities of occupation-specific

7

In this Chapter, I use a slightly simplified version of Shaw’s (1984) model to capture the skills
transferability only from the most previous occupation. The original Shaw’s model captures the skills
transferability from all previous occupations.

30

investment in the current and previous occupations, where the weight is the transferability
ij

gj

parameter (γ ,…, γ ).
j

For simplicity, the complete individual’s occupational history, Kt , can be estimated in
the reduced form for each individual in the following way:
j
j
i
Occtt = (occ_tenuret(j) ) + ˆ ij (occ_tenuret(i) )

(2.2)

ij

γ is a proxy for the “skills transferability” from occupation i to j (either by Shaw’s or
j

i

Ormiston’s approach, developed in Chapter 1). occ_tenuret(j) and occ_tenuret(i) are the years
of tenure in the current occupation j and the first previous occupation i, respectively. Thus,
j

Occtt represents an occupational investment that includes skills from the current occupation j
j

and some proportion of skills from the first previous occupation i. In addition, Occtt+1 = 1+
j

j

Occtt for individuals who do not change occupation (occupational stayers) and Occtt+1 =1+
j

ˆ ij * Occtt for individuals who change occupation (occupational switchers).
Parent (2000) constructs two types of (industry) tenures, including continuous and noncontinuous spells. The continuous spell is measured by the consecutive number of years an
individual has been in the same industry, including tenure with the current firm. If an individual
changes firms to take another job in the same industry, he/she still adds one more year in that
industry, whereas if he/she gets another job in a different industry, his/her tenure in that industry
will reset to zero. On the other hand, for the non-continuous spells, if an individual changes to a
new job that is not in the same industry as the job he/she just left but is in the same industry as
some other previous job, then the industry tenure is not reset to zero. Instead, it is reset to the

31

level of industry tenure of that prior job. The main difference between these two types can be
considered as reflecting different depreciation rates of (industry) human capital, in which the
continuous spells correspond to the case that the (industry) specific human capital depreciates
rapidly after leaving an industry. On the other hand, the non-continuous spells can help increase
the variance in the tenure that is required to identify the tenure coefficient (Parent, 2000). In line
with the ideas of skills transferability that assumes less skills depreciation, in the estimation of
my proposed occupational human capital measure, I focus only on the non-continuous spells.
Further details on the construction of tenure measures and other variables can be found in
Appendix B.
To illustrate, consider a worker with ten years of market experience. This worker had
worked at occupation A for six years before entering his current occupation (occupation B).
Assume that the skills transferability parameter from occupation A to B is 0.3. Thus, his current
occupational investment in occupation B is 5.8 [4.0 + (0.3)*(6.0)].
To summarize, this measure of the occupational investment variable is a function of the
number of years in each occupation (occupational tenure), and the transferability of skills among
occupations. Note that if instead we assume the skills transferability is zero (i.e., occupational
skills are depreciated completely during the occupational switches), then the occupational
j

investment (Occtt ) will be equal to the years of experience in the current occupation, which is
just the occupational tenure that has been applied in the prior studies (for example, Sullivan,
2006; Zangelidis, 2007; Kambourov and Manovskii, 2009). The subsequent section describes the
empirical framework used in this analysis to examine the workers’ wages process.

32

3.2 Wage Determination
First, I estimate a wage equation that considers the effects of general, firm-, and industryspecific human capital using the OLS method as follows.
lnWijdt = β1firm_tenureijt + β2experit + β3ind_tenureidt + γ’Xijdt + υijdt

(2.3)

where lnWijdt is the natural logarithm of real hourly wage ($1979) of worker i, working at firm j,
industry d, and in time period t. firm_tenureijt denotes the current firm tenure (in level and
squared terms), experit is the “actual” labor market experience ( in level and squared terms).
ind_tenureidt denotes the 2- or 3-digit industry tenure at industry d (in level and squared terms),
and Xijdt is a vector of control variables containing worker and job characteristics including years
of education, age, marital status, major occupation and industry dummies, SMSA, year dummies,
region, a dummy variable indicating whether an individual works at a current firm more than one
8

year, and the skills intensity at the current occupation including SVP, TQ, or JobZone, (see
Appendix C for the discussion and estimation of these skills intensity measures). For a complete
definition and construction of control variables, see Appendix B. υijdt is a standard iid error term
capturing unobserved individual characteristics, such as innate ability or motivation, and it is
assumed to be uncorrelated with all the regressors. If the general, firm-, and industry-specific
skills positively contribute to the wage determination, then β1, β2, and β3 are expected to be
positive.

8

The reason for including this variable is that the first year of work with an employer typically requires
more investment in job-related skills (Parent, 2000). In particular, this variable would be expected to be
positive if investment in firm-specific skills rapidly increases at the beginning of a job.

33

Next, in order to examine the role of occupation-specific skills in the accumulated work
experience, consider the following wage equation.
lnWijdqt = β1firm_tenureijt + β2experit +β3ind_tenureidt + β4occ_tenureiqt+ γ’Xijdqt + υijdqt (2.4)
where occ_tenureiqt represents the 2- or 3-digit occupational tenure of individual i at current
occupation q as of time t (in level and squared terms). Hence, β4 represents the effect of
occupational tenure on the average of workers’ wages. If longer occupational tenure increases
workers’ wages, then β4 will be positive as well as the coefficients of other variables
representing workers’ stock of human capital. If labor market experience is completely general,
then a change in occupations or industries should not matter (i.e., the magnitude of coefficient of
experience should be slightly affected). In contrast, if a portion of the experience effect reflects
occupational or industry specific skills, then adding a control for occupational (industry) tenure
should reduce the magnitude of experience. In addition, if the occupational (industry) tenure
plays a significant role in wage determination, then we would expect that including occupation
(industry) tenure into the wage equation would also decrease the magnitude of the firm tenure’s
effect on wages, since some proportion of the estimated firm tenure effect would be attributed to
both industry and/or occupation tenures that workers have accumulated in work, not only firmspecific skills. In other words, we would expect the magnitude of β1 and β2 in Equation 2.4 to be
smaller than that of β1 and β2 in Equation 2.3.
To address the issue of different degrees of skills transferability across occupations,
Equation 2.4 replaces the occupational tenure measure with the measure of occupational
investment as follows.

34

lnWijdqt = β1firm_tenureijt + β2experit + β3ind_tenureidt + β4Occtiqt+ γ’Xijdqt + υijdqt (2.5)
where Occtiqt is the proposed measure of occupational investment (both in level and squared
terms). In other words, β4 is the estimated rate of return to one year of occupational investment
relative to occupational tenure. Thus, if this new measure of occupational human capital
contributes to the wage determination, then we would also expect β4 to be positive.
However, the return to all of the tenure measures as well as occupational investment may
be affected by the presence of various forms of unobserved heterogeneity across individuals and
across job matches. These unobserved differences across individuals such as innate ability,
motivation, as well as quality matches of the job, may bias the estimated return to human capital
measures. For example, if individuals with high unobserved ability who earn a high wage are
more likely to stay in the same occupation to obtain occupational skills, then the estimated return
to occupational tenure or occupational investment based on a simple OLS regression will be
biased up. Unfortunately, it is very difficult to eliminate all sources of potential biases introduced
by these unobserved heterogeneity. To my best knowledge, there is no “best” way to eliminate
all of the potential biases. In this chapter, I apply the fixed effect estimation (FE) to eliminate
some sources of unobserved heterogeneity. The major advantage of FE method is on the ability
to control for all characteristics of individuals in the model that is constant over time even
without measuring them. FE method eliminates potentially large sources of bias and instead
makes comparisons within individuals (Allison, 2005)
In summary, the analysis of occupational investment’s effect on the workers’ wages
relies on the careful measurement of occupational investment, including skills transferability,
and I propose that moving from the occupational tenure measure to my proposed measure of

35

occupational investment incorporating skills transferability across occupations will correct for
the measurement error and unbiasedness as well as provide more precise estimates of the
contribution to wages. The next section explains the main dataset used in this analysis.

4. Data
The main dataset in this chapter is the National Longitudinal Survey of Youth (NLSY). It
is a longitudinal dataset (1979–2008) that contains detailed information about the employment
and education experiences of a nationally representative sample of young men and women
(12,826 individuals) who were born between 1957 and 1964. The participants were between the
ages of 14 and 21 when first interviewed in 1979 (and thus were 35–42 years old in 2000). The
participants were interviewed, typically in their homes, annually through 1994 and are currently
interviewed on a biennial basis. In this analysis, I use the data from 19 survey years, 1979–2000.
The primary feature of this dataset is that it provides a weekly employment record that contains
information about the durations of employment spells, including the starting and ending dates for
all jobs, along with the wages, hours of work, and 3- and 2-digit U.S. Census occupation and
industry codes for each job. These occupation and industry codes for each job are used to create
a series of occupation and industry tenure variables for each individual in the sample. The main
advantage of this dataset is that it covers the early labor market experience when most change
occurs and when career decisions are likely to be most important.
This chapter uses the NLSY data ranging from 1979 to 2000 and focuses on white males
9

who are above 18 years old. These sample restrictions closely follow those imposed in Parent’s

9

This study uses only the data through 2000 to reduce the inconsistency of the coding format. Since 2000,
both occupation and industry codes have been changed from the 1980 Census code to the 2000 census
code.

36

(2000) study, except that the occupation and industry switches within a particular firm are also
considered the genuine mobility.

10

Individuals who are self-employed, serving in the military at

any time, or employed in the agricultural sector or public sector are all excluded from the sample
in that particular year. Further, individuals who leave the survey forever after the first interview
in 1979 such that there has been no information of their work histories are excluded from the
sample. I also delete individuals who return to school on a full-time basis within six years.
Details on sample restriction are provided in Appendix Table B1.
Based on the complete labor market history of the individuals since the first time they
entered the labor market, I was able to construct all human capital measures, including actual
labor market experience, firm tenure, industry tenure, occupational tenure, and occupational
investment. In particular, the firm tenure variable measures the length of time a worker has
worked with his/her current or most recent employer until the date of interview. This study also
uses an actual experience variable as opposed to the potential experience variable used in many
prior studies, and it measures the total work experience an individual has accumulated in the
labor market up to the date of interview. Both occupational and industry tenures (at the 2- and 3digit levels) measure the years an individual has been working in a particular occupation or
industry.
Since the measure of occupational (industry) tenure is one of the key variables in this
analysis, it is important to realize that there are problems of measurement error inherent in
identifying “true” occupational switches. The occupational coding procedure in the NLSY (and

10

Some individuals report changes in industry and/or occupation without changing jobs (firm). While
these changes are more likely to happen for occupation changes within firms, it is almost impossible to
change industries within a firm (Parent, 2000). Also, there is no way to differentiate between real
occupation or industry changes and coding errors in the data. Parent (2000) solves this problem by
restricting the industry code to be the same within a firm.

37

also in other datasets such as the CPS) uses coders to assign particular occupation codes from the
information given by survey participants. The information usually includes a brief description of
the job. There is evidence of substantial measurement error in assigning these codes, especially
at the more detailed 3-digit level (Kambourov and Manovskii, 2009). The errors in the NLSY
occupation and industry codes may contaminate the analysis and lead to a false interpretation of
occupational (industry) human capital effect on wages. Going from the 3- to 2- digit code might
help reduce to some extent the measurement error. Thus, in this analysis, I construct occupation
(industry) tenure at both 3- and 2- digit levels.
Observations with missing data on the variables used in the wage equation are excluded
for the particular sample year. In some cases, the occupation and industry codes are missing
because individuals refused to answer the question about the type of work they have performed.
Also, they could be missing because errors were made in processing the descriptions of
individuals’ jobs (Neal, 1999). After deleting all missing information on the main variables used
in the analysis, the sample is an unbalanced panel containing a total of 26,818 observations on
2,309 individuals.

5. Empirical Results
5.1 Descriptive Statistics
In this section, I first describe occupational mobility among employed workers from 1979
to 2000. In this case, I define occupational mobility as a fraction of employed workers who
report a current occupation differently from their most recent job in the previous year. Figure 1
shows that, on average, the percentages of employed workers who in any two consecutive years
change the 3-, 2- and 1-digit occupations are approximately 62.5, 51.4, and 41.8 percent,

38

respectively and are generally falling over time as the cohort ages. Note that the discrete drop in
mobility around 1992–1993 is likely a data artifact associated with a change from paper- to
computer-based survey administration.

11

The information in Figure 1 has two important

implications. First, it illustrates that we do observe that people change more often within 1-digit
occupations, and most often at the 3-digit level. Second, it suggests that changes in occupation
are very common and tend to occur early in a worker’s career, especially for young workers (i.e.,
they are not only ‘job shopping’, but also ‘occupation shopping’). Less occupational mobility
over time might be because the accumulation of occupational skills makes occupational switches
increasingly costly. Also, workers might be able to gradually locate better and better
occupational matches as time spent in the labor market increases (Gathmann and Schönberg,
2010).
Table 6 presents the descriptive statistics of the key variables of this study. It indicates
that the sample of white males earns an average of 5.7 per hour in 1979 dollars. In this sample,
workers stay at the same 3-digit level occupation for an average of 2.45 years, which is lower
than the corresponding industry tenure. The fact that occupational tenure is lower than industry
tenure at both 3- and 2- digit levels suggests that workers are more likely to move across
occupations than industries. Regarding general and firm-specific human capital, the average firm
tenure is around 3.41 years, and workers have spent, on average, 8 years in the labor market.
Note that the firm tenure is slightly higher than both occupational and industry tenure. This
interesting finding would be missed if I did not allow for occupational and industry movement
within firms. An average 3-digit occupational investment is higher than the occupational tenure
by about one year. This emanates from an additional value of skills transferability from the
11

Industry mobility patterns are quite similar to Figure 1 even though a higher percentage of workers
tend to stay in the same industry.

39

previous occupation. On average, the individual in this sample changes firms about four times
between these periods. The average number of years of education in this sample is low, around
13 years, because the analysis focuses on individuals who started their careers early in the labor
market. (See Appendix Table B2 for the complete descriptive statistics of variables used in this
analysis.)
Table 7 presents the correlation among the key human capital variables. In general, it
indicates that the human capital measures are positively correlated. For example, the correlation
between firm tenure and labor market experience is around 0.54, while the correlation between
firm tenure and occupational tenure is somewhat lower, around 0.37. Again, this discrepancy
appears because I allow occupational movement within firms to occur in the data. From an
econometric standpoint, since these human capital measures are highly correlated with one
another, the upward bias in one variable is likely to induce a downward bias in the other
estimates of the tenure slope as well (Altonji and Shakotko, 1987). Therefore, it requires some
careful interpretation of these measures. The following section presents empirical results
examining the role of occupational human capital on wage determination.

