by hui xiong - university of toronto t-space...hui xiong doctor of philosophy graduate department of...

Human Capital and Occupational Mobility

by

Hui Xiong

A thesis submitted in conformity with the requirementsfor the degree of Doctor of PhilosophyGraduate Department of Economics

University of Toronto

c© Copyright 2015 by Hui Xiong

Abstract

Human Capital and Occupational Mobility

Hui Xiong

Doctor of Philosophy

Graduate Department of Economics

University of Toronto

2015

This thesis contains three chapters. Chapter 1 uses underutilized SIPP to analyze occupa-

tional mobility in the U.S. from 1988 to 2003. I propose two additional mobility rates to do

robustness check, with careful treatment of the coding error. I classify all occupational switches

into three categories: horizontal, vertical and special. I find horizontal switches dominate the

other two in shares; aging decreases mobility while education’s role ambiguous. I examine the

interaction between occupational mobility and labor market status. Developing an algorithm

to extract nonemployment information from SIPP, I find most occupational switchers do not

experience nonemployment between jobs, similar to job-changing occupational stayers, but du-

ration variation is less for the former. As time goes by, the job-job mobility fraction declines

for both.

Chapter 2 utilizes SIPP to uncover additional occupational mobility facts. I find occupa-

tional behavior exhibits strong persistence among nonemployed as well as employed workers;

occupational switchers do not always switch to an occupation similar to previous one; and av-

erage length of transition duration between jobs varies with previous occupation. Motivated

by these facts I build a directed search model that includes aggregate and idiosyncratic shocks,

occupational human capital and search frictions. The model can account for the bulk of data. I

use the model to study relative importance of idiosyncratic vs. aggregate shocks, and barriers to

occupational mobility. I find idiosyncratic shocks are much more important than aggregate ones

in generating mobility; fixed mobility costs and search frictions constitute significant barriers

to mobility while transfer loss of occupational human capital does not.

Chapter 3 studies returns to occupational human capital assuming all occupations are dis-

tinct and occupational human capital is partially transferable. I name the associated tenure

variable “General Occupational Tenure” and propose an empirical Transfer Rate function that

ii

relates its transferable portion with occupation distance. Combining SIPP and DOT, I perform

a generalized wage regression under 1-, 2-, and 3-digit occupational classifications and find three

patterns: returns to General Occupational Tenure demonstrate great variation across occupa-

tions; fixed returns generally dominate variable returns; and they are negatively correlated.

Finally I generalize the result to a larger family of Transfer Rate functions.

iii

Acknowledgements

I am grateful to Gueorgui Kambourov and Diego Restuccia for their continuous support and

guidance. I thank Michelle Alexopoulos, Margarida Duarte, Burhanettin Kuruscu, Robert

McMillan, Shouyong Shi, Aloysius Siow, Ronald Wolthoff, Shintaro Yamaguchi and seminar

participants at the Canadian Economic Association 46th Annual Conference, Rimini Confer-

ence in Economics and Finance 2012, and University of Toronto for their helpful comments.

Financial support from Gueorgui Kambourov’s research fund and Shouyong Shi’s Social Sci-

ences and Humanities Research Council of Canada Research Grant is gratefully acknowledged.

All remaining errors are mine.

iv

Contents

1 The U.S. Occupational Mobility from 1988 to 2003:

Evidence from SIPP 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Overview of SIPP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.3 Key Concepts of Occupational Mobility . . . . . . . . . . . . . . . . . . . . . . . 6

1.3.1 Three Types of Occupational Switches . . . . . . . . . . . . . . . . . . . . 6

1.3.2 Measures of Occupational Mobility . . . . . . . . . . . . . . . . . . . . . . 7

1.4 Occupational Mobility in SIPP . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

1.4.1 Horizontal Switches Dominate Other Occupational Switches . . . . . . . . 10

1.4.2 Occupational Mobility at Different Times . . . . . . . . . . . . . . . . . . 10

1.4.3 Occupational Mobility in Different Age-Education Subgroups . . . . . . . 13

1.5 Nonemployment Intervened in Occupational Switches . . . . . . . . . . . . . . . 14

1.5.1 Nonemployment Fractions . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

1.5.2 Nonemployment Duration Distributions . . . . . . . . . . . . . . . . . . . 17

1.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2 A Directed Search Model of Occupational Mobility 38

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

2.2 Distance between Occupations and General Occupational Tenure . . . . . . . . . 42

2.2.1 Occupations and Occupational Classification under SIPP . . . . . . . . . 42

2.2.2 Distance between Occupations and General Occupational Tenure . . . . . 43

2.2.3 Occupation-Specific Returns . . . . . . . . . . . . . . . . . . . . . . . . . . 44

2.3 Occupational Mobility Patterns in the Labor Market . . . . . . . . . . . . . . . . 46

2.3.1 Occupational Mobility Distributions . . . . . . . . . . . . . . . . . . . . . 47

2.3.2 Transition Time Distributions . . . . . . . . . . . . . . . . . . . . . . . . . 48

2.3.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

2.4 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

2.4.1 Model Environment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

2.4.2 Value Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

2.5 Numerical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

v

2.5.1 Direct Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

2.5.2 Calibration Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

2.5.3 Calibration Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

2.5.4 Model Fit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

2.5.5 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

2.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

3 General Occupational Tenure and Its Returns 79

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

3.2 KM Wage Regression Revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

3.3 The Concept of General Occupational Tenure and Its Returns . . . . . . . . . . . 85

3.3.1 Occupation Distance and General Occupational Tenure . . . . . . . . . . 85

3.3.2 Estimating Occupation-Specific Returns to the General Occupational Tenure 88

3.4 Data and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

3.4.1 Survey of Income and Program Participation . . . . . . . . . . . . . . . . 89

3.4.2 Dictionary of Occupational Titles . . . . . . . . . . . . . . . . . . . . . . . 91

3.4.3 Principal Component Analysis . . . . . . . . . . . . . . . . . . . . . . . . 92

3.4.4 Occupational and Industrial Classifications . . . . . . . . . . . . . . . . . 95

3.5 Empirical Results of Returns to the General Occupational Tenure . . . . . . . . . 96

3.5.1 Regression Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

3.5.2 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

3.6 One Limiting Case and Convergence . . . . . . . . . . . . . . . . . . . . . . . . . 99

3.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

4 Appendices 112

4.1 Procedures for Constructing Tenure Variables . . . . . . . . . . . . . . . . . . . . 112

4.2 1990 Census of Population Occupation Classification System . . . . . . . . . . . 114

4.3 1980 Standard Occupational Classification System . . . . . . . . . . . . . . . . . 128

Bibliography 147

vi

Chapter 1

The U.S. Occupational Mobility

from 1988 to 2003:

Evidence from SIPP

1.1 Introduction

The returns to labor market experience have long been a research interest in macro and

labor economics. Earlier works (e.g. Mincer (1974)) attach importance to workers’ general

human capital, mainly education and overall labor market experience. Later writers stress

the significance of firm-specific (e.g. Topel (1991)) or industry-specific (e.g. Neal (1995)

and Parent (2000)) human capital. Recent studies show that human capital tends to be

occupation-specific. For instance, Kambourov and Manovskii (2009b) report that other

things being equal, a five-year occupational tenure is linked with a 12% to 20% increase in

wages; and if occupational tenure is taken into account, industry or job tenure is of relatively

little importance for explaining the wage level.

This new finding is particularly interesting in that the occupation-specific human cap-

ital is closely tied with other macroeconomic phenomena. For example, Kambourov and

Manovskii (2009a) calibrate a model to match the level and the change of occupational mo-

bility and it accounts quite well for the level and the change of within-group wage inequality.

In Kambourov (2009) and Ritter (forthcoming), occupation-specific human capital plays an

important role in the context of international trade.

Given that the occupation-specific human capital is important, and that most of the pa-

pers mentioned above stress the loss of human capital during the occupational switch process,

one question is why workers change occupations, and how they change. Unfortunately, these

issues are not very well addressed. Indeed, there is fairly small literature on the occupational

1

Chapter 1. The U.S. Occupational Mobility from 1988 to 2003:Evidence from SIPP2

mobility. Moreover, most existing models focus on young people’s occupation-shopping ac-

tivities (e.g. Neal (1999)), very few papers study prime age workers’ occupational mobility,

which is more relevant to the studies aforementioned. This is partly due to the insuffi-

cient empirical research on this important issue. Without learning key facts and patterns of

workers’ occupational mobility, there is little to say about its underlying mechanisms.

In the limited empirical papers, there seem to be two well known facts concerning the U.S.

occupational mobility. The first is that its level is considerably high. Vella and Moscarini

(2004) report an annual rate of 8% at the 3-digit level based on the CPS March files from

1976 to 2000. Kambourov and Manovskii (2008) use the PSID original and retrospective files

from 1968 to 1997 to calculate the occupational mobility per year: 13% at the 1-digit level,

15% at the 2-digit level, and 18% at the 3-digit level. Some differences exist between the two

studies. The former includes only individuals employed at both time t and time t-1 in the

sample, while the later also covers those who are unemployed in the previous period. More

importantly, although devoting much effort to solving the endogeneity problem, the former

does not take into account the coding error in occupation data, whereas the latter puts

tremendous effort in correcting coding errors using an extra retrospective file, which makes

its result more reliable. In this paper, I control the coding error carefully by verifying other

relevant variables and find a 3-digit annual mobility rate ranging from 14.26% to 15.22%,

which is close to that reported by Kambourov and Manovskii (2008). Moreover, I break

down all the occupational shifts into the ones with human capital destruction and the ones

without. It turns out the former category constitutes a dominant share of all occupational

switches.

The second fact is that the U.S. occupational mobility increases in the 1990’s than in

the 1960’s. Parrado et al. (2007) find that the fraction of workers who do not change

occupations declines from 38.0% in 1969-80 to 36.4% in 1981-92, and this is the case for

both male (from 35.6% to 34.0%) and female (from 50.0% to 42.5%). Their results are

obtained from PSID 1968-1992, and are not free from coding error. More reliable results

come again from Kambourov and Manovskii (2008). They report a significant increase in

the U.S. occupational mobility in late 1990’s than in late 1960’s, from 10% to 15% at the

1-digit level, from 12% to 17% at the 2-digit level, and from 16% to 20% at the 3-digit level.

Based on my data and sampling period, I find that the 3-digit yearly mobility rate rises from

10% in early 1990’s to 18% in late 1990’s and then drops gradually to 13% in early 2000’s.

Despite the above consensus, there are still certain key facts not very clear. One in-

teresting issue is the relationship between occupational mobility and labor market status.

Nonemployment (unemployment and/or out of labor force) seems important in occupational

mobility studies. Markey and Parks (1989), using 1987 CPS, report that the occupational

mobility during 1986-87 is 9.9%, among which 12.5% occupational changes are involuntary,


and the median unemployment spell for switchers (25 and older) is 7.5 weeks. Ideally data

can tell us the pattern of occupational switches: job-job or job-unemployment-job; if the

latter, how long the unemployment spell would be.

The data I use is the Survey of Income and Program Participation (SIPP), which is largely

underutilized by economists. SIPP is a large panel data set provided by the U.S. Census

Bureau. It has several exclusive advantages over other panel data sets. Specifically, it con-

tains rich labor market data (e.g. primary and secondary employers/industries/occupations,

starting/ending date of a job, weekly/monthly labor market status, etc.). In addition, it has

a high interview frequency design (every 4 months). Given SIPP records two occupations for

each worker in each period, I propose and calculate two extended versions of occupational

mobility rate to check the robustness of my findings. Moreover, SIPP’s high frequency of

interviewing and recording enables me to examine, in addition to conventional annual mo-

bility rate, other mobility rates of shorter time intervals, e.g. monthly mobility rate and

4-month mobility rate. Furthermore, I take advantage of SIPP’s large sample to look at mo-

bility differences across age-education subgroups. It is found that as age increases, workers’

occupational mobility declines. However, the education’s impact is ambiguous.

SIPP’s two features are also very helpful for me to investigate the nonemployment-related

questions mentioned above. These features enable me to observe a worker’s occupation affil-

iation and labor market activity details within a year. Specifically, I investigate the nonem-

ployment duration between two jobs for both occupational switchers and stayers who change

jobs. Though available, the duration information is not easy to extract. I develop a sophis-

ticated algorithm to obtain the nonemployment duration distributions in the units of as fine

as weeks. I find that most occupational switchers do not experience nonemployment between

jobs, very similar to job changers without involving an occupational switch. However, the

duration variation is less in the former group than in the latter group. And as time goes by,

the fraction of job-job mobility decreases for both groups.

The rest of paper is organized as follows: Section 1.2 introduces some background in-

formation of SIPP; Section 1.3 discusses the paper’s key concepts concerning occupational

mobility; Section 1.4 computes and analyzes various occupational mobility rates; Section 1.5

discusses the nonemployment during occupational switches; and I conclude in section 1.6.

1.2 Overview of SIPP

SIPP is designed to collect detailed information on income, employment, and participation in

the various government transfer programs of the U.S. civilian noninstitutionalized population

who are at least 15 years old. Using a two-stage complex sampling method, SIPP selects

a nationally representative sample of households. Once a sample is chosen, SIPP tracks all


the sample members (even if they move) and interviews them and the individuals who live

with them every 4 months. SIPP is administered in panels, and each panel consists of a

new sample. Within a SIPP panel, all the sample members are interviewed every 4 months;

each round of interview is called a wave. The whole sample is divided into 4 similar size

subsamples; each of them is called a rotation group. In each month, only one rotation group

is interviewed, with the information collected regarding the previous 4 months. The month

when the interview is held is called the interview month, whereas the months on which

the information is gathered are called the reference months. Therefore, in a given wave

the interview month and the reference months vary chronologically for different rotation

groups. Table 1.1 uses the 1996 Panel as an example to demonstrate the concepts of wave,

rotation group and reference month. Initially SIPP plans on starting a new panel of some

20,000 households each year and continuing a panel for 8 waves, or 32 months, but the actual

sample size, the starting time and the panel duration vary due to budget constraint and other

factors. There are 14 panels so far with the first one the 1984 Panel and the latest one the

2008 Panel. The number of Wave 1 eligible households varies from 12,425 (the 1986 Panel)

to 44,200 (the 2004 Panel), and the panel duration varies from 3 waves (the 1989 Panel) to

15 waves (the 2008 Panel). SIPP changes significantly from the 1996 Panel on: it abandons

the time-overlapping panel design and increases the panel size as a remedy; it introduces

computer-assisted interviewing to improve data consistency; it modifies variable names and

variable attributes drastically; and it reforms data editing and imputation procedures. For

convenience, I refer to the panels after the 1996 redesign (from the 1996 Panel onwards) as

new panels and the previous panels old panels. SIPP offers 3 kinds of public use files: core

wave files, topical module files, and longitudinal research files. Core wave files only contain

information on a given wave and are released when that wave is finished. Aside from the core

questions asked repetitively in all the waves for a panel, some supplemental questions are

asked in each interview. These questions are of different topics and the topic varies across

waves. The respondents’ answers to the topical questions are summarized in the topical

module files. The longitudinal research files contain information on all the waves of a panel

and are not available until the interviewing for that whole panel is completed. This paper

makes use of only longitudinal research files.

This paper uses data from 7 SIPP panels: Panels 1988, 1990, 1991, 1992, 1996, and 2001.1

To make my results comparable with Kambourov and Manovskii (2008)’s, I impose similar

sample restrictions on the data. That is, the male workers aged 23-612, who are not self-

1When the paper’s first draft is written, the 1988 Panel is the earliest panel available on SIPP’s officialweb site. And the then latest panel, the 2004 Panel, is still under editing. I don’t use the 1989 Panel, becauseit is very short (3 waves) and the Census Bureau never produces its longitudinal file.

2Individuals out of this age range may stay in the sample part of the time. For instance, an individualmay be younger than 23 at wave 1, but when he turns 23 in some later wave, he enters into the sample. The


or dual-employed and who do not work for the government. My sample restrictions differ

from Kambourov and Manovskii (2008)’s in one dimension: I do not require the sampled

individuals to be household heads. They employ the restriction simply because only the

household heads have occupation affiliation data available in the PSID, while in SIPP every

member has this information recorded. Another point worth noting is that, the individuals

who constitute my sample must be SIPP respondents who participate in the first wave

interview, or the original sample members. As mentioned above, SIPP interviews all the

original sample members and the individuals who live with them at the interview time. SIPP

drops the latter from the sample once they stop residing with the original sample members.

Therefore, these people enter and exit SIPP sample irregularly and their information is

discontinuous and incomplete. So I exclude them from my sample. Table 1.4 lists the

starting reference month, the ending reference month, the number of waves, and the number

of observations for each of the 7 samples I select.

Compared with the Current Population Survey (CPS), the longitudinal feature of SIPP

obviously makes it more appropriate to study workers’ occupation-shifting behavior over

time. The CPS has its sampled members 4 months in the survey, then 8 months out, and

4 months in again, and finally dropped permanently, which is by nature designed for cross-

sectional studies. Moreover, instead of tracing individuals, the CPS chooses to track ad-

dresses, which is a more serious problem for investigating individuals’ occupational changes.

There exist several other longitudinal surveys, but none of them is as suitable as SIPP con-

cerning my study purposes, in a variety of aspects. The Panel Study of Income and Dynamics

(PSID), started in 1968, provides much longer panel data than SIPP. But PSID has about

only 5,000 households tracked since then, a much smaller sample size than SIPP. Another

advantage of SIPP over the PSID is SIPP’s higher survey frequency: SIPP interviews sample

members every 4 months while the PSID does it annually, making SIPP suffer less recall er-

rors. Finally SIPP data are much richer and detailed than the PSID data (including the job

and labor force data interested in this paper), not only because of more frequent recording

but also due to its more comprehensive design. The National Longitudinal Survey of Youth

(NLSY) targets some particular year born cohorts, who are first interviewed as children or

young adults, and is therefore not as representative of the whole U.S. labor market as SIPP.

same rule applies to individuals who become older than 61 within the panel duration: they exit the sampleat the age of 61.


1.3 Key Concepts of Occupational Mobility

1.3.1 Three Types of Occupational Switches

As recommended in Kambourov and Manovskii (2008) and Vella and Moscarini (2004), I

focus on the 3-digit level occupational classification. The essential reason is that, compared

with its 1-digit and 2-digit level counterparts, 3-digit occupational classification is more

relevant to the conveyance or destruction of occupation-specific human capital during the

switch process, which I care about in the current research. And SIPP has been adopting

the 3-digit level occupational classification since it was started in 1984. SIPP’s occupational

classification system is almost the same as the 1990 Census of Population classification3,

which in turn is developed from the 1980 Standard Occupational Classification (SOC 1980).

The appendix lists the 1990 Census occupational classification system.

Depending on whether there exists destruction of occupation-specific human capital in

the switching process, I classify all the occupational switches into two broad categories, based

upon the textual description of every occupation title: no-loss switches and loss switches.

When talking about no-loss switches, I assume 100% occupation-specific human capital can

be transferred from the source occupation to the target one. Generally there are two types

of no-loss switches: (1) moving up or down the career ladder and (2) switches between

occupations requiring almost the same knowledge and skills. An example of the former type

is a promotion from a sales worker to a sales manager. On the other hand, if an individual

turns from an economist to an economics professor in college, I regard it as a second type

no-loss switch. Specifically, I refer to the first type no-loss switches as vertical switches, and

the second type no-loss switches as special switches. At last, all the other switches (loss

switches) are called horizontal switches.

I do the following to break down all the possible SIPP occupational switches into 3 classes:

vertical, special, and horizontal. Consider the vertical switches first. In practice, I restrict

this class of occupational switches only to adjacent up-moving changes, i.e. workers to

supervisors, supervisors to managers. Why do I rule out down-moving changes? Since high-

level positions generally demand more sophisticated skills than low-level ones, if an individual

moves up his or her career ladder, all the knowledge accumulated at the low-level position

can serve as the foundation for learning the high-level position skills. However, if moving

down, much of the complex knowledge is no longer useful for the low-level position, which

generally requires only simple, practical and repetitive operations. In reality, a worker’s

3There are slight differences between SIPP’s classification and the 1990 Census classification. Specifically,2 occupations in the 1990 Census classification, 003 (legislators) and 016 (postmasters and mail superinten-dents), do not exist in SIPP’s classification; lawyers (178) and judges (179) are distinct occupations in the1990 Census classification, while they are combined into one occupation, lawyers and judges (178), in SIPP’sclassification.


reservation wage is going up as occupational tenure increases, and I observe far more up-

going movements than down-going ones.4 The reason I do not count jumping promotions

in (e.g. workers to managers directly) is that, it is not common in reality; and even if

it happens, because a worker mainly performs concrete tasks while the main content of a

manager’s work involves management. The required skills by the two kinds of jobs differ

considerably and do not overlap much. Of course, saving computational cost is another very

important consideration for introducing these two exclusions.

One good feature of the SOC 1980 and hence SIPP’s occupational classification, is that

in many occupation groups, supervisory positions are listed first, followed by the occupations

supervised. This special structure makes it easier for me to identify vertical switches. How-

ever, two groups do not have this desired structure: the managerial and professional specialty

occupations; and the technical, sales, and administrative support occupations. Therefore,

on the one hand I fully take advantage of this vertical design of SIPP’s classification, and

on the other I put more effort in finding vertical switches within those two groups, e.g. 204

(Dental hygienists) to 085 (Dentist).

In the spirit of SOC 2000, an improved version of SOC 1980, I regard apprentices and

assistants as occupations associated with occupation-specific human capital accumulation,

but not helpers and aides (too general knowledge). This implies that up-moving switches

involving apprentices and assistants are classified as vertical changes, while those related

with helpers and aides are included in horizontal switches.

I am very conservative in identifying special switches, again, to minimize subjectivity.

This class of changes mainly consists of two categories: (1) switches between research posi-

tions and their corresponding teaching positions, e.g. 166 (economists) and 119 (economics

teachers, postsecondary) and (2) switches between private household positions and their

corresponding service positions, e.g. 404 (cooks, private household) and 436 (cooks).5

Tables 1.2 and 1.3 list all the possible vertical and special switches under SIPP’s occu-

pational classification, respectively. I have identified 250 possible vertical changes and 44

possible special changes. All other switches, as long as not appearing in either of the two

tables, are classified as horizontal switches.

1.3.2 Measures of Occupational Mobility

Standard Definition of Occupational Mobility

Firstly I define occupational mobility in the same manner as Kambourov and Manovskii

(2008) do, that is, the proportion of currently employed workers who report a current occu-

4Some may not think these arguments convincing. However, as far as the broad and very broad definitionsof occupational change (see Subsection 1.3.2) are concerned, it is no longer potentially problematic.

5This is deemed as a flaw of SOC 1980, and SOC 2000 improves on it.


pation different from their most recently reported previous occupation. I call it the standard

definition of occupational mobility. Since SIPP records up to 2 occupations, primary and

secondary, for each sample member at any given time,6 I restrict my attention to the primary

occupation for this moment (under the standard definition).

Since the PSID’s interview interval is one year, the occupational mobility calculated

by Kambourov and Manovskii (2008) is annual mobility. However, SIPP interviews its

sample members every 4 months, and therefore I can compute the four-month occupational

mobility. For convenience, I call it wave occupational mobility. Moreover, old SIPP panels

record respondents’ occupation affiliation month by month, or 4 primary occupations and 4

secondary occupations in each wave (in contrast, new SIPP panels record occupations wave

by wave, or one primary occupation and one secondary occupation in each wave), which

implies that I can also calculate monthly mobility for old panels. In order to compare my

results with Kambourov and Manovskii (2008)’s, I need calculate yearly mobility. For new

panels, I compare current wave’s occupation with the occupation 3 waves before; if the

source occupation is not available (respondents unemployed, out of the sample, refusing to

answer, missing value, etc.), I move one wave backward rather than one year (or 3 waves)

backward, in order not to waste information. Similarly, for old panels, I compare current

month’s occupation with the occupation 12 months before; if the source occupation is not

available, I move one month backward rather than one year (or 12 months) backward. I

calculate wave mobility for old panels in the same spirit.

Coding error is a big concern when one tries to use survey data. For instance, Kam-

bourov and Manovskii (2008) control for the PSID coding error by the use of its Retro-

spective Occupation-Industry Supplemental Data Files, which unfortunately do not exist for

SIPP. Here I apply the approach proposed by Hill (1994). In particular, when I observe an

occupational switch (no matter it is horizontal, vertical, or special), I check whether there

is an associated change in employer, industry, weekly working hours, and hourly pay. Once

I observe one of the 4 changes takes place, I deem the occupational switch reliable and refer

to it as a backed switch. Otherwise I regard the occupational switch spurious, caused much

likely by the coding error. Table 1.5 lists the backing rates for the 3 types of occupational

switches as well as for the overall occupational switches for different panels.7 It can be seen

that all the backing rates are impressively high8, which demonstrates that the occupation

6More accurately, SIPP records up to 2 jobs, primary and secondary, and the jobs’ occupation affiliationsfor each respondent at any given time. The primary job either generates more income or has longer workinghours than the secondary job. But the decision on which job is primary and which job is secondary is subjectto an interviewer’s discretion.

7These backing rates are associated with the annual occupational mobility.8The backing rates for vertical switches are relatively low. Since promotions are very likely to take place

within a firm than across firms, I might not be able to observe anticipated changes in employer or in industry,or even in weekly working hours. The most possible change I can see should be an increase in hourly pay.


affiliation data in SIPP are considerably reliable. One possible reason is the dependent cod-

ing method where the coding staff have a respondent’s SIPP occupation history at hand

when coding, which SIPP adopts as early as with the 1986 Panel. Given the approach I take

to control for coding errors, the standard definition of occupational mobility would be: the

backed proportion of currently employed workers who report a current occupation different

from their most recently reported previous occupation.

Extensive Definitions of Occupational Mobility

The standard definition of occupational mobility obviously has its limitations. Suppose a

worker works in Occupation A initially, and switches to Occupation B temporarily, and then

switches back to Occupation A. According to my standard definition, there are 2 occupational

changes regarding this worker. However, in terms of the loss of occupation-specific human

capital, the second switch appears not destructive and might involve no loss at all. To address

this issue, I propose the broad definition of occupational mobility. Continue to focus on the

primary occupation, an occupation pool is constructed for each worker. In particular, all

the primary occupations in history (till the previous period) enter into this occupation pool.

As one can imagine, as time goes by, a worker’s occupation pool tends to expand. When

identifying the type of an occupational switch, I assume that no change supersedes vertical

change, which in turn supersedes special change, which finally supersedes horizontal change.

That is, examine the current primary occupation and one’s occupation pool, whenever I can

find an element exact the same as the current occupation, I conclude that this worker does

not change his occupation at the time being, even if some other element can form a vertical

pair, or a special pair, or a horizontal pair with the current occupation. Only when no

element can be found the same as the current occupation, do I start to search for an element

in the pool to constitute a vertical pair with the current occupation. Depending on whether

this endeavor succeeds, the process may end or proceed to the next round.

So far the information on the secondary occupation is not made use of. In the data, it is

not uncommon that a worker switches back and forth between the primary and secondary

occupations, which intuitively should cause no loss of occupation-specific human capital. I

extend the broad definition of occupational mobility to the very broad definition of occu-

pational mobility with the help of secondary occupation information. The basic idea is the

same as the broad definition of occupational mobility. The only difference is that, when

constructing one’s occupation pool, his secondary occupations in history are also included.

What the 3 definitions in common is that, when identifying the type of the occupa-

However, the information on hourly pay is not widely available in SIPP. For instance, in the 2001 Panel, onaverage less than 20% of the respondents report their hourly pay rates. In the calculation, I regard lackinginformation as unbacked.


tional change, I only investigate the primary occupation as far as the current occupation

side is concerned. One reason is that in SIPP the primary occupation is more important

than the secondary one. Another is that adding the secondary occupation to the current

occupation side would make the judgment rule unnecessarily complicated and hence increase

computational cost significantly.

When applying the broad and very broad definitions of occupational mobility, whether

an occupational switch is backed or not would be no longer relevant. Since one element

in the occupation pool can sometimes be linked with 2 jobs in history, when this element

happens to be the one side which forms a no-change pair, or a vertical or special switch pair

with the current primary occupation, there is no convincing way to tell which job supersedes

the other, and therefore it is difficult to find a reference point.

1.4 Occupational Mobility in SIPP

1.4.1 Horizontal Switches Dominate Other Occupational Switches

The first issue examined is the distribution of occupational mobility. Do horizontal, vertical,

and special switches always coexist? If yes, how important is each of them? Tables 1.6 to

1.8 show the shares of 3 types of occupational switches, under different definitions, for the 7

selected SIPP samples.9

The tables clearly show the relative share of each individual occupational switch type.

On average, horizontal switches account for more than 95% of all the occupational switches,

dominating the other two types. Vertical switches have a share around 3%, which is quite

small, and special switches 0.7%, which is trivial. This result is robust across all the 3

definitions of occupational mobility.10 Given the structure of occupational mobility, in the

rest of the paper, I focus mainly on the horizontal mobility, in addition to the overall mobility.

1.4.2 Occupational Mobility at Different Times

The panel-wide average occupational mobility rates provide us with cross-sectional informa-

tion. I am equally concerned with the occupational mobility in the time-series dimension.

That is, how does the mobility rate evolve as time goes by?

Since old and new panels differ greatly in many aspects, I apply different methods to

compute their occupational mobility rates. As mentioned before, a worker’s occupation

affiliation is a monthly variable in old panels, but a wave variable in new panels. This

9These shares are associated with the annual occupational mobility.10The conclusion also holds robustly when I vary the time interval, i.e., calculating compositional shares

based on the wave and monthly mobility.


implies that I can calculate yearly, wave, and monthly mobility rates for old panels, but only

yearly and wave mobility rates for new panels.

In general, SIPP provides enough information to calculate the mobility rate associated

with a given calendar month in the sample period. However, one should be aware that the

time concept is clearer for old panels than for new panels. Since the occupation affiliation is

a monthly variable in old panels, to calculate, for instance, the annual rate, one needs only to

look at the occupational information in a given calendar month and 12 months before (the

corresponding sequential month numbers would differ across rotation groups). But when

the occupation affiliation is a wave variable as it is in new panels, it is ambiguous in what

exact occupation a worker works in a given month, since the time distributions of the two

occupations recorded over a given wave are not well documented, and moreover, SIPP might

drop some worked occupations and record only two occupations in a wave for new panels.

Therefore, for old panels, a straightforward approach is used to compute the occupational

mobility for a given month. Specifically, given a calendar month, I map it to the sequential

month numbers for different rotation groups individually, and then calculate the mobility

rate for each group, and finally average them out. For new panels, I assume implicitly that

the occupation affiliation points to the first reference month in each wave, which implies that

only one rotation group is used to compute the mobility rate for a given calendar month.

For instance, to calculate the mobility for December 1996, only Rotation Group 1 is used;

to calculate the mobility for January 1997, only Rotation Group 2 is used (please refer to

Table 1.1).

One important feature of old SIPP panels is that they have some time overlapping in the

panel duration. By this design, SIPP is essentially enlarging its sample size in the overlapping

period. To exploit this advantage of old panels, if possible, I average the mobility rate for a

given calendar month, using sample sizes as weights. New SIPP panels, nevertheless, don’t

have the overlapping design any longer, and hence I don’t average the results.

Figures 1.1 to 1.6 plot annual, wave, and monthly rate series according to different

mobility definitions for the overall mobility and the horizontal mobility. Note that several

time gaps exist in the yearly and wave mobility series11, due to the unavailability of reference

observations in calculation. For instance, the 1988 Panel ends with Dec. 1989 (last calendar

month) and thus the last month for which I can compute a mobility rate is Dec. 1989.

However, the 1990 Panel starts with Oct. 1989 (first calendar month). To compute its

annual mobility, I have to begin with its 13th observations for the first interviewed rotation

group, and these observations are associated with Oct. 1990. Therefore, a gap between Dec.

11The 3 gaps for the yearly mobility series are Dec. 1989 to Oct. 1990, Dec. 1995 to Dec. 1996, and Nov.1999 to Oct. 2001. The wave mobility series has 3 gaps as well, which are: Dec. 1989 to Feb. 1990, Dec.1995 to Apr. 1996, and Nov. 1999 to Feb. 2001.


1989 and Oct. 1990 in the annual mobility series has emerged. Similarly, a narrower gap,

from Dec. 1989 to Feb. 1990 appears in the wave mobility series. The monthly mobility

series, however, does not have a similar gap, simply because the first available month in this

series based on the 1990 Panel is Nov. 1989, which is prior to Dec. 1989. As mentioned above,

for the overlapping two months (Nov. and Dec. 1989), a weighted average is calculated as

the final result.

It is clear in Figures 1.1 and 1.2 that starting from the early 1990’s, the annual mobility

goes up gradually till the late 1990’s, which verifies Kambourov and Manovskii (2008)’s

finding, and then levels off (or even mildly declines) afterwards, generally consistent with

Vella and Moscarini (2004)’s result. Yet, there seems no overall trending for the whole sample

period.

As can be seen in the figures that there are a few obvious outliers for the annual and

wave mobility series,12 which, though do not significantly affect the mean values of the corre-

sponding mobility rates, increase individual series’ variances appreciably. With the outliers

excluded, Table 1.9 lists average mobility rates for various series. As anticipated, the hori-

zontal mobility rates are very close to their overall counterparts, since the horizontal switch

is the dominant type among all 3 occupational switches. No matter what time intervals are

considered, annual, wave, or monthly, the magnitudes of mobility rates are similar under the

three definitions, which shows that these numbers are quite robust. The annual mobility,

for instance, is around 15% concerning all three definitions, roughly consistent with Kam-

bourov and Manovskii (2008)’s finding (18%)13. And the wave mobility is about 7%. As the

time interval declines (i.e. from annual to wave, from wave to monthly), the mobility series’

variation increases nevertheless. Taking the overall mobility as an example and considering

the annual rate, the coefficients of variation for the standard, broad and very broad mobility

are all equal to 0.14. But for the wave rate, the 3 values turn to 0.14, 0.16, and 0.16, respec-

tively. Finally as far as the monthly rate is concerned, the results become 0.30, 0.31, and

0.34, respectively. The same pattern applies to the horizontal mobility as well. A possible

reason is, as the time interval declines, the random factors that may cancel out one another

to a large extent in the relatively long time spans (e.g. a year, or a wave), would start to

play noticeable roles, which results in the fact that the coefficient of variation for a monthly

rate is considerably larger than that of its annual or wave counterpart. Therefore, I would

12The annual mobility outliers are Dec. 1996 (2.6 to 3.0 times the average) and Apr. 1997 (2.6 to 2.8times the average), under all the 3 definitions for both overall and horizontal mobility. Similarly, there isone outlier, Aug. 1996 (4.5 to 6.0 times the average), for the wave mobility. A probable reason is that forthe 1996 Panel, the occupation affiliation data are relatively inaccurate for Rotation Group 1’s first 2 waves,making those mobility rates which use these 2 waves as references unusually high.

13Their sample period is from 1968 to 1997. And they report the mobility rate of 20% in the late 1990’s.But as my figures show, the mobility tends to decrease after that period, which would average down themobility level from 20% even if I were to use the PSID data.


concentrate on the annual and wave mobility henceforth.

Comparing the annual mobility and the wave mobility in Table 1.9, one finds that the

former is slightly more than twice but far less than 3 times the latter for both overall and

horizontal series. Since one year consists of 3 waves, this indicates that some workers keep

changing occupations after their first occupational switch, otherwise on average the annual

mobility would be roughly 3 times the wave mobility. This finding echoes Vella and Moscarini

(2004)’s result that a residual persistence exists in the occupational-matching process: some

less-lucky and poorly matched workers keep changing their occupations.

1.4.3 Occupational Mobility in Different Age-Education Subgroups

I break down each of the 7 selected SIPP samples into 6 age-education subgroups. Along the

age dimension, there are 3 categories: young-age group (23–35), middle-age group (36–48),

and old-age group (49–61). According to an individual’s education attainment, he falls either

in low-education group (high school and less) or in high-education group (some college and

college). Following the same method in Subsection 1.4.2, I compute various annual and wave

occupational mobility rates for every age-education subgroup, according to types (overall and

horizontal) and definitions (standard, broad, and very broad).

As in Subsection 1.4.2, the magnitudes of mobility rates are similar under the three

definitions, no matter what time interval is concerned, for a given age-education subgroup;

as the time interval decreases from annual to wave, the mobility series’ variation increases;

the horizontal mobility rates are very close to their corresponding overall mobility rates;

and the patterns of both the annual and wave series resemble that of their whole-sample

counterparts in Subsection 1.4.2, for all the 6 age-education subgroups: climbing up slowly

in the 1990’s, leveling out and declining gradually afterwards, showing no general trend in

the sample period.

First consider age’s impact on occupational mobility. Since human capital is largely

occupation-specific and occupational switches cause losses of occupational human capital

(horizontal switches dominate the other two types), as age increases and occupational hu-

man capital accumulates, the opportunity cost of changing one’s occupation will go up.

Therefore, the occupational mobility should decline with age. My results confirm this intu-

ition just like Kambourov and Manovskii (2008)’s do. As two examples, Figures 1.7 and 1.8

depict the annual overall mobility under standard definition and the wave horizontal mobil-

ity under very broad definition for all the 6 age-education subgroups, respectively. It is clear

that the occupational mobility indeed declines with age whatever education group a worker

belongs to. In each panel of Figures 1.7 and 1.8, an age group forms a stratum for itself

and separates one another coarsely. However, the demarcation between the middle-age and

old-age groups becomes ambiguous in the 2000’s for the low-education workers, which might


indicate that a high school graduate reaches the peak of his learning curve earlier nowa-

days than in the past, perhaps because the high school education is increasingly general

and thus decreasingly helpful in terms of building a worker’s occupational human capital.

Another pattern is that the within-group variation increases with age. For instance, the

coefficients of variation for the young-age, middle-age, and old-age groups in Figure 1.8’s

top panel (low-education group) are 0.20, 0.27, and 0.43, respectively, and for the bottom

panel (high-education group), 0.20, 0.32, and 0.49, respectively. This indicates that the

young-age workers’ occupation-switching behavior is more uniform across time than other

two age groups’. It could be the case that young-age workers are mainly influenced by the

occupational matching process, while middle- and old-age workers are affected more by the

macroeconomic conditions (e.g. occupational shocks). The above two patterns are common

in all the mobility series calculated in this subsection.

I continue by investigating the influence of education attainment on the occupational

mobility. Different from Kambourov and Manovskii (2008) who uncover that the college

educated workers exhibit lower occupational mobility than the less-educated, I find no simple

patterns in this regard. In particular, for middle-age workers, a college-education lowers one’s

occupational mobility; whereas for old-age workers, a college-education plays an exactly

opposite role. For the young-age group, however, the evidence is mixed. For instance, in

Figure 1.7 Group 114’s average mobility is 19.92%, less than that of Group 215 (20.31%);

conversely, in Figure 1.8 Group 1’s average mobility is 7.87%, greater than that of Group 2

(7.60%). My finding appears more relevant to that of Vella and Moscarini (2004), who claim

that the college effect is ambiguous.

1.5 Nonemployment Intervened in Occupational Switches

SIPP provides detailed information on workers’ labor market status in new panels. From

Panel 1996 on, individuals’ weekly and monthly labor market states are recorded. However,

panels prior to 1996 are weak in this regard. Hence, I put my focus on new panels in this

section.

1.5.1 Nonemployment Fractions

I examine how nonemployment (unemployment and/or out of labor force) relates to occu-

pational shifts in two steps. The first step is a natural extension of Section 1.4. Specifically,

I ask how many occupational switchers experience nonemployment between the source and

target occupations. As a comparison, I compute this fraction for the job changers who

14They are young-age low-education workers.15They are young-age high-education workers.


nonetheless do not switch their occupations. A very important consideration in calculat-

ing these statistics is the sample size. Different from the statistic of occupational mobility,

which is based on a considerably large sample containing all the original SIPP members

who satisfy my sample restriction conditions (see Table 1.4: Sample Size), the sample size

(denominator) shrinks dramatically for the statistics of nonemployment fraction. Take the

1996 Panel as an example, the average sample size for computing the annual mobility rates

is 6808. However, on average there are 1137 backed horizontal switchers, 39 backed vertical

switchers, and 4 backed special switchers, who constitute the samples based on which the

nonemployment fractions of horizontal, vertical, and special switchers, respectively, are cal-

culated. It is obviously not appropriate to compute the nonemployment fractions of vertical

and special switchers on the basis of above two very small samples. Since the nonemploy-

ment time distributions of horizontal and vertical switchers appear much different,16 it is

also not sensible to group these two distinct classes of occupational switchers together and

calculate the “overall” nonemployment fraction of occupational switchers. Therefore, I calcu-

late only horizontal switchers’ nonemployment fraction, together with the above-mentioned

nonemployment fraction of the job changers who do not switch their occupations. Again,

restricted by the relatively small sample, I need pool observations from all the 4 rotation

groups together in the computation, which implies that I am unable to calculate a statistic

that corresponds to a definite calendar month as in Section 1.4. Since in order to do that

especially for the new panels, rotation group-wise statistics are indispensable. But here I

have to combine different rotation groups to enlarge the sample size. So chronologically

speaking, all the nonemployment fractions are based on waves in this subsection (as is the

same case in the following subsection for the same reason), and caution should be exercised

in explaining the results whenever there involves a time dimension.

For new panels, in computing the standard annual mobility in Section 1.4, a worker’s

current occupation is compared with the one 3 waves before,17 so as to determine whether

he is an occupational switcher or not, and if yes, what type this switch is. Therefore, my

extended exercise would be to check whether the worker experiences any nonemployment

16See Subsection 1.5.2.17Essentially I am caring about whether any nonemployment is involved between the adjacent 2 occupations.

However, some occupational switches could take place in the intervening 2 waves. If this is the case, thenonemployment period associated with the intervening 2 waves should be irrelevant with the source and targetoccupations that constitute the annual mobility. Despite this weakness, I continue with the nonemploymentfraction based on annual mobility, for the following three reasons. (1) Annual mobility is the most often usedstatistic in the literature, and the annual mobility based nonemployment fraction is its natural extension. (2)The time span of 2 waves (or 8 months) is not problematically long so that further occupational changes arenot very likely to occur. (3) I use this statistic just to get a general picture of the nonemployment relatedto occupational shifts, and another more accurate measure is used in Subsection 1.5.2. However, to reduceinaccuracy, I restrict my attention to the standard definition in Section 1.5. Since under broad and verybroad definitions, it is very likely that the source occupation is in history and more than 3 waves apart andthus potentially more problematic.


during the intervening 2 waves. The SIPP variable I make use of is the Monthly Employment

Status Recode (MESR), and it has a finer classification than the conventional three-class

categorization (employed, unemployed, and out of labor force). MESR classifies a worker’s

monthly employment status into one of the following 8 classes.

1: with job entire month, worked all weeks.

2: with job entire month, missed one or more weeks but not because of a layoff.

3: with job entire month, missed one or more weeks because of a layoff.

4: with job part of month, but not because of a layoff or looking for work.

5: with job part of month, some time spent on layoff or looking for work.

6: no job in month, spent entire month on layoff or looking for work.

7: no job in month, spent part of month on layoff or looking for work.

8: no job in month, no time spent on layoff or looking for work.

Following Ryscavage (1989) I adopt 2 definitions of unemployment, a limited one (MESR

equal to 6 or 7) and a comprehensive one (MESR equal to 3, 5, 6, or 7). MESR equaling 8

would be classified as out of labor force. The judgment rule is straightforward: if the limited

definition of unemployment is taken, all the 8 MESR’s for the intervening 2 waves are

examined one by one (MESR is a monthly variable and subject to change across months); as

long as a value of 6 or 7 is observed, the worker is believed to have experienced unemployment

during the switch; by the same token, a value of 8 leads to the conclusion that the worker

leaves the labor force for some time; only when all the MESR’s take on a value other than

6, 7, or 8 do I conclude that there is no nonemployment intervened in the switching process.

Note that being unemployed and being out of labor force are not mutually exclusive, that

is, it could be the case that a worker experiences both unemployment and out of labor force

(subsequently) in the intervening 2 waves.

Table 1.10 lists various measures of nonemployment fraction for the backed horizontal

occupational switchers and for the job changers who nonetheless do not switch their occu-

pations. Three findings emerge. (1) No matter which of the 5 measures is considered, the

nonemployment fraction is very similar between the occupational switchers and the occupa-

tional stayers. (2) The majority of occupational switchers do not experience nonemployment

when they change occupations: they just move directly from the source occupation to the

target one. Likewise, more than 50% of the occupational stayers switch directly between

employers, without experiencing any unemployment or out of labor force period. This in-

dicates that on-the-job search is extremely important for both types of switching behavior.

(3) The fraction of occupational switchers who experience intervening nonemployment rises

in the 2001 Panel than in the 1996 Panel, as is also true for occupational stayers.


1.5.2 Nonemployment Duration Distributions

Due to the limitations of nonemployment fractions (see Footnote 17 for details), I proceed

by investigating the nonemployment time intervened in the two adjacent occupations, which

is undoubted a more direct and accurate statistic in examining the importance of nonem-

ployment to occupational changes. And SIPP is exceptionally suitable for this computation.

Compared with other often used panel data, SIPP’s high frequency of interviewing (every 4

months) and recording (in terms of occupation affiliation, every 4 months for new panels and

every month for old panels) obviously stands it out.18 More importantly, SIPP records work-

ers’ Weekly Employment Status Recode (WKESR, from which MESR is derived), which, on

the one hand makes SIPP users more confident in its labor force data’s reliability, and on the

other enables researchers to measure nonemployment time in the units of as fine as weeks.

Despite these desirable features, surprisingly, SIPP has never been used to study the nonem-

ployment time distributions during occupational switches. One possible reason is that SIPP

is not well known among researchers; another might be that the algorithm to compute this

statistic is somewhat involved.

The basic idea is to first identify the ending date for the source occupation (“date ending”

henceforth) and the starting date for the target occupation (“date starting” henceforth), and

then to examine each of the WKESR’s in between so as to calculate the total numbers of

unemployment weeks and out of labor force weeks.19 It follows that, for new panels, the

computation would be based on wave occupational changes since it is the two adjacent

occupations that are of interest, which is different from that in Subsection 1.5.1.

The information on a “date starting” or a “date ending” is not always available in SIPP:

take the 1996 Panel for instance, the average responding rates to “date starting” and “date

ending” questions are 85% and 4%20, respectively, for all the sample members. Even if it

is available, I need further check the information’s consistency. Recall that the occupation

affiliation is a wave variable in new panels. Consider the source occupation first and call its

corresponding wave the source wave. Denote the source wave’s first day “date A” and its last

day “date B”. Consistency requires that “date ending” fall in between “date A” and “date

B”, obviously. If this condition is violated, I regard “date ending” illegitimate and do not

18Although I cite unemployment spell statistics from studies based on the CPS in Section 1.1, as arguedbefore, the CPS is not suited for this research purpose, due to its non-longitudinal nature. Readers aresometimes prone to doubt those results’ reliability.

19SIPP records “date starting” and “date ending” data in terms of calendar time. However, WKESR’s areorganized according to their sequential month number (relative to the starting reference month, or Month 1,of a given panel) and sequential week number (1 to 5). It is necessary to analyze a “date starting” or a “dateending” and to transform it into the corresponding sequential month number and sequential week number,which requires some effort.

20Intuitively, the responding rate of “date ending” should be comparable to the occupational mobility rate(in this case, the wave mobility rate of about 7%). A 4% overall responding rate translates into a respondingrate of 57% for occupational switchers. The reason for this low availability rate, however, is not very clear.


use it in the subsequent computation. Then move on to the target occupation and similarly

call its corresponding wave the target wave. Denote the target wave’s last day “date D”.

Again, consistency would require that “date starting” fall in between “date A” and “date

D”. Likewise, its violation would lead to the ignorance of “date starting” subsequently. In

addition to the above two basic consistency conditions, there is another consistency condition:

“date ending” should be no later than “date starting”. If this condition does not hold, there

is, however, no convincing way to tell which of “date starting” and “date ending” is invalid.

Given that the availability rate is always higher for “date starting” than for “date ending”,

I just assume that violation of the third consistency condition results in the nullity of “date

ending” and the validity of “date starting”.

It follows that, whether a “date starting” or a “date ending” is usable will depend on

both its availability and its validity. According to the usability of “date starting” and “date

ending”, I break down all the occupational switches into 4 groups. In Group 1, both “date

starting” and “date ending” are usable. I start with “date starting” and move backwards

until “date ending” is reached,21 to examine each WKESR in between. In Group 2, only

“date starting” is available and valid. Thus I start by “date starting” and move backwards

until the first WKESR suggesting the status of employment is reached. By this it is implicitly

assumed that this first WKESR indicates the ending of the source occupation. However, if

no such WKESR exists, the investigation stops at “date A”. The approach is symmetric for

Group 3, in which only “date ending” is usable. I start by “date ending” and move forwards

until the first WKESR which suggests the status of employment is reached. If I cannot find

such a WKESR, I stop at “date D”. In Group 4, neither “date starting” nor “date ending”

can be used. Consider the time interval between “date A” and “date D”, it is anticipated

that a pattern of employment– nonemployment– employment should arise somewhere.22 The

principle therefore is to locate this structure first and then to identify the nonemployment

period in the middle. It does not matter where to start, “date A” or “date D”, and I choose

the former in my approach. Intuitively, the more information of the survey is made use of to

compute a statistic, the more confidence I have in the result. In this sense, I hope as many

as possible observations fall in Group 1 and as few as possible in Group 4. Fortunately the

samples behave nicely in this regard. For instance, the 1996 Panel’s sample has the following

composition: 33.81% for Group 1, 56.43% for Group 2, 4.67% for Group3, and 5.09% for

Group 4.

Like MESR, WKESR has a finer classification than the conventional three-class catego-

rization. WKESR classifies a worker’s weekly employment status into one of the following 5

21Equivalently, one can start with “date ending” and move forwards until “date starting” is reached. Bothmethods will yield the same result.

22It could be the case that this structure is preceded by some nonemployment time and/or followed bysome nonemployment time.


classes.

1: with job or business, working.

2: with job or business, absent without pay, but not on layoff.

3: with job or business, absent without pay, on layoff.

4: no job or business, looking for work or on layoff.

5: no job or business, not looking for work and not on layoff.

To be compatible with Ryscavage (1989), I also propose two definitions of unemployment

based on WKESR, a limited one (WKESR equal to 4) and a comprehensive one (WKESR

equal to 3 or 4). WKESR equaling 5 would be classified as out of labor force.

Tables 1.11 to 1.13 list the nonemployment time distributions for horizontal occupational

switchers, occupational stayers (job changers), and vertical occupational switchers, respec-

tively, based on the data of Wave 2, the 1996 Panel. Each provides a typical example of

its own kind. In particular, horizontal occupational switchers have a very similar nonem-

ployment time distribution to that of occupational stayers. No matter what measure is

considered, the majority of both do not experience any intervening nonemployment period

during the switching process. The feature is more salient as far as the out of labor force

duration is concerned. This verifies the finding in Subsection 1.5.1, but with a more rigorous

measure.23 On the other hand, the nonemployment time distribution for vertical occupa-

tional switchers appears very different: it is far less spread than that for the above two classes

of workers. Vertical occupational switchers tend to cluster around zero nonemployment and

some very limited number of medium-length nonemployment time spans. Because of this

significant difference and the very small sample size of vertical switchers, I choose to put this

group aside and focus only on horizontal switches.

To get a more general picture, I classify the nonemployment duration into 5 categories

according to its length: no interruption (zero week), short (less than a month, or 1-4 weeks),

medium (more than a month but less than a quarter, or 5-13 weeks), long (more than a

quarter but less than a year, or 14-52 weeks), and very long (more than a year, or 53+

weeks). Tables 1.14 and 1.15 show the average nonemployment time distributions under the

above five-group classification of horizontal switchers and occupational stayers (job chang-

ers) for the 1996 Panel and the 2001 Panel, respectively. Note that the two definitions of

unemployment yield very similar unemployment duration distributions, which makes the

two associated nonemployment duration distributions analogous as well. For both horizon-

tal switchers and occupational stayers (job changers), most of them do not experience any

nonemployment periods in the switching process and this feature is most pronounced for the

23Even when I compute these two distributions based on data later than Wave 2 (namely, occupationalswitchers and occupational stayers in Waves 3, 4, 5, etc.), although the upper bounds of support increase,the conclusion still holds qualitatively.


out of labor force duration distribution; the number of workers who experience a very long

interruption (53+ weeks) is trivial; and the remaining workers are distributed roughly evenly

in the other three interruption groups. Comparing the 1996 Panel and the 2001 Panel,24

it is observed that the number of workers falling in the no interruption group is declining,

while that of workers experiencing a long interruption time is rising considerably, for both

horizontal switchers and occupational stayers (job changers). This pattern is more salient

for the two nonemployment duration distributions.

At last, I calculate the mean nonemployment duration under different measures, associ-

ated with their coefficients of variation. As three examples, Figures 1.9 to 1.11 plot backed

horizontal switchers and occupational stayers (job changers)’ mean unemployment durations

(limited definition), mean out of labor force durations, and mean nonemployment durations

(comprehensive definition), respectively, together with their corresponding coefficients of

variation. Despite many similarities between horizontal switchers and occupational stayers

discussed above, the graphs show some interesting differences. It is clear that in most cases

horizontal switchers have a longer mean nonemployment duration than occupational stayers

do. However, the variation is always smaller for the former than for the latter. It could be

the case that many occupational switchers cannot afford a long nonemployment duration

for a desired job and are forced to change their occupation to make ends meet. Although

the time concept is vague in this section, one can still see a general rising trend in all the

figures: the mean interruption time is increasing for both groups of workers. My previous

finding, that the no interruption group is shrinking while the long interruption time group

expanding, naturally leads to this result.

1.6 Conclusion

This paper uses SIPP, an underutilized data set to analyze the occupational mobility in the

U.S. from 1988 to 2003. Exploiting SIPP’s detailed information on workers’ occupation, I

propose and calculate various extended versions of occupational mobility rate to do robust-

ness check, with careful treatment of the coding error. Unlike works that treat occupational

mobility homogeneously, I classify all occupational switches into three categories: horizontal,

vertical and special. Numerous mobility rates are computed according to different defini-

tions, categories, time intervals, and subgroups. I find that, in terms of shares, horizontal

switches dominate vertical and special ones at all times; that the mobility level and trend are

generally consistent with other empirical works; and that aging decreases the occupational

mobility while education’s role ambiguous. Moreover, I examine the interaction between

24Although it is warned earlier that one should be careful when comparing results across time in thissection, these two panels are separated quite apart chronologically (they even do not have overlapping time),and hence the panel-wise statistics can be compared meaningfully.


occupational mobility and labor market status, taking advantage of SIPP’s high interview

frequency and rich labor market information recording. I develop an algorithm to extract

nonemployment information between jobs from SIPP. I find that most occupational switch-

ers do not experience nonemployment between jobs, very similar to job changers without

involving an occupational switch, but the duration variation is less in the former group than

in the latter group. As time goes by, the employment-to-employment mobility fraction is

declining for both groups.

In the job turnover literature, two important indicators are (gross) mobility and net

mobility (one-half of sum of the absolute changes in employment shares of different estab-

lishments). They shed light on the mechanisms accounting for the occupational mobility

here as well. If the gross mobility is comparable with the net mobility, then it is the occupa-

tional shock that matters: occupations that receive good shocks expand and induce mainly

labor inflows, whereas occupations that receive bad shocks contract and induce mainly la-

bor outflows. On the other hand, if the gross mobility dominates the net mobility, then it

is the matching process that matters: there are workers entering into and exiting from an

occupation at the same time and the two effects cancel out each other a great deal.

Kambourov and Manovskii (2008)’s results show that both mechanisms above seem at

work. In the 1960’s, the gross mobility is 16% and the net one is 9%. In the 1990’s, the gross

mobility is 20% and the net one is 13%. In both cases, the former is greater than the latter,

but not by a significant amount (less than twice the latter). Therefore, a theoretical model

of prime age workers’ occupational switch needs to include occupation-level shocks and the

matching process. In addition, my findings in this article suggest that search also plays an

important role. On the one hand, on-the-job search seems to be a common practice as most

workers do not experience nonemployment between the source and target occupations. On

the other, the fact that mean nonemployment duration is on the rise implies that search

frictions become more serious than before.


Table 1.1: 1996 Panel: Rotation Groups, Waves, and Reference Months

ReferenceMonth

Rotation Group ReferenceMonth

Rotation Group

1 2 3 4 1 2 3 4

Dec-95 W1 1 Dec-97 W7 1Jan-96 W1 2 W1 1 Jan-98 W7 2 W7 1Feb-96 W1 3 W1 2 W1 1 Feb-98 W7 3 W7 2 W7 1Mar-96 W1 4 W1 3 W1 2 W1 1 Mar-98 W7 4 W7 3 W7 2 W7 1Apr-96 W2 1 W1 4 W1 3 W1 2 Apr-98 W8 1 W7 4 W7 3 W7 2May-96 W2 2 W2 1 W1 4 W1 3 May-98 W8 2 W8 1 W7 4 W7 3Jun-96 W2 3 W2 2 W2 1 W1 4 Jun-98 W8 3 W8 2 W8 1 W7 4Jul-96 W2 4 W2 3 W2 2 W2 1 Jul-98 W8 4 W8 3 W8 2 W8 1

Aug-96 W3 1 W2 4 W2 3 W2 2 Aug-98 W9 1 W8 4 W8 3 W8 2Sep-96 W3 2 W3 1 W2 4 W2 3 Sep-98 W9 2 W9 1 W8 4 W8 3Oct-96 W3 3 W3 2 W3 1 W2 4 Oct-98 W9 3 W9 2 W9 1 W8 4Nov-96 W3 4 W3 3 W3 2 W3 1 Nov-98 W9 4 W9 3 W9 2 W9 1Dec-96 W4 1 W3 4 W3 3 W3 2 Dec-98 W10 1 W9 4 W9 3 W9 2Jan-97 W4 2 W4 1 W3 4 W3 3 Jan-99 W10 2 W10 1 W9 4 W9 3Feb-97 W4 3 W4 2 W4 1 W3 4 Feb-99 W10 3 W10 2 W10 1 W9 4Mar-97 W4 4 W4 3 W4 2 W4 1 Mar-99 W10 4 W10 3 W10 2 W10 1Apr-97 W5 1 W4 4 W4 3 W4 2 Apr-99 W11 1 W10 4 W10 3 W10 2May-97 W5 2 W5 1 W4 4 W4 3 May-99 W11 2 W11 1 W10 4 W10 3Jun-97 W5 3 W5 2 W5 1 W4 4 Jun-99 W11 3 W11 2 W11 1 W10 4Jul-97 W5 4 W5 3 W5 2 W5 1 Jul-99 W11 4 W11 3 W11 2 W11 1

Aug-97 W6 1 W5 4 W5 3 W5 2 Aug-99 W12 1 W11 4 W11 3 W11 2Sep-97 W6 2 W6 1 W5 4 W5 3 Sep-99 W12 2 W12 1 W11 4 W11 3Oct-97 W6 3 W6 2 W6 1 W5 4 Oct-99 W12 3 W12 2 W12 1 W11 4Nov-97 W6 4 W6 3 W6 2 W6 1 Nov-99 W12 4 W12 3 W12 2 W12 1Dec-97 W6 4 W6 3 W6 2 Dec-99 W12 4 W12 3 W12 2Jan-98 W6 4 W6 3 Jan-00 W12 4 W12 3Feb-98 W6 4 Feb-00 W12 4

NOTES: The cell entry W1 1 represents Wave 1, Reference Month 1. For Rotation Group 1, the referencemonths for Wave 1 are Dec-95 through Mar-96. (Source: SIPP Users’ Guide, 3rd Ed., Table 2-2)


Table 1.2: Vertical Occupational Switches in SIPP

023–025: 007 337–344: 305 486–489: 485027: 008 337: 023 495, 496: 494

028–033: 009 348, 353: 306 498: 497034: 013 354–378: 307 505–549: 503035: 018 379: 303 506: 505106: 084 404: 433 563–565: 553204: 085 405: 448 564: 563207: 095 407: 448 567, 569: 554213: 055 413–415: 006 569: 567214: 056 416, 417: 413 575–577: 555215: 057 418–424: 414 576: 575218: 063 425–427: 415 579–584: 556223: 078 433: 017 585, 587: 557224: 073 434–444: 433 587: 585229: 064 439: 404, 436 614–617: 613234: 178 443: 435 634–699: 628243: 013 445: 085 635: 634

253–285: 243 449–455: 448 639: 637305: 007 457–469: 456 654: 653

308, 309: 304 473: 475 804–814: 803327: 028, 029 474: 476 844–859: 843

328: 027 477: 475, 476 865–889: 864

329: 164 479–484: 477

NOTES: There are 250 possible pairs in total, with the source occupation code before the colon and thetarget occupation code after the colon.

Table 1.3: Special Occupational Switches in SIPP

004: 005 118: 167 404: 436005: 004 119: 166 405: 449064: 129 125: 168 406: 466069: 116 129: 064 407: 453073: 115 133: 083 436: 404

077, 079: 136 136: 077, 079 445: 204078: 114 166: 119 449: 405083: 133 167: 118 453: 407114: 078 168: 125 466: 406115: 073 204: 445 804–809: 804–809

116: 069

NOTES: There are 44 possible pairs in total, with the source occupation code before the colon and the targetoccupation code after the colon.


Table 1.4: Overview of Selected Samples

Original Panel Starting Month Ending Month Number of Waves Sample Size

1988 Oct. 1987 Dec. 1989 6 5,2041990 Oct. 1989 Aug. 1992 8 9,8151991 Oct. 1990 Aug. 1993 8 6,4711992 Oct. 1991 Mar. 1995 10 8,8481993 Oct. 1992 Dec. 1995 9 8,8351996 Dec. 1995 Feb. 2000 12 8,507

2001 Oct. 2000 Dec. 2003 9 8,285

Table 1.5: Backing Rates for Selected Samples (%)

Panel Overall Horizontal Vertical Special

1988 94.09 94.48 84.95 95.621990 94.63 94.66 92.91 1001991 95.54 95.62 91.17 1001992 95.96 96.21 87.96 1001993 96.08 96.18 91.84 93.851996 95.79 95.84 93.79 1002001 95.99 96.14 90.04 100

Average 95.44 95.59 90.38 98.5


Table 1.6: Shares of Horizontal, Vertical and Special Switches: Standard Definition (%)

Panel Horizontal Vertical Special

1988 95.52 2.92 1.561990 96.05 3.25 0.71991 96.93 2.29 0.781992 96.86 2.83 0.311993 96.8 2.82 0.381996 96.67 3 0.332001 97.19 2.49 0.32

Average 96.57 2.8 0.63

Table 1.7: Shares of Horizontal, Vertical and Special Switches: Broad Definition (%)


1988 94.71 3.42 1.881990 95.81 3.56 0.621991 96.76 2.44 0.811992 96.4 3.36 0.241993 96.33 3.14 0.531996 95.88 3.3 0.422001 96.52 3.11 0.37

Average 96.06 3.19 0.7

Table 1.8: Shares of Horizontal, Vertical and Special Switches:Very Broad Definition (%)


1988 94.76 3.5 1.741990 95.72 3.65 0.621991 96.42 2.55 1.031992 96.28 3.46 0.271993 96.34 3.15 0.511996 95.65 3.86 0.492001 96.37 3.22 0.41

Average 95.93 3.34 0.72


Table 1.9: Average Rates for Overall Mobility and Horizontal Mobility (%)

DefinitionOverall Mobility Horizontal Mobility

Annual Wave Monthly Annual Wave Monthly

Standard15.22 7.10 1.79 14.70 6.88 1.74(2.13) (0.99) (0.54) (2.09) (0.99) (0.52)

Broad14.77 6.66 1.62 14.18 6.40 1.55(2.03) (1.03) (0.50) (1.97) (1.02) (0.48)

Very 14.26 6.03 1.34 13.67 5.78 1.29Broad (1.93) (0.96) (0.46) (1.86) (0.95) (0.44)

NOTES: Outliers are excluded (see Footnote 12 for details). In parentheses are standard deviations.

Table 1.10: Average Nonemployment Fractions for Panels 1996 and 2001 (%)

Panel 1996 Panel 2001

Switcher Stayer Switcher Stayer

Unempl (lim) 15.07 17.35 23.36 21.48(2.10) (1.89) (3.90) (4.77)

Unempl (comp) 22.69 26.7 31.53 30.67(2.08) (1.50) (3.45) (3.71)

Out 11.91 13.52 17.34 18.43(2.87) (3.11) (2.76) (2.05)

Nonempl (lim) 22.59 25.77 33.9 33.98(3.58) (2.65) (3.99) (4.99)

Nonempl (comp) 29.36 33.92 40.68 41.91(3.50) (2.58) (3.69) (3.99)

NOTES: For columns, Switcher refers to the backed horizontal occupational switchers, and Stayer the occupa-tional stayers who change their jobs (employers). For rows, Unempl (lim) represents unemployment (limiteddefinition), Unempl (comp) unemployment (comprehensive definition), Out out of labor force, Nonempl (lim)unemployment (limited definition) or out of labor force, and Nonempl (comp) unemployment (comprehensivedefinition) or out of labor force. All statistics are associated with the annual occupational mobility under thestandard definition. In parentheses are standard deviations.


Table 1.11: Nonemployment Duration Distributions for Backed Horizontal Switchers:Panel 1996, Starting Wave 2

Unempl (lim) Unempl (comp) Out Nonempl (lim) Nonempl (comp)

Num.of Freq. Num.of Freq. Num.of Freq. Num.of Freq. Num.of Freq.Weeks (%) Weeks (%) Weeks (%) Weeks (%) Weeks (%)

0 66.6 0 65.54 0 79.85 0 54.21 0 53.151 5.31 1 5.86 1 5.55 1 8.7 1 9.252 3.04 2 3.04 2 3.83 2 3.3 2 3.33 3.46 3 3.46 3 0.96 3 4.63 3 4.634 1.57 4 1.57 4 4.44 4 4.41 4 4.415 3.06 5 3.06 5 0.66 5 3.56 5 3.566 1.25 6 1.25 6 1.5 6 1.66 6 1.667 2.82 7 3.33 7 0.47 7 1.78 7 2.298 1.79 8 1.79 8 0.61 8 2.44 8 2.449 0.94 9 0.94 9 0.27 9 1.91 9 1.9110 1.72 10 1.72 10 1.28 10 2.43 10 2.4312 1.69 12 1.69 12 0.58 11 0.36 11 0.3614 1.83 14 1.83 12 2.85 12 2.8517 0.66 17 0.66 13 0.47 13 0.4719 0.27 19 0.27 14 1.47 14 1.4720 1.95 20 1.95 17 0.66 17 0.6621 0.22 21 0.22 18 0.36 18 0.3623 0.38 23 0.38 19 0.27 19 0.2725 1.16 25 1.16 20 1.64 20 1.6426 0.27 26 0.27 21 0.22 21 0.22

22 0.57 22 0.5723 0.38 23 0.3824 0.31 24 0.3125 1.16 25 1.1626 0.27 26 0.27

NOTES: Unempl (lim) represents unemployment (limited definition), Unempl (comp) unemployment (com-prehensive definition), Out out of labor force, Nonempl (lim) unemployment (limited definition) or out of laborforce, and Nonempl (comp) unemployment (comprehensive definition) or out of labor force. All frequenciesare associated with the wave occupational mobility under the standard definition.


Table 1.12: Nonemployment Duration Distributions for Occupational Stayers:Panel 1996, Starting Wave 2



0 74.33 0 74.33 0 88.72 0 67.78 0 67.781 3.36 1 3.36 1 1.83 1 4.2 1 4.22 2.19 2 2.19 2 2.08 2 2.14 2 2.143 4.17 3 4.17 3 0.85 3 4.3 3 4.34 4.6 4 4.6 5 1.3 4 5.1 4 5.16 1.41 6 1.41 6 0.84 5 1.3 5 1.37 2.01 7 2.01 7 1.32 6 0.57 6 0.579 0.71 9 0.71 8 1.28 7 2.34 7 2.3412 0.56 12 0.56 9 1.22 8 0.98 8 0.9816 2.86 16 2.86 10 0.56 9 1.93 9 1.9317 0.62 17 0.62 11 0.72 11 0.7218 0.82 18 0.82 12 0.84 12 0.8421 1.31 21 1.31 14 0.56 14 0.5625 1.06 27 1.06 16 2.86 16 2.86

17 0.62 17 0.6218 0.82 18 0.8220 0.56 20 0.5621 1.31 21 1.3125 1.06 27 1.06


Table 1.13: Nonemployment Duration Distributions for Backed Vertical Switchers:Panel 1996, Starting Wave 2



0 64.52 0 64.52 0 76.67 0 64.52 0 64.526 12.15 6 12.15 1 23.33 6 12.15 6 12.157 23.33 7 23.33 8 23.33 8 23.33



Table 1.14: Nonemployment Duration Distributions for Horizontal Switchers andOccupational Stayers: Panel 1996 (%)

Num.of Unempl (lim) Unempl (comp) Out Nonempl (lim) Nonempl (comp)

Weeks SW ST SW ST SW ST SW ST SW ST

0 73.29 75.82 71.62 74.84 77.81 80.14 60.43 63.82 58.89 63.06(4.17) (4.50) (4.20) (4.02) (3.83) (5.50) (4.78) (6.52) (4.99) (6.14)

1–4 9.54 8.97 10.58 9.52 10.22 8.69 13.53 13.61 14.42 13.99(2.16) (3.51) (2.52) (3.74) (2.46) (2.18) (3.19) (2.17) (3.56) (2.06)

5–13 9.52 7.37 9.84 7.6 5 5.18 11.82 9.52 12.26 9.67(2.26) (2.67) (2.29) (2.63) (1.71) (1.95) (3.10) (2.32) (3.05) (2.30)

14–52 7.33 7.72 7.65 7.91 6.06 5.13 12.45 11.76 12.65 12.01(1.49) (2.59) (1.50) (2.41) (2.81) (3.12) (3.18) (4.04) (3.37) (3.92)

53+ 0.32 0.13 0.32 0.13 0.91 0.87 1.78 1.28 1.78 1.28(0.43) (0.30) (0.43) (0.30) (0.76) (1.22) (1.36) (1.50) (1.36) (1.50)

NOTES: SW refers to the backed horizontal occupational switchers, and ST the occupational stayers who changetheir jobs (employers). Unempl (lim) represents unemployment (limited definition), Unempl (comp) unemployment(comprehensive definition), Out out of labor force, Nonempl (lim) unemployment (limited definition) or out oflabor force, and Nonempl (comp) unemployment (comprehensive definition) or out of labor force. All frequenciesare associated with the wave occupational mobility under the standard definition. In parentheses are standarddeviations.

Table 1.15: Nonemployment Duration Distributions for Horizontal Switchers andOccupational Stayers: Panel 2001 (%)

Num.of Unempl (lim) Unempl (comp) Out Nonempl (lim) Nonempl (comp)

Weeks SW ST SW ST SW ST SW ST SW ST

0 65.67 69.13 63.1 66.93 72.49 72.2 50.91 53.69 48.38 51.78(5.37) (6.33) (4.01) (5.73) (5.25) (4.71) (5.23) (6.64) (4.33) (6.17)

1–4 10.32 9.85 11.94 10.77 8.64 7.94 12.01 13.44 13.59 14.36(2.63) (2.80) (2.11) (2.38) (2.36) (1.79) (2.94) (3.89) (3.39) (3.88)

5–13 9.98 10.49 10.72 11.37 7.99 7.92 13.43 12.99 14.18 13.42(2.20) (2.74) (2.34) (3.18) (2.34) (2.72) (3.49) (1.72) (3.78) (2.41)

14–52 13.38 10.14 13.58 10.55 10.23 9.29 21.29 18.33 21.49 18.88(5.55) (4.70) (5.48) (4.85) (3.99) (3.47) (6.40) (5.63) (6.32) (5.89)

53+ 0.65 0.39 0.65 0.39 0.64 0.66 2.36 1.55 2.36 1.55(0.95) (0.67) (0.95) (0.67) (0.58) (1.06) (2.17) (1.77) (2.17) (1.77)

NOTES: SW refers to the backed horizontal occupational switchers, and ST the occupational stayers who changetheir jobs (employers). Unempl (lim) represents unemployment (limited definition), Unempl (comp) unemployment(comprehensive definition), Out out of labor force, Nonempl (lim) unemployment (limited definition) or out oflabor force, and Nonempl (comp) unemployment (comprehensive definition) or out of labor force. All frequenciesare associated with the wave occupational mobility under the standard definition. In parentheses are standarddeviations.


Figure 1.1: Annual Occupational Mobility: Overall Mobility (%)

NOTES: mobas: standard definition; mobab: broad definition; mobavb: very broad definition.

Figure 1.2: Annual Occupational Mobility: Horizontal Mobility (%)

NOTES: mob1s: standard definition; mob1b: broad definition; mob1vb: very broad definition.


Figure 1.3: Wave Occupational Mobility: Overall Mobility (%)


Figure 1.4: Wave Occupational Mobility: Horizontal Mobility (%)



Figure 1.5: Monthly Occupational Mobility: Overall Mobility (%)


Figure 1.6: Monthly Occupational Mobility: Horizontal Mobility (%)



Figure 1.7: Annual Overall Mobility (Standard Def.) by Age and Education Level (%)

NOTES: Low-education workers are in top panel and high-education workers in bottom panel. Outliers areexcluded (see Footnote 12). mobas: the overall occupational mobility under standard definition; grp1: thegroup with young-age and low-education; grp2: the group with young-age and high-education; grp3: thegroup with middle-age and low-education; grp4: the group with middle-age and high-education; grp5: thegroup with old-age and low-education; grp6: the group with old-age and high-education.


Figure 1.8: Wave Horizontal Mobility (Very Broad Def.) by Ageand Education Level (%)

NOTES: Low-education workers are in top panel and high-education workers in bottom panel. Outliers areexcluded (see Footnote 12). mob1vb: the horizontal occupational mobility under very broad definition; grp1:the group with young-age and low-education; grp2: the group with young-age and high-education; grp3: thegroup with middle-age and low-education; grp4: the group with middle-age and high-education; grp5: thegroup with old-age and low-education; grp6: the group with old-age and high-education.


Figure 1.9: Mean Unemployment Duration (Limited Def.) and Coeff. of Variation

NOTES: Mean unemployment duration (limited definition), in the units of weeks, is in top panel and itscoefficient of variation in bottom panel. lim0: mean unemployment duration for occupational stayers; lim1:mean unemployment duration for horizontal switchers; cv: coefficient of variation.


Figure 1.10: Mean Out of Labor Force Duration and Coeff. of Variation

NOTES: Mean out of labor force duration, in the units of weeks, is in top panel and its coefficient of variationin bottom panel. out0: mean out of labor force duration for occupational stayers; out1: mean out of laborforce duration for horizontal switchers; cv: coefficient of variation.


Figure 1.11: Mean Nonemployment Duration (Comprehensive Def.)and Coeff. of Variation

NOTES: Mean nonemployment duration (comprehensive definition), in the units of weeks, is in top paneland its coefficient of variation in bottom panel. co0: mean nonemployment duration for occupational stayers;co1: mean nonemployment duration for horizontal switchers; cv: coefficient of variation.

Chapter 2

A Directed Search Model of

Occupational Mobility

2.1 Introduction

Recent economic studies demonstrate an important role the occupational mobility plays in

accounting for various interesting economic issues. For instance, Kambourov and Manovskii

(2009a) calibrate a model to match the level and the change of occupational mobility and it

accounts quite well for the level and the change of within-group wage inequality. Hoffmann

(2010) argues that a worker’s occupational mobility is of exceptional significance among

the worker’s labor market mobility dynamics, which determines an individual’s life-cycle

earnings pattern.

Following Keane and Wolpin’s (1997) seminal paper on young workers’ occupational (and

educational) choice, most empirical studies take a micro view and focus on matching a model

to the individual choice pattern in the data.1 The model usually has a very large number

of parameters to pin down, to achieve a perfect fit to the data. Therefore, it’s not very

easy to distinguish the relative importance of each parameter and to understand what role

each factor plays, due to the complicated correlations and interactions among parameters.

This paper looks at data from a macro perspective with the focus on the aggregate mobility

patterns, in an attempt to address two questions: on the one hand, what are the main

determinants that induce/generate the aggregate occupational mobility; on the other, what

are the major barriers to the mobility. In building a parsimonious model which contains

only a small set of variables, it is relatively easy for me to identify the key factors and to

find out the answer.

In this paper, I utilize the unique interview structure of the longitudinal Survey of Income

1See, for example, Hoffmann (2010), Sullivan (2010a) and Yamaguchi (2010).

38

Chapter 2. A Directed Search Model of Occupational Mobility 39

and Program Participation (SIPP) data to uncover novel interesting facts on occupational

mobility, in addition to those well documented in the literature. In particular, SIPP records

workers’ detailed information on occupation and other employment variables every 4 months,

which enables me to identify a worker’s occupational switch and his or her change in labor

market status within a year’s time. More specifically, I examine 5 broad occupational ag-

gregates (Professional, Technical, Service, Craft, and Operators) and study in detail the

mobility patterns among them.2 The four facts are generally descriptive of two aspects of

workers’ occupational behaviors. The first aspect concerns workers’ choice “direction”: given

an individual’s source occupation (and other characteristics), what target occupation (could

be the same as the source) does the person select to work in. I calculate transition matrix

across occupations and employment states, denoted as the mobility distributions, and find

three salient features regarding mobility distributions. The second aspect is related with the

transition time between a worker’s source and target occupations (again, they could be the

same). The transition time is defined as a period during which a worker is unemployed, not

in the labor force, working part-time, or working full-time in a transitory job. It is a gen-

eralization of the concept of nonemployment (unemployment and/or not in the labor force)

in the context of occupational mobility.3 Obviously, the length of transition time contains

useful information on ease of entry into a particular occupation. I calculate transition time

distributions for various source occupations and point out one important feature of this class

of distributions.

In particular, I find that occupational behavior exhibits strong persistence not only among

employed workers but also among workers in transition; a worker becomes less likely to switch

occupation with the increase in occupational human capital; occupational switchers do not

always switch to an occupation similar to their previous one; and the average length of

transition duration varies with their previous occupation.

Research that involves occupational mobility often makes use of two prototypical models:

Lucas Jr. and Prescott (1974) and Keane and Wolpin (1997), with the former a search model

and the latter a dynamic discrete choice structural model with no search frictions. Casual

observation reveals that when an individual faces occupational choice, he or she considers

pros and cons in each candidate occupation and makes the decision to maximize his or her

personal interests. This is the basic feature of a discrete choice model. And to be shown

soon, among the facts this paper tries to address, some are related with the unemployment

2SIPP uses the U.S. Census occupational classification system to record a respondent’s occupational affil-iation, which contains around 500 occupation titles. In principle, I can study the mobility patterns at thismore disaggregative level. However, due to computational constraint, the model in this paper can only besolved and simulated at the aggregate level. In order to examine whether the model can replicate key factsfound in the data, the 5-class classification is used.

3I use this broader than convention concept to stress that a worker’s occupational change is a seriouscareer decision.


spell, or more accurately, the transition duration. Search models are usually used to account

for unemployment, among other labor market phenomena. Therefore, each of the two models

has a main merit which makes it suitable for modeling certain aspects of the occupational

mobility. The current paper combines one with the other.

Motivated by the facts found in data, I construct a directed search model of occupational

mobility. Unlike in Lucas Jr. and Prescott (1974) where all unemployed workers apply for

all jobs with a positive probability (go to all the submarkets with a positive probability4),

workers in my model observe all the relevant variables in each occupation, and given their

individual characteristics, most importantly their current occupation (or latest one if unem-

ployed) and occupational human capital, apply to only one target occupation, the one which

brings them the most benefits. In Keane and Wolpin (1997), unemployment is denoted as

home production, which is a worker’s voluntary choice. While in the current paper, un-

employment (or more accurately, transition) is caused by exogenous separation and search

frictions, and is therefore involuntary.

The model includes both aggregate and idiosyncratic shocks. The aggregate shock affects

a set of occupation-specific variables: job-finding rates, displacement rate, and occupational

returns. They jointly characterize an individual occupation and are key to workers’ decision-

making, and are public information to all the agents. The idiosyncratic shock is used to

capture individual-level factors that affect a worker’s occupational choice, such as personal

interests, family responsibilities, and health conditions, and turns out to be of exceptional

importance.

As in Keane and Wolpin (1997), human capital is included in the model. However,

what’s different here is that occupational human capital is not strictly occupation-specific

as in a typical Roy-type model. Instead, it is partially transferable across occupations, and

the transferability depends on the similarity between the source and target occupations. I

apply the task-based approach used in Gathmann and Schonberg (2010) to measure the

distance between a pair of occupations. In this sense, the occupational human capital in the

paper is general as well as occupation-specific. The transfer loss of human capital apparently

constitutes an obstruction to the occupational mobility. But as numerical exercise shows,

quantitatively, it is not very significant. Sullivan (2010a) also contains both human capital

and occupational mobility in his model, and his main purpose is to compare the relative

importances of the two in accounting for workers’ earnings and lifetime utility. In this paper,

I find the two factors are interrelated: removing the transfer loss of human capital will lead

to an increase in occupational mobility. On the other hand, search frictions are important

in the model, as their existence is necessary in generating unemployment to unemployment

4More accurately, not all the submarkets see newcomers: the submarkets experiencing a very bad shockdon’t attract unemployed workers.


transitions, or more accurately, transition to transition mobility.

Like most sophisticated dynamic discrete choice structural models, this model does not

admit a closed-form solution analytically, and is thus solved numerically. In particular,

I calibrate the model to match exclusively the mobility distributions. And it turns out

that the model can also account for a large fraction of transition time distributions. In

all, it explains 65% of the former and 76% of the latter. Furthermore, the model can

match reasonably well the emphasized facts. It is then used to study (i) the importance of

idiosyncratic vs. aggregate shocks, and (ii) the barriers to occupational mobility. I perform

numerical experiments by removing the relevant variables, one at a time, from the model and

examine how the model performs. I find that idiosyncratic shocks are the main determinant

of occupational mobility whereas aggregate shocks are unimportant. Further, fixed mobility

costs and search frictions constitute significant barriers to mobility while the transfer loss of

occupational human capital is only of modest importance quantitatively.

My work also contributes to the fast-growing directed search literature.5 It’s novel to

study the occupational mobility in a directed search framework. Search is directed in the

sense that when a worker makes the occupational choice decision, he or she observes relevant

information on all occupations and therefore applies optimally, instead of sending his or

her applications at random. Based on the general result in Menzio and Shi (2010), the

stochastic directed search model in this paper admits a Block Recursive Equilibrium (BRE).

In principle, the model should have as one of its state variables the infinite-dimensional

distribution of workers across different occupations and employment states, which makes the

computation of an equilibrium extremely expensive, if not infeasible. However, according to

Menzio and Shi (2010), a special equilibrium in which workers’ value and policy functions are

independent of the distribution of workers, or a BRE, exists. I compute one such equilibrium

at a relatively low cost. And in solving the model numerically, I obtain a set of estimates of

occupation-specific job-finding rates. To my knowledge, there does not exist similar estimates

in the literature and they are therefore novel.

The rest of the paper is organized as follows. Section 2.2 introduces some key concepts

of the paper and applies them in the wage regression. Section 2.3 documents the stylized

facts found in the data. Section 2.4 describes the model. In Section 2.5, I do the calibration

and numerical experiments. Conclusions are in the last section.

5See, for instance, Moen (1997), Acemoglu and Shimer (1999), Burdett et al. (2001), Shi (2009), andMenzio and Shi (2011).


2.2 Distance between Occupations and General Occupational

Tenure

In this section, a measure to determine the distance between a pair of occupations is intro-

duced. Associated with the measure, is the Transfer Rate, or the fraction of occupational

human capital that is transferable across occupations. Accordingly, a new relevant tenure

variable, the General Occupational Tenure, is used to measure occupational human capital

in this context.

2.2.1 Occupations and Occupational Classification under SIPP

Occupation is a name or title that is assigned to a certain class of work duties an individual

performs at work. In general, people group jobs of similar work content, job tasks, and skill

requirements to a single occupation title. There exist many such occupational classifications,

with the most popular ones: the Standard Occupational Classification (SOC), the Dictionary

of Occupational Titles (DOT) and the Occupational Network Database (O*NET). They

classify occupations and give detailed descriptions and qualifications to each occupation. All

of them take a hierarchical structure: a large number of narrowly defined occupations are

grouped into a smaller number of broader occupational aggregates, and they are further

categorized into a even smaller number of higher level occupational aggregates, and this

process may proceed further, depending on an individual classification system’s design.

This paper makes use of the data from the Survey of Income and Program Participation

(SIPP). SIPP is designed by the U.S. Census Bureau to collect detailed information on

income, employment, and government transfer programs participation of the U.S. civilian

noninstitutionalized population. It selects a nationally representative sample of households

and interviews them in every 4 months (called a wave). The high interview frequency is very

advantageous to me in that I can observe a worker’s occupational mobility and employment

state change within a year. SIPP is administered in the so-called panels, and each panel is a

new sample. I use SIPP’s 1996 panel (SIPP96 henceforth) which spans from December 1995

to February 2000 and covers a total of 95398 respondents,6 one of SIPP’s largest samples.

SIPP96 utilizes the U.S. 1990 Census Occupational Classification System, which in turn

builds upon the SOC 1980 version. SIPP96’s occupation table consists of 501 finest titles,

which are grouped into 13 major groups and finally 6 summary groups. Constrained by

the computational capacity, I use the highest level aggregates, the 6 occupational summary

groups (1-digit occupations). But the proposed framework and methodology can in princi-

ple be applied to any level of occupational classification, and are hence general. Due to the

6The actual data come from a subsample of SIPP96. The main restrictions include: male, aged between18 and 64, not disabled, and not self-employed. And the size of effective sample is 26421.


special nature of farming related occupations, they are not considered in the paper. So a

classification of 5 broad occupations is used as follows:

1. Managerial and Professional Specialty Occupations, and henceforth Professional for

short.

2. Technical, Sales, and Administrative Support Occupations, and henceforth Technical

for short.

3. Service Occupations, and henceforth Service for short.

4. Precision Production, Craft, and Repair Occupations, and henceforth Craft for short.

5. Operators, Fabricators, and Laborers, and henceforth Operators for short.

2.2.2 Distance between Occupations and General Occupational Tenure

Intuitively, when a worker changes occupation, he or she usually incurs some loss of occu-

pational human capital. This is because each occupation has its specific requirements of

knowledge, skill, and proficiency. When a worker switches between a pair of occupations of

very different requirements, the loss is heavy, or equivalently, the fraction of occupational

human capital that can be transferred is small. It’s no wonder when a recruiter interviews a

job candidate, an inevitable question is: do you have any experience in this field? Only the

experience in the interviewed occupation or in a close one is of interest, and the experience

in a distant occupation does not really matter.

Though intuitive, it is not easy to measure the distance between occupations in practice.

The point of finding such a measure is that it can help quantify the amount of occupational

human capital conveyed (or destructed, the other side of a same coin) in an occupational

switch. The closer the two occupations are, the higher is the Transfer Rate, and the farther,

the lesser. I find the measure used in Gathmann and Schonberg (2010) is intuitively appealing

and applicably convenient. The measure takes a task-based approach: each occupation is

differentiated by the set of tasks it employs and the degree of intensity of every task deployed.

In an occupational space of n tasks, a single occupation is represented by a vector O of n

dimensions (O1, O2, . . . , On), where Oi is the degree of intensity of task i. Strictly speaking,

an occupation is a ray that stems from the origin of the vector space and goes through the

point that corresponds to the aforementioned vector.7 The distance between two occupations

is measured by the angle formed by the two corresponding rays. The bigger the angle is, the

farther the pair of occupations are from each other. With all the elements in an occupation

vector to be nonnegative, the distance measure lies in [0, π/2]. As a demonstration, Figure 2.1

shows a switch example from Occupation O to Occupation O′ in a 2-dimensional vector space

7That is, what matters is the relative intensity across tasks, not the absolute magnitude of intensity indices.Different points on a ray stand for different normalization methods of task intensities for a same occupation.


(namely, each occupation employs only 2 tasks). In the figure, the angle formed by the source

and target occupations, θ measures the distance between the two occupations.

When a worker changes occupation from O to O′ with the distance θ ∈ [0, π/2], the

Transfer Rate of human capital regarding Occupation O is a strictly decreasing function

of θ. Intuitively, this decreasing function appears to be convex: in general, occupational

switch stands for a serious change in one’s career and is costly. Even when switching to a

relatively close new occupation, the loss of occupational human capital is considerable. A

large fraction is lost as a consequence of initial deviation from the source occupation, and

the impact of further deviations tends to be comparatively small. Empirically, I find the

following convexly decreasing function is satisfactory:

TransRate(θ) = (− 2

πθ + 1)5 (2.1)

As Figure 2.2 shows, Equation (2.1) is essentially a convex transformation from a linear

function of θ, f(θ) = − 2πθ + 1.8 When there’s no occupational switch (θ = 0), Transfer

Rate equals 1, that is, 100% of occupational human capital can be transferred. When the

farthest possible switch takes place (θ = π/2), Transfer Rate equals 0, namely, nothing is

transferable.

Practically, this measure can be computed readily using existing data. Gathmann and

Schonberg (2010) offer information on the degree of intensity of 3 basic tasks, manual,

analytic, and interactive, that are utilized in 64 occupations (Table A1 in their paper). I

aggregate them into my 5 occupations. Table 2.1 lists pairwise distances and Transfer Rates

under the paper’s occupational classification. It shows that the closest non-self pair is Craft

and Operators, and the farthest pair is Professional and Operators, which is in line with

our intuitions. In what follows, the array of Transfer Rates in Table 2.1 is simply called the

Transfer Matrix.

2.2.3 Occupation-Specific Returns

Following convention in the literature, I use tenure to measure the occupational human

capital. Because occupational human capital can be partially transferred across occupations

in this context, it is general as well as occupation-specific. Whereas in most papers, zero

transferability is assumed. Therefore I call the occupational tenure in my paper General

Occupational Tenure to distinguish them. Conceptually, General Occupational Tenure is

between the conventional occupational tenure and general human capital, e.g. labor market

work experience, with it being broader than the former and narrower than the latter. To

8Various Transfer Rate Functions are tried and this one yields the best calibration result. The calibrationis discussed in Section 2.5.


apply this new tenure concept in a wage regression, I use a very general framework as follows:

logw = β1I1 + · · ·+ β5I5 + βEduEdu + βEduSqEdu2

+ βExpWorkExp + βExpSqWorkExp2 + βEmpEmpTen

+ βEmpSqEmpTen2 + βIndIndTen + βIndSqIndTen2

+ βOcc1I1 ×GenOccTen + · · ·+ βOcc5I5 ×GenOccTen

+ βOccSq1I1 ×GenOccTen2 + · · ·+ βOccSq5I5 ×GenOccTen2

+X ′B + ζ (2.2)

In the above regression, logw is the natural log of real wage; Ii is the indicator function for

Occupation i, and it equals 1 if the worker examined works in Occupation i and 0 otherwise;

Edu is a worker’s years of schooling; WorkExp, EmpTen, and IndTen are a worker’s work

experience, employer tenure, and industrial tenure, respectively; GenOccTen is a worker’s

General Occupational Tenure; and finally X is a vector of indicator functions that control

for a worker’s race, marital status, region, 1-digit industry, whether being unionized and the

interview group9.

In the wage regression, regressors Edu, WorkExp, EmpTen, IndTen, and GenOccTen take

a quadratic form. To solve the endogeneity problem, I exploit SIPP’s panel data structure

and follow Altonji and Shakotko (1987), using WorkExp, EmpTen, and IndTen’s deviations

from mean as their instruments. The regression is run in an overlapping manner, every time

data from 3 consecutive waves are being used. So each wave’s data are used for 3 regressions.

Please note that GenOccTen is special and I do not instrument for it. Because occupational

tenure is transferable in the current framework, a worker’s large amount of human capital in

his current occupation could result from the fact that he does really well in the occupation;

however, it is also possible that he is actually a mediocre practitioner in the field but inherits

a lot of human capital from his previous occupations. In this sense, the endogeneity coming

from occupational match quality is less of the concern. By interacting occupational dummies

with the constant, GenOccTen, and its squared term, occupation-specific intercept, linear,

and quadratic coefficients can be obtained. In combination, they provide information on the

returns in a particular occupation.

Table 2.2 lists the main results for the wage regression. Each column stands for an

individual regression, with the title indicating what data are used. For instance, Wave 2

implies that the particular regression is based on data starting from Wave 2, namely, data

from Waves 2, 3, and 4.10 The numbers in the table are estimated coefficients on the

9SIPP divides all respondents into 4 groups and calls them rotation groups.10Recall that every regression makes use of data from 3 consecutive waves, and the regression is run in an

overlapping manner.


regressors, and the regressors are listed in the first column. The stars next to an estimate

indicate its significance level, with single star implying 5%, double star 1%, and triple star

0.1%. Number of observations is on the last row.

As can be seen in Table 2.2, coefficients on WorkExp, EmpTen, IndTen and their square

terms are in most cases not significant in the presence of GenOccTen, whereas GenOc-

cTen and its square are almost always significant for all 5 occupations (except the squared

GenOccTen for Occupation 1, Professional). In some sense this is good news. If the wage

information is needed in a model, I do not need to keep track of various tenure variables.

Instead, only the variable GenOccTen is required, or to be more careful, the education level

Edu being added. Table 2.2 also shows that occupation-specific returns consist of 3 com-

ponents, the intercept term βi, the linear term βOcci× GenOccTen, and the quadratic term

βOccSqi ×GenOccTen2, where i ∈ 1, 2, 3, 4, 5. So for each occupation, the log real wage is

a quadratic function of the General Occupational Tenure, given a worker’s other character-

istics. As a worker’s General Occupational Tenure increases, his wage rate first goes up, and

then reaches the peak and finally declines.11 Furthermore, among the 3 return terms, the

intercept component is much more important than the other 2 terms quantitatively. Take

Occupation 2 (Technical) in Regression Wave 2 as an example. Combining the linear and

quadratic effects, a worker with 30 years of General Occupational Tenure (the peak time for

Technical wage returns) can get a return of 0.406, strictly dominated by the return from

the intercept term, 1.385. Figure 2.3 plots occupation-specific intercepts against time and

Table 2.3 lists coefficient of correlation for pairwise occupational intercept returns. As can

be seen, in general the intercept returns of 5 occupations are positively correlated, moving in

the same direction at all times. Motivated by this feature, I assume there exists an aggregate

shock in the economy. The aggregate shock, for simplicity, may take on 2 values: Good (g)

or Bad (b). When a good shock hits the economy, the intercept component is higher and

hence the entire occupational returns are higher in all occupations, than when a bad shock

hits the economy. This issue will be explored further in Section 2.4.

2.3 Occupational Mobility Patterns in the Labor Market

In this section, I discuss some key descriptive statistics obtained from SIPP96 and show

what stylized facts can be extracted from them.

11Occupation 1, Professional is an exception, because its second order coefficient is insignificant. So aProfessional worker sees his wage increase linearly with his General Occupational Tenure.


2.3.1 Occupational Mobility Distributions

As mentioned before, mobility distributions concern the flow distributions across occupations

and employment states. Figures 2.4 and 2.5 depict them graphically. Figure 2.4 plots all the

possible flows for workers who are in transition12 at time t− 1, or off-job search workers. At

time t, some of them may be employed in the same occupation as their source occupation,

namely the latest occupation one works in when he is employed; some may be employed in

other occupations; others may continue to stay in transition. Similarly, Figure 2.5 shows all

the possible flows for workers who are employed at time t− 1, or on-the-job search workers.

The potential destinations are identical in Figure 2.5 as in Figure 2.4.

To be more concrete, Tables 2.4 to 2.7 display workers’ occupational mobility distribu-

tions for off-job search workers in good time, off-job search workers in bad time, on-the-job

search workers in good time, and on-the-job search workers in bad time, respectively. Each

table is divided into 2 parts, with the left part under the title “Source” and the right part

“Target” or “Result”, implying that a worker works in Occupation x at time t−1, and works

in Occupation y at time t.13 For off-job search workers, their target occupations can be ob-

served, which are the occupations they work at time t. However, for an on-the-job search

worker, although his occupation at time t is known, it might not be his target, because he

may have a bad luck and has his application for the desired occupation declined, and thus

goes back to the time t− 1 occupation. Hence in tables for on-the-job search workers, “Re-

sult” rather than “Target” is used to show this difference. Furthermore, every table’s right

part consists of 3 groups of data according to skill levels: low (with the General Occupational

Tenure less or equal to 14 years), medium (with the General Occupational Tenure between

14 and 28 years), and high (with the General Occupational Tenure greater than 28 years).

Moreover, the 5 columns under each skill level refer to the 5 occupations worked at time t.

Depending on their distances from the source occupation, the 5 occupations are listed in

the order from near to far, with the leftmost column closest to the source occupation and

the rightmost farthest. In combination, Tables 2.4 to 2.7 list the flow percentages in each

case. For instance, the number “15.19” on the first row in Table 2.4 indicates that, in good

time among off-job search workers with low skill levels who used to be working in Occupa-

tion 1 before time t− 1 and who find jobs at time t, 15.19% of them shift to Occupation 2.

Finally, to emphasize, in the tables the dominant flows (greater than 50%) and important

flows (greater than 10%) are marked with double stars and single stars, respectively. At this

moment, ignore all the data in parentheses. In what follows, I call workers on each row in

12Recall that a worker is in transition if he is unemployed, and/or not in the labor force, and/or doingpart-time work, and/or doing full-time transitory work.

13Strictly speaking, this is the case for on-the-job search workers, who work at both time t− 1 and time t.For off-job search workers, the table considers that a worker used to be working in Occupation x beforetime t− 1 and works in Occupation y at time t.


a table, or equivalently workers who share a same source occupation, a group, and I further

call them a subgroup if they share a same source occupation and a same skill level.

From the above tables, I find 3 stylized facts with regard to the occupational mobility

distributions:

Fact 1 Workers’ occupational behavior demonstrates strong persistence. In other words,

the majority of them continue to work in an occupation the same as their previous occupa-

tion. As can be seen from Tables 2.4 and 2.5, more than half of the off-job search workers

continue to work in their source occupation, no matter how skilled they are, regardless of the

aggregate economic conditions. This pattern is even sharper for on-the-job search workers.

Tables 2.6 and 2.7 show that at least 98% of workers in all the subgroups end up with the

same occupation in time t as in time t− 1. Keane and Wolpin (1997) and Sullivan (2010a)

report a similar finding, but only for on-the-job search workers.

Fact 2 As a worker’s General Occupational Tenure increases, he becomes less likely

to switch to another occupation. This tendency generally holds for workers of all source

occupations and under both good and bad aggregate shocks, especially starker for off-job

search workers. Take off-job search workers whose source occupation is 1 (Professional) as

an example, in good times, 68.57% of them choose a target occupation of Professional when

their General Occupational Tenure is less than 14 years; this fraction goes up to 89.74%

for those whose General Occupational Tenure is between 14 and 28 years; the percentage

continues to climb and 100% of them choose not to switch occupations if we consider only

the workers with the General Occupational Tenure greater than 28 years.

Fact 3 For workers who do change their occupations14, they do not always switch to

the closest neighbor occupation. While the groups of occupational switchers with source

occupations Professional, Craft, and Operators tend to switch to the nearest occupation,

their counterparts with the source occupation Technical are likely to select distant target

occupations, and Service workers are inclined to shift to the occupation of middle distance.

2.3.2 Transition Time Distributions

Mobility distributions alone, however, are not complete in describing the occupational mobil-

ity patterns in the labor market. They only consider the workers who hold a job at time t15.

What’s missing here is the pattern for workers who do not work for at least 2 consecutive pe-

riods. For these workers, the intervening transition time between the 2 jobs (2 occupations)

is a nontrivial statistic.

Table 2.8 and Figure 2.6 are informative with regard to the transition time distributions.

14The flows that are both qualitatively and quantitatively important are those with a single star in Tables2.4 and 2.5.

15They may be with or without a job at time t− 1.


Table 2.8 lists the mean transition times for workers of all source occupations under both

good and bad aggregate shocks.16 For each source occupation, numbers on the top row

are in the units of waves, the reference period of SIPP interview, where one wave equals 4

months; numbers on the bottom row are the equivalent mean transition times in the units

of months, calculated from the top row numbers. For this moment, ignore the numbers

in parentheses. Figure 2.4 graphically demonstrates the transition time distributions across

different subgroups and under different aggregate shocks. For this moment just pay attention

to Part(a) and ignore Part(b). Each part consists of 10 panels, with the top row depicts

situations under good shocks and bottom row bad shocks. The 5 panels on each row, from

left to right, correspond to Occupations 1 to 5: Professional, Technical, Service, Craft, and

Operators, respectively. The 3 stacked bars in every panel represent 3 different skill levels

from left to right: low, medium, and high.17 Each bar has 3 sections and they, from bottom

to top, show the fractions of workers who experience short (4 months), medium (8 months

or 1 year), and long (more than 1 year) transition periods, respectively.

According to Table 2.8 and Figure 2.6, a fact concerning the transition time distributions

is:

Fact 4 The average length of transition time depends on a worker’s source occupation,

and follows such an order from long to short: Professional, Technical, Service, Craft, and

Operators, no matter under what economic conditions. This cross-sectional difference is

clear in Table 2.8. And Figure 2.6 provides more details: the mean transition time is mainly

influenced by the fractions of 2 groups of workers: the top end who experience long transition

periods and the bottom end who experience short durations. An occupation sees a longer

mean transition time than another either because it has more workers in its top end, or

because it has less workers in its bottom end; the “middle class” is relatively even across

occupations.

2.3.3 Discussion

The 4 facts extracted from data provide valuable insights for understanding the occupational

mobility in the labor market. Fact 1 emphasizes the importance of including occupational

stayers, otherwise the picture might be incomplete. Therefore a synthetic approach is appro-

priate for analyzing workers’ occupational behavior: they always aim to maximize individual

interests by making the best occupational choice, no matter what they end up to be: occu-

16The mean transition times seem much longer than the mean unemployment duration reported in manyempirical papers. Recall that I include the time a worker spends on out of the labor force, doing part-timejobs, and doing full-time transitory jobs in the transition. The mean unemployment duration calculated fromSIPP96 is 16.21 weeks, consistent with the results reported by other authors.

17Recall from Tables 2.4 to 2.7 that, a worker is low skilled if his General Occupational Tenure is less thanor equal to 14 years, medium skilled if between 14 and 28 years, and high skilled if more than 28 years.


pational switchers or occupational stayers. Yet on-the-job search and off-job search should

be distinguished, as quantitatively a big difference can be seen in Fact 1. Facts 2 and 4

indicate that, among an individual worker’s many characteristics, his source occupation and

General Occupational Tenure are very relevant to his optimal occupational choice. Fact 3

implies that the distance between occupations is an important factor when a worker makes a

decision, but may not be the unique factor. The aggregate shocks are motivated by the wage

regression results. However, inclusion of only aggregate shocks seems inadequate. Further

investigation of Tables 2.4 to 2.7 discloses that a same subgroup of workers’ optimal choices

always differ, indicating that idiosyncratic shocks also matter.

To summarize, in an occupational mobility model, both occupational switchers and stay-

ers are to be considered; but on-the-job search and off-job search should be treated differently;

source occupation, General Occupational Tenure, aggregate shocks and individual shocks are

significant variables; and distance between occupations is an important factor.

2.4 The Model

This section describes the model and explains how the results obtained from the wage re-

gression can be used in the model.

2.4.1 Model Environment

The economy is populated by a large amount of workers, who derive utility only from con-

sumption. Time is discrete. The workers discount future utility at rate β and die with

probability ρ in every period. So the effective discount factor for workers is β = β(1 − ρ).

A worker is endowed with a fix number T units of time in each period. There are N oc-

cupations in the economy that are indexed by i, where i ∈ I ≡ 1, 2, . . . , N. If a worker

works in Occupation i at time t, his labor income is witT , where wit is the real wage rate in

Occupation i at time t. If he is not employed, he receives benefits from the government (not

explicitly modeled) with the amount bT , where b is a positive constant. There is no tech-

nology of borrowing or saving in this economy, so a worker consumes all his labor income or

government benefits in each period, given a strictly increasing instantaneous utility function

u(). Leisure or labor disutility is not modeled, so a worker seeks to maximize

∞∑t=0

βtu(ct)

where ct is the worker’s autarky consumption at time t.

Human capital is general as well as occupation-specific, and is measured by the General

Occupational Tenure. As demonstrated in the wage regression, General Occupational Tenure


is the most important determinant of the wage rate, wit = w(sit, ψt), where sit is the General

Occupational Tenure in Occupation i at time t, and ψt is a vector of all the other variables

that affect the wage. In the wage regression, they are the vector Xt. General Occupational

Tenure is partially transferable across occupations: when a worker switches occupation from

i to j, a fraction δij ∈ [0, 1] of General Occupational Tenure can be carried with the worker,

that is, sjt = sitδij . δij can be large or small, depending on the distance between occupations.

The closer the 2 occupations are , the larger is δij . A matrix composed of δ’s with δij being

its entry at i’s row and j’s column is called a Transfer Matrix. Transfer Matrix is symmetric

with 1’s on its diagonal. General Occupational Tenure si increases by τ when a worker works

in Occupation i for one period.

Workers meet occupations with search frictions and search is directed. When a worker

applies to an occupation, he knows the probability with which he can get a job offer in that

occupation. This probability is called the job-finding rate of that occupation. Job-finding

rates are occupation-specific and vary across occupations. For a given occupation, the job-

finding rate is different for on-the-job search and off-job search workers, with the rate higher

for the former. In reality, it is usually easier for a job searcher to find work on-the-job than

off-job, perhaps due to the difference in social networking, available information, financial

resources, so on and so forth. Each period, workers also separate exogenously from work in all

occupations. The probability of an exogenous separation is called the displacement rate, and

like the job-finding rate, it differs across occupations. Therefore, an individual occupation is

fully characterized by its wage scheme (most importantly, the General Occupational Tenure),

job-finding rates (on-the-job and off-job) and displacement rate. They are well known to all

the workers and workers make the optimal choice in applying to occupations.

The model period is chosen to be the same as SIPP’s reference period, which is 4 months.

Each period has 3 stages, with the first being search and matching stage. At this stage, all

on-the-job search workers and eligible off-job search workers18 send out applications to their

target occupation.19 For a worker applying to Occupation i on-the-job, he succeeds in

matching with target occupation with the job-finding rate pi; for an off-job search worker,

his job-finding rate is p′i, with p′i < pi. The on-the-job search worker switches to his target

occupation if he gets the job offer, otherwise he goes back to his source occupation and

matches with it with certainty. The same is for the off-job search worker if he obtains the

job offer, but in case of a failure, he receives benefits from the government and waits to search

off-job again in the next period. At the second stage, production and consumption take place.

18See Stage 3 for details.19This model assumes that all eligible workers search with probability one in a period. Given the length

of model period (4 months), this assumption is reasonable. Menzio and Shi (2011) calibrate a directed jobsearch model to the U.S. economy and find that monthly on-the-job search probability is 1 and monthlyoff-job search probability is 0.833 for all workers.


Matched workers work in their occupations and get paid. All workers consume what they

get. At the last stage, matched workers separate with their occupations exogenously with

the displacement rate qi. Those separated cannot search immediately and must stay not

employed for one period.

2.4.2 Value Functions

The economy is assumed to be in the stationary state, and hence all the value functions are

without a time subscript. The value function for on-the-job search workers is as follows

V (i, sit,Ωt) = maxVstay(i, sit,Ωt), Vswitch(i, sit,Ωt) (2.3)

where,

Vstay(i, sit,Ωt) =

u(witT ) + β(1− qit)EV (i, sit + τ,Ωt+1) + βqitEU(i, sit + τ,Ωt+1),

Vswitch(i, sit,Ωt) =

maxj 6=i,j∈I

pjt[u(wjtT ) + β(1− qjt)EV (j, sitδij + τ,Ωt+1)

+ βqjtEU(j, sitδij + τ,Ωt+1)]

+ (1− pjt)[u(witT ) + β(1− qit)EV (i, sit + τ,Ωt+1)

+ βqitEU(i, sit + τ,Ωt+1)]− φ

in which Ωt denotes the wage schemes (or the set of variables as well as coefficients in the

wage regression) for all occupations at time t, E represents the conditional expectation based

on information contained in Ωt, and φ is a fixed mobility cost in terms of utils, which is a

one-time cost and is incurred only when a worker prepares to switch occupation on-the-job.

In principle, the state variable vector in all the above value functions, and hence the

corresponding policy function, should contain the infinite-dimensional distribution of workers

across different occupations and employment states. However, as shown in Menzio and

Shi (2010), a general class of stochastic directed search models admits a Block Recursive

Equilibrium in which agents’ value and policy functions are independent of the distribution

of workers. In this paper, I consider such a BRE and therefore the distribution object is not

included.20

20Although my model is one-sided equilibrium in nature while Menzio and Shi (2010) is a two-sided equi-librium model, the occupation-specific wage schemes and job-finding rates can be deemed as derived fromthe firm (occupation)’s side of a two-sided equilibrium model. There is no human capital in Menzio and Shi(2010) while it exists in this paper, but firms (occupations) do not face uncertainty in this dimension becauseI assume a large number of workers and the law of large numbers just removes this uncertainty.


For an off-job search worker, his value function is

W (i, sit,Ωt) = maxWstay(i, sit,Ωt),Wswitch(i, sit,Ωt) (2.4)

where,

Wstay(i, sit,Ωt) =

p′it[u(witT ) + β(1− qit)EV (i, sit + τ,Ωt+1)

+ βqitEU(i, sit + τ,Ωt+1)]

+ (1− p′it)[u(bT ) + βEW (i, sit,Ωt+1)],

Wswitch(i, sit,Ωt) =

maxj 6=i,j∈I

p′jt[u(wjtT ) + β(1− qjt)EV (j, sitδij + τ,Ωt+1)

+ βqjtEU(j, sitδij + τ,Ωt+1)]

+ (1− p′jt)[u(bT ) + βEW (i, sit,Ωt+1)]− φ′

in which φ′, like φ is a fixed mobility cost in terms of utils, which is a one-time cost and is

incurred only when a worker prepares to switch occupation off-job.

Finally, for workers who are not employed, or in transition, the value function takes the

form

U(i, sit,Ωt) = u(bT ) + βEW (i, sit,Ωt+1) (2.5)

For an on-the-job search worker with source occupation i and General Occupational

Tenure sit, he chooses between to stay and to switch, and to what target occupation j to

apply if the latter. If he stays, he earns wage from occupation i, and gets displaced with

probability qit at the end of the period. If he elects to switch to occupation j, with probability

pjt he gets the job and then works in occupation j to earn labor income, and gets displaced

with probability qjt at the end of the period; however, when he fails to get a job offer in his

target occupation, he returns to work in occupation i, and later separates exogenously with

occupation i with probability qit. Note the evolution of General Occupational Tenure: when

transfer happens, a fraction δij is carried on, and when production takes place, an amount of

τ is added. A disutility φ is incurred for a worker to choose an occupation different from his

source one. It is used to capture the preparation cost when one seeks to switch occupation.

An off-job search worker faces the same decision problem as his on-the-job search coun-

terpart. The difference is that, at this time, he has to apply even when he elects to be

an occupational stayer, and he no longer has a safety net when his application fails (going

back to work in the source occupation). In this sense, all occupations appear to be more

homogeneous. When a worker searches off-job, he gets an offer with the occupation-specific


job-finding rate. If he succeeds, he produces in the target occupation and then proceeds to

the following separation stage; if he fails, he receives benefits and gets ready to search off-job

in the following period. Again, a fixed mobility cost is incurred when one tries to switch

occupation off-job (φ′).

A worker in transition cannot search. He receives benefits from the government and waits

to conduct off-job search in the period that follows. Please note that workers enter transition

by displacement. Not to work is not a choice in the model. To concentrate on main issues,

I assume that the government can perfectly monitor workers and announces that those who

do not bother to apply are not eligible for the benefits. Because leisure is not valued and

utility comes solely from consumption, no one will choose to be idle in this context.

I assume a log utility function in what follows.21 This assumption helps simplify the

model significantly. It can be shown that the constant log T shifts value functions in a parallel

fashion and does not affect workers’ optimal choice. So it can be removed from the model.

Plugging the log wage function in the wage regression into the value functions, one finds a

set of variables as follows: a worker’s race, rotation group number, education level, marital

status, region, 1-digit industry, and whether being unionized. Among them, the first two are

constants and hence can be removed safely. The remaining variables, especially education

level22 and marital status, tend not to change often and can be deemed as close to constants.

So I remove them as well. To cover the effects of all non-constant variables in the wage

equation including error terms, an i.i.d. idiosyncratic shock ε is introduced. In addition to

this, it is also used to capture the individual-level factors that affect a worker’s occupational

choice decision, such as health condition, personal interest, and family commitment. More

over, to make the model simple, instead of drawing ε’s for all candidate occupations, it is

drawn only for the source occupation. So, it actually reflects a difference effect: a very

large positive number indicates that a worker really wants to stay in the current occupation;

while a negative number with a large absolute value implies strong incentives to switch. The

idiosyncratic shock is assumed to be normally distributed, with mean zero and standard

deviation σ and σ′ for on-the-job and off-job search workers, respectively.

So far the terms remain in Ωt are General Occupational Tenure sit and its coefficients.

Recall that in Section 2.2 an aggregate shock is proposed. I denote it as χt here. The

introduction of χ is to save me the effort of keeping track of the complex evolution of the

coefficients on occupational returns. Further, I assume that ε is independent of χ.

The simplified value functions are listed as follows:

On-the-job search value function

21Though not standard, some authors use log utility functions in the studies of labor market, e.g. Sullivan(2010a), Pavan (2011), and Yamaguchi (2012).

22The sample is selected such that workers have a stable attachment with the labor market. Basically, theyhave finished their education stage.


V (i, sit, χt, εt) = maxVstay(i, sit, χt, εt), Vswitch(i, sit, χt, εt) (2.6)

where,

Vstay(i, sit, χt, εt) =

log(wi(sit, χt)) + β(1− qit)EV (i, sit + τ, χt+1, εt+1)

+ βqitEU(i, sit + τ, χt+1, εt+1) + εt,

Vswitch(i, sit, χt, εt) =

maxj 6=i,j∈I

pjt[log(wj(sit, χt)) + β(1− qjt)EV (j, sitδij + τ, χt+1, εt+1)

+ βqjtEU(j, sitδij + τ, χt+1, εt+1)]

+ (1− pjt)[log(wi(sit, χt)) + β(1− qit)EV (i, sit + τ, χt+1, εt+1)

+ βqitEU(i, sit + τ, χt+1, εt+1)]− φ

Off-job search value function

W (i, sit, χt, εt) = maxWstay(i, sit, χt, εt),Wswitch(i, sit, χt, εt) (2.7)

where,

Wstay(i, sit, χt, εt) =

p′it[log(wi(sit, χt)) + β(1− qit)EV (i, sit + τ, χt+1, εt+1)

+ βqitEU(i, sit + τ, χt+1, εt+1)]

+ (1− p′it)[log(b) + βEW (i, sit, χt+1, εt+1)] + εt,

Wswitch(i, sit, χt, εt) =

maxj 6=i,j∈I

p′jt[log(wj(sit, χt)) + β(1− qjt)EV (j, sitδij + τ, χt+1, εt+1)

+ βqjtEU(j, sitδij + τ, χt+1, εt+1)]

+ (1− p′jt)[log(b) + βEW (i, sit, χt+1, εt+1)]− φ′

Transition value function

U(i, sit, χt, εt) = log(b) + βEW (i, sit, χt+1, εt+1) (2.8)

The aggregate shock χt takes on two values in the model: Good (g) or Bad (b). The

value of g or b is determined with the help of Table 2.2. The average intercept is calculated

for each occupation and is used as a cutoff point. The periods in which the intercepts are


above the cutoff point are called good times, otherwise bad times. This is done to all 5

occupations, and not surprisingly the conclusions are generally consistent, a pattern already

shown by Table 2.3 and Figure 2.3. Then I calculate mean occupation-specific intercepts

in good times βig and in bad times βib. I do the same for linear coefficient βOccTeniχ and

quadratic coefficient βOccTenSqiχ , χ ∈ g, b. Based on the above parameters, the log wage in

the value functions can be determined as follows:

log(wi(si, χ)) = βiχ + βOccTeniχ si + βOccTenSqiχ s2i

log(wj(si, χ)) = βjχ + βOccTenjχ δijsi + βOccTenSqjχ (δijsi)2,

where χ ∈ g, b.Please note that the labor demand side (occupation side) is not explicitly modeled in the

paper. Instead, I take an indirect approach: including the occupation-specific job-finding

rates and wage schemes in the model. They are key equilibrium objects: in a two-sided

equilibrium model, both sides take them as given and make optimal decisions, and these

decisions indeed generate the job-finding rates taken as given. In the current paper, they

are obtained through calibration, and are functions of the aggregate shock.

2.5 Numerical Analysis

Like most dynamic discrete choice structural models, the model does not have an closed-

form analytical solution and is solved numerically. In this section, I discuss how the model

is parameterized and solved, and then analyze the results.

2.5.1 Direct Estimation

Recall that the model period is chosen to be the same as SIPP’s reference time, 4 months or

1/3 year. Except otherwise stated, all the time variables in the paper are expressed in the

units of years. In Section 2.2, the coefficients on occupation-specific returns are estimated

through the wage regression. The intercept coefficient βiχ, linear coefficient βOccTeniχ , and

quadratic coefficient βOccTenSqiχ vary with occupations and aggregate economic conditions,

where i ∈ 1, 2, 3, 4, 5, χ ∈ g, b. And they are evaluated according to the method presented

in Section 2.4, on the basis of wage regression estimates.

Furthermore, the following parameters come from direct inference from data. The dis-

count rate β is set to be 0.9870 to match an annual interest rate of 4%. The death probability

ρ equals 0.007246 so that a worker’s potential working life in the labor market is 46 years

(aged 18–64). Jointly they imply that the value of effective discount rate β is 0.98. The


per-period increment of General Occupational Tenure when one remains in an occupation τ

is equal to 1/3 (year). I choose b, the invariant benefits, to be 0.4629, so that the benefits

are 36%23 of the average wage rate in the economy (ignoring the linear and quadratic com-

ponents of the returns). The aggregate shock χ is assumed to be i.i.d. and data show that

it takes on g with the probability 0.625 and b 0.375.24 The occupation-specific displacement

rate is estimated by the displacement flow divided by its corresponding employment stock

for each occupation. All the estimates in this subsection are summarized in Table 2.9.

The table shows that the cross-sectional variation in the intercept component of occupa-

tional returns is similar under both aggregate shocks, and so is the order of magnitudes, from

big to small: Professional, Craft, Technical, Operators, and Service. A negative shock brings

the intercept return slightly down in all 5 occupations with the average decline of 4.09%.

The cross-occupational difference of the less important return parts, linear and quadratic,

also does not see an obvious variation in good times than in bad times, and a bad shock

causes the linear return coefficients uniformly higher (except going down in Professional)

and the quadratic ones uniformly lower (except both zero in Professional). For occupation-

specific displacement rates, the cross-sectional variation has little changes across aggregate

economic conditions, but a negative shock’s impact is very different across occupations: the

displacement rates in Technical, Craft, and Operators are relatively stable, while Professional

experiences an 18.16% increase and Service an even sharper one, 47.03%.

As a reminder, the Transfer Matrix, which is needed when I discount the General Occu-

pational Tenure for switchers, can be found in Table 2.1.

2.5.2 Calibration Strategy

The parameters that cannot be evaluated in the above process include: the occupation-

specific job-finding rates piχ’s (on-the-job) and p′iχ’s (off-job), where i ∈ 1, 2, 3, 4, 5, χ ∈g, b; fixed mobility costs φ (on-the-job) and φ′ (off-job); and the standard deviations of

idiosyncratic shocks σ (on-the-job) and σ′ (off-job).25 To keep the model tightly param-

eterized, I assume p′iχ = γpiχ, γ ∈ (0, 1). That is, the job-finding rates for off-job search

workers are uniformly lower than their counterparts for on-the-job search workers, by a con-

stant factor γ. To summarize, there are 15 values to be determined in total: piχ’s where

i ∈ 1, 2, 3, 4, 5, χ ∈ g, b, γ, φ, φ′, σ and σ′. I denote the set of unknown parameters Θ

and evaluate Θ through calibration.

In particular, I calibrate the model to match the starred flows in Tables 2.4 to 2.7, or

23Kletzer and Rosen (2008) calculate the U.S. average replacement rate (average weekly benefits as a shareof average weekly earnings) on the basis of Department of Labor data. The number is 0.36 for the years 1975to 2004.

24In another version of the model, a Markov process is assumed. The numerical results have hardly changed.25The means of idiosyncratic shocks are assumed to be zero for workers of both search types.


mobility distributions. The calibration consists of 2 steps. In the first step, given a set of

values for the 15 parameters, the model is solved numerically. It is easy to show that the

model can be expressed as a system of 2 nested functional equations: on-the-job search value

function V and off-job search value function W . Using value function iteration, one function

can be solved given another function. Therefore, I keep performing value function iterations

for V and W alternately until the system converges. In Step 2, I simulate the model to get

the statistics corresponding to the starred ones in Tables 2.4 to 2.7, and compute the sum

of squared differences between the 2 sets of statistics based on the data and on the model.

The whole process is repeated until the difference is minimized by the simplex method and

the resultant Θ are the desired values. The algorithm is demonstrated in Figure 2.7.

2.5.3 Calibration Results

Table 2.10 lists the calibration results. As shown in the table, the job-finding rates for

on-the-job search workers differ across occupations, varying from 30.99% (Professional) to

53.36% (Service) in good times and from 27.40% (Technical) to 32.85% (Craft) in bad times:

the cross-sectional difference is bigger under good shocks than under bad shocks. A negative

aggregate shock makes the job-finding rate significantly lower for all occupations except Pro-

fessional: the Professional job-finding rate declines by 8.29%, whereas the average decrease

for the other 4 occupations is 35.54%. As a consequence of the drop, the mean transition du-

ration goes up in all occupations, a feature can be found in Table 2.8. The job-finding rates

for off-job search workers are about 80% of the corresponding rates for on-the-job search

workers: the gap is moderate.

The fixed mobility cost for an on-the-job switch is 4.92, roughly equivalent to 21.3 hours

of hourly wage for an average worker with 30 years of General Occupational Tenure; whereas

the cost for an off-job switch is 2.34, only about 1.6 hours of hourly wage for a same average

worker. The reason for a higher on-the-job mobility cost might be that preparing for a new

career while holding a full-time job is more challenging than with no job at hand, both

physically and intellectually, and thus incurs higher disutility. The standard deviation of the

idiosyncratic shock for an employed worker is 8.65, which is around 4.74% of the average

lifetime utility for an employed worker with 30 years of General Occupational Tenure, while

for a worker in transition, the shock’s standard deviation is higher at the level of 11.33,

or roughly 6.19% of the average lifetime utility for a worker in transition with 30 years of

General Occupational Tenure. It’s not very clear why an off-job search worker has a stronger

intentive to leave or to stay in his source occupation than an on-the-job search worker does,

and maybe it is related with some psychological issues.


2.5.4 Model Fit

To evaluate the model’s performance, I do 2 checks. The first check is to see how much

the model can account for the 2 sets of distributions, occupational mobility (Tables 2.4

to 2.7) and transition time (Figure 2.6). The simulations are conducted to generate both

distributions. The mobility distributions based on model simulation are presented in Tables

2.4 to 2.7 with numbers in parentheses. And the transition time distributions from model

simulation are displayed graphically as Part (b) in Figure 2.6, from which the mean transition

time is computed for each occupation and demonstrated in parentheses in Table 2.8. In

comparing model-based and data-based distributions, I can tell what fraction of the data

distributions can be accounted for by the model. Specifically, if a mass probability in the

data distributions takes on a value x, and its corresponding value in the model distributions

is y, and then y/x × 100% is the fraction accounted for. In many cases, y is no more than

x and this ratio is less than or equal to 100%. Under circumstances where y > x, the

formula (1 − y)/(1 − x) × 100% is used: x is over accounted for because its complement,

1 − x26, is under accounted for. It turns out that the model can account for 65.46% of the

mobility distributions and 76.30% of the transition time distributions. Note that the former

are targeted when the model is calibrated, but in no time do I try to match the latter. In

this sense, the model does a good job.

Secondly, I examine the 4 occupational mobility facts discussed in Section 2.3 one by one.

As for Fact 1, the model successfully generates the occupational choice persistence. As shown

by the parenthesized numbers in Tables 2.4 to 2.7, generally more than 70% of workers work

in their source occupation. But it also has a limitation: the on-the-job search persistence

seems not significantly stronger than that in the off-job search as in data, due to the fact that

the model largely underestimates the persistence for employed workers in all occupations.

The model does not work well in reproducing Fact 2, that the tendency to switch occupation

declines as workers accumulate the General Occupational Tenure. As can be seen in Tables

2.4 and 2.5, the model generates too high stayer fractions for low-skill subgroups and too

low stayer fractions for medium- and high-skill subgroups. The model fits Fact 3 fairly

well in that it accurately predicts the major occupational targets for occupational switchers

in 3 occupations, Professional, Craft, and Operators, but the outflow directions for source

occupations Technical and Service are missed. As far as Fact 4, the ordering of occupation-

specific mean transition times is concerned, the model’s performance is satisfactory. The

order it generates is: Professional, Technical, Craft, Operators, and Service, which are listed

in Table 2.8. Except Service, the other 4 occupations are put in the right place. However,

one problem is that the model underreports the length of average transition period for all

26All numbers in the distributions are percentage points and no more than 1. When y > x, x is necessarilyless than 1.


occupations. An investigation of Figure 2.6 reveals the reason: the model-based fraction of

workers who spend medium transition times is too high in each occupation, while that of

workers experiencing long transition durations is too low.

Despite the limitations discussed above, the model is reasonable. In general, it captures

the main features found in the data, especially given its highly parsimonious nature. Em-

pirical occupational choice models often have a large set of parameters to estimate. For

instance, Keane and Wolpin (1997) have more than 80 parameters to pin down, and Sulli-

van (2010a) estimates nearly 200 model variables. In this paper, only 15 parameter values

are obtained through solving the model. More importantly, the purpose of this paper is to

find out which shock, aggregate or idiosyncratic, is more important in understanding the

occupational mobility, and what factors may act to obstruct the mobility. With only a small

number of key variables in the model, it is relatively easy to identify the role each factor

plays and to arrive fast at the answer.

2.5.5 Experiments

With the structural model at hand, I do 2 numerical experiments to answer the questions

raised at the very beginning in the paper. The first question is: how important are idiosyn-

cratic versus aggregate shocks in understanding the occupational mobility? To address this

question, I start with the benchmark model, and first take away idiosyncratic shocks to see

how much the modified model can account for the mobility and transition duration distribu-

tions, and then do a similar exercise by removing only aggregate shocks to see what happens.

Recall that aggregate shocks affect the following occupation-specific variables: job-finding

rates, displacement rates, intercept, linear, and quadratic returns. By removing aggregate

shocks, I mean letting each of these variables take on the value of its average across aggregate

shocks. Table 2.11 lists the exercise’s results. It shows the fraction of occupational mobility

distributions that can be accounted for under each circumstance.27 As already known, with

both shocks present the benchmark model can account for 65.46% of the mobility distribu-

tions. However, removing the idiosyncratic shock makes the fraction drop sharply to 10.91%.

In contrast, in the absence of aggregate shocks, the fraction accounted for sees almost no

change (65.45%).28

As a demonstration, Figure 2.8 compares the policy functions for the off-job search Craft

workers at bad times under 2 scenarios: with and without idiosyncratic shocks. Panel (a)

of the figure shows the workers’ best occupational choice when there exist idiosyncratic

shocks. In solving the benchmark model, I discretize the idiosyncratic shock ε into 5 values,

27Subsection 2.5.4 discusses the method on how I calculate the fraction of a data-based distribution ac-counted for by the model.

28Because neither of the modifications brings noticeable change to the fraction of transition time distribu-tions the model can account for, the result reports are omitted.


ε1 < ε2 < ε3 < ε4 < ε5 with ε3 = 0, ε1, ε2 negative and ε4, ε5 positive. Recall that a

very negative value indicates strong incentives to switch occupation while a very positive

one to stay. In Panel (a), if a worker is hit by ε1 or ε2, his optimal target is Occupation 1

(Professional) when his General Occupational Tenure is less than 5 years and Occupation 5

(Operators, an occupation closer to Craft than Professional) when he has more than 5 years

of General Occupational Tenure. However, if he is hit by other shocks, he will choose to

stay in Craft (Occupation 4). As opposed to Panel (a), Panel (b) depicts the worker’s best

choice when there is no idiosyncratic shock. His policy function is quite simple in this case:

never to switch regardless of his skill level, as if he were always hit by ε3 in the world of

idiosyncratic shocks. As Figure 2.8 shows, when the idiosyncratic shock is taken away from

the model, there will be too little mobility generated, and hence the explained fraction of

mobility distributions drops dramatically.

The first experiment shows that it is idiosyncratic shocks, or more accurately, the very

negative realizations of idiosyncratic shocks that induce workers to switch occupation. A

following question naturally is: what are the barriers to the occupational mobility. In Ex-

periment 2, I examine 3 factors, fixed mobility costs, search frictions, and the transfer loss of

General Occupational Tenure to assess their quantitative importances in this respect. The

experiment is conducted in the same spirit as in Experiment 1. Once again, I start with the

benchmark model, and deviate from it by first taking away fixed mobility costs, then search

frictions, finally the transfer loss, one factor at a time, to see how the model performs. The

experiment results are displayed in Table 2.12.

To remove fixed mobility costs, I let both φ and φ′ equal zero. As Table 2.12 shows,

if the fixed mobility costs are taken away, there is a moderate decrease in the explained

fraction of the mobility distributions from 65.5% to 49.6%, while the explained fraction of

the transition time distributions keeps essentially unchanged. Figure 2.9 further illustrates

that, the removal of mobility costs results in an excess of occupational mobility. It compares

the policy function for the on-the-job search Operators at good times under 2 scenarios: with

and without the fixed mobility cost, in a similar manner as in Figure 2.8. Consider workers

who face a zero value of the idiosyncratic shock. With the presence of the mobility cost, an

operator’s best choice is to stay as shown in Panel (a). In contrast, if the mobility cost is

eliminated, workers can afford to switch to Occupation 4 (Craft) when they have a General

Occupational Tenure less than 15 years or greater than 30 years, as depicted in Panel (b).

So taking fixed mobility costs away will cause the model to generate too much mobility,

opposite to the case of the removal of idiosyncratic shocks, but both of them will make the

explained fraction of mobility distributions decline.

Next, I make all the job-finding rates equal one for both on-the-job and off-job search

workers, that is, piχ = 1, γ = 1, i ∈ 1, 2, 3, 4, 5, χ ∈ g, b, to eliminate search frictions. As


opposed to the fixed mobility costs, search frictions’ impact is mainly on the transition time

distributions. Without search frictions, the fraction of duration distributions the model can

account for simply drops to zero, or there is no transition at all.29 This result is no surprise

since search frictions are the unique mechanism that generates transition in the model.

Search frictions, however, does not seem to have obvious effects on mobility distributions:

elimination of them leads to almost no change in the fraction of mobility distributions the

model can account for.

To make all the occupational switches without a loss of General Occupational Tenure,

I let all the elements in the Transfer Matrix equal one, namely, δij = 1, i, j ∈ 1, 2, 3, 4, 5.Like fixed mobility costs, the transfer loss of General Occupational Tenure basically sees its

influence on the mobility distributions. If it is taken away, there is a modest decline in the

explained fraction of mobility distributions from 65.6% to 60.9%, and no apparent change

in the explained fraction of transition time distributions. Figure 2.10 demonstrates the role

that transfer loss plays in obstructing the occupational mobility. To focus on main issues,

suppose there are neither mobility costs nor search frictions in the economy. Moreover, also

assume there is no aggregate shock or exogenous separation. The figure plots the log wage as

a function of the General Occupational Tenure for 2 occupations, Technical and Operators.

Consider an operator with 30 years of General Occupational Tenure. His current log wage is

1.77 at Point A. If he thinks of switching to Technical, without a transfer loss he will carry

100% or 30 years of General Occupational Tenure to the new occupation and receive a log

wage of 1.82, which is represented by Point B. However, in case the transfer loss exists, he

can transfer only 16% or 4.8 years of General Occupational Tenure to Technical, according

to the Transfer Matrix, and thus earn a log wage of 1.58, indicated by Point C. Hence he

will give up the idea of switching. It’s obvious that the transfer loss constitutes a barrier

to the occupational mobility. But in the model it is less important than the fixed mobility

costs in a quantitave sense.

2.6 Conclusions

In this paper, I document four stylized facts on the occupational mobility, making use of

SIPP’s unique interview structure. Motivated by these facts, a directed search model is

built to investigate the mechanism of aggregate mobility across occupations and employ-

ment states. The model includes both aggregate and idiosyncratic shocks, and contains

occupational human capital measured by the General Occupational Tenure as well as search

frictions. The model can account for 65% of the occupational mobility distributions and 76%

29Strictly speaking, all separated workers experience one period of transition. Because the model assumesthat a worker cannot search immediately after being displaced and has to wait for one period.


of the transition time distributions found in the data. And it also captures the main features

of the four emphasized facts. To examine what role each factor plays in generating the mo-

bility, I conduct two groups of numerical experiments. I show that idiosyncratic shocks are

the main determinant of occupational mobility whereas aggregate shocks are unimportant,

and that fixed mobility costs and search frictions constitute significant barriers to the mo-

bility while the transfer loss of General Occupational Tenure is only of modest importance

quantitatively.

In the model, the aggregate shock affects an array of occupation-specific variables: job-

finding rates (for employed workers and workers in transition); displacement rate; intercept,

linear, and quadratic occupational returns. Some of them are estimated directly from data,

and others are obtained from model calibration. As shown in the paper, a negative shock

generally lowers job-finding rates, raises displacement rate, and makes the intercept and

quadratic returns go down while the linear component go up for all occupations. But it

is hard to figure out how the aggregate shock exerts its influence on those occupation-

specific variables in a one-sided equilibrium model. Indeed, they are equilibrium objects in

an economy (especially the job-finding rates and return components). To study the process

of how the variables achieve their equilibrium levels after an aggregate shock hits, a two-sided

equilibrium model like in Menzio and Shi (2010) is needed.

Another issue deserves further investigation is the nature of idiosyncratic shocks, which

are the main determinant of occupational mobility. It is found in the paper that the shock’s

variance is larger for workers in transition than for workers employed, but the reason remains

unclear. One conjecture is that a worker’s individual shock is correlated with his transition

duration. The model abstracts from private savings and does not track workers’ benefits

collection history. It is assumed that a worker in transition can receive social benefits in-

finitely. But of course in reality, there is a time limit. When this time limit approaches, an

unemployed worker may compromise and become more willing to work in a new occupation.

This might be a direction to pursue in further studies.


Table 2.1: Distance/Transfer Rate For Occupation Pairs

Source Target1 2 3 4 5

1 0/1 0.10/0.72 0.39/0.24 0.53/0.13 0.58/0.102 0.10/0.72 0/1 0.30/0.35 0.44/0.20 0.48/0.163 0.39/0.24 0.30/0.35 0/1 0.17/0.56 0.24/0.444 0.53/0.13 0.44/0.20 0.17/0.56 0/1 0.07/0.805 0.58/0.10 0.48/0.16 0.24/0.44 0.07/0.80 0/1

NOTES: In each cell, there are 2 numbers. The left one refers to the distance between the source and targetoccupations and is in the units of radians. The right one is the Transfer Rate for the pair. Occupations 1 to 5correspond to Professional, Technical, Service, Craft, and Operators, respectively.

Table 2.2: Results of Wage Regression

Wave2 Wave3 Wave4 Wave5 Wave6 Wave7 Wave8 Wave9

EmpTen -0.1090* -0.0296 -0.0628 -0.0582 -0.0669 -0.0260 -0.1420** -0.0749EmpTenSq -0.000057 -0.000302 -0.000215 -0.000370 -0.000333 0.000529 0.000305 -0.000744WorkExp 0.0886 0.1920*** 0.1060 0.0595 0.0872 0.0592 0.1210* 0.0795WorkExpSq -0.000170 -0.000189 -0.000165 -0.000331 -0.000328 -0.000104 -0.000343 0.000133IndTen 0.0365 -0.1220 -0.0102 0.0282 0.0229 0.0036 0.0659 0.0208IndTenSq -0.000346 -0.000398 -0.000102 0.000213 0.000252 -0.000739 -0.000905 -0.000285Edu 0.0184 0.0127 0.0132 0.0184 0.0270* 0.0305** 0.0245* 0.0231EduSq 0.001100 0.001270 0.001140 0.000900 0.000611 0.000453 0.000671 0.000675I1 1.604*** 1.677*** 1.609*** 1.579*** 1.599*** 1.648*** 1.671*** 1.665***I2 1.385*** 1.505*** 1.471*** 1.423*** 1.457*** 1.510*** 1.539*** 1.517***I3 1.307*** 1.403*** 1.335*** 1.281*** 1.309*** 1.332*** 1.367*** 1.333***I4 1.443*** 1.577*** 1.572*** 1.523*** 1.528*** 1.526*** 1.532*** 1.520***I5 1.372*** 1.501*** 1.479*** 1.439*** 1.444*** 1.455*** 1.485*** 1.463***I1 OccTen 0.0190 0.0247* 0.0284** 0.0230** 0.0211* 0.0179* 0.0180* 0.0176*I2 OccTen 0.0271*** 0.0257*** 0.0289*** 0.0304*** 0.0255*** 0.0182*** 0.0149** 0.0143**I3 OccTen 0.0218** 0.0222*** 0.0299*** 0.0308*** 0.0263*** 0.0225*** 0.0207*** 0.0228***I4 OccTen 0.0341*** 0.0305*** 0.0287*** 0.0283*** 0.0279*** 0.0288*** 0.0289*** 0.0270***I5 OccTen 0.0224*** 0.0191*** 0.0191*** 0.0181*** 0.0193*** 0.0188*** 0.0161*** 0.0164***I1 OccTenSq -0.000265 -0.000455 -0.000482 -0.000283 -0.000240 -0.000202 -0.000261 -0.000302I2 OccTenSq -0.000452** -0.000435** -0.000540*** -0.000613*** -0.000469** -0.000266 -0.000182 -0.000167I3 OccTenSq -0.000434 -0.000455* -0.000734*** -0.000785*** -0.000631** -0.000519** -0.000524* -0.000598**I4 OccTenSq -0.000666*** -0.000589*** -0.000557*** -0.000557*** -0.000549*** -0.000589*** -0.000599*** -0.000561***I5 OccTenSq -0.000320** -0.000259* -0.000270** -0.000255* -0.000283** -0.000276** -0.000222* -0.000262**

Obs 8813 9037 9177 9268 9364 9441 9423 9407

NOTES: Each column stands for an independent regression, with the title indicating what data are used. Forinstance, Wave 2 implies that the particular regression is based on data starting from Wave 2, namely, data fromWaves 2, 3, and 4. The numbers in the table are estimated coefficients on the regressors, which are listed in the firstcolumn. The stars next to an estimate indicate its significance level, with single star implying 5%, double star 1%,and triple star 0.1%. Number of observations is on the last row.


Table 2.3: Coefficient of Correlation for Pairwise Intercept Returns

Occupation Occupation1 2 3 4 5

1 1 · · · · · · · · · · · ·2 0.8460 1 · · · · · · · · ·3 0.8642 0.7014 1 · · · · · ·4 0.3405 0.6063 0.5667 1 · · ·5 0.6176 0.8332 0.7212 0.9301 1

NOTES: Occupations 1 to 5 correspond to Professional, Technical, Service, Craft, and Operators, respectively.

Table 2.4: Occupational Mobility Distributions for Off-Job Search Workers in Good Time

Source Target

GenOccTen ≤ 14 14 < GenOccTen ≤ 28 GenOccTen > 28

1 1∗∗ 2∗ 3 4 5 1∗∗ 2 3 4 5 1∗∗ 2 3 4 568.57 15.19 6.47 3.70 6.08 89.74 8.41 0 1.85 0 100 0 0 0 0(84.62) (9.23) (0) (6.15) (0) (72.73) (27.27) (0) (0) (0) (58.33) (41.67) (0) (0) (0)

2 2∗∗ 1 3 4∗ 5∗ 2∗∗ 1 3 4 5 2∗∗ 1 3 4 560.72 5.24 7.76 12.71 13.56 85.79 4.53 0 5.28 4.40 84.11 8.66 0 7.23 0(83.67) (16.33) (0) (0) (0) (79.07) (20.93) (0) (0) (0) (91.67) (8.33) (0) (0) (0)

3 3∗∗ 4 5∗ 2 1 3∗∗ 4 5 2 1 3∗∗ 4 5 2 166.95 7.19 19.89 4.50 1.47 85.26 3.10 5.48 2.67 3.49 100 0 0 0 0(76.74) (21.71) (0) (0) (1.55) (91.18) (8.82) (0) (0) (0) (83.33) (16.67) (0) (0) (0)

4 4∗∗ 5∗ 3 2 1 4∗∗ 5∗ 3 2 1 4∗∗ 5 3 2 173.26 18.39 0 6.13 2.22 79.20 10.46 1.65 4.45 4.23 96.50 0 3.50 0 0(78.80) (8.29) (0) (0) (12.90) (72.53) (27.47) (0) (0) (0) (82.14) (17.86) (0) (0) (0)

5 5∗∗ 4∗ 3 2∗ 1 5∗∗ 4 3 2 1 5∗∗ 4 3 2 169.12 13.70 5.17 9.98 2.04 89.03 8.40 2.57 0 0 83.91 6.90 9.19 0 0(74.53) (18.24) (0) (0) (7.23) (80.26) (19.74) (0) (0) (0) (86.96) (13.04) (0) (0) (0)

NOTES: The table is divided into 2 parts, with the left part Source and the right part Target, implying that a workerused to be working in occupation Source before time t − 1 and works in occupation Target at time t. GenOccTenrefers to the General Occupational Tenure. The table’s right part consists of 3 groups of data according to the skilllevel: low, medium, and high, depending on the length of GenOccTen which is in the units of years. The 5 columnsunder each skill level represent the 5 occupations worked at time t. Depending on their distances from the sourceoccupation, the 5 occupations are listed in the order from near to far, with the leftmost column closest to the sourceoccupation and the rightmost farthest. Given a source occupation, all workers of the same skill level constitute asubgroup. The numbers are the flow percentages for a given subgroup. Numbers without parentheses are from dataand those in parentheses from model simulation. For numbers coming from data, those no less than 10% are markedwith single stars and 50% double stars. Occupations 1 to 5 correspond to Professional, Technical, Service, Craft,and Operators, respectively.


Table 2.5: Occupational Mobility Distributions for Off-Job Search Workers in Bad Time

Source Target


1 1∗∗ 2∗ 3 4 5∗ 1∗∗ 2 3 4 5 1∗∗ 2 3 4 558.02 20.98 2.67 6.64 11.69 92.92 7.08 0 0 0 100 0 0 0 0(71.43) (28.57) (0) (0) (0) (76.92) (23.08) (0) (0) (0) (33.33) (66.67) (0) (0) (0)

2 2∗∗ 1 3 4∗ 5∗ 2∗∗ 1 3 4 5∗ 2∗∗ 1 3 4 5∗

56.07 6.83 1.36 15.25 20.48 74.11 6.52 2.47 6.20 10.70 81.22 0 0 0 18.78(73.42) (26.58) (0) (0) (0) (82.76) (17.24) (0) (0) (0) (80.00) (20.00) (0) (0) (0)

3 3∗∗ 4∗ 5∗ 2 1 3∗∗ 4 5 2 1 3∗∗ 4 5 2 167.55 13.02 10.38 9.06 0 95.75 0 0 4.25 0 100 0 0 0 0(82.14) (17.86) (0) (0) (0) (55.56) (44.44) (0) (0) (0) (100) (0) (0) (0) (0)

4 4∗∗ 5∗ 3 2 1 4∗∗ 5 3 2 1 4∗∗ 5 3 2 173.85 19.03 1.83 3.93 1.36 93.49 2.29 0 2.20 2.01 91.39 2.53 3.84 2.24 0(70.53) (18.95) (0) (0) (10.53) (68.63) (31.37) (0) (0) (0) (67.86) (32.14) (0) (0) (0)

5 5∗∗ 4∗ 3 2 1 5∗∗ 4 3 2 1 5∗∗ 4 3 2 170.41 14.72 2.59 8.71 3.57 88.03 6.52 0 2.78 2.66 100 0 0 0 0(81.76) (13.21) (0) (0) (5.03) (80.39) (19.61) (0) (0) (0) (100) (0) (0) (0) (0)

NOTES: The table is divided into 2 parts, with the left part Source and the right part Target, implying that a workerused to be working in occupation Source before time t − 1 and works in occupation Target at time t. GenOccTenrefers to the General Occupational Tenure. The table’s right part consists of 3 groups of data according to the skilllevel: low, medium, and high, depending on the length of GenOccTen which is in the units of years. The 5 columnsunder each skill level represent the 5 occupations worked at time t. Depending on their distances from the sourceoccupation, the 5 occupations are listed in the order from near to far, with the leftmost column closest to the sourceoccupation and the rightmost farthest. Given a source occupation, all workers of the same skill level constitute asubgroup. The numbers are the flow percentages for a given subgroup. Numbers without parentheses are from dataand those in parentheses from model simulation. For numbers coming from data, those no less than 10% are markedwith single stars and 50% double stars. Occupations 1 to 5 correspond to Professional, Technical, Service, Craft,and Operators, respectively.

Table 2.6: Occupational Mobility Distributions for On-The-Job Search Workers in Good Time

Source Result


1 1∗∗ 2 3 4 5 1∗∗ 2 3 4 5 1∗∗ 2 3 4 598.86 0.15 0.22 0.61 0.16 99.47 0 0 0 0.53 100 0 0 0 0(79.25) (18.14) (0) (2.61) (0) (75.00) (25.00) (0) (0) (0) (84.71) (15.29) (0) (0) (0)

2 2∗∗ 1 3 4 5 2∗∗ 1 3 4 5 2∗∗ 1 3 4 599.08 0.07 0.21 0.22 0.42 99.84 0 0 0 0.16 100 0 0 0 0(77.18) (22.82) (0) (0) (0) (77.36) (22.64) (0) (0) (0) (73.48) (26.52) (0) (0) (0)

3 3∗∗ 4 5 2 1 3∗∗ 4 5 2 1 3∗∗ 4 5 2 199.52 0.08 0.20 0.20 0 100 0 0 0 0 100 0 0 0 0(77.86) (22.14) (0) (0) (0) (76.66) (23.34) (0) (0) (0) (79.49) (20.51) (0) (0) (0)

4 4∗∗ 5 3 2 1 4∗∗ 5 3 2 1 4∗∗ 5 3 2 199.48 0.45 0 0 0.07 99.77 0.14 0.03 0 0.05 99.89 0.11 0 0 0(76.88) (15.24) (0) (0) (7.88) (76.16) (23.84) (0) (0) (0) (77.42) (22.58) (0) (0) (0)

5 5∗∗ 4 3 2 1 5∗∗ 4 3 2 1 5∗∗ 4 3 2 199.49 0.24 0.13 0.12 0.03 99.79 0 0.06 0.15 0 100 0 0 0 0(77.55) (22.45) (0) (0) (0) (76.63) (23.37) (0) (0) (0) (74.75) (25.25) (0) (0) (0)

NOTES: The table is divided into 2 parts, with the left part Source and the right part Result, implying that aworker works in occupation Source at time t − 1 and in occupation Result at time t. GenOccTen refers to theGeneral Occupational Tenure. The table’s right part consists of 3 groups of data according to the skill level: low,medium, and high, depending on the length of GenOccTen which is in the units of years. The 5 columns under eachskill level represent the 5 occupations worked at time t. Depending on their distances from the source occupation,the 5 occupations are listed in the order from near to far, with the leftmost column closest to the source occupationand the rightmost farthest. Given a source occupation, all workers of the same skill level constitute a subgroup. Thenumbers are the flow percentages for a given subgroup. Numbers without parentheses are from data and those inparentheses from model simulation. For numbers coming from data, those no less than 10% are marked with singlestars and 50% double stars. Occupations 1 to 5 correspond to Professional, Technical, Service, Craft, and Operators,respectively.


Table 2.7: Occupational Mobility Distributions for On-The-Job Search Workers in Bad Time

Source Result


1 1∗∗ 2 3 4 5 1∗∗ 2 3 4 5 1∗∗ 2 3 4 599.80 0 0 0.20 0 100 0 0 0 0 100 0 0 0 0(77.73) (20.34) (0) (1.93) (0) (83.14) (16.86) (0) (0) (0) (78.72) (21.28) (0) (0) (0)

2 2∗∗ 1 3 4 5 2∗∗ 1 3 4 5 2∗∗ 1 3 4 599.72 0 0 0.10 0.18 100 0 0 0 0 100 0 0 0 0(74.48) (25.52) (0) (0) (0) (78.82) (21.18) (0) (0) (0) (77.11) (22.89) (0) (0) (0)

3 3∗∗ 4 5 2 1 3∗∗ 4 5 2 1 3∗∗ 4 5 2 199.06 0.14 0.18 0.39 0.24 100 0 0 0 0 100 0 0 0 0(77.06) (22.94) (0) (0) (0) (81.95) (18.05) (0) (0) (0) (63.64) (36.36) (0) (0) (0)

4 4∗∗ 5 3 2 1 4∗∗ 5 3 2 1 4∗∗ 5 3 2 199.52 0.28 0.13 0.06 0 99.94 0.06 0 0 0 100 0 0 0 0(78.85) (15.89) (0) (0) (5.26) (77.17) (22.83) (0) (0) (0) (76.35) (23.65) (0) (0) (0)

5 5∗∗ 4 3 2 1 5∗∗ 4 3 2 1 5∗∗ 4 3 2 199.54 0.23 0.16 0.07 0 100 0 0 0 0 100 0 0 0 0(77.78) (21.64) (0) (0) (0.58) (78.40) (21.60) (0) (0) (0) (76.86) (23.14) (0) (0) (0)

NOTES: The table is divided into 2 parts, with the left part Source and the right part Result, implying that aworker works in occupation Source at time t − 1 and in occupation Result at time t. GenOccTen refers to theGeneral Occupational Tenure. The table’s right part consists of 3 groups of data according to the skill level: low,medium, and high, depending on the length of GenOccTen which is in the units of years. The 5 columns under eachskill level represent the 5 occupations worked at time t. Depending on their distances from the source occupation,the 5 occupations are listed in the order from near to far, with the leftmost column closest to the source occupationand the rightmost farthest. Given a source occupation, all workers of the same skill level constitute a subgroup. Thenumbers are the flow percentages for a given subgroup. Numbers without parentheses are from data and those inparentheses from model simulation. For numbers coming from data, those no less than 10% are marked with singlestars and 50% double stars. Occupations 1 to 5 correspond to Professional, Technical, Service, Craft, and Operators,respectively.

Table 2.8: Mean Transition Time for All Source Occupations

Source Good Time Bad Time

1 4.11(3.09) 5.28(3.47)16.44(12.36) 21.12(13.88)

2 3.97(2.85) 4.58(3.07)15.88(11.40) 18.32(12.28)

3 3.56(2.54) 4.22(2.59)14.24(10.16) 16.88(10.36)

4 3.22(2.68) 4.07(2.93)12.88(10.72) 16.28(11.72)

5 3.14(2.66) 4.00(2.91)12.56(10.64) 16.00(11.64)

NOTES: For each source occupation, numbers on the top row are in the units of waves, the reference period of SIPPinterview, where one wave equals 4 months; numbers on the bottom row are the equivalent mean transition timesin the units of months, calculated from the top row numbers. Numbers without parentheses are based on the data,while numbers in parentheses are from model simulation. Occupations 1 to 5 correspond to Professional, Technical,Service, Craft, and Operators, respectively.


Table 2.9: Parameters Obtained Not Through Calibration

Occupation-specific Returns

β1g 1.654 β1b 1.594β2g 1.508 β2b 1.422β3g 1.354 β3b 1.299β4g 1.545 β4b 1.498β5g 1.477 β5b 1.418βOccTen1g 0.0213 βOccTen

1b 0.0210βOccTen2g 0.0204 βOccTen




5b 0.0199

βOccTenSq1g 0 βOccTenSq

1b 0

βOccTenSq2g -0.000318 βOccTenSq

2b -0.000511


3b -0.000617


4b -0.000591


5b -0.000286

General Parameters

β 0.9870 ρ 0.007246τ 0.3333 b 0.4629

Aggregate Shock Probabilities

Pg 0.625 Pb 0.375

Occupation-specific Displacement Rates

q1g 0.1586 q1b 0.1874q2g 0.1318 q2b 0.1286q3g 0.1163 q3b 0.1710q4g 0.0914 q4b 0.0860q5g 0.0891 q5b 0.0904

NOTES: Occupations 1 to 5 correspond to Professional, Technical, Service, Craft, and Operators, respectively.Both βOccTenSq1g and βOccTenSq1b equal zero because the coefficient on I1 OccTenSq is insignificant in all wageregressions, see Table 2.2.


Table 2.10: Parameters Obtained Through Calibration

p1g 0.3099 p1b 0.2842p2g 0.4355 p2b 0.2740p3g 0.5336 p3b 0.2882p4g 0.4312 p4b 0.3285p5g 0.4594 p5b 0.2974φ 4.92 φ′ 2.34σ 8.65 σ′ 11.33γ 0.7933

NOTES: Occupations 1 to 5 correspond to Professional, Technical, Service, Craft, and Operators, respectively.

Table 2.11: Fraction of Occupational Mobility Distributions Accounted For:Experiment 1 (%)

Model FractionBenchmark 65.46

Without idiosyncratic shocks 10.91Without aggregate shocks 65.45

Table 2.12: Fraction of Distributions Accounted For: Experiment 2 (%)

Model Mobility DurationBenchmark 65.46 76.30

Without Mobility Costs 49.55 76.45Without Search Frictions 65.16 0TransferRates = 100% 60.91 76.68

NOTES: Mobility and Duration refer to the occupational mobility distributions and transition durationdistributions, respectively.


Figure 2.1: Distance Between Occupations: 2-Task Case

NOTES: The distance between source occupation O and target occupation O′ is measured by θ, the angleformed by them. The bigger θ is, the farther the 2 occupations are from each other. θ ∈ [0, π/2].

Figure 2.2: Transfer Rate Function

NOTES: The Transfer Rate function y = (− 2πx+1)5 is convexly decreasing. It is transformed from the linear

function y = − 2πx+ 1.


Figure 2.3: Occupation-Specific Intercepts

NOTES: The figure plots occupation-specific intercepts against time. The intercept returns of 5 occupationsare positively correlated, moving in same directions at all times. Occupations 1 to 5 correspond to Professional,Technical, Service, Craft, and Operators, respectively.


Figure 2.4: Flow Chart for Off-Job Search Workers

NOTES: Workers who are in transition at time t − 1 with source occupation i may end up at time t beingemployed in their source occupation i, or being employed in a different occupation j, j 6= i, or continuing tostay in transition.


Figure 2.5: Flow Chart for On-the-Job Search Workers

NOTES: Workers who are employed at time t − 1 with source occupation i may end up at time t beingemployed in their source occupation i, or being employed in a different occupation j, j 6= i, or flowing totransition.


Figure 2.6: Transition Time Distributions

NOTES: This figure demonstrates the transition time distributions across different subgroups and underdifferent aggregate shocks. Part(a) is based on data and and Part(b) model simulation. Each part consists of10 panels, with the top row depicts situations under good shocks and bottom row bad shocks. The 5 panelson each row, from left to right, correspond to Occupations 1 to 5: Professional, Technical, Service, Craft,and Operators, respectively. The 3 stacked bars in every panel represent 3 different skill levels from left toright: low, medium, and high. Each bar has 3 sections and they, from bottom to top, show the fractionsof workers who experience short (4 months), medium (8 months or 1 year), and long (more than 1 year)transition periods, respectively.


Figure 2.7: Calibration Algorithm


Figure 2.8: Policy Function: Off-Job Search Craft Workers in Bad Times

NOTES: ε’s 1 to 5, from small to large, are 5 realizations of the idiosyncratic shock. Occupations 1 to 5correspond to Professional, Technical, Service, Craft, and Operators, respectively.


Figure 2.9: Policy Function: On-The-Job Search Operators in Good Times

NOTES: The idiosyncratic shock ε takes on the value of ε3, or zero. Occupations 1 to 5 correspond toProfessional, Technical, Service, Craft, and Operators, respectively.


Figure 2.10: Transfer Loss during an Occupational Switch

NOTES: The figure plots the log wage as a function of the General Occupational Tenure for 2 occupations,Technical and Operators. Consider an operator with 30 years of General Occupational Tenure. His currentlog wage is 1.77 at Point A. If he thinks of switching to Technical, without a transfer loss he will carry 100%or 30 years of General Occupational Tenure to the new occupation and receive a log wage of 1.82, which isrepresented by Point B. However, in case the transfer loss exists, he can transfer only 16% or 4.8 years ofGeneral Occupational Tenure to Technical, according to the Transfer Matrix, and thus earn a log wage of1.58, indicated by Point C. Hence he will give up the idea of switching.

Chapter 3

General Occupational Tenure and

Its Returns

3.1 Introduction

How general is human capital? This question has interested economists for half a century and

sparked a large amount of research. The answer to this question is important, as different

answers have quite different implications for an economy as well as for an individual when

a job market mobility takes place. The switches of employers, industries, and occupations

are so frequent when we observe a given labor market, or when we observe a worker’s life

cycle. If the nature of human capital is general, then the transfer cost is small for both the

individual and the economy, because a large part of it can be transferred during the switch

process and thus a big destruction or waste can be avoided; in contrast, if the human capital

is largely specific, then the transfer is costly and excessive mobility may be detrimental to

the whole economy as well as to an individual. Indeed, the specificity of human capital sheds

light on issues like lifecycle inequality (Sullivan (2010a)), wage inequality (Kambourov and

Manovskii (2009a)), growth difference (Wasmer (2004)), trade effects (Ritter (forthcoming)),

and contract designs (Gibbons and Waldman (2006)), etc.

Various studies examine this topic using different methodologies and data (for instance,

see Becker (1964) and Mincer (1974) for the discussions of general human capital, i.e. ed-

ucation and labor market experience; see Bartel and Borjas (1981), Altonji and Shakotko

(1987), Abraham and Farber (1987), Topel (1991), and Altonji and Williams (2005) for

the studies of firm specificity of human capital; see Neal (1995) and Parent (2000) for ex-

plorations of industrial specificity of human capital). The recent study of Kambourov and

Manovskii (2009b) (KM henceforth) argues for the occupational specificity of human capital:

they find that when the occupational tenure is accounted for, employer tenure and industrial

79

Chapter 3. General Occupational Tenure and Its Returns 80

tenure play a very little role. In their study, the treatment of occupational human capital is

standard as in the literature: it is accumulated within an occupation and it gets completely

destroyed during a switch. Two more recent studies provide new insights about the nature

of occupational human capital with the help of task-based approach. Gathmann and Schon-

berg (2010) and Yamaguchi (2012) assume that there exist a small number of fundamental

tasks, which are utilized in every occupation but with different intensiveness. For example,

a set of basic tasks may include analytical task, interpersonal task, and motor task. In other

words, each occupation is a specific use of the bundle of fundamental tasks. Gathmann and

Schonberg (2010) argue that the occupational human capital is more general than previ-

ously considered and it is transferable across occupations, just because it is comprised of

several kinds of task-specific human capital and each individual task-specific human capital

is accumulable and transferable. Yamaguchi (2012) demonstrates that great heterogeneity

exists across occupations and the source of this heterogeneity is the different utilization of

the task bundle. He explicitly models and estimates, as functions of the task intensiveness

combination, the rewarding structure, skill accumulation, and non-pecuniary preference in

every occupation and finds that the model fits data very well.

In this paper, I redo KM’s exercise using the data of Survey of Income and Program

Participation (SIPP) and verify their main finding. This justifies the continued focus on

the occupational human capital, in contrast to the employer-specific or industrial human

capital. The conventional view deems occupations uniformly distinct, which in some sense is

a simplifying strategy that stresses the homogeneous aspect of occupational human capital

(equally non-transferable). However, Yamaguchi (2012) shows that occupations are so dif-

ferent in so many dimensions and we need take their heterogeneity serious. Both Gathmann

and Schonberg (2010) and Yamaguchi (2012) prove that the task-based approach provides

a useful lens through which one can examine this heterogeneity. This article applies the

task-based approach to study the heterogeneous returns across occupations. Specifically,

I generalize KM’s framework in two directions. First, I assume that occupational human

capital, measured by tenure, is partially transferable and the transferability depends on the

similarity between the source and target occupations. Second, returns to occupational tenure

are allowed to be different and are thus occupation-specific.

Under the assumption of partial transferability, the occupational human capital is some-

thing lying between the completely specific human capital, like the firm-specific human

capital, and the completely general human capital, like the labor market work experience,

and is therefore not only specific but also general. And so should be the corresponding

occupational tenure that is used to measure the occupational human capital. I give it a new

name, “General Occupational Tenure”, to distinguish it from the conventional occupational

tenure that assumes uniform non-transferability and to stress its transferable feature.


Conceptually, my “General Occupational Tenure” is similar to the “task tenure” in Gath-

mann and Schonberg (2010). But empirically, we use very different methods to track it.

Gathmann and Schonberg (2010) decompose the conventional occupational tenure into in-

dividual task-associated tenures and keep track of them. when the task tenure is needed,

they synthesize them to get the result. While in this article, I apply a simpler black-box

strategy. In particular, I propose an empirical Transfer Rate function which relates the trans-

ferable portion of General Occupational Tenure and the similarity between occupations, or

occupation distance. At anytime, given the General Occupational Tenure and the associated

occupation, the new General Occupational Tenure is readily obtainable if a target occupation

is told (so the occupation distance is determined).

I adopt an angle measure based on the task approach advocated in Gathmann and

Schonberg (2010) to quantify occupation distance. To calculate this occupation distance,

the key is the task intensiveness data. I use the Dictionary of Occupational Titles (DOT) to

obtain task intensiveness information and apply the principal component analysis to generate

task intensiveness indices. During the process, the augmented April 1971 Current Population

Survey (CPS) is used to convert the DOT’s ordinal scores into the cardinal-flavored values

required by the proposed angle measure of occupation distance. Later, the CPS file is also

used to match the occupation titles between the SIPP and the DOT.

Yamaguchi (2012) shows that great return difference could exist among occupations and

therefore I augment KM’s wage regression by allowing for more flexible occupational het-

erogeneity. Specifically, I allow for occupation-specific returns not only in terms of constant

coefficients, but also in terms of linear and quadratic coefficients on the General Occupational

Tenure. In doing so, I can observe different return structures for dozens of 1- and 2-digit

occupations and hundreds of 3-digit occupations. Traditional dynamic discrete choice struc-

tural models admit return heterogeneity when modeling occupational choices. However, due

to the heavy computational burden imposed by the curse of dimensionality, they are forced

to choose only a few occupations in the model.1 In my reduced-form framework, I can have

a much larger occupation set than structural models do.

By examining occupation-specific returns for 1-, 2-, and 3-digit occupations, I find three

common patterns hold for all the occupational classifications. First, there is considerable

variation of the returns to General Occupational Tenure across occupations. Second, among

the components that constitute the total return, the intercept part in general dominates the

combination of linear and quadratic parts. Third, the intercept part is inversely related with

the combination of linear and quadratic parts and thus a tradeoff exists between them.

1For instance, Keane and Wolpin (1997) consider 5 alternatives which include schooling, home production,and 3 occupations; Hoffmann (2010) has 3 occupational choices plus unemployment; Sullivan (2010a) includesschooling, unemployment, and 5 occupations; Xiong (2012) considers 5 occupations.


The above conclusions are based on the specific functional form assumption made in the

current paper. To generalize it, I first consider an extreme case where occupational human

capital is strictly non-transferable and so the General Occupational Tenure reduces to the

conventional occupational tenure. The occupation-specific returns for 1-, 2-, and 3-digit

occupations are investigated for this extreme case and the three patterns are found still

valid. Then I show that a special family of convexly decreasing Transfer Rate functions

converge to the considered extreme case as the discounting during a switch becomes heavier.

Provided that the initial and limiting cases share the same properties, these properties should

apply to all the cases in between. Thus the three patterns tend to be a general result rather

than just a specific case.

This article also contributes to the literature by helping reconcile two different views on

the specificity of human capital. On the one hand, KM suggest that human capital tends

to be occupation-specific. On the other, Gathmann and Schonberg (2010) and Poletaev and

Robinson (2008) find that task-specific human capital is the most important source of wage

growth. Using the task-approach to analyze occupational human capital, I show in this

article that there is essentially no conflict between the two views, because an occupation is

equivalent to a specific usage of the basic task bundle. In this sense, there is no difference

between occupation specificity of human capital and task specificity of human capital.

In a very relevant paper, Sullivan (2010b) demonstrates that great heterogeneity exists

not only in returns to occupational tenure, but also in returns to industrial tenure across

occupations. He also extends the KM wage regression in two dimensions: by allowing for

within-firm occupational mobility and by running wage regressions independently occupation

by occupation. Whereas in this paper I augment the KM wage regression by allowing for

occupational human capital’s partial transferability and occupation-specific return structure

of returns to only occupational tenure variables. Our second extensions point to the same

direction but theoretically his is obviously more general. However, constrained by the data,

his exercise can only be performed at the 1-digit level. While my regressions are run under

1-, 2-, and 3-digit levels. So empirically speaking, the conclusions in this article are more

robust. Indeed, there is a tradeoff between theoretical flexibility and empirical robustness.

Despite the differences, both our exercises conclude that return structures differ a lot among

various occupations.

The rest of the paper is organized as follows. In Section 3.2, I run wage regressions in

a KM framework and compare my results and theirs. Section 3.3 introduces the notion of

General Occupational Tenure and sets up a new econometrical framework to estimate its

returns in every individual occupation. Section 3.4 discusses data sets that are used and

relevant procedures. In Section 3.5, I show the main empirical results and discuss them. I

turn to the estimation and discussion of one limiting case in Section 3.6. Conclusions are in


the last section.

3.2 KM Wage Regression Revisited

Kambourov and Manovskii (2009b) perform the wage regression as follows:

logw = β0 + βEduEdu + βEduSqEdu2

+ βExpWorkExp + βExpSqWorkExp2 + βExpCbWorkExp3

+ βEmpEmpTen + βEmpSqEmpTen2 + βOJOJ

+ βIndIndTen + βIndSqIndTen2 + βIndCbIndTen3

+ βOccOccTen + βOccSqOccTen2 + βOccCbOccTen3

+X ′B + ζ (3.1)

In the above regression, logw is the natural log of real wage; Edu is a worker’s years of

schooling; OJ (Old Job) is a dummy variable which equals one if one’s employer tenure is

equal to or greater than one year and zero otherwise; WorkExp, EmpTen, IndTen, and Oc-

cTen are a worker’s work experience, employer tenure, industrial tenure, and occupational

tenure, respectively; and finally X consists of the following regressors: 1-digit occupation

dummies, 1-digit industry dummies, union dummy, marital status dummies, year dummies,

region dummies, current and lagged county level unemployment rates. To solve the en-

dogeneity problem, the authors apply Altonji and Shakotko (1987)’s instrumental variable

method that is widely used in the literature (e.g. Parent (2000), Gathmann and Schonberg

(2010)). In particular, they use WorkExp, EmpTen, OJ, IndTen, and OccTen’s deviations

from mean as their instruments. Take EmpTen as an example: suppose EmpTen is the

average employer tenure the worker has with the current employer, then EmpTen is instru-

mented by ˜EmpTen = EmpTen − EmpTen and similarly EmpTen2 is instrumented by˜EmpTen2 = EmpTen2 − EmpTen2. Finally, given the nature of panel data, the error terms

are allowed to be serially correlated for a given individual. In summary, KM use an IV-GLS

method.

Due to the differences in data and sample restrictions2, this paper modifies the above

regression equation slightly. In particular, the threshold for OJ to take on unity is 4 months,

instead of one year, because the data I use have a frequency of 4 months while KM use annual

PSID (Panel Study of Income Dynamics) data.3 This paper’s regression does not include

year dummies as my data spread a relatively short time span (4 years from 1996 to 2000)

2A detailed description of the data along with the sample restrictions is in Section 3.4.1.3The regressions are also performed with an OJ of the one-year threshold just as KM do, and the results

are very similar.


compared to KM’s (25 years from 1968 to 1993). And county level unemployment rates are

not included as well, for there is no county level residence information in my data. On the

other hand, my econometrical model contains two sets of extra dummy variables: one set

controls for race and the other set controls for interview group number. While KM restrict

their sample to white people, I do not impose such a restriction. My data are divided into

four interview groups (called Rotation Groups) and information collected from each Rotation

Group is based on a different reference time; KM’s PSID data do not have such a structure.

Table 3.1 lists the coefficient estimates for the wage regression in a KM framework.

The three columns from left to right correspond to 1-, 2-, and 3-digit occupational (and

industrial) classifications, respectively4. As the table shows, if a wage regression controls

for a worker’s occupational tenure, other tenure variables, specifically employer tenure and

industrial tenure, are no longer important: the estimated coefficients are not significant at

the conventional significance levels. This finding is similar to that in KM and supports the

occupational specificity of human capital.

Table 3.2 demonstrates the returns to 2, 5, and 8 years of various tenure variables: oc-

cupational tenure, industrial tenure, and employer tenure, respectively, assuming everything

else being equal. Again the results are reported under 1-, 2-, and 3-digit occupational (and

industrial) classifications. As can be seen in the table, in the presence of occupational tenure

in a wage regression, the returns to industrial tenure and employer tenure are of minor im-

portance and their p-values are much larger than their occupational tenure counterparts,

consistent with the coefficient estimates in Table 3.1, and with the finding in KM5. And

Table 3.2 shows that returns to occupational tenures are hump-shaped with the peak ap-

pearing at 5 years. Specifically, 5 years of occupational tenure are associated with a wage

increase of 2.4% to 3.2%. However, KM report that the returns to 2-, 5-, and 8-year oc-

cupational tenures are increasing monotonically and 5 years of occupational tenure would

increase a worker’s wage by 8.02% to 11.97%. One possible reason for this difference may

be the different sampling times of the two data sets.

To summarize, I verify KM’s finding using the SIPP data in this section: among different

classes of specific human capital, occupational human capital is the most important one. So

I set my focus on this class of human capital throughout the current article.

4For a detailed discussion of occupational and industrial classifications, please refer to Section 3.4.4.5Please refer to their Table 3.2 (Section B).


3.3 The Concept of General Occupational Tenure and Its Re-

turns

3.3.1 Occupation Distance and General Occupational Tenure

The KM framework assumes that all occupations are uniformly distinct, and so if a worker

switches from one source occupation to any non-self target occupation, the loss of occupa-

tional human capital is the same: 100%. The task-based approach (e.g., Gathmann and

Schonberg (2010) and Yamaguchi (2012)) takes an alternative view of occupations and dis-

tances across them. Specifically, it assumes that there are a small number of elementary

tasks, for instance, a set of two tasks: cognitive task and motor task. These tasks are fun-

damental in that they are used in every occupation, but with different combination of task

intensivenesses. A unique combination of task intensivenesses distinguishes one particular

occupation from all the other occupations and therefore defines an occupation. For exam-

ple, a computer programmer’s position requires mainly the cognitive task; a construction

worker’s position demands very intensive motor task; while a cook’s position may be in the

middle: it needs some cognitive task but not as intensive as a computer programmer, and

some motor task but not as intensive as a construction worker. More formally, suppose there

exist n basic tasks and the intensiveness index of a given task can be expressed as a real

number between 0 and 1 (after some normalization) and these indices are comparabe across

occupations, and then every point in the cube [0,1]n ⊂ Rn denotes an occupation.

Given the above definition of an occupation, occupations are no longer uniformly distinct.

A subset of occupations may be more similar to one another than those outside of the

subset, because they have similar combinations of task intensivenesses; some occupations

might be very dissimilar to one another, because they have very different combinations of

task intensivenesses. It follows that the transfer loss of occupational human capital is not

always 100%. It should depend upon the similarity of the source and target occupations. If

they differ a lot, the loss is supposed to be big; if they are really similar, the loss should be

small. In practice, there are different ways to measure the similarity, or “distance” between

a pair of occupations. For instance, Yamaguchi (2012) uses the euclidean distance in a

task space to denote the occupation distance. While in Gathmann and Schonberg (2010),

the authors essentially measure the angle formed by the origin-source occupation ray and

the origin-target occupation ray and consider it the occupation distance.6 In this article, I

follow Gathmann and Schonberg (2010) because I find it extremely intuitive. Specifically, it

reflects the idea that it is the multi-tasking ability, or the ability to handle multiple tasks

simultaneously that is valued in every occupation. When an upgrade or promotion takes

6The actual measure they use is one minus cosine of the angle.


place, the performance requirements rise for all the tasks at the same time. Similarly, when

a downgrade or demotion takes place, the performance requirements fall for all the tasks at

the same time. In these two scenarios, it seems reasonable to assume zero loss of occupational

human capital. This is indeed the case when Gathmann and Schonberg (2010)’s measure is

used, because the source and target occupations lie on the same ray and the distance angle

is zero. Figure 2.1 illustrates the angle measure graphically under the assumption of R2.

With all the task intensiveness indices ranging from zero to one, the angle distance

ranges from 0 (the shortest distance) to π/2 (the longest distance). Please note that it

is important that the task intensiveness indices are cardinal and are comparable across

occupations, because only when the two conditions are met at the same time are the units of

intensiveness, in the task space of occupations, consistent along a given axis and consistent

across axes so that the angle measure is not twisted and as a result, meaningful.

The task-based approach looks at occupations in a new perspective, and thus the occu-

pational human capital, which is measured by tenure, should also be modified accordingly.

Because now in the new framework, when one switches occupation he or she will carry a

fraction of the occupational human capital (measured by tenure) from the source occupation

to the target occupation: the loss is not 100% any longer. Therefore, this tenure needs a new

name, so as to distinguish itself from the conventional occupational tenure concept in the

literature. Gathmann and Schonberg (2010) call it “task tenure”, while I name it “General

Occupational Tenure”, which seems more appropriate. First, what it really measures is a

tenure associated with a particular occupation, more specifically, a target occupation, not

a tenure associated with a particular task. In fact, Gathmann and Schonberg (2010) track

task-specific human capital for every task and, when calculating this literal task tenure the

authors always decompose the occupational tenure first. Second, the adjective “general”

stresses its transferable nature: a portion of it is valued by both source and target occupa-

tions. In this sense, it is general, though not completely general.

How do we calculate the General Occupational Tenure? There are straightforward ways

and black-box ways. Gathmann and Schonberg (2010) use a straightforward method. Sup-

pose a worker is observed to work in an occupation for a period of time. Then this occu-

pational tenure is decomposed into the tenures associated with all the individual tasks, and

each task-related tenure is tracked separately. At the time when there is a need to calculate

the General Occupational Tenure, one composes all the individual task-related tenures to

get the result. This decomposition-composition cycle repeats itself again and again. With

being straightforward as its advantage, this method is a little bit tedious if the number of

basic tasks is big and a worker switches occupation frequently in his lifecycle. Alternatively,

a black-box method is used in this paper. In particular, I assume there exists a Transfer

Rate function decreasing in the occupation distance, which yields what fraction of the oc-


cupational tenure can be transferred from the source to the target occupation given any

pair of occupations. This empirical object offers me a handy tool to track one’s General

Occupational Tenure without tracking his or her individual task-related tenures and avoids

frequent decomposition-composition manipulations. Conceptually, it is reasonable to assume

that the Transfer Rate function is convexly decreasing in the occupation distance with the

following intuition: occupational switches constitute a serious change in an individual’s ca-

reer path; even a small deviation from the source occupation implies a tremendous shift in

the multi-tasking requirement and therefore to a large extent the previous working aptitude

is no longer useful; however, as the deviation becomes bigger, the marginal cost of occupa-

tional switch is decreasing because the bulk of cost has already been incurred by the initial

movements. In some sense, this is analogous to the marginal utility’s evolution when one is

saturating his or her desires.

The specific choice of Transfer Rate function is an empirical issue. This article follows

Xiong (2012) and assumes that

TransRate(θ) = (− 2

πθ + 1)5 (3.2)

Equation (3.2) is convexly decreasing in the occupation distance and based upon the linear

function of f(θ) = − 2πθ + 1. When the occupation distance takes on the smallest value,

namely, θ = 0, the Transfer Rate equals 1, that is, 100% of occupational tenure can be

transferred. When, in theory, the farthest possible switch takes place, namely, θ = π/2, the

Transfer Rate equals 0, namely, nothing is transferable. Xiong (2012) finds that the above

funtional form yields a good calibration result to match his model-generated statistics with

the data. Hence, I take it as the baseline functional form.7 To be more concrete, let me show

an example given in Gathmann and Schonberg (2010).8 They assume there are two basic

tasks called analytical and manual. A worker works in Occupation A (with analytical and

manual intensive indices 0.5 and 0.5, respectively) for one year and then switches to Occu-

pation B (with analytical and manual intensive indices 0.3 and 0.7, respectively). According

to their equations, at the time when the switch takes place, the “task tenure” associated

with Occupation A is 1 × 0.5 × 0.5 + 1 × 0.5 × 0.5/((0.5)2 + (0.5)2) = 1, and the “task

tenure” associated with Occupation B is 1×0.5×0.3+1×0.5×0.7/((0.3)2 +(0.7)2) = 0.862.

Therefore, the Transfer Rate is 0.862/1 = 0.862. If the same transfer happens in my frame-

work, the angle measure of distance between Occupations A and B is arccos(0.5 × 0.3 +

0.5 × 0.7/(√

(0.5)2 + (0.5)2√

(0.3)2 + (0.7)2)) = 0.380. Plug θ into Equation (3.2) and we

get a Transfer Rate of 0.250. It turns out that my framework discounts a switcher’s General

7Numerous Transfer Rate functional forms are experimented on and a relevant discussion is to be foundin Section 3.6.

8See P.16 at Section IIID in their paper.


Occupational Tenure more heavily than that in Gathmann and Schonberg (2010).

With a Transfer Rate function at hand, it is simple to trace a worker’s General Occu-

pational Tenure. Please note that when one talks about the General Occupational Tenure,

there is always a corresponding occupation it is associated with.9 To illustrate how to cal-

culate the General Occupational Tenure, suppose one worker starts his or her career path

by entering Occupation A, and he or she accumulates the General Occupational Tenure (as-

sociated with Occupation A) one for one when he or she works in Occupation A. At some

point in time, this worker switches to a new occupation, Occupation B. Then he or she starts

working with some endowed General Occupational Tenure (associated with Occupation B).

To calculate this endowment, we multiply his or her General Occupational Tenure associated

with Occupation A, with the Transfer Rate determined by the occupation distance between

the source occupation A and the target occupation B. Then on top of the endowment, the

worker accumulates the General Occupational Tenure (associated with Occupation B) one

for one when he or she works in Occupation B. And the process repeats itself until the end

of the worker’s career path. At any time, we need track only two objects: the General Occu-

pational Tenure and its associated occupation. Thus the black-box method saves one a lot

of efforts because there is no longer a need to track all the tenures associated with individual

tasks.

3.3.2 Estimating Occupation-Specific Returns to the General Occupa-

tional Tenure

Because the General Occupational Tenure is necessarily affiliated with an occupation, its

returns should be occupation-specific. Empirically, I modify Equation (3.1) and use the

following econometric model to perform the wage regression:

logw = β1I1 + · · ·+ βnIn + βEduEdu + βEduSqEdu2

+ βExpWorkExp + βExpSqWorkExp2 + βEmpEmpTen

+ βEmpSqEmpTen2 + βOJOJ + βIndIndTen + βIndSqIndTen2

+ βOcc1I1 ×GenOccTen + · · ·+ βOccnIn ×GenOccTen

+ βOccSq1I1 ×GenOccTen2 + · · ·+ βOccSqnIn ×GenOccTen2

+X ′B + ζ (3.3)

In the above regression, as in Equation (3.1), logw is the natural log of real wage; Edu is a

worker’s years of schooling; OJ is a dummy variable which takes on unity if one’s employer

tenure is equal to or greater than 4 months and zero otherwise; WorkExp, EmpTen, and

9More specifically, it is the “target” occupation.


IndTen are a worker’s work experience, employer tenure, and industrial tenure, respectively;

and X consists of the following regressors: 1-digit industry dummies, union dummy, marital

status dummies, region dummies, race dummies, and Rotation Group dummies. In addition,

Ii is the indicator function for Occupation i, and it equals 1 if the worker examined works

in Occupation i and 0 otherwise; GenOccTen is the General Occupational Tenure. And

I continue to apply an IV-GLS approach that uses WorkExp, EmpTen, OJ, IndTen, and

GenOccTen’s deviations from mean as their instruments and allows for serial correlation in

the error term for any given individual.

Implicitly, there is a key difference between wage regressions (3.1) and (3.3). As argued

before, Equation (3.1) sees occupations uniformly distinct and therefore emphasizes their ho-

mogeneous side with the focus on the return’s time-series dimension, or the wage increment

as time passes by. In contrast, Equation (3.3) starts with the view that every occupation is

unique with its special task intensiveness combination, and so stresses occupations’ heteroge-

neous side with the focus on the return’s cross-section dimension, or the absolute magnitude

differences among occupations. Please note in the framework of Equation (3.3), for a given

occupation i, its General Occupational Tenure’s return, in the units of log real wages, has

3 parts: the constant part, βi; the linear part, βOcciGenOccTen; and the quadratic part,

βOccSqiGenOccTen2. Borrowing terms from the fixed cost and the variable cost, I call the

constant part the fixed return and the combination of linear and quadratic parts the variable

return, for the latter depends upon the magnitude of General Occupational Tenure while

the former does not. I will continue with discussions of the regression’s empirical results in

Section 3.5.

3.4 Data and Methods

In this section, I introduce the data that are used to run the wage regression under Equa-

tion (3.3) and that are used to retrieve the occupational characteristics, and the method to

construct task intensiveness indices for individual occupations, in addition to that used to

define 1-, 2-, and 3-digit occupational and industrial classifications.

3.4.1 Survey of Income and Program Participation

SIPP is designed by the U.S. Census Bureau to collect detailed information on income,

employment, and government transfer programs participation of the U.S. civilian nonin-

stitutionalized population. It selects a nationally representative sample of households and

tracks them for several years. SIPP is administered in panels: from time to time, SIPP

selects a new sample called a panel and keeps track of respondents in that panel. Within a

SIPP panel, all the respondents are interviewed every 4 months (called a wave). The detailed


information on individual’s personal characteristics, family composition, assorted incomes,

insurance coverage, program participation, employment and/or business, assets owned is

recorded. Initially, the U.S. Census Bureau plans to start a new panel of around 20,000

households each year and continue a panel for 32 months, but the actual sample size and the

panel duration vary significantly. There are 14 panels so far with the first one the 1984 Panel

and the latest one the 2008 Panel. The number of sampled households varies from 12,425

to 44,200, and the panel duration varies from 12 months to 60 months. SIPP undergoes

an overhaul in 1996 with two most eminent reforms. First, it introduces computer-assisted

interviewing and as a result the data consistency improves greatly for the panels after 1996

than earlier panels. Second, it abandons the overlapping time design, that is, several panels

(with different starting times and ending times) are operated at the same time. As a remedy,

sample size increases significantly for panels after 1996 than those before 1996.

SIPP data have two unique advantages over other widely-used labor market panel data,

such as PSID and National Longitudinal Survey of Youth (NLSY) in serving this paper’s

study purpose. First, SIPP has a higher interview frequency (3 times per year) while most

other popular surveys interview respondents annually. Thus SIPP provides richer labor mar-

ket dynamics information, which in particular enables me to identify occupational mobility

that takes place in the middle of a year. Second, SIPP asks respondents their occupational

tenure in the first wave while other surveys don’t. This information is of vital importance

for the current project as respondents’ direct answer to this question is more accurate than

any indirect imputation that is forced to be applied when other data sets are used.

The choice of the 1996 panel of SIPP (SIPP1996 henceforth) for my econometrical ex-

ercise is based on the following considerations. First, panels after 1996 have a higher data

quality due to the introduction of computer-assisted interviewing. Second, to estimate re-

turns for hundreds of (3-digit level) occupations, large sample size is necessary: there should

be sufficiently many observations in each individual occupation cell to guarantee identifi-

cation. So, recent panels are preferable to earlier panels. Third, the task intensiveness

information for each occupation comes from the Dictionary of Occupational Titles (DOT),

which was released in the 1970’s. So there is a time gap between SIPP data and the DOT.

To make the estimation sensible, we want to minimize the time gap. In this sense, earlier

panels are more suitable than recent panels. As far as all above three factors are concerned,

SIPP1996 is the best compromise.

I impose the sample restrictions on SIPP1996 as follows: male, aged between 18 and

64, not disabled, and not self-employed. For a given worker, only when the following three

conditions are satisfied is his person-wave observation qualified for the wage regressions in

Sections 3.2 and 3.3.2: he is working on a full-time job, that is, the weekly working hours are

no less than 35 hours; his nominal hourly wage is no less than 4.25 dollars, the U.S. federal


minimum wage rate in 1995; and moreover, he holds such a job for at least two waves so

that the IV-GLS method can be applied. In the end, the sample consists of 6,832 individuals

with 45,320 person-wave observations. Summary statistics are listed in Table 3.3. A detailed

description on how to construct various tenure variables can be found in Appendix 4.1.

3.4.2 Dictionary of Occupational Titles

DOT is a large data set created by the U.S. Department of Labor to provide standardized

occupational information for the purposes of matching job applicants with job vacancies.

It contains rich information on requirements and features for over 12,000 finely defined

occupations found in the U.S. labor market. Major part of the the DOT data come from

the job analysts through on-site observation of occupations when they are performed, and

for those that are difficult to observe the data come from surveying related professional

and trade associations. The first edition of DOT was released in 1939, and in 1949, 1965,

and 1977 the following editions II, III, and IV (latest edition) were publicized. For a given

occupation, up to 62 characteristics are recorded which fall into one of 7 broad categories:

worker functions, general education development, specific vocational preparation, aptitudes,

temperaments, physical demands, and environmental conditions.

Many characteristics are recorded using a multi-point rank system. The variable Rea-

soning gives a typical example (descriptions come from U.S. Department of Labor (1972)).

This variable describes an occupation’s requirement on workers’ ability to perform reasoning

tasks and takes on the integer value from 1 (simplest) to 6 (most difficult) with details as

follows:

1. Apply commonsense understanding to carry out simple one- or two-step instructions.

Deal with standardized situations with occasional or no variables in or from these situations

encountered on the job.

2. Apply commonsense understanding to carry out detailed but uninvolved written or

oral instructions. Deal with problems involving a few concrete variables in or from standard-

ized situations.

3. Apply commonsense understanding to carry out instructions furnished in written,

oral, or diagrammatic form. Deal with problems involving several concrete variables in or

from standardized situations.

4. Apply principles of rational systems to solve practical problems and deal with a vari-

ety of concrete variables in situations where only limited standardization exists. Interpret a

variety of instructions furnished in written, oral, diagrammatic, or schedule form.

5. Apply principles of logical or scientific thinking to define problems, collect data, estab-

lish facts, and draw valid conclusions. Interpret an extensive variety of technical instructions

in mathematical or diagrammatic form. Deal with several abstract and concrete variables.


6. Apply principles of logical or scientific thinking to a wide range of intellectual and

practical problems. Deal with nonverbal symbolism (formulas, scientific equations, graphs,

musical notes, etc.) in the most difficult phases. Deal with a variety of abstract and concrete

variables. Apprehend the most abstruse clauses of concepts.

Other characteristics are recorded by a binary variable. Take the variable Climb as an ex-

ample. If the occupation involves its workers’ climbing movement, then the variable takes

on unity and zero otherwise.

3.4.3 Principal Component Analysis

I use the DOT to derive task intensiveness indices for individual occupations. However, if we

consider every characteristic variable captures an individual task, then the task space has too

many dimensions and it will impose too heavy a burden on computation. In fact, if taking

a closer look at all the DOT characteristic variables, one finds many of them are closely

correlated and essentially measure a same thing. For instance, it seems acceptable to say

that the Mathematical variable and the Numerical variable both measure an occupation’s

skill requirement to crunch numbers. In literature, economists use the technique of Principal

Component Analysis (PCA)10 to summarize the DOT information and to lower the task

space dimension. The PCA is based on just the assumption that the information contained

in a large number of variables can be represented by a small number of synthesized variables,

and a set of weight coefficients (called factor loadings) are estimated so that the variation in

the original data is maximized in the framework of synthesized variables.

There exist two different PCA approaches used by previous studies, based on different

assumptions. The first approach assumes that a subset of DOT variables measures only one

task and not other tasks. The second approach assumes that all DOT variables measure all

tasks and these tasks are orthogonally distributed. The former approach requires a priori

knowledge on the nature of DOT variables and fundamental tasks. While this knowledge

is not required by the latter approach, it is sometimes difficult to assign a meaningful task

name to a synthesized variable. Research that applies the first approach include Autor et al.

(2003), Bacolod and Blum (2010) and Yamaguchi (2012). Studies like Ingram and Neumann

(2006) and Poletaev and Robinson (2008) use the second approach. There is no general

conclusion on which approach is better than the other, as can be seen by the fact that

researchers use both methods. This article follows Yamaguchi (2012) closely and takes the

first approach. In Yamaguchi (2012), the author’s assumptions are reasonable, and moreover

he also performs a robustness check using the second approach, which yields very similar

results.

10or factor analysis, a closely related technique, which in practice yields very similar results as the PCAthough conceptually the two techniques are different.


In particular, I follow Yamaguchi (2012) in assuming that there exists a set of two funda-

mental tasks: cognitive task and motor task.11 Moreover, I choose the same subset of DOT

variables as in Yamaguchi (2012) for individual fundamental tasks. Specifically, 11 DOT

variables are assumed to measure only the cognitive task: Data, People, Reasoning, Math-

ematical, Language, Intelligence, Verbal, Numerical, Influencing People, Accepting Respon-

sibility for Direction, and Dealing with People; 15 DOT variables are assumed to measure

only the motor task: Things, Motor Coordination, Finger Dexterity, Manual Dexterity, Eye-

hand-foot Coordination, Spatial Perception, Form Perception, Color Discrimination, Setting

Limits, Tolerance or Standards, Strength, Climb, Stoop, Reach, Talk, and See.

Please note that all DOT characteristics are order variables or dummy variables, and

are thus ordinal. However, as discussed in Section 3.3.1, to use the desired angle measure

of occupation distance, it is important that the intensiveness indices are cardinal for each

individual occupation and are comparable across tasks. In Autor et al. (2003) and Yamaguchi

(2012)12, the authors tackle the issue carefully and use a specific data set to help transform

the original ordinal DOT scores to some cardinal values. The key data they use is the

augmented April 1971 CPS file released by the National Academy of Sciences (2001), in

which experts assign individual DOT occupation codes and characteristics to the 60,441

respondents in the sample. After getting the principal components from the DOT, they

convert them into cardinal-flavored percentile scores13 with the help of employment weights

in the augmented April 1971 CPS file. I tackle the issue in the spirit of their strategy but

with a slight modification. In particular, I convert the original DOT ranking scores into

percentile scores14 using the employment weights in the augmented April 1971 CPS file

before doing PCA. This way I assure that all the obtained task intensive indices fall in the

range of [0, 1] and have a percentile meaning. As mentioned early and to be discussed in

the next subsection, DOT is a much-finer occupational classification than 1-, 2-, and 3-digit

occupational classifications. With DOT-level occupational task indices at hand, it is easy to

aggregate them into the 1-, 2-, and 3-digit levels and calculate corresponding angle distance

measures.

11The choice of the task set is essentially an art and subject to researchers’ discretion. Autor et al. (2003)consider 4 tasks: nonroutine analytic, nonroutine interactive, routine cognitive, and routine manual. InBacolod and Blum (2010), the set consists of cognitive, motor, people, and physical strength. Ingram andNeumann (2006) elect to use a set of 4 tasks: intelligence, fine motor, coordination, and strength. WhereasPoletaev and Robinson (2008)’s task set has 3 elements: general intelligence, fine motor, and physical strength.The specific choice depends firstly on the research objective. Researchers also have other considerations: thenumber of basic tasks should not be too small, otherwise some useful information contained in the DOTwill be wasted; on the other hand, the number should not be too large, either, or the economic model’scomputational cost would be too high.

12He follows the method in Autor et al. (2003).13As a normalization, the scores are divided by 100 so that the results range from 0 to 1.14Again, they are normalized by a division with 100.


Table 3.4 lists summary statistics of the cognitive and motor intensiveness indices for

1-, 2-, and 3-digit occupations, respectively.15 Take 3-digit occupations as an example, the

physician’s position (84) requires most intensive cognitive skill, while the garbage collector’s

position (875) needs least intensive cognitive skill; as for the motor task, the electrical and

electronic equipment assembler’s position (683) is most demanding, whereas the religious

worker, not elsewhere classified (177) is least challenging.

Note that various occupations demonstrate different intensiveness combinations of the

cognitive and motor tasks. The following four 3-digit occupation titles provide four extreme

examples. Veterinarians (86) have high index numbers on both dimensions: 0.970 of cognitive

intensiveness index and 0.980 of motor intensiveness index (this order of task intensiveness

indice is followed by subsequent examples). The position of ushers (462) is an opposite

example with both index numbers low (0.270 and 0.443). Religious worker, not elsewhere

classified (177) gives us an example of high cognitive index and low motor index (0.944,

0.235). Lastly, textile sewing machine operators (744) see a combination of low cognitive

index and high motor index (0.291, 0.915).

Yamaguchi (2012) shows that great heterogeneity of task human capital exists in a given

1-digit occupation title. He finds that there is considerable task complexity variation of

3-digit occupations within a 1-digit occupational aggregate. He decomposes the total task

complexity variance into the within-group and between-group variances and finds that the

former accounts for more than 50% of the total variance for both the cognitive and motor

tasks. He continues to draw 3-digit occupations that belong to two different 1-digit occu-

pational groups on a same scatter plot and finds a significant overlap. So he argues that

the idea of a certain 1-digit occupation is uniformly more skill-demanding than the other

is debatable. Table 3.4 in the current article reinforces the above finding. Note that the

1-, 2-, and 3-digit occupations that require the most intensive motor task are construction

and extractive occupations (16), veterinarians (27), and electrical and electronic equipment

assemblers (683), respectively. However, the 3-digit occupation 683 does not belong to the

2-digit occupational group 27, and moreover, the 2-digit occupation 27 does not belong to

the 1-digit occupational aggregate 16, either.

Table 3.5 lists summary statistics on the angle measure of occupation distance based on

Equation (3.2). They are in the units of radians. The average non-self distances for 1-, 2-,

and 3-digit classifications are 0.256, 0.298, and 0.456, respectively. For 1-digit occupations,

the largest distance exists between social scientists, social workers, religious workers, and

lawyers (4) and handlers, equipment cleaners, helpers, and laborers (20), and the smallest

15The results come from the PCA analysis of the April 1971 CPS file. However, some occupation titles donot exist in the CPS data. So the number of observed titles are sometimes less than that listed in the variousclassifications. For 1-digit occupations, we have 20 observed vs. 20 listed; for 2-digit occupations, we have 56observed vs. 58 listed; and for 3-digit occupations, we have 423 observed vs. 501 listed.


distance between registered nurses, pharmacists, dietitians, therapists, and physician’s assis-

tants (7) and technologists and technicians, except health (10). Among 2-digit occupations,

the position of lawyers and judges (21) and the position of handlers, equipment cleaners, and

laborers (87) lie farthest to each other, while the position of teachers, except postsecondary

institutions (23) and the position of insurance, securities, real estate and business service

sales occupations (41) lie closest to each other. For 3-digit occupations, the position of re-

ligious workers, not elsewhere classified (177) and the position of garbage collectors (875)

generate the maximum distance, whereas the position of chemical technicians (224) and the

position of general office clerks (379) constitute the minimum distance. The distances of

some above pairs may not seem intuitive. But please bear in mind that this distance mea-

sure is based on the intensive combination of basic tasks, not the knowledge occupations

make use of. And it is very difficult, if not impossible, to compare the distances among

several sets of knowledge.

3.4.4 Occupational and Industrial Classifications

An occupational (industrial) classification is a collection of occupation (industry) titles and

is usually organized using a hierarchical structure. For instance, the U.S. Census 1990 occu-

pational classification, based upon which SIPP codes its respondents’ occupation affiliation,

lists 501 finest occupational titles, and they are aggregated into 13 Major Groups and fur-

ther into 6 Summary Groups. Because SIPP 1996 Panel is the main data set which is

used in the current paper and upon which the wage regressions are run, I take the 501-title

classification as a reference and call it 3-digit classification. The Census 1990 occupational

classification in turn is built upon the 1980 Standard Occupational Classification (SOC1980)

system. SOC1980 is an occupational classification of 4 layers: 664 Units, 224 Minor Groups,

58 Major Groups and finally 20 Divisions. In this article, I take the 20 Divisions and the

58 Major Groups as 1-digit and 2-digit occupations, respectively. Because the Census 1990

occupational classification is derived from SOC1980, the crosswalk is readily available. Ap-

pendix 4.2 lists the Census 1990 occupational classification and Appendix 4.3 shows the

SOC1980 system.

Similarly, SIPP adopts the U.S. Census 1990 industrial classification as the reference to

code its respondents’ industry affiliation. The collection consists of 235 finest industrial titles,

and they are aggregated into 13 Major Groups. I take the 235-title classification as a reference

and call it 3-digit classification. The Census 1990 industrial classification is developed from

the 1987 Standard Industrial Classification (SIC1987) system, which, analogous to SOC1980,

holds a four-layer hierarchical structure: 1503 Industries, 504 Industry Groups, 82 Major

Groups and finally 10 Divisions. The current paper takes the 10 Divisions and the 82 Major

Groups as 1-digit and 2-digit industries, respectively. Again, since the Census 1990 industrial


classification is derived from SIC1987, the crosswalk is readily available.

There is a need to match the occupational titles in the SIPP1996 and in the DOT: the

former is the main data set on which wage regressions are based, and the latter is the source

to extract task intensiveness indices. I solve this technical difficulty in an indirect manner.

The key is the augmented April CPS file. This data set can bridge the gap since in this file

every worker’s occupation affiliation is coded using both DOT classification and the 1977

Standard Occupational Classification (SOC1977). Recall that the Census 1990 occupational

classification, which SIPP1996 adopts as its benchmark, is developed from SOC1980. And

SOC1980 is a revised version of SOC1977 and they are actually very similar. Therefore, I

have been able to construct a crosswalk between them16, so that the SIPP data and DOT

are finally linked.

3.5 Empirical Results of Returns to the General Occupational

Tenure

3.5.1 Regression Results

Recall that the KM wage regression framework is modified to accommodate the new concept

of General Occupational Tenure and to estimate its returns.17 The new wage regression

inherits an important feature from the KM regression, that is, in the presence of occupational

tenure variables (now the General Occupational Tenure) in a wage regression, the returns to

the industrial tenure and the employer tenure are not important. As an example, Table 3.6

lists the coefficient estimates for the regression under 1-digit occupational classification. The

coefficients on employer tenure, industrial tenure and their squared terms are in general

insignificant with very large p-values. This is also true for regressions under 2- and 3-digit

occupational classifications.

As discussed in Section 3.3.2, the returns to the General Occupational Tenure is naturally

occupation-specific and consist of the fixed and variable parts. For any given occupation,

the fixed return is the estimated intercept, and the variable return involves the General

Occupational Tenure and its squared terms. Table 3.7 lists the empirical results.18 First,

there is considerable variation of the returns to General Occupational Tenure among different

occupations. The mean of General Occupational Tenure is roughly 13 years for all three

occupational classifications and so I use it as a benchmark value. The summary statistics

16The crosswalk between SOC1980 and SOC1977 is available upon request.17Refer to Section 3.3.2 for details.18All the values are calculated on the basis of point estimates of coefficients on the General Occupational

Tenure-related variables in the wage regression Equation (3.3), with insignificant estimates (10% significancelevel) changed to zeros.


are calculated for a worker who has 13 years of General Occupational Tenure, for 1-, 2-

, and 3-digit occupational classifications19. Take 3-digit occupations as an example, as

Section A of Table 3.7 shows, given 13 years of General Occupational Tenure, the largest

return a 3-digit occupation generates is 14.60 while the smallest -14.41. Compared to the

mean return(0.88), its standard deviation (1.65) is big. Second, among the two components

that constitute the total return, the fixed return in general dominates the variable return.

Section B of Table 3.7 lists the fixed return. In Section C, the variable return is calculated

according to 3 different levels of the General Occupational Tenure: 6 years (roughly 25

percentile of the General Occupational Tenure for all three occupational classifications),

12 years (50 percentile), and 19 years (75 percentile). It is evident that the mean fixed

return is significantly larger than the mean variable return at all three levels, under all

three occupational classifications. Moreover, the remark is strengthened by the fact that

the fixed return has a smaller standard deviation than the variable return. Third, the fixed

return and the variable return is inversely related and thus a tradeoff exists between the two

components. Section D shows the coefficient of correlation between the fixed and variable

returns at the 3 different levels that appear in Section C for 1-, 2-, and 3-digit occupational

classifications. As can be seen in the table, in almost all cases, the two returns are strongly

negatively correlated.

3.5.2 Discussion

The proposed extended wage regression in this paper, namely, using the General Occu-

pational Tenure to replace the conventional occupational tenure, starts from a conceptual

innovation, where occupations are analyzed using the task-based approach. But empirically,

is General Occupational Tenure a better alternative to conventional occupational tenure?

This subsection takes a closer look at this issue.

Firstly let us look at the distribution of conventional occupational tenure for a given level

of General Occupational Tenure. Recall that the General Occupational Tenure comes from

two sources: heritage from the previous occupations, and accumulation from the current

occupation. If the accumulation part accounts for a big chunk, then the substitution of

the General Occupational Tenure for the conventional occupational tenure will not make

much difference. On the other hand, if the heritage part plays an important role, then the

replacement is meaningful.

Table 3.8 lists some summary statistics of the conventional occupational tenure, namely

the accumulation part, given a specific General Occupational Tenure level, for 1-, 2-, and

19The number of identifiable occupations is listed on the last row of Table 3.7. Some occupations’ returncoefficients cannot be identified because too few observations fall in their cells. This problem is especiallypronounced for the 3-digit occupational classification due to its large number of occupation titles. In the end,19 1-digit occupations, 37 2-digit occupations, and 274 3-digit occupations are identified.


3-digit occupations. In particular, for each General Occupational Tenure level, the table

shows the mean and the coefficient of variation of the conventional occupational tenure.

Moreover, the table also contains a statistic which equals the 50 percentile of the conventional

occupational tenure divided by the given General Occupational Tenure. This statistic reveals

an upperbound share of the accumulated part for half of the workers (or equivalently, one

minus this statistic reveals a lowerbound share of the inherited part for half of the workers).

And again, three representative General Occupational Tenure levels are considered in the

table: 6 years (25 percentile), 12 years (50 percentile), and 19 years (75 percentile).

Table 3.8 provides supportive evidence of the important role the heritage part plays. For

instance, given 6 years of General Occupational Tenure, the mean accumulated part is only

around 2.7 years for all three occupational classifications, or roughly 40% of the General

Occupational Tenure. Moreover, the conventional occupational tenure demonstrates consid-

erable dispersion with the coefficients of variation varying around 0.80 for all classifications.

Furthermore, half of the workers accumulate less than one third of the given 6 years of Gen-

eral Occupational Tenure, or equivalently, half of the workforce obtains at least two thirds

of the General Occupational Tenure from heritage. Although the table shows that as the

General Occupational Tenure increases, importance of the accumulated part rises with its

dispersion shrinking, the inherited part’s role still cannot be ignored.

Secondly, I run some nested wage regressions which include both General Occupational

Tenure and conventional occupational tenure as independent variables. It is found that, in the

presence of General Occupational Tenure variables, the estimated coefficients on conventional

occupational tenure variables are basically not significant. This finding tends to support that

variation of the General Occupational Tenure is more important in accounting for variation

of the log real wage and hence the General Occupational Tenure is a more suitable variable

in a wage regression than the conventional occupational tenure.

More specifically, two groups of nested wage regressions are performed. The first group

is a constrained nested wage regression by adding the General Occupational Tenure variable

and its squared and cubed terms in the regression function (3.1). It is constrained in the sense

that the return coefficients are restricted to be same across occupations. The second group

is an unconstrained nested wage regression by adding the conventional occupational tenure

variable and its squared term in the regression function (3.3). It is unconstrained because like

(3.3), the return coefficients are occupation-specific and can be different across occupations.

However, there is a limitation for the nested regressions: Altonji and Shakotko (1987)’s

instrumental variable method cannot be used in this context. Because the instrument for the

General Occupational Tenure is always the same as that for the conventional occupational

tenure, the number of instruments are less than the number of endogenous variables and

the system is underidentified. Therefore, only GLS is applied to take care of individual-


level serial correlations. It is evident that the estimates would be biased. However, the

aim of this exercise is not to obtain a set of consistent estimated coefficients. Due to that

estimated coefficients on the General Occupational Tenure variables and on the conventional

occupational tenure variables tend to be biased in the same direction, the biased estimates

can still provide some clues on which occupational tenure variables, General or conventional,

is more relevant in a wage regression.

Table 3.9 lists coefficient estimates on conventional occupational tenure variables and

General Occupational Tenure variables for the constrained nested wage regression under all

three occupational classifications. In general, the coefficient estimates on General Occu-

pational Tenure variables are significant; whereas the coefficient estimates on conventional

occupational tenure variables are not. Table 3.10 reports summary statistics for the un-

constrained nested wage regressions. Because the return coefficients are occupation-specific

and the numbers of occupations are big under 2- and 3-digit classifications, the table only

reports the percentages of significant estimates for different classes of tenure variables. It

can be seen that for 1- and 3-digit occupations, the percentage of significant estimates is

higher for General Occupational Tenure variables than for conventional occupational tenure

variables, but for 2-digit occupations, the conclusion is reversed. Jointly, Tables 3.9 and

3.10 tend to convey the information that the wage variability is more closely related with

variability of the General Occupational Tenure and therefore this new tenure variable is a

more appropriate independent variable in a wage regression.

Recall that magnitude of the General Occupational Tenure depends crucially on the

Transfer Rate function. Given a specific Transfer Rate function, Equation (3.2), we have the

previous three conclusions. A natural question is, do they still hold under other Transfer

Rate function assumptions? To answer this question, I consider an extreme case and show

that a family of convexly decreasing Transfer Rate functions converge to it as the discounting

becomes more and more heavy. For this limiting case, the three patterns are still valid. This

tends to support a yes answer to the question raised, at least for the special set of Transfer

Rate functions.

3.6 One Limiting Case and Convergence

Consider the conventional assumption in the occupational literature, namely, occupational

human capital is specific and not transferable across occupations. This simplifying scenario

constitutes a limiting case of the General Occupational tenure. When the discounting of a

Transfer Rate function becomes extremely heavy, an individual suffers 100% loss of occu-

pational human capital, measured by tenure, when he or she switches occupation. In this

case, the wage regression (3.3) would reduce to an occupation-specific version of KM wage


regression (3.1).20

I perform the occupation-specific version of KM wage regression for this extreme case and

report the empirical results in Table 3.11, which has exactly the same layout as Table 3.7. It

is obvious that all three conclusions based on Table 3.7 continue to hold for Table 3.11, the

extreme case. Firstly, the variation of returns to (General) Occupational Tenure is still large

across occupations. As Table 3.11’s Section A shows, while 1- and 2-digit occupations see

slightly less spreading in their return distributions than in Table 3.7, the 3-digit occupation’s

return demonstrates much more variation. Secondly, in most cases, the fixed return is

significantly larger than the variable return. Comparing the numbers in Section B and in

Section C, we find this is true for all three classifications. Thirdly, the fixed return and the

variable return is still negatively correlated in Table 3.11 (Section D), though not as strongly

as in Table 3.7 and so the tradeoff continues to exist.

As argued in Section 3.3.1, it is intuitively appealing to assume a convexly decreasing

Transfer Rate function. In fact, the baseline Transfer Rate function TransRate(θ) = (− 2πθ+

1)5 is a convex transformation of the linear function TransRate(θ) = − 2πθ + 1 which yields

1 when θ equals 0 and 0 when θ equals π/2. It is very easy to make this linear function

more or less convex by changing its exponent, and so I focus on the family of Transfer Rate

functions that are convexly transformed from the above linear function. In particular, I raise

the power of the linear function to 3, 7, 11, and 15, respectively to achieve increasingly heavy

discounting, in addition to 5, the baseline value. The extreme case can be reached by setting

the power to plus infinity, where TransRate(θ) equals 1 when θ equals 0 and 0 when θ takes

on all the other values.

Table 3.12, again under all three occupational classifications, lists the statistics based on

various Transfer Rate function assumptions and clearly demonstrates a converging tendency.

Each row indicates a Transfer Rate function and the numbers in the leftmost column are the

values of the exponents for the corresponding Transfer Rate functions. There are 4 columns

of statistics for a given occupational classification: from left to right, the first column lists

the average non-self transfer rate which shows the degree of discounting, and as the value

of the exponent turns bigger the discounting becomes heavier and thus the mean transfer

rate smaller; columns 2 to 4 list the euclidean distances between the point estimates based

on a given Transfer Rate function with the indicated exponent value and the point esti-

mates based on the extreme-case Transfer Rate function (exponent equal to plus infinity)

for occupation-specific constant coefficients, linear coefficients, and quadratic coefficients,

respectively. Recall that for the wage regression of General Occupational Tenure, Equa-

tion (3.3), the return for an individual occupation takes a quadratic structure. Because the

constant, linear, and quadratic coefficients display obviously different orders of magnitude,

20More accurately, the cubed terms on the right hand side of the KM regression equation are removed.


the three groups’ distances are calculated separately. As can be seen in Table 3.12, with the

exponent value approaching plus infinity, the distances of the constant, linear, and quadratic

coefficients become smaller and smaller, for 1-digit and 2-digit occupational classifications.

Under 3-digit occupational classification, the pattern firmly holds for the constant distance

and generally holds for the linear distance. However, the expected convergence does not

appear for the quadratic distance, given the experimented values of exponents. It could be

the case that the converging pattern resumes for the quadratic distance as larger exponents

are tried.

Tables 3.11 and 3.12, in conjunction with Table 3.7, reveal an appealing finding. With the

increase of the exponent value, the investigated family of Transfer Rate functions converge

to the limiting-case Transfer Rate function with a plus infinity exponent. And the limiting

case shares the same three features with the baseline model. Therefore, it is reasonable to

argue that the three patterns of the returns to General Occupational Tenure apply to the

whole family of Transfer Rate functions. It may not be too mistaken to conjecture that these

results will also hold for a larger set of convexly decreasing Transfer Rate functions, an even

larger generalization.

The generalization basically lowers importance of specific choice of the baseline exponent

value, 5. As mentioned in Section 3.3.1, the value of 5 helps yield desirable calibration results

in Xiong (2012). In this section, I further show that this is a satisfactory choice in a goodness-

of-fit sense. Table 3.13 lists the Root Mean Squared Errors (MSE) for various choices of the

exponent value in the Transfer Rate function under all three occupational classifications. In

general, a smaller Root MSE indicates better goodness-of-fit of a regression.21 According

to Table 3.13, it seems that as the exponent rises or the discounting becomes heavier, the

Root MSE displays an inverse U shape for 1-digit occupations, a monotonically increasing

trend for 2-digit occupations, and a U shape for 3-digit occupations. Because intuitively it

is believed that a Transfer Rate function is convexly decreasing, the exponent value should

be greater than unity. Among the choices listed in the table, 5 is acceptable under all three

occupational classifications in that it helps generate a relatively low Root MSE in all three

scenarios.

3.7 Conclusion

In this article, I study the returns to occupational human capital under the assumption

that all occupations are uniquely distinct and that occupational human capital is partially

transferable. I name the associated tenure variable “General Occupational Tenure” and

21The extended wage regression (3.3) does not contain a constant regressor and thus the conventional Rsquared cannot be used here.


propose an empirical Transfer Rate function that relates its transferable portion with the

occupation distance. Combining SIPP data and task information from the DOT, I perform

a generalized wage regression under 1-, 2-, and 3-digit occupational classifications and find 3

common patterns: returns to the General Occupational Tenure demonstrate great variation

across occupations; the fixed return generally dominates the variable return; and the two

are always negatively correlated. Finally I generalize this result to show that they actually

apply to a large family of convexly decreasing Transfer Rate functions by showing that as the

discounting becomes heavier these functions converge to a limiting case where the 3 patterns

hold.

In the current paper, I start by repeating KM’s econometrical exercise using the SIPP

data and get a result analogous to theirs: including a worker’s occupational tenure in a wage

regression makes other tenure variables less important. This lends support to the view that

human capital tends to be occupation-specific rather than firm-specific or industry-specific.

I continue to extend KM’s framework in two important aspects: the occupation-specific

returns are allowed and the occupational tenure is assumed to be partially transferable.

The two extensions are actually based on one fundamental deviation from the traditional

simplifying assumption that all occupations are uniformly distinct and so the occupational

human capital is equally non-transferable. The new underlying assumption stresses the het-

erogeneity existing across occupations. Because occupations are very different, the returns

should be occupation-specific. Because occupations are not uniformly distinct, some occu-

pations are closer to each other and others are farther, and thus the occupation distance

should determine how transferable the occupational human capital is between a given pair

of occupations. Then the Transfer Rate function comes into play in tracking the General

Occupational Tenure.

I face two technical challenges in this project. Firstly, in constructing the favored distance

measure, it is important that the intensiveness indice are cardinal and comparable across

tasks. I use the augmented April 1971 CPS file to tackle this problem. This data set codes

individuals’ occupations using both the DOT and other popular occupational classifications

and it contains the DOT rank information for every respondent. Using the employment

distribution across occupations, I compute the percentile of each DOT score for each respon-

dent. Then I use the percentile-based value to replace the ordinal DOT score. Secondly, to

link the SIPP data on which the wage regression is based and the DOT information where the

task intensiveness indice come from, I must find a crosswalk between the two occupational

classification systems. Although there is no direct crosswalk, I use an indirect approach.

SIPP adopts the Census 1990 classification which in turn is based on the SOC1980, whereas

April 1971 CPS contains SOC1977 codes. The SOC1980 is an update of SOC1977 and they

are similar, so I am able to construct a crosswalk between them.


Note that the limiting case for the convergence is a special version of the KM wage

regression. It assumes absolute no-transfer of human capital across occupations, same as

KM do. But it allows for occupation-specific occupational returns. In this sense, it is a

generalization of KM and KM provide an “average” estimate of occupational returns. But

obviously, Equation (3.3) is a further generalization.

In future research, if one can prove that a good shock to occupational returns dissi-

pates quickly as occupation distance increases and can find an empirical relationship exists

between occupational return shocks and occupation distances, then a scalar (occupation

distance) can replace the occupation title in an economic model, and therefore a discrete

occupational choice model can be turned into a continuous occupational choice model. As

a result, computational constraint will no longer be a major concern and the occupation set

in an model can be greatly enlarged to be closer to the real economy.


Table 3.1: KM Wage Regressions Estimates

1-Digit 2-Digit 3-Digit

EmpTen -0.00197 -0.00188 -0.000857(0.00203) (0.00209) (0.00212)

EmpTenSq 0.0000581 0.0000527 0.0000173(0.0000821) (0.0000854) (0.0000872)

WorkExp 0.0848∗∗∗ 0.0856∗∗∗ 0.0858∗∗∗

(0.00799) (0.00763) (0.00782)WorkExpSq -0.00148∗∗∗ -0.00156∗∗∗ -0.00156∗∗∗

(0.000340) (0.000335) (0.000344)WorkExpCb 0.00000962∗ 0.0000106∗∗ 0.0000101∗∗

(0.00000504) (0.00000502) (0.00000512)IndTen 0.00431 0.00155 0.00178

(0.00669) (0.00671) (0.00723)IndTenSq -0.000607 -0.000325 -0.000501

(0.000573) (0.000581) (0.000587)IndTenCb 0.00000717 0.00000148 0.00000626

(0.0000123) (0.0000124) (0.0000121)OccTen 0.0102 0.0127∗∗ 0.0112

(0.00628) (0.00642) (0.00688)OccTenSq -0.00118∗∗ -0.00139∗∗ -0.00112∗

(0.000562) (0.000572) (0.000574)OccTenCb 0.0000235∗ 0.0000282∗∗ 0.0000224∗

(0.0000122) (0.0000123) (0.0000120)OJ 0.00353 0.00328 0.00327

(0.00277) (0.00281) (0.00288)

Observations 45320 44098 42346Individuals 6832 6680 6467

NOTES: The dependent variable is log real wage. Other covariates include an intercept, years of schoolingwith squared term, 1-digit occupation and industry dummies, union dummy, marital status dummies, regiondummies, race dummies, and Rotation Group dummies. IV-GLS estimation method is used. Refer toRegression Equation (3.1) in the text. Standard errors are in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗

p < 0.01.


Table 3.2: Returns to Tenure, KM Wage Regressions


2 years 5 years 8 years 2 years 5 years 8 years 2 years 5 years 8 yearsOccupation .016 .024 .018 .020 .032 .027 .018 .031 .029

(.135) (.231) (.474) (.064) (.119) (.285) (.123) (.176) (.294)Industry .006 .007 -.001 .002 -.000 -.008 .002 -.003 -.015

(.583) (.742) (.978) (.874) (.992) (.773) (.897) (.907) (.624)Employer -.004 -.008 -.012 -.004 -.008 -.012 -.002 -.004 -.006

(.329) (.324) (.322) (.362) (.353) (.347) (.677) (.663) (.649)

NOTES: P-values are in parentheses. Returns to various tenures are calculated based on the coefficientestimates of Regression Equation (3.1).

Table 3.3: Descriptive Statistics

Mean Std. Dev.

Age 38.30 10.61Years of schooling 12.23 2.27Percent married (%) 66.30Percent unionized (%) 28.47Percent white (%) 86.10WorkExp (yrs) 19.40 10.62EmpTen (yrs) 5.48 7.54OccTen (yrs)

1-digit 9.31 9.532-digit 9.19 9.523-digit 8.93 9.53

IndTen (yrs)1-digit 9.55 9.542-digit 9.29 9.543-digit 9.19 9.55

GOccTen (yrs)1-digit 13.86 9.102-digit 13.36 9.083-digit 13.49 9.14

NOTES: WorkExp, EmpTen, OccTen, IndTen, and GOccTen refer to labor market work experience, employertenure, occupational tenure, industrial tenure, and General Occupational Tenure, respectively. In principle,the 1-digit mean of GOccTen should be the largest, and the 3-digit mean the smallest, among the three.However, when wage regressions are performed, some occupations have too few observations in the cell andthus the corresponding workers are deleted. So the mean GOccTen is calculated based on 3 different samplesfor the 3 occupational classifications. Specifically, the 2-digit sample is a subset of the 1-digit sample, andthe 3-digit sample is a subset of the 2-digit sample.


Table 3.4: Summary Statistics of Task Intensiveness Indices

Cognitive Task


Max 1 (6) 1 (26) 1 (84)Min .257 (20) .256 (87) .194 (875)Mean .737 .736 .646Std. dev. .218 .227 .229

Motor Task


Max 1 (16) 1 (27) 1 (683)Min .456 (4) .354 (21) .235 (177)Mean .798 .711 .742Std. dev. .164 .186 .174

Obs 20 56 423

NOTES: Results come from the PCA analysis of the augmented April 1971 CPS file. In parentheses arecorresponding occupation codes. In particular, for 1-digit occupations, 6 refers to Health Diagnosing andTreating Practitioners; 20 refers to Handlers, Equipment Cleaners, Helpers, and Laborers; 16 refers to Con-struction and Extractive Occupations; and 4 refers to Social Scientists, Social Workers, Religious Workers,and Lawyers. For 2-digit occupations, 26 refers to Physicians and Dentists; 87 refers to Handlers, Equip-ment Cleaners, and Laborers; 27 refers to Veterinarians; and 21 refers to Lawyers and Judges. For 3-digitoccupations, 84 refers to Physicians; 875 refers to Garbage Collectors; 683 refers to Electrical and ElectronicEquipment Assemblers; and 177 refers to Religious Workers, n.e.c. (not elsewhere classified). In the data,the numbers of observed occupation titles for 1-, 2-, and 3-digit occupational classifications are 20, 56, and423, respectively.

Table 3.5: Summary Statistics of Angle Occupation Distances (in radians)


Max .804 (4 20) .894 (21 87) 1.050 (177 875)Min .007 (7 10) .0001 (23 41) .00000315 (224 379)Mean .256 .298 .456Std. dev. .175 .211 .278

NOTES: Theoretically, the angle occupation distance ranges from 0 to π/2. In parentheses are correspondingoccupation pairs in terms of occupation codes. In particular, for 1-digit occupations, 4 refers to SocialScientists, Social Workers, Religious Workers, and Lawyers; 20 refers to Handlers, Equipment Cleaners,Helpers, and Laborers; 7 refers to Registered Nurses, Pharmacists, Dietitians, Therapists, and Physician’sAssistants; and 10 refers to Technologists and Technicians, Except Health. For 2-digit occupations, 21 refersto Lawyers and Judges; 87 refers to Handlers, Equipment Cleaners, and Laborers; 23 refers to Teachers,Except Postsecondary Institutions; and 41 refers to Insurance, Securities, Real Estate and Business ServiceSales Occupations. For 3-digit occupations, 177 refers to Religious Workers, n.e.c. (not elsewhere classified);875 refers to Garbage Collectors; 224 refers to Chemical Technicians; and 379 refers to General Office Clerks.


Table 3.6: Generalized Wage Regression Estimates, 1-Digit

EmpTen -0.000281 GOccTen17 -0.0101 GOccTenSq19 0.000314(0.0021) (0.0125) (0.0003)

EmpTenSq -0.0000635 GOccTen18 -0.0157∗ GOccTenSq20 -0.000356(0.0001) (0.0093) (0.0003)

WorkExp 0.0779∗∗∗ GOccTen19 -0.0104 Occ1 0.819∗∗∗

(0.0067) (0.0108) (0.1423)WorkExpSq -0.000909∗∗∗ GOccTen20 0.0110 Occ2 1.028∗∗∗

(0.0001) (0.0109) (0.3471)IndTen 0.00330 GOccTenSq1 0.000955∗∗ Occ3 0.128

(0.0037) (0.0004) (0.3747)IndTenSq -0.000406∗∗∗ GOccTenSq2 -0.000661 Occ4 0.534

(0.0001) (0.0010) (0.3983)GOccTen1 -0.0308∗∗ GOccTenSq3 -0.00203 Occ5 1.334∗∗∗

(0.0121) (0.0012) (0.2522)GOccTen2 0.00505 GOccTenSq4 -0.00232 Occ7 1.595∗∗∗

(0.0379) (0.0029) (0.2576)GOccTen3 0.102∗∗ GOccTenSq5 0.00124∗∗ Occ8 0.493∗

(0.0452) (0.0005) (0.2697)GOccTen4 0.0411 GOccTenSq7 0.000931 Occ9 0.719∗∗∗

(0.0850) (0.0007) (0.2347)GOccTen5 -0.0704∗∗∗ GOccTenSq8 -0.000855 Occ10 0.933∗∗∗

(0.0182) (0.0009) (0.1782)GOccTen7 -0.0621∗∗ GOccTenSq9 -0.00209∗∗ Occ11 0.665∗∗∗

(0.0269) (0.0010) (0.1348)GOccTen8 0.0268 GOccTenSq10 0.000865∗ Occ12 0.554∗∗∗

(0.0346) (0.0005) (0.1380)GOccTen9 0.0479 GOccTenSq11 0.000546 Occ13 0.548∗∗∗

(0.0318) (0.0004) (0.1212)GOccTen10 -0.0255 GOccTenSq12 0.0000617 Occ14 0.414∗∗∗


(0.0130) (0.0004) (0.1246)GOccTen12 -0.00183 GOccTenSq14 -0.000348 Occ16 0.771∗∗∗

(0.0133) (0.0004) (0.1235)GOccTen13 -0.0145 GOccTenSq15 0.000476∗ Occ17 0.727∗∗∗

(0.0118) (0.0003) (0.1408)GOccTen14 0.0157 GOccTenSq16 -0.0000282 Occ18 0.724∗∗∗

(0.0163) (0.0003) (0.1159)GOccTen15 -0.0230∗∗ GOccTenSq17 0.000359 Occ19 0.584∗∗∗


(0.0099) (0.0003) (0.1207)

Obs 45314

NOTES: The dependent variable is log real wage. Other covariates include years of schooling with squaredterm, Old Job dummy, 1-digit industry dummies, union dummy, marital status dummies, region dummies,race dummies, and Rotation Group dummies. IV-GLS estimation method is used. Occi, GOccTeni, andGOccTenSqi indicate the coefficients before Ii, Ii×GenOccTen, and Ii×GenOccTen2, respectively. Refer toRegression Equation (3.3) in the text. Return coefficients for Occupation 6 (Health Diagnosing and TreatingPractitioners) cannot be identified due to too few observations. Standard errors are in parentheses. ∗ p < 0.10,∗∗ p < 0.05, ∗∗∗ p < 0.01.


Table 3.7: Returns to General Occupational Tenure, TransRate(θ) = (− 2πθ + 1)5


A. Total return (13 yrs, in log real wages)Max 1.324 2.073 14.596Min 0 0 -14.414Mean .647 .669 .881St. dev. .286 .394 1.649

B. Fixed return (in log real wages) .703 .639 .840(.382) (.516) (1.465)

C. Variable return (in log real wages)6 yrs -.029 -.0001 -.030

(.196) (.256) (.720)12 yrs -.052 .024 .024

(.386) (.478) (1.652)19 yrs -.073 .083 .193

(.611) (.711) (4.032)

D. Corr(fixed, variable)6 yrs -.772 -.730 -.79312 yrs -.753 -.711 -.59419 yrs -.716 -.667 -.312

Obs 19 37 274

NOTES: Standard deviations are in parentheses. The mean of General Occupational Tenure is roughly 13years for 1-, 2-, and 3-digit occupational classifications. And the 25, 50, and 75 percentiles of General Occupa-tional Tenure are roughly 6 years, 12 years, and 19 years, respectively, for all three occupational classifications.Corr(fixed, variable) denotes the coefficient of correlation between the fixed and variable returns at a givenGeneral Occupational Tenure level. The numbers of occupations whose returns are identifiable in a wageregression are 19, 37, and 274 for 1-, 2-, and 3-digit occupations, respectively.

Table 3.8: Summary Statistics of OccTen for a Given GOccTen


6 yrs GOccTenMean(%) 2.75(42.3%) 2.69(41.3%) 2.64(40.6%)CV 0.79 0.79 0.8250pctl OccTen/GOccTen 35.9% 30.8% 30.8%



NOTES: OccTen and GOccTen refer to the conventional occupational tenure and the General OccupationalTenure, respectively. CV denotes the coefficient of variation. The 25, 50, and 75 percentiles of GeneralOccupational Tenure are roughly 6 years, 12 years, and 19 years, respectively, for all three occupationalclassifications.


Table 3.9: Constrained Nested Wage Regressions Estimates


OccTen 0.00163 -0.00872 -0.00592(0.00570) (0.00568) (0.00589)

OccTenSq 0.000243 0.00107∗∗ 0.000946∗

(0.000480) (0.000477) (0.000491)OccTenCb -0.00000279 -0.0000169∗ -0.0000156

(0.00000964) (0.00000940) (0.0000100)GOccTen 0.0159∗∗∗ 0.0182∗∗∗ 0.0157∗∗∗

(0.00572) (0.00552) (0.00572)GOccTenSq -0.00100∗∗ -0.00131∗∗∗ -0.00106∗∗

(0.000407) (0.000410) (0.000423)GOccTenCb 0.0000136∗ 0.0000197∗∗ 0.0000150∗

(0.00000771) (0.00000799) (0.00000840)

Observations 47935 46708 43073Individuals 8405 8254 7713

NOTES: The dependent variable is log real wage. Other covariates include an intercept, years of schoolingwith squared term, employer tenure with squared term, work experience with cubed and squared terms,industrial tenure with cubed and squared terms, 1-digit occupation and industry dummies, union dummy,marital status dummies, region dummies, race dummies, Old Job dummy, and Rotation Group dummies.GLS estimation method is used. Standard errors are in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.

Table 3.10: Fractions of Significant Estimates for Unconstrained Nested Wage Regressions(%)


OccTen 10.53 18.92 35.90OccTenSq 15.79 32.43 34.43GOccTen 26.32 13.51 38.83GOccTenSq 36.84 16.22 39.93

NOTES: The dependent variable is log real wage. Covariates include years of schooling with squared term,employer tenure with squared term, work experience with squared term, industrial tenure with squared term,1-digit industry dummies, union dummy, marital status dummies, region dummies, race dummies, Old Jobdummy, and Rotation Group dummies; and following occupation-specific variables: intercept, conventionaloccupational tenure with squared term, and General Occupational Tenure with squard term. GLS estimationmethod is used.


Table 3.11: Returns to General Occupational Tenure, Limiting Case


A. Total return (9 yrs, in log real wages)Max 1.317 1.526 38.291Min .330 0 -19.534Mean .714 .666 .754St. dev. .236 .271 3.005

B. Fixed return (in log real wages) .734 .673 .828(.218) (.261) (.830)

C. Variable return (in log real wages)1 yr .004 .001 .0003

(.033) (.041) (.100)5 yrs .004 .002 -.020

(.147) (.174) (.986)15 yrs -.099 -.038 -.218

(.408) (.363) (8.427)

D. Corr(fixed, variable)1 yr -.421 -.411 -.6765 yrs -.461 -.439 -.33015 yrs -.483 -.446 -.104

Obs 19 37 274

NOTES: Standard deviations are in parentheses. The results are for the limiting case where the GeneralOccupational Tenure is assumed to be not transferable. In this scenario, the mean of General OccupationalTenure is roughly 9 years for 1-, 2-, and 3-digit occupational classifications. And the 25, 50, and 75 per-centiles of General Occupational Tenure are roughly 1 year, 5 years, and 15 years, respectively, for all threeoccupational classifications. Corr(fixed, variable) denotes the coefficient of correlation between the fixed andvariable returns at a given General Occupational Tenure level. The numbers of occupations whose returnsare identifiable in a wage regression are 19, 37, and 274 for 1-, 2-, and 3-digit occupations, respectively.


Table 3.12: Euclidean Distances to Limiting Case’s Coefficients

Exp1-Digit 2-Digit 3-Digit

Trans D cons D lin D qudr Trans D cons D lin D qudr Trans D cons D lin D qudr3 .617 1.365 .091 .0076 .574 2.263 .187 .0128 .591 20.228 1.899 .63595 .478 1.148 .069 .0074 .439 2.025 .151 .0127 .456 14.406 2.052 .62907 .386 .813 .063 .0061 .353 1.817 .142 .0085 .369 12.485 2.410 .682411 .272 .735 .034 .0021 .254 1.581 .137 .0066 .265 9.743 1.817 .697015 .206 .702 .032 .0016 .199 1.494 .131 .0063 .206 8.410 1.715 .7088

NOTES: Each row corresponds to a Transfer Rate function. Exp is the value of the exponent for a given TransferRate function, i.e. t in TransRate(θ) = (− 2

πθ + 1)t, with 5 the baseline value in this article. Trans denotes the

average non-self transfer rate which shows the degree of discounting, and as the value of the exponent turns bigger thediscounting becomes heavier and thus the mean transfer rate smaller. D cons, D lin, and D qudr list the euclideandistances between the point estimates based on a given Transfer Rate function with the indicated exponent valueand the point estimates based on the extreme-case Transfer Rate function (exponent equal to plus infinity) foroccupation-specific constant coefficients, linear coefficients, and quadratic coefficients, respectively.

Table 3.13: Root MSEs for Generalized Wage Regressions

Exp1-Digit 2-Digit 3-Digit

Trans√MSE Trans

√MSE Trans

√MSE

1 .837 .4134 .810 .4203 .820 .47813 .617 .4183 .574 .4218 .591 .45645 .478 .4231 .439 .4259 .456 .45677 .386 .4267 .353 .4292 .369 .459311 .272 .4306 .254 .4329 .265 .463515 .206 .4321 .199 .4344 .206 .4656∞ 0 .4293 0 .4350 0 .4648

Obs 45314 44024 39795

NOTES: Each row corresponds to a Transfer Rate function. Exp is the value of the exponent for a given TransferRate function, i.e. t in TransRate(θ) = (− 2

πθ + 1)t, with 5 the baseline value in this article. Trans denotes the

average non-self transfer rate which shows the degree of discounting, and as the value of the exponent turns biggerthe discounting becomes heavier and thus the mean transfer rate smaller.

√MSE is the Root Mean Squared Error

for the generalized wage regression with a given exponent value. In general, a smaller Root MSE indicates bettergoodness-of-fit of a regression.

Chapter 4

Appendices

4.1 Procedures for Constructing Tenure Variables

The sample restrictions on SIPP1996 are as follows: male, aged between 18 and 64, not

disabled, and not self-employed. For a given worker, only when the following three conditions

are satisfied is his person-wave observation qualified for the wage regressions: he is working

on a full-time job, namely, the weekly working hours are no less than 35 hours; his nominal

hourly wage is no less than 4.25 dollars, the U.S. federal minimum wage rate in 1995; and

moreover, he holds such a job for at least two waves so that the IV-GLS method can be

applied. For a given individual, his labor market information is examined wave by wave.

I first construct WorkExp, the labor market work experience. In the data, some workers

are observed to enter the labor market at as early as 15 years old. But some occupations,

especially under 3-digit classifications, have explicit or implicit restrictions to young workers

under 18. So in this paper, only work experience after 18 years old are considered. I initialize

a worker’s WorkExp by Age - 6 - Edu if his schooling years are no less than 12 and by Age

- 18 otherwise (Edu is the years of schooling). After that, as long as the worker is observed

to work full time for a wave, his WorkExp is increased by 1/3 year.

The second constructed tenure variable is EmpTen, the employer tenure. For a current

incumbent worker, I initialize his EmpTen using his job’s start date information. Specifically,

it equals the start date of the interviewing wave minus the start date of the job. After that,

as long as the employment relationship remains, the EmpTen is incremented by 1/3 year for

every passing wave. If a worker is observed to start working for a new employer in a given

wave, I initialize his EmpTen to zero. The subsequent EmpTen should equal the working

time in the starting wave: I use the wave’s end date to subtract the job’s start date. After

that, as long as the employment relationship remains, the EmpTen is incremented by 1/3

year for every passing wave. I exclude transitory full-time jobs from the regressions, which

112

Chapter 4. Appendices 113

are defined by EmpTen less than 8 months (2 waves). Because the IV-GLS method requires

a minimum of 2 observations for a given employer, but transitory job holders have only one.

Then I consider OccTen, the conventional occupational tenure. Recall that SIPP asks

respondents for their occupational tenure information directly in the first wave, and I ini-

tialize OccTen with that value. After that, as long as the occupation affiliation does not

change, I increase the OccTen in each following wave by the corresponding EmpTen, even

when the employer changes. In case a worker is observed to start a new occupation, the

OccTen is reinitialized to zero. And the tracking rule stays the same as before. Note that

OccTen naturally has at least 2 observations for any given occupation, as it inherits this

feature from EmpTen.

I continue to construct IndTen, the industrial tenure. Unlike OccTen, the industrial

tenure information is never solicited from its respondents by SIPP. I initialize IndTen with

OccTen’s first value. After that, as long as the industry affiliation does not change, I increase

the IndTen in each following wave by the corresponding EmpTen, even when the employer

changes. In case a worker is observed to start working in a new industry, the IndTen is

reinitialized to zero. And the tracking rule is the same as before. Analogous to OccTen,

IndTen naturally has at least 2 observations for any given industry, as it inherits this feature

from EmpTen.

I do a consistency check after the tenure variable initializations. Logically, WorkExp

should be no less than OccTen (IndTen) and OccTen (IndTen) should be no less than

EmpTen. I take WorkExp as a reference, as it is derived from Age and Age should generally

be recorded accurately. If OccTen’s (IndTen’s) initial value is greater than the corresponding

WorkExp, I reevaluate OccTen (IndTen) to WorkExp. Similarly, if EmpTen’s initial value is

greater than the corresponding OccTen, I reevaluate EmpTen to OccTen.

Lastly, I construct GOccTen, the General Occupational Tenure, with the help of OccTen.

Theoretically, I need to know a worker’s occupation history since his entry into the labor

market to obtain his GOccTen. However, for a large number of workers, I observe them

only in the middle of their career path in the SIPP. To initialize GOccTen, I multiply the

difference of initial WorkExp and initial OccTen with the average non-self Transfer Rate.

The idea is that, before one’s first observed occupational tenure, he is assumed to have done

some “average” occupational switch. The way to track GOccTen is very simple, as discussed

in the text, when an occupational switch takes place, we discount the current GOccTen

using the Transfer Rate determined by the occupation distance between the source and

target occupations to get the new GOccTen; when there is not an occupational switch, the

GOccTen is incremented by the actual working time in the interviewing wave: the job’s end

date minus the wave’s start date if an employer switch happens, 1/3 year otherwise.


4.2 1990 Census of Population Occupation Classification Sys-

tem

The list presents the occupational classification developed for the 1990 Census of Population

and Housing. There are 501 categories for the employed with 1 additional category for the

experienced unemployed and 3 additional categories for the Armed Forces. These categories

are grouped into 6 summary groups and 13 major groups. The classification is developed

from the 1980 Standard Occupational Classification (SOC1980). “n.e.c.” is the abbreviation

for not elsewhere classified. In parentheses are corresponding SOC1980 codes. (Source: SIPP

1993 Panel, Longitudinal File Codebook, Appendix A-4)

1990

Census Occupation category

code

MANAGERIAL AND PROFESSIONAL SPECIALTY OCCUPATIONS

Executive, Administrative, and Managerial Occupations

3 Legislators (111)

4 Chief executives and general administrators, public administration (112)

5 Administrators and officials, public administration (1132-1139)

6 Administrators, protective services (1131)

7 Financial managers (122)

8 Personnel and labor relations managers (123)

9 Purchasing managers (124)

13 Managers, marketing, advertising, and public relations (125)

14 Administrators, education and related fields (128)

15 Managers, medicine and health (131)

16 Postmasters and mail superintendents (1344)

17 Managers, food serving and lodging establishments (1351)

18 Managers, properties and real estate (1353)

19 Funeral directors (pt 1359)

21 Managers, service organizations, n.e.c. (127, 1352, 1354, pt 1359)

22 Managers and administrators, n.e.c. (121, 126, 132-1343, 136-139)

Management Related Occupations

23 Accountants and auditors (1412)

24 Underwriters (1414)

25 Other financial officers (1415, 1419)

26 Management analysts (142)

27 Personnel, training, and labor relations specialists (143)

28 Purchasing agents and buyers, farm products (1443)

29 Buyers, wholesale and retail trade except farm products (1442)

33 Purchasing agents and buyers, n.e.c. (1449)

34 Business and promotion agents (145)

35 Construction inspectors (1472)

36 Inspectors and compliance officers, except construction (1473)


37 Management related occupations, n.e.c. (149)

Professional Specialty Occupations

Engineers, Architects, and Surveyors

43 Architects (161)

Engineers

44 Aerospace (1622)

45 Metallurgical and materials (1623)

46 Mining (1624)

47 Petroleum (1625)

48 Chemical (1626)

49 Nuclear (1627)

53 Civil (1628)

54 Agricultural (1632)

55 Electrical and electronic (1633, 1636)

56 Industrial (1634)

57 Mechanical (1635)

58 Marine and naval architects (1637)

59 Engineers, n.e.c. (1639)

63 Surveyors and mapping scientists (164)

Mathematical and Computer Scientists

64 Computer systems analysts and scientists (171)

65 Operations and systems researchers and analysts (172)

66 Actuaries (1732)

67 Statisticians (1733)

68 Mathematical scientists, n.e.c. (1739)

Natural Scientists

69 Physicists and astronomers (1842, 1843)

73 Chemists, except biochemists (1845)

74 Atmospheric and space scientists (1846)

75 Geologists and geodesists (1847)

76 Physical scientists, n.e.c. (1849)

77 Agricultural and food scientists (1853)

78 Biological and life scientists (1854)

79 Forestry and conservation scientists (1852)

83 Medical scientists (1855)

Health Diagnosing Occupations

84 Physicians (261)

85 Dentists (262)

86 Veterinarians (27)

87 Optometrists (281)

88 Podiatrists (283)

89 Health diagnosing practitioners, n.e.c. (289)

Health Assessment and Treating Occupations

95 Registered nurses (29)

96 Pharmacists (301)

97 Dietitians (302)


Therapists

98 Respiratory therapists (3031)

99 Occupational therapists (3032)

103 Physical therapists (3033)

104 Speech therapists (3034)

105 Therapists, n.e.c. (3039)

106 Physicians assistants (304)

Teachers, Postsecondary

113 Earth, environmental, and marine science teachers (2212)

114 Biological science teachers (2213)

115 Chemistry teachers (2214)

116 Physics teachers (2215)

117 Natural science teachers, n.e.c. (2216)

118 Psychology teachers (2217)

119 Economics teachers (2218)

123 History teachers (2222)

124 Political science teachers (2223)

125 Sociology teachers (2224)

126 Social science teachers, n.e.c. (2225)

127 Engineering teachers (2226)

128 Mathematical science teachers (2227)

129 Computer science teachers (2228)

133 Medical science teachers (2231)

134 Health specialties teachers (2232)

135 Business, commerce, and marketing teachers (2233)

136 Agriculture and forestry teachers (2234)

137 Art, drama, and music teachers (2235)

138 Physical education teachers (2236)

139 Education teachers (2237)

143 English teachers (2238)

144 Foreign language teachers (2242)

145 Law teachers (2243)

146 Social work teachers (2244)

147 Theology teachers (2245)

148 Trade and industrial teachers (2246)

149 Home economics teachers (2247)

153 Teachers, postsecondary, n.e.c. (2249)

154 Postsecondary teachers, subject not specified

Teachers, Except Postsecondary

155 Teachers, prekindergarten and kindergarten (231)

156 Teachers, elementary school (232)

157 Teachers, secondary school (233)

158 Teachers, special education (235)

159 Teachers, n.e.c. (236, 239)

163 Counselors, educational and vocational (24)

Librarians, Archivists, and Curators


164 Librarians (251)

165 Archivists and curators (252)

Social Scientists and Urban Planners

166 Economists (1912)

167 Psychologists (1915)

168 Sociologists (1916)

169 Social scientists, n.e.c. (1913, 1914, 1919)

173 Urban planners (192)

Social, Recreation, and Religious Workers

174 Social workers (2032)

175 Recreation workers (2033)

176 Clergy (2042)

177 Religious workers, n.e.c. (2049)

Lawyers and Judges

178 Lawyers (211)

179 Judges (212)

Writers, Artists, Entertainers, and Athletes

183 Authors (321)

184 Technical writers (398)

185 Designers (322)

186 Musicians and composers (323)

187 Actors and directors (324)

188 Painters, sculptors, craft-artists, and artist printmakers (325)

189 Photographers (326)

193 Dancers (327)

194 Artists, performers, and related workers, n.e.c. (328, 329)

195 Editors and reporters (331)

197 Public relations specialists (332)

198 Announcers (333)

199 Athletes (34)

TECHNICAL, SALES, AND ADMINISTRATIVE SUPPORT OCCUPATIONS

Technicians and Related Support Occupations

Health Technologists and Technicians

203 Clinical laboratory technologists and technicians (362)

204 Dental hygienists (363)

205 Health record technologists and technicians (364)

206 Radiologic technicians (365)

207 Licensed practical nurses (366)

208 Health technologists and technicians, n.e.c. (369)

Technologists and Technicians, Except Health

Engineering and Related Technologists and Technicians

213 Electrical and electronic technicians (3711)

214 Industrial engineering technicians (3712)

215 Mechanical engineering technicians (3713)

216 Engineering technicians, n.e.c. (3719)

217 Drafting occupations (372)


218 Surveying and mapping technicians (373)

Science Technicians

223 Biological technicians (382)

224 Chemical technicians (3831)

225 Science technicians, n.e.c. (3832, 3833, 384, 389)

Technicians; Except Health, Engineering, and Science

226 Airplane pilots and navigators (825)

227 Air traffic controllers (392)

228 Broadcast equipment operators (393)

229 Computer programmers (3971, 3972)

233 Tool programmers, numerical control (3974)

234 Legal assistants (396)

235 Technicians, n.e.c. (399)

Sales Occupations

243 Supervisors and proprietors, sales occupations (40)

Sales Representatives, Finance and Business Services

253 Insurance sales occupations (4122)

254 Real estate sales occupations (4123)

255 Securities and financial services sales occupations (4124)

256 Advertising and related sales occupations (4153)

257 Sales occupations, other business services (4152)

Sales Representatives, Commodities Except Retail

258 Sales engineers (421)

259 Sales representatives, mining, manufacturing, and wholesale (423, 424)

Sales Workers, Retail and Personal Services

263 Sales workers, motor vehicles and boats (4342, 4344)

264 Sales workers, apparel (4346)

265 Sales workers, shoes (4351)

266 Sales workers, furniture and home furnishings (4348)

267 Sales workers; radio, TV, hi-fi, and appliances (4343, 4352)

268 Sales workers, hardware and building supplies (4353)

269 Sales workers, parts (4367)

274 Sales workers, other commodities (4345, 4347, 4354, 4356, 4359,4362, 4369)

275 Sales counter clerks (4363)

276 Cashiers (4364)

277 Street and door-to-door sales workers (4366)

278 News vendors (4365)

Sales Related Occupations

283 Demonstrators, promoters and models, sales (445)

284 Auctioneers (447)

285 Sales support occupations, n.e.c. (444, 446, 449)

Administrative Support Occupations, Including Clerical

Supervisors, Administrative Support Occupations

303 Supervisors, general office (4511,4513,4514,4516,4519,4529)

304 Supervisors, computer equipment operators (4512)

305 Supervisors, financial records processing (4521)


306 Chief communications operators (4523)

307 Supervisors; distribution, scheduling, and adjusting clerks (4522, 4524-4528)

Computer Equipment Operators

308 Computer operators (4612)

309 Peripheral equipment operators (4613)

Secretaries, Stenographers, and Typists

313 Secretaries (4622)

314 Stenographers (4623)

315 Typists (4624)

Information Clerks

316 Interviewers (4642)

317 Hotel clerks (4643)

318 Transportation ticket and reservation agents (4644)

319 Receptionists (4645)

323 Information clerks, n.e.c. (4649)

Records Processing Occupations, Except Financial

325 Classified-ad clerks (4662)

326 Correspondence clerks (4663)

327 Order clerks (4664)

328 Personnel clerks, except payroll and timekeeping (4692)

329 Library clerks (4694)

335 File clerks (4696)

336 Records clerks (4699)

Financial Records Processing Occupations

337 Bookkeepers, accounting, and auditing clerks (4712)

338 Payroll and timekeeping clerks (4713)

339 Billing clerks (4715)

343 Cost and rate clerks (4716)

344 Billing, posting, and calculating machine operators (4718)

Duplicating, Mail and Other Office Machine Operators

345 Duplicating machine operators (4722)

346 Mail preparing and paper handling machine operators (4723)

347 Office machine operators, n.e.c. (4729)

Communications Equipment Operators

348 Telephone operators (4732)

353 Communications equipment operators, n.e.c. (4733, 4739)

Mail and Message Distributing Occupations

354 Postal clerks, ext. mail carriers (4742)

355 Mail carriers, postal service (4743)

356 Mail clerks, ext. postal service (4744)

357 Messengers (4745)

Material Recording, Scheduling, and Distributing Clerks

359 Dispatchers (4751)

363 Production coordinators (4752)

364 Traffic, shipping, and receiving clerks (4753)

365 Stock and inventory clerks (4754)


366 Meter readers (4755)

368 Weighers, measurers, checkers and samplers (4756, 4757)

373 Expediters (4758)

374 Material recording, scheduling, and distributing clerks, n.e.c. (4759)

Adjusters and Investigators

375 Insurance adjusters, examiners, and investigators (4782)

376 Investigators and adjusters, except insurance (4783)

377 Eligibility clerks, social welfare (4784)

378 Bill and account collectors (4786)

Miscellaneous Administrative Support Occupations

379 General office clerks (463)

383 Bank tellers (4791)

384 Proofreaders (4792)

385 Data-entry keyers (4793)

386 Statistical clerks (4794)

387 Teachers aides (4795)

389 Administrative support occupations, n.e.c. (4787, 4799)

SERVICE OCCUPATIONS

Private Household Occupations

403 Launderers and ironers (503)

404 Cooks, private household (504)

405 Housekeepers and butlers (505)

406 Child care workers, private household (506)

407 Private household cleaners and servants (502, 507, 509)

Protective Service Occupations

Supervisors, Protective Service Occupations

413 Supervisors, firefighting and fire prevention occupations (5111)

414 Supervisors, police and detectives (5112)

415 Supervisors, guards (5113)

Firefighting and Fire Prevention Occupations

416 Fire inspection and fire prevention occupations (5122)

417 Firefighting occupations (5123)

Police and Detectives

418 Police and detectives, public service (5132)

423 Sheriffs, bailiffs, and other law enforcement officers (5134)

424 Correctional institution officers (5133)

Guards

425 Crossing guards (5142)

426 Guards and police, exc. public service (5144)

427 Protective service occupations, n.e.c. (5149)

Service Occupations, Except Protective and Household

Food Preparation and Service Occupations

433 Supervisors, food preparation and service occupations (5211)

434 Bartenders (5212)

435 Waiters and waitresses (5713)

436 Cooks (5214. 5215)


438 Food counter, fountain and related occupations (5216)

439 Kitchen workers, food preparation (5217)

443 Waiters/waitresses assistants (5218)

444 Miscellaneous food preparation occupations (5219)

Health Service Occupations

445 Dental assistants (5232)

446 Health aides, except nursing (5233)

447 Nursing aides, orderlies, and attendants (5236)

Cleaning and Building Service Occupations, except Household

448 Supervisors, cleaning and building service workers (5241)

449 Maids and housemen (5242,5249)

453 Janitors andcleaners (5244)

454 Elevator operators (5245)

455 Pest control occupations (5246)

Personal Service Occupations

456 Supervisors, personal service occupations (5251)

457 Barbers (5252)

458 Hairdressers and cosmetologists (5253)

459 Attendants, amusement and recreation facilities (5254)

461 Guides (5255)

462 Ushers (5256)

463 Public transportation attendants (5257)

464 Baggage porters and bellhops (5262)

465 Welfare service aides (5263)

466 Family child care providers (pt 5264)

467 Early childhood teachers assistants (pt 5264)

468 Child care workers, n.e.c. (pt 5264)

469 Personal service occupations, n.e.c. (5258, 5269)

FARMING, FORESTRY, AND FISHING OCCUPATIONS

Farm Operators and Managers

473 Farmers, except horticultural (5512-5514)

474 Horticultural specialty farmers (5515)

475 Managers, farms, except horticultural (5522-5524)

476 Managers, horticultural specialty farms (5525)

Other Agricultural and Related Occupations

Farm Occupations, Except Managerial

477 Supervisors, farm workers (5611)

479 Farm workers (5612-5617)

483 Marine life cultivation workers (5618)

484 Nursery workers (5619)

Related Agricultural Occupations

485 Supervisors, related agricultural occupations (5621)

486 Groundskeepers and gardeners, except farm (5622)

487 Animal caretakers, except farm (5624)

488 Graders and sorters, agricultural products (5625)

489 Inspectors, agricultural products (5627)


Forestry and Logging Occupations

494 Supervisors, forestry, and logging workers (571)

495 Forestry workers, except logging (572)

496 Timber cutting and logging occupations (573, 579)

Fishers, Hunters, and Trappers

497 Captains and other officers, fishing vessels (pt 8241)

498 Fishers (583)

499 Hunters and trappers (584)

PRECISION PRODUCTION, CRAFT, AND REPAIR OCCUPATIONS

Mechanics and Repairers

503 Supervisors, mechanics and repairers (60)

Mechanics and Repairers, Except Supervisors

Vehicle and Mobile Equipment Mechanics and Repairers

505 Automobile mechanics (pt 6111)

506 Automobile mechanic apprentices (pt 6111)

507 Bus, truck, and stationary engine mechanics (6112)

508 Aircraft engine mechanics (6113)

509 Small engine repairers (6114)

514 Automobile body and related repairers (6115)

515 Aircraft mechanics, ext. engine (6116)

516 Heavy equipment mechanics (6117)

517 Farm equipment mechanics (6118)

518 Industrial machinery repairers (613)

519 Machinery maintenance occupations (614)

Electrical and Electronic Equipment Repairers

523 Electronic repairers, communications and industrial equipment (6151, 6153, 6155)

525 Data processing equipment repairers (6154)

526 Household appliance and power tool repairers (6156)

527 Telephone line installers and repairers (6157)

529 Telephone installers and repairers (6158)

533 Miscellaneous electrical and electronic equipment repairers (6152, 6159)

534 Heating, air conditioning, and refrigeration mechanics (616)

Miscellaneous Mechanics and Repairers

535 Camera, watch, and musical instrument repairers (6171,6172)

536 Locksmiths and safe repairers (6173)

538 Office machine repairers (6174)

539 Mechanical controls and valve repairers (6175)

543 Elevator installers and repairers (6176)

544 Millwrights (6178)

547 Specified mechanics and repairers, n.e.c. (6177, 6179)

549 Not specified mechanics and repairers

Construction Trades

Supervisors, Construction Occupations

553 Supervisors; brickmasons, stonemasons, and tile setters (6312)

554 Supervisors, carpenters and related workers (6313)

555 Supervisors, electricians and power transmission installers (6314)


556 Supervisors; painters, paperhangers, and plasterers (6315)

557 Supervisors; plumbers, pipefitters, and steamfitters (6316)

558 Supervisors, construction n.e.c. (6311, 6318)

Construction Trades, Except Supervisors

563 Brickmasons and stonemasons (pt 6412, pt 6413)

564 Brickmason and stonemason apprentices (pt 6412, pt 6413)

565 Tile setters, hard and soft (pt 6414, pt 6462)

566 Carpet installers (pt 6462)

567 Carpenters (pt 6422)

569 Carpenter apprentices (pt 6422)

573 Drywall installers (6424)

575 Electricians (pt 6432)

576 Electrician apprentices (pt 6432)

577 Electrical power installers and repairers (6433)

579 Painters, construction and maintenance (6442)

583 Paperhangers (6443)

584 Plasterers (6444)

585 Plumbers, pipefitters, and steamfitters (pt 645)

587 Plumber, pipefitter, and steamfitter apprentices (pt 645)

588 Concrete and terrazzo finishers (6463)

589 Glaziers (6464)

593 Insulation workers (6465)

594 Paving, surfacing, and tamping equipment operators (6466)

595 Roofers (6468)

596 Sheetmetal duct installers (6472)

597 Structural metal workers (6473)

598 Drillers, earth (6474)

599 Construction trades, n.e.c. (6467, 6475, 6476, 6479)

Extractive Occupations

613 Supervisors, extractive occupations (632)

614 Drillers, oil well (652)

615 Explosives workers (653)

616 Mining machine operators (654)

617 Mining occupations, n.e.c. (656)

Precision Production Occupations

628 Supervisors, production occupations (67, 71)

Precision Metal Working Occupations

634 Tool and die makers (pt 6811)

635 Tool and die maker apprentices (pt 6811)

636 Precision assemblers, metal (6812)

637 Machinists (pt 6813)

639 Machinist apprentices (pt 6813)

643 Boilermakers (6814)

644 Precision grinders, filers, and tool sharpeners (6816)

645 Patternmakers and model makers, metal (6817)

646 Lay-out workers (6821)


647 Precious stones and metals workers (Jewelers) (6822, 6866)

649 Engravers, metal (6823)

653 Sheet metal workers (pt 6824)

654 Sheet metal worker apprentices (pt 6824)

655 Miscellaneous precision metal workers (6829)

Precision Woodworking Occupations

656 Patternmakers and model makers, wood (6831)

657 Cabinet makers and bench carpenters (6832)

658 Furniture and wood finishers (6835)

659 Miscellaneous precision woodworkers (6839)

Precision Textile, Apparel, and Furnishings Machine Workers

666 Dressmakers (pt 6852, pt 7752)

667 Tailors (pt 6852)

668 Upholsterers (6853)

669 Shoe repairers (6854)

674 Miscellaneous precision apparel and fabric workers (6856, 6859, pt 7752)

Precision Workers, Assorted Materials

675 Hand molders and shapers, except jewelers (6861)

676 Patternmakers, lay-out workers, and cutters (6862)

677 Optical goods workers (6864, pt 7477, pt 7677)

678 Dental laboratory and medical appliance technicians (6865)

679 Bookbinders (6844)

683 Electrical and electronic equipment assemblers (6867)

684 Miscellaneous precision workers, n.e.c. (6869)

Precision Food Production Occupations

686 Butchers and meat cutters (6871)

687 Bakers (6872)

688 Food batchmakers (6873,6879)

Precision Inspectors, Testers, and Related Workers

689 Inspectors, testers, and graders (6881, 828)

693 Adjusters and calibrators (6882)

Plant and System Operators

694 Water and sewage treatment plant operators (691)

695 Power plant operators (pt 693)

696 Stationary engineers (pt 693, 7668)

699 Miscellaneous plant and system operators (692, 694, 695, 696)

OPERATORS, FABRICATORS, AND LABORERS

Machine Operators, Assemblers, and Inspectors

Machine Operators and Tenders, Except Precision

Metalworking and Plastic Working Machine Operators

703 Lathe and turning machine set-up operators (7312)

704 Lathe and turning machine operators (7512)

705 Milling and planing machine operators (7313, 7513)

706 Punching and stamping press machine operators (7314, 7317,7514, 7517)

707 Rolling machine operators (7316, 7516)

708 Drilling and boring machine operators (7318, 7518)


709 Grinding, abrading, buffing, and polishing machine operators (7322, 7324, 7522)

713 Forging machine operators (7319, 7519)

714 Numerical control machine operators (7326)

715 Miscellaneous metal, plastic, stone, and glass working machine operators (7329, 7529)

717 Fabricating machine operators, n.e.c. (7339, 7539)

Metal and Plastic Processing Machine Operators

719 Molding and casting machine operators (7315, 7342, 7515,7542)

723 Metal plating machine operators (7343, 7543)

724 Heat treating equipment operators (7344, 7544)

725 Miscellaneous metal and plastic processing machine operators (7349, 7549)

Woodworking Machine Operators

726 Wood lathe, routing, and planing machine operators (7431,7432. 7631, 7632)

727 Sawing machine operators (7433, 7633)

728 Shaping and joining machine operators (7435, 7635)

729 Nailing and tacking machine operators (7636)

733 Miscellaneous woodworking machine operators (7434, 7439, 7634. 7639)

Printing Machine Operators

734 Printing press operators (7443, 7643)

735 Photoengravers and lithographers (6842, 7444, 7644)

736 Typesetters and compositors (6841, 7642)

737 Miscellaneous printing machine operators (6849, 7449, 7649)

Textile, Apparel, and Furnishings Machine Operators

738 Winding and twisting machine operators (7451, 7651)

739 Knitting, looping, taping, and weaving machine operators (7452, 7652)

743 Textile cutting machine operators (7654)

744 Textile sewing machine operators (7655)

745 Shoe machine operators (7656)

747 Pressing machine operators (7657)

748 Laundering and dry cleaning machine operators (6855, 7658)

749 Miscellaneous textile machine operators (7459, 7659)

Machine Operators, Assorted Materials

753 Cementing and gluing machine operators (7661)

754 Packaging and filling machine operators (7462, 7662)

755 Extruding and forming machine operators 7463, 7663)

756 Mixing and blending machine operators (7664)

757 Separating, filtering, and clarifying machine operators (7476, 7666, 7676)

758 Compressing and compacting machine operators (7467, 7667)

759 Painting and paint spraying machine operators (7669)

763 Roasting and baking machine operators, food (7472, 7672)

764 Washing, cleaning, and pickling machine operators (7673)

765 Folding machine operators (7474, 7674)

766 Furnace, kiln, and oven operators, ext. food (7675)

768 Crushing and grinding machine operators (pt 7477, pt 7677)

769 Slicing and cutting machine operators (7478, 7678)

773 Motion picture projectionists (pt 7479)

774 Photographic process machine operators (6863, 6868, 7671)


777 Miscellaneous machine operators, n.e.c. (pt 7479, 7665, 7679)

779 Machine operators, not specified

Fabricators, Assemblers, and Hand Working Occupations

783 Welders and cutters (7332, 7532, 7714)

784 Solderers and brazers (7333, 7533, 7717)

785 Assemblers (772, 774)

786 Hand cutting and trimming occupations (7753)

787 Hand molding, casting, and forming occupations (7754, 7755)

789 Hand painting, coating, and decorating occupations (7756)

793 Hand engraving and printing occupations (7757)

795 Miscellaneous hand working occupations (7758, 7759)

Production Inspectors, Testers, Samplers, and Weighers

796 Productioninspectors, checkers, and examiners (782, 787)

797 Production testers (783)

798 Production samplers and weighers (784)

799 Graders and sorters, ext. agricultural (785)

Transportation and Material Moving Occupations

Motor Vehicle Operators

803 Supervisors, motor vehicle operators (8111)

804 Truck drivers (8212-8214)

806 Driver-sales workers (8218)

808 Bus drivers (8215)

809 Taxicab drivers and chauffeurs (8216)

813 Parking lot attendants (874)

814 Motor transportation occupations, n.e.c. (8219)

Transportation Occupations, Except Motor Vehicles

Rail Transportation Occupations

823 Railroad conductors and yardmasters (8113)

824 Locomotive operating occupations (8232)

825 Railroad brake, signal, and switch operators (8233)

826 Rail vehicle operators, n.e.c. (8239)

Water Transportation Occupations

828 Ship captains and mates, except fishing boats (pt 8241, 8242)

829 Sailors and deckhands (8243)

833 Marine engineers (8244)

834 Bridge, lock, and lighthouse tenders (8245)

Material Moving Equipment Operators

843 Supervisors, material moving equipment operators (812)

844 Operating engineers (8312)

845 Longshore equipment operators (8313)

848 Hoist and winch operators (8314)

849 Crane and tower operators (8315)

853 Excavating and loading machine operators (8316)

855 Grader, dozer, and scraper operators (8317)

856 Industrial truck and tractor equipment operators (8318)

859 Miscellaneous material moving equipment operators (8319)


Handlers, Equipment Cleaners, Helpers, and Laborers

864 Supervisors, handlers, equipment cleaners, and laborers, n.e.c. (85)

865 Helpers, mechanics and repairers (863)

Helpers, Construction and Extractive Occupations

866 Helpers, construction trades (8641-8645, 8648)

867 Helpers, surveyor (8646)

868 Helpers, extractive occupations (86.5)

869 Construction laborers (871)

874 Production helpers (861, 862)

Freight, Stock, and Material Handlers

875 Garbage collectors (8722)

876 Stevedores (8723)

877 Stock handlers and baggers (8724)

878 Machine feeders and offbearers (8725)

883 Freight, stock, and material handlers, n.e.c. (8726)

885 Garage and service station related occupations (873)

887 Vehicle washers and equipment cleaners (875)

888 Hand packers and packagers (8761)

889 Laborers, except construction (8769)

MILITARY OCCUPATIONS

903 Commissioned Officers and Warrant Officers

904 Non-commissioned Officers and Other Enlisted Personnel

905 Military occupation, rank not specified

EXPERIENCED UNEMPLOYED NOT CLASSIFIED BY OCCUPATION

909 Last worked 1984 or earlier


4.3 1980 Standard Occupational Classification System

The SOC1980 consists of 20 Divisions, 58 Majors, 224 Minors, and 664 Units for the employed

workers. In addition, there are one Division/Major/Minor/Unit for the Military Occupations

and one Division/Major/Minor/Unit for the Miscellaneous Occupations. In the list, 2-, 3-,

and 4-digit codes refer to Majors, Minors, and Units, respectively. Titles without numerical

codes correspond to Divisions. (Source: U.S. Bureau of Labor Statistics, SOC Information

Desk)

Code Title

Executive, Administrative and Managerial occupations

11 Officials and Administrators, Public Administration

111 Legislators

112 Chief Executives and General Administrators

113 Officials and Administrators, Government Agencies

1131 Judicial, Public Safety and Corrections Administrators

1132 Human Resources Program Administrators

1133 Natural Resources Program Administrators

1134 Rural, Urban, and Community Development Program Administrators

1135 Public Finance, Taxation, and Other Monetary Program Administrators

1139 Officials and Administrators, Public Administration, Not Elsewhere Classified

12-13 Officials and Administrators, Other

121 General Managers and Other Top Executives

122 Financial Managers

123 Personnel and Labor Relations Managers

124 Purchasing Managers

125 Managers; Marketing, Advertising, and Public Relations

126 Managers; Engineering, Mathematics, and Natural Science

127 Managers; Social Sciences and Related Fields

128 Administrators; Education and Related Fields

1281 Administrators; Colleges and Universities

1282 Administrators; Elementary and Secondary Education

1283 Administrators; Education and Related Fields, Not Elsewhere Classified

131 Managers; Medicine and Health

132 Production Managers, Industrial

133 Construction Managers

134 Public Utilities Managers

1341 Communication Operations Managers

1342 Transportation Facilities and Operations Managers

1343 Electricity, Gas, Water Supply, and Sanitary Services Managers

1344 Postmasters and Mail Superintendents

135 Managers; Service Organizations

1351 Managers; Food Serving and Lodging Establishments

1352 Managers; Entertainment and Recreation Facilities

1353 Managers; Property and Leasing


1354 Managers; Membership Organizations

1359 Managers, Service Organization, Not Elsewhere Classified

136 Managers; Mining, Quarrying, Well Drilling, and Similar Operations

137 Managers; Administrative Services

139 Officials and Administrators; Other, Not Elsewhere Classified

14 Management Related Occupations

141 Accountants, Auditors, and Other Financial Specialists

1412 Accountants and Auditors

1414 Underwriters

1415 Loan Officers

1419 Other Financial Officers

142 Management Analysts

143 Personnel, Training, and Labor Relations Specialist

144 Purchasing Agents and Buyers

1442 Buyers, Wholesale and Retail Trade, except Farm Products

1443 Purchasing Agents and Buyers, Farm Products

1449 Purchasing Agents and Buyers, Not Elsewhere Classified

145 Business and Promotions Agents

147 Inspectors and Compliance Officers

1472 Construction Inspectors

1473 Inspectors and Compliance Officers, except Construction

149 Management Related Occupations, Not Elsewhere Classified

Engineers, Surveyors and Architects

16 Engineers, Surveyors and Architects

161 Architects

162-3 Engineers

1622 Aerospace Engineers

1623 Metallurgical and Materials Engineers

1624 Mining Engineers

1625 Petroleum Engineers

1626 Chemical Engineers

1627 Nuclear Engineers

1628 Civil Engineers

1632 Agricultural Engineers

1633 Electrical and Electronic Engineers

1634 Industrial Engineers

1635 Mechanical Engineers

1636 Computer Engineers

1637 Marine Engineers and Naval Architects

1639 Engineers, Not Elsewhere Classified

164 Surveyors and Mapping Scientists

1643 Land Supervisors

1644 Cartographers

1649 Surveyors and Mapping Scientists, Not Elsewhere Classified

Natural Scientists and Mathematicians

17 Computer, Mathematical, and Operations Research Occupations


171 Computer Scientists

1712 Computer Systems Analyst

1719 Computer Scientists, Not Elsewhere Classified

172 Operations and Systems Researchers and Analysts

1721 Operations Researchers and Analysts

1722 Systems Researchers and Analysts, Except Computer

173 Mathematical Scientists

1732 Actuaries

1733 Statisticians

1739 Mathematical Scientists, Not Elsewhere Classified

18 Natural Scientists

184 Physical Scientists

1842 Astronomers

1843 Physicists

1845 Chemists, Except Biochemists

1846 Atmospheric and Space Scientists

1847 Geologists

1849 Physical Scientists, Not Elsewhere Classified

185 Life Scientists

1852 Forestry and Conservation Scientists

1853 Agricultural and Food Scientists

1854 Biological Scientists

1855 Medical Scientists

Social Scientists, Social Workers, Religious Workers, and Lawyers

19 Social Scientists and Urban Planners

191 Social Scientists

1912 Economists

1913 Historians

1914 Political Scientists

1915 Psychologists

1916 Sociologists

1919 Social Scientists, Not Elsewhere Classified

192 Urban and Regional Planners

20 Social, Recreation, and Religious Workers

203 Social and Recreation Workers

2032 Social Workers

2033 Recreation Workers

204 Religious Workers

2042 Clergy

2049 Religious Workers, Not Elsewhere Classified

21 Lawyers and Judges

211 Lawyers

212 Judges

Teachers, Librarians, and Counselors

22 Teachers; College, University and Other Postsecondary Institutions

2212 Atmospheric, Earth, Marine, and Space Science Teachers


2213 Biological Science Teachers

2214 Chemistry Teachers

2215 Physics Teachers

2216 Natural Science Teachers, Not Elsewhere Classified

2217 Psychology Teachers

2218 Economics Teachers

2222 History Teachers

2223 Political Science Teachers

2224 Sociology Teachers

2225 Social Science Teachers, Not Elsewhere Classified

2226 Engineering Teachers

2227 Mathematical Science Teachers

2228 Computer Science Teachers

2231 Medical Science Teachers

2232 Health Specialties Teachers, Not Elsewhere Classified

2233 Business, Commerce and Marketing Teachers

2234 Agriculture Teachers

2235 Art, Drama, and Music Teachers

2236 Physical Education Teachers

2237 Education Teachers

2238 English Teachers

2242 Foreign Language Teachers

2243 Law Teachers

2244 Social Work Teachers

2245 Theology Teachers

2246 Trade and Industrial Teachers

2247 Home Economics Teachers

2249 Teachers; Postsecondary, Not Elsewhere Classified

23 Teachers, Except Postsecondary Institutions

231 Prekindergarten and Kindergarten Teachers

232 Elementary School Teachers

233 Secondary School Teachers

235 Teachers; Special Education

236 Instructional Coordinators

239 Adult Education and Other Teachers, Not Elsewhere Classified

24 Vocational and Educational Counselors

25 Librarians, Archivists, and Curators

251 Librarians

252 Archivists and Curators

Health Diagnosing and Treating Practitioners

26 Physicians and Dentists

261 Physicians

262 Dentists

27 Veterinarians

28 Other Health Diagnosing and Treating Practitioners

281 Optometrists


283 Podiatrists

289 Health Diagnosing and Treating Practitioners, Not Elsewhere Classified

Registered Nurses, Pharmacists, Dietitians, Therapists, and Physician’s Assistants

29 Registered Nurses

30 Pharmacists, Dietitians, Therapists, and Physicians Assistants

301 Pharmacists

302 Dietitians

303 Therapists

3031 Respiratory Therapists

3032 Occupational Therapists

3033 Physician Therapists

3034 Speech Pathologists and Audiologists

3039 Therapists, Not Elsewhere Classified

304 Physicians Assistants

Writers, Artists, Entertainers, and Athletes

32 Writers, Artists, Performers, and Related Workers

321 Authors

322 Designers

323 Musicians and Composers

324 Actors and Directors

325 Painters, Sculptors, Craft-Artists and Artist-Printmakers

326 Photographers

327 Dancers

328 Performers, Not Elsewhere Classified

329 Writers, Artists, and Related Workers; Not Elsewhere Classified

33 Editors, Reporters, Public Relations Specialist, and Announcers

331 Editors and Reporters

3312 Editors

3313 Reporters

332 Public Relations Specialists and Publicity Writers

333 Radio, Television and Other Announcers

34 Athletes and Related Workers

Health Technologists and Technicians

36 Health Technologists and Technicians

362 Clinical Laboratory Technologists and Technicians

363 Dental Hygienists

364 Health Record Technologists and Technicians

365 Radiological Technologists and Technicians

366 Licensed Practical Nurses

369 Health Technologists and Technicians, Not Elsewhere Classified

Technologists and Technicians, Except Health

37 Engineering and Related Technologists and Technicians

371 Engineering Technologists and Technicians

3711 Electrical and Electronic Engineering Technologists and Technicians

3712 Industrial Engineering Technologists and Technicians

3713 Mechanical Engineering Technologists and Technicians


3719 Engineering Technologists and Technicians, Not Elsewhere Classified

372 Drafting Occupations

373 Surveying and Mapping Technicians

3733 Surveying Technicians

3734 Cartographic Technicians

3739 Surveying and Mapping Technicians, Not Elsewhere Classified

38 Science Technologists and Technicians

382 Biological Technologists and Technicians, except Health

383 Chemical and Nuclear Technologists and Technicians

3831 Chemical Technologists and Technicians

3832 Nuclear Technologists and Technicians

3833 Petroleum Technologists and Technicians

384 Mathematical Technicians

389 Science Technologists and Technicians; Not Elsewhere Classified

39 Technicians; Except Health, Engineering, and Science

392 Air Traffic Controllers

393 Radio and Related Operators

396 Legal Technicians

397 Programmers

3971 Programmers, Business

3972 Programmers, Scientific

3974 Programmers, Numerical, Tool and Process Control

398 Technical Writers

399 Technicians, Not Elsewhere Classified

Marketing and Sales Occupations

40 Supervisors; Marketing and Sales Occupations

401 Supervisors; Sales Occupations, Insurance, Real Estate and Business Services

402 Supervisors; Sales Occupations, Commodities Except Retail

403 Supervisors; Sales Occupations, Retail

41 Insurance, Securities, Real Estate and Business Services Sales Occupations

412 Insurance, Real Estate, and Securities Sales Occupations

4122 Insurance Sales Occupations

4123 Real Estate Sales Occupations

4124 Securities and Financial Services Sales Occupations

415 Business Service Sales Occupations

4152 Business Services, Except Advertising, Sales Occupations

4153 Advertising and Related Sales Occupations

42 Sales Occupations; Commodities Except Retail

421 Sales Engineers

423 Technical Sales Workers and Service Advisors

4232 Technical Sales Workers, Aircraft

4233 Technical Sales Workers, Agricultural Equipment and Supplies

4234 Technical Sales Workers, Electronic Equipment

4235 Technical Sales Workers, Industrial Machinery, Equipment, and Supplies

4236 Technical Sales Workers, Medical and Dental Equipment and Supplies

4237 Technical Sales Workers, Chemicals and Chemical Products


4239 Technical Sales Workers, Not Elsewhere Classified

424 Sales Representatives

4242 Sales Representatives; Commercial and Industrial Equipment and Supplies

4243 Sales Representatives; Garments and Related Textile Products

4244 Sales Representatives; Motor Vehicles and Supplies

4245 Sales Representatives; Pulp, Paper, and Paper Products

4246 Sales Representatives; Farm Products and Livestock

4249 Sales Representatives; Not Elsewhere Classified

43 Sales Occupations Retail

434-5 Salespersons, Commodities

4342 Salespersons; Motor Vehicles, Mobile Homes, and Supplies

4343 Salespersons; Musical Instruments and Supplies

4344 Salespersons; Boats and Marine Equipment and Supplies

4345 Salespersons; Sporting Goods

4346 Salespersons; Garments and Textile Products

4347 Salespersons; Books, Stamps, Coins, and Stationery

4348 Salespersons; Furniture and Home Furnishings

4351 Salespersons; Shoes

4352 Salespersons; Radio, Television, High Fidelity, and Household Appliances

4353 Salespersons; Hardware

4354 Salespersons; Cosmetics, Toiletries, and Allied Products

4356 Salespersons; Jewelry and Related Products

4359 Salespersons; Not Elsewhere Classified

436 Sales Occupations; Others

4362 Sales Clerks

4363 Counter Clerks

4364 Cashiers

4365 News Vendors

4366 Street Vendors, Door-to-Door Sales Workers, and Related Occupations

4367 Salespersons; Parts

4369 Sales Occupations; Services, Not Elsewhere Classified

44 Sales Related Occupations

444 Appraisers and Related Occupations

445 Demonstrators, Promoters, and Models

446 Shoppers

447 Auctioneers

449 Sales Occupations; Other, Not Elsewhere Classified

Administrative Support Occupations, Including Clerical

45 Supervisors; Administrative Support Occupations, Including Clerical

4511 Supervisors; General Office Occupations

4512 Supervisors; Computer and Peripheral Equipment Operators

4513 Supervisors; Secretaries, Stenographers and Typists

4514 Supervisors; Information Clerks

4516 Supervisors; Correspondence Clerks and Order Clerks

4519 Supervisors; Record Clerks

4521 Supervisors; Financial Record Processing Occupations


4522 Supervisors; Duplicating, Mail and Other Office Machine Operators

4523 Chief Communications Operators

4524 Supervisors; Mail and Message Distribution Clerks

4525 Supervisors; Material Recording, Scheduling, and Distributing Clerks

4528 Supervisors; Adjusters, Investigators, and Collectors

4529 Supervisors; Miscellaneous Administrative Support Occupations

46-47 Administrative Support Occupations, Including Clerical

461 Computer and Peripheral Equipment Operators

4612 Computer Operators

4613 Peripheral Equipment Operators

462 Secretaries, Stenographers and Typists

4622 Secretaries

4623 Stenographers

4624 Typists

463 General Office Occupations

464 Information Clerks

4642 Interviewing Clerks

4643 Hotel Clerks

4644 Reservation Agents and Transportation Ticket Clerks

4645 Receptionists

4649 Information Clerks, Not Elsewhere Classified

466 Correspondence Clerks and Order Clerks

4662 Classified-ad Clerks

4663 Correspondence Clerks

4664 Order Clerks

469 Record Clerks

4692 Personnel Clerks, Except Payroll and Timekeeping

4694 Library Clerks

4696 File Clerks

4699 Record Clerks, Not Elsewhere Classified

471 Financial Record Processing Occupations

4712 Bookkeepers and Accounting and Auditing Clerks

4713 Payroll and Timekeeping Clerks

4715 Billing Clerks

4716 Cost and Rate Clerks

4718 Billing, Posting, and Calculating Machines Operators

472 Duplicating, Mail and Other Office Machine Operators

4722 Duplicating Machine Operators

4723 Mail Preparing and Handling Machine Operators

4729 Office Machine Operators, Not Elsewhere Classified

473 Communication Equipment Operators

4732 Telephone Operators

4733 Telegraphers

4739 Communications Equipment Operators, Not Elsewhere Classified

474 Mail and Message Distributing Occupations

4742 Postal Clerks, Except Mail Carriers


4743 Mail Carriers, Post Office

4744 Mail Clerks, Except Post Office

4745 Messengers

475 Material Recording, Scheduling, and Distributing Clerks

4751 Dispatchers

4752 Production and Planning Clerks

4753 Traffic, Shipping, and Receiving Clerks

4754 Stock and Inventory Clerks

4755 Meter Readers

4756 Weighers, Measures, and Clerks

4757 Samplers

4758 Expediters

4759 Materials Recording, Scheduling, and Distributing Clerks, Not Elsewhere Classified

478 Adjusters, Investigators, and Collectors

4782 Insurance Adjusters, Examiners, and Investigators

4783 Investigators and Adjusters, Except Insurance

4784 Clerks, Social Welfare

4786 Bill and Account Collectors

4787 License Clerks

479 Miscellaneous Administrative Support Occupations, Including Clerical

4791 Bank Tellers

4792 Proof Readers

4793 Data Entry Keyers

4794 Statistical Clerks

4795 Teacher Aides

4799 Administrative Support Occupations, Including Clerical, Not Elsewhere Classified

Service Occupations

50 Private Household Occupations

502 Day Workers

503 Launderers and Ironers

504 Cooks, Private Household

505 Housekeepers and Butlers

506 Child Care Workers, Private Household

507 Private Household Cleaners and Servants

509 Private Household Occupations, Not Elsewhere Classified

51 Protective Service Occupations

511 Supervisors; Service Occupations, Protective

5111 Supervisors; Firefighting and Fire Prevention Occupations

5112 Supervisors; Police and Detectives

5113 Supervisors; Guards

512 Firefighting and Fire Prevention Occupations

5122 Fire Inspection and Fire Prevention Occupations

5123 Firefighting Occupations

513 Police and Detectives

5132 Police and Detectives, Public Service

5133 Correctional Institution Officers


5134 Sheriffs, Bailiffs, and Other Law Enforcement Officers

514 Guards

5142 Crossing Guards

5144 Guards and Police, Except Public Service

5149 Protective Service Occupations, Not Elsewhere Classified

52 Service Occupations, Except Private Household and Protective

521 Food and Beverage Preparation and Service Occupations

5211 Supervisors; Food and Beverage Preparation Service Occupations

5212 Bartenders

5213 Waiters and Waitresses

5214 Cooks, Except Short Order

5215 Short-order Cooks

5216 Food Counter, Fountain and Related Occupations

5217 Kitchen Workers, Food Preparation

5218 Waiters’/Waitresses’ Assistants

5219 Miscellaneous Food and Beverage Preparation Occupations

523 Health Service Occupations

5232 Dental Assistants

5233 Health Aides, Except Nursing

5236 Nursing Aides, Orderlies, and Attendants

524 Cleaning and Building Service Occupations, Except Private Households

5241 Supervisors; Cleaning and Building Service Workers

5242 Maids and Housemen

5244 Janitors and Cleaners

5245 Elevator Operators

5246 Pest Control Occupations

5249 Cleaning and Building Service Occupations, Not Elsewhere Classified

525-6 Personal Service Occupations

5251 Supervisors; Personal Service Occupations

5252 Barbers

5253 Hairdressers and Cosmetologists

5254 Attendants, Amusement and Recreation Facilities

5255 Guides

5256 Ushers

5257 Public Transportation Attendants

5258 Wardrobe and Dressing Room Attendant

5262 Baggage Porters and Bellhops

5263 Welfare Service Agents

5264 Child Care Workers, Except Private Household

5269 Personal Service Occupations, Not Elsewhere Classified

Agricultural, Forestry and Fishing Occupations

55 Farm Operators and Managers

551 Farmers (Working Proprietors)

5512 General Farmers

5513 Crop, Vegetable, Fruit and Tree Nut Farmers

5514 Livestock, Dairy, Poultry and Fish Farmers


5515 Horticulture Specialty Farmers

552 Farm Managers

5522 Managers; General Farm

5523 Managers; Crop, Vegetable, Fruit and Tree Nut Farm

5524 Managers; Livestock, Dairy, Poultry and Fish Farm

5525 Managers; Horticulture Specialty Farm

56 Other Agriculture and Related Occupations

561 Farm Occupations, Except Managerial

5611 Supervisors, Farm Workers

5612 General Farm Workers

5613 Field Crop and Vegetable Farm Workers (Hand)

5614 Orchard and Vineyard and Related Workers (Hand)

5615 Irrigation Workers

5616 Farm Machinery Operators

5617 Livestock Workers

5618 Marine Life Cultivation Workers

5619 Nursery Workers

562 Related Agricultural Occupations

5621 Supervisors; Related Agricultural Workers

5622 Groundskeepers and Gardeners, Except Farm

5624 Animal Caretakers, Except Farm

5625 Graders and Sorters; Agricultural Products

5627 Inspectors; Agricultural Products

57 Forestry and Logging Occupations

571 Supervisors; Forestry and Logging Workers

572 Forestry Workers, Except Logging

573 Timber Cutting and Related Occupations

579 Logging Occupations, Not Elsewhere Classified

58 Fishers, Hunters, and Trappers

583 Fishers

584 Hunters, and Trappers

Mechanics and Repairers

60 Supervisors; Mechanics and Repairers

61 Mechanics and Repairers

611 Vehicle and Mobile Equipment Mechanics and Repairers

6111 Automobile Mechanics

6112 Bus and Truck Engine, and Diesel Engine Mechanics

6113 Aircraft Engine Mechanics

6114 Small Engine Repairers

6115 Automobile Body and Related Repairers

6116 Aircraft Mechanics (Except Engine Specialists)

6117 Heavy Equipment Mechanics

6118 Farm Equipments Mechanics

613 Industrial Machinery Repairers

614 Machinery Maintenance Occupations

615 Electrical and Electronic Equipment Repairers


6151 Communications Equipment Repairers

6152 Electric Motor, Transformer, and Related Repairers

6153 Electric and Electronic Repairers, Commercial and Industrial Equipment

6154 Data Processing Equipment Repairers

6155 Electronic Repairers, Home-entertainment Equipment

6156 Household Appliance and Power Tools Repairers

6157 Telephone Line Installer and Repairers

6158 Telephone Installers and Repairers

6159 Miscellaneous Electrical and Electronic Equipment Repairers

616 Heating, Air-conditioning, and Refrigeration Mechanics

617 Miscellaneous Mechanics and Repairers

6171 Camera, Watch, and Other Precision Instrument Repairers

6172 Musical Instrument Repairers and Tuners

6173 Locksmiths and Safe Repairers

6174 Office Machine Repairers

6175 Mechanical Controls and Valve Repairers

6176 Elevator Installers and Repairers

6177 Riggers

6178 Millwrights

6179 Mechanics and Repairers, Not Elsewhere Classified

Construction and Extractive Occupations

63 Supervisors; Constructions and Extractive Occupations

631 Supervisors; Construction

6311 Supervisors; Overall Construction

6312 Supervisors; Brickmasons, Stonemasons, and Hard Tile Setters

6313 Supervisors; Carpenters and Related Workers

6314 Supervisors; Electricians and Power Transmissions Installers

6315 Supervisors; Painters, Paperhangers, and Plasterers

6316 Supervisors; Plumbers and Pipefitters and Steamfitters

6318 Supervisors; Other Construction Trades

632 Supervisors; Extractive Occupations

64 Construction Trades

641 Brickmasons, Stonemasons, and Hard Tile Setters

6412 Brickmasons

6413 Stonemasons

6414 Tile Setters, Hard

642 Carpenters and Related Workers

6422 Carpenters

6424 Drywall Installers

643 Electricians and Power Transmissions Installers

6432 Electricians

6433 Electrical Power Installers and Repairers

644 Painters, Paperhangers, and Plasterers

6442 Painters (Construction and Maintenance)

6443 Paperhangers

6444 Plasterers


645 Plumbers, Pipefitters and Steamfitters

646-7 Other Construction Trades

6462 Carpet and Soft Tile Installers

6463 Concrete and Terrazzo Finishers

6464 Glaziers

6465 Insulation Workers

6466 Paving, Surfacing, and Tamping Equipment Operators

6467 Rail and Track Laying Equipment Operators

6468 Roofers

6472 Sheetmetal Duct Installers

6473 Structural Metal Workers

6474 Drillers, Earth

6475 Air Hammer Operators

6476 Pile Driving Operators

6479 Construction Trades, Not Elsewhere Classified

65 Extractive Occupations

652 Drillers, Oil Well

653 Explosive Workers

654 Mining Machine Operators

656 Extractive Occupations, Not Elsewhere Classified

Precision Production Occupations

67 Supervisors; Precision Production Occupations

68 Precision Production Occupations

681-2 Precision Metal Workers

6811 Tool and Die Makers

6812 Precision Assemblers (Metal)

6813 Machinists

6814 Boilermakers

6816 Precision Grinders, Filers, and Tool Sharpeners

6817 Patternmakers and Model Makers (Metal)

6821 Lay-out Workers

6822 Precision Hand Molders and Shapers (jewelers)

6823 Engravers

6824 Sheet Metal Workers

6829 Miscellaneous Precision Metal Workers

683 Precision Woodworkers

6831 Patternmakers and Model Makers, Wood

6832 Cabinet Makers and Bench Carpenters

6835 Furniture Finishers

6839 Miscellaneous Precision Woodworkers

684 Precision Printing Occupations

6841 Precision Typesetters

6842 Precision Lithographers and Photoengravers

6844 Bookbinders

6849 Miscellaneous Precision Printing Occupations

685 Precision Textile, Apparel and Furnishings Workers


6852 Tailors and Dressmakers, Hand

6853 Upholsterers

6854 Shoemakers and Leather Workers and Repairers

6855 Precision Laundering, Cleaning, and Dyeing Occupations

6856 Apparel and Fabric Patternmakers

6859 Miscellaneous Precision Apparel and Fabric Workers

686 Precision Workers; Assorted Materials

6861 Precision Hand Molders and Shapers (Except Jewelers)

6862 Precision Patternmakers, Lay-out Workers and Cutters

6863 Detail Design Painters and Decorators

6864 Optical Goods Workers

6865 Dental Laboratory Technicians

6866 Gem and Diamond Working Occupations

6867 Precision Electrical and Electronic Equipment Assemblers

6868 Photographic Process Workers

6869 Miscellaneous Precision Workers, Not Elsewhere Classified

687 Precision Food Production Occupations

6871 Butchers and Meat Cutters

6872 Bakers

6873 Batchmakers (Candymakers, Cheesemakers, Etc.)

6879 Miscellaneous Precision Food Workers

688 Precision Inspectors, Testers, and Related Workers

6881 Precision Inspectors, Testers, and Graders

6882 Precision Adjusters and Calibrators

69 Plant and System Operators

691 Water and Sewage Treatment Plant Operators

692 Gas Plant Operators

693 Power Plant Operators

6931 Stationary Engineers

6932 Power Plant and Systems Operators, except Stationary Engineers

694 Chemical Plant Operators

695 Petroleum Plant Operators

696 Miscellaneous Plant or System Operators

Production Working Occupations

71 Supervisors; Production Occupations

73-74 Machine Setup Operators

731-2 Metal Working and Plastic Working Machine Setup Operators

7312 Lathe and Turning Machine Setup Operators

7313 Milling and Planning Machine Setup Operators

7314 Punching and Shearing Machine Setup Operators

7315 Extruding and Drawing Machine Setup Operators

7316 Rolling Machine Setup Operators

7317 Press and Brake Machine Setup Operators

7318 Drilling and Boring Machine Setup Operators

7319 Forging Machine Setup Operators

7322 Grinding, Abrading, Buffing, and Polishing Machine Setup Operators


7324 Lapping and Honing Machine Setup Operators

7326 Numerical Control Machine Setup Operators

7329 Miscellaneous Metalworking and Plastic Working Machine Setup Operators

733 Metal Fabricating Machine Setup Operators

7332 Welding Machine Setup Operators

7333 Soldering and Brazing Machine Setup Operators

7339 Miscellaneous Fabricating Machine Setup Operators

734 Metal and Plastic Processing Machine Setup Operators

7342 Molding and Casting Machine Setup Operators

7343 Plating and Coating Machine Setup Operators

7344 Heating Equipment Machine Setup Operators

7349 Miscellaneous Metal and Plastic Processing Machine Setup Operators

743 Woodworking Machine Setup Operators

7431 Lathe and Turning Machine Setup Operators

7432 Router and Planer Machine Setup Operators

7433 Sawing Machine Setup Operators

7434 Sanding Machine Setup Operators

7435 Shaping and Joining Machine Setup Operators

7439 Miscellaneous Woodworking Machine Setup Operators

744 Printing Machine Setup Operators

7443 Printing Press Setup Operators

7444 Photoengraving and Lithographing Machine Setup Operators

7449 Miscellaneous Printing Machine Setup Operators

745 Textile Machine Setup Operators

7451 Winding and Twisting Machine Setup Operators

7452 Knitting and Weaving Machine Setup Operators

7459 Textile Machine Setup Operators, Not Elsewhere Classified

746-7 Assorted Materials: Machine Setup Operators

7462 Packaging and Filling Machine Setup Operators

7463 Extruding and Forming Machine Setup Operators

7467 Compressing and Compacting Machine Setup Operators

7472 Roasting and Baking Machine Setup Operators

7474 Folding Machine Setup Operators

7476 Still, Clarifying, and Precipitating Machine Setup Operators

7477 Crushing, Grinding and Polishing Machine Setup Operators

7478 Slicing and Cutting Machine Setup Operators

7479 Miscellaneous Machine Setup Operators

75-76 Machine Operators and Tenders

751-2 Metal Working and Plastic Working Machine Operators and Tenders

7512 Lathe and Turning Machine Operators and Tenders

7513 Milling and Planning Machine Operators and Tenders

7514 Punching and Shearing Machine Operators and Tenders

7515 Extruding and Drawing Machine Operators and Tenders

7516 Rolling Machine Operators and Tenders

7517 Press and Brake Machine Operators and Tenders

7518 Drilling and Boring Machine Operators and Tenders


7519 Forging Machine Operators and Tenders

7522 Grinding, Abrading, Buffing, and Polishing Machine Operators and Tenders

7529 Miscellaneous Metalworking and Plastic Working Machine Operators and Tenders

753 Metal Fabricating Machine Operators and Tenders

7532 Welding Machine Operators and Tenders

7533 Soldering and Brazing Machine Operators and Tenders

7539 Miscellaneous Fabricating Machine Operators and Tenders

754 Metal and Plastic Processing Machine Operators and Tenders

7542 Molding and Casting Machine Operators and Tenders

7543 Plating and Coating Machine Operators and Tenders

7544 Heating Equipment Machine Operators and Tenders

7549 Miscellaneous Metal and Plastic Processing Machine Operators and Tenders

763 Woodworking Machine Operators and Tenders

7631 Lathe and Turning Machine Operators and Tenders

7632 Router and Planer Machine Operators and Tenders

7633 Sawing Machine Operators and Tenders

7634 Sanding Machine Operators and Tenders

7635 Shaping and Joining Machine Operators and Tenders

7636 Nailing and Tacking Machine Operators and Tenders

7639 Miscellaneous Woodworking Machine Operators and Tenders

764 Printing Machine Operators and Tenders

7642 Typesetting and Composing Machine Operators and Tenders

7643 Printing Machine Operators and Tenders

7644 Photoengraving and Lithographing Machine Operators and Tenders

7649 Printing Machine Operators and Tenders, Not Elsewhere Classified

765 Textile, Apparel and Furnishings Machine Operators and Tenders

7651 Winding and Twisting Machine Operators and Tenders

7652 Knitting and Weaving Machine Operators and Tenders

7654 Textile Cutting Machine Operators and Tenders

7655 Textile Sewing Machine Operators and Tenders

7656 Shoe Machine Operators and Tenders

7657 Pressing Machine Operators

7658 Laundering and Dry Cleaning Machine Operators and Tenders

7659 Miscellaneous Textile Machine Operators and Tenders

766-7 Machine Operators and Tenders; Assorted Materials

7661 Cementing and Gluing Machine Operators and Tenders

7662 Packaging and Filling Machine Operators and Tenders

7663 Extruding and Forming Machine Operators and Tenders

7664 Mixing and Blending Machine Operators and Tenders

7665 Cooling and Freezing Equipment Operators and Tenders

7666 Separating and Filtering Machine Operators and Tenders

7667 Compressing and Compacting Machine Operators and Tenders

7668 Boiler Operators and Tenders (Low Pressure)

7669 Coating, Painting, and Spraying Machine Operators and Tenders

7671 Photographic Processing Machine Operators

7672 Roasting and Baking Machine Operators and Tenders


7673 Washing, Cleaning and Pickling Equipment Operators and Tenders

7674 Folding Machine Operators and Tenders

7675 Furnace, Kiln, and Oven Operators and Tenders

7676 Still, Clarifying, and Precipitating Operators and Tenders

7677 Crushing, Grinding and Polishing Machine Operators and Tenders

7678 Slicing and Cutting Machine Operators and Tenders

7679 Miscellaneous Machine Operators and Tenders, Not Elsewhere Classified

77 Fabricators, Assemblers, and Hand Working Occupations

771 Welders and Solderers

7714 Welders and Cutters

7717 Solderers and Brazers

772 Assemblers

774 Fabricators, Not Elsewhere Classified

775 Hand Working Occupations

7752 Hand Sewing Occupations

7753 Hand Cutting and Trimming Occupations

7754 Hand Molding and Casting Occupations

7755 Hand Forming And Shaping Occupations

7756 Hand Painting, Coating and Decorating Occupations

7757 Hand Engraving And Printing Occupations

7758 Hand Grinding and Polishing Occupations

7759 Miscellaneous Hand Working Occupations

78 Production Inspectors, Testers, Samplers, and Weighers

782 Production Inspectors, Checkers and Examiners

783 Production Testers

784 Production Samplers and Weighers

785 Graders and Sorters, Except Agricultural

787 Production Expediters

Transportation and Material Moving Occupations

81 Supervisors; Transportation and Material Moving Occupations

811 Supervisors; Motorized Equipment Operators

8111 Supervisors; Motor Vehicle Operators

8113 Railroad Conductors and Yardmasters

812 Supervisors; Materials Moving Equipment Operators

82 Transportation Occupations

821 Motor Vehicle Operators

8212 Truck Drivers, Tractor-trailer

8213 Truck Drivers, Heavy

8214 Truck Divers, Light (Including Delivery and Route Drivers)

8215 Bus Drivers

8216 Taxicab Drivers and Chauffeurs

8218 Driver-Sales Workers

8219 Other Motor Transportation Occupations, Not Elsewhere Classified

823 Rail Transportation Occupations

8232 Locomotive Operating Occupations

8233 Railroad Brake, Signal, and Switch Operators


8239 Rail Vehicle Operators, Not Elsewhere classified

824 Water Transportation Occupations

8241 Ship Captains and Mates

8242 Boat and Barge Operators

8243 Sailors and Deckhands

8244 Marine Engineers

8245 Bridge, Lock, Lighthouse Tenders

825 Airplane Pilots and Navigators

828 Transportation Inspectors

83 Materials Moving Occupations, Except Transportation

831 Materials Moving Equipment Operators

8312 Operating Engineers

8313 Longshore Equipment Operators

8314 Hoist and Winch Operators

8315 Crane and Tower Operators

8316 Excavating and Loading Machine Operators

8317 Grader, Dozer, and Scraper Operators

8318 Industrial Truck and Tractor Equipment Operators

8319 Miscellaneous Materials Moving Equipment Operators

Handlers, Equipment Cleaners, Helpers and Laborers

85 Supervisor; Handlers, Equipment Cleaners, Helpers, and Laborers

86 Helpers

861 Helpers; Machine Operators and Tenders

8611 Helpers; Metalworking and Plastic Working Machine Operators and Tenders

8614 Helpers; Metal and Plastic Processing Machine Operators and Tenders

8615 Helpers; Woodworking Machine Operators and Tenders

8616 Helpers; Printing Machine Operators and Tenders

8617 Helpers; Textile, Apparel and Furnishings Machine Operators and Tenders

8618 Helpers; Machine Operators and Tenders, Assorted Materials

8619 Helpers; Precision Production Occupations and Setup Operations

862 Helpers; Fabricators and Inspectors

863 Helpers; Mechanics and Repairers

8632 Helpers; Vehicle and Mobile Equipment Mechanics and Repairers

8633 Helpers; Industrial Machinery Repairers

8635 Helpers; Electrical and Electronic Equipment Repairers

8637 Helpers; Miscellaneous Mechanics and Repairers

864 Helpers; Construction Trades

8641 Helpers; Brickmasons, Stonemasons, and Hard Tile Setters

8642 Helpers; Carpenters and Related Workers

8643 Helpers; Electricians and Power Transmission Installers

8644 Helpers; Painters, Paperhangers, and Plasterers

8645 Helpers; Plumbers, Pipefitters and Steamfitters

8646 Helpers; Surveyor’s

8648 Helpers; Other Construction Trades

865 Helpers; Extractive Occupations

87 Handlers, Equipment Cleaners and Laborers


871 Construction Laborers

872 Freight, Stock, and Materials Movers; Hand

8722 Garbage Collectors

8723 Stevedores

8724 Stock Handlers and Baggers

8725 Machine Feeders and Offbearers

8726 Freight, Stock, and Materials Movers, Not Elsewhere Classified

873 Garage and Service Station Related Occupations

874 Parking Lot Attendants

875 Vehicle Washers and Equipment Cleaners

876 Miscellaneous Manual Occupations

8761 Hand Packers and Packagers

8769 Manual Occupations, Not Elsewhere Classified

Military Occupations

91 Military Occupations

Miscellaneous Occupations

99 Miscellaneous Occupations

Bibliography

Abraham, Katharine G. and Henry S. Farber (1987), “Job duration, seniority, and earnings.”

American Economic Review, 77, 278–29.

Acemoglu, Daron and Robert Shimer (1999), “Efficient unemployment insurance.” Journal

of Political Economy, 107, 893–928.

Altonji, Joseph G. and Robert A. Shakotko (1987), “Do wages rise with job seniority?”

Review of Economic Studies, 54, 437–459.

Altonji, Joseph G. and Nicolas Williams (2005), “Do wages rise with job seniority? a re-

assessment.” Industrial and Labor Relations Review, 370–397.

Autor, David H., Frank Levy, and Richard J. Murnane (2003), “The skill content of recent

technological change: An empirical exploration.” The Quarterly Journal of Economics,

1279–1333.

Bacolod, Marigee P. and Bernardo S. Blum (2010), “Two sides of the same coin us residual

inequality and the gender gap.” Journal of Human Resources, 45, 197–242.

Bartel, Ann P. and George J. Borjas (1981), “Wage growth and job turnover: An empirical

analysis.” In Studies in labor markets, 65–90, University of Chicago Press.

Becker, Gary S. (1964), “Human capital: a theoretical analysis with special reference to

education.” National Bureau for Economic Research, Columbia University Press, New

York and London.

Burdett, Kenneth, Shouyong Shi, and Randall Wright (2001), “Pricing and matching with

frictions.” Journal of Political Economy, 109, 1060–1085.

Gathmann, Christina and Uta Schonberg (2010), “How general is human capital? a task-

based approach.” Journal of Labor Economics, 28, 1–49.

Gibbons, Robert and Michael Waldman (2006), “Enriching a theory of wage and promotion

dynamics inside firms.” Journal of Labor Economics, 24, 59–108.

147

BIBLIOGRAPHY 148

Hill, Daniel H. (1994), “The relative empirical validity of dependent and independent data

collection in a panel survey.” Journal of Official Statistics, 10, 359–359.

Hoffmann, Florian (2010), “An empirical model of life-cycle earnings and mobility dynam-

ics.” Mimeo, University of British Columbia.

Ingram, Beth F. and George R. Neumann (2006), “The returns to skill.” Labour Economics,

13, 35–59.

Kambourov, Gueorgui (2009), “Labour market regulations and the sectoral reallocation of

workers: The case of trade reforms.” Review of Economic Studies, 76, 1321–1358.

Kambourov, Gueorgui and Iourii Manovskii (2008), “Rising occupational and industry mo-

bility in the united states: 1968–1997.” International Economic Review, 49, 41–79.

Kambourov, Gueorgui and Iourii Manovskii (2009a), “Occupational mobility and wage in-

equality.” Review of Economic Studies, 76, 731–759.

Kambourov, Gueorgui and Iourii Manovskii (2009b), “Occupational specificity of human

capital.” International Economic Review, 50, 63–115.

Keane, Michael P. and Kenneth I. Wolpin (1997), “The career decisions of young men.”

Journal of Political Economy, 105, 473–522.

Kletzer, Lori G. and Howard Rosen (2008), “Reforming unemployment insurance for the

twenty-first-century workforce.” Path to prosperity: Hamilton Project ideas on income

security, education, and taxes, 63.

Lucas Jr., Robert E. and Edward C. Prescott (1974), “Equilibrium search and unemploy-

ment.” Journal of Economic Theory, 7, 188–209.

Markey, James P. and William Parks (1989), “Occupational change: pursuing a different

kind of work.” Monthly Labor Review, 3–12.

Menzio, Guido and Shouyong Shi (2010), “Block recursive equilibria for stochastic models

of search on the job.” Journal of Economic Theory, 145, 1453–1494.

Menzio, Guido and Shouyong Shi (2011), “Efficient search on the job and the business cycle.”

Journal of Political Economy, 119, 468–510.

Mincer, Jacob (1974), “Schooling, experience, and earnings. human behavior & social insti-

tutions no. 2.”

BIBLIOGRAPHY 149

Moen, Espen R. (1997), “Competitive search equilibrium.” Journal of Political Economy,

105, 385–411.

Neal, Derek (1995), “Industry-specific human capital: Evidence from displaced workers.”

Journal of labor Economics, 653–677.

Neal, Derek (1999), “The complexity of job mobility among young men.” Journal of Labor

Economics, 17, 237–261.

Parent, Daniel (2000), “Industry-specific capital and the wage profile: Evidence from the

national longitudinal survey of youth and the panel study of income dynamics.” Journal

of Labor Economics, 18, 306–323.

Parrado, Eric, Asena Caner, and Edward N. Wolff (2007), “Occupational and industrial

mobility in the united states.” Labour Economics, 14, 435–455.

Pavan, Ronni (2011), “Career choice and wage growth.” Journal of Labor Economics, 29,

549–587.

Poletaev, Maxim and Chris Robinson (2008), “Human capital specificity: evidence from the

dictionary of occupational titles and displaced worker surveys, 1984–2000.” Journal of

Labor Economics, 26, 387–420.

Ritter, Moritz (forthcoming), “Offshoring and occupational specificity of human capital.”

Review of Economic Dynamics.

Ryscavage, Paul (1989), Measuring spells of unemployment and their outcomes. US Bureau

of the Census.

Shi, Shouyong (2009), “Directed search for equilibrium wage–tenure contracts.” Economet-

rica, 77, 561–584.

Sullivan, Paul (2010a), “A dynamic analysis of educational attainment, occupational choices,

and job search.” International Economic Review, 51, 289–317.

Sullivan, Paul (2010b), “Empirical evidence on occupation and industry specific human

capital.” Labour Economics, 17, 567–580.

Topel, Robert (1991), “Specific capital, mobility, and wages: Wages rise with job seniority.”

Journal of Political Economy, 145–176.

U.S. Department of Labor (1972), Handbook for Analyzing Jobs. U.S. Government Printing

Office.

BIBLIOGRAPHY 150

Vella, Francis and Giuseppe Moscarini (2004), “Aggregate worker reallocation and occupa-

tional mobility in the united states: 1976-2000.” In 2004 Meeting Papers, 183, Society for

Economic Dynamics.

Wasmer, Etienne (2004), “Interpreting europe and u.s. labor market differences: The speci-

ficity of human capital investments.” American Economic Review, 96, 811–831.

Xiong, Hui (2012), “A directed search model of occupational mobility.” Mimeo, University

of Toronto.

Yamaguchi, Shintaro (2010), “Career progression and comparative advantage.” Labour Eco-

nomics, 17, 679–689.

Yamaguchi, Shintaro (2012), “Tasks and heterogeneous human capital.” Journal of Labor

Economics, 30, 1–53.

by hui xiong - university of toronto t-space...hui xiong doctor of philosophy graduate department of...

Documents