$3... ...

. FE,

, . .
113.2.
:s

3&3"...

.. . Va. is.
fl V y ‘ \ra-u .*
. ‘ ”$9.43 3 .. :«hfia
. . , . , , , ‘ . .dzfiwmv.
22.1,. . .. . . V
.fifi ..
4.3..

.o

a}; .
.61"? .u

.u
r

. . x 64.4%
, ‘ . .. .

an: «Azwu . . 9mm

,xzfllni .

‘.
v.1At . .
El
.«V
(.,

 

 

 

2a:

\

 

This is to certify that the
dissertation entitled

"Two Essays on Work and Education"

presented by

Soomyung Jang
has been accepted towards fulfillment

of the requirements for

Ph.D. degree in Economics

I V‘sz'v

 

 

Major p fessor

Date June 16, 2000

 

MS U is an Affirmative Action/Equal Opportunity Institution 0-12771

 

LIBRARY
Michlgan State
Universlty

 

 

 

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before. date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE I DATE DUE DATE DUE

APARL‘IAOAQUIEJ
' U

 

 

Usfiw

‘Q

 

 

 

 

 

 

 

 

 

 

 

 

 

mm mm.“

 

 

Two Essays on Work and Education

By

Soomyung Jang

A DISSERTATION

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

DOCTOR OF PHILOSOPHY

Department of Economics
2000

 

In (it
academic 3?“.
3510.110“), 1‘
is non-linear
academic :11“

mid-36w} 50

q

families in 6

Third. IOCa] I

unemi‘bfmc

ABSTRACT
TWO ESSAYS ON WORK AND EDUCATION
By

Soomyung J ang

Using samples of high school seniors from two data sets, the NLS-72 and HS&B,
this study examines the determinants of employment during high school and explores the
effects of employment during high school on college attendance and the choice of college
major.

In the examination of the determinants of high school employment, we focus on
academic ability, socio-economic status, and local economic conditions. The findings are
as follows: first, the relationship between high school employment and academic ability
is non-linear. Students with medium academic ability work most and students with high
academic ability work moderate hours if they work. Second, students from families with
mid-level socio-economic status work more hours and hold more jobs than students from
families in either the bottom socio-economic status or the top socio-economic status.
Third, local economic conditions, such as the employment growth rate and the
unemployment rate are decisive factors in determining employment during high school.
Fourth, in 1980, students with high academic ability and high socio-economic status
participated in the labor force more intensively than similar high school students did in
1972.

In Chapter 2, we utilize multinomial logit and probit models to explore the effects

of high school employment on college attendance, and on the choice of college major. AS

' i

 

m '91s mus *

1
sjna

3118mm {it

 

ml more K
1;, us its in
36531:» 3%}
LL16 Chdagict
Com
strongl} 1)}.

lah -. . '
ukéh x hk‘K-l,

fisociatior‘

Tm-

in previous studies, we find that high school work experience, in general, is negatively
related to college attendance, particularly four-year college attendance. However, this
negative relationship does not necessarily mean that work experience affects college
attendance negatively since students who do not intend to attend college may choose to
work more. Controlling for such endogeneity, work experience for the students in 1972
appears to increase college attendance, while in 1980, it affects college attendance
negatively. This change in the relationship over time seems to be related to changes in
the characteristics of student workers between 1972 and 1980.

Conditional on attending college, we find that high school work experience is
strongly associated with choice of major. Students who work during their senior year in
high school are more likely to choose majors that lead to high-paying jobs later. The
longer students work during their senior year, the more likely they are to choose business
or engineering as a major, versus majoring in liberal arts or education curricula. This
association is stronger among males than females.

The relationship between high school work experience and the choice of college
major does not appear to be attributable to certain unobservable characteristics, such as a
greater tendency to work, that leads some students to choose ‘practical majors’ which
facilitate future employment. Rather, the association seems to occur because high school
work experience influences the choice of college major.

These findings contribute to our understanding of why there are positive effects of
high school work experience on later earnings, despite its negative association with post-

secondary education.

To my beloved ones, Hyunju, Doore and Sunna

iv

ACKNOWLEDGEMENTS

Being born
Giving birth
Both are fear and labor

With joy of new life

As I begin my career as a fledgling economist, I would like to thank my
committee members for their help in guiding me through this dissertation. The chair of
my committee, Professor Byron W. Brown guided my work with patience and devotion.
His strong support and encouragement were invaluable for helping me get through this
program peacefully. He has my very highest respect both as a teacher and as a person.

My special thanks go to Professor Lalonde who lefi this school while advising
me. Despite the long distance between us, he constantly gave me advice and comments
through letters, e-mail, and telephone calls. His professional direction, along with his
trust and confidence, were a priceless source of support that enabled me to finish this
program.

As the final members of my committee, Professors Woodbury and Plank made
substantial contributions to my dissertation through their timely comments and thoughtful
critiques. I truly appreciate their help and support. Without my committee members’
help, this thesis would not have been possible, although I am responsible for any

remaining mistakes.

v.1

 

uhO£3\€C

 

hardship is
daughters if.
lam so \cr‘;
Elixir)? 3350 3
been great c
m} brother.
la“. B} US;

.‘\l$\
hndsuppyq

Stud}. 3022;

“cl
I m:
l0 e-
ln 1}

film

I am greatly indebted to my wife, Hyunju and my daughters, Doore and Sunna
who gave me tremendous support and love, and who have endured the most difficult
hardship due to my studies in the US. Since returning to Korea, my wife has raised my
daughters all by herself while working. From the depths of my heart, I give thanks to her.
I am so very grateful that my daughters have grown to be such wonderful children. My
thanks also go to my parents and my parents-in—law. Their support and patience have
been great encouragement for me while I finished my degree. I would also like to thank
my brother, Soochan and his family for their warm support and comfort. My brother-in-
law, Byungwook’s, generous support was extraordinarily helpful and greatly appreciated.

Also, I would like to thank Pastor Lee and his church for their deep friendship and
kind support. I would like also to thank friends of mine who helped me throughout my
study. Some of them are: Chirok Han and his family, Sungwon Kim and his wife, Jess
Reaser, Jason Nantan, Mike Khoury, Chieng Cheng, Donggeun Kim, Heeson Choi, ,
Dong Wook Han, Ali Berker, Professor Taekho Kwon, Eunjong Shin and J intae Kim, Pat
Sedan, my special friends and workers.

If today I were a farmer,

In my orchard, I would pick up

well ripened apples.

I would give a basket full of apples

to each one of you.

In the future,

fruits from my life will be given to each one of you.

vi

 

 

LIST OF J

l
CRAFT?

 

 

 

TABLE OF CONTENTS

LIST OF TABLE ............................................................ x

CHAPTER I
THE DETERMINANTS OF HIGH SCHOOL EMPLOYMENT

[-1. Introduction .............................................................. l

1.1. Previous studies on the effects of high school employment on later
earnings and employment ............................................ 2

1.2. Major Findings ........................................................ 20

I-2. Previous Studies on the Determinants of High School Employment
.................................................................................... 25

[-3. Data, Characteristics of High School Employment and Strategy

3.1. Data ..................................................................... 32
3.2. Characteristics of High School Employment ..................... 36
3.3. Strategy ................................................................. 40

[-4. Estimation Results

4.1 Over-time comparisons .............................................. 51
4.2 Job Availability ....................................................... 56
4.3.Academic Ability ..................................................... 67
4.4.Family Socio-economic Backgrounds ............................. 76
4.5. Other Variables ........................................................ 79

vii

 

 

 

[-5. COM

Referenn

CHAPTIT
HIGH S(
AND THI
1“. [Mn

"-2. High
Off
2.]. II

.

.

‘ ~
\

“ H:

[L30 ECODI
3.1. M‘

3.2. pr.

 

[-5. Conclusion ............................................................... 81
References ..................................................................... 88

CHAPTER II

HIGH SCHOOL WORK EXPERIENCE, COLLEGE ATTENDANCE,
AND THE CHOICE OF COLLEGE MAJOR

11-]. Introduction ............................................................. 92

11-2. High School Work Experience, College Attendance, and the Choice

Of College Major

2.1. High School Work Experience and College Attendance ......... 102

2.2. High School Work Experience and the Choice of College Major
............................................................................ 109
11-3. Econometric Models ..................................................... 120
3.1.MNLmodel ............................................................ 121
3.2. Probit models ........................................................... 126
3.2.1. Standard Probit Model .............................................. 127

3.2.2. Probit Models with a Continuous EndogenouesVariable .. . ....127

3.3 Data ...................................................................... 134

11-4. Estimation Results

4.1. College Attendance Choice ............................................ 141
4.1.1. The MNL Model Results ............................................ 142
4.1.2. Probit Results ......................................................... 150

viii

 

4.2. C
4.2.1.
4.2.2.
4.2.3.

"-5. Con

Reference

 

 

4.2. College Major Choice .................................................. 158
4.2.1. Vocational Field Versus Academic Field ........................ 158

4.2.2. Major Choices among College Students in the Academic

Field ................................................................... 161
4.2.3. Probit Results ........................................................ 178
11-5. Conclusion ................................................................. 186
Reference ......................................................................................... 191
ix

 

 

Table l- l I

 

 

 

Table 1.3-.

-—-l
CA)
J...
(b
a)

LIST OF TABLES

Table 1-1-1 The effect of high school employment on later earning ...... 3
Table 1-2-1 Determinants of High School Employment .................. 26
Table 1-3-1 NLS-72 and HS& B ............................................. 35
Table 1-3-2 Working Hours Distribution and Average Test Score ......38
Table 1-3 -3 Working Hours Distribution and Average Test Score:

During Senior Year .............................................. 38
Table 1-3-4 Working Hours Distribution and Average Test Score:

During Junior Year .............................................. 38

Table 1-3-5 Working Experience During Junior and Senior Years

And Average Test Score. ....................................... 39
Table 1-3-6 Job Characteristics ............................................. 39
Table 1-3-7 Earnings ......................................................... 39
Table 1-3-8 Summary Statistics of Local Economics Indicators ....... 43
Table 1-3-9 ..................................................................... 44
Table 1-3-10 Race Difference of Working Experience .................... 44
Table 1-3-11 Race Differences in Work Experience: Male ............... 44
Table 1-3-12 Race Differences in Work Experience: Female ............ 44

Table 1-3-13 Male working Hours Distribution During Senior Year

By Test Score Quartiles ....................................... 49

 

 

 

 

 

Table 1-3

Table 1-3

Table 1-4-

Table 1-4-

Table 1-4-

Table 1-3-14

Table 1-3-15

Table 1-3-16

Table 1-4-1-1

Table 1-4-1-2

Table 1-4-2-1

Table 1-4-2-2

Table 1-4-2-3

Table 1—4-2—4

Table 1-4-2-5

Table 1-4-2-6

Table 1-4-3-1

Working Hours Distribution During Senior Year by
Test Score Quartiles ............................................. 49
Working Hours Distribution During Senior Year by

SES ................................................................ 50

Working Hours Distribution During Senior Year by

SES ................................................................ 51
Probit For Job Holding .......................................... 54
Tobit For Working Hours ....................................... 55

The Effects of Local Economic Conditions on Job-
Holdings(Probit model) male:4064 ........................... 59
The Effects of Local Economic Conditions on Hours
Worked(Tobit model) male:4064 ............................. 60
The Effects of Local Economic Conditions on Job-
Holdings(Probit model) femalez4966 ........................ 60
The Effects of Local Economic Conditions on Hours
Worked(Tobit model) female: 4966 ........................... 61
....................................................................... 64
....................................................................... 64
The marginal effects of test score on the different

Working status in the senior year ............................. 69

xi

 

 

Table l-4

Table l-4

Table 1-4

Table 1-4.
Table lul-
Table 1-4-
Table 1-4-
Table 1-4-.
Table 1-4-

Table 2.3.}

 

 

Table 1-4-3-2

Table 1-4-3-3

Table 1-4-3-4

Table 1-4-3-5

Table 1-4-3-6

Table 1-4-3-7

Table 1-4-3-8

Table 1-4-4-1

Table 1-4-4-2

Table 2-2-1-1

Table 2-2-1-2

Table 2-2-1-3

Table 2-2-1-4

Table 2-2-1-5

Table 2-2-2-1

Table 2-2-2-2

Table 2-2-2-3

Table 2-2-2-4

The marginal effects of test score on the different

Working status ................................................... 69
Probit for job-holdings during the senior year ............... 71
Tobit for working hours during the senior year ............. 71
Male ............................................................... 73
Female ............................................................ 74
Male ............................................................... 74
Female ............................................................ 74
Family Socio-Economic Status ................................ 78
Family Socio-Economic Status ................................ 78

Review of Previous Studies on the Effects of High

School work Experience on College ........................ 103
Working Hours and First College Choices: Male ......... 106
Working Hours and First College Choices: Female ....... 106
Working Hours and First College Choices: Male ..........107
Working Hours and First College Choices: Female ....... 107
Working Hours and College Study Field: Male ............ 1 14
Working Hours and College Study Field: Female .........114
Working Hours and College Study Field: Male ............ 1 14
Working Hours and College Study Field: Female ........J 14

xii

 

Table 2-2
Table 2-2
Table 2-2
Table 2-2

Table 2-3~'

Table _-3-

 

Table 2-

J)

‘
- 1

Table 24-
Table 2-4-

Table 2-4-

Table 2-2-2-5 Working Hours and Major Choice: Male .................. 118
Table 2-2-2-6 Working Hours and Major Choice: Female ................ 118
Table 2-2-2-7 Working Hours and Major Choice: Male ................... 1 19
Table 2-2-2-8 Working Hours and Major Choice: Female ................ 1 19
Table 2-3-3-1 Summary Statistics for sub-sample of college students

In Academic Field ............................................. 138
Table 2-3-3-2 SES Distribution for the sub-sample of college students

In the Academic field(weighted) ............................. 141
Table 2—3-3-3 SES Distribution for full samples (weighted) ............... 141
Table 2-4-1-1 College Attendance Choice: Male ............................ 144
Table 2-4-1-2 College Attendance Choice: Female ......................... 145
Table 2-4-1-3 Changes in the Mean Probability of Attending Two or

Four Year College ............................................. 145
Table 2-4-1-4 Changes in the Mean Probability of Attending Two or

F our Year College .............................................. 147
Table 2-4-1-5 College Attendance Choice: Male ............................ 148
Table 2-4-1-6 College Attendance Choice: Female ......................... 149
Table 2-4-1-7 Male: Probit for College Attendance ........................ 154
Table 2-4—1-8 Female: Probit for College Attendance ..................... 155
Table 2-4-1-9 Male: Probit for College Attendance ...................... 156

xiii

Table 2-4

  
 

Table 2--1

Table 2-4.

Table 2-4.

Table 1-3.

 

Table 2--

Table 2-4-1-10 Female: Probit for College Attendance ................... 157

Table 2-4-2-1
Table 2-4-2-2
Table 2-4-2-3
Table 2—4-2—4
Table 2-4-2—5
Table 2-4-2-6
Table 2-4-2-7
Table 2-4—2-8
Table 2-4-2-9
Table 2-4-2-10
Table 2-4-2-11

Table 2-4-2-12

Table 2-4-2-13

Probit for Vocational Major Versus Academic
Major ........................................................... 160

Probit for Vocational Major Versus Academic

Major ........................................................... 160
Male: MNL Results for Major Choice ..................... 165
Male: MNL Results for Major Choice ..................... 166

Male: Changes in the Mean Probability of Choosing

Each Major .................................................... 166
Female: MNL Results for Major Choice ................... 167
Female: MNL Results for Major Choice ................... 168

Female: Changes in the Mean Probability of Choosing

Each Major .................................................... 168
Male: MNL Results for Major Choice ..................... 171
Male: MNL Results for Maj or Choice ..................... 172

Male: Changes in the Mean Probability of Choosing

Each Major .................................................... 172
Male: Weighted MNL Results for Choice of each

Major ........................................................... 173

Male: Changes in the Mean Probability of Choosing

xiv

 

Table 2-4-

Table 2-4-;

 

Table 2-4-:

Table 2‘4-2

Table 2‘4‘n

~

Table 2‘4‘:

Table App:

Table 2-4—2-14

Table 2-4-2-15

Table 2-4-2-16

Table 2-4-2-17

Table 2-4-2-18

Table 2-4-2-19

Table 2-4-2-20

Table 2-4-2-21

Table 2-4-2-22

Table 2-4-2-23

Each Major with Weights ................................... 173
Female: Male, MNL Results for Major Choice .......... 175
Female: MNL Results for Major Choice .................. 176
Female: Changes in the Mean Probability of Choosing
Each Major ................................................... 176
Female: Weighted MNL Results for Major Choice with
Weights ......................................................... 176
Female: Changes in the Mean Probability of Choosing
Each Major .................................................... 177
Probit models: Practical majors and non-technical majors
Male and Female ............................................. 180
Probit models: Practical majors and non-technical majors
Male and Female ............................................. 180
Male practical major cases: Endogeneity issues .......... 185
Probit for the practical choice: Instruments and practical
Major choice .................................................. 186
OLS for Working Hours During the Senior Year:

Instruments and Working Hours ......................... 186

Table Appendix 1 Reservation wages, Test Scores and Country Economic

Conditions among whites ................................... 62

XV

Table A;

 

 

Table Appendix 2 Reservation wages, Test Scores and Country Economic
Conditions among white College Student in Academic

Field ......................................................... 183

xvi

 

DII‘

 

I. Introduc

0T IO “T‘S’L \\
“Emilee
A5 0111' Tex ;

ILLS pli 5 11:1 \

CHAPTER 1
DETERMINANTS OF HIGH SCHOOL EMPLOYMENT

1. Introduction

Working during high school has been an important research topic for economists
and educators because of its policy implications for transition from high school to work
or to post secondary education. Most studies have focused on the effects of such working
experience on subsequent earnings, subsequent employment and subsequent schooling.
As our review below reveals, many of the studies indicate that high school employment
has positive effects on later employment and earnings (Ruhm 1997, Carr 1996, Meyer
and Wise 1982, Spephenson 1981, Stern and Kakata 1989, Marsh 1991). Some of these
studies explain these positive effects through observable or unobservable individual
heterogeneity (Light 1999 and Hotz et al 1999).

Other studies found some negative effects, especially of long working hours, on
educational aspects such as higher absenteeism from school (Lillydahl 1990), high drop
out rate (Marsh 1991, Steel 1991),lower college attendance (Carr, Wright and Brody
(1996), lower long-term educational attainment (Carr, Wright and Brody 1996).

This study focuses on two other important issues related to high school
employment. First, in chapter one, we investigate the determinants of high school
employment. We claim that by focusing on the determinants of high school employment,
we are able to understand better why high school employment has such positive effects on
later earnings and employment, and/or why it has negative effects on educational

outcomes. Second, in chapter two, we explore the effects of high school employment on

up?

 

  
    
   
  

pea-second.
We contend '
effects of. it:
educational a

In Its»
SCHOOl Sniff“),
the determing:

“‘6 find in 0;;

1.1 Praious
employ ment

The 1
another in 3<

post-secondary educational choices such as college attendance and college major choices.
We contend that in this investigation, we also are able to understand the two contrasting
effects of high school employment on later earnings (or employment) and later
educational aspects.

In this introduction, first, by reviewing previous studies on the effects of high
school employment on later earnings or later employment,l we explain why investigating
the determinants of high school employment is important, and then, we summarize what

we find in our investigation on the determinants of high school employment.

1.1 Previous studies on the effects of high school employment on later earnings and
employment.

The previous studies on the effects of high school employment differ from one
another in several important respects. In our review, we focus on the different
econometric methods, different measures of high school work experience, and control
variables included in their analyses. This review is summarized in Table 1-1-1. We
concentrate on the most recent studies (for other studies, see Ruhm 1997).

In most of these studies, the following wage (employment) equation is estimated
by the ordinary least squares method (OLS).2

Y =X,,j.oz+Z,.7+HIT/,0,6+a,,j (1-1)

u,

 

‘ In the second chapter, we review those studies on the effects of high school employment on later
education.
2 Also educationl attainment measured in terms of continuous variable can be estimated in this fiamework.

 

    

Trbk l-—l-l:l
alto: iuzr \.
33-: a1; 1);:

Sea: .\ 2:74.“

.hl I‘ \‘\‘
'.i.
IV”- l" 7‘.
l- I. '\C

a

._-

I- :"“‘«o'

I "
l \"'“I h
l

TH‘.

|: V ’-

' ‘
i \i‘ a-

.
i
l{ V
I l‘h‘. ‘.-"
‘
\ a.
— ‘l-l.
.
l-.
I \ .9
l " r-e \
l MSW-ff I...

liq-A" ._ .
(b “4~..“
. l y
l ”\Q\c.

I “..o

l

g?!”
N;'
In‘ L '- Vi\
:l.‘_'.*vlgl .‘
- II \C
\‘0
H

 

Table 1—1-1: The effects of high school employment on later earnings‘

 

 

 

 

Author Data, Samples Econometric Methods and Measurement of High School Key Findings
and and Dependent Specifications. Work Experience/Control
Source Variable Variables"
Carr, 1. NLSY. 1.0LS 1. The number of weeks I. for every additional hour
Wright, 2. 16-19 in In Y. = x_ p + 71.1 + 8. worked or the number of hours worked as a teenager in
and 1979, enrolled in .‘ ’ . ' ' per week, in 1978 (including 1978, one can expect to
Brody high school as of 2' Linear function ofhrgh summer). regardless of the cam $1.87 more in 1990.
(1996) May 1, 1979 SChOOI work experience high school grades in which 2. Positive effects on later
3. Earnings in measure. each student were enrolled at employment.
I990 (12 years that time. 3. significant negative
later after high 2. Hours worked during 1990 effects on educational
school and years of schooling attainment.
graduation (238) included
Ruhm I. NLSY. l. OLS, IV and treatment effect 1. two different measures. One l.students who work 10 to
(1997) 2. Those who method"". is hours worked during the 20 hours during their senior
were high school In y" _-_ X, fl + 7H, reference week in each grade year earn much more
freshmen or 2 and the other is the average (15~20°/o) annually 6-9
sophomores in 'l' 5H: + 5: hours per week in each grade. years later than students
1979, remained 2, Quadratic functions of high 2. Work experience and who have not worked and
enrolled in school work experience_ schooling afier high school are obtain
school through 3, Work experience in each not included. 2. Students who work
their senior year. high school grade are separate during high SChOOl work
588 males and in the estimation equation. more later than their
561 females. counterparts. (iii) Student
3. Average workers have higher
hourly wages occupational status later on.
over the 3-year 3. 1V method and
period 1988-90: treatment effects method
6-9 years after are not successful.
high school
Light 1.NLSY 1. GLS and lV/GLS (Hausman l. (i)Cumulative hours worked 1. High school students
(1999) 2. Male high and Taylor l981)"*" in grade 11 and grade 12 (only who work intensively tend
school graduates 1n Y = X. a + 2 fl excluding summer time) to take more vocational
who have not " " ' divided by 1800 in order to courses while taking fewer
attended college + 7H,. + 8" make it a year unit, assuming courses in other subjects.
during the first 2' (1) Linear function ofhigh that 25 hours per week on the 2. They work far more
nine years after school work experience average is one year of high intensively after high
high school measure. (ii) categorical school work experience. (ii) school graduation.
graduation. variable for average working the average number of hours (iii) After controlling for
(685 male and hours per week (iii) Int era cti on per week in grades ”-12. the above two factors, high
5689 between years ofhi h h 1 Three dummies 1-10, ”-20, school employment proves
_ g so 00 .
observations) work experience and the 21 or more are made for this to have a small and
3. Hourly wages dummies of years after high measure. relatively short-lived effect
for each year school on post-school wages.
during the 9 3. Basically no distinction 2. Work experience after high 3. A young man who works
years after high between working in the junior school and Number of Credits an average of 2: ther week
school . - - in each subject: in grades 11 an 1 earns
graduation. year and working m the senior (a)Humanities/social studies about 6% more than his

 

 

year. However, only work
experience during the senior
year is included in some
estimation.

 

(b) mathematics/natural
science (c)vocational subjects
((1) other subjects

nonemployed counterparts
6 years after high school
graduation. The estimated
premium is smaller at other
points in his early career.

 

 

 

 

.
5

as gt'l‘

. . l \ a . .. .. T
\ ¢ .. . . s

h ... c .. as \ \ c
a. . .... an \ “E -1 a. x

.u. ‘ ¢ \
1‘ ... o..
. v : ..
..-. a a

 

 

 

 

hdeyer
and
VVEe
(1982)

1. NLS-72 (1972
High School
Seniors) .

2. High school
graduates who
are not enrolled
in college.

3. Weeks
worked or wage
rates in the year
in one, two,
three and four
years after high
school
graduation.

1. The following wage or
weeks worked equation is
jointly estimated with the
college attendance equation
because of potential bias due to
the possible correlation

between A" and 8,.
111),“ = Xtifl+7Hi'
+ aA“ + a,

where A“ is college

attendance choice.

2. Categorical variable for the
average hours worked during
senior year.

1. Categorical variables which
represent the average work
hours for either paid or non-
paid jobs during the senior
year: categories are for weeks
worked: 1-5, 6-10, 11-15, 16-
20, 21-25, 26 to 30, and 31 or
more.