5.2 OLS Regression Results of the Wage Equation
The empirical results from the basic OLS regression are reported in Table 8. Column (1)
presents the standard wage equation with wages as a function of human capital and includes the
actual labor market experience (general human capital), firm tenure (firm-specific human capital),
and industry tenure (industry-specific human capital). The OLS coefficient estimates for
2

2

experience and experience , firm tenure and firm tenure , and industry tenure and industry
2

tenure are 0.0567 and -0.0008, 0.0207 and -0.0008, and 0.0244 and -0.0009, respectively. As
40

we can see, the coefficients of these three components of human capital are all statistically
significant at the 1-percent level. This suggests that wages increase with labor market experience,
firm tenure, and industry tenure though at a decreasing rate across workers, and that the workers’
investment in the general, firm-, and industry-specific human capital each make distinct
contributions to the workers’ wages.
Column (2) adds the occupational tenure into the wage equation. The OLS coefficient
2

estimates for occupational tenure and occupational tenure are 0.0299 and -0.0009. In this
specification, the occupational tenure has a fairly strong and positive effect on workers’ wages.
Also, the quadratic term of occupational tenure is negative and statistically significant at the 5percent level, indicating evidence of a diminishing effect of occupational tenure on wages. In
addition, the effect of occupational tenure on wages is larger than that of industry and firm
tenures. Compared to Column (1), including occupational tenure measures into the wage
equation also lowers the coefficient estimates of experience, firm tenure, and industry tenure as
2

2

well (i.e., the coefficient estimates for experience and experience firm tenure and firm tenure
2

and industry tenure and industry tenure are reduced from 0.0567 and -0.0008 to 0.0528 and 0.0007, from 0.0207 and -0.0008 to 0.0183 and -0.0007, and from 0.0244 and -0.0009 to 0.0170
and -0.0007, respectively). Following Altonji and Shakotko (1987), the relationship between
tenures and wages is summarized by calculating the effect of the first two and five years of
tenures on the wages. This is reported in Table 9. For example, the first five years of
occupational tenure based on OLS increase wages by 12.6 percent; the first five years of industry
tenure increase wages by 6.9 percent; the first five years of firm tenure increase wages by 7.3
percent, and the first five years of labor market experience increase wages by 24.6 percent.

41

Column (3) of Table 8 includes the occupational skills transferability measure into the
wage equation. The coefficient of skills transferability is 0.0127 and statistically significant at
the 1-percent level. This indicates the important effect of previous occupational skills on wages
across workers. In particular, two additional years of occupational skills from the previous
occupation, on average, increase the current wages by 2.6 percent, controlling for the current
occupational tenure. Note that including occupational skills transferability only slightly lowers
the coefficient estimates of occupational tenure.
Column (4) to (6) independently includes the measure of skills intensity in the current
occupation, either SVP, TQ, or JobZone, into the wage equation. As we can see, each of these
three skills intensity measures produces a fairly similar effect on wages, and it is statistically
significant at the 1-percent level. For example, the SVP coefficient translating into the elasticity
at the mean equals 0.09[0.0177*5.13]. In other words, a 10-percent increase in the SVP at the
mean increases wages by about 9 percent. This indicates the important effect of the occupational
skills acquisition on determining wages across workers. Note that there is similar change in other
coefficient estimates when adding each of these three skills intensity measures. Therefore, for
ease of presentation, I will present the results hereafter using only the SVP measure.
Column (7) replaces the occupational tenure measure with the occupational investment
2

measure. The coefficient estimates of occupational investment and occupational investment are
0.0291 and -0.0007. As we can see, the linear term of occupational investment is positive and
statistically significant at the 1-percent level while the squared term is negative and statistically
significant at the 10-percent level. This indicates that the occupational skills from the previous
occupation are successfully transferred to the current occupational skills so that the investment in
both current and previous occupational skills increases the workers’ wages, though at a
42

decreasing rate. The more total experience the worker has in the occupation profile, the higher
his/her wages will be. Column (8) includes the SVP measure into the wage equation. As we can
see, the coefficient of SVP is still positive and statistically significant at the 5-percent level, and
there is little change to other coefficient estimates, which means that the skills intensity is
independently important for wage determination after controlling for other human capital
measures.

5.3 FE Regression Results of the Wage Equation
As mentioned earlier, there is the possibility of unobserved heterogeneity bias across
individuals and jobs that occurs when (pooled) cross-sectional results are interpreted as a true
estimate for the average individual over time. Table 10, therefore, presents the FE results. As we
can see, except for labor market experience, other FE coefficient estimates of human capital
measures are smaller than those of OLS. This upward bias of OLS coefficient estimates is
probably from the positive correlation between the unobserved worker and job characteristics
and tenure measures that also affect wages. In particular, compared to Column (3) of Table 8,
Column (3) of Table 10 shows that the FE coefficient estimates for occupational tenure and
2

occupational tenure are reduced from 0.0311and -0.0009 to 0.0160 and -0.0006, and the
coefficient estimates for occupational skills transferability is reduced from 0.0127 to 0.0059,
with a smaller t-statistic but still statistically significant at the 10-percent level. In particular, two
years of skills accumulation from the previous occupation, on average, increase the workers’
wages by about 1.2 percent, all else equal. Despite about a 50 percent reduction in the coefficient
estimate, the skills transferability from the previous occupation is still an important source of
workers’ own wages over time.
43

Column (4) presents that the FE coefficient estimate of skills intensity, SVP, is slightly
lower than that of OLS (i.e., 0.0177 vs. 0.0117) but still statistically significant at the 1-percent
level. This indicates that the skills intensity of the current occupation has a significant effect on
the workers’ own wages over time even after eliminating the unobserved heterogeneity.
Column (5) shows that the FE coefficient estimates of occupational investment and
2

occupational investment are 0.0130 and -0.0004. As we can see, the coefficient estimate of the
linear term of occupational investment is statistically significant at the 1-percent level while that
of the squared term is statistically insignificant. The F-test for the significance of both terms of
occupational investment, however, indicates that they are highly jointly significant (F=83.05,
p<0.0001). This emphasizes the role of occupational investment as one of the primary sources of
variance in workers’ own wages over time although the diminishing effect of occupational
investment in this specification is not supported. Column (6) presents the FE coefficient
estimates of occupational investment together with SVP on wages. The coefficient estimate of
SVP is positive and still statistically significant at the 5-percent level. In particular, a 10-percent
increase in SVP at the mean increases wages by about 6 percent.
Returning to Table 9, the upper part of the table compares the OLS and FE accumulated
effects of the first two and five years of human capital measures. In particular, the first two and
five years of occupational tenure, based on OLS, increase wages by 5.6 and 12.6 percent while
the FE estimate drops to around 2.7 and 5.8 percent, respectively. The coefficient estimate of
industry tenure is somewhat smaller than that of occupational tenure. The first two and five years
of industry tenure, based on OLS, increase wages by 3.1 and 6.9 percent, while the first two and
five years of industry tenure, based on FE, increase wages by 2.2 and 4.6 percent. The first two
and five years of firm tenure, based on OLS, increase wages by 3.4 and 7.3 percent while the FE
44

estimate drops to around 2.8 and 5.7 percent. Note that while the returns of tenures are important,
they are not as large as the return to general labor market experience. The OLS return to two and
five years of labor market experience are 10.3 and 24.6 percent while the FE return to 2 and 5
years of general labor market experience are 19.9 and 47.8 percent. This large increase in the FE
return to labor market experience is probably from the strong correlation between labor market
experience and other tenure measures, which implies that the upward bias in the tenure measures
resulting from the unobserved heterogeneity will result in a downward bias in the experience
estimates (Altonji and Shakotko, 1987). The lower part of Table 9 reports the cumulative effect
of human capital, emphasizing occupational investment. As we can see, the return to
occupational investment relative to two and five years of occupational tenure based on OLS is
7.3 and 16.3 percent, while the return to occupational investment based on FE drops by over 50
percent relative to the OLS to around 3.2 and 7.1 percent. As mentioned earlier, this large drop in
the FE estimate is probably attributed to the bias from the positive correlation between
occupational investment and unobserved heterogeneity that affects wages.
To compare the effect of occupational tenure with occupational investment on workers’
wages, Table 9 also shows that the overall effect of the first five years of occupational tenure is
to raise wages by 5.8 to 12.6 percent while the occupational investment results in a wage
increase of 7.1 to 16.3 percent over the first five years. As we can see, the return to occupational
investment is higher than that of occupational tenure, indicating the significant effect of
occupational skills transferred from the previous occupation and current occupational skills on
wages.
In summary, the OLS and FE estimates of the wage equation indicate the significant
effect of the occupational skills transferability on wages. In particular, two additional years of

45

occupational skills from the previous occupation, on average, increase the workers’ wages by 1.2
to 2.4 percent controlling for the current occupational tenure, and the return to (3-digit)
occupational investment relative to five years of occupational tenure is around 7.1 to 16.3
percent. Regarding other components of human capital, the OLS and FE estimates indicate that
labor market experience increases wages by 10 to 20 percent during the first two years of work
and 22 to 48 percent during the first five years; the firm tenure increases wages by 2 to 3 percent
during the first two years of work and 5.5 to 7 percent during the first five years of work; the
industry tenure increases wages by 2 to 3 percent during the first two years of work and 4 to 7
percent during the first five years. As we can see, the large reduction in the tenure estimates is
accompanied by a large offsetting increase in the labor market experience. This is expected
because the fairly strong and positive correlation between labor market experience and tenure
measures, shown in Table 7, implies that the upward bias in the tenure estimates resulting from
unobserved heterogeneity will result in a downward bias in the OLS estimate of labor market
experience. The wage rising with the labor market experience could come from the general skills
accumulation over time or from the labor market search (Altonji and Shakotko, 1987).

5.4 Comparison of Shaw’s and Ormiston’s Skills Transferability
Table 11 reports the comparison of the OLS coefficient estimates of Shaw’s and
Ormiston’s skills transferability at the 2-digit level. Column (1) and (3) show that the coefficient
estimate of Ormiston’s skills transferability is 0.0057, and it is statistically significant at the 5percent level, while the Shaw’s skills transferability estimate is negative and statistically
insignificant. This implies that the Shaw’s skills transferability estimate at the 2-digit level does
not contribute to wages across workers after controlling for the current occupational tenure.

46

Column (2) shows that the linear estimate of occupational investment based on Ormiston’s
approach is 0.0280 and it is statistically significant at the 1-percent level while the squared term
of occupational investment is -0.0007, and it is statistically significant at the 10-percent level.
Column (4) shows that the coefficient estimates of occupational investment based on Shaw’s
approach are 0.0311 and -0.0011, and they are statistically significant at the 5-percent level. This
indicates that although the Shaw’s skills transferability in itself, on average, does not contribute
to wages across workers, the total occupational investment still has a significant effect on wages.
Table 12 reports the comparison of the FE estimates of Shaw’s and Ormiston’s
occupational skills transferability at the 2-digit level. In contrast to the OLS estimates in Table
11, the FE estimates of both Shaw’s and Ormiston’s skills transferability are negative and
statistically insignificant, indicating that occupational skills from the previous occupation might
not contribute to the workers’ own wages over time. On the other hand, Column (2) shows that
the linear estimate of occupational investment based on Ormiston’s approach is 0.0116 and it is
statistically significant at the 1-percent level while the squared term of occupational investment
is -0.0004, and it is statistically significant at the 10-percent level. Column (4) shows that the
coefficient estimates of occupational investment based on Shaw’s approach are 0.0155 and 0.0007, and they are statistically significant at the 5-percent level. Although the coefficients of
occupational investment relatively drop compared to the OLS estimate, the occupational
investment estimates from both methods are statistically significant.
Table 13 reports the results including both Shaw’s and Ormiston’s occupational
investments into the same equation. Column (2) presents the OLS result. On the one hand, the Ftest for the significance of Ormiston’s occupational investment (both linear and quadratic forms)
indicates the non-statistical significance (F=1.25, p=0.28), which implies that Ormiston’s

47

occupational investment does not add information into the model once controlling for Shaw’s
occupational investment. On the other hand, the F-test for the significance of Shaw’s
occupational investment (both linear and quadratic forms) indicates that they are highly jointly
significant (F=16.66, p<0.0001), which implies that Shaw’s occupational investment adds
information to the model of Ormiston’s occupational investment. Column (5) presents the FE
result. The F-test for the significance of Ormiston’s and Shaw’s occupational investments
indicate that both types of skills transferability are jointly significant to each other (F=4.6,
p=0.01 and F=12.07, p<0.0001, respectively).
Table 14 reports the overall effect of the first two and five years of Ormiston’s and
Shaw’s occupational investment. In particular, the OLS return of occupational investment
relative to the first two and five years of occupational tenure based on Ormiston’s method are
7.09 and 15.81 percent, while the FE return to the first two and five years drops to around 2.84
and 6.13 percent. Similarly, the OLS return to the first two and five years of occupational
investment based on Shaw’s method are 6.03 and 13.33 percent while the FE return to the first
two and five years drops to around 2.90 and 6.19 percent. As we can see, both approaches
produce a fairly similar effect on workers’ wages.

6. Conclusion and Future Research
Most previous studies of the labor market assume that skills are either completely general
(e.g., years of education and labor market experience) or specific to a particular firm (e.g., firm
tenure). Recent studies, however, suggest that skills might be specific to occupation, and they are
more likely to be an important source of wage determination. In this chapter, I apply Shaw’s
(1984) concept of occupational human capital to create a measure of occupational investment
48

incorporating skills transferability as a proxy for occupational human capital. The key findings of
this chapter suggest that the occupational skills from previous jobs can also affect the workers’
wages at the current job and that the occupational investment measure is one of the important
determinants of workers’ wages over the lifecycle. In particular, controlling for the tenure at the
current occupation, two years of occupational skills transferred from the previous occupation, on
average, increase wages by 1.0 to 2.6 percent, and the return to occupational investment relative
to the first five years of occupational tenure at the 3-digit level is around 7.1 to 16.3 percent.
Since the (3-digit) occupational skills transferability is statistically significant based on both OLS
and FE, it implies that occupational skills transferability not only helps to explain differences in
wages across workers (relative wages) but also is necessary for the wage growth of the worker as
well. In addition, the finding shows that the occupational skills intensity at the current occupation
is also important for determining the workers’ wages. In particular, a 10-percent increase in SVP
at the mean increases current wages by about 6 to 9 percent. Moreover, consistent with Altonji
and Shakotko’s (1987), I also find that general labor market experience accounts for a large share
of wages. This indicates the importance of general skills accumulation, which is emphasized in
human capital theory, and the labor market search, which is emphasized in the search model
(Altonji and Shakotko, 1987).
One of the limitations of this study comes from the possibility of an additional
endogeneity problem. FE estimation does not eliminate all of the potential biases created by the
correlation between tenure variables and the error term in the wage equation. In particular, FE
only eliminates the bias from the unobserved factors that are constant over time. However, there
are still other unobserved factors, such as the job quality match, that are not constant over time,
and they are likely to be correlated with the tenure measures and wages. It is possible that a

49

worker could get into a “good” firm that always pays higher wages, so he/she is more likely to
stay in that firm longer. If this is the case, then the FE-estimated coefficients on firm tenure
would be biased up. Since the firm tenure is positively and highly correlated with other tenure
measures, the FE estimates of other tenures, including occupational tenure and occupational
investment, would be biased down. In other words, the estimated coefficients of occupational
tenure and investment would be the lower bound of the true effect of occupational human capital.
While not without limitation, the findings in this chapter significantly enhance our
understanding of the importance of occupational human capital and skills transferability in wage
determination. My results suggest that the occupational investment is one of the key sources of
variation in wages.
For the future research, the concept of occupational human capital is relevant for studies
of wage inequality (e.g., gender wage gap, or racial wage disparities). Specifically, the gender
wage gap literature considers the role of post-school human capital investment operationalized as
accumulation of general skills (years of market experience), the returns to firm tenure, and the
return to job shopping over the career. However, there has been no research that directly
incorporates the idea of occupational human capital into the gender wage differences. It is well
known that men and women are concentrated in different occupations throughout the labor force
and have different patterns of lifecycle labor supply. Thus, in future research, I can also examine
differences in patterns of occupational change and skills transferability as well as skills intensity
by gender and the extent to which these differences contribute to the gender wage gap.
Similarly, the idea of occupational human capital could apply to a study of the public and
private wage differentials. The U.S. labor force studies generally indicate that public workers’
earnings, on average, are higher than private workers’ earnings. While the private workers’ pay

50

and benefits have stagnated, public workers’ compensation advantage has grown from $30,415 in
2000 to $61,998 in 2009, which is more than double what private sector workers earn (Cauchon,
2010). The clarification for the existence of these wage differences is still a puzzle. There are
multiple factors proposed to explain the likelihood of positive public wage differentials. One
explanation emanates from the difference in union bargaining power. Belman et al. (1997) contend
that public sector unions may have a greater ability than private sector unions to raise their relative
wage rates because of the relatively low elasticity of demand for labor in the public sector.
Another explanation comes from the segmented labor market model. The recent evidence shows
that the distribution of workers across occupations shared by public and private sectors are very
different (Belman and Heywood, 2004). Those occupations in the public sectors are much more
educated and more likely to be in professional or technical positions than those in the private
sector. Despite taking into account those above factors, there is still a significant residual to be
explained. In particular, there is relatively little research examining the role of occupational human
capital. The public wage premium might exist because of the lower relative occupational mobility
or higher skills transferability of the public sector labor workforce, which in turn will contribute to
higher earnings.
Also, future research may benefit from the application of occupational skills
transferability measure to the modeling of other outcome variables as well. For example, this
paper only examines the effect of occupational skills transferability on wages and does not
consider the effect on other types of benefits. Thus, it might be interesting for the future research
to investigate the effect of occupational human capital, especially the skills transferability, on
total compensation, which also includes other benefits such as the health insurance and pension
(Podgursky and Swaim, 1987).

51

Figure 1. Occupational Changes among Employed Workers at the 1-, 2-, and 3-digit Occupations
from 1979–2000

50
40
30
20

Percent(%)

60

70

Occupational Mobility (1979-2000)

1980

1985

1990
Year
3-digit
1-digit

52

1995
2-digit

2000

Table 6. Descriptive Statistics of Key Variables (1979–2000)
Variable
Real hourly wage ($1979)
Age
Education
Occupation tenure (3-digit)
Occupation tenure (2-digit)
Occupational investment (3-digit)
Occupational investment (2-digit, Shaw’s
approach)
Occupational investment (2-digit,
Ormiston’s approach)
Industry tenure (3-digit)
Industry tenure (2-digit)
Firm tenure
Experience
SVP
Number of firms changed
Number of Observations
Number of Individuals

Mean
5.70
27.43
12.22
2.45
3.19
3.29

Std. Dev.
3.61
5.82
2.31
2.46
3.11
2.80

3.30

3.12

4.27
3.40
3.96
3.41
8.12
5.14
3.71

3.42
3.49
3.85
3.91
5.33
1.71
2.57
26,818
2,309

Table 7. Correlations among Key Human Capital Measures

Experience
Firm tenure
Industry tenure
Occupational tenure
Occupational
investment

Occupational Occupational
Experience
Firm tenure Industry tenure tenure
investment
1.000
0.559
1.000
0.633
0.603
1.000
0.579
0.465
0.581
1.000
0.687

0.524

53

0.640

0.922

1.000

Table 8. OLS Estimates of the Wage Equation
Independent variable
Experience (in years)
Experience

2

Firm tenure (in years)
Firm tenure

2

Industry tenure (in
years)
2
Industry tenure

(1)
0.0567
(10.22)

(2)
0.0528
(9.49)

(3)
0.0510
(9.10)

(4)
0.0503
(9.00)

(5)
0.0513
(9.18)

(6)
0.0502
(8.44)

(7)
0.0491
(8.56)

(8)
0.0484
(8.45)

-0.0008 -0.0007 -0.0007 -0.0007 -0.0007 -0.0007 -0.0007 -0.0007
(-3.71) (-3.55) (-3.49) (-3.38) (-3.55) (-2.99) (-3.13) (-3.03)
0.0207 0.0183 0.0180 0.0178 0.0167 0.0178 0.0179 0.0177
(5.92)
(5.29)
(5.16)
(5.09)
(4.79)
(4.64)
(5.12)
(5.06)
-0.0008
(-3.68)

-0.0007
(-3.33)

-0.0007
(-3.26)

-0.0007
(-3.21)

-0.0006
(-2.89)

-0.0007
(-2.66)

-0.0007
(-3.23)

-0.0007
(-3.19)

0.0244
(6.79)

0.0170
(4.73)

0.0168
(4.66)

0.0167
(4.64)

0.0170
(4.72)

0.0165
(4.26)

0.0174
(4.85)

0.0173
(4.83)

-0.0009
(-3.73)

-0.0007
(-2.88)

-0.0007
(-2.97)

-0.0007
(-2.96)

-0.0007
(-3.05)

-0.0007
(-2.7)

-0.0007
(-3.11)

-0.0007
(-3.1)

0.0299
(7.04)

0.0311
(7.15)

0.0306
(7.04)

0.0307
(7.05)

0.0281
(5.88)

-0.0009
(-2.32)

-0.0009
(-2.24)

-0.0009
(-2.2)

-0.0009
(-2.19)

-0.0006
(-1.33)

0.0127
(3.06)

0.0118
(2.84)

0.0120
(2.85)

0.0095
(2.21)
0.0291
(6.31)

0.0286
(6.19)

-0.0007
(-1.98)

-0.0007
(-1.95)
0.0175
(3.85)

26,818

26,818

Occupation tenure (in
years)
2
Occupation tenure

Occupational
transferability
Occupational
investment
Occupational
2
investment
SVP

0.0177
(3.91)

TQ

0.0294
(4.96)

JobZone
#Observations

26,818

26,818

26,818

2

26,818

26,515

0.0459
(5.70)
23,041

Adj. R
0.4013 0.4064 0.4069 0.4076 0.4067 0.4168 0.4067 0.4074
Note: t-statistics are in the parentheses. Not included independent variables are years of education, age,
marital status, major occupation and industry dummies, SMSA, year dummies, region dummies, and a
dummy variable indicating whether an individual works for a current firm more than one year.

54

Table 9. Cumulative Returns to Human Capital Measures

Variable
Firm tenure
Occupational tenure
Industry tenure
Total experience

Firm tenure
Occupational investment*
Industry tenure
Total experience

OLS
Column(2) of Table 8
2 years
5 years

FE
Column(3) of Table 10
2 years
5 years

0.034
(0.008)
0.056
(0.010)

0.073
(0.023)
0.126
(0.031)

0.028
(0.006)
0.027
(0.008)

0.057
(0.018)
0.058
(0.024)

0.031
(0.008)
0.103
(0.012)

0.069
(0.024)
0.246
(0.033)

0.022
(0.006)
0.199
(0.016)

0.046
(0.019)
0.478
(0.042)

Column(7) of Table 8
2 years
5 years
0.033
0.071
(0.008)
0.073
(0.015)
0.032
(0.008)
0.096

(0.023)
0.163
(0.047)
0.069
(0.024)
0.229

Column(5) of Table 10
2 years
5 years
0.028
0.056
(0.006)
0.032
(0.011)
0.023
(0.006)
0.195

(0.018)
0.071
(0.036)
0.048
(0.019)
0.468

(0.012)
(0.034)
(0.016)
(0.043)
Note: standard deviations are in the parentheses.
* The return to occupational investment is calculated by using the relative
number of current occupational tenures and the skills transferability.

55

Table 10. FE Estimates of the Wage Equation
Independent variable
Experience (in years)
Experience

(1)
0.1058
(14.24)

Firm tenure

2
Industry tenure

(6)
0.0996
(13.33)

-0.0013
(-6.45)
0.0160
(5.79)

-0.0013
(-6.36)
0.0158
(5.72)

-0.0013
(-6.31)
0.0156
(5.65)

-0.0013
(-6.18)
0.0158
(5.72)

-0.0013
(-6.13)
0.0157
(5.66)

-0.0010
(-5.40)
0.0157
(5.64)

-0.0009
(-5.22)
0.0125
(4.53)

-0.0009
(-5.18)
0.0125
(4.50)

-0.0009
(-5.11)
0.0124
(4.47)

-0.0009
(-5.20)
0.0132
(4.77)

-0.0009
(-5.13)
0.0132
(4.76)

-0.0007
(-3.38)
0.0148
(4.52)

-0.0007
(-3.45)
0.0160
(4.75)

-0.0007
(-3.46)
0.0159
(4.73)

-0.0007
(-3.69)

-0.0007
(-3.71)

-0.0006
(-2.12)
0.0059
(1.92)

-0.0006
(-2.10)
0.0054
(1.77)
0.0130

Industry tenure (in years)

(5)
0.1000
(13.37)

-0.0006
(-2.10)

2

(4)
0.1003
(13.48)

-0.0008
(-3.89)

Firm tenure (in years)

(3)
0.1006
(13.53)

-0.0014
(-6.79)
0.0169
(6.14)

2

(2)
0.1021
(13.77)

0.0129

(3.71)

(3.66)

-0.0004
(-1.31)

-0.0003
(-1.27)
0.0115
(3.43)
26,818

Occupation tenure (in years)
2
Occupation tenure

Occupational transferability
Occupational investment
Occupational investment

2

SVP
#Observations

26,818

26,818

2

26,818

0.0117
(3.48)
26,818

26,818

Adj. R
0.3088
0.3100
0.3101
0.3101
0.3099
0.3104
Note: t-statistics are in the parentheses. Not included independent variables are years of education,
age, marital status, major occupation and industry dummies, SMSA, year dummies, region
dummies, and a dummy variable indicating whether an individual works for a current firm more
than one year.

56

Table 11. Comparison of Shaw’s and Ormiston’s Skills Transferability (OLS)
OLS
Ormiston's approach
(1)
(2)
0.0453
0.0418
(12.12)
(10.73)
-0.0005
(-3.45)
0.0158
(5.80)

Experience

2

Firm tenure (in years)
Firm tenure

2

Industry tenure (2-digit, in years)
2
Industry tenure

SVP
Occupation tenure (2-digit, in
years)
2
Occupation tenure

Occupational transferability
Occupational investment
Occupational investment

-0.0005
(-2.84)
0.0159
(5.79)

-0.0006
(-3.53)
0.0160
(5.87)

-0.0005
(-3.30)
0.0160
(5.85)

-0.0006
(-3.57)
0.0195
(7.50)

-0.0006
(-3.66)
0.0418
(10.73)

-0.0006
(-3.56)
0.0197
(7.55)

-0.0006
(-3.57)
0.0196
(7.52)

-0.0008
(-5.01)
0.0180
(5.93)

Independent Variable
Experience (in years)

Shaw's approach
(3)
(4)
0.0471
0.0456
(12.62)
(12.29)

-0.0005
(-2.84)
0.0177
(5.82)

-0.0008
(-4.93)
0.0182
(6.01)

-0.0008
(-4.93)
0.0184
(6.06)

0.0311
(11.10)

0.0297
(10.72)

-0.0010
(-5.31)
0.0057
(2.02)

-0.0010
(-5.36)
-0.0174
(-1.09)
0.0280
(8.69)

2

0.0311
(10.99)

-0.0007
(-3.61)
26,818

-0.0011
(-5.53)
26,818

#Observations
26,818
26,818
2
Adj. R
0.4084
0.4075
0.4084
0.4083
Note: t-statistics are in the parentheses. Not included independent variables are years
of education, age, marital status, major occupation and industry dummies, SMSA, year
dummies, region dummies, and a dummy variable indicating whether an individual
works for a current firm more than one year.

57

Table 12. Comparison of Shaw’s and Ormiston’s Skills Transferability (FE)
FE
Ormiston's approach
(1)
(2)
0.0982
0.0963
(13.04)
(12.64)

Experience

2

Firm tenure (in years)
Firm tenure

2

Industry tenure (2-digit, in years)
2
Industry tenure

SVP
Occupation tenure (in years)
2
Occupation tenure

Occupational transferability

Shaw's approach
(3)
(4)
0.0987
0.0973
(13.11)
(12.98)

-0.0012
(-5.70)
0.0140
(5.04)

Independent Variable
Experience (in years)

-0.0012
(-5.37)
0.0142
(5.09)

-0.0012
(-5.75)
0.0140
(5.04)

-0.0012
(-5.60)
0.0140
(5.03)

-0.0008
(-4.49)
0.0151
(5.38)

-0.0008
(-4.60)
0.0163
(5.83)

-0.0008
(-4.48)
0.0150
(5.37)

-0.0008
(-4.50)
0.0150
(5.37)

-0.0007
(-4.02)
0.0123
(3.68)
0.0147
(5.13)

-0.0008
(-4.36)
0.0120
(3.57)

-0.0007
(-4.01)
0.0122
(3.65)
0.0144
(5.17)

-0.0007
(-4.03)
0.0124
(3.68)

-0.0006
(-3.09)
-0.0012
(-0.39)

-0.0006
(-3.06)
-0.0210
(-1.01)

Occupational investment
Occupational investment

0.0116
(3.39)
-0.0004
(-1.67)
26,818

2

0.0155
(5.55)
-0.0007
(-3.18)
26,818

#Observations
26,818
26,818
2
Adj. R
0.3113
0.3106
0.3114
0.3113
Note: t-statistics are in the parentheses. Not included independent variables are years of
education, age, marital status, major occupation and industry dummies, SMSA, year
dummies, region dummies, and a dummy variable indicating whether an individual works
for a current firm more than one year.

58

Table 13. Comparison of Additional Information from Shaw’s and Ormiston’s Occupational
Investment
Independent Variable
Occupational investment
(Shaw's)
Occupational
investment2(Shaw's)

(1)

(3)

FE
(5)

(4)

0.0311
(10.99)

0.0276
(4.62)

0.0155
(5.55)

-0.0011
(-5.53)

-0.0011
(-2.32)

-0.0007
(-3.18)

(6)

0.1029
(2.62)
-0.0153
(-4.49)

Occupational investment
(Ormiston's)

0.0037
(0.53)

0.0280
(8.69)

-0.0202
(-2.64)

0.0116
(3.39)

0.0001
(0.18)

-0.0007
(-3.61)

0.0016
(3.03)

-0.0004
(-1.67)

26,818

26,818

26,818

26,818

26,818

26,818

0.4083

0.4084

0.4075

0.3113

0.3119

0.3106

Occupational
investment2(Ormiston's)

#Observations
2
Adj. R

OLS
(2)

Note: t-statistics are in the parentheses. Not included independent variables are experience, firm
tenure, industry tenure, years of education, age, marital status, major occupation and industry
dummies, SMSA, year dummies, region dummies, and a dummy variable indicating whether an
individual works for a current firm more than one year.

Table 14. Comparison of Cumulative Returns to Shaw’s and Ormiston’s Occupational
Investment

Variable
Occupational investment
(Ormiston) 2-digit

OLS
2 years
0.0709
(0.0101)
0.0603

FE
5 years
0.1580
(0.0305)
0.1333

2 years
0.0284
(0.0107)
0.0290

5 years
0.0613
(0.0326)
0.0619

Occupational investment
(Shaw) 2-digit
(0.0067)
(0.0199)
(0.0067)
(0.0201)
* The return to occupational investment is calculated by using the relative number of current
occupational tenures and the skills transferability.

59

Chapter 3
OCCUPATIONAL HUMAN CAPITAL AND EARNINGS LOSSES OF DISPALCED
WORKERS: DOES THE DEGREE OF SIMILARITY BETWEEN PRE- AND POSTDISPLACEMENT OCCUPATIONS MATTER?

1. Introduction
The costs of job displacement have received considerable attention from policy makers
and researchers in previous decades. The recent recession and the resulting dramatic increase in
involuntary displacement have increased interest in this issue. Millions of prime-age workers
have been displaced, and those fortunate enough to find employment have often suffered
substantial reductions in earnings. Studies of the U.S. labor force have indicated that displaced
workers suffer a large and persistent loss of earnings (for example, Ruhm, 1991; Jacobson et al.,
1993; Kletzer and Fairlie, 2003). Ruhm (1991) found that the weekly earnings of displaced
workers were 16 percent lower than those of non-displaced workers one year following the
displacement. Four years after displacement, their earnings remained 14 percent below their
former earnings. Given the large and ongoing losses suffered by displaced workers, it is
important to determine the factors affecting the magnitude of earnings losses associated with
displacement.
The goal of this chapter is to investigate the effect of accumulated human capital, and
particularly occupational human capital, on earnings losses. Specifically, to what extent does the
similarity between pre- and post-displacement occupations affect the earnings losses of the
displaced workers? I hypothesize the earnings losses associated with displacement are affected
by the degree to which occupation-specific skills can be transferred across occupations. Among
workers who switch occupations following displacement, I predict those who switch to a closer
60

(more similar) occupation will suffer smaller earnings losses than those who switch into a
dissimilar occupation, as they lose fewer of their occupation-specific skills. Unlike previous
studies of job displacement, I use a continuous measure of occupational skills transferability
based on Shaw’s and Ormiston’s approaches developed in Chapter 1 to measure the similarity
between pre- and post-displacement occupations of reemployed displaced workers.
This chapter uses the data from the 2004, 2006, 2008, and 2010 Displaced Worker Survey
(DWS) biennial supplement to the Current Population Survey (CPS) to estimate the portion of
earnings losses attributable to lack of occupational skills transferability. This dataset contains the
retrospective data on both pre- and post-displacement labor market circumstances, including
occupations, for a large sample of workers who lost their jobs over the period of 2001 to 2009.
The main findings of this chapter show that post-displacement earnings losses are highly
correlated with the degree of similarity between pre- and post-displacement occupations.
Displaced workers who find jobs in occupations similar to their previous jobs, as measured by
occupational skills transferability, suffer smaller earnings losses than those who find less similar
jobs. For example, using Ormiston’s measure, displaced workers whose new jobs measure 10
percentage points closer to their previous job on the occupational similarity measure experience
approximately 4 percent less earnings losses. In addition, this relationship is non-linear: higher
skills transferability reduces the earnings losses at a decreasing rate.
These findings provide important insight into the determination of earnings losses
following job displacement. First, from a theoretical perspective, this research contributes to the
human capital literature by emphasizing the importance of occupational human capital and the
idea that occupational skills are (partially) transferable across occupations. From a policy
perspective, it provides valuable guidance for the development of government programs to help

61

displaced workers. For example, the Trade Adjustment Assistance (TAA) program provides
trade-displaced workers with retraining, job-search assistance, a health care tax credit, and an
additional period of unemployment benefits at a total cost of $2 billion annually (White, 2009;
Baicker and Rehavi, 2004). The objective of job-search assistance is to help displaced workers
return to work as soon as possible in order to increase their short-run earnings. In contrast, my
results suggest that a focus on helping workers find occupations similar to their pre-displacement
jobs may help reduce the earnings losses of displaced workers.
The structure of this chapter is as follows. Section 2 summarizes the literature on the role
of human capital in earnings losses due to displacement. Section 3 sets up the methodological
framework. Section 4 describes the main dataset and related variables. Section 5 discusses the
descriptive statistics. Section 6 presents the main findings and robustness tests based on
Ormiston’s approach at the 3-digit level as well as the comparison of Shaw’s and Ormiston’s
approaches at the 2-digit level. The last section concludes and discusses some of the implications
for policy makers and for future research.

2. Literature Review
In general, a displaced worker is defined as someone who is involuntarily separated from
his or her job as a consequence of economic factors and business decisions of the employer that
are beyond the control of the workers. Those who lose jobs because of individual job performance
or who decide to voluntarily quit are not considered displaced workers. Studies of the U.S. labor
force generally agree that displaced workers earn less in their post-displacement occupations than
their original jobs; however, they differ in their estimates of the amount of earnings losses and
provide little insight into the sources of earnings losses. For example, Jacobson et al.(1993) used

62

quarterly administrative data from the state of Pennsylvania and found that the displaced workers
who are 20 years old or older suffered earnings losses of more than 40 percent one year following
displacement. Six years later, their earnings were still reduced by 25 percent. Using the National
Longitudinal Survey of Youth (NLSY), Kletzer and Fairlie (2003) found that young displaced
male and female workers (19 to 27 years old) have lower earnings by approximately 9 percent five
years after displacement. The differences in estimates across studies could be due to differences in
the data used or the methodology, or the discrepancies could simply reflect unexplained variance
in the earnings losses workers experience.
There are many factors in the previous literature that were proposed and tested empirically
to explain the earnings losses of displaced workers, including human capital attributable to a job,
unionization, job quality match, and local market conditions (Topel, 1991; Carrington, 1993;
Kletzer, 1998). In this section, I review studies based on the human capital theory. The idea that
human capital is an important determinant of earnings is not new. The human capital model
proposed by Becker (1964) categorizes human capital skills into two types. Although these two
types are widely known throughout the economics literature, it is useful for the purpose of
defining occupational human capital to revisit these original distinctions. The first type of human
capital in Becker's model is “general human capital,” which is defined as skills that are productive
and completely transferable across different firms. The second is “firm specific human capital,”
which is defined as skills that are productive (unique) at only one firm and not transferable across
firms. Typically, workers acquire these specific skills through on-the-job training by the employer,
familiarity with employers’ processes, production lines, or business culture. As the amount of skill
increases with tenure at a given firm, earnings also rise. On the one hand, if human capital skills
are “general” in the case of being transferable across firms, then the effect of displacement on

63

earnings losses should be minor. On the other hand, if firm-specific skills are important for
determining earnings, then workers with high levels of firm-specific skills who are displaced from
their jobs are more likely to experience large and possibly persistent earnings losses (Kletzer,
1998). There is some empirical evidence to support the idea that workers with more specific
human capital experience greater earnings losses. Most previous studies find that earnings
following job displacement is strongly correlated with pre-displacement years of work experience
and job tenure (for example, Podgursky and Swaim, 1987; Topel, 1990; Carrington, 1990; Faber,
1993). Faber (1993) used the Displaced Workers Survey (DWS) and found that each additional
year of pre-displacement firm tenure is related to additional earnings losses of 1to1.3 percent.
Using a different methodology, Topel (1991) found that when a worker with 10 years of tenure is
displaced, post-displacement wages were approximately 25 percent lower.
More recent literature on job displacement has started to explore the importance of other
dimensions of human capital and to emphasize that workers’ earnings likely depend on skills that
are not completely general or firm-specific but instead specific to the worker’s industry or
occupation. Several studies have found that some skills are transferable within but not between
industries (for example, Podgursky and Swaim, 1987; Carrington, 1993; Neal, 1995). In particular,
the post-displacement earnings of workers who change industries are lower than those who could
find a new job in the same industry. For example, Neal (1994) found that displaced workers who
switch between 1-digit industries following displacement, on average, suffer greater wage losses
than those who find new jobs in their pre-displacement industry. In addition, the postdisplacement wages among displaced workers who stay in their old industry increase more sharply
with pre-displacement tenure and years of labor market experience than among those who switch
industries. In a related study, Carrington and Zaman (1994) found that there is inter-industry

64

variation in the costs of job displacement. For instance, their results indicated that workers who
were displaced from the manufacturing and mining industries, on average, suffered greater wage
losses than those who were displaced in the service industry.
There are also several studies that have considered the importance of occupational human
capital in explaining earnings losses (Carrington, 1990; Topel, 1990; Kambourov and Manovskii,
2009). For example, Topel (1990) used the Panel Study of Income Dynamics (PSID) from 1968 to
1985 and found that the workers who changed 1-digit occupations after displacement had a shortrun reduction in annual earnings of approximately 17.2 percent during the first two years after
displacement, while displaced workers who changed between 1-digit industries experienced only a
7 percent reduction in annual earnings. In addition, Kambourov and Manovskii (2009)
demonstrated that:
“On average, a currently employed worker who was displaced from a job in the preceding
5 years suffers 15 percent reduction in weekly earnings. However, those who stay in the
same occupation after the displacement only suffer 6 percent drop in their weekly earnings,
even after controlling for the pre-displacement firm tenure whereas those who switch their
occupation experience an 18 percent drop.” (pg. 63)
Thus, there is a growing body of evidence that some human capital is specific to occupation and
industry. These empirical results are taken as evidence to support the idea that displaced workers
who regain employment in the same occupation or industry are more likely to continue using more
of their human capital, as so suffer lower earnings losses.
The aforementioned studies usually characterize occupations or industries upon
reemployment as either “the same” or “different,” generally using a dichotomous variable and
either the 1-, 2- or 3-digit level industrial or occupational classifications. This approach

65

inherently assumes that workers who could not find a job in their previous occupational or
industrial category will lose all of their occupation- or industry-specific skills. However, in
reality, at least some skills are likely to be transferred across occupations. For example, an
engineer who is displaced from his old job and gets a new job as a clerk should be able to apply
some portion of his engineering skills to a clerical job.
The idea of occupational skills transferability was first introduced by Shaw (1984). She
defined the concept of occupational human capital as “individual investment in skills which
relates to one particular occupation, and he/she would be able to transfer a certain amount of
his/her occupation skills during the occupational switch depending on the “transferability”
between those two occupations” (pg. 320–321). Ormiston (2006) later developed an occupational
skills transferability index and showed that there is a positive relationship between skills
transferability and occupational movement as well as post-displacement earnings among bluecollar workers.

12

In particular, he found that displaced blue-collar workers are more likely to

choose a new occupation with high skills transferability. Building on Shaw’s definition and the
above findings, I apply the idea that skills specific to a given occupation are transferable to
similar occupations but have limited transferability to dissimilar occupations. Therefore, workers
should experience less (more) earnings losses when reemployed in occupations that is more (less)
similar to their old jobs.

12

His study mainly focuses on blue-collar occupations that include: transportation and material moving;
production; installation, maintenance, and repair; construction and extraction; and farming, fishing, and
forestry.

66

3. Empirical Methodology
As a first approximation, this chapter assumes that workers in the sample were
exogenously displaced, and then a subset of them was randomly assigned to a new job. Of those
who end up in a new job, I assume some are in occupations that are similar to the predisplacement occupation, and others are not.

13

If these assumptions are true, then ordinary least

square (OLS) regression will provide unbiased and consistent estimators. Equation 3.1 is the
baseline specification for estimating the impact of occupational similarity on earnings losses.
’

ΔlnWi = β0 + β1Transferi,lc + γ Xi

+ εi

(3.1)

ΔlnWi is the difference in the natural logarithm of real weekly earnings between pre- and post14

displacement jobs for individual i,

and negative values of this variable indicate larger earnings

losses. Transferi,lc is the skills transferability between pre- and post-displacement occupations
(i.e., from occupation l to occupation c) at both 2- and 3-digit levels.

15

Xi is a vector of control

variables including age, gender, marital status, race, years of tenure at the lost job, and other
worker characteristics and work history variables (see Appendix Table D1 for a complete
summary of all independent variables). Studies in this literature often control for labor market
experience using the proxy variable potential experience (age-education-six) since the DWS
lacks the measure of an actual year of market experience. However, given the change in the
measure of education in the DWS after the 1992 survey from an actual number of years of
13

However, these workers might not represent a random sample of displaced workers because the
probability of being displaced may differ across occupations.
14
The real weekly earnings are measured in $2010. Also, since the dependent variable is in the log form,
the estimated coefficients could be interpreted as percentage changes in earnings losses due to the job
displacement.
15
As described earlier, Ormiston’s skills transferability is estimated at both the 2- and 3-digit levels
while Shaw’s skills transferability is estimated only at the 2-digit level.

67

education to education level, I opt for using worker age and education level instead. In addition,
I include the unemployment rate during the year of the survey across states and a dichotomous
variable indicating the year of displacement as proxies for labor market conditions that could
16

affect the opportunities for displaced workers to find new jobs.

εi is an error term with mean

zero and constant variance, and it is assumed to be independent and identically distributed across
individuals.
Thus, β1 in Equation 3.1 is the estimated effect of skills transferability between pre- and
post-displacement occupations on earnings losses. If the transferability of occupational skills is
important for determining the earnings losses, post-displacement earnings should be higher for
reemployed workers who find occupations more similar to their old jobs. This means β1 is
anticipated to be positive.
Even if greater occupational similarity does reduce earnings losses on average, there is
reason to expect diminishing returns to similarity. Specifically, the return to one additional
percentage point of skills transferability will be higher for displaced workers who find postdisplaced employment in an occupation with a low value of skills transferability (i.e. in a very
dissimilar occupation) than those who have already found a closer match. To test this hypothesis,
I estimate the following model:
2

’

ΔlnWi = β0 + β1Transferi,lc + β2Transfer i,lc + γ Xi

+ εi

(3.2)

In particular, if the above hypothesis is correct, β2 should be negative.

16

Carrington (1993) introduced more diverse measures of local market conditions for pre-displacement
occupations and industry markets for workers, and he found that these local market conditions can also
explain some of the wage losses encountered by displaced workers.

68

Next, I compare the contribution of the skills transferability measure (i.e., a continuous
variable bounded between 0 and 1) to the dichotomous occupational change measure commonly
used in previous studies of occupational human capital. The dichotomous measure is equal to 1 if
an individual changes occupations and 0 otherwise. I hypothesize that the continuous measure of
occupational skills transferability will add information to the model and explain more of the
variation in earnings losses than the dichotomous measure. To test this hypothesis, I estimate the
following regressions:
’

ΔlnWi = β0 + δ1SAMEOCCi + γ Xi

+ εi

(3.3)
2

’

ΔlnWi = β0 + β1Transferi,lc + β2Transfer i,lc+ δ1SAMEOCCi + γ Xi

+ εi

(3.4)

According to the above hypothesis, one would expect that the inclusion of skills
2

transferability (Transferi,lc and Transfer i,lc) in Equation 3.4 would decrease the magnitude of
the coefficient on the dichotomous measure (SAMEOCCi) as compared to the estimate from
Equation 3.3. In addition, the adjusted R2 of Equation 3.3 should be lower than that of Equation
3.2.
The preceding analysis relies on the two key assumptions laid out at the beginning of this
section: (1) workers are exogenously displaced and (2) a random subset of workers is (randomly)
assigned to new jobs. If these assumptions do not hold, the magnitudes of the estimated
coefficients from the OLS regression might be misleading because of the selection bias. In
particular, the sample in this analysis only includes displaced workers who had become
reemployed by the time of the survey. Therefore, the estimated relationship between the skills
transferability across occupations and earnings losses in this sample may be quite different from
the relationship observed in the labor force as a whole, which includes both reemployed workers
69

and workers who remain unemployed. To account for possible selection bias, I will use two
methods. First, I use the sample restriction suggested in Neal’s (1995) study, and second, I will
use the Heckman selection correction method. The details and advantages and disadvantages of
each strategy are discussed along with the results.

4. Data
To examine the effect of the transferability of occupational human capital skills on
earnings losses following job displacement, I used the 2004, 2006, 2008, and 2010 Displaced
Workers Survey (DWS), which is a supplement to the January Current Population Survey
17

(CPS).

The DWS has been widely used in studies of displaced workers because of the

comprehensiveness of its questions about the incidence and cost of job displacement. In
particular, displaced workers in the DWS are identified based on a question in the CPS that asks,
“During the last three calendar years, did you lose a job or leave one because: 1) the plant or
company is closed or moved, 2) your position or shift is abolished, or 3) there is insufficient
work?” If the answer from the respondent is “yes,” they will be asked a series of questions
regarding the lost job and current job (if they have been reemployed during the current
survey).

18

Since each survey year collects information on job loss that includes three years of

job displacement prior to the survey date, all four surveys provide information about workers
who lost their jobs from 2001 (for the 2004 survey) to 2009 (for the 2010 survey).

17

The data is publicly available at Center for Economic and Policy Research. 2011. CPS Displaced
Worker Uniform Extracts, Version 1.02. Washington, DC (http://www.ceprdata.org/cps/dws_prog.php).
18
If more than one job was lost during the pre-displacement period, information was obtained from the
job held for the longest time.

70

One disadvantage of this dataset is that it is cross-sectional, which means that all predisplacement job information is gathered through retrospective questions in the DWS.
Retrospective data is subject to “recall bias.” In general, recall bias arises from the fact that
people usually forget things that happened in the past. Studies suggest that a person’s memory
decays exponentially with time and that people are more likely to remember more important
events from the past (Evans and Leighton, 1995). For the DWS, displaced workers are asked to
recall job history information over the past three years. This raises concerns that respondents
may either understate or overstate the effects of job displacement.

19

However, a major advantage

of this dataset, particularly when compared to other datasets such as the NLSY and PSID, is that
it is a very large dataset drawn on a random sample of over 60,000 households. It is weighted to
be representative of the U.S. population, and the measures in the dataset are straightforward
(Kletzer, 1998). In addition, the three-year interval of information collected on job displacement
in the DWS also provides the data to determine variations in the amount of time since
displacement at the time of interview (Carrington, 1990).
The DWS provides pre- and post-displacement 2- and 3-digit U.S. census occupation
codes and weekly earnings that I used in my regression analysis. Specifically, the pre- and postdisplacement 2- and 3-digit occupations are compared to attach the value of skills transferability.
The sample in this analysis includes individuals who are between the ages of 20 and 62 who
were reemployed at the time of the survey.

20

There were 27,540 displaced between 2001 and

2006. Of those displaced workers, 20,713 were excluded from this sample because they were not
19

For a more careful discussion about the recall bias of DWS, see Evans and Leighton (1995), and
Kletzer (1998).
20
The typical reason to exclude workers aged above 62 because these workers will, in general, be
qualified for the social security retirement payments and therefore might encounter the different set of
constraints in the labor marker than the younger workers (Podgursky and Swaim, 1987).

71

reemployed by the time of survey, or they did not supply data on the key variables such as preand post-displacement occupations. I also excluded displaced workers who served in the military
or were self-employed for either pre- or post-displacement jobs.

21

My final sample, therefore,

consists of 6,827 individuals.

5. Descriptive Statistics
Appendix Table D1 presents the summary statistics of the variables used in this analysis
for the full sample pooling across the 2004 to 2010 surveys. The first row of Appendix Table D1
shows the mean changes in real weekly earnings (in $2010) between pre- and post-displacement
jobs. On average, displaced workers earned 21 percent less when reemployed than in their predisplacement job. The 21 percent translates to around a $121 loss in weekly earnings. The typical
employee had worked for their former employer for five years, and 89.5 percent of workers in
the sample worked full-time at the lost job. Educational attainment is divided into five categories:
less than high school, high school, some college, college, and advanced degree. One third of the
workers in the sample had a high school diploma, while another third had completed some
college education. A majority of workers in this sample were white and relatively old (around 40
years old). Sixty percent of the sample received health insurance from their former employer and
43 percent qualified for unemployment insurance benefits. Appendix Table D2 presents
descriptive statistics of these displaced workers compared to that of the full workforce from the
CPS. To examine the implications of sample exclusion criteria, I compared the analysis sample
to the full sample of workers in the CPS. With the exception of average weekly earnings, the
workers in the displaced workers sample did not differ greatly from the full labor force. While
21

The reason to exclude self-employed workers, especially at the pre-displacement job is because it is
difficult to see how the idea of an exogenous displacement could be applied to self-employed workers.

72

the average weekly earnings at the current job of displaced workers were substantially lower for
displaced workers than the full workforce (i.e., $685.01 vs. $789.89), the differences in gender,
race, and age were quite small between the two samples.
Table 15 presents the proportion of reemployed workers who were displaced between
1991 and 2009 and found a different (3-digit) occupation from their previous job. Most displaced
workers who were reemployed at the survey date (from 1994 to 2010) could not find the same 3digit occupation in reemployment. Close to 70 percent of the reemployed workers ended up
working in a different occupation from their old job.
Next, Figure 2 displays the distribution of occupational skills transferability among
reemployed displaced workers. In particular, it shows that there is substantial variation in skills
transferability. The average skills transferability is around 75 percent, which means that
displaced workers, on average, found a job to which they could transfer 75 percent of their
previous occupational skills.
To determine the relationship between earnings losses and the similarity between preand post-displacement occupations, measured by skills transferability, Figure 3 presents the
average changes in the natural logarithm of real weekly earnings (in $2010) by the level of
Ormiston’s skills transferability at the 3-digit level. Although the mean loss resulting from job
displacement for the full sample is approximately 21 percent, Figure 3 shows that there is a wide
dispersion across different levels of skills transferability. It indicates a positive relationship
between occupational skills transferability (at the 3-digit level) and the decline of earnings losses.
In particular, it shows that higher skills transferability from the previous occupation to the new
occupation can help reduce earnings losses due to job displacement. For example, at a skills
transferability level between 0.2 and 0.3 points, the average earnings losses were approximately

73

26 percent; at skills transferability between 0.3 and 0.4, they are around 21 percent. In addition,
Figure 3 illustrates that as the skills transferability increased, the slope of earnings losses became
flatter, illustrating the possibility of a non-linear relationship between skills transferability and
earnings losses. Thus, Figure 3 provides descriptive evidence to support the hypotheses outlined
above, and they motivate the idea of assessing the importance of occupational human capital and
inter-occupational variation as a determinant of earnings losses as a result of job displacement.

6. Estimation Results
6.1 OLS Regression Results
While Figure 3 suggests that the ability to transfer occupational skills to a new job can
reduce earnings losses following job displacement, the relationship may change when I control
for observable characteristics. The empirical results based on Ormiston’s skills transferability at
the 3-digit level from the OLS regression shown in Table 16 highlight the effects of key
independent variables on earnings losses of displaced workers. The parameter of interest is the
22

coefficient on the transferability of occupational human capital (Transferi,lc).

Column (1) of Table 16 is the standard model of occupational human capital and earnings
losses, and it is similar to those used by prior studies. In this model, occupational human capital
is captured by a dichotomous variable indicating whether the displaced worker remained at the
same 3-digit occupation (SAMEOCCi). This estimated coefficient represents the effect of
occupation-specific human capital on earnings losses if workers lose all occupation-specific
skills when displaced and reemployed in a new occupation. Consistent with the previous studies,

22

The results in Table 15 are unweighted. I also apply the weighted regression using the CPS final
sampling weights, and the results do not differ importantly from those shown in Table 15.

74

the regression result suggests that, on average, staying at the same 3-digit occupation reduces
earnings losses by 10.50 percent, all else equal. In addition, staying at the same 3-digit industry
also reduces earnings loss by 8.30 percent, all else equal.

23

The expected earnings loss of

displaced workers who stay in their occupation and industry at the sample mean of worker
characteristics is approximately 8.04 percent; the expected earnings loss of workers who only
stay in their industry but change occupation is approximately 18.52 percent; the expected
earnings loss of workers who only stay in their occupation but change industry is approximately
16.38 percent, and the expected earnings loss of workers who are displaced from both occupation
and industry is approximately 26.86 percent. This implies that skills that are specific to
occupation and industry are important to the determination of earnings losses.
In column (2), I substitute the skills transferability (Transferi,lc)—the continuous
measure of occupational human capital which ranges from 0 to 1—for the dichotomous variable
of switching occupations (SAMEOCCi) to ascertain the effect of the skills transferability across
occupations. There is a very strong association between occupational skills transferability and
earnings losses. In particular, estimates suggest a 10 percentage points increase in skills
transferability is associated with a 3.6 percent reduction in real weekly earnings holding other
factors fixed, and the estimated coefficient is statistically significant at the 1 percent level. Also,
the magnitude of the coefficient on the dichotomous industry change variable is reduced to 6.40
percent. Thus, this finding indicates a positive and strong relationship between the decline of
earnings losses and occupational matches after job displacement. For example, consider the
estimated earnings losses of two displaced workers, both of whom worked as HR managers. One
23

2

The adjusted R and estimated coefficients of other variables in the model are also consistent with the
previous studies.

75

finds a new job as a sales manager, and the other becomes a baker. The skills transferability from
the HR manager position to the sales manager position is equal to 0.8661 (86.61 percent of the
HR skills transferred to the sales manager position), and the skills transferability from the HR
manager position to the baker occupation is equal to 0.1593 (15.93 percent of the HR skills
transferred to the baker occupation). According to these estimates, the expected earnings loss due
to the loss of skills transferability of the worker who was displaced from the HR manager and
reemployed as the sales manager was lower than the expected earnings loss of the other, who
switched from the HR manager position to the baker occupation, by 28.0 percentage points.
In addition to the effect of occupational skills transferability, there is also a strong
relationship between pre-displacement work history and subsequent earnings losses. For example,
the coefficient of firm tenure at the lost job is negative and statistically significant. This
highlights the importance of specialized skills that accumulate over time within a firm and are
lost when changing jobs. Specifically, two additional years of tenure at a previous firm, on
average, increased earnings losses by 1.60 percent, holding other variables fixed. This finding is
consistent with other studies of the importance of firm-specific human capital that have found
workers will lose all the skills specific to a particular firm when they change to a new employer
(Podgursky and Swaim, 1987; Carrington, 1990;Topel, 1990).
The coefficient on individual workers’ union membership at a pre-displacement job is
negative and statistically significant at 1 percent level, indicating that unionized workers suffer
larger earnings losses than the nonunionized workers. This is consistent with Khun and
Sweetman’s (1999) study, which also found a strong relationship between post-displacement
earnings losses and union status. The earnings losses of displaced workers who collected
unemployment insurance benefits were significantly greater than those who did not collect the

76

benefit by 4.50 percent, all else equal. Also, displaced workers whose UI benefit had already
been exhausted suffer greater earnings losses than those whose benefit was still available.
The coefficients of the remaining independent variables have the expected signs. For
example, displaced workers working full-time at the lost job were more likely to suffer greater
earnings decline by 31.40 percent relative to what they earned before they were displaced. The
earnings losses of workers who were informed in advance about their job displacement were not
statistically significant from those who were not informed in advance. Also, as mentioned above,
2

we use age (and age ) together with the education level as a proxy for the number of years of
2

market experience. The coefficients of both age and age are statistically significant at the 5
percent level, which implies that the number of years of general market experience, on average,
help lower the earnings losses.
As indicated in Figure 3, the relationship between the occupational skills transferability
and earnings losses might not be linear. Column (3), therefore, represents the results from
2

Equation 3.4, which includes the quadratic form of skills transferability (Transfer i,lc ). The
results show that both linear and quadratic terms of skills transferability are individually
statistically significant at the 5 percent level. The F-test for the significance of both linear and
quadratic forms of skills transferability indicates that these two variables are highly jointly
significant (F=15.84, p<0.0001). In addition, the quadratic term has a negative value, implying
that skills transferability reduces earnings losses at a decreasing rate.
Following Equation 3.3, column (4) includes both types of measures of occupational
2

human capital (i.e., Transferi,lc, Transfer i,lc and SAMEOCCi). The results show that when
2

skills transferability (Transferi,lc and Transfer i,lc) is included in the model together with the
77

dichotomous variable indicating switching occupations (SAMEOCCi), the coefficient of
SAMEOCCi drops from 0.105 to -0.004 and becomes statistically insignificant while the
magnitude of linear coefficient of skills transferability is still large and statistically significant at
2

the 5 percent level. Compared with the standard model in column (1), the adjusted R increases
from 7.78 percent to 8.49 percent in column (4), indicating that skills transferability captures a
slightly greater variation in earnings losses than the dichotomous variable of switching
occupations. The F-test for the significance of both linear and quadratic forms of skills
transferability also indicates that these two variables are highly jointly significant (F=10.34,
p<0.0001), which implies that both forms of skills transferability add information to the model.
In other words, compared with the result in column (3), adding the dichotomous variable
(SAMEOCCi) in column (4) does not add any information into the model (i.e., the adjusted R

2

does not change). This finding confirms the above hypothesis that skills transferability across
occupations is a better measure of occupational human capital than the dichotomous variable of
switching occupations. In other words, the effect of the dichotomous variable is captured by the
skills transferability so that measuring the occupational human capital in terms of the degree of
similarity between two occupations is more important than whether or not workers can find the
same occupations as their previous jobs.

6.2 Robustness Tests
The results from the previous section present a strong relationship between earnings
losses due to job displacement and occupational similarity as measured by skills transferability.
However, as mentioned earlier, OLS estimates may be subject to selection bias because the
78

process of reemployment is not random and wages are only observed for workers who are
reemployed in the data. The relationship between skills transferability and earnings losses in a
sample of displaced workers who are reemployed at the survey date might be quite different
from the relationship observed in the labor force as a whole. For example, those who were not
reemployed may have had different unobserved characteristics affecting their earnings from
those of reemployed group. If this is the case, the OLS method could produce biased and
inconsistent estimates.
One way to reduce the source of selection bias is to restrict the sample to workers who
were displaced at least one year before the survey date because those workers displaced in the
most recent years are less likely to be reemployed and hence report their weekly earnings (Neal,
1995). Table 17 presents the OLS results for this group of workers. Even though this restriction
lowered the sample size by approximately 37 percent, the estimated transferability effect is very
similar to those reported in Table 16. For example, column (2) shows that increasing the skills
transferability by 10 percentage points, on average, lowers earnings losses by 3.8 percent, all else
equal.
The more conventional correction for selection bias is to apply Heckman’s (1979) twostep procedure. The process of this procedure applied to this context is as follows. First, I
estimate the following probit regression of the probability of being reemployed at the survey date.
*

Empt i = β´Z + υi

(3.5)

*

where Empt i is the latent variable for being reemployed at the survey date for worker i. We can
*

observe workers i being reemployed if Empt i > 0 and hence can observe all his/her postdisplacement characteristics. Z is a vector of worker characteristics and work history. υi is the

79

error term of unobserved worker characteristics with mean zero and constant variance, and it is
assumed to be independent and identically distributed across individuals.
In general, Z should include all variables from the earnings losses regression and also
contain at least one additional variable that is excluded from the earnings losses regression. If Z
does not contain the excluded variable(s), then the identification of the model will rely
completely on the functional form assumptions. Therefore, to satisfy this condition, at least one
variable that affects the reemployment decision but does not directly affect earnings losses
should be included in Z. The excluded variable proposed in this analysis is a dichotomous
variable indicating whether the individual has a child under the age of two or five (ch02, ch05). I
hypothesize that displaced workers who have children under two years old (ch02) or under five
years old (ch05) might be less likely to seek reemployment conditioning on the other observed
worker characteristics, but it should not directly influence their earnings reductions. Childcare
responsibilities might increase the opportunity cost of getting a job and lower the probability of
being employed (Podgursky and Swaim, 1987; Carrington, 1990).
The second step of this procedure is an OLS regression of log difference in real weekly
earnings that includes the estimated coefficients from the probit regression (inverse of the Mills
ratio). The results of Heckman’s (1979) two-step procedure are reported in Table 18.
Column (1) presents Heckman’s two-step estimates without the excluded variable, and
column (2) presents the estimates with two additional excluded variables (ch02 and ch05). Note
that as the inverse of the Mills ratios are far from statistically significant, I can reject the
hypothesis of sample selection implying that the corrections for selection bias contribute little to
the results. The comparison of the corrected sample selection and OLS estimators shows that
most of the coefficients, including skills transferability, are not greatly affected by correction for

80

selection.

24

However, some other coefficients, such as the coefficient of industry switches are

sensitive to the estimation method. This implies that the effect of skills transferability on the
post-displacement earnings losses does not appear to be biased by selection into
reemployment.

25

Thus, the results corrected for selection bias also support the hypothesis that

displaced workers are compensated for occupation-specific skills, and workers who find
occupations that are more similar to their pre-displacement jobs will suffer less earnings losses
than those who find dissimilar occupations, thus confirming the importance of occupational
human capital as a major determinant of earnings losses.

6.3 Comparison of the Effect of Shaw’s and Ormiston’s Skills Transferability
This subsection discusses the results comparing Shaw’s and Ormiston’s skills
transferability at the 2-digit level. Similar to Table 16, column (1) of Table 19 shows the result of
standard model of occupational human capital and earnings losses using a dichotomous variable
of occupational switches at the 2-digit level. The coefficient of occupational switches is slightly
higher than that at the 3-digit level (i.e., the coefficient estimate of staying at the same 3-digit
occupation is 0.105 while the coefficient estimate of staying at the 2-digit occupation is 0.116).
In particular, on average, staying at the same 2-digit occupation lowers earnings losses by 11.6
percent, all else equal. Next, column (2) and (3) present the effect of Ormiston’s skills
transferability on earnings losses at the 2-digit level. In general, the results look very similar to
24

One of the reasons that the results do not change so much in this specification may be because of the
weakness of the instrument or the excluded variation (ch02, ch05) in the identification of the participation
bias term. Unfortunately, there does not appear to be another attractive alternative.
25
The other source of bias emanated from the omitted variables within reemployed displaced workers,
such as luck, innate ability, and motivation, from the reduced form regression (Equation 3.1). If the
worker who has more innate ability or motivation is the one to find a more similar post-displacement
occupation and also has lower earnings loss, then the estimated coefficient will be biased and inconsistent.

81

those at the 3-digit level. For example, if the 2-digit skills transferability increases by 10
percentage points, on average, it reduces real weekly earnings losses by 3.3 percent, holding
other factors fixed, and the estimated coefficient is statistically significant at the 1 percent level.
Column (4) presents the regression results based on Shaw’s skills transferability. It shows that if
the Shaw’s skills transferability increases by 10 percentage points, on average, it reduces real
weekly earnings losses by 1.2 percent, holding other factors fixed. As we can see, compared to
Ormiston’s skills transferability, the magnitude of Shaw’s skills transferability is approximately
3 times smaller than that of Ormiston’s coefficient. This probably could be explained by the
argument in Chapter 1 that the return to Shaw’s skills transferability is contaminated by other
social factors and by the market demand; therefore, the real return to Shaw’s skills transferability
would be lower. Column (5) shows the result of the non-linear relationship between earnings
losses and Shaw’s skills transferability. As we can see, the coefficients of both level and square
terms are statistically insignificant. However, the F-test for the significance of both linear and
quadratic forms of skills transferability indicates that these two variables are highly jointly
significant (F=6.85, p<0.001).
To summarize, Table 19 shows that the rates of return to both approaches of skills
transferability are statistically significant in the linear case although the magnitude of Shaw’s
skills transferability is lower than that of Ormiston’s skills transferability. However, while
Ormiston’s skills transferability indicates evidence of non-linear relationship between earnings
losses and skills transferability, Shaw’s skills transferability does not.

82

7. Conclusions and Implications for Future Research
This chapter highlights the role of occupational human capital in determining the
earnings losses after job displacement; in particular, the degree to which occupation-specific
skills are transferable across occupations depends on the similarity between two given
occupations. The key finding of this chapter is that occupational human capital, and more
specifically, skills transferability between occupations, is an important determinant of changes in
weekly earnings following displacement. In particular, the ordinary least square (OLS) estimates
based on Ormiston’s approach at the 3-digit level present that, on average, displaced workers
whose new jobs measure 10 percentage points closer to their previous job on the occupational
similarity measure experience approximately 3.6 percent less earnings losses ceteris paribus.
This implies that the degree of similarity between pre- and post-displacement occupations has a
strong effect on how displaced workers will suffer from earnings losses both in terms of
significance and the magnitude of the effect. In addition, the relationship between occupational
skills transferability and earnings losses are not linear: a higher degree of skills transferability
lowers earnings losses, but the decline attenuates as skills transferability increases. Moreover,
previous studies using a dichotomous variable indicating whether displaced workers find the
same occupation imply convexity (in some sense, an infinite amount of convexity) in the
relationship between earnings losses and occupational skills transferability. However, the results
presented in this chapter suggest that previous results are wrong. Instead, the relationship
between earnings losses and occupational skills transferability is concavity. In terms of the
comparison of Shaw’s and Ormiston’s skill transferability at the 2-digit level, The findings show
that the rates of return to both approaches of skills transferability are statistically significant in
the linear case, although the magnitude of Shaw’s skills transferability is lower than that of

83

Ormiston’s skills transferability. In addition, both Ormiston’s and Shaw’s skills transferability
indicate evidence of non-linear relationship between earnings losses and skills transferability.
This chapter’s findings on earnings losses due to job displacement have important
implications for researchers, policy makers, and the public. For future research, it sheds light on
the role of occupational human capital in determining earnings losses after job displacement. In
fact, the results show that displaced workers also receive compensation that is not completely
general or specific to a particular firm, but rather specific to occupation. The findings from this
chapter highlight the importance of occupational specific human capital in the employment
relationship; losing these types of skills might result in substantial losses to the displaced
workers’ earnings. In addition, previous research on this topic has used a dichotomous variable
indicating whether displaced workers remained in the same occupation as their pre-displacement
job and ignored the fact that occupational skills could be (partially) transferable across
occupations.
However, using the continuous variable of skills transferability, the results from this
chapter indicate that if displaced workers could find occupations that are very similar, if not
exactly the same as their old jobs, they can transfer most of their previous occupational skills to
the new job and their earnings losses would be reduced. Moreover, the findings indicate that the
measure of skills transferability is a better measure of occupational human capital than the
dichotomous variable of switching occupations. Thus, those performing future research
regarding occupational human capital should consider applying this measure to other outcome
variables as well. For example, this chapter only examines the effect of occupational skill
transferability on the loss of earnings between pre- and post-displacement occupations and does
not consider the effect on other types of benefits. Thus, it might be interesting for future research

84

to investigate the effect of occupational human capital, especially skills transferability, on total
compensation, which also includes other benefits such as health insurance and pensions
(Podgursky and Swaim, 1987).
From a policy perspective, these findings provide valuable guidance for the development
of government programs to help displaced workers. The government has spent an enormous
amount on helping this group of workers. For example, the 1988 Economic Dislocation and
Worker Adjustment Assistance Act provides one billion dollars annually for retraining displaced
workers (Carrington and Zaman, 1994). In addition, the amount of money the unemployment
insurance benefit (UI) allocates to displaced workers each year is tremendously high, around $76
billion in 2009 and $58 billion in 2010. The government has also promoted various new and
expensive programs including job-search assistance and job counseling programs to help eligible
displaced workers find new jobs. Specifically, the objective of this job-search assistance is to
help displaced workers return to work as soon as possible in order to increase their short-run
earnings. The results of this chapter point toward a program to retrain displaced workers while
maintaining their occupation-specific skills; in particular, the program would assist those who
work in occupations that do not have many similar counterparts. To assist those workers who
have a hard time finding a new occupation with high skills transferability, the training program
might also build new occupational skills so that workers could have occupations to choose from
by increasing skills transferability. In addition, these results suggest that a displaced worker
assistance program would focus on matching displaced workers with jobs that are similar, if not
exactly the same, as their pre-displacement occupations to help reduce the short-run earnings
losses of displaced workers. Moreover, it also suggests that earnings losses are likely to be

85

greatest for those in occupations where there is a low chance of re-employment in a similar
occupation. Therefore, it lays out a basis for differential aid by occupation.

86

Table 15. Percentage of Reemployed Workers Who Change Occupations
Full Sample

1994

1996

1998

2000

2002

2004

2006

2008

2010

Different
Occupation
(percent)

69.74

71.12

71.26

67.43

69.35

65.52

63.88

65.60

61.72

0

.1

.2

.3

.4

Figure 2. The Fractions of Displaced Workers by Level of Skills Transferability

<0.2

0.2-0.4

0.4-0.6

Skills transferability (3-digit)

87

0.6-0.8

1

-.4

-.3

-.2

-.1

0

Figure 3. The Average Change in ln (real weekly earnings) by Level of Occupational Skills
Transferability between Pre- and Post-Displacement Occupations

<0.2

0.2-0.3 0.3-0.4 0.4-0.5 0.5-0.6 0.6-0.7 0.7-0.8 0.8-0.9
Skill transferability (3-digit)
Change in ln(real weekly earnings)

88

Fitted values

1

Table 16. The Earnings Losses Regression
Dependent Variable: Difference in log of real weekly earnings
($2010; OLS)

Independent Variable
Same Occupation (0/1)
Same Industry (0/1)

(1)
0.105
(6.28)
0.083
(4.73)

(2)

(3)

0.064
(3.62)
0.0036
(9.37)

0.072
(4.03)
0.0093
(3.12)

(4)
-0.004
(-0.09)
0.072
(4.06)
0.0091
(2.21)

-0.008
(-5.08)
-0.121
(-3.64)
0.014
(0.84)

-0.008
(-4.91)
-0.122
(-3.69)
0.011
(0.62)

-0.00004
(-1.99)
-0.008
(-4.91)
-0.123
(-3.70)
0.010
(0.60)

-0.00004
(-1.19)
-0.008
(-4.91)
-0.123
(-3.70)
0.010
(0.60)

-0.001
(-0.07)
-0.043
(-2.43)
-0.194
(-6.64)
-0.313
(-9.08)
0.015
(2.54)

-0.004
(-0.20)
-0.045
(-2.51)
-0.186
(-6.38)
-0.314
(-9.10)
0.014
(2.47)

-0.003
(-0.15)
-0.045
(-2.53)
-0.185
(-6.36)
-0.315
(-9.12)
0.015
(2.49)

-0.003
(-0.15)
-0.045
(-2.53)
-0.185
(-6.36)
-0.315
(-9.12)
0.015
(2.49)

-0.0002
(-3.02)
-0.089
(-4.35)
6,827

-0.0002
(-2.97)
-0.092
(-4.51)
6,827

-0.0002
(-2.98)
-0.094
(-4.58)
6,827

-0.0002
(-2.98)
-0.094
(-4.58)
6,827

0.0778

0.0845

0.0850

0.0849

Transferability*100
2

Transferability

Firm Tenure (at the lost job)
Union Status (at the lost job, 0/1)
Advanced Notice (0/1)
Health Insurance (at the lost job,
0/1)
UI Benefit (0/1)
UI Ended (0/1)
Full-time Job (at the lost job, 0/1)
Age
2

Age

Female
N
2
Adj. R

Note: t-statistics are in the parentheses. Not included as independent variables are 10 major occupation
dummies, 10 major industry dummies, marital status, US citizen dummy, race dummies (i.e., black,
white, Hispanic), dummy of geographic moves, unemployment rate at year of survey across state, year
of displacement, and 4 education levels.

89

Table 17. The Earnings Losses Regression Excluding Workers Who Were Displaced at Least
One Year before the Survey Date
Dependent Variable: Difference in log of real weekly earnings
($2010; OLS)

Independent Variable
Same Occupation (0/1)
Same Industry (0/1)

(1)
0.108
(4.75)
0.099
(3.99)

(2)

(3)

0.079
(3.25)
0.0038
(7.99)

0.087
(3.54)
0.0105
(3.27)

(4)
-0.015
(-0.28)
0.088
(3.55)
0.0096
(2.12)

-0.010
(-5.87)
-0.136
(-3.61)
-0.003
(-0.17)

-0.010
(-5.74)
-0.138
(-3.69)
-0.007
(-0.34)

-0.00005
(-2.09)
-0.010
(-5.73)
-0.138
(-3.70)
-0.007
(-0.35)

-0.00004
(-1.07)
-0.010
(-5.72)
-0.139
(-3.70)
-0.007
(-0.36)

-0.006
(-0.27)
0.007
(0.28)
-0.206
(-7.03)
-0.295
(-7.93)
0.013
(1.83)

-0.009
(-0.38)
0.007
(0.28)
-0.196
(-6.71)
-0.297
(-8.02)
0.011
(1.68)

-0.008
(-0.35)
0.007
(0.29)
-0.194
(-6.66)
-0.299
(-8.06)
0.012
(1.69)

-0.008
(-0.35)
0.007
(0.29)
-0.194
(-6.66)
-0.299
(-8.06)
0.012
(1.69)

-0.0002
(-2.45)
-0.093
(-4.10)
4,308

-0.0002
(-2.31)
-0.094
(-4.17)
4,308

-0.0002
(-2.31)
-0.096
(-4.26)
4,308

-0.0002
(-2.31)
-0.096
(-4.25)
4,308

0.0882

0.0969

0.0976

0.0974

Transferability*100
2

Transferability

Firm Tenure (at the lost job)
Union Status (at the lost job, 0/1)
Advanced Notice (0/1)
Health Insurance (at the lost job,
0/1)
UI Benefit (0/1)
UI Ended (0/1)
Full-time Job (at the lost job, 0/1)
Age
2

Age

Female
N
2
Adj. R

Note: t-statistics are in the parentheses. Not included as independent variables are 10 major occupation
dummies, 10 major industry dummies, marital status, US citizen dummy, race dummies (i.e., black,
white, Hispanic), dummy of geographic moves, unemployment rate at year of survey across state, year
of displacement, and 4 education levels.

90

Table 18. Heckman’s Two-Step Estimation of the Earnings Losses
Dependent Variable: Difference in log of real weekly
earnings ($2010)
No Excluded variable

Independent Variable
Same Industry (0/1)
Transferability*100
Firm Tenure (at the lost job)
Union Status (at the lost job,
0/1)
Advanced Notice (0/1)
Health Insurance (at the lost job,
0/1)
UI Benefit (0/1)
UI Ended (0/1)
Full-time Job (at the lost job,
0/1)
Age
2

Age

Female
Inverse Mill Ratio
N

Excluded variable: ch02, ch05

OLS

(1)
0.129
(1.77)
0.0038
(6.08)
-0.007
(-3.53)

(2)
0.089
(0.85)
0.0038
(4.99)
-0.007
(-3.14)

(3)
0.064
(3.62)
0.0036
(9.37)
-0.008
(-4.91)

-0.074
(-1.83)
0.034
(1.48)

-0.065
(-1.41)
0.048
(1.80)

-0.122
(-3.69)
0.011
(0.62)

-0.039
(-1.15)
-0.046
(-0.99)
-0.228
(-6.18)

-0.022
(-0.51)
-0.094
(-1.37)
-0.237
(-5.49)

-0.004
(-0.20)
-0.045
(-2.51)
-0.186
(-6.38)

-0.442
(-11.68)
0.015
(2.28)

-0.465
(-10.23)
0.017
(2.05)

-0.0002
(-2.86)
-0.087
(-3.19)
-0.159
(-0.69)
12,984

-0.0003
(-2.65)
-0.087
(-2.81)
0.026
(0.08)
10,239

-0.314
(-9.10)
0.014
(2.47)
0.0002
(-2.97)
-0.092
(-4.51)

6,827

Note: t-statistics are in the parentheses. Not included as independent variables are 10 major occupation
dummies, 10 major industry dummies, marital status, US citizen dummy, race dummies (i.e., black,
white, Hispanic), dummy of geographic moves, unemployment rate at year of survey across state, year
of displacement, and 4 education levels.

91

Table 19. Comparison of the Return to Shaw’s and Ormiston’s Skills Transferability

Independent Variable
Same Occupation (0/1)

Dependent Variable: Difference in log of real weekly earnings
($2010; OLS)
(1)
(2)
(3)
(4)
(5)

0.116
(7.27)
0.0033
(9.24)

Ormiston transfer*100
2
Ormiston transfer *100

0.0141
(3.32)
-0.0001
(-2.55)
0.0012
(7.29)

Shaw transfer*100

0.0125
(1.26)

2
Shaw transfer *100

Same Industry (0/1)
Firm Tenure (at the lost job)
Union Status (at the lost job, 0/1)
Advanced Notice (0/1)
Health Insurance (at the lost job, 0/1)
UI Benefit (0/1)
UI Ended (0/1)
Full-time Job (at the lost job, 0/1)
Age
Age

2

Female
N
2
Adj. R

0.087
(5.31)
-0.009
(-5.26)
-0.101
(-3.18)
0.012
(0.73)
-0.014
(-0.78)
-0.040
(-2.38)
-0.198
(-7.02)
-0.312
(-9.41)
0.014
(2.53)

0.077
(4.75)
-0.008
(-5.17)
-0.101
(-3.19)
0.011
(0.65)
-0.016
(-0.92)
-0.040
(-2.38)
-0.194
(-6.91)
-0.314
(-9.44)
0.013
(2.41)

0.082
(4.99)
-0.008
(-5.14)
-0.101
(-3.19)
0.010
(0.58)
-0.015
(-0.86)
-0.040
(-2.39)
-0.193
(-6.89)
-0.313
(-9.43)
0.013
(2.39)

0.087
(5.31)
-0.009
(-5.25)
-0.102
(-3.18)
0.012
(0.73)
-0.014
(-0.78)
-0.040
(-2.38)
-0.198
(-7.02)
-0.312
(-9.41)
0.014
(2.54)

-0.0001
(-1.14)
0.089
(5.38)
-0.009
(-5.24)
-0.102
(-3.21)
0.012
(0.73)
-0.013
(-0.77)
-0.040
(-2.37)
-0.198
(-7.01)
-0.312
(-9.39)
0.014
(2.57)

-0.0002
(-3.09)
-0.0805
(-4.19)
7,469

-0.0002
(-2.98)
-0.0817
(-4.26)
7,469

-0.0002
(-2.95)
-0.0844
(-4.37)
7,469

-0.0002
(-3.09)
-0.0806
(-4.19)
7,469

-0.0002
(-3.12)
-0.0817
(-4.25)
7,469

0.0793

0.0829

0.0837

0.0794

0.0795

Note: t-statistics are in the parentheses. Not included as independent variables are 10 major occupation
dummies, 10 major industry dummies, marital status, US citizen dummy, race dummies (i.e., black,
white, Hispanic), dummy of geographic moves, unemployment rate at year of survey across state, year
of displacement, and 4 education levels.

92

Appendices

93

APPENDIX A
Appendix Table A1. Knowledge, Skill, and Ability Categories across Occupations as Defined by
O*NET
Knowledge
Administration and
Management
Biology
Building and Construction
Chemistry
Clerical
Communications and Media
Computers and Electronics
Customer and Personal
Service
Design
Economics and Accounting
Education and Training
Engineering and Technology
English Language
Fine Arts
Food Production
Foreign Language
Geography
History and Archeology
Law and Government
Mathematics
Mechanical
Medicine and Dentistry
Personnel and Human
Resources
Philosophy and Theology
Physics
Production and Processing
Psychology
Public Safety and Security
Sales and Marketing
Sociology and Anthropology
Telecommunications
Therapy and Counseling
Transportation

Skill

Ability

Active Learning
Active listening
Complex Problem Solving
Coordination
Critical Thinking
Equipment Maintenance
Equipment Selection

Arm-hand Steadiness
Auditory Attention
Category Flexibility
Control Precision
Deductive Reasoning
Depth Perception
Dynamic Flexibility

Installation
Instructing
Judgment and Decision
Making
Learning Strategies
Management of Financial
Resources
Management of Material
Resources
Management of Personnel
Resources
Mathematics
Monitoring
Negotiation
Operation and Control
Operation Monitoring
Operation Analysis
Persuasion
Programming

Dynamic Strength
Explosive Strength

Quality Control Analysis
Reading Comprehension
Repairing
Science
Service Orientation
Social Perceptiveness
Speaking
System Analysis
System Evaluation
Technology Design
Time Management

Memorization
Multilimb Coordination
Near Vision
Night Vision
Number Facility
Oral Comprehension
Oral Expression
Originality
Perceptual Speed
Peripheral Vision
Problem Sensitivity

94

Extent Flexibility
Far Vision
Finger Dexterity
Flexibility of Closure
Fluency of Ideas
Glare Sensitivity
Gross Body Coordination
Gross Body Equilibrium
Hearing Sensitivity
Inductive Reasoning
Information Ordering
Manual Dexterity
Mathematical Reasoning

Appendix Table A1. Knowledge, Skill, and Ability Categories across Occupations as Defined by
O*NET (Cont’d)
Knowledge

Skill
Troubleshooting
Writing

95

Ability
Rate Control
Reaction Time
Response Orientation
Selective Attention
Sound Localization
Spatial Orientation
Speech Clarity
Speech Recognition
Speed of Closure
Speed of Limb Movement
Stamina
Static Strength
Time Sharing
Trunk Strength
Visual Color
Discrimination
Visualization
Wrist-finger Speed
Written Comprehension
Written expression

APPENDIX B
Description of the Main Variables in the NLSY Dataset
In Chapter 2, the analysis focuses on the so-called “CPS” job (firm), which is defined as
the current or most recent employer at the date of interview. In each survey year, an individual
can have up to five different jobs (employers), and the one that is the current or most recent
would be assigned as a CPS job. The detailed job information for each CPS job, regardless of
whether it is a full- or part-time job, has been used to construct the key variables of this analysis.
The following explains in more detail the key variables used in this analysis.
1. Firm Tenure: Firm tenure is measured based on the actual start date to stop date at each
interview and added together between survey periods. For example, one individual starts
working for Firm A on June 1, 1980 and is interviewed on June 31, 1980. He is still working for
that firm on August 2, 1981, when he is interviewed again. He continues to work for the same
firm until April 5, 1982, when he changes jobs. He is interviewed again on August 10, 1982. A
cumulative firm tenure at Firm A in number of weeks could be constructed as follows:
Tenure 1= [Tenure from June 1, 1980 to June 31, 1980 at the 1980 interview]
Tenure 2 = Tenure 1 + [Tenure from July 1, 1980 to August 2, 1981 at the 1981 interview]
Tenure 3 = Tenure 1 + Tenure 2 + [Tenure from August 3, 1981 to April 5, 1982, the date that he
left this firm at the 1982]
where Tenure1 is the current firm tenure in the year 1980, Tenure 2 is the firm tenure in the year
1981, and Tenure 3 is the firm tenure in the year 1982.
Note that there is a possibility of double-counting employers. Up until 1998, the
employers were only tracked between two contiguous interview years; therefore, an individual
who works for a particular firm during the first year, leaves that firm next year, and then returns

96

to work at the same firm later after a year or more would be determined to work for a new firm
during the second tenure since the previous tenure would have been out of the tracking scopes.
As a result of double-counting employers, there is a possibility of firm tenure with a single firm
being calculated as tenure with two separate employers, and unfortunately there is no systematic
way to measure how often this event might occur. After 1998, however, the computer-based
interview system began to keep track of all employers so that it recognized when an individual
returned to the same employer that he/she left a number of years earlier. In addition, there is also
a gap within the reported start to stop date (a “between-job” gap) for a specific firm in which an
individual is not actively working for that employer, but does not consider himself completely
disassociated from that firm during these periods. Thus, the total tenure with that firm will not
account for these gaps because they occur before the individual has reported an actual stop date
for association with that firm, so these weeks of gaps would be considered as part of his tenure
for that firm.
2. Actual Labor Market Experience: This measures the total years of work experience in a
labor market accumulated as a full-time or part-time employee since the first interview up to the
time of the NLSY interview. One advantage of the NLSY is that it is a longitudinal data that
reports the number of weeks worked for each year in the sample and obtains this information
retrospectively for the years preceding the sample. This allows us to construct the measure of
actual labor market experience, which is the key variable in this analysis. Specifically, I used the
question that asks for the “number of weeks worked since the last interview” of each individual.
Then, I added them together and divided this number by 52 to obtain the yearly actual labor
market experience for each individual.

97

3. Occupation and Industry Tenure: For the 3-digit level of occupation and industry tenure
measures, I used the CPS occupation/industry of individual, which is defined as “the current or
most recent occupation/industry that the individual has during the survey week.” Given the CPS
occupation/industry for each individual across year, I can construct both occupation and industry
tenures directly by comparing each individual’s occupation and industry year by year. If the
occupation/industry between two consecutive years were reported as being the same, then I
would assume that individual worked for that occupation/industry for the whole year. In that case,
I can add the number of weeks worked since the last interview (divided by 52) into the
occupation/industry tenure. Note that it is possible that an individual holds more than one
occupation/industry within the same firm during the time of the interviews, but only the
occupation/industry reported at the interview would be the current/most recent
occupation/industry. In addition, in some cases, the occupation and industry codes are missing
because individuals declined to give information about the type of work they performed. In other
cases, the codes are missing because errors were made during processing the descriptions of the
individuals’ jobs. These errors would create “invalid skips” in the occupation and industry data.
In addition, in many cases, the codes are missing because the individuals have never been asked
about certain characteristics of a particular job. This happened in some, but not all, cases where
jobs either involved less than 10 hours of work per week or less than 9 weeks of actual work.
These missing codes would create “valid skips” in the data (Neal, 1999).
4. Hourly Wage: Hourly wage is calculated if the individual indicates that he is paid on an
hourly basis. For those individuals who are not paid hourly, the coder calculates the hourly wage
based on the respondent’s time unit of pay and the usual hours worked per day or week.

98

5. Married: For simplicity, this dummy variable takes the value of “1” if an individual is
married and “0” otherwise (i.e., single, divorced, separated, widowed).
6. Regional dummies: these dummies include the North East, North Central, West, and South.
7. SMSA: This dummy variable takes the value of “1” if an individual is living in a SMSA and
“0” otherwise.
8. Major Occupation Dummies: Thirteen dummy variables represent the 1-digit 1980 census
occupations as follows:
-Executive, administrative, and managerial
-Professional specialty
-Technicians and related support
- Sales
-Administrative support, including clerical
-Private household services
-Protective services
-Services, except protective and household
-Farming, forestry, and fishing
-Precision production, craft, and repair
-Machine operators, assemblers, and inspectors
-Transportation and material moving equipment
-Handlers, equipment cleaners, helpers, and laborers
9. Major industry dummies: These 13 dummy variables represent the 1-digit 1980 census
industry as follows:
-Agriculture, forestry, and fisheries

99

-Mining
-Construction
-Manufacturing durable goods
-Transportation and communications
-Wholesale trade
-Retail trade
-Finance, insurance, and real estate
-Business and repair services
-Personal services
-Entertainment and recreation services
-Professional and related services
10. Dummy variable of old firm: A dummy variable equals “1” if the firm tenure is greater than
1 and “0” otherwise. The reason for including this variable is that the first year of work with an
employer typically requires more investment in job-related skills (Parent, 2000).

100

Appendix Table B1. Sample Construction
12,686

NLSY total sample
Deletions
Military sample
Female
Non-white
School returners
Leaving the survey forever after
the first interview

1,280
5,827
1,795
966

Remaining sample
11,406
5,579
3,784
2,818

35

2,783

Missing occupation more than
13 years

228

2,555

Being in the military at any time
Firm tenure at 1979 > 2 years

197
49

2,358
2,309

101

Appendix Table B2. Descriptive statistics of all variables from 1979-2000
Variable
Real hourly wage ($1979)
Age
Education
Percent married
Hours worked per week
Occupation tenure (3-digit)
Occupation tenure (2-digit)
Occupational investment (3-digit)
Occupational investment (2-digit, Shaw’s approach)
Occupational investment (2-digit, Ormiston’s
approach)
Industry tenure (3-digit)
Industry tenure (2-digit)
Firm tenure
Firm tenure >1 (0/1)
Number of firms changed
Labor market experience
SVP (0/1)
TQ (0/1)
Job zone (0/1)
Married (0/1)
Live in North East (0/1)
Live in North Central (0/1)
Live in South (0/1)
Live in SMSA (0/1)
Major Occupation
Executive and Managerial (0/1)
Professional (0/1)
Technicians (0/1)
Sales (0/1)
Administrative support (0/1)
Private household service (0/1)
Protective service (0/1)
Service, except protective (0/1)

102

Mean
5.70
27.43
12.22
0.48
42.29
2.45
3.19
3.29
3.30

Std. Dev.
3.61
5.82
2.31
0.50
11.48
2.46
3.11
2.80
3.12

4.27
3.40
3.96
3.41
0.66
3.71
8.12
5.14
1.53
2.57
0.48
0.18
0.28
0.32
0.75

3.42
3.49
3.85
3.91
0.47
2.57
5.33
1.71
1.10
0.97
0.50
0.39
0.45
0.47
0.43

0.09
0.06
0.03
0.08
0.07
0.00
0.01
0.09

0.28
0.23
0.17
0.28
0.25
0.02
0.10
0.29

Appendix Table B2. Descriptive statistics of all variables from 1979-2000 (Cont’d)
Variable
Farming, forestry, and fishing (0/1)
Craft, precision production (0/1)
Machine operators (0/1)
Transportation (0/1)
Handlers (0/1)
Major industry
Agriculture, forestry, and fisheries (0/1)
Mining (0/1)
Construction (0/1)
Manufacturing Durable goods (0/1)
Manufacturing Nondurable goods (0/1)
Transportation, communications (0/1)
Wholesale trade (0/1)
Retail trade (0/1)
Finance, insurance, and real estate (0/1)
Business and repair services (0/1)
Personal services (0/1)
Entertainment and recreation services (0/1)
Professional and related services (0/1)
Number of Observations
Number of Individuals

Mean
0.05
0.24
0.11
0.08
0.10
0.05
0.02
0.14
0.16
0.09
0.08
0.05
0.18
0.03
0.09
0.02
0.02
0.07

103

Std. Dev.
0.21
0.43
0.31
0.28
0.30
0.21
0.13
0.35
0.37
0.28
0.27
0.21
0.39
0.18
0.28
0.15
0.13
0.26
26,818
2,309

APPENDIX C
The Estimations of Occupational Skills Intensity
Besides occupational skills transferability, the skills intensity—the level of skills
acquisition of occupational human capital—is also another aspect of occupational human capital
neglected in the past literature. Most studies on the wage determination process assume there is
the same rate of acquisition of occupational tenure across various occupations although the level
of skills acquisition of occupational human capital is largely determined by the type of
occupation held. Some occupations provide relatively limited prospects for capital acquisition,
while some occupations provide extensive human capital acquisition opportunities. Thus, in
Chapter 2, I also control for the skills intensity at the current occupation in the wage equation. In
order to estimate this occupational skills intensity measure, I use the following variables.
The first variable is called the Standard Vocational Preparation (SVP) from the O*NET
dataset. The SVP provides information regarding the required level of knowledge and skills for a
particular occupation, rated on a nine-point scale.

26

Specifically, the definition of SVP is “the

amount of time required by a typical worker to learn the techniques, acquire the information, and
develop the facility needed for average performance in a specific job-worker situation” (Oswald
et al., 1999, pg. 4). It includes training acquired in a school, work, military, institutional, or
vocational environment, but excludes schooling without specific vocational content. Appendix
Table C1 represents the rating scale that is used to specify the value of SVP.
The average score of SVP over 467 (3-digit) occupations shown in Appendix Table C2 is
approximately 5.74. The average SVP of financial manager is around 7.66 (i.e., on average, a

26

The SVP is originally estimated based on the Standard Occupational Classification (SOC) code, and
then it is matched to the U.S. Census code for this analysis.

104

financial manager requires two to over four years to develop the skills in this occupation) while
the SVP of the construction laborer is 3.26 (i.e., on average, a construction laborer requires one
to three months to develop the skills in this occupation). This implies that the skills intensity of
financial manager is much higher than that of construction laborer.
The primary advantage of this measure is that the level of the SVP requirement for an
occupation can be directly linked to the level of specific occupational education and training
achieved by an individual. In addition, it allows individuals to prepare to be able to perform
effectively in an occupation (Oswald et al., 1999). On the other hand, its main disadvantage is
that it is highly correlated with education (i.e., more educated individuals might take a shorter
time period to acquire needed skills), and also it is, to some extent, an arbitrary measure based on
estimates made by researchers of the Bureau of Employment Security for the 4000 jobs of the
Dictionary of Occupational Titles (DOT) (Shaw, 1981).
The second variable of occupational skills intensity is from the direct measure of on-thejob training, labeled as “TQ,” available in the 1985 and 1993 PSID. It is calculated in fractions of
year based on the mean response of a representative national sample between the ages of 18 to
64. In particular, individuals, including both household heads and wives, respond to the question:
“On a job like yours, how long would it take the average new person to become fully trained and
qualified?” (Duncan and Hoffman, 1979, pg. 596) This question is designed to measure the
volume of training attached to the current job and acquired on the job.

27

The main advantage of

this variable is that it is formed by those individuals who actually work in the occupation, and it
appears less correlated with education (Shaw, 1981). However, the TQ could be an imprecise

27

The question uses the term “the average new person” rather than “you” in order to minimize the
responded training differences due to skills or experience unique to that individual (Duncan and Hoffman,
1979).

105

proxy of skills intensity because even the training in two jobs with equal training time periods
could have different degrees of intensity (Duncan and Hoffman, 1979). Appendix Table C2
demonstrates the distribution of the average TQ scores of 14 selected occupations. On average, it
takes 1.92 years for individuals to be fully trained and qualified in their current occupations.
Financial managers require approximate 3.64 years, while bakers require only 0.93 years to
become fully trained, and the automobile mechanics need around 1.33 years.
The third variable is called a JobZone, also from the O*NET dataset. It is a five-level
scale based on data from job incumbents and occupational experts (OEs) regarding the levels of
education, experience, and training needed for work in their occupations as displayed in
Appendix Table C3. Each occupation is assigned to one of five JobZones—groups of
occupations that require similar amounts of vocational preparation. Occupations in JobZone 1
require the least vocational preparation (i.e., the least skills intensity); occupations in JobZone 5
require the most vocational preparation (i.e., the highest skills intensity).
Appendix Table C2 demonstrates that the average JobZone score of financial managers is
4.50, which implies the considerable to extensive preparation needed for work in this occupation.
On the other hand, bank tellers are in JobZone 2, implying some preparation needed for work in
this occupation.
In some cases, the estimates of SVP, TQ, and JobZone are difficult to interpret. For
example, one might feel suspicious that the TQ measure for workers at amusement and
recreation facilities is higher than that of economists. Also, cashiers require 0.17 years while
barbers require about 1.50 years. Another concern is whether the training time required to
estimate these measures to become an average; for example, a locomotive engineer could begin

106

his career when he moves up from fireman, when he takes his initial job with the railroad, or at
some other time (Scoville, 1966). Therefore, these estimates require some careful interpretation.
Comparing the SVP, TQ, and JobZone, Appendix Table C4 provides the correlations
among these three variables. In general, the correlation among SVP, TQ, and JobZone are fairly
high and positive. For example, the correlation between SVP and TQ is 0.67, while the
correlation between TQ and JobZone is around 0.50. Thus, they are hypothesized to be a proxy
of the occupational skills intensity.

107

Appendix Table C1. The SVP Levels and Definition
SVP level
1

Approximate required
training time in years
0.00

2

0.04

3

0.17

4

0.38

5

0.75

6

1.50

7

3.00

8

7.00

9

12.00

108

Appendix Table C2. The Average SVP, TQ, and JobZone Scores for 14 Selected Occupations
Occupation (1980 Census code)
All occupations (Mean)
Financial manager (7)
manager and administrator (19)
Architects (43)
Librarian (164)
Economists(166)
cashier (276)
Bank teller (383)
Barbers (457)
Automobile mechanics (505)
Baker (687)
Rolling machine operators (707)
Construction laborers( 869)
Truck drivers (804, 805)
Attendants, amusement and
recreation facilities (459)

SVP
5.74
7.66
7.02
7.92
6.40
6.92
2.97
5.04
6.02
6.38
6.37
3.72
3.26
3.54

TQ
1.92
3.64
1.90
5.96
2.12
4.34
0.18
0.82
0.74
1.33
0.93
0.08
1.17
0.65

JobZone
2.87
4.50
3.80
4.50
5.00
4.50
1.50
2.00
3.00
3.00
2.00
3.00
1.50
-

3.58

4.57

1.78

Appendix Table C3. The JobZone Levels and Definition
JobZone

Preparation Level

1

Little or No preparation needed

2

Some preparation needed

3

Medium preparation needed

4

Considerable preparation needed

5

Extensive preparation needed

Appendix Table C4. Correlations among Three Skills Intensity Measures
Skills intensity measure
SVP
TQ
JobZone

SVP
1.00
0.67
0.69

109

TQ

JobZone

1.00
0.50

1.00

Appendix D
Appendix Table D1. Descriptive Statistics of Displaced Workers (2004–2010)
Variable
Difference in the natural logarithm of real weekly earnings (in
$2010)
Difference in the real weekly earnings (in $2010)
Weekly earning at the lost job
Weekly earning at the current job
Occupational skills transferability
Same Occupation (0/1)
Same Industry (0/1)
Female (0/1)
Age
Less than high school (0/1)
High school (0/1)
Some college degree (0/1)
College degree (0/1)
Advanced degree (0/1)
Years of firm tenure at the lost job
Union status at the lost job (0/1)
Married (0/1)
Advanced notice (0/1)
Health insurance at the lost job (0/1)
UI benefit (0/1)
UI ended (0/1)
US citizen (0/1)
Moved geographically (0/1)
Full-time at the lost job (0/1)
White (0/1)
Black (0/1)
Hispanic (0/1)

110

Mean

Std. Dev.

-0.21
-121.15
773.68
685.01
75.44
0.33
0.29
0.44
39.43
0.07
0.33
0.34
0.20
0.07
4.81
0.09
0.56
0.36
0.58
0.43
0.14
0.93
0.15
0.90
0.75
0.09
0.12

0.68
454.86
625.93
575.42
21.95
0.47
0.45
0.50
11.25
0.26
0.47
0.47
0.40
0.25
6.23
0.28
0.50
0.48
0.49
0.50
0.35
0.26
0.35
0.31
0.44
0.28
0.32

Appendix Table D1. Descriptive Statistics of Displaced Workers (2004–2010) (Cont’d)
Variable
Year of displacement =2009 (0/1)
Year of displacement =2008 (0/1)
Year of displacement =2007 (0/1)
Year of displacement =2006 (0/1)
Year of displacement =2005 (0/1)
Year of displacement =2004 (0/1)
Year of displacement =2003 (0/1)
Year of displacement =2002 (0/1)
Year of displacement =2001(0/1)
State unemployment rate
Major Occupation
Management (0/1)
Professional (0/1)
Service (0/1)
Sales (0/1)
Clerical (0/1)
Farming, fishing, and forestry (0/1)
Construction (0/1)
Installation, maintenance, and repair (0/1)
Production (0/1)
Transportation (0/1)
Major industry
Agriculture (0/1)
Mining (0/1)
Manufacturing (0/1)
Wholesale and retail trade (0/1)
Information (0/1)
Financial (0/1)
Professional and business services (0/1)
Educational and health services (0/1)
Leisure and hospitality (0/1)
Other services (0/1)

Mean
0.11
0.10
0.14
0.07
0.13
0.07
0.18
0.10
0.10
6.23
0.14
0.17
0.11
0.10
0.15
0.01
0.10
0.05
0.11
0.07

0.35
0.37
0.32
0.30
0.36
0.07
0.30
0.22
0.31
0.25

0.01
0.01
0.20
0.15
0.04
0.08
0.13
0.11
0.08
0.03

111

Std. Dev.
0.32
0.30
0.35
0.25
0.34
0.26
0.38
0.30
0.30
2.31

0.08
0.09
0.40
0.35
0.19
0.27
0.34
0.31
0.27
0.18

Appendix Table D2. Descriptive Statistics for CPS and DWS (2004–2010)

Variable
Female (0/1)

CPS
Mean Std. Dev.
0.51
0.50

Weekly pay at the current job ($)
Age
Less than high school (0/1)
High school (0/1)
Some college (0/1)
College (0/1)
Advanced college (0/1)
Married (0/1)
White (0/1)
Black (0/1)

789.89
40.21
0.080
0.30
0.29
0.22
0.11
0.59
0.74
0.09

Number of workers

47,480

112

628.23
11.92
0.27
0.46
0.46
0.41
0.31
0.49
0.44
0.29

DWS
Mean Std. Dev.
0.44
0.50
685.01
39.43
0.07
0.33
0.34
0.20
0.07
0.56
0.75
0.09
6,827

575.42
11.25
0.26
0.47
0.47
0.40
0.25
0.50
0.44
0.28

References

113

REFERENCES
Allison, P. D. (2005). Fixed effects regression methods for longitudinal data using SAS. Cary,
NC: SAS Institute Inc.
Altonji, J., and Shakotko, R. (1987). Do wages rise with job seniority? Review of Economic
Studies. 54, 473–59.
Altonji, J., and Williams, W. (1997). Do wages rise with job seniority? A Reassessment.
Industrial & Labor Relations Review, 58, 370–397.
Abraham, K. G., and Farber, H. S. (1987). Job duration, seniority and earnings. American
Economic Review, 77, 278–297.
Baicker, K., Rehavi M. M. (2004). Policy watch: Trade adjustment assistance. Journal of
Economic Perspectives, 18, 239–255.
Bartel, A. P., and Borjas, G. J. (1981). Wage growth and job turnover: An empirical analysis.
in S. Rosen (Ed.)., Studies in labor markets. New York: National Bureau of Economic
Research.
Becker, G. S. (1964). Human capital. Chicago: University of Chicago Press.
Belman, D., Heywood, J. S., and Lund, J. (1997). Public sector earnings and the extent of
unionization.” Industrial and Labor Relations Review, 50, 610–628.
Belman, D., and Heywood, J. S. (2004). Public sector comparability: The role of earnings
dispersion. Public Finance Review, 32, 567–587.
Borjas, G. J. (1981). Mobility and lifetime wages. Industrial and Labor Relations Review, 34,
365–376.
Carrington, W. J. (1990). Specific human capital and worker displacement (Dissertation).
University of Chicago, Chicago.
Carrington, W. J. (1993). Wage losses for displaced workers: Is it really the firm that matters?
Journal of Human Resources, 28, 435–462.
Carrington, W. J., and Zaman, A. (1994). Interindustry variation in the costs of job displacement.
Journal of Labor Economics, 12, 243–275.
Cauchon, D. (2010, August). Federal workers earning double their private counterparts.
USA Today. Retrieved from http://www.usatoday.com/money/economy/income/2010-0810-1Afedpay10_ST_N.htm
Duncan, G. J., and Hoffman, S. (1979). On-the-job training and earnings differences by race
and sex. The Review of Economics and Statistics, 61, 594–560.

114

Evans, D. S., and Leighton, L. S. (1995). Retrospective bias in the displaced worker surveys.
Journal of Human Resources, 30, 386–396.
Goldsmith, H. and Veum J. R. (2002). Wages and the composition of experience. Southern
Economic Journal, 69, 429–433.
Heckman, J. (1979). Sample selection bias as specification error. Econometrica, 47, 153–161.
Jacobson, L. S., LaLonde, R., and Sullivan, D. (1993). Earnings losses of displaced workers.
American Economic Review, 83, 685–709.
Kambourov, G., and Manovskii, J. (2009). Occupational specificity of human capital.
International Economic Review, 50, 63–115.
Kletzer, L. G. (1998). Job displacement. Journal of Economic Perspectives, 12, 115–136.
Kletzer, L. G., and Fairlie, R. W. (2003). The long-term costs of job displacement for young
adult workers. Industrial and Labor Relations Review, 56, 682–698.
Kuhn, P. and Sweetman, A. (1998). Wage loss following displacement: The role of union
coverage. Industrial and Labor Relations Review, 51, 384–400.
Mincer, J. (1974). Schooling, experience, and earnings. New York, NY: Columbia University
Press.
Neal, D. (1995). Industry-specific human capital: Evidence from displaced workers. Journal
of Labor Economics, 13, 653–677.
Neal, D. (1999). The complexity of job mobility among young men. Journal of Labor
Economics, 17, 237–261.
Neal, D. (1998). The link between ability and specialization. Journal of Human Resources, 33,
173–200.
Ormiston, R.A. (2006). The role of occupation-specific human capital in economic analysis.
(Dissertation). Michigan State University, East Lansing.
Oswald, F., Campbell, J., McCloy, R., Rivkin, D., and Lewis, P. (1999). Stratifying
occupational units by specific vocational preparation. O*NET Consortium, 1–26.
Parent, D. (2000). Industry-specific capital and the wage profile: Evidence from the national
longitudinal survey of youth and the panel study of income dynamics. Journal of Labor
Economics, 18, 306–323.
Podgursky, M., and Swaim, P. (1987). Job displacement and earnings loss: Evidence from the
displaced worker survey,” Industrial and Labor Relations Review, 41, 17–29.
Ruhm, C. J. (1991). Are workers permanently scarred by job displacements? American
Economic Review, 81, 319–324.
115

Schmieder, J. (2007). Returns to tenure: Is specific human capital acquired on the job? Working
Paper.
Gathmann, C., and Schönberg, U. (2010). How general is human capital? A task-based
approach. Journal of Labor Economics, 28, 1–49.
Scoville, J. E. (1966). Education and training requirements for occupations. The Review of
Economics and Statistics, 48, 387–394.
Shaw, K. L. (1981). The income effects of occupational change and the investment in
occupational skills (Dissertation). Harvard University.
Shaw, K. (1984). A formulation of the earnings function using the concept of occupational
investment. Journal of Human Resources, 19, 319–340.
Sullivan, P. (2006). Empirical evidence on occupation and industry specific human capital.
MPRA Paper No. 863, University Library of Munich, Germany.
Topel, R. (1990). Specific capital and unemployment: Measuring the costs and consequences of
job loss. Carnegie-Rochester Conference Series on Public Policy, 33, 181–214.
Topel, R. (1991). Specific capital, mobility, and wages: Wages rise with job seniority.
Journal of Political Economy, 99, 145–176.
White, R. (2009). Displacement-related earnings losses: Evaluating trade adjustment assistance
and wage insurance. Labor Studies Journal, 34, 219–234.
Willis, R. (1986). Wage determinants: A survey and reinterpretation of human capital
earnings functions. In O. Ashenfelter & R. Layard (Eds.), Handbook of Labor Economics
(pp. 525–602). Amsterdam: North-Holland Publishing Co.
Zangelidis, A. (2008). Occupational and industry specificity of human capital in the British
labor market. Scottish Journal of Political Economy, 55, 420–443.

116