Categories are for wages are]-
15, 16-30, and 31 or more

2. school years and work
experience are controlled for
in the wage equation
estimates.

1. Those who worked
between 16 and 20 hours,
and those who worked over
30 hours in the senior year
earn 12%, 18% more
respectively than those who
did not work at all.

2. An additional five hours
per week in the senior year,
up 20, is associated with
1.5 more weeks worked per
year in each of the four
post-high school years.

 

Hotz,
Xu,
Tienda,
Ahituv
l 999

 

1.NLSY
2.Separate
samples for
Whites, Blacks,
Hispanics.
Those males
who were
between the ages
of 13-16 in
1978, and who
would be
between 24-27
at the 1989
interview.

3. Wages in each
working

status( j ) at age
17, 22, and 28-
(1) if j =1,
working while in
school, (2)

if _] =2, working
in non-school,
part-time
employment, (3)
if j =3, working
in full-time
employment.

1. (i) OLS (ii) simultaneous
estimation of wage equations,
with selection equation to
correct for (contemporaneous)
selectivity of wages, associated
with different types of
employment. (iii) Unified
dynamic, selection model
(Cameron and Heckman 1991)

lnY =X,,a+Z,.,6

it}
+ u “j

The selection equation is as
follows:

Selection into employment
status 1: school and part time
only) or status (3) : full-time
employment ) or status (4):
part-time employment)

VI]: = max!“ {V11} }9

l

V

“I - utility if individual i
chooses state ft from six states
at age tto max his utility.

V"j = Xuy + Z6 + a,”

2. The number of years of work
attending high school are
entered in the linear form

 

 

1.Years of working while
attending high school. No
distinctions across work in
different grades. No measure
of work intensity. Students
who worked are treated as
homogenous.

2. After making six, mutually-
exclusive activity categories,
”*", they construct measures
of (i) number of years of work
while attending high school.
(ii) number of years of work
while attending college. (iii)
number of years of part-time,
non-school-related, work. (iv)
number of years of flail-time
work. (v) number of years of
accumulated number of years
attending school and not
working (vi) number of grades
completed. (vii) number of
years spent enlisted in the
military. All these variables
are included.

 

By comparing the returns to
a year of working while
attending high school to the
return to a year of
schooling without working,
they find that after
controlling for unobserved
heterogeneity, there are no
positive effects of working
while attending high school
on wages.

 

Notes: " This summary table focuses on the earnings equations. Those studies on the effects of high school
employment on later schooling are not summarized in this table. " The common control variables: an ability measure (
test scores), and family background variables are included in all the studies. Thus, not reported. *"The instrumental
variables used for identification in Ruhm’s study are geographic characteristics (local unemployment, the region of the
county, SMSA and urbanity ) in the senior year. ""The instrumental variables used for identification in Light study
are (i)(Deviations from within-person means of each time-varying regressor) + (the within-person means of each
exogenous regressor) (ii) per capita family income during the senior year of high school.(iii) unemployment rate
during the senior year of high school.
(iv) a dummy which indicates ‘urban’ if student lives in that year (v) a dummy (the high school offers a distributive
education program. "”* The following are the categories: (1)school only, (2) school and part-time only (3) full-time
employment. (4) part-time employment ( and no school attendance) (5) enlisted in military. (6) other non-school and
non-work activity.

 

 

  
   
   

where I, \.
ended: at a
status I, j l. s.
exr‘enence d.
.l'“, represent
reticular err:
high schwl. ;
change 0\ er ‘
char-amen Silt
time Hid] " U‘.
and employ, i

In th

“hen high 3

he . Wk.
l-aT‘ - S
at"? 691:»

where Y

m. is the hourly wage or earnings (or a measure of employment such as weeks

worked) at a certain year (t) after high school graduation, in a particular employment

status ( j ), such as working while attending college; HW, is a measure of working
experience during high school, such as the average hours worked during the senior year;3

X ”I represents a vector of certain characteristics of an individual iat the year t in a

particular employment status ( j), such as the previous working experience except during
high school, and the years of schooling; Z , represents other control variables which do not

change over time, such as an ability measure, family background, race, or geographical
characteristics. However, most studies estimate a single wage equation at one point in
time without distinguishing employment status. Thus, in most cases, we can omit the time
and employment status subscripts.

In the studies that estimate the single wage (employment) equation by OLS and
when high school work experience is specified linearly", the positive and significant

values of the coefficient ,6 is interpreted as evidence of the positive effects of high
school employment on later earnings or employment (Spephenson 1981, Stern and
Kakata 1989, Marsh 1991, Carr 1996). In these studies, the sub-scripts t and j are

assumed to be fixed in the equation (1-1). The most recent such study is Carr et a1.
(1996), which investigates the long-term effects of high school employment on earnings,

employment, and educational attainment, 10 years after high school graduation. Their

 

3 Some studies specify this high school work experience in the quadratic form or a categorical variable
which represents working hour categories and in those cases, this term a vector of variables which are
related to high school work experience

4 When high school work experience is specified in the quadratic form and the effects of two terms of high
school work experience are different, the two effects can be adjusted to see the net effect.

 

   
  
   

md} inc UF-
[hc Nalltlnai
manna} h.
annual earn:
with RSI: 0’.
find that each
number of co:

Home
school \\ earl-t
regardless of
maple xx ere
the junior _\e
Was enrol led
Moreover. ti

“rim
3&1“; a

\T\3C

‘
. ”’Hc cw
W-L , \‘Ilf’.
._ kl” a? '_ t“
u cc. -

study includes a rich set of controls for background characteristics. Using a sub-sample of
the National Longitudinal Survey of Youth (NLSY), Carr et al. (1996) show that for every
additional hour worked during high school in 1978, one can expect about $2.00 more in
annual earnings for wage and salary earners in 1990. Also, they estimate the equation
with years of school completed 10 years after high school as the dependent variable, and
find that each additional 100 hours worked in high school in 1978 reduce by .03 to.04 the
number of completed years of post-secondary schooling.

However, their study has some limitations. First of all, their measure of high
school work experience is imprecise because they use the total hours worked only in 1978
regardless of what grade students were enrolled in. Because the respondents of their
sample were 15-18 years old in 1978, hours worked can represent either hours worked in
the junior year or in the senior year or other grades, depending on what grade a student
was enrolled in 1978. Then, hours worked in other grades are naturally omitted.
Moreover, they do not distinguish between working during the academic year and
summer vacation.5 . Second, they do not control for working experience after high
school.6 We can expect that high school work experience is positively related to later
work experience. Thus, high school work experience should indirectly increase later
earnings through later working experience. Since Carr et a1 omit working experience after
high school, we can not see the direct effects of high school working experience on later

earnings. Third, a more fundamental problem with this kind of OLS estimation is a

 

5 Most other studies distinguish working during the academic year and summer vacation and focus on the
work experience during the academic year. Because, during the summer vacation, there is not much
academic obligation from students, these studies do not regard this experience as a special work experience
which affects their academic performance or work productivity.

44.4

 

    
   
  

rcssible sci
because he l‘.
betueen H55
experience ii

To >‘4.

melt OdIIVH

Where V r..-
, A

of r~ .

possible selection bias. If a student works during high school and earns more later
because he has a higher ability which is not observable, this will cause correlation

between HW, and a, , and a positive bias in the estimates of the effects of working

experience during high school on wage or earnings.

To solve this endogeneity problem, several studies use the instrumental variable
method (IV) (Lillydahl 1990; Ruhm 1997; Light 1999). In the first stage regression,
typically, the following reduced form of working hours equation is estimated,

HW, =X,07r+Z,¢+S,K+u, (1-2)
where S, represents the instrument variable for identification; HW, represents a

continuous variable which measures high school work experience; and X "I and Z , are
the same as in (1-1). After this equation is estimated, the predicted value of HW, instead
of HW, itself is put into the second stage regression for the earnings (wage or
employment) equation.

Two main studies which use IV are Ruhm (1997) and Light (1999).7 With the
NLSY data set, Ruhm (1997) utilizes both IV and OLS method when estimating single
wage or weeks worked equations. His sample is those who were high school freshmen or
sophomores in 1979, and who remained enrolled in school through their senior year and
were interviewed in 1991. The main dependent variable is the average annual earnings

over the 3-year period of 1988-90 divided by hours, or weeks employed (for the

employment equation). There is no distinction according to employment status. Thus, the

 

6 Carr et al include hours worked during 1990. However, it is not a correct measure of work experience after
high school.

 

  
   
   

sub-script 1 .

{Wei Rah:
is bears \\ or.-
n‘orked per t.
and the junior
regression am

To Cu:
background t
there is a sly

Eater earnin gs

the.“ Scalar}

sub-script tis fixed and the sub-script j can be dropped from equations (1-1) and (1-2).

Ruhm’s measurement of high school employment is much more accurate than Carr et al’s
(1996). Ruhm uses two different sources of information on high school employment. One
is hours worked during the week prior to the survey data. The other is average hours
worked per week from the work history file in NLSY. Hours worked in the senior year
and the junior year are separately recorded and specified in the quadratic form in the
regression analysis.

To control for the possible observable selectivity bias, Ruhm includes extensive
background variables with the OLS method. From OLS regressions, he concludes that
there is a significant positive effect of hours worked in the senior year of high school on
later earnings. For example, those who worked between 10 and 20 hours per week during
their senior year earn 6-16 % annually more and have higher occupational status later
than those who did not work at all.

To control for the possible unobservable selectivity bias, Ruhm uses geographic
characteristics (local unemployment, the region of the county, SMSA and lived in urban
area during high school) as instruments. However, his IV estimates generate large
standard errors for the ‘hours worked’ coefficients. Even though the coefficients of hours
worked during the senior year by the IV method are much larger than the corresponding
OLS estimates, the standard errors of those coefficients are much larger than the

corresponding OLS standard errors.8 resulting in imprecise measurement of the effect of

 

7 Lillydahl (1990) does not describe his method very well. He does not specify what his instruments are.
Thus, here, we do not review his study.

a For an example, one of Ruhm’s 1V estimates of the coefficient of hours worked during the senior year is
.1093 while its standard error is .2625. One comparable OLS estimates of the coefficient is .0168 while its
standard error is .0066.

 

bagh school .
and schoolzri

Li git:
she uses a re

NLSY 3am;

  
  

sewn-dam} er:
restriction m:
“PIMP-em >"

schooling is n

high school employment. Another problem with Ruhm’s study is that working experience
and schooling after high school graduation are not controlled for.9

Light’s study (1999) is quite different from Ruhm (1997) in several respects. First,
she uses a restricted male sample of non-college bound high school graduates from the
NLSY sample. Her sample consists of high school graduates who report no post-
secondary enrollment during their first 9 year following graduation. This sample

restriction means that we can drop the sub-script j in the equation (1-1) because

employment status is regarded as the same in the relation to college attendance, and that
schooling is naturally controlled for because all individuals in the sample are high school
graduates.

Second, her measure of high school employment and how she controls for it in her
analyses differ from Ruhm’s (1997). One of Light’s measures of high school work
experience is the cumulative number of hours worked during the junior and senior
academic years divided by 180010 and is specified in linear form in the regression
analysis. She also uses work experience during the senior year divided by 900 and
specified in linear form.11 To control for non-linearity, she also includes in her
specification an alternative measure of high school employment consisting of three
dummy variables indicating whether the average number of hours per week worked in

grades 11 and 12 is between 1 to 10 hours, 11 to 20 hours, or more than 20 hours. Later,

 

9 Ruhm(1997) uses also the treatment effects method. In the first stage regression, the probit model is run,
where job-holding in the senior year is the dependent variable. In the second stage regression, the inverse
Mill’s ratio from the probit model is included as an additional covariate. Another interesting thing is that
Ruhm never attempts to control for the possible selection bias in the weeks worked equation.

‘0 By doing so, she converts high school work experience to “one-year unit”, assuming that an individual
who works 25 hours per week on the average for two, 36-week long academic years earns one year of high
school work experience.

 

   
    

she also in
junior and x
to see if the
Thin-

least Squat-e
method 11V (
logarithm of
irst 9 years
together b}
experience
harnariities
addition 11‘
uIlOlc‘Sen'a

'I 9. ‘
INCHEITC

7:.
Hr

she also includes interaction terms between high school work experience during the
junior and senior year and dummy variables indicating years since high school graduation
to see if the effect of high school work experience changes over time.

Third, Light estimates multiple wage equations by the method of Generalized
Least Squares (GLS) and a variant of the instrumental variables/generalized least squares
method (IV/GLS Hausman and Taylor 1981). The dependent variable is the natural
logarithm of the CPI-deflated hourly wages earned on jobs held in each year during the
first 9 years after high school graduation. Thus, nine wage equations are estimated
together by the GLS. To control for the observed heterogeneity, she includes work
experience after high school graduation and the number of high school credits in
humanities/social studies, mathematics/natural sciences, and vocational subjects in
addition to other various background and ability variables. To control for the
unobservable heterogeneity, she utilizes IV/GLS with the same instrument variable for
identification in each equation. 12

By comparing GLS and IV/GLS estimates, Light concludes that an individual
who works 25 hours per week on the average during his junior and senior years of high
school, earns 6% more per hour than one who does not work at all, while an individual
who works 10 hours per week on the average in his senior year and zero hours in his

junior year earns 1.2% more per hour. However, these positive effects are short-lived as

evidenced by the fact that these wage premiums fall to zero within 8 years of high school

 

” However, she never uses working experience during the junior year and senior year separately in the
regression analysis.

'2 The instrumental variables are 1) the deviations from within-person means of each time-varying regressor
plus the within-person means of each exogenous regressor; 2) per capita family income during the senior
year of high school; 3) local unemployment rate during the senior year of high school; 4) a dummy variable

10

graduation. Thus, the direct effect of high school employment on later earnings turns out
to be small and relatively short-lived compared to Ruhm’s findings. One thing to notice
in comparing the IV/GLS estimates with the GLS estimates, is that the coefficients of
hours worked during high school by IV/GLS are much larger than those by GLS, which
suggests that high school employment intensity is negatively correlated with the
unobserved, personal heterogeneity.

There are other studies which are concerned about another possible bias in the
estimates of the effects of high school employment on later earnings. For many high
school graduates, 1 to 4 years after high school graduation is still the transition period
from schooling to work. During this period, individuals need to make choices about
whether to attend college and whether to work during their college years if they attend
college, and so on. Thus, some studies are concerned about the possible bias in the
estimates of effects of high school employment on later earnings (employment) due to the
effects of such individuals’ choices of employment or schooling. For example, estimating
the unbiased effects of high school work experience in the earnings equation for non-
college bound high school graduates is difficult if an individual’s college attendance
decision is related to his wage rates or if college attendance is related to working
experience during high school.13 To control for such selection bias, wage equations are
estimated with the selection equation. Meyer and Wise (1982), and Hotz et a1 (1999) are

examples of such studies.

 

which indicates ‘urban’ if student live in a city in that year; 5)a dummy variable which indicates whether the
high scth offers a distributive education program or not.

'3 Many studies show that high school working is negatively related to college attendance (Carr, Wright and
Brody 1996, Marsh 1991.)

 

 

When Meyer and Wise (1982) estimate the effects of high school employment on
the wages and weeks worked during the first four years after high school graduation for
those not in college in each year, they estimate simultaneously the following college
attendance equation along with equation (1-1).14

61.. =1 ifQ.. 20
q“ = 0 otherwise
Q1: =Xiyl+Zig+Wp+qu (1'3)
where q“ is college attendance status at time I; Q” is an individual’s utility for school

attendance at time t ; and W represents a vector of other variables which could affect
school attendance choice. ‘5

When Meyer and Wise (1982) estimate the effects of high school employment in,
their measure of high school employment is a categorical variable for hours worked in
jobs held during the high school senior year. After controlling for bias due to
Simultaneous decisions on working and college attendance, Meyer and Wise find a
Significant positive effect of working hours during the senior year of high school on later
earnings for high school graduates who were not attending college at the year. For
example, individuals who worked 16-20 hours per week are estimated to earn annually
about 12% more than those who did not work at all in high school even after work

experience after high school is controlled for. They claim that “depending upon the

amount of work in high school and estimated weeks worked based on other

 

" Thus, we can rewrite the equation (1-1) as In Y” = X n. ,6 + 7H,. + aA". + 8,. , where A". is college

attendance choice. Meyer and Wise are concerned about the possible correlation between errors in the equation (1- l)
and (1-3).

l2

 

~..n\

Ci

(1"

 

characteristics, the estimated ‘effect’ on annual earnings of high school work could be as
high as 30 or 35%” (p279). These effects are much larger than Light’s estimate16 and
even larger than Ruhm’s estimates.

However, the Meyer and Wise study uses a different data set, NLS-72, while most
studies use the NLSY. The respondents of NLS-72 are an earlier cohort than the NLSY
sample. Another study which uses the 1966-74 Youth in Transition Study data (YTS), the
sample of which is an earlier cohort than the NLSY sample, also shows that those
employed during high school earned about 27% more than others five years after
graduating from high school (Mortimer and Finch 1986).17 However, we do not know
whether the changes in the estimated gains from high school work experience arise from
whether the selection into student employment is vary across cohorts or whether these
gains have truly changed over time.

Another study which utilizes a similiar simultaneous equation approaches as
Meyer and Wise is Hotz, et al. (1999). They use the sample of NLSY males who were
between the ages of 13-16 in 1978 and who were between 24-27 in 1989. By tracing each
individual’s school and work experience during the period between 1978 and 1989, Hotz,
et al. (1999) create the following Six mutually-exclusive activities statuses: (1)school

only, (2) school and part-time only, (3) full-time employment, (4) part-time employment

 

'5 Meyer and Wise do not include the high school employment measure in the control variables for the
college attendance equation because of their concern about the possible reverse causation.

'6 Meyer and Wise’s estimates are comparable with Light’s in the senses that (1) both the samples are non-
college bound high school graduates even though Light’s sample is more restricted, and (2) both working
experience are controlled for. However, they control for different sources of selection bias. Light controls
for selection bias directly related to high school work experience by IV/GLS method while Meyer and
Wise control for selection bias rising from college attendance choices by simultaneous equation method.

'7 We do not review their study in detail because (1) their study focus on the other outcomes such as
academic achievement and educational attainment, (2) their study relies on the mean difference analysis
(ANOVA) instead of regression analysis; moreover their study does not control for selection bias.

l3

 

nee-r-

 

it ‘~.

without school attendance, (5) enlisted in military, and (6) other non-school and non-
work activity. They then measure the number of years of work while attending high
school, the number of years of work while attending college, the number of years of part-
tirne, non-school-related, work, the number of years of full-time work, the number of
years of attending school and not working, the number of grades completed, and the
number of years spent enlisted in the military. All these variable are included as
independent variables in the equation (1-1). Thus, high school work experience is defined
as the number of years of work while attending high school.

Then, the wage equations in the form of (1-1) in the employment categories of the
above named (2), (3) and (4) at age 17, 22, 27 are simultaneously estimated with the
status choice equations. Since more than two alternatives are given, the status choice
equation becomes complicated as follow. The individual ichooses status k from the six
feasible categories of status named above.

V”r =maxj€J {VIU},

l

where V"j is the utility if individual i chooses state k from six states at age I. The utility

function can be defined thus:

V,” =X,,y+Z,5+nuj (1-4)

where X n is a vector of age specific variables, and Z , is a vector of time constant
individual specific variables. First, error terms in (1-1) and (1-4) are assumed to be
correlated only across j (to correct for contemporaneous selection bias for different types

of employment). Later, these error terms are allowed to be correlated across j and t (to

correct for dynamic selection bias).

l4

 

Tue:

 

([1

(I:

’77

After equations (1-1) and equation (1-4) are estimated simultaneously, the rate of
return to the years of work while attending high school is compared to the rate of return to
the years of attending school and not working. Hotz, et al. find that, when the bias due to
simultaneous decisions on these choices at a certain point in time is controlled for, the
rate of return to the years of work while attending high school is much higher than the
rate of return to the years of schooling only. However, when, the dynamic selection bias is
controlled for, there is no larger return to the years of work while attending high school
than the years of schooling only. Hotz, et al. conclude that the positive effect of high
school employment on later earnings is Spurious and arises from heterogeneity across
individuals. These results greatly contrast with Ruhm’s results. In another respect, these
results also contrast with Ruhm’s IV and Light’s IV/GLS results, in which the rate of
return to high school employment becomes larger when the unobserved individual
heterogeneity is controlled for. Hotz, et al. (1999) implies that the high school work
experience is positively correlated with unobserved, personal heterogeneity, while Ruhm
(1997) and Light (1999) suggest that high school work experience is negatively correlated
with unobserved, personal heterogeneity.

However, we need to pay attention to the basis of Hotz, et al.’s (1999) findings.
Firstly, Hotz, et al. (1999) do not distinguish schooling without work in high school from
schooling without work in college. Thus, comparing the rate of return to the years of
work while attending high school to the rate of return to the years of schooling only might
not be appropriate to identify the effect of high school employment. Secondly, as Light
(1999) points out, student workers are treated as homogenous group regardless of how

many hours they work. Working a few hours and working more than 20 hours while in

15

 

 

(I!

(+-

(1"

tr

\4

~A1

school is treated the same. Also, working is treated as the same regardless of what grades
students work in. Working a couple of hours per week Should be distinguished from
working 20~30 hours per week. F ourthly, like Light (1999), the working experience after
high school is controlled for in this study by the measures of various experiences. Thus,
there still exists the possibility that high school employment has some positive effects on
later employment even if there is no such effect on wages.

Our review of the previous research on the effect of high school employment on
later earnings reveals that the effects of high school employment are very complex and
the results vary greatly across studies. In this chapter, we focus on the determinants of
high school employment rather than its effects. By summarizing issues raised by the
above review, we contend that identifying the determinants of high school employment is
important in understanding the later effects of high school employment. The important
issues in the above review as follows.

First, definitions and measurements of high school work experience differ across
studies. Different Specifications, not to mention different econometric methods, seem to
affect the results. To find the correct estimates the effects of high school employment on
later earnings or employment, above all we need to define high school work experience
clearly because the different measures of high school employment could capture the
different characteristics of student workers who have different high school work
experience. Ruhm(1997) uses hours worked per week in the senior year and the junior
year separately'8 and specifies these hours in the quadratic form in the regression analysis.

Meyer and Wise (1982) use the average hours worked per week during only the senior

16

 

l .u.‘ c .c «N. \ .LJ. 0PM .. x... .. . .. u . .

 

year and uses a categorical variable in the regression analysis. Carr, et al. (1996) use
hours worked in 1978 regardless of grades, for those who were 15-18 in 1978, and
specifies these hours in linear form in the regression analyses. Hotz, et al. (1999) treat all
student workers as homogenous group regardless of how many hours they work and
specifies them in linear form. Light (1999) is more careful in measuring high school work
experience and specifies it in various functional forms. However, her sample is very
restricted.

Second, most studies which find lower or no positive effects of high school
employment on later earnings,19 control for working experience after high school (Light
1999 and Hotz et a1 1999). By contrast Ruhm (1997) and Carr et a1 (1996) who use the
same NLSY data set do not control for working experience and find large positive effects
of high school employment.20 These results together imply that even when the direct
effect of high school employment on later earnings is small, there still are strong positive
effects of high school employment on later employment opportunities. Then, we need to
explain why such positive effects occur. However, there after no studies which attempt to
control for bias in the estimates of the effects of high school employment on after-school
employment.

Third, the results also seem to depend on the samples used different studies. Some

studies use NLS-72, which is for the cohort of high school seniors in 1972 or some use

 

'8 As mentioned earlier, Ruhm uses hours worked during the reference week or hours worked per week on
the average.

'9 When we compare only the results by a similar method such as OLS or GLS results across studies,
inclusion or exclusion of after-high school work experience affects the results greatly.

2° Using NLS-72 data set, Meyer and Wise (1982) find large and significant positive effects of high school
work experience on later earnings even after they control for the working experience. However, since they
are using different data set, we do not know how much controlling after-school work experience reduces the
positive effects.

YTS, which is for the 2 to 3 years earlier cohort than the NLS-72 cohort, while other
studies use NLSY which is the sample of those who were born between 1957 to 1964 and
are 14 to 21 years old on January 1 1979. The positive effects of high school work
experience on later earnings for earlier cohorts seem to be larger than for more recent
cohorts. High school employment has not changed in both terms of job-holding and
working hours since late 1960 (Ruhm 1997). However, the characteristics of student
workers could have changed. The changes in the characteristics of student workers could
have caused the changes in the effects. In our study, by using two comparable data sets
across time, NLS-72 and the HS&B senior data set, we investigate any changes in the
characteristics of student workers from 1972 to 1980.

Fourth, as mentioned earlier, studies which try to control for the selectivity bias
imply two opposite possibilities in the relationship between high school employment and
the unobserved individual heterogeneity, depending on the econometric methods. If the
unobserved individual heterogeneity represents a kind of individual productivity, the
relationship could be either positive or negative. Examining the relationship between high
school employment and observable characteristics might help to understand better the
true relationship between high school employment and the unobserved individual
heterogeneity. For example, if high school employment is positively correlated with a
ability measure which is observable, high school employment is might positively
correlated with the unobserved individual heterogeneity.

In identifying determinants of high school employment, we focus on the following
elements which could be related to high school employment: the local economic

conditions, an academic ability measure, and family background, and difference in the

18

 

 

tiers of h. ~

Jegree Firs:

,.
s 1 Q - .- '
rotations. a .

 

 

I ' I -
putts “111‘. >‘.'.

r
9m“ \ ' ,3 0 «
Sir-rilc\m\‘ 1‘

effects of these factors on high school work experience between two years, 1972 and
1980. By doing so, we address several issues raised from our literature review to a certain
degree. First of all, if high school student workers live in areas with good economic
conditions, a factor which causes higher demand for their labor, and they also live in
places with similar conditions after they graduate from high school, the gains in
employment and earnings could be spuriously related to high school work experience;
i.e., the employment (or wage) and the labor market conditions are strongly correlated.
Actually, most studies investigate the short-term effects of high school work experience
on earnings as we see in our Table 1-1-1 (1 to 5 years after high school graduation),
which could be problematic because the local economic conditions might not change
dramatically during a short period.

Second, examining the relationship between high work experience and academic
ability (or family background) could explain why different measures (or different
Specifications) of high school work are related to different results in estimates of its
effects to some degree. For example, students with higher academic ability ( and/or better
family background) are less likely to hold jobs in earlier grades because of their own or
their parents’ commitment on their future education.21 If students with higher ability (or
better socio-economic status,) work more intensively during their senior year, the larger
effects of working hours during the senior year than working hours in the junior year, on
later earnings and employment (when they are separately entered in the earnings equation

as Ruhm 1997 ) could be simply selection bias. Or there is a possibility that students with

 

2' Since main required activities for high school students are academic activities, working while in high
school might be so different from working without such academic duties. College preparation is also an

19

 

 

 

 

I' ‘

51 Us."

“43,4881
5

(U
r‘
(b
..,

o
.-

r2
.-

9
i b

"(J

“1"“
Law:

’“4

‘1
mitt:

 

higher academic ability ( and/or better family background) might work moderate hours
even if they work. Thus, the non-linear specifications of high school work experience
might just capture the hidden non-linearity relationship between working hours and
academic ability.

Third, by observing any changes in academic ability and family backgrounds of
high school workers between two years, 1972 and 1980, and comparing the different
results in the estimates of effects of high school employment across cohorts by the
previous studies, we can suggest one possible answer why there is difference in the
effects of high school employment on later earnings. If more students with higher social
status and higher academic ability work more in more recent years than before, and they
have the smaller positive effects for high school work experience, the changes in positive
effects of high school employment might come from other sources.22

Fourth, careful observations of the relationship between those observable
characteristics such as test scores or family socio-economic backgrounds, and high school
employment might help to understand the relationship between those unobservable

characteristics and high school employment.

1. 2 Major Findings
Using two comparable, nationally representative high school senior data sets,

NLS-72 (National Longitudinal Study: the High School Class of 1972), and HS&B (High

School and Beyond), we observe several important characteristics of the determinants of

 

issue related to working during high school. Most students could finish their college preparation by their
junior or early senior year.
22 We discuss this issue further in the conclusion.

20

 

,.
5
p

,.
IAI‘
\

.‘\ —,
sil'

high school employment. Firstly, factors that indicate job availability, such as local labor
market or economic conditions, race and community type, are very important in the
determinations of job-holdings and working hours during high school senior and junior
year. Secondly, academic ability is related to employment during the senior year in non-
linear fashion. Students with the highest and lowest academic ability tend to work fewer
hours than those who are in the middle range of academic ability. Thirdly, more able
students tend work in the senior year rather than in the junior year. F ourthly, students
from white middle class families work most. Students from low and upper class tend to
work less. Fifthly, more able and more affluent students hold jobs more in 1980 than in
1972, and work more intensively in 1980 than in 1972. These finding are explained in
more detail as follows.

(1) With respect to factors for job availability: with the HS&B senior group data,
which has the labor market indicator file, we find that for both males and females, county
unemployment rates are negatively correlated to job-holdings or working hours during
high school. Other market indicators such as the percentage of people employed within a
county, the growth rate of employment, the average per capita income, and its growth
rate, have a positive impact on high school senior student employment.

Race could be an indicator for job availability. Blacks, Hispanics and Asians are
much less likely to work during high school than Whites and they work less if they do
work. Since youth unemployment rates are much higher and persistent among these
minority groups, especially Blacks, the significant and negative effects of being Black on
job-holdings and working hours during high school senior year certainly shows the lower

job availability for them.

21

Community type is also important for job availability, for females, those who live
in urban and suburban areas work more than those who live in rural areas. For males,
community types do not matter possibly because students who live in rural areas also
have as much work as those in urban and suburban due to family farming. Considering
these relationships, we can infer that high school employment is strongly related to job
availability. High school student workers or high students who work intensively live in
the areas (or conditions) where there are more jobs around them or more demand for their
labor.

(2) With respect to the ability measures (standardized test scores): The linear
relationship between high school employment in the senior year and test scores has
changed noticeably between 1972 and 1980. In 1980, for both males and females,
students with high test scores have higher probability of job-holdings or of working heavy
hours after other factors are controlled for, while in 1972, students with high test scores
were less likely to hold jobs or working longer hours than others. In 1980, the positive
relationship between test score and senior year employment gets stronger when ‘working
in junior year’ is controlled for because ‘working in junior year’ is negatively related to
test scores. Thus, more senior students with higher academic ability tend to participate in
labor force and employment in 1980 than in 1972.

However, in both years, the relationship between test scores and employment
(job-holdings and hours worked in the senior year) seems to be nonlinear rather than
linear, possibly because the students with lower test scores have difficulty finding jobs
and those with high test scores tend to concentrate on their academic activities during

high school. Thus, students with high test scores tend to work moderate hours if they

22

 

  
  
  

Wile \Vl‘it‘l
inioabinai
lllt’COt’lllCl:
anti is i
ii test scores
Tris non-hr.
equations.
131‘
ll”. imlilb
COl’.dl'.lOl'ls
0i SOClo-e
l’illll. \io
SCHIOT R;

ailment f.

work. When we put test scores and test scores squared together as independent variables
into a binary dependent variable models of job-holding or a tobit model for hours worked,
the coefficient of test score is always significant and positive, while that of test scores
squared is almost always significant and negative. Results from a categorical specification
of test scores in the those regressions also reflect the non-linearity of the relationship.
This non-linearity make it difficult to control for ability bias in the earnings or wage
equations.

(3) With respect to family background: in both 1972 and 1980, students who are
in families with medium socio-economic backgrounds work more than those from low or
high socio-economic status after other factors such as test scores, local economic
conditions, and previous working experience are controlled for. However, the importance
of socio-economic status in the determination of high school employment gets reduced in
1980. More student from affluent families seem to work in 1980 than in 1972 in their
senior year even after all other factors are controlled for. However, students from more
affluent family still seem to work moderate hours even if they work.

(4) With respect to working in the junior:23 The working in the junior year and test
scores seems to be linearly and negatively related.24 Students with higher academic ability
are less likely to hold jobs during their junior year and they are more likely to work fewer
hours even when they hold jobs than students with lower or medium academic ability.
With other factors held constant, the coefficients of test scores are negative and
significant in the linear form of test scores, while the coefficients of test scores and test

scores squared are not significant when the quadratic form of test scores are specified.

 

2’ We have the information on working in the junior year only in HS&B.

23

 

 

Trip, to 11.:
ii the limo:

hfl

 

factors are co
Toning ex;
dezearrnaa
thither \\ o

VIP“) .l‘nzr
“.11.; um.
5

t
lieerogentz
5

iiih’ls Oi‘lll

lie dixuss the F

Also students from families with high socio-economic backgrounds are less likely to
begin to work in the junior while students from middle class family still work most even
in the junior year.

Another important fact regarding working in the junior year is that after other
factors are controlled for, those who work in the junior year work more in the senior year.
Working experience during junior year is a single predominant factor in the
determinations of job-holdings and working hours in the senior year. We are not sure
whether working experience during junior year help students to find jobs and work more
during their senior year, or whether those who work in both years have an unobservable
heterogeneity such as a tendency to work.

(5) With respect to other factors: in addition to these influences, there are some
important factors, such as whether the mother works or not when students are in high
school, whether the respondents have children or not, which high school programs they
are in.

In the next section, previous studies on the determinants of high school
employment are briefly reviewed, and in the third section, our data and strategy are
described. In the fourth section, the results are explained, and in the conclusion, the
implications of our findings are discussed in the context of the previous research on its

effects of high school employment on later earnings and employment.

 

2’ We discuss the possible casual effects relationship between these factors later.

24

 

 

ll’rhious '

Tilt".

 

H1 1 . l 1“
X3001 .‘lu'u‘s

 
   

l I I
this $11.11.
‘13 l. $023.1.

sales is to

Engage \1

tenor of

2. Previous Studies on the determinants of high school employment.

There are a few previous studies which investigate the determinants of high
school students’ employment, such as job-holdings and/or hours worked. Even within
these small number of studies on this issue, the determinants of high school employment
is a secondary concern and is only briefly examined because the main concern of these
studies is to explore the effects of high school employment on later earnings, employment

or schooling.

These studies mostly utilize two basic schemes: either the OLS with hours worked
during high school as the dependent variable25 or the binary dependent variable (probit or
logit) framework with job-holding as the dependent variable. In the OLS method, the
following simple high school work experience equation is estimated.

H , = ,6l (ability) + ,62 (familyfc) + ,63 (labormkt) + ,64 (others) + u, (2-1)

where H , is a continuous measure of high school work experience, such as the

average working hours per week during high school; ability is test scores; familyfc is a

vector of family background variables, such as race, family income and parents’
education; labormkt is a vector of local labor market indicators, such as local
unemployment rate; and others is a vector of other factors which could possibly affect

high school employment.

With the same independent variables, the following binary dependent analysis is
also exploited,

JH,’ =XB+e

25

 

 

Ioehcidmg
“V1 J!" "1 ..
Lk‘v‘L‘JLuK ,‘A‘
lurk-cw

d‘;"‘..’\ 1" ‘
.>L‘A “H‘-

JH, =1 if JH,’ is larger than zero

= 0 otherwise

(2-2)

where JH,‘ is an underlying latent variable which indicates individual’s utility from job-

holding during high school; X is the vector of all the observable factors which can affect

job-holding utility such as test scores, family background variables, labor market

conditions; e is a continuously distributed variable independent of X ; and the

distribution of e is symmetric about zero. Depending the assumptions regarding the

distribution, the probit or logit model is estimated.

In our review of previous studies, we pay attention to how authors measure high

school employment (or working hours) differently, and whether they include labor market

conditions as control variables, and whether the relationships between high school

employment and ability measures, or high school employment and family background,

are specified as linear or non-linear. Why these factors are important is explained in the

course of this study. These studies are summarized in table 1-2-1.

Table 1-2-1: Determinants of High School Employment.

 

Author and Source

Data and Method

Findings

 

 

 

Marsh (1991) HS&B Sophomore data: Hours worked is modestly related to
The correlation between background background variables such as family socio-
variables (and/or ability variables) and economic status. Negative and insignificant
hours worked. relationship between hours worked and test
OLS for the average hours worked during scores.
the last three years of high school.

Ruhm (I997) NLSY. High local unemployment rate is negatively

 

OLS for the hours worked during the senior
year and Probit model for job-holding
during the senior year.

 

related to job-holdings. Standardized test
scores is positively but insignificantly related
to job-holdings. Blacks have lower probability
to hold jobs. Family income is positively
related tojob holdings.

 

 

25 Tobit model instead of OLS is also used (See Stern et al 1997)

26

 

34:13. N- -

Sm: wt

 

«1' ,k
Ll“ FIG “:5.

N)“. "- 9—
lit um Eu“.

 

i

”OHS ll 0,

to mix

‘v
xi
5

.m- ’.
n‘ts‘t‘k‘l \e

Wilt;

 

 

 

Dustman, Rajaj and NCDS from England. Regional unemployment rate, attending
Ssoest (1996) Hours worked equation is estimated as a private school, and being handicapped
part of a three equation structural model. significantly negative effects on hours worked.
Two other equations are: school success The number of younger siblings, a working
equation and school leaving decision mother, and ability measures have positive
equation. effects on hours worked.
Stern, Finkelstein, NCRVE data. Blacks and Hispanics are less likely to be
Urquiola and Logit forjob-holdings during the senior employed and tend to work less hours.
Cagampang (1997) year, and Tobit for hours worked are Students whose parents are high school
estimated. graduates tend to work more.

 

 

 

 

Using the HS&B sophomore data, Marsh (1991) explores the effects of
background variables (such as socio-economic status, race, sex, attending public schools,
and mother working) and sophomore outcomes (such as test score, grades, home work,
honors and so on) on average hours worked during the years in high school, and on the
hours worked in each year separately. He concludes that working is only modestly related
to background variables, especially to family socio-economic status. Also, he finds a
negative relationship between sophomore outcomes (especially grade and test scores) and
working hours,26 which could imply either that those who work more during high school
are less qualified,27 or that ‘working’ affects high school students’ performance
negatively.

However, his research has several weaknesses and can be criticized on several
grounds. First of all, Marsh basically relies on correlation analysis instead of regression
analyses such as the OLS or the binary dependent variable method.28 Thus, he assumes
only linear relationships between hours worked during high school and socio-economic

status, and academic ability. However, there is a high possibility that those relations could

 

2‘ However, using the same data set, Crawford( 1997) finds a strong positive relationship between test score
and working hours. The best guess is that it is because Crawford uses working hours in the senior year as
the dependent variable while Marsh(l99l) uses the averaged hours in three grades.

27 Also, he finds negative effects of total hour worked (or sophomore and junior working hours) on senior
outcomes. However, he finds that there is no such negative effects of working in senior year.

27

ix FLOR-k.

”ht tart
mJLn “JD
5

. p, l ..
i0? 62le i;

 

. .', y
mg . . -
and!“ “In":
I s
. . ~ .
{remit} are
. ‘ I“‘ .
>Ptklilka‘l‘n

this are

1):“ wt}

d1“; JO .
tilinp‘l‘k

be non-linear. Students from lower socio-economic status or with lower academic ability
might tend to work less possibly because they have less access to jobs or lower demands
for their labor. Also, students from higher socio-economic status or with high academic
ability might tend to work less possibly because they have less need to work because their
parents are well-off or they want to concentrate on academic activity. Thus, nonlinear
specifications might be crucial in identifying how socio-economic status and/or academic
ability are related to high school employment.

Second, Marsh’s measure of high school employment does not capture accurately
the intensity of work efforts by high school students. His ‘total hours worked’ variable is
the weighted mean of hours worked during the last three years of high school.29 Also, the
distinction between working in the senior year and working in other grades seems to be
important because most students finish their college entrance preparations in their junior
year or early in their senior year, completing standardized tests, etc, SAT tests or
finalizing decisions regarding whether to go to college or not. Therefore, more seniors
seem to participate in the labor force and work more intensively than juniors. Also,
juniors seem to work rather sporadically compared to seniors, evidenced by the fact that

many more seniors work than juniors during a given week (Ruhm 1997) even though the

 

28 Even though Marsh presents simple correlation and standardized beta weights in detail, he does not report
the OLS regression results in detail, only presenting the percentage of explained variance in hour worked by
background variables, sophomore outcomes and hours worked in lower grades.

29 First, Marsh gives weights to hours worked at the most recent job during senior year according to when
they last worked: 1.0 if worked last week; .667 if worked within the past month; .5 if worked within the
past three months; .1 if worked since the start of school. For junior and sophomore year, students indicated
the average number of hours worked during the period. Then, he averages the number of hours over three
years. However, the hours worked during the junior and sophomore year do not represent the averaged
hours worked during the whole year. They indicate the hours worked during the period student held jobs.
Thus, he gives too much weight to the junior and sophomore year. Ruhm(1997) shows that the hours
worked during the junior is much fewer than senior year on the average, while the HS&B Table 3-4 shows
that junior year students have worked once almost as much as senior year students.

28

, rants;

 

hush;
Tab}: H 5.1
later cm":
mi \\ jg: ‘,
pricier a‘
M;

and is...

 

Ni‘ipa‘,;0
HS&Bin
(Wfitp
are ke§ c'
differen

SIUJents

dag he
and ed“
Ofempl,
eqUatiOn
This. he
Wu [Z

_‘..'3 Marsh “50

percentage of students who have ever worked through their whole junior year is almost as
high as the percentage of those who have ever worked in their whole senior year (See
Table HS&B 3-4). Also, considering that many studies which find the positive effects on
later earnings are mostly based on senior year working experience (Ruhm 1997, Meyer
and Wise 1982, Stern and Nakata 1989), employment in the senior year needs to receive
particular attention.30

Marsh only controls for a very few background variables such as sex, race (Black
and Hispanics), socio-economic status (based on both parents’ education, father’s
occupation and family income). He does not utilize the labor market indicator file in
HS&B in identifying the determinants of high school employment. As Freeman and Wise
(1982) point out, the levels of aggregate economic activity or local economic well-being
are key determinants of youth employment. Further, while there seem to be clear
differences in the determinants across gender, Marsh does not separate male and female
students.

In many ways, Ruhm (1997) overcomes the flaws of Marsh’s study. Using NLSY
data, he explores the effects of high school employment on later earnings, employment
and educational attainment. In his investigation, he attempts to identify the determinants
of employment (job-holdings and hours worked) to control for selection bias in the wage
equations by using the treatment effect and instrument variables method (IV method).
Thus, he estimates an OLS equation (2-1) for working hours and a probit model for job-
holding (2-2). He concentrates on job-holdings and hours worked during senior year

instead of the average hours worked and also includes the mother’s and father’s education

 

3° Marsh also investigates the determinants of hours worked in three grades(sophomore, junior, and senior).

29

 

L‘ C3183?
- ’ ‘ ~
i312) t0 U .
I V ' l'"
harm. .

At.

 

 

tobho‘tiité
for Blah
Fm; me.
We IAI-
Q‘Ll’i'ii \an;
H,
WHO
WI IS tr
36mm \
Lllt’il' jUI
COHUQI‘

din“ 'V

I,”

as categorical variables. Also, he includes dummies for areas with high unemployment
rates to control for local labor market conditions and also considers other extensive
background variables.

According to Ruhm’s study (he reports only the results of the probit model for
job-holding in the senior year), senior year employment probabilities are relatively low
for Blacks and those living in areas with high unemployment rates (more than 12%).
Family income has a positive effect on job-holdings. The sign of one standardized test
score (AF QT - Armed Forces Qualification Test) is positive but insignificant, and few
other variables have significant effects.

However, Ruhm does not control for working in the junior year as a determinant
of job-holding during the senior year. Controlling for the work experience in the junior
year is important especially when we want to see the selectivity of the working during the
senior year in terms of academic ability. It is because, if those students who work during
their junior year have lower academic ability and more accesses to jobs, without
controlling for the work experience during the junior year, the selectivity of working
during the senior year in terms of academic ability might not be capture.

Also, unemployment rates are the only local labor market indicator and ‘local’ is
not specified. Above all, any non-linear forms of test scores specifications are not
attempted. Thus, Ruhm fails to capture the plausible non-linear relationship of test scores
and high school employment.

Using the National Child Development Study (N CDS) data from England,

Dustmann, Rajaj and Ssoest (1996) analyze the determinants of working hours, school

 

separately but unfortunately does not report any detailed results, mentioning that he has the similar results.

30

h OF“. '

success and school leaving decision simultaneously, even allowing error terms in three
equations to be correlated. Thus, the equation (2-1) is part of a three equation structural
model. For both male and female students, regional unemployment rate for school
leavers, attending private schools, and being handicapped have significantly negative
effects on working hours. The number of younger siblings, whether the mother works,
and ability measures all have a positive effect. Also, male students work more if the
father’s occupation is farmer. Students work less if parents encourage them to go college.
Female students work less if their parents have higher skills. Some of these findings are
consistent with our findings which we describe later even though the results are based on
British data. However, Dustrnann et a1 (1996) do not distinguish working experience by
grades and have not attempted any non-linear forms of ability measures in the
specifications of working hours.

Using data by the National Center for Research in Vocational Education
(NCRVE), Stern et al (1997) also briefly examine the determinants of job-holdings and
hours worked in their investigation of the effects of high school employment, especially
Co-operative education, which ties work to school as part of vocational education, on
later earnings. Logit model for job-holdings in the senior year, and, instead of OLS, Tobit
model for hours worked per week in the senior year are exploited.31 Only a few
independent variables, such as sex, ethnicity variable (Black or Hispanics), parents’
income, ‘both parents’ variable which indicates whether a student has both mother and
father, parents’ education, essay score are included. Stern et al (1997) find that Blacks

and Hispanics are less likely to be employed and tend to work less hours. Students whose

 

3 ' Because some students do not work at all, they assume that the distribution of hours worked is truncated.

3l

 

rm—
1d,,

dot:
5L;

9.1,,
llu

GIOUj

the hi

’ESCKJI

lest. T

parents are high school graduates tend to work more hours. However, this study is so
brief that we can not identify the full determinants of high school employment.

Our review shows that previous studies on the determinants of high school
employment are very brief compared to its complexity, and the investigation of its
determinants were a secondary concern even in these studies. However, as we see in the
review of studies on its effects, without understanding the characteristics or determinants
of high school employment itself, it is very hard to identify the effects correctly. In the

next section, we describe our data, the characteristics of high school employment, and our

econometric models.

3. Data, Characteristics of High School Employment, and Strategy
3.1 Data.

The current study uses two different data sets: NLS-72 and the HS&B Senior
Group data. NLS-72 is a longitudinal survey that originally sampled 22,652 seniors from
the high school class of 1972. It contains extensive family background variables, as well
as a standardized test score (N LS test score). We restrict the sample to those who
responded to the first and second follow-up surveys (1973, July 1974), and took the NLS
test. The sample sizes are 6,763 for males and 7,013 for females respectively.

HS&B is also a longitudinal survey that originally sampled 28,000 seniors and
over 30,000 sophomores from 1,100 high schools in 1980. HS&B contains almost the
same variables as NLS-72. HS&B has better features for studying the determinants of
high school employment because it contains a labor market indicator file, which shows

economic activities of areas such as county, SMSA (Standard Metropolitan Statistical

32

 

I’-

V
i" '
\UF

Der

Area), and states. Also, it contains the information on working experience in the junior
year. We use the senior sample and restrict it to those who responded to the first and
second follow-ups and took the NLS test. The sample sizes are 4,063 for males and 4,966
for females.

With these two data sets, we are able to study differences over time in these two
senior groups. Also, with the HS&B sample, we identify the determinants of high school
employment of almost the same group as the NLSY sample, which includes those who
were 14-19 years old in 1979, and which has been extensively used to study the effects of
high school employment on later employment and earnings. Thus, the current study might
help us to understand why some studies find positive effects of high school employment
of this group on later earnings, or why the results of those studies on the effects of high
school experience of this same group on later earnings differs. The sample sizes we use
here with these two data sets are around 10,000, which is much larger than the NLSY
samples of about 1,000.

In NLS-72, information on high school employment was obtained from responses
to the question “on the average over the school year, how many hours per week do you
work in a paid or unpaid job?” Then, hours worked were recorded as a categorical
variable. In HS&B, students reported how many hours they worked in most recent paid
jobs and the hours worked recorded as a categorical variable in their senior year as well as
during their junior year. Using this information, ‘work,’ ‘heavy work,’ ‘moderate work,’
and ‘working hours’ variables are created. ‘Work’ represents those who have ever worked
during their senior year, ‘heavy work’ represents those who worked more than 15 hours

per week for the NLS-72 samples, and 15 hours or more for HS&B samples, ‘moderate

33

 

 

Thurs-e

 

work’ represent those who work 6-20 hours per week for the NLS-72 sample and 5-21
hours for the HS&B sample. ‘Working hours’ variable is the midpoint of the categorical
variable of hours worked.32 Using the information on work experience in the junior year
in the HS&B senior data, ‘work junior’ and ‘working hours junior’ are also created by the
same pattern.

The reasons for investigating the determinants of ‘heavy work’ or ‘moderate
work’ in addition to those of job-holding are as follows. First, since academic activities
such as homework, and preparation for classes require much of students’ time and efforts
during high school, working long hours may not be compatible with such academic
duties. Thus, the characteristics of students who work heavily during high school might
be different from others. Second, since some studies indicate that students who work
moderately have the highest benefits from high school employment, it might be
interesting to see
whether those students have distinctive characteristics from others. For example, the
quadratic form of hours worked per week in the equations of later earnings as in Ruhm
(1997) implies the highest beneficial effects for the moderate work. Also, many studies
which investigate the effects of high school employment on educational aspects indicate
such beneficial effects as higher educational level (D’Amico 1984, Turner 1994), high
academic achievement (Lillydahl 1990), higher future school enrollment rate (Steel
1991)

Except for the information on high school employment, both of these data sets

also contain extensive background variables, information on high schools, high school

 

’2 For example, when a student answers the average working hours per week is between 21 and 25, the

34

 

achexemc
miles L
iron. the:
813615 in,
Because 1.1
others. st.
MCI-fry;
Charles.
Stilton.

Table 1.3

achievement, and a standardized test score. Means and standard deviations of the

variables used in our regression analyses are tabulated in the following Table 1-3-1.

From the table, we can tell that in the HS&B sample, the minority groups, especially

Blacks and Hispanics are over-represented, which is due to the original sample design.

Because these minority groups, on the average, have lower socio-economic status than

others, students whose family socio—economic backgrounds are in the bottom quartile are

over-represented.33 Before we explain our regression analysis methods, we summarize the

characteristics of high school employment from these two data sets in the following sub-

 

 

section.
Table 1.3—1 (NLS-72 and HS&B)
Variable Definition Mean and Standard Mean and Standard
Name Deviation : NLS-72 Deviation: HS&B
-_._~ -.. Male Female Male Female ..
(6763) (7013) (4064) (4966)
WORK 0-1 dummy variable that equals to one if .82 .70 .72 .67
student work in senior (.39) (.46) (.45) (.47)
WORKING Number of working hours during high 14.8 10.4 14.4 1 1.4
HOURS school senior year (11.05) ( 10.01) (12.4) (11.2)
GRADE Average point of grade during high school. 2.03 2.43 2.75 2.97
(0-4 scale) (1.02) (.99) (.75) (.69)
TEST SCORE NLS test score 50.9 50.2 49.6 48.3
(8.67) (8.65) (9.28) (8.67)
SES- 0-1 dummy variable that equals one if .48 .48 .45 .43
MEDIUM student’s $133 is in 2"d and 3rd quartile (.50) (.50) (.50) (.49)
SES-HIGH 0-1 dummy variable that equals one if .26 .22 .21 .18
student’s SES is within 4‘“ quartile (.44) (.41) (.41) (.38)
BLACK 0-1 dummy variable that equals one if .09 .13 .21 .25
student is black (.29) (.34) (.41) (.43)
HISPANICS 0-1 dummy variable that equals one if .04 .04 .25 .24
student is hispanic (.20) (.20) (.43) (.43)
ASIAN 0-1 dummy variable that equals one if .01 .01 .04 .03
student is asian (.11) (.10) (.18) (.17)
NATIVE 0-1 dummy variable that equals one if .01 .01 .02 .01
AMERICAN student is native american (.1 1) (.10) (.14) (.12)

 

‘working hours’ is 23.
33 Since it is highly plausible that high school work experience is related to socio-economic status or race,
this over-representation can cause some bias in our regression analysis. To solve this problem, we attempted
two methods. First, we run the same regressions only for Whites. Second, we use the weights given in the
data sets. Since the results with these methods are not much different, we do not report these results.

 

iEL30§

\l‘ W1 Rh

5173’ RE 1‘\

\CRTH

SlBLlNGS Number of siblings 2.14 2.25 3.29 3.32
(1.83) (1.84) (2.80) (2.56)
MOMWORK O-l dummy that equals to one if mother .58 .60 .70 .72
work when student in high school. (.49) (.49) (.46) (.45)
ANY 0-1 dummy variable that equals one if .10 .09 .004 .027
DEPENDENT student has any dependent (.31) (.28) (.064) (.16)
NO 0-1 dummy variable that equals to one if .09 .09 .12 .1 1
ENGLISH english is not spoken at home. (.29) (.28) (.32) (.32)
HIGH 0-1 dummy variable that equals one if .45 .40 .40 .40
SCHOOL student’s high school program is academic (.50) (.49) (.49) (.49)
HIGH 0-1 dummy variable that equals one if .20 .29 .23 .25
SCHOOL student’s high school program is vocational (.40) (.45) (.42) (.43)
SMALL . 0-1 dummy variable that equals one if .27 .28
TOWN“l student lives in small town during senior (.44) (.45)
URBAN 0-1 dummy variable that equals one if .25 .27 .26 .27
student lives in urban area during senior (.43) (.44) (.44) (.44)
SUBURBAN 0-1 dummy variable that equals one if .24 .22 .45 .45
student lives in suburban during senior (.43) (.41) (.50) (.50)
NORTH 0-1 dummy variable that equals one if .28 .26 .25 .24
CENTRAL student lives in north central during senior (.45) (.44) (.43) (.43)
SOUTH 0-1 dummy variable that equals one if .32 .33 .37 .38
student lives in south during senior year (.47) (.47) (.48) (.48)
WEST 0-1 dummy variable that equals one if .18 .17 .19 .19
student lives in west during senior year (.39) (.37) (.39) (.39)

 

Notes: "' In HS&B data, this dummy indicates whether the respondent has any children or not.
“ In HS&B data, most small town residents seem to be classified into suburban residents.

3.2. Characteristics of High School Employment.

From Table 1-3-2 and Table 1-3-3, 4, 5, 6and 7 we can characterize high school

employment in this study as follows: (1) the majority of senior students have worked at

least once during their senior year, though female students work much less than male

students. Table 1-3-2 shows that in 1972, 81.38% of male and 70.49% of female students
have worked either paid jobs or non-paid jobs during their senior year. Table 1-3-3 and 4
show that in 1980, 72.02% of male students and 66.53% of female students have worked
paid jobs during their senior year.

(2) Students work longer hours in 1980 than in 1972 if they work. 49.77 % of

male students and 33.32 % of female students in this study worked 16 hours or more

36

during their high school senior year inl972, while in 1980, 69 % of male students and
41.6% of female students worked 15 hours or more. (3) A high turn over rate is expected
among high school senior workers. From Table 1-3-5, only 51.3% of male students and
46.7 % of female students worked during a given week (a reference week) while 72.0%
of males and 66.5% of worked at least once a year.

(4) Many students work in lower grades also. 71.7 % of male students and 58.7%
of female students worked at least once during their junior year (see Table 1-3-4).
Considering the fact that senior students work much more than junior or sophomore
students during a given week (reference week) from the CPS data,34 such similar
employment probability of male students across two years can indicate a higher turn over
rate among juniors than seniors.

(5) From the Table 1-3-6 and 7 we see that jobs students hold during their senior
year are mostly low-paying, low-skilled, low quality and private service oriented, and
very casual jobs. Most students earned less than $4.00 per hour in 1980. Among workers,
12.3% of male students and 19% of female students worked as “waiter or waitress in a
restaurant or drive-in” and 18.4% of male students and 23.4% of female students work as
“store clerk or salesperson” and male students tend work more on the farm (6.36%) and
as “skilled trade” or “other manual labor” (13.7%), while female students work more on
“baby-sitting or child-care” (11.7%) and as “office or clerical” workers (18.4%). These
characteristics of senior jobs are consistent with Stern and Nakata’s findings with NLSY

(1989). They find that “high school students are over-represented in sales and clerical, as

 

3‘ Employment probability for senior students during the CPS reference week from 1968 to 1988 is between
35% and 45% while for junior students is between 24% and 35%. Average working hours per week among

37

iii! a in 10'
limit in
them to an

In 11

9‘." ‘ h '
Skhu‘l\aki\

well as in low-skilled labor, service, and farm jobs” (p197). In most cases, these jobs
provide almost no on—the-job training. From these job characteristics, it is hard to expect
them to accumulate a high quality of human capital.35

In the following sub-section, we explain the econometric methods and

specifications of high school employment.

 

 

 

 

 

 

 

 

 

Table 1-3-2: Workirfi Hours Distribution and Average Test Score (NLS-72)
Male Ave. Test Score Female Ave. Test Score

None 1245 (18.4%) 51.6 2126 (30.3%) 49.6

Less than 6 Hours 730 (10.8%) 52.7 920 (13.1%) 51.3

6 to 10 hours 797 (1 1.8%) 51.3 931 (13.3%) 50.0

11 to 15 hours 609 (9.%) 52.0 741 (10.6%) 51.8

16to 20 hours 915(13.5%) 51.5 900(12.8%) 51.4

21 to 25 hours 804 (11.9%) 51.1 609 (8.7%) 49.9

26 to 30 hours 607 (9.0%) 49.2 353 (5.0%) 48.5

More than 30 hours 1056 (15.6%) 48.0 433 (6.2%) 47.9

Total 6763 50.9 7013 50.2

Table 1-3-3: Working Hours Distribution and Average Test Score :During Senior Year (HS&B)

- Male Avc-lsaésprcwwlismals Ave- Test Score

None 1 137(27.98%) 48.74 1662(33.47%) 46.47

1-4 hours 240( 5.91%) 48.21 444( 8.94%) 48.45

5-14 hours 627( 15.43%) 51.23 793(15.97%) 49.90

15-21 hours 807(19.86%) 51.04 1077(21.69%) 49.93

22-29 hours 567( 13.95%) 51.62 561(1 1.30%) 49.64

30-34 hours 282( 6.94%) 48.64 206( 4.15%) 47.73

More than 30 hours 404( 9.94%) 47.15 223( 4.49%) 46.10

Total 4064 49.647(9.273) 4966 48.338(8.674)

Table 3-4 :Working Hours Distribution and Average Test Score During Junior Year. (HS&B)
Male Ave. Test Score Female Ave. Test Score

None 1 1 14(27.66%) 51.21 2017(40.91%) 47.81

1-4 hours 323( 8.02%) 50.96 487( 9.88%) 49.43

5-14 hours 605(15.02%) 50.34 734( 14.89%) 49.91

15-21 hours 781(19.39%) 50.00 847(17.18%) 49.33

22-29 hours 515(12.79°/o) 49.55 412(8.36%) 48.30

30-34 hours 268( 6.66%) 46.77 192(3.89) 46.22

More than 30 hours 421(10.45%) 45.35 24l(4.89) 44.84

Missing 37 43.58 36 42.86

Total 4064 49.647(9.273 4966

 

Note: the percentage is calculated among non-missing observations.

 

working high school seniors are between 17 and 20 hours while for juniors they are between 14 and 17
hours during the same period.(See Ruhm 1997)
35 However, around 80 % of them say “job encourages good work habits” 1n the HS&B senior group data.

38

 

 

 

7:313”, “_i'.
l is .-

_—.———-—'

" ‘ ,. \ ,.
“if" v.1; l in
l

1101 3):.“ l3:

“0713.10: ‘. -:
511‘.“
\1 11:71 31”.: '

in. ".25: 31;.
“in 5:20!)

Y“ 1‘ \ "’"‘»f

bl... .g

Utter \11.‘
”71C or L
(in-(rs

s

l-Q

Table 1-3-5: Working Experience During Junior and Senior years and Ave. Test Score

(HS&B)

 

 

 

 

.2... Male Female -.--
Number(%) Ave. Test Score (Std.) Number(%) Ave. Test Score (Std.)

Work Senior Year 600 (14.8%) 50.98(9.21 1) 1061(21 .4%) 48.49(8.447)

(Will/k Both Years 2327 (57.3%) 49.75(8.943) 2243(45.2%) 49.654(8.569)

Work Junior Year 586 (14.4%) 46.67(9.580) 670(13.5%) 45.804(8.549)
Iggllvork Both Years 551(13.6%) 50.95(9.683) 992(20.0%) 46.91(8.652)

Work Last Week 2083(51.3%) 50.53(8.813) 2317(46.7%) 50.051(8.386)

Work Senior Year 2927(72.0%) 50.00(9.01 1) 3304(66.5%) 49.28(8.546)

Work Summer 3461(86.0%) 49.68(9.27) 3615(73.2%) 48.36(8.67)

Work Junior Year 2913(71.7%) 49.13(9.157) 2913(58.7%) 48.77(8.715)

 

 

 

 

 

 

 

39

Table 1-3-6: Job Characteristics. (HS&B)
-1196 of Work _. ._-..Mal¢...-., .. .....f_e_mal¢- .

Lawn Work or Odd Jobs 1 18(4.2°/o) l8(.6%)

Waiter or Waitress 3474(12.3%) 609( 19.0%)

Babysitting or Child Care l8(.6%) 375(1 1.72%)

Farm or Agricultural Work l80(6.4%) 27(.8%)

Factory Work, Unskilled or Semi-skilled l34(4.7%) 64(2.0%)

Skilled-trade 283( 10.0%) 31(.97%)

Other Manual Work 389(13.7%) 51(1.6%)

Store Clerk or Salesperson 521(18.4%) 749(23.41%)

Office or Clerical 86(3.0%) 588(18.4%)

Hospital or Health 72(2.5%) 186(5.8%)

Others 678(23.9S%) 497( 15.53%)
Total 2826 3195

Table 1-3-7: Earnings ($1980) (HS&B)

Hourly Wage on the Job Male Female

Less than $1.50 37(1.3%) 221(6.8%)

$1.50 to $1.99 40(1.4%) 141(4.3%)

$2.00 to $2.49 97(3.4%) 185(5.7%)

$2.50 to $2.89 142(4.9%) 197(6.0%)

$2.90 to $3.09 484(16.7%) 497(15.2%)

$3.10 to $3.49 1309(45.2%) 1527(46.8%)

$3.50 to $3.99 341(11.8%) 278(8.5%)

$4.00 per hour or more 445(15.4%) 218(6.7%)

2895 3264

 

 

 

3.3. Strateg

10 i
tailed. i0
moderate 1
m not.

in the sec:

}..

1‘ .

[111 5,11 \Ob
*1..l

Maul}

hodrs t0

Th? mg

“here 1

“oiling

“here (I

3.3. Strategy

To identify the determinants of high school employment, two methods are
utilized. For the choices of ‘work,’ for the choice of ‘heavy work,’ and for the choice of
‘moderate work,’ we run a binary dependent variable model (probit model). This is the

same model for the job-holdings used in the previous studies (Ruhm 1997), and described
in the section 2. It is rewritten here for convenience.

Y,’ = XB + e

Y, = 1 if Y,’ is larger than zero

= 0 otherwise. (3-1)

For the investigation of the determinants of working hours, we utilize the two-
limit tobit models instead of the OLS method because the working hours variables are
actually truncated at the longest working hours recorded (30 hours for NLS-72 and 35
hours for HS&B), and also, we can regard them to be truncated at zero working hours.
The model for working hours then becomes

Y,' = X ,B + v,
where Y,’ is the latent variable, for example, utility from working hours. We denote the
working hours observed by Y, ,

Y, = 0 if Y,’ 50

= Y,’ if OSY,’ sUL
= UL if Y,' SUI. (3-2)
where UL is the upper limit, 30 hours for NLS-72 and 35 hours for HS&B. When the

dependent variables are truncated, the estimates of parameters from the usual OLS

4o

 

 

(IA.

.

1,.

sli
~

(I)

regression are inconsistent. Thus, under the assumption that the error term v, is normally

distributed with zero mean, we can consistently estimate the parameters with a tobit
model with two limits36 (see Maddala 1983).

In these regression analyses, males and females are separated because there are
big gender differences in high school employment in terms of both job-holdings or
working hours as we see in Table 1-3-2 and Table 1-3-3. As mentioned earlier, factors
that indicate job availability, ability and family socio-economic background are focused
on.

For job availability, we pay attention to local economic conditions, community
types, and race. Freeman and Wise (1982) summarize the youth labor market problems:

( 1) youth employment is more sensitive to cyclical movements; (2) the unemployment
rate is high among those who live in areas with high poverty levels; (3) the growth rate of
personal income is positively related to youth employment. Thus, we can expect youth
employment to be strongly related to levels of economic activity and the economic well-
being of the areas they live in. Even though their summaries are related to the general
youth employment including both young people in school and out of school, influences of
local market conditions on high school employment would be very similar and we can
expect lower job-holdings and fewer working hours among high school students who live
in areas with lower economic activities (high local unemployment, lower income growth
rates) and lower per capita income.

In HS&B, the labor market file provides the information on local (county)

economic conditions such as the unemployment rate, employment growth rate, average

 

3" We also utilize a tobit model with one limit. However, the results are almost the same.

41

'n“..

T’:J

 

psicapil;

iii-l" il‘

L...“ ill

m- l e ,
ultil'hsl \,

ix: ,
‘Jrr‘fi
\“l‘\l\ll\

cindiiiii
315011115
1.16165:
, '3 {(113,
{0: {1‘9
Curio:
liitr‘i‘.
iii-i
’0 ace
14113
am

SCllo

Stud:-

per capita income, its growth rate in the county, the percentage of people employed
within the county, and the mean wage in the manufacturing industries in the area.37

However, these county-level economic indicators might not represent all the
market conditions. The labor market conditions each individual student faces might be
different even within a county. A reservation wage could indicate more specific market
conditions each individual student faces. Usually, when the reservation wage is high and
all other conditions are the same, the probability of being employed is lower. However, if
the reservation wage and employment are influenced by the level of economic activity in
the county, the reservation wage will be higher in the areas where there is high demand
for high school employment, and could represent the other market conditions which
cannot be captured in these county economic indicators. Thus, we test whether the
reservation (minimum) wages during high school affect high school employment. In the
HS&B survey, Students are asked “What is the lowest hourly wage you would be willing
to accept for a job while still in high school” and answered the question as a categorical
variable (from below $1.5 to $4.00 or more by every $5). We make a dummy variable for
above $3 categories and a dummy for those who will not work at any wage during high
school, and thus, have no reservation wage.

Means and standard deviations of these variables are tabulated in the Table 138,

the standard deviations of the economic indicator variables are large. For instance, the

average unemployment rate is 7.5% with standard deviation 2.64 across counties where

students live.

 

37 Unfortunately, NLS-72 do not provide the information on the local market conditions.

42

 

l ’ 0’

1.1 L

r

m" '- m e \
L.-.1,.‘ \

aiailabij,
"568103

313‘: “hi

. FOf [hf 1":
:10ng Der 1
3851;". -

Table 1-3-8.: Summary Statistics of Local Economic Indicators (HSs&B)

 

 

Variable Name Definition Male Female
U Rate county unemployment rate 7.5 7.6
(2.64) (2.64)
Employment Growth county employment growth rate 1.00 1.00
(3.74) (3.85)
Income Growth county income growth rate 10.5 10.5
(4.22) (4.22)
County Income county log average income 9.1 1 9.1 1
(.24) (.24)
% Employed percent of people employed in county 42.7 42.8
(5.9) (6.0)
Mean Wage Mean wage per hour in manufacturing industries in SMSA 8.21 8.17
(1.54) (1.49)
Reservation Wage 0-1 dummy which is equal to zero if student’s reservation is .77 .67
larger than $3.00 (.42) (.47)
Will Not Work 0-1 dummy which is equal to zero if student will not work at .02 .02
any wage (.12) (.13)

 

Community type, such as rural, urban or suburban areas can indicate job
availability for students. Considering the fact that most students work in the service sector
(see Table 1-3-6), we can expect more jobs available for students who live in urban and
suburban areas than for those who live rural areas.

In terms of job-holdings, racial minorities work much less.38 From the second
column of the Table 1-3-10, in 1980 59.0% non-white and 70.7% white students have
worked at least once during their junior year. The difference is 11.7%. This racial gap is
bigger during their senior year (14.8% difference) and is even bigger for the reference
week (18.9% difference). As we see from Table 1-3-11 and 12, these gaps are largest for
Blacks. Black males work even less during their senior year than their junior year. Also,
in 1972, Blacks’ job-holding rate in their senior year is much lower than other races even

though the difference is much smaller than in 1980 (see Table 1-3-9).

 

3' For those who are employed, racial differences in working hours are much smaller. The average working
hours per week for the employed Whites, Blacks, Hispanics and Asians are 19, 18, 19, and 17 in the HS&B
sample and 17, 15, 17 and 16 hours in the NLS-72 sample respectively.

43

 

llOlllLlC

iii to c

1
OUT? 1111

Whether race itself indicates job availability is very controversial. If a student

from a certain race works less, they might do so simply because they want to work less

due to cultural differences. However, there also might exist less demand for Blacks and

other minorities either because they have less ability or because employers have a racial

bias against them. Minority groups have higher unemployment rates even though their

labor force participation rates are much lower. They might even be discouraged to join in

the labor force because there is not much demand for them (See Table 1-3-11 and 12).

 

 

 

 

Table 1-3-9 (NLS-72)
Whites Black Hispanics Asians

Mal?“ 83.0 71.1 788ww ~78‘lfin

Female 72.3 57.5 59.7 57.8

Table 1-3-10: Race Difference of Working Experience (HS&B)

 

Work in Senior

 

Work in Ju919r______W9r19..1:%$t_\1/ssk_ _ -

 

 

 

 

 

 

 

White 76.8 70.7 58.7

Minority 62.0 59.0 39.8

Table 1-3-11: Race Differences in Work Experience (%) Male (HS&B)

Race Labor Force Part Unemployment Work in Senior Work in Junior Work Last Week
Rate

White

(1942) 83.8 12.9 78.1 75.28 59.3

Black

(867) 75.7 26.9 63.3 68.51 41.0

Hispanic

(1022) 78.6 16.5 71.0 70.16 47.4

Asian

(143) 69.2 16.8 51.3 55.9 37.8

Table 1-3-12: Race Differences in Work Experience (%) Female (HS&B)

Race Labor Force Part Unemployment Work in Senior Work in Junior Work Last Week
Rate .._.- _ _ _.......- .. - ._--._- ~..._...-. 5.--“... __-._ -._~_,.,..-_.-- _.,.___ .. . _ .,_._

White

(2334) 81.7 12.0 75.7 66.8 58.2

Black

(1223) 70.6 24.8 54.7 50.1 33.7

Hispanic

(1180) 72.3 16.9 61.9 50.4 38.9

Asian

(152) 67.8 11.2 61.8 60.5 40.1

 

44

 

11'
icons: 3.1

{$330333

‘ inn,
L0LL.ul:
5

picture, 1

afasurc
mm

Wit?)

SGM)

19861.

Cases n

With respect to the relationship between the ability measure (standardized test
scores) and high school work experience, we focus on the following two points: (i) the
relationship between different working experiences and academic ability, especially
differences between working in the junior year and working in the senior year. (ii) non-
linearity of the relationship of test scores to job-holdings and working hours.

Students took the same test in NLS-72 and HS&B. To measure different types of
cognitive skills, the test contain six different categories: vocabulary, reading, math,
picture, formula and mosaic area. We use the average test score in the first three areas.

As we see in the review of studies on the effect of high school work experience,
the measures or definitions of it are quite different across studies. Those different
measures and/or specifications of the work experience in the earnings (or employment)
equations seem to derive different results across studies. We suspect that these different
measures and specifications might capture the relationship between high school
employment and academic ability differently.

The negative effects of high school employment on psychological and educational
aspects such as higher drop out rate, higher rates of delinquency, drug and alcohol abuse,
and so on, are mostly based on students’ working experience in the junior and/or the
sophomore year (Greenberger and Steinberg 1980, Marsh 1991, Mortimer and Finch
1986). However, the large positive effects on later earnings and employment are in many
cases more related to working experience in the senior year.

Table 1-3-5 shows that those who work only in their senior year have the highest
average test scores, while those who work only in their junior year have the lowest

average test score. Also the same table indicates that the average test score of those who

45

 

 

ii: “(iii
110116.31 11
senior it
smells

iii-iii: '
like.) 51‘
543351 .i
11316 11‘s
“allied

“Willi;

itfienr
model
”10161:

J

Sam:

are working during the reference week is slightly higher than that of those who have ever
worked in their senior year. All these facts together could imply that employment in the
senior year(especially, working in the reference week) is more selective toward better
students or better workers than in any other grades. Also, the 1-3-9 and Table 1-3-3
indicate that those students who work 6~20 hours have the highest average test scores.
Many studies show that the beneficial effects of high school work experience is the
largest among those who work moderate hours. However, students who works long hours
have the lowest average test scores. We suspect that the quadratic specification of hours
worked in the earnings equation as in Ruhm (1997) could capture this selectivity of these
working moderate hours instead of the true effects of working experience.

After controlling for other variables, to examine the linear relationship between
different work status and academic ability, the probit model for ‘work’, ‘heavy work’,
‘moderate work’ for both the 1972 and 1980 samples are run. Also we run the probit
models for ‘work’ during the reference and ‘work’ in the junior year for the 1980
samples are run. The test scores in these regressions are specified in the linear form.
Thus, only the first order of test scores is included.

Then, we concentrate on the relationship between senior year work experience and
academic ability. The tobit models for working hours as well as the probit models for job-
holdings are utilized. Including just test scores in independent variables in regression
analyses might not be enough to detect the selection bias in terms of ability. When
students have the same academic ability, those who have worked in their junior year
might have advantages in getting jobs, or in working longer hours over those who do not

have previous working. This is because they might have more information on the job

46

 

“'4 .4
dteui’mlt.
123381017
1 ’ i <1,
“with U.
S

1

1mm 0
a l, ‘
€331.01
1‘1 ‘v 1'5.
“Hum,“

il‘l.ll\ 3Q

011 \\' of};

of .“(lfl

ilfierm 3‘,

SO "44'. 11c
liming “0
5C01' 85, 11 9 8i

market and might have more access to jobs or because employers might value their
previous working experience.

Thus, to see the true selectivity of the high school senior employment in terms of
academic ability, controlling for previous working experience might be important in the
regression analyses. 39 A dummy variable ‘work junior’ which represents those who have
worked during their junior year or ‘working hours junior’ which indicates hours worked
during the junior year is included in the regressions.40 By excluding and including this
dummy or the hours worked variable in the regression analyses, we want to see how the
effect of ability measures on working in the senior year changes before and after
controlling for the previous working experience. Then, we can see whether there is any
ability selectivity of employment in the senior year. Unfortunately, we have information
on working in jtmior year only with the HS&B data.

We also examine the non-linearity of the relationship of test scores and ‘work’ or
or ‘working hours’ during the senior year. Previous studies assume only the linear
relationship between high school employment and academic ability and find almost no
significant relationship between these two variables. However, if the relationship is non-
linear, especially quadratic, a linear specification of test scores in the regressions of high
school employment might not capture their true relationship.

As briefly mentioned earlier in our review of the previous studies on the

determinants of high school employment, the work selection process during high school

 

39 So far, we assume that test scores represent each individual’s academic ability without being affected by
previous working experience. However, even if we assume that previous working experience affect test
scores, we also need to control for previous working experience. Without controlling for previous working
experience, negative or no relationship between working in senior and test scores could be the results from

47

is com;

prod.

3

aid

(9

the) ca
car's),

mics:

“Oil
lelal‘
Case
P05:

the

{/75 rel

11
4‘ Elf"
‘ l“ 151':

is complicated in terms of academic ability, because high school years are a transition
period, and because high school students have to spend at least 35 hours or more on
academic activities. Students with low test scores seem to work less probably because
they can not find jobs (even if they are less likely to go college, thus more like to work
early). Students with high test scores might tend to work probably because they want to
invest more human capital in academic areas. Also, from an employer’s perspective,
those jobs might not require much higher academic ability beyond a certain level. These
tendencies can cause a non-linearity effect of test scores on job-holdings or working
hours during high school.

Table 1-3-12 and Table 1-3-14 indicate that, except for males in 1972, mean
working hours by test score quartiles show a possible non-linear, probably quadratic
relationship between working hours and test scores.41 Also, except in the HS&B female
case, the percentages of students who have not held jobs by test score quartiles indicate a
possible non-linearity.42 Students whose test scores are in the 2nd and 3rd quartiles have
the highest job-holding rate. From Table 1-3-9 and Table 1-3-3, we can see another
aspect of this non-linearity. Students who work moderate hours (5-20 hours) have a
higher average test score than others. To capture a non-linear relationship, a quadratic
function of test scores or categories of test scores is included in the probit equations for

job-holdings and the tobit equations for hours worked.

 

the reverse causation because of strong correlation between working in senior year and working in junior

year.
4° We also include the working experience during the summer between the senior year and the junior year.
“ Even in 1972, the portion of students who have not worked at all implies this nonlinear relationship.

‘2 In this case, hours worked show the non-linearity.

48

 

\ne

16558.1
1:011
111015
111031
I} 1025
’1 11331.

More in

\me
1 4 1),
M11 1
15-31
1,1,,
311.3,;
Vim»:
110125
11628
Hear

\
\Olfi

C0111

Tablel- 3-13: Male Working Hours Distribution During Senior Year by Test Score Quartiles: (N LS-72)

 

 

 

 

 

 

 

 

 

male female
1 st 2nd 3rd 4th 1 st 2nd 3rd 4th
_,__.. W « ___-_..._9_uel:t_i!__. Quanflwflyartflm,QaauL.,_Quartil Quart“ Quart“ 011211111..-
None 18.6 15.5 16.9 22.4 35.2 29.1 26.8 29.8
Less than6 Hours 8.9 8.0 11.1 15.0 11.7 11.7 14.1 15.4
6 to 10 hours 10.7 11.7 12.0 12.6 14.3 12.3 12.4 14.0
11 to 15 hours 7.2 8.5 10.0 10.1 7.0 10.8 12.3 12.5
16 to 20 hours 11.3 14.1 14.6 14.0 9.4 13.8 14.9 13.5
21 to 25 hours 10.9 12.2 13.2 11.3 8.7 8.9 9.3 7.9
26 to 30 hours 10.3 11.0 9.0 5.8 5.6 6.4 4.7 3.2
More than 30 22.2 18.9 13.3 8.7 8.0 7.1 5.6 3.7
hours
Mean Working
Hours 16.1 16.4 14.7 12.1 10.1 11.2 10.8 9.5
Note: all units are % except the mean working hours.
Table 1-3-14: Working Hours Distribution During Senior Year by Test Score Quartiles (HS&B)
Male Female
l st 2nd 3rd 4th 1 st 2nd 3rd 4th
_ _ 9139411-....Quartjlmguanil Q1141?“ Qua?!“ Quart“ Quartil Quart“ ___.
None 33.6 25.2 25.5 26.14 42.5 33.7 27.0 25.4
1-4 hours 8.0 5.4 3.9 5.7 9.8 7.2 8.0 10.6
5-14 hours 12.7 14.7 16.3 18.4 13.1 15.6 15.7 21.2
15-21 hours 15.3 19.5 22.17 23.2 16.0 23.6 24.3 26.0
22-29 hours 10.5 16.3 16.0 14.4 8.0 11.7 16.4 10.8
30-34 hours 6.9 9.0 6.7 5.5 4.6 3.9 4.4 3.5
More than 30 13.0 10.2 9.6 6.7 6.0 4.3 4.2 2.6
hours
Mean Working
Hours 13.60 15.58 15.20 13.84 9.97 11.65 13.13 11.73

 

Note: all units are % except the mean working hours.

Family background factors could be important for working during high school

because they would affect students’ access to jobs though parents’ businesses or social

contacts, and/or affect students’ work ethics ( Rees and Gray 1982). However, as seen

earlier in our review, previous studies find that few family background factors

significantly influence high school employment (Marsh 1991, Ruhm 1997, Dustrnann et

al 1996). In our investigation of the influences of family background factors, we are

particularly interested in their possible non-linear relations to high school employment.

Marsh (1991) assumes only a linear relationship between hours worked and socio-

49

 

 

scorn
flail)
ample,
marlin:
to 28:1

11'1””
Nun

61100111

be" .
11\

19811 ;

'milc
lnClU1

4nd {1

economic status (SES). Even though other studies include categorical variables for some
family background factors in their regression analyses of the determinants of high school
employment, they do not investigate the possible non-linearity fully. However, as
mentioned earlier, there is a possibility that SES and high school employment are related
to each other in a non-linear pattern. Students whose SES is low, have less access to jobs.
Students whose SES is high do not need to work because their parents are well off and
encourage them to concentrate on academic activities.

Table 1—3-15 and Table 1-3-16 show a possible non-linearity in the relationship
between the mean working hours and the percentage of job-holdings except in case of
1980 female case (the first row in Table 1-3-16). Those whose SES is in the 2nd and 3lrd
quartiles work most in terms of job-holdings and working hours. In order to capture a
possible non-linearity of the relationship, we make 3 SES categories and includes the
‘middle’ and ‘top’ categories in the regression. For detailed family backgrounds, we
include categorical variables for family income, father’s education, mother’s education

and father’s occupation.

 

 

 

Table 1-3-152Working Hours Distribution During_Senior Year by SES (N LS-72)
Male Female
Bottom Middle Top Bottom Middle Top
None 19.6 16.5 20.7 34.9 28.0 29.2
Less than 6 Hours 9.1 9.6 14.6 11.3 16.0 16.0
6 to 10 hours 13.2 10.7 12.4 15.2 13.7 13.7
11 tolS hours 8.0 9.1 9.7 8.6 11.1 11.1
16 to 20 hours 13.3 13.7 13.4 9.9 14.4 14.4
21 to 25 hours 9.4 13.6 11.3 7.4 7.5 7.5
26 to 30 hours 9.4 9.6 7.5 5.2 3.6 3.6
More than 30 hours 18.0 17.3 10.4 7.4 4.5 4.5
Mean Working
Hours 14.9 15.8 12.9 9.8 11.1 9.7

 

Note: ‘Bottom’ represents the first quartile, ‘Middle’ represents the second and third quartiles, ‘Top’
represents the fourth quartile. All units are % except the mean working hours.

50

Table To

.———

iii

111 hours
i11 hour
l“: llO.‘
“'20 30;;
11.1-311101
110:: Titer.
.1121“. ll 1'
Haiti
K
Vote '8(

. ,,
'39’2531‘...

{9151683

'1‘ E2
*1: C
diffeie
m0}e
5851C 16
11011101:
0116i [it

higher pn

Table 1-3-16:Working Hours Distribution During Senior Year by SES (HS&B)

 

 

 

Male Female

Bottom Middle Top Bottom Middle Top
None 31.7 25.3 27.0 40.1 30.3 25.4
1-4 hours 7.1 5.0 5.8 8.9 8.4 10.5
5-14 hours 13.6 16.4 16.5 14.2 16.1 19.9
15-21 hours 17.6 20.1 23.3 20.3 22.9 22.3
22-29 hours 12.8 15.5 12.8 8.0 13.3 14.4
30-34 hours 6.6 7.0 7.5 3.7 4.6 4.2
More than 30 hours 10.6 10.7 7.1 5.0 4.5 3.3
Mean Working
Hours 13.7 15.2 14.1 10.2 12.3 12.3

 

Note: ‘Bottom’ represents the first quartile, ‘Middle’ represents the second and third quartiles, ‘Top’
represents the fourth quartile. All units are % except the mean working hours.

Other included independent variables are dummies for whether the mother works
for a paid job during the student’s high school years, whether student has his (or her) own
children (or dependents in 1972), and whether English is spoken at home; the number of
siblings; region categories and high school program categories, such as academic,
general, and vocational.

In the following section, we summarize and explain the major results in our

regression analyses.

4. Estimation Results.

4-1: Over-time comparisons: In this subsection, we summarize the similarities and
differences in the regression results of the probit models and tobit models between the
two years (see Table 1-4-1-1 and 2. From now on, we refer to these regressions as the
basic regressions). We can observe quite similar influences of background variables on
job-holdings and working hours in the senior year in both 1972 and 1980. First, after
other factors are controlled for, students whose SES are in the 2“d an 3rd quartile have

higher probability of holding a job and work more hours than those whose SES are either

51

 

especial
hours or
510113111:
program
male an

R18.

except
lloik IT

betucc

m6 11m
Wife
JOb-hol
CQelllC;
Slander.
HOHele
Iobir of,
mamas
academic

land )0 11 0

in the 1St or 4‘h quartile, except for the 1980 female case. Second, even after other
variables including test scores are controlled for, job-holding rates for minorities,
especially for Blacks, are much lower than for Whites, and minorities work much fewer
hours on the average. Third, male students who are in academic programs work less than
students in the general program, while both male and female students in vocational
programs work more than those in the general program. Fourth, if the mother works, both
male and female students work more in terms of job-holdings and working hours in both
years.

There are other similarities. The more siblings students have, the more they work
except in the case of males in 1980. Students who have any dependents or any children
work more except in the case of female in 1980. There seems to be no linear relationship
between GPA and high school work experience.

We can also observe the differences overtime. First, when we put the test scores in
the linear form, the relationship has dramatically changed over-time. In 1972, the
coefficient of test scores is negative and significant for males in the probit estimation for
job-holdings and is zero for females. In the tobit models for working hours, the
coefficients are negative and significant for both male and females. In the male case, one
standard deviation of test scores decreases more than a one and half working hour.
However, in 1980, the coefficients of test scores are positive and significant (except in the
tobit of working hours for male.) A increase of one standard deviation of test scores
increases almost one working hour. This change indicates that more students with higher
academic ability participate in the labor force. Second, students in the 4th quartile SES

tend to work more in terms of job-holdings and working hours in 1980 than 1972. The

52

 

IHESIS

at

5:1

coefficients of SES-HIGH in the tobit model for working hours are negative in 1972 even
though they are not significant. However, in 1980. they are positive for both male and
females. In the 1980 female case, female students in the SES-HIGH work 1.6 hours more
than those in the bottom SES quartile. More affluent female students work in 1980 than
1972. Third, the effects of the community type seem to have changed. Students who live
in urban and suburban areas work more in 1980 than 1972 compared to those students
who live in rural areas.

These findings might be explained by the secular rise of materialism among high
school students and increases in prices consumption goods for high school students, such
as toys or clothes (see Greenberge and Steinberger 1986). In 1972, students in poor
families might have worked due to their families’ economic demand or students with
lower academic ability might have worked during high school because they had decided
to go labor market directly after high school graduation. However, in 1982, students even
with high academic ability or students from affluent families might have begun to work
because they had to spend more money on their consumption. Given a certain demand for
high school student workers, higher participation of students from middle class families
or students with higher academic ability might have driven out those students in the
bottom in terms of academic ability or family backgrounds.

There are other noticeable changes. In 1980, students from homes where English
is not spoken work less in terms of both job-holding and hours worked, while there is no

such difference in 1972. Students who have dependents seem to work 3 hours more in

53

Table I-4-l-l: Probit For Job Holding .

(NLS-72 & Table HS&B)

 

 

 

Probit
Variables 72 Male 72 Female 80 Male 80 Female
,. WW 16763) 17013) (4064L 1:19.615)
CONSTANT 1.197‘” .148 .374" -.435"
(.154) (.135) (.186) (.170)
GRADE .022 .030 -.055 .032
(.022) (.020) (.035) (.033)
TEST SCORE -.007” -.000 .007“ .011‘“
(.003) (.003) (.003) (.003)
SES-MEDIUM‘ .137‘" .095" .098‘ .099"
(.048) (.040) (.051) (.044)
SES-HIGH .065 .036 .005 .162’"
(.058) (.052) (.066) (.062)
BLACK -.513"”'" -.374"" -.391"'" -.43l"‘"
(.067) (.056) (.064) (.055)
HlSPANlCS -.303"" -.325"" -.l30“ -.178"'”
(.099) (.085) (.065) (.060)
ASIAN -.250 -.413" -.518"‘" -.388”"‘
(.162) (.154) (.120) (.116)
NATIVE AMERICAN -.118 -.113 -.472"’” -.356"
(.107) (.163) (.146) (.156)
SIBLINGS .041‘“ .051‘" -.005 .022"
(.010) (.009) (.010) (.010)
MOMWORK .132*" .I 16"” .l48""‘ .131""
(.039) (.034) (.048) (.043)
ANY DEPENDENT .156" .137" .247 -.554""
(CHILD) (.065) (.060) (.346) (.115)
NO ENGLISH .126‘I .013 -.157“ -.219""
(.069) (.058) (.076) (.069)
HIGH SCHOOL ACADEMIC -.l66‘” .044 -.138"'" -.036
(.046) (.043) (.055) (.049)
HIGH SCHOOL .I43"" .138”"‘ .047 .164‘"
VOCATIONAL (.054) (.042) (.058) (.050)
SMALL TOWN -.204"" .069
(.055) (.046)
URBAN -.273‘” -.003 .035 .198“"
(.055) (.047) (.060) (.053)
SUBURBAN -.293"" .086" .078 .175‘"
(.057) (.051) (.053) (.047)
NORTH CENTRAL .083 .191 "" .042 .154‘
(.053) (.047) (.066) (.060)
SOUTH .085" -.071 .019 -.027
(.051) (.045) (.063) (.055)
WEST .1 19” .028 .045 .095
(.059) (.053) (.073) (.065)
R Square .0347 .0244 .0296 .0502
(1 square) (223.9) (210.4) (142.4) (317.5)

 

Notes: (a) Standard errors are in the parentheses. (b) ‘, ". 1'" means ‘significant’ at 10% level, at 5% , and 1% level.

54

 

" - 1
UN).
"7
1:51 ‘

we

HIS-Pl.

.151 z)

Table 1-4-1-2: Tobit For Working Hours

(NLS-72 & Table HS&B)

 

Variables 72 Male 72 Female 80 Male 80 Female
(9.7.6.3.). _ . ..<7Q.'3.)_-_-_,______ ._ 149911-..- 1426.9)-..
CONSTANT 22.762‘" 5802"" 12.893‘" -.372
(1.650) (1.552) (.2648) (2.224)
GRADE -.156 .100 -.637 -.095
(.235) (.232) (.496) (.437)
TEST SCORE -.184"" -.059“ .01 1 .097"
(.032) (.031) (.046) (.040)
SES-MEDIUM" 1648‘" .874‘ 1.676" 1.727'"
(.513) (.462) (.725) (.585)
SES-HIGH -.612 -.607 .002 1.607“I
(.628) (.598) (.934) (.791)
BLACK -8.365"" 6572”" -6.746”“ -5.558“"
(.771) (.665) (.918) (.740)
HlSPANlCS 4019“" 4797"" -2.1 1 I" 4.997”
(1.087) (1.021) (.909) (.782)
ASIAN -3. 169‘ -4.077" -9.383*” -5.249“‘”
(1.808) (1.877) (1.823) (1.541)
NATIVE AMERICAN -1.641 -2.766 -4.761'"' 4.997”
(1.846) (1.858) (2.195) (.782)
SIBLINGS 409"" .711‘"I -.131 360*"
(.110) (.103) (.152) (.131)
MOMWORK 1.298‘" 2248"" 2.852‘" 1798""
(.418) (.389) (.687) (.568)
ANY DEPENDENT 2842"" 2753*” 9.650" -6.420‘”
(CHILD) (.670) (.681) (4.745) (1.677)
NO ENGLISH .124 .713 -l.194 -2.349”
(.714) (.669) (1.117) (.946)
HIGH SCHOOL ACADEMIC -3.008"'” .176 3548“" -.838
(.494) (.494) (.779) (.638)
HIGH SCHOOL 2.608‘" 2557"“ 1.479“ 2543""
VOCATIONAL (.545) (.478) (.815) (.654)
SMALL TOWN -.798 1701""
(.553) (.524)
URBAN -.l46 1451"“ 1.510‘I 1418”"
(.573) (.538) (.851) (.701)
SUBURBAN -287 2.228’" 2.057"""'I 2996“"
(.590) (.571) (.742) (.617)
NORTH CENTRAL 1.306" 2073‘“ .564 1.690"
(.567) (.519) (.926) (.756)
SOUTH 2.370‘" -.475 1.202 -.l40
(.563) (.517) (.891) (.722)
WEST .618 -.627 .052 L422"
(.640) (.606) (1.025) (.830)
R Square .0113 .0078 .0076 .0087
(x squareL (475.6) (324.1) (190.6) (256.5)

 

Note: see notes in Table 1-4-1-1.

55

mFQIQ

1972. In 1980, male students who have children work 10 hours more than others while
female students who have any children work 6 hours less than others. This striking
contrast between males and females might be due to the welfare support system for the
single mothers in 19803.

From now on, we concentrate in more detail on factors that indicate job
availability, ability and family socio-economic background and only report the

coefficients of the related variables in each focus.

4-2: Job availability: (1) Local economic conditions: The effects of various local
economic conditions on job-holdings and working hours in the senior year, are tabulated
separately in Table 1-4-2-1, 2, 3, and 4. All other control variables in the ‘basic’
regressions are included in these regressions and the coefficients of them are not reported.
The unemployment rate is one of most important factors which determines high school
employment. The coefficients of the unemployment rate are significant and negative in
most cases. One percent increase of the county unemployment rate reduce .5 to 1%
probability of being employed at the mean values of control variables. (Marginal effects
are not reported here). Without controlling for other economic indicators, a one percent
increase of the county unemployment rate reduces a half hour working hour for males
(see column 1 in Table 1-4-2-2) and one third working hour for females (see column 1 in
Table 1-4-2-4).

Other indicators which represent the level of local aggregate economic activity are
the percentage of employed people within a county, employment growth rate, and per
capita income growth rate. The coefficients of these variables are in many cases positive

and significant in the regressions of job-holdings and working hours for both male and

56

 

(‘2

CE

0-

female students (column 7 in Table 1-4-2-1, columns 2, 6, and 7 in Table 1-4-2-2,
columns 3 and 4 in Table 1-4-2-3, and columns 3, 4, 6, 7, 9 in Table 1-4-2-4 ).

The level of affluence of the county a student lives in is very important for
females. The coefficients of log per capita income are significantly positive in all of the
specifications for females except columns 9 in Tables 1-4-2-3 and 4. One unit increase of
logged per-capita income increases work time by 4-5 hours for female students. However,
for male students, its effects are relatively small when unemployment rate is controlled
for.

These tables seem to show some gender differences in influences of these various
economic conditions on high school senior employment. When these variables are
included all together in the regression, unemployment rates and growth rates of
employment in a county, or the percentage of the employed within a county seem to be
more important in male students’ employment while per capita income and its growth
rates seem to be more important in female students’ employment (see columns 7 in
Tables 1-4-2-1,2,3, and, 4). Because of the correlation between these variables, we can
not argue that the gender differences are obvious. However, the gender difference in
characteristics of jobs students hold (Table 1-3-6) might be related to this phenomenon
because female students’ jobs are more strongly related to consumption rather than to
production.43 When income rises, people spend more money on private services and this
causes higher demand for student (especially, the female student) labor in the service

SCCIOI'.

 

’3 Female students’ jobs are a little more concentrated in baby-sitting, waitress and sales-person.

57

However, as mentioned earlier, these county economic indicators might not
capture the market conditions each individual faces. When the reservation wage is
included as an independent variable, the regression results display a highly positive
relationship between employment and reservation wage among high school students. The
results are shown in columns (9) in all the Tables 1-4-1,2,3, and, 4. Students with higher
than a $3.00 reservation wage work about 5 hours more for females and 7 hours more for
males than those who have lower reservation wage. Also the effect of this high
reservation wage is to increase the probability of being employed by 8-12% at the mean
values of the control variables. (not reported.) Considering the high labor force
participation among high school students, we can conclude that those who are willing to
work with a lower reservation wage simply can not find jobs or employers who want to
hire them. When the reservation wage is controlled for, almost all other labor market
variables become insignificant. Also, the effects of reservation wage reduce even the size
of community type effects (not reported here). Sweeping away all the effects of other
market indicators means strong correlation between local market conditions and
reservation wages, and could mean that students would have a higher reservation wage
because there is higher demand for them.44

However, there could be a compounding effect in the coefficient of the high
reservation wage dummy: students can have higher reservations just because they work.
This problem might be serious when a reservation wage is related not to market demand

but to a student’s ability. However, there are evidences which show that reservation

 

4’ Mean wage in manufacturing industries are not significant even though it is negatively related with high
school employment. However, the mean wage hardly represent the high school students wage. Most
students earn 24 dollars per one hour.

58

wages represent market demand students face. First, when we include the high reservation
wage dummy, the coefficients of test scores in these regressions do not change (not
reported) while the coefficients of all other market indicators and region or community
types becomes smaller and insignificant. Second, we tabulate the relationship between
reservation wages and test scores and the relationship between reservation wages and
county economic conditions among Whites in Appendix Table 1.45 Appendix Table 1
also shows that reservation wages are related to market demand rather than to individual
ability. If reservation wages are related to students’ ability, students with higher academic
ability might demand higher reservation wages. However, there is almost no differences
in the average test scores across reservation wage categories. However, appendix Table 1
shows that better county economic conditions are related to higher reservation wages.
Thus, it is very plausible that the reservation wages indicate the market demand

individual students face.46

Table 1-4-2-1: The Effects of Local Economic Conditions on Job-holdings (Probit model) Female: 4046 (HS&B)

 

 

 

 

Market (1) (2) (3) (4) (5) (6) (7) (8) (9)
Variables
U Rate -.0'333'¥TWI - - - W" 1.015 - -.014
(.008) 031“" 032*” 032*“ 028*” 027*" (.010) 035*” (.011)
(.009) (.009) (.009) (.009) (.009) (.009)
Employme .004 .004 .006 .007 .007
nt Growth (.006) (.006) (.006) (.007) (.006)
Income -.001 -.003 -.005 -.003 -.006
Growth (.006) (.007) (.143) (.007) (.007)
County .207“ .239" .105 .046
Income (.126) (.130) (.143) (.145)
% .012" .01 1"
Employed (.005) (.005)

 

’5 The reason we restrict the sample to Whites only is because there can be some relationship between race

and reservation wage.

‘6 Another interesting point in Appendix Table a is that there is no correlation between ‘will not work’ and

test scores, and ‘will not work’ and county economic conditions. Thus, the choice of ‘will not work’ seems

to be random choices by students.

59

Mean .019 .007

Wage (.018) (.006)

Reservatio 330*“

n Wage (.052)

Will Not -

Work 876""
(.179)

R Square .0326 .0326 .0326 .0327 .0331 .0334 .0334 .0328 .0522

 

Notes: (a) All other regressors in NLS-72 & Table HS&B 4-1-1 are included and the coelficients of them are not
reported. (b) see notes in the basic regressions (NLS-72 & Table HS&B 4-1-1 and 2).

Table 1-4-2-2: The Effects of Local Economic Conditions on Hours worked (tobit model) Male: 4064 (HS&B)

 

Market (1) (2) (3) (4) (5) (6) (7) (8) (9)
Variables
’WLJRau: ‘":”"WWWWW"IW"""W"""1"”WWWWWWL" ""‘W""}&§§¥¥¥'WW":335;?WWW2316¥¥"WWL‘WW"WWW5566¥¥"WW

503*" 474"“ .480‘” 461"" (.130) (.131) (.151) 487*" (.149)
(.121) (.122) (.122) (.123) (.126)

Employm .147“ .128 .147“ .150“ .160‘

ent (.084) (.087) (.088) (.088) (.086)

Growth

Income .120 .085 3.017“ .066 .012

Growth (.092) (.095) (1.861) (.098) (.097)

County 2921‘ .052 2.217 .939

Income (.1814) (.097) (2.041) (2.018)

% .071 .050

Employe (.075) (.073)

(1

Mean -.115

Wage (.251)

Reservati 6.935‘"

on Wage (.745)

Will Not -1 1.751“

Work (2.860)

R Square .0083 .0083 .0084 .0085 .0084 .0086 .0086 .0083 .0137

 

Note: See notes in Table 1-4-2-1.

Table 1-4-2-3: The Effects of Local Economic Conditions on Job-holdings (Probit model) Female: 4966 (HS&B)

 

 

 

 

 

Market (1) (2) (3) (4) (5) (6) (7) (3) (9)

Variables

u Rate -.029m .0593“ -.028”"‘ “628*" -13—2'05 463i)“ -.016* -.030m -.017*
(.008) (.003) (.003) (.003) (.008) (.003) (.009) (.008) (.010)

Employm -.002 -.005 .004 .003

ent (.005) (.006) (.005) (.005)

Growth

60

_“"‘I‘

 

 

 

Income .010“ .012" -.002 -.002 -.003
Growth (.006) (.006) (.005) (.005) (.005)
County 352"" .311“ .256" .202
Income (.117) (.122) (.135) (.136)
% .008 .009 .007
Employe (.006) (.006) (.006)
d
Mean .012
Wage (.017)
Reservati 219*"
on Wage (.042)
Will Not -1.044
Work (.163)
‘0‘
R Square .0525 .0525 .0530 .0531 .0539 .0542 .0543 .0526 .0685
Note: See notes in Table 1-4-2-1.
Table 1-4-2-4: The Effects of Local Economic Conditions on Hours worked (tobit model) Female: 4966 (HS&B)
Market (I) (2) (3) (4) (5) (6) (7) (3) (9)
Variables
WU Rate WWZW3W2‘13‘W'W.’3Wi§i¥*" W'Lf’z’ésfii'W"9.238;5"":3773W -. i 73 W -.087' -.323"* -.1 10
(.100) (.100) (.100) (.100) (.107) (.109) (.124) (.103) (.122)
Employm .017 -.035 -.000 .001 .001
ent (.067) (.069) (.070) (.070) (.069)
Growth
Income 216*" 226*” .166“ .185" .131‘
Growth (.076) (.078) (.081) (.082) (.081)
County 5488““ 4.772" 3.712" 2.427
Income (1.535) * (1.759) (1.735)
(1.593)
% .087 .052
Employe ( .061) (.060)
d
Mean .008
Wage (.209)
Reservati 5.209"
on Wage ‘
(.542)
Will Not -13.630
Work (2.356)
#1!
R Square
.0090 .0090 .0093 .0093 .0094 .0096 .0097 .0098 .0148

 

Note: See notes in Table 1-4-2-1.

61

Appendix Table 1: Reservation Wages, Test Scores and County Economic Conditions among Whites.

 

 

Reservation Reservation Reservation Reservation Will not
wage equal to wage above Wage Wage above work
$2.00 or less $2.00 and less between $3.50
than $3.00 $3.00 and
Test Scores 53.4 52.8 53.2 52.3 52.3
(8.67) (8.09) (7.99) (8.59) (8.97)
County
Unemployment 8.06 7.99 7.31 6.95 7.75
Rate _ (2.68) (2.71) (2.63) (2.40) (2.87)
Employment
Growth rates .76 .80 .93 1.29 .78
(3.36) (3.74) (2.94) (3.26) (3.47)
Percentage of
People employed 41.7 42.12 43.8 44.7 42.4
(6.18) (5.97) (5.83) (5.87) (6.42)
Log County Per
Capita Income 8.45 8.81 10.26 10.95 9.41
(4.79) (4.41) (4.09) (3.94) (4.88)

 

Tables 1-4-1-1 and 2 show a lower probability of job-holdings and less working
hours for racial minorities when they have the same other conditions. From now on, we
concentrate on the effects of being minorities on only job-holdings not working hours
because the average working hours among job-holders are similar across races (see
footnote 36).

The marginal effects of being Black, Hispanic or Asian on job-holdings are
tabulated in Table 1-4-2-5 and 6 for each sample. In 1972, after GPA, test scores, SES,
community types, regions, and other background variables are controlled for, Blacks,
Hispanics and Asians have much lower probability of being employed. Black males and
Hispanic males have 14% and 9% lower probability of being employed respectively at the
mean values of the independent variables, while Black females, Hispanics females, and
Asian females have 14%, 12%, and 15% lower probability. (See columns 1 in Table 1-4-
2-5). Instead of SES, detailed family background variables; categorical variables for

mother’s education, father’s education, father’s occupation, and family income are added

62

to the regressors, the results do not change. (see columns 2 in Table 1-4-2-5). It turns out
that after other conditions are controlled for, on the average, Black male students work
over 6 hours less than White counterpart, Black female students work over 4 hours less
than White counterparts, and Hispanic student work 3 or 4 hours less than White students
(see the basic regression Table 1-4-1-2 ).

Many of these trends remain unchanged in 1980 except for Hispanic and Asian
males. Table 1-4-2-6 show that also in 1980, under the same conditions as columns 1 in
Table 1-4-2-5), Blacks and Asian students have much lower probability of job-holdings.
At the mean values of control variables, Black students, Hispanic students and Asian
students have 4~20% lower probability of working (see columns 1 in Table 1-4-2-6).
These probabilities do not change when more control variables are added. In column 2,
the county unemployment, the percentage of people employed within the county, the
average per capita income, the employment growth rate, the income growth rate, high
reservation dummy, and ‘will not work’ dummy are added. In column 3, in addition to
those variables, detailed family background variables are added and in column 4, a
dummy which indicates whether a student has worked during his or her junior year is also
added. However, except for the column 4, there are no big changes in the coefficients of
‘Blacks’, ‘Hispanics’ and ‘Asians’. However, minority status is expected to affect the
junior year job-holdings. Under the same conditions, Black male students and Asian male
students work 3~4.5 hours less and 4.5~6 hours less than White counterpart respectively
and Black and Asian female students work 2.5~3.5 hours less than White counterparts.

(see the basic regressions in Table 1-4-1-2).

63

However, in 1980, the gap in job-holdings between Whites and Hispanic narrows.
At the mean values of the control variables, the probability of being employed for
Hispanics was 8~12% lower than Whites in 1972, while it is 2~7% lower in 1980. Also,
in 1972, Hispanic senior students worked 3-4 hours less than White students. In 1980, the
difference becomes around one hour. We do not know whether demand for Hispanics
youth has changed or the labor force participation of Hispanic students has changed.
However, the narrower race gap of high school employment between Whites and
Hispanics in 1980 is consistent with Marsh’s finding (1991). Another interesting change
is that the coefficients of Asian dummy for males is much lower than in 1972, which
means that the job-holding probability for the Asian males in 1972 was lower than in

1980.

Table 1-4-2-5 The Effects of Being a minority on job-holdings (NLS-72)

 

 

 

Races Male Female
(1') WWiziWWWWW'('0Ww (2)
Blacks -.157"‘" -.159"‘" -.138*" -.136"‘"
(.023) (.023) (.021) (.022)
Hispanics -.088"* -.101”" -.120”"‘ -.121”*
(.032) (.032) (.033) (.033)
Asians -.072 -.075 -.155*** -.152""‘
(.051) (.051) (.061) (.061)
R Square .0347 .0426 .0244 .0271

 

Notes: (a) In column (I), all other variables in the basic regressions are included. (b) In column (2) detailed family
backgrounds instead of SES categories are included. (c) See notes in the basic regressions (Table 1-4-1-1 and 2).

 

 

Table 1-4-2-6 (HS&B)
Races Male Female
(I) (2) (3) <4) (I) (2) T3) (4) "
Blacks -.l38"* -.136*” -.139"* -.116*** -.161"“‘ -.169*** -.163”* 431*"
(023) (024) (024) (024) (021) (022) (022) (022)

64

:QIQ

 

Hispanics -.044" -.O37‘l —.040"' -.024 -.066*“ -.057‘”'I -.055“ -.037"‘

(022) (.023) (023) (023) (.022) (.023) (023) (023)
Asians 492*" -.198"‘ -.151m -.147m -.148*" -.152*" -.l48"* -.154"*

(047) (048) (058) (049) (046) (047) (047) (048)
R Square .0296 .0523 .0564 .1070 .0502 .0683 .0733 .1109

 

Notes: (a) In column (1), all other variables in the basic regressions are included. (b) In column (2), in addition, all the
economic indicators are included. (0) In column (3), detailed family backgrounds instead of SES categories are
included. In column (4), ‘work in the junior year’ is included. ((1) See notes in the basic regressions (Table 1-4-1-1 and
2).

The significance of negative effects of race, especially for Blacks, on job-
holdings and working hours does not mean necessarily less job availability for their
group. There could be supply side effects if Blacks, Asians and Hispanics could have
tendency not to work or not to search as vigorously enough for jobs due to some cultural
effects. The supply side effect is likely to be larger among Asians considering relatively
low unemployment rate and low labor force participation rate among them (Tables 1-3-
10, 11 and 12). However, a twice as high unemployment rate among Blacks than among
Whites (see Table 1-3-10, 11 and 12) with much less labor force participation rate (which
could be caused by low demand for their labor) indicates lower job availability for them.
Also, since other variables (an ability measure, SES, and local economic conditions) are
controlled for in the regression analyses, the significant negative effect of being Black
implies that jobs are not available for Blacks. When we restricted the sample to those who
are in the labor force, we have almost the same results (not reported in Tables), which
strongly implies that minorities, especially Blacks, want to have jobs but jobs are not
available for them as much as for Whites.

(3) Community type: Another sign of the strong positive correlation between high

school employment and job availability is found in the influences of community types on

65

job-holdings and working hours. Four categories of community type for NLS-72 (‘rural,’
‘small city town,’ ‘urban,’ and ‘suburban’ ) and three categories for HS&B (‘rural,’
‘urban,’ and ‘suburban’) are used. The category of ‘rural’ is excluded in the regression
analyses. (We do not make separate tables for the community effects. Please see the basic
regressions in Table 1-4-1-1 and 2). We expect students who live in urban and suburban
areas work more than those who live in rural areas under the same conditions of other
factors because students’ jobs are concentrated in the private service sector.

In 1972, in the binary dependent model regressions for job-holdings, all the
coefficients of community type categories are negative and significant for males, and for
females, almost all the coefficients are positive and significant. (see the first and second
columns in Table 1-4-1-1). However, in terms of working hours, male students who live
in the urban and suburban areas work almost as much as those who live in rural areas.
Female students who live in urban and suburban areas work much more than those who
live in the rural areas (see the first and second columns in Table 1-4-1-2).

In 1980, when almost the same regressions as in the year 1972 are run, the results
are almost the same even after local unemployment rates are controlled for. For male
students, categories of ‘urban’ and ‘suburban’ are positively related to ‘working hour’
while they are not related to job-holdings. (see the first columns in Table 1-4-1-1 and 2).
For female students, these categories are positively related to both of the two dependent
variables (see the second columns in Table 1-4-1-1 and 2) One possible explanation of
these results in both 1972 and 1980 is that the male students who live in the rural areas
might have more jobs on the farms. However, their working hours might be shorter

because farm work must be done during daytime. For females, students who live in urban

66

and suburban areas work much more than those who live in rural areas because female
students do not have many jobs on the farms but have jobs in the service sector which is
concentrated in the urban and suburban areas.47

Considering all the effects of race, various economic indicators and community
types on job-holdings and ‘working hours’, we can conclude that high school students
work because they have more opportunities to work and more jobs available around them
and more demand for their labor while almost all students are willing to work.48 If student
workers keep living in these favorable circumstances after high school, their gains in later

earnings and employment are not surprising.

4-3: Academic ability: As we mentioned in the previous section, with respect to the
relationship between the ability measure (standardized test scores) and working
experience during high school, we have focused on the following two points: (i) the
relationship between different working experiences and academic ability, especially the
differences between working in junior year and working in senior year in their relations to
academic ability. (ii) non-linearity of the relationship of test scores to job-holdings and
working hours.

(i) The relationship between different working experiences and academic ability:
The marginal effects of test scores on the different working status at the mean values of
control variables are tabulated in Table 1-4-3-1 and Table 1-4-3-2. In these regressions,

test scores are specified in the linear function. As mentioned earlier, the relationship has

 

’7 However, when all other labor market conditions including the reservation wage dummy and ‘will not
work’ dummy are controlled for, the significance of the coefiicients disappears.
’8 Students who will not work at any reservation wage during high school is less than 2% in 1980.

67

changed over time. In 1972, ‘work’ and ‘work more than 15 hours’ are negatively related
to test scores. A one standard deviation of test scores decrease the probability of working
or being employed byl .5% , the probability of working more than 15 hours by 4% for
males. For females, it decreases the probability of working more than 15 hours by 1.7%
and has no effects on the probability of working. In 1980, a one standard deviation of test
scores increases the probability of working or being employed by 2% for male and 3.5%
for female.

However, in both years, test scores are positively related to working moderate
hours (6-20 hours for the NLS-72 samples and 5-21 hours for the HS&B samples). In
1972, a one standard deviation of test score increase the probability of working between
6-20 hours by 1.7% for males when the relationship between test score and ‘work’ or
‘heavy work’ is negative. In 1980, the positive relationship between test score and
‘moderate work’ is much stronger than between test score and ‘work’ or ‘heavy work’.
When we assume that working hours do not affect academic ability, we can conclude that
the students with good academic ability work moderate hours. Then, the positive effects
of moderate working hours is not surprising. The selection effects of this moderate work
could show up in the quadratic function of working hours as the positive coefficients of
working hours and the negative coefficients of the working hours squared.

Table 1-4-3-2 also shows other interesting points. Students who work during the
reference week have higher academic ability than those who have worked at least once
during the senior year (compare the column ‘work’ and the column ‘work reference
week’.) Considering the high tum-over rates among the high school students, measuring

the high school work experience by the hours worked during the reference might have a

68

more serious selectivity problem (See Ruhm 1997). Working in the junior year is

negatively related to test scores. Comparing the ‘work’ column and the ‘work junior’

column shows the dramatic change in the relationship. Also working during summer time

is negatively related to test scores even if the absolute value of the coefficient is much

smaller in this case. If we measure the high school work experience by the average hours

worked during junior and senior year rather than by the hours worked during only the

senior year, we can expect the lower effects of high school working experience on later

earnings (See Light 1999 and Hotz 1999).

Table 1—4-3-1: The marginal effects of test score on the different working status in the senior year. (NLS-72)

 

 

Work Work More than 15 Work 6~20 Hours
-______._-.._-__--_--_---___,__.__~_.______-_____...-~___i _-___119urs
% of Students 81.6 50.1 34.3
M The Coefficient of Test -.0017" -.005"” .0020"
Scores (.0007) (.001) (.0009)
% of Students 69.7 32.7 36.4
F The Coefficient of Test -.0000 -.0020"“" .0012
Scores (.0010) (.0010) (.0010)

 

Note: (a) All other variables in the basic regressions are included and are not reported. (b) See notes in the basic

regressions (Table 1-4-1-1 and 2). (c) M: Male sample: F: Female sample

 

 

Table 1-4-3-2: The marginal effects of test score on the different working status (HS&B)
Work Work more Work 5-21 Work Work Junior Work
than 14 hours Reference Summer
__g_____ - ~_ .- _______ hours“ W _> __ _ _ _' Week --_-___-
%of 72.0 50.7 35.3 51.3 71.7 86.0
Students
M The .0024” .0012 0048"" 0039“" -.0052"”" -.0018“
Coefficient (.0009) (.0012) (.001 1) (.0012) (.0011) (.0008)
of Test
Scores
% of 66.6 41.6 37.7 46.7 58.7 73.2
Students
F The 0040"" .0035‘" .0060‘" 0048“" -.0018 .0000
Coefficient (.0011) (.0012) (.0011) (.0012) (.0012) (.0010)
of Test
Scores

 

Note: (a) All other variables in the basic regressions are included and are not reported. (b) See notes in the basic

regressions (Table 1-4-1-1 and 2). (c) M: Male sample; F: Female sample

69

Next, we focus on the relationship between academic ability and work experience
during the senior year because high school students work most intensively during the
senior year”, and because many studies on the effects of high school employment focus
on the working experience during the senior year and find the positive effects. Here, first,
we look at how the working experience in the junior year, and summer affect job-
holdings and hours worked during the senior year and how controlling for these previous
work experiences change the relationship between the work experience and test scores. At
the mean values of the control variables, the marginal effects of previous working
experience and test scores on job-holdings in the senior year are tabulated in the Table 1-
4-3-3 and those effects on working hours in the senior are tabulated in the Table 1-4-3-4.
In the column 1 and 2, the same other control variables are included as in the ‘basic
regressions’ Table 1-4-1-1 and 2. In the column 3, in addition, all the economic
conditions including the reservation wage are controlled for.

These tables show that students who have worked during either the junior year or
summer between the junior year and senior year have stronger tendency to work during
their senior year. After controlling for the job-holding experience during summer,
students who have worked during their junior year have 19~24 % higher probability of
working in their senior year. Students who have worked during the summer have 10~14
% higher probability of working in their senior year. Also, working hours during the
senior year are affected by the working hours during the junior year and summer. The
coefficients of working hours in the junior year and summer in the tobit models for

workings hours in the senior year are in the range between .2~. 4. When the ‘work

 

’9 See footnote 32.

7O

junior’ dummy and ‘work summer’ instead of hours worked during those periods are

plugged into the regressors, the results show that those who have worked during the

senior year work around 8-9 hours more than those who have not worked and those who

have worked during the summer work 6-7 hours more than those who have not worked

(the results are not reported.) Another observation we can make from these tables is that

when working or working hours in the junior year is controlled for, the already positive

coefficients of test scores on ‘work’ or ‘working hours’ become bigger, especially for

male working hours.50 It is because the relationship of working during the junior is

negatively related as we see in the column 4 in Table 1-4-3-2.

 

 

 

Table 1-4—3-3 Probit for job-holdings during the senior year. (HS&B)
.2. Male -Miemls
(l) (2) (3) (l) (2) (3)
Test Score .0024" .0041 ""' 0039*" .0040‘" 0047"“ 0042""
(.0011) (.0011) (.0011) (.0011) (.0011) (.0011)
Work Junior 237*” .228‘" .189"* .186‘”
(.018) (.019) (.015) (.016)
Work .136”" .124‘" 418”" .105*"
Summer (.024) (.024) (.018) (.018)
R Square .0296 .0956 .1 102 .0501 .1003 .1130
LR Chi Square 142.43 461.72 530.78 317.26 635.04 715.36

 

Notes: (a) In columns (1) and (2), other control variables in the basic regressions are included. (b)
In column (3), in addition to those variables. all the economic indicators in the last column in the Table 1-4-2-6 are
included. (c) See notes in the basic regressions (Table 1-4-1-1 and 2).

 

 

Table 1-4-3-4 Tobit for working hours during the senior year. (HS&B)
Male Female
(1) (2) (3) (1) (2) (3)
Test Score .0077 .0931 *‘1‘ .0752" .0929” .0991 *" .0818"
(.0462) (.0430) (.0428) (.0403) (.0376) (.0375)
Working Hours .369‘" 355"" 305*" 291""
Junior (.026) (.026) (.023) (.023)
Work Hours 266*" 245*" 230*" 209*"
Summer (.024) (.024) (.018)
R Square .0076 .0298 .0335 .0086 .0281 .0314
LR Chi Square 189.52 746.17 839.15 255.88 831.36 930.75

 

Notes: (a) See the notes Table 1-4-3-3.

 

5° Summer work does not seem to be related to test scores. Including or excluding ‘work summer’ does not
change the coefficients of test scores.

71

(ii) Nonlinear relationship between high school employment and test scores: To
examine the non-linearity of the relationship, first, we specify test score in the quadratic
function. The coefficients of test scores and test scores squared in the probit model for
job-holdings and tobit model for working hours are tabulated in Table 1-4-3-3 for
males and Table 1-4-3-4 for females and Table 1-4-3-5 for males and Table 1-4-3-6 for
females. In this specification of test scores, the coefficients of test scores are always
positive and significant and those of test scores squared always negative and significant
(see columns 2 in each tables.) However, the sizes of these coefficients are quite different
between two years. The coefficients of test scores in 1982 are much larger than in 1972
while the coefficients of test scores squared in 1982 are much smaller than in 1972,
meaning that more students with high academic ability hold jobs and work more hours in
1980.

To examine the nonlinear effects more specifically, we make a categorical
variable of test scores with four categories: one category for each test score quartile and
the first quartile dummy is dropped and the same regressions are run. The results show
that in 1972, students whose test scores are in the 3rd and 4‘11 quartiles work as much as or
less than students in the first quartile (see columns 3 in Table 1-4-3-3 and 4). However, in
1980, even students in the 4‘h test score quartiles work much more than students in the
first quartile except for the case of male working hours, and work almost as much as
students in the 2nd and 3rd quartiles (see columns 3 in Table 1-4-3-5 and 6). When more

categories for test scores are made and the same regressions are run, the results show that

72

-r-r'.‘u.‘

the top 10% of students in the test score distribution work almost as much as the bottom
10% and work less than those who are in the middle (not reported here).

Using the HS&B data, we also examine a non-linearity of the effect of test scores
on job-holdings and working hours in the junior year by including test scores and test
scores squared, or by including categories of test scores quartiles (the later results are not
reported). It turns out that there is not such a non-linear relationship but only a negative
linear relationship shown in Table 1-4-3-2. The higher test scores a student has, the lower
is his probability of working in his junior year. The lower test scores a student has, the
more hours he works in the junior year. If the causal effect runs from test scores to
working experience, it can imply that students with lower academic ability work

intensively in the earlier grades.5 I

 

 

Table 1-4-3-5: Male. (NLS-72)
Probit for work Tobit for Working hours
(1) (2) (3) (1) (2) (3)
Test Score -.007" 075"" -.185“" 769*”
(.003) (.023) (.032) (.250)
Test Score Squared -.0008""‘ -.0094‘”
(.0002) (.0024)
2nd Test Quartile .0973 .149
(.006) (.589)
3rd Test Quartile .058 -1397...
(.062) (.654)
4"” Test Quartile -. 100 3792*"
(.070) (.743)
X square 223.93 236.99 234.04 475.58 490.34 484.51
R Square .0347 .0367 .0362 .0113 .0116 .0115

 

Notes: (a) All other control variable are included as in the basic regression and are not reported. (b) The first
columns in each regression are the same as in the basic regression and reported here for comparison. (c) See notes in
the basic regressions (Table 1-4-1-1 and 2).

 

5 ' If work experience during high school afi‘ects test scores, it is hard to interpret these results. This is (i)
because the effects of working during the senior year seem to be quite different from those of working
during the junior year, and (ii) because the effects of working experience during the senior year in 1980
seem to be different fi'om those in 1972.

73

Table 1-4-3-6: Female

(NLS-72)

 

 

Probit for work Tobit for Working hours
(1) (2) (3) (1) (2) (3)
Test Score -.0001 .0549" -.0589* 1.067‘"
(.0028) (.0203) (.0315) (.2349)
Test Score Squared -.0005"‘"' -.0111"”""'
(.0002) (.0023)
2"d Test Quartile .0751 .800
(.0472) (.541)
3rd Test Quartile .1 138” .605
(.0539) (.6131)
4‘h Test Quartile -.0062 .1. 165*
(.0622) (.7117)
X square 210.39 217.82 219.85 324.10 347.51 335.40
R Square .0244 .0253 .0255 .0078 .0084 .0081

 

Notes: (a) All other control variable are included as in the basic regression and are not reported. (b) The
first columns in each regression are the same as in the basic regression and reported here for comparison.
See notes in the basic regressions (Table 1-4-1-1 and 2).

 

 

Table 1-4-3-7: Male (HS&B)
Probit for work Tobit for Working hours
(I) (2) (3) (1) (2) (3)
Test Score 007*" .126‘” .008 1.9803"*
(.003) (.025) (.046) (.3612)
Test Score Squared -.0012"'" -.020”*
(.0002) (.004)
2"“ Test Quartile .173m 2391"
(.065) (.933)
3rd Test Quartile .210m 1.997"
(.070) (.999)
4‘“ Test Quartile 185." .962
(.079) (1.127)
X square 142.43 165.72 147.81 189.52 219.86 197.95
R Square .0296 .0344 .0307 .0076 .0088 .0097

 

Notes: All other control variable are included as in the basic regression and are not reported. (b) The first
columns in each regression are the same as in the basic regression and reported here for comparison. See
notes in the basic regressions (Table 1-4-1-1 and 2).

 

74

 

Table 1-4-3-8: Female (HS&B)
Probit for work Tobit for Working hours
(I) (2) (3) (l) (2) (3)
Test Score .011 .088‘” .093" 1.849‘"
(.003) (.024) (.040) (.317)
Test Score Squared -.0008*” -.018""‘
(.0002) (.003)
2"" Test Quartile .167""’ 2.864m
(.053) (.716)
3rd Test Quartile .243": 4009*"
(.060) (.805)

4‘h TeSt Quartile 262"" 2.599m

(.073) (.943)
x square 317.26 327.49 323.28 255.88 287.06 278.53
R Square .0501 .0517 .0511 .0086 .0097 .0094

 

Notes: (a) All other control variable are included as in the basic regression and are not reported. (b) The
first columns in each regression are the same as in the basic regression and reported here for comparison.
See notes in the basic regressions (Table 1-4-1-1 and 2).

In sum, the relationship between test scores and high school senior employment is
very complex. In 1980, students with high academic ability seem to work more than in
1972. Employment in the senior year is more selective in terms of academic ability than
employment in any other grades. The relationship between high school employment in the
senior year and test scores is non-linear.

Due to this complexity, it is difficult to interpret the positive effects of high school
working experience on later earnings and employment after high school. It is difficult to
control selection bias in the later earnings equations (or employment equations) and to
find the true effects of high school employment on later earnings or employment.
However, from the above results, we can draw some implications with respect to
controlling for the selection bias. First, we need to distinguish working in the senior year
from working experiences in earlier grades. Second, including test scores as independent
variables in later earnings equations (or employment equations) is not enough to control
for the selection bias regarding test scores because there is not enough linear correlation
between high school working hours and test scores. Also, if we want to use the
instrumental variable method or treatment effect to control for the selection bias, we need
to include non-linear forms of test scores in the first stage regressions (high school

employment decision equation or high school working hours equation). This is because

75

the linear form of test scores in the high school employment equation can not capture the
true relationship between high school employment and test scores. Third, we need to
consider the period or year when samples are drawn. The regression results might be

different depending on the sampling years.

4-4: Family socio-economic status: As mentioned in the subsection 4-1, SES (family
socio-economic status) affects job-holdings and working hours during high school in a
nonlinear pattern as we expect.52 In 1972, for males and females, more students from the
middle SES hold jobs than those from the bottom and top SES after other variables are
controlled for. Students from the top SES hold jobs almost at the same rate as students
from the bottom SES (see the first two columns in Table 1-4-1-1). In terms of working
hours, students from the middle SES also work the most. However, students whose SES
is in the top work almost as much as students whose SES is in the bottom (see the first
two columns in Table 1-4-1-2).

In 1980, in terms of job—holdings, the same trends as in 1972 remain only for
males (see the last columns in Table 1-4-1-1.) Male students from the middle SES hold
jobs the most. However, in the female case, students from top SES hold jobs more than
others. In terms of working hours, Male students from the middle SES work most and
females from the middle and top SES work the most. (see the last columns in Table 1-4-
1-2.)

To see the effects of family background in more detail, we put categorical

variables of mother’s education, father’s education, father’s occupation, and family

76

income as independent variables in the probit and tobit models with other control
variables in the basic regressions. The coefficients of these variables are tabulated in
Table 1-4-4-1 and Table 1-4-4-2. First, mother’s education level seems to be less
important than other family background factors. In 1972, female students whose mothers
have more than four year college education seem to work less in terms of working hours.
However, except that, no coefficients of mother education categorical variables are
significant. Second, father’s education affects students’ working even after father’s
occupation is controlled for. In 1972, male students whose fathers have more than 4 year
college education work less in terms of job-holdings and working hours and male
students whose fathers just graduated from high school seem to work the most. However,
in 1980, female students whose fathers have more than 4-year college education work the
most in both terms of working hours and job holdings. Third, when the father is a farmer,
student work most in terms of job-holdings and working hours except for the 1980 female
case. Fourth, students whose family incomes are in the middle seem to work the most.
The non-linear relationship between high school employment and SES seems to be driven
mostly by the family income effects.

Students from the middle SES group tend to work more than students from the
bottom and top SES group. However, this non-linear influence of family background is
weaker in 1980 than in 1970 because more students from high SES seem to work in 1980.
Also, in 1980, it is even much weaker for females than for males because female students

with fathers who graduated from college work more than others. Also, this non-linearity

 

5‘ In NLS-72 and HS&B, SES is made out of family income, parents’ education, and father’s occupation
Duncan score. For detailed description, see NLS-72 Data File Users Manual.

77

tendency of the effects of SES is very evident in the job-holdings and working hours in

the junior year (not reported).

 

 

 

Table 1-4-4-1: Family Socio-economic Status (NLS-72)
Probit for work Tobit for Working hours
Male Female Male Female
Mother Education .010 -.001 .434 -.145
High School Graduate (.050) (.041) (.525) (.468)
Mother Education -.076 -.022 -l.158 -.313
Some College (.065) (.054) (.691) (.615)
Mother Education .048 -.081 -.845 -2.109”*
More Than Four Year College (.077) (.068) (.823) (.773)
Father Education -.053 .1 10” -.757 1.113"
High School Graduate (.058) (.042) (.524) (.483)
Father Education -.031 .066 -1.044 .323
Some College (.062) (.053) (.648) (.596)
Father Education -.171“ .080 -2.964"‘" .391
More Than Four Year College (.074) (.066) (.788) (.741)
Professional .158" -.025 .078 -.500
(.072) (.065) (.775) (.740)
Management .174‘" .057 1.214"I .307
(.062) (.056) (.657) (.631)
Farmer 396”" .234‘" 5.787‘" 2.233’"
(.103) (.078) (.945) (.865)
Sales and Clerical .163‘” .025 .836 .398
(.067) (.056) (.708) (.655)
Craftsman .131"‘" .027 .864“ .157
(.050) (.043) (.525) (.491)
Family Income($6.000-$9,000 .182‘" -.023 1.503" -.066
in 1972) (.068) (.057) (.719) (.653)
Family Income($9,000-$12.000 .091 .032 1.344" .958
in 1972) (.068) (.060) (.730) (.679)
Family Income($12,000- .240‘" .108 2.884‘” 1.489"
$15,000 in 1972) (.077) (.068) (.798) (.763)
Family Income(Above $15000 .041 .003 1.815” .405
in 1972) (.072) (.065) (.771) (.741)
X square 275.43 233.36 553.34 351.90
R Square .0426 .0271 .0131 .0085

 

Notes: (a) all other variables except SES categories are included and not reported. (b) See notes in

(Table 1-4-1-1 and 2).

the basic regressions

 

 

Table 1-4-4-2: Family Socio-economic Status (HS&B)
Probit for work Tobit for
.. , - - - ._______ -__-__--___------_..W.9r_k.i_08.heurs- __
Male Female Male Female

Mother Education -.014 .043 1.022 .903

High School Graduate (.056) (.048) (.794) (.636)

Mother Education .076 .024 .897 .580

Some College (.085) (.032) (1.173) (.922)

Mother Education .058 -.077 .731 -.312

More Than Four Year College (.871) (.078) (1.228) (1.005)

Father Education .072 .099“ .210 .854

78

High School Graduate (.060) (.052) (.840) (.670)
Father Education .038 -.035 -.254 -.089
Some College (.090) (.075) (1.244) (.983)
Father Education -.032 ..248"" -1.721 2.291 ”
More Than Four Year College (.087) (.082) (1.210) (1.029)
Professional -.1 l3 .1 15 -l .051 .752
(.073) (.077) (1.637) (.949)
Management .071 .067 .706 1 .499
(.082) (.078) (1.122) (.977)
Farmer .223‘" -.120 4.705‘“ -2.245
(.123) (.104) (1.661) (1.421)
Sales and Clerical .090 .055 .351 .108
(.104) (.092) (1.415) (1.168)
Craftsman .093 .085 .988 1.735"
(.067) (.064) (.929) (.815)
Family Income($7,000- -.039 .1 15 .806 1.47]
$11,999 in 1980) (094) (.075) (L366) (1026)
Family Income($12,000- .108 .135" 3.092" 1.983"
$15,999 in 1980) (093) (.077) 0377) (1040)
Family Income($16,000- .067 134* 2.566" 2.057"
$19,999 in 1980) 1-096) (079) (1374) (1055)
Family Income($20,000 to .067 .174” 3.322" 2417“
$24,999 in 1980) (096) (082) (1-374) (1-090)
Family Income($25,000 to -.028 .076 2.177 1.378
$37,999 in 1980) (098) (084) (L408) (L116)
Family Income($38,000 or .076 .087 3.079" 1.131
more in 1980 £103) (092) (L456) ( 1-206)
x square 162.32 347.53 215.15 285.34
R Square .0337 .0549 .0086 .0096

 

Notes: Notes: (a) all other variables except SES categories are included and not reported. (b) See notes in the basic

regressions (Table 1-4-1-1 and 2).

4-5: Other variables: When the mother works when students are in the high school, both

male and female students tend to hold jobs more and work more during high school than

others (Table 1-4-1-1 and 2). Greenberger and Steinberg (1986) argue that “the

employment of individual mother may serve as a model for the employment of her own

children and others.” (p36) Also, they point out that the adult women’s increased

involvement in the labor force has changed the day time ecology of the household. Thus,

increases in women’s labor force participation in more recent years might have caused

high school student to work more than before.

79

The number of siblings affects high school employment positively and in most
cases, significantly. However, in 1980, the number of siblings does not affect male
student employment and even the sign of the coefficient of the number of siblings
becomes negative though it is not significant in most cases.

If students have their own dependents (1972) or children (1980), they work more
in terms of job-holdings and hours worked in general. In 1972, having dependents affects
significantly and positively the job-holding and hours worked for both male and female
students. In 1980, male students with children still work more. However, female students
with children hold jobs less and work less hours. We suspect that this change in the
impact of having own children on female employment may occur due to the change in the
way the welfare system supports single mothers over this period of time.

There is no significant difference in student’s employment depending on whether
English is spoken at home or not.

Some other variables affect high school employment. One of them is the high
school program. High school students in the academic program work less and students in
the vocational program work more than those in general program in most cases. Students
in academic programs might be encouraged not to wok, while students in vocational
programs might be encouraged to work. This relationship implies that school can be a

determinant in high school employment.

80

5. Conclusion.

In the conclusion, we discuss the implications of our findings in the context of
previous research on the effects of high school employment on later earnings and
employment. First, our review of the studies on the effects of high school employment on
later earnings indicates that the measurements of high school work experiences are
different across studies, and the regression results from these studies seem to be sensitive
to the measurements and specifications of high school work experience in the earnings
equations. Our study might explain why different measurements or specifications lead to
different results. It seems to be, in a certain degree, related to the complexity of the
relationship between high school employment and academic ability. In terms of test
scores, employment during the senior year, especially during the reference week, is the
most selective. However, job-holdings and working hours during the senior year are non-
linearly related to test scores, while job-holdings and working hours during the junior
year are negatively and linearly related to test scores. Students with medium academic
ability seem to work the most during their senior year. Students with high academic
ability seem to work moderate hours if they hold jobs.

From these observations, we can expect that the different measurements of high
school work experience might capture students’ different abilities by different working
experience. If we measure only the work experience during the senior year, we might
expect large positive effects of high school work experience on later earnings. Or if we
measure work experience in each grades separately, we might expect large positive
effects for the work experience in the senior year, and small positive effects for the work

experience during other grades. If we measure high school work experience as the

81

average working experience during the high school years, we expect smaller positive
effects of high school employment than when we measure work experience only during
the senior year.

Mentioned earlier, Ruhm (1997) distinguishes working hours in the senior year
from working hours in the junior year and puts them separately in the quadratic function
in earnings equations. The coefficients of the first order term of senior year working
hours are positive and significant, while those of the second order term are negative and
significant. Also, all the coefficients of the first and second order terms of working hours
in the junior year are not significant. The large positive first order coefficients of working
hours in the senior year seem to capture the highest average academic abilities of students
who work moderate hours during their senior year. We also find that working in the
reference is more selective. When Ruhm uses the average working hours per week from
the work history file in NLSY, the positive effect is too sarnller.

Light (1999), who uses the average work experience during the junior and senior
years and put them in the linear function in the earnings, finds smaller positive effects
than Ruhm (1997). Hotz et al (1999), who regard all working experience as homogeneous
regardless of working hours and grades, find no effect when all the selection bias is
controlled for.53

Second, due to the complexity of the relationship between academic ability and
high school employment, it is very difficult to control for the selection bias. Some simple

ways to control for selection bias of high school employment in the later earnings or

 

53 However, we are not sure that such findings in Light and Hotz et a1 is directly related to their different
measurements of high school employment because they control for working experience after high school
and use different econometric methods.

82

employment equations regarding ability are (1) just including ability measures (test
scores) in the later earnings or employment equations, (2) including the inverse Mill’s
ratio in those equations after a binary dependent variable model of work choice during
high school is run (the treatment effects method), or (3) including the estimated high
school work experience after the reduced form of the work experience is estimated by a
OLS (instrument variables method). However, without thorough and full investigation
into how and what factors affects high school employment, these efforts can not
guarantee successful control for selection bias.

Just including test scores in later earnings or employment equations can not
capture the selection bias of high school employment because there is no linear
relationship between high school employment and test scores. It only captures the linear
relationship between later earning and test scores. However, when we use the IV method
or treatment effect method, we need to include the non-linear forms of test scores in the
first stage regressions. By doing so, we are able to capture the non-linear relationship
between high school employment and ability measure. Especially, if we use the quadratic
form of test scores with IV method, we might be able to identify the high school
employment equations because we have an extra variable-test scores squared- in the first
stage regressions. It is especially true when we are not able to identify high school
employment equations because there is no other variables which are not included in the
second stage regressions but still affect high school employment.

Third, in studies which directly investigate the effects of high school employment

on later employment, strong evidence of the positive effects on later employment can be

83

found54 (Meyer and Wise 1982 and 1984; Ruhm 1997; Carr 1997; Marsh 1991). Also,
those studies that find lower or no such positive effects on later earnings always control
for working experience after high school graduation (Light 1999 and Hotz et a1 1999),
which presents also other evidence for a strong correlation between high school
employment and working experience after high school. Even if there is no direct positive
effect of high school employment on later earnings, there still exists indirect positive
effects on later earnings by enhancing the later working experience.

Our current study provides some explanation for why there are such positive
effects of high school employment on later employment (and probably later earnings).
Students who have more favorable market conditions seem to work more. The demand
for student labor seems to be a more decisive factor than the supply of student labor.
Students who live in counties with good economic conditions, such as low unemployment
rates, high employment grth rates, high per capita income and fast income growth,
hold more jobs and work more than others. Even these county economic indicators do
not seem to capture the demand that an individual student might face as we see the very
large positive effect of a high reservation wage. Students who have more favorable
market conditions during high school are more likely to live and work in the similar
conditions after high school. Considering the two other factors which can affect job
availability, race and community type, these favorable conditions for working high school
students might intensify. While many studies control for race, only a few studies try to
control for the initial market conditions in which students have lived during their high

school years. One way to do so is including the local market conditions of the high school

 

5” However, there no studies which investigate the effect of high school employment on later employment

84

years in the later earnings equations is. Ruhm (1997) includes only the initial local
unemployment rates as a categorical variable in the later earnings equations. However,
including local unemployment rates is minimally adequate because, as our study suggests,
they do not represent appropriately the market demand each individual student faces.

Fourth, the family socio-economic status affects high school employment in a
non-linear way. Students whose SES is in the middle work the most. If a student with a
high SES works, he or she generally works moderate hours. This makes it very hard to
control for selection bias in terms of family backgrounds.

Fifth, Ruhm (1997) summarizes the changes in the characteristics of high school
employment over-time. He wrote: “Neither the frequency of student job-holding nor the
work hours of employed students have changed much since the late 19608, with any time
trend being dwarfed by cyclical fluctuations” (746-747). We also compare the
characteristics of high school employment between the years, 1972 and 1980, and find
that there is not much difference in terms of students job-holdings and working hours
between these two years. However, we find one big change in the determinants of high
school employment over-time, which is that, students with higher academic ability and
higher SES worked significantly more in 1980 than in 1972.

This observation might be very important in interpreting the effects of high school
employment on later earnings or employment because the positive effects of high school
employment seems to have decreased over-time. As mentioned earlier, the positive
effects of high school employment seem to be larger for earlier cohorts (NLS-72 cohorts

or YTS cohorts) than for later cohorts (the NLSY cohort or the HS&B cohort). In more

 

by controlling for the selection bias except Meyer and Wise (1982).

85

recent cohorts, more students with higher academic ability and higher SES participate in
the labor force and work more. Then, without controlling for the selectivity bias, the
positive effects of high school experience must be larger in more recent cohorts than in
old cohorts. However, existing studies show the opposite possibility. This cohort
observation could imply that the changes in the effects of high school employment might
depend on the changes in underlying factors, such as changes in the distribution of the
different industries or occupations.55 We need to investigate this issue further.

With various econometric methods, most studies focus on controlling for the
unobserved heterogeneity in the investigation of the effects of high school employment.
They ignore these complexities of its measurements and determinants. In this study, we
summarize the existing studies on the effects of high school employment on later earnings
and employment, and explore the possible determinants of high school employment. If
the unobservable heterogeneity is related to this observable heterogeneity such as test
scores, local economic conditions and family backgrounds, these relationship between
high school employment and the observable heterogeneity might help us to identify the
underlying relationship between high school employment and later earnings or
employment. While Ruhm (1997) and Light (1999) suggest the negative relationship
between high school employment and the unobservable heterogeneity, Hotz et al (1999)
suggest imply the positive relationship. Our study suggests that there might not be such
clear relationship.

However, the implications of our study should be qualified by its own limitations.

\

$5 The true effects of high school employment might change over time. However, considering the job
Characteristics high school students hold, it is very hard to think about such possibility.

86

First, NLS-72 and HS&B do not provide the amount of working experience in detail, and
thus, our measure of working experience is still not accurate. For an example, working
experience during the junior year in HS&B is very inaccurate because the average
working hours during the year are ambiguous. They could represent the average working
hours while students hold jobs, or the average working hours during the junior year
depending upon how students answer to the question “during the school year before this
one, about how many hours per week on the average did you work for pay outside your
own home?”. The information in NLS-72 is much more ambiguous because working
experience includes both paid and non-paid jobs.

Second, if working experience in the senior year influences test scores, identifying
the selectivity in high school employment in the senior year becomes more complicated.
We do not pursue this issue thoroughly.

However, with these qualifications, our study suggests that when we investigate
the effects of high school working experience, we need to pay attention to how to the
measurement of the high school work experience, what factors determine it, and how they

are related to high school working experience.

87

REFERENCES

88

References

Crawford, David L., Johnson, Amy W., Summers, Anita A. 1997 “Schools and Labor Market Outcomes”,
Economics of Education Review, Vol. 16, No. 3, pp. 255-269

Carr, Rhoda V., Wright, James D., and Brody, Charles J. 1996, “Effects of High School Work Experience a
Decade Later: Evidence from the National Longitudinal Survey”, Sociology of Education 1996.,
Vol.69(.1 anuary):66-8 1 .

D’amico, Ronald. 1984. “Does Employment During high School lmpair Academic Progress?” Sociology of
Education 57; 152-64

Dustmann, Christian. , Rajah, Najma, and Soest, Arthur van, 1996, “Part-time work, School Success, and
School Leaving”, Discussion Papers in Economics, Department of Economics, University College London.

Freeman, Richard B. and Wise, David A. 1982, “The Youth Labor Market Problem: Its Nature, Causes, and
Consequences” pp 1-16 The Youth Labor Market Problem: Its Nature, Causes, and Consequences edited by
Freeman and Wise.

Greenberger, Ellen and Laurence Steinberg. 1986 When Teenagers Work: The Psychological aand Social
costs of Adolescent Employment. New York: Basic Books.

Hausman, J. A., & Taylor, W. E. (1981) “Panel Data and Unobservable Individual Effects”, Econometrica,
Vol.49, No. 6, 1377-1398

Lillydahl, Jane H. “Academic Achievement and Part-Time Employment of High School Students”, Journal
of Economic Education; V21 n3 summer 1990, pp. 307-16

Maddala, G. S., “Limited-dependent and Qualitative Variables in Econometrics “ Econometric Society
Monographs No. 3, Cambridge University Press, 1983.

Marsh, H. W. 1991. “Employment during High School: Character Building or Subversion of Academic
Goals?” Sociology of Education 64: 172-89.

Meyer, Robert H. and David A. Wise. 1982. “High School Preparation and Early Labor Force Experience.”
Pp. 277-347 in The Youth Labor Market Problem: Its Nature, causes and Consequences, edited by Richard
B. F reemand and David A. Wise. Chicago: University of Chicago Press.

Meyer, R. H. and Wise, D. A.(l984) “The transition from school to work: the experiences of black and
whites” Research in Labor Economics 6, 123-176

Mortimer, Jeylan T. and Michael D. Finch. 1986 “The Effects of Part Time Work on Adolescent Self
concept and Achievement” Pp. 66-89 in Becoming A Worker, edited by Kathryn M. Borman and Jane
Reisman. Norwood, NJ: Ablex.

Rees, Albert and Gray, Wayne. 1982 “Family Effects in Youth Employment” pp 453-466 in The Youth
Labor Market Problem: Its Nature, Causes, and Consequences edited by Freeman and Wise.

Ruhm C.(l995) The extent and consequences of high school employment. Journal of Labor Research 16,
293-303

Ruhm(1997): 7. Is High School Employment Consumption or Investment?.
Journal of Labor Economics. Pub Date: October 1997

89

Steel, L. (1991): “Early work experience among white and non-white youths: implications for subsequent
enrollment and employment” Youth and Society 22, 419-447

Steinberg, Laurence, and Dombusch, Sanford M. “Negative correlates of part-time Employment During
Adolescence: Replication and Elaboration.” Developmental Psychology 27 (March 1991): 304-13.

Steinberg, Laurence; Fegley, Suzanne; and Dombusch, Sanford M. “Negative impact of Part-Time work on
Adolescent Adjustment: Evidence from a Longitudinal Study.” Developmental Psychology 29(March
1993): pp. 171-80.

Stern, David, Finkelstein, Neal., Urquiola, Migue1., and Cagampang, Helen., 1997 “What Difference Dose

It Make 1f School and Work are Connected? Evidence on Co-operative Education in the United States”,
Economics of Education Review, Vol. 16, No. 3,

90

High School Work Experience, College Attendance, and the
Choice of College Major

91

Chapter [I
High School Work Experience, College Attendance, and the
Choice of College Major

1. Introduction.

As we observed in Chapter 1, the majority of high school students work during
high school and about 50 percent of students work more than 10 hours in a week. This
work experience during high school has been found to have influential effects on later
employment, earnings, occupation and educational attainment (Ruhm 1997, Carr et a1
1996, Light 1999). This study explores the relationship between high school employment
and an individual’s college major as well as college attendance.

Previous studies indicate that working during high school is related to a higher
occupational status (Ruhm 1995, 1997, and Mortimer and Finch 1986) as well as to
higher earnings later (Ruhm I997, Carr 1996, Meyer and Wise 1982). However, high
school employment seems to be negatively related to post-secondary educational
attainment if there is any relationship between high school employment and post-
secondary education (Carr et a1 1996, Mortimer and Finch 1986.) Even without
controlling for the later educational attainment, high school work experience is still
positively related to higher occupational attainment (Ruhm 1997).56 Why do high school
student workers obtain higher occupational status and higher earnings later with lower
educational attainment? Considering that the jobs which students hold during high school
are mostly low-paying, low-skilled, low quality and private service oriented, and very

casual jobs, we might not expect student workers to enhance their productivity directly

92

through this work experience, and we might not expect such work experience to help
them to obtain high occupational status later.

We suspect that high school work experience influences students’ choices for later
education and training, which, in turn, might enhance their productivity and help them to
obtain higher occupational status. High school years as well as a few years after high
school serve as a very important transition period from school to work or high school to
college. During this period, young people make very important choices such as college
attendance, their college major, or if they decide to go the job market directly,
determining their occupation or job. Choices made during this period, especially about
education or training, can have fundamental impacts on their future. If students work
during high school, this work experience may have effects on these choices.

Exploring the relationship between high school work experience and individuals’
college major choice among college attendees might help understand a possible link
between high school work experience and higher occupational status or higher earnings at
later time because there are large differences across majors in terms of educational
contents, and there are also large differences in the later earnings and occupational status
across degrees according to the different college majors. There are no previous studies
which empirically investigate the relationship between high school work experience and

. . 7
college major choices.5

 

5° Another study also shows a high probability of transition from low-paying job experience to high-paying
jobs for young men (Gritz and MaCurdy 1992.)

57 Even though Greenberg and Steinberg (1986) strongly suggest the possible correlation between the
increase in high school employment and the increase in the portion of college students who major in
business and engineering, they did not analyze such a relationship with actual data.

93

In this study, we also explore the relationship between high school work
experience and college attendance. By doing so, we might explain in part why high
school work experience is positively related to higher later earnings and occupational
status, while it is negatively related to post-secondary educational attainment.

We use two comparable high school senior data sets, the National Longitudinal
Study of the High School Class of 1972 (NLS-72) for 1972 high school seniors and High
School and Beyond (HS&B) for 1980 high school seniors. For each data set we create
two separate samples according to respondents’ gender because major choices differ
greatly between genders. With these data sets, we investigate and compare the effects of
high school work experience on the first college major choice as well as on college
attendance between 1972 and 1980 and between genders.

First, we investigate the relationship between high school employment and
college attendance choices among ‘no college’, two-year colleges, and four-year colleges
by using the multi-nominal logit model (MNL). The dependent variable is students’
choices of college attendance during the first fall semester after their high school
graduation (October 1972 for NLS-72 samples and October 1980 for HS&B samples).
The key independent variables are ‘working hours’ during high school senior year;
‘work,’ which indicates whether students worked or not during their senior year; or
‘heavy work’ which indicates whether students worked 16 or more hours for 1972
samples, and 15 or more hours for 1980 samples. For the HS&B sample, we have two
additional variables which indicate high school work experience: ‘work junior’ which
indicates whether students worked or not during their junior year and ‘working hours

junior.’

94

We can summarize our findings with the MN L model for the college attendance
choice. (1) Work experience, especially heavy work experience is negatively related to
college attendance, especially to four-year college attendance. (2) The negative
relationship between work experience during the senior year and college attendance is
more evident in 1972 than in 1980. However, even in 1980, there is a strong negative
relationship between the work experience during the junior year and college attendance
for both four-year and two-year college. (3) The negative relationship is stronger among
males than females.

In the MNL models for college attendance choice, the coefficients of ‘work’,
‘heavy work’ and ‘working hours’ are in almost all cases negative and significant in the
four-year college attendance. In 1972, ‘work’ is on the average associated with a 3.2
percent fall in the probability of attending four-year college for males only. ‘Heavy work’
is associated with a 3.4 percent fall for males and a 3.0 percent fall for females , in the
probability of attending four-year colleges on average.

However, in 1980, the negative relationship is weaker than in 1972. ‘Heavy work’
is associated with a 2.3 percent fall in the male students’ probability of attending four-
year colleges and a 2.1 percent fall in the female students’ probability of attending four-
year colleges, However, when work experience during the junior year is controlled for,
almost all the significance of negative coefficients of ‘work’ ‘heavy work’ and ‘working
hours’ in the senior year disappear while ‘work junior’ and ‘working hours junior’ are
significantly and negatively related to both two-year and four-year college attendance.
‘Work junior’ is associated with a 1 to 4 percent fall in the probability of attending either

two-year or four-year colleges. One interesting finding is that, in the male case, work

95

experience during in the junior year is more negatively related to two-year college
attendance than to four-year college attendance.

Second, we also utilize the probit model, as many previous studies have used the
binary dependent framework to see the relationship between high school employment and
college attendance (Carr et a1 1996, Meyer and Wise 1982, Marsh 1991.) In this
framework, the dependent variable is equal to one if a student attends either a two-year or
four-year college and otherwise, is equal to zero.

Our main purpose of using the probit model is either to control for or test the
possible endogeneity in the relationship between high school work experience and college
attendance, because students who plan not to go to college could work more during high
school. In the MN L model, testing or controlling for the endogeneity is not easy due to
computational difficulty. With the two stage probit method suggested by Lee (1981), we
control for the endogeneity related to continuous independent variables with some
limitations. We also test endogeneity with the two stage probit model suggested by Rivers
and Vuong (1988).58

The results from the standard probit model confirm the findings of the previous
studies. Working heavily during high school is negatively associated with college
attendance. Except for the female case in 1980, ‘heavy work’ and ‘working hours’ in the
senior year are negative and significant. In 1972, for the male case, ‘work’ and ‘heavy
work’ are respectively associated with a 5 percent and a 7 percent fall in the probability
of attending college at the mean values of other control variables. In the female case, only

the ‘heavy work’ is associated with a 5 percent fall in the probability at the mean values

 

’8 We could also use the MLE method. We do not attempt the MLE method.

96

of other control variables. In 1980, ‘heavy work’ is associated with a 2.5 to 4 percent fall
in the probability of attending college for males, depending on the controlling for work
experience during the junior year while the coefficients of ‘work’ are not significant. In
the female case, without controlling for work experience in the junior year, there is no
such significant relationship.

The results from the two stage probit models are mixed. In 1972, after controlling
for the endogeneity, work experience during the senior year affects college attendance
positively even though endogeneity test shows there is no significant endogeneity with
respect to working hours. However, in 1980, even afler controlling for the endogeneity
with respect to the relationship between ‘working hours’ and college attendance, working
hours during high school affect college attendance negatively and the endogeneity test
also shows that there is no significant endogeneity with respect to working hours.59

For the relationship between high school employment and college major choice,
first we classify all college enrollees by their study areas- academic field and vocational
field, -and apply the standard probit model to this choice. The dependent variable is equal
to one if student is in the vocational field and otherwise, is equal to zero.

Our brief investigation shows that, except for the female case in 1972, after
controlling for academic ability and other background variables, there is no significant

relationship between high school work experience and the choice between vocational

field and academic field in both years. In 1972, female students who worked during their

 

’9 These findings could imply that, in 1972, high school students who decided not go college tend to work
longer, while in 1980, students’ working hours were irrelevant to students’ plans for college attendance. We
explain the reasons later.

97

senior were more likely to choose the vocational field than those who did not work at all,
and the longer they worked, the more likely they are to choose the vocational field.

Second, we restrict our samples to college students in the academic field and
apply the multi-nominal logit model (MNL) to their specific major choices. Maj or choice
during the first fall semester after high school graduation (October 1972 for NLS-72
samples and October 1980 for HS&B samples) is the dependent variable. We designate
six major categories; business, engineering (including computer science and architecture),
health science, pre-professional (pre-law or pre-medicine), science (including
mathematics) and liberal arts (including other majors).

Our estimates indicate that even after different cognitive skills such as verbal and
mathematics skills and courses taken in each subject during high school are controlled
for, high school work experience is strongly associated with students’ major choices.
Students with work experience during their high school senior year are more likely to
choose majors which lead to high-paying jobs later. The longer students work during their
senior year, the more they are likely to choose business or engineering as a major and they
are less likely to choose liberal arts or education. In the MNL model, for the most cases,
the coefficients of ‘work’, ‘heavy work’ and ‘working hours’ during the senior year are
significantly positive in the log odds ratio between the major choices of business or
engineering and the choice of ‘liberal arts and other’ majors.

This association is stronger in 1980 than in 1972, stronger among males than
females, and stronger among Whites than minorities. In 1972, male students who worked
during their senior year have a 4.6 percent higher probability of choosing business and a

2.4 percent lower probability of choosing liberal arts or education than those who did not

98

work at all. Male students with work experience have a 5 percent higher probability of
choosing a business major than others. A ten hour increase in working hours reduces the
probability of choosing liberal arts by a 2.5 percent and increases the probability of
choosing business majors by a 2.4 percent. In the female case, the relationship is much
weaker than in the male case. The coefficient of ‘work’ is not significant in all the major
choices. However, ‘heavy work’ is positively associated with choosing a business major.
‘Heavy work’ increase the probability of choosing business by a 2 percent and decreases
the probability of choosing liberal arts by about a 3 percent.

In 1980, the coefficients of ‘work’, ‘heavy work’ and ‘working hours’ are positive
in the choices of business and/or engineering majors for both males and females. In the
male case, ‘work’ is associated with about a 5 to 8 percent fall in the probability of
choosing a liberal arts major and about a 5 to 7 percent rise in the probability of choosing
a business major.60 ‘Heavy work’ is associated with a 5 to llpercent decrease in the
probability of choosing a liberal arts major, a 6 to 8 percent increase in the probability of
choosing a business major, and a 3 to 10 percent increase in the probability of choosing
an engineering major. A ten hour increase in working hours is associated with a 3 to 5
percent decrease in the probability of choosing a liberal arts major and a 2 to 3 percent
increase in the probability of choosing a business major and a 1 to 5 percent increase in

the probability of choosing an engineering major.

 

”0 This association between working experience during the senior year and students’ tendency to choose
business or engineering majors is much stronger among White students. When we restrict the sample to
White only for the HS&B samples, ‘work’ or ‘heavy work’ are associated with a 7~10 percent increase in
the probability of choosing a business major and a 4~5 percent increase in the probability of choosing an
engineering major while they are associated with a 8~10 percent decrease in the probability of choosing
liberal arts and others majors. Also, we use the weights to accommodate the sampling design in the HS&B
sample, and the associations are almost as strong as when we use the White only sample.

99

In the female case, only the coefficients of ‘heavy work’ and ‘working hours’ are
positive and significant in the choices of business and engineering. ‘Heavy work’ is
associated with a 3 to 4 percent decrease in the probability of choosing a liberal arts major
and a 3 to 4 percent increase in the probability of choosing a business majors and a 2 to
4 percent increase in the in the probability of choosing an engineering major. A ten
working hour increases is associated with about a 1 percent decrease in the probability of
choosing liberal arts major and a .4 to 1 percent increase in the in the probability of
choosing a business major and a 1 to 2 percent increase in the in the probability of
choosing an engineering major.61

However, this association between high school employment and college major
choice does not necessarily mean that work experience influences students’ college major
choice. For example, if students have a strong tendency to work in the labor market, they
might work during high school and choose majors that will facilitate their chances for
work in the future. Controlling for this endogeneity is very difficult in the MNL model.
Thus, once again, to test whether this association comes from such endogeneity, we
utilize the two stage probit models. We regard business and engineering (including
computer science and architecture) as ‘practical majors.’ We also test whether high
school working experience causes students to avoid ‘the non-technical majors’- liberal
arts and education majors- with the same binary dependent framework.

The probit model results confirm the MNL results, showing that high school work

experience is significantly and positively associated with choosing business or

 

6' Another interesting finding is that in the HS&B female case, ‘work junior’ is positively associated with
choosing liberal arts and negatively associated with choosing business and engineering majors, regardless of

100

engineering major and are significantly and negatively associated with choosing a ‘non-
technical major’ such as liberal arts, or education.

The results from the two stage probit models for the HS&B male sample indicate
that this strong association between high school work experience and college major
choice is not caused by a certain characteristics of student workers-such as tendency to
work. After controlling for the endogeneity, the coefficients of working hours are still
positive and significant in the choice of practical major. Also, the endogeneity test show
that there is no such endogeneity in the relationship between high school working hours
in the senior year and ‘practical college major’ choice.

In sum, working during high school is negatively associated with college
attendance, especially four-year college attendance. However, among college enrollees in
the academic field, high school student workers tend to choose business or engineering
majors and avoid liberal arts and education. High school work experience seems to
influence student workers to choose such majors over other majors.

In the next section, we briefly review studies that investigate the effects of high
school work experience on college attendance, and studies that are related to our current
study on the relationship between high school work experience and the choice of college
major. In the same section, we also summarize what our data show without controlling
for other variables. In section 3, the MNL method and the probit models are described. In
section 4, regression results are explained. In section 5, the conclusion of our study is

made and further research on the issue is suggested.

 

holding work experience during the senior year constant. In the male case, there is no such relationship

101

2. High School Work Experience, College Attendance and the Choice of College
Major.
2-1. High School Work Experience and College Attendance.

Many previous studies which investigate the effects of high school employment
on post-secondary education focus on the relationship between high school work
experience and college enrollment (or post secondary educational attainment). Most of
these studies find lower college enrollment rates among high school student workers,
especially among those who have worked heavily during high school. Table 2-2-1-1
summarizes these previous studies.

Using NLS-72, Meyer and Wise (1982) show that, after holding other background
variables constant, the hours worked during the senior year are negatively related to
college enrollment in the first four years after high school graduation. Meyer and Wise
show that those who worked more than 20 hours per week in their senior year are
considerably less likely to be in college than those who did not work at all. Using the
HS&B sophomore data set, Marsh (1991) also shows that the total hours worked during
the senior, junior and sophomore years are negatively related to attending college in 1 to 2
years after high school graduation.62

However, among studies which use the NLSY data set, the results of their

investigations on the relationship between high school employment and later college

 

found between work experience in the junior year and college majors.

6‘ However, Marsh’s measure of hours worked is not accurate as shown in Chapter 1. Also his variables
which indicates college attendance is problematic because he seems to add up the numerical values of the
categorical variable which represents college student status at several points in time after high school. In
that categorical variable, ‘0’, ‘1’ and ‘2’ indicate respectively that the respondent was not a student, was a
part-time student, or was a full-time student at a post-secondary institution at each of four points in time:
February 1982, October 1982, October 1983, February 1984.

102

attendance differ from one another, even though all of them also indicates a negative

relationship between heavy work during high school and post-secondary education. 63

Ruhm (1997) shows that only in the female case, working more than 10 hours per week

during the senior year is negatively related to grade completed and college graduation.

However, Carr et al (1996) show that there is a significant and negative relationship

between high school employment, college attendance and college completion, especially

for males. Steel (1991) shows that when both the job-holdings and hours worked are

entered into the college enrollment equation, job-holding is positively related to future

enrollment, while hours worked is negatively related to future enrollment.‘54

Table 2-2-1-1 Review of Previous Studies on the Effects of High School Work Experience on College

 

 

 

Attendance.

Authors Data, Method, Definitions/Specifications of Key Findings

, Dependent Variable High School Employment

Sources

Meyer (1) NLS-72 Hours worked per week Person who worked more than 20 hours
and (2) Probit model categories: 1-5, 6-10, 1 1-15, per week were considerably less likely to
Wise simultaneously estimated 16-20, 21-25, 26 to 30, 31 or be in college in any of the four years
(1982) with weeks worked (or more. than those who worked less. (10% lower

 

wage) equation.

(3) Nonattendance status
in college during the first
four years. Each year
separately.

 

Hours worked in linear form

 

at the mean value of control variables)
Persons who worked less than 20 hours
per week were also less likely to be in
college during the first year after high
school than those who did not work at
all.

The hours worked per week is negatively
and significantly related to college
attendance.

 

 

(’3 These studies also show the relationship between high school employment and high school graduation
differently. D’Amico (1984) shows that as the percent of weeks that students worked more than 20 hours
increases during the junior year, female white students were more likely to drop out. She also shows that as
the percent of weeks that student worked 1 to 20 hours increases during the junior year, minority female,
white female and white male students were less likely to drop out. The percent of weeks that students
worked more than 20 hours during the sophomore year is negatively related to high school graduation only
for white males. However, Ruhm(1997) shows that senior work experience is positively related to high
school graduation. Carr et a1 (1996) show that there is no such relationship between high school
employment and graduation.
6" Even though their measure of enrollment is not necessarily the enrollment in colleges, the significant and
positive sign of the coefficient of the job-holdings in the enrolhnent regression seems to show that students
who are employed during their senior year have higher academic ability and better family background.

103

 

 

 

 

 

 

 

Marsh (1) HS&B sophomore Total hours worked during the College attendance is negatively related
(1991) data, 1980, 1982, and senior, junior and sophomore to total hours worked.
1984 interviews. year.
(2) OLS for ‘university’ ‘ Linear form.
DAmic (l)NLSY-1979-82 The percent weeks worked The percent weeks worked more than 20
0 interviews; high school more than 20 hours/week and hours during the junior year is positively
(1984) students (in 1979): White 1-20 hours/week during the related to drop-out only for female
male and minority male, sophomore year and/or the Whites.
White female and junior year.; linear forms and The percent weeks worked 1-20 hours
minority female are each year experience during the junior year is negatively
separately sampled. exclusively. related to drop-out for minority females,
(2) Probit model Average hours worked per White females and white males.
(3) Drop-out: those whose week during the high school The percent weeks worked more than 20
highest grade completed year (junior year or sophomore hours during the sophomore year
as of 1982 was less than year); increased the probability of drop-out
1 1 years and who had not Quadratic form, each year only for white males.
attended school within the experience exclusively. Slight non-linear relationship
last year (significance problem)
Steel (1) NLSY, 1979-81 Both employment status and For Whites, job-holdings increases
(1991) interviews, l7-18-year hours worked per week in 1979 future enrollment while hours worked is
olds in 1979. Whites, regardless of grades are entered negatively related to the future
Hispanics and Blacks are together. Thus, employment enrollment rate.
separate. status and hours worked can
(2) Tobit model represent the work experience
(3) Months of enrollment even in college.
between 1980 and 1981
Ruhm (1) NLSY, those who Hours worked during the Senior work experience is positively
(1997) were high school reference week during the associated with high school graduation
freshmen or sophomores sophomore year, the junior year rates, while there is no statistically
in 1979, remained and the senior year are significant relationship between work
enrolled in school through separately entered and are experience either during the junior year
their senior year. 588 specified in the quadratic form. or sophomore year.
males and 561 females In the cases of the highest For females, those who worked more
(2)Probit model and OLS grade completed and the than 10 hours during the senior year
(3) the probability of high probability of finishing 4 or have completed less grades and lower
school graduation and the more years of college, only the college graduation rates.
probability of finishing 4 hours worked during the senior For males, there is no such relationship.
or more years of college year are entered.
by 1991; the highest
grade completed in 1991.
Carr, (1) NLSY. 16-19 in Hours worked in 1978 divided Hours worked is negatively related to
Wright 1979, enrolled in high by 100. the highest grade completed, especially
and school as of May 1, 1979 Linear form of hours worked. for males. 100 hours working in 1978
Brody (2) Logistic regression for reduced around .03~.04 years of
(1996) the probability of high schooling by 1991.

 

school graduation, 1-3
years of college, 4 or
more years of college;
OLS for the highest grade
completed by 1991.

 

 

No relationship between hours worked
and the probability of high school
graduation.

Significant and negative relationship
between hours worked and college
attendance, and college completion after
controlling for other background
variables, particularly for males,.

 

* Marsh’s definition of ‘university’ is sum of activity variables indicating that the respondent was not a
student (0), was a part-time student (1), or was a full-time student (2) at a post-secondary institution at each
of four points in time: February 1982, October 1982, October 1983, and February 1984.

104

 

These differences in their results seem to be related to the different measurement
of high school work experience and different specification of high school work
experience in the regression analyses. Ruhm measures hours worked during the reference
week in the sophomore, junior and senior year separately as high school work experience
and puts them separately in the quadratic form in the regression analyses, while Carr et a1
measure high school work experience as hours worked in 1978 and put it in a linear form
in the regression analyses. Steel measures the average hours worked per week in 1979 as
high school experience, and a dummy for the job-holding status and hours worked
together as independent variables in the regression analyses.

The general negative relationship between high school employment and college
enrollment is usually explained by the assumption that working takes away students’ time
and efforts from academic activities, and thus, hindering students from attending colleges.
However, there is a possibility that students who plan not to go to college work longer
and earlier during high school.

Using the NLS-72 and HS&B data sets, we briefly reinvestigate the relationship
between high school employment and college attendance because, decisions about college
attendance might be closely related to decisions on the major choices, which are our main
focus in this study. When we investigate how high school employment is associated with
college attendance, we distinguish four-year college attendance from two-year college.
All the previous studies on this issue regard these two different types of post-secondary
institutions as the same. However, there are distinctive characteristics of two-year and
four-year college. Most two-year college majors and courses are skill-oriented and

technical. Also, tuition and academic criteria for admission are different from one

105

another. There is a likely possibility that high school work experience might influence
high school seniors’ decision about the college type as well as college enrollment because
the work experience might change students’ views on college education and might
influence academic preparation for college.65

We tabulate the relationship between working hours during the high school senior
year and college attendance (choices) in the Table 2-2-1-2 and 3 and Table 2-2-1-4 and 5.
The tables also show some features of the relationship between high school work
experience and college attendance. First, in 1972, students who worked heavily during
their senior year were much less likely to attend college than those who worked

moderately or those who did not work at all (see the first rows of tables).

 

 

 

 

 

 

Table 2-2-1-2: Working Hours and First College Choices : Male (NLS-72)
None less 6-10 1 1-15 16-20 21-25 26-30 More Total Total
than 6 hours hours hours hours hours than 30 (%) #r
____________ _ h99§-..-_______-- _-___-_-_-_--.____ “whiten“... _.___--._ .____._--
No College
(%) 49.3 47.3 53.6 52.7 57.1 61.3 64.9 75.7 57.9 3915
Two-Year
College (%) 12.4 12.6 14.3 12.5 13.0 13.8 13.3 8.8 12.4 840
Four-Year
College (%) 38.3 40.1 32.1 34.8 30.0 24.9 21.8 15.5 29.7 2008
Total Number 1245 730 797 609 915 804 607 1056 6763
Table 2-2-1-3: Working Hours and First College Choices :Female (NLS-72)
None less 6-10 1 1-1 5 16-20 21 -25 26-30 More Total Total
than 6 hours hours hours hours hours than 30 % #
hours hours
No College (%)
60.3 53.0 56.4 57.6 59.9 67.7 70.8 75.1 60.6 4248
Two-Year Collge
(%) 10.9 13.2 11.5 11.7 12.6 11.2 9.6 7.9 11.4 796
Four-Year
College (%) 28.8 33.8 32.1 30.6 27.6 21.2 19.6 17.1 28.1 1969
Total Number 2126 920 931 741 900 609 353 433 7013

 

 

6’ The two-year versus four-year college enrollment decision itself has been a topic in the economics of
education (see Rouse 1994, Alexander and Pallas 1987).

106

Table2-2-1-4: Working Hours and First College Choices: Male (HS&B)

 

 

 

 

 

 

 

None 1-4 5-14 1 5-21 22-29 30-34 More Total Total
hours hours hours hours hours than 35 % fl
_- -9996
No College (%)
49.1 47.7 40.1 44.7 53.8 56.4 59.9 49.0 1931
Two-Year Collge
(%) 16.9 16.2 18.8 17.3 18.3 19.1 15.7 17.5 688
Four-Year
College (%) 34.0 36.2 41.1 38.0 27.9 24.5 24.4 33.5 1322
Total Number 374 235 608 779 552 273 394 3941
Table 2-2-1-5: Working Hours and First College Choices: Female (HS&B)
None 1-4 5-14 15-21 22-29 30-34 More Total Total
hours hours hours hours hours than 35 % #
_-_-__ 119.96
No College (%)
50.5 44.9 38.3 40.7 48.8 51.5 56.4 46.1 2225
Two-Year Collge
(%) 16.9 17.0 17.2 19.9 17.3 18.0 22.3 17.9 867
Four-Year
College (%) 32.6 38.1 44.5 39.5 33.9 30.5 21.4 36.0 1739
Total Number 1613 430 767 1052 549 200 220 4831

 

Further, the relationship is almost linear except for the first two time categories.66 Since
working occupies students’ time and effort, and may hinder students’ academic
achievement during high school, this relationship is not surprising. However, this result
might occur because students who have decided not to go to college work longer hours
while in high school.

Second, we find that this relationship between work experience and college
attendance is much weaker in 1980 than 1972, especially for females. The probability that

a female student who worked 30-34 hours per week during her high school senior year

 

6" In 1972, students who had no work experience were less likely to attend college than those who had
worked less than 6 hours among males and those who have worked 1-10 hours among females. However, as
we note in the chapter one, those who had no work experience during their high school senior year did not
worked because jobs were not available for them because of either poor ability or market conditions.

107

would attend college is almost as high as that of a female student who did not work at all.
In 1980, as we see in chapter 1, more affluent and more academically able students
participate in the labor market than in 1972. This weakened negative relationship could
have resulted from a negative/positive relationship between working hours and ability, or
the negative/positive relationship between working hours and families’ socio-economic
status.

Third, there remains a strong negative correlation between four year college
attendance and working hours. At the same time, there is no such relationship between
working hours and two-year college attendance. It is possible that high school students
who have worked might choose two-year colleges because many two-year college major
fields and courses are skill-oriented and technical.

We also can tell that many of those who have worked during high school go to
two-year or four-year colleges. Even among students who have worked more than 30
hours per week, more than 25 percent in 1972 and more than 40 percent in 1980 went to
college right after high school graduation. Thus, many college students have some work
experience during high school. This work experience could have some relationship with

the courses they take and their majors during their first college years.

108

2-2. High School Work Experience and the Choice of College Major.

Students’ college majors serve as one of the determining factors of their future
income, occupation, and even future social lives. Previous studies show that college
students in general are rational and consider their expected future earnings as one
criterion when they choose their college majors (Freeman 1976, Berger 1988).67 Grogger
and Bide (1995), using the same data sets we use in this study, NLS-72 and HS&B,
estimate major-specific wage premiums of college graduates relative to earnings of high
school graduates. They show that among male college graduates, males who have
majored either in business or engineering have the highest returns to their college degrees
in 1978, and afterwards, they enjoy the highest earnings grth rates as a function of time
and (or) experience relative to earnings of high school graduates.

For female college graduates, science majors, business majors, and engineering
majors have the highest earnings. Science majors, business majors, and engineering
majors have earned respectively 32, 21, and 17 percent more than high school graduates
in 1978, while social science majors, education majors, arts as well as other majors
respectively earned, 12, 10, and 12 percent more than high school graduates in the same
year. Afterwards, the growth rates of wage premium as a function of time and (or)
experience are highest among business and engineering majors.

Other studies also show that individuals choosing more practical and skill oriented

fields such as business, engineering or computer science enjoy significant wage premium

 

‘57 However, using the NLS-72 data set, Eide and Waehrer (1998) find that in the female case, the expected
wage is a negative determinant for female undergraduate major choice, and the option value of majors
related to later graduate school attendance is a positive determinant in major field choice.

109

over those majoring in less technical fields such as liberal arts and education.(Altonji
1993, Grubb 1993, and Rumberger and Thomas 1993.)

The question is whether work experience during high school is related to college
major choice. In particular, our question is whether high school student workers have a
stronger tendency than non-workers to choose business or engineering majors. A recent
study (Light 1999) shows that high school student workers among non-college bound
high school graduates tend to take more vocational courses such as type writing,
automobile mechanics, accounting, and health related courses during high school than
non-workers. This study also shows that these practical courses are related to higher later
earnings. If work experience makes non-college bound high school student workers more
sensitive to future earnings and more practical in choosing high school courses, it would
make college-bound students more sensitive to future earnings and more practical in
choosing their college courses. College courses that students take are bounded to a certain
degree by the choice of their major field.

If students’ work experience is associated with their taking more practical courses
or education, why and how does it makes students more practical? There are several
Possible explanations for this phenomenon. First, Greenberg and Steinberg (1986) clearly
link high school employment to an increase in college students’ enrollment in ‘lucrative’

maj Ors. They argue that working experience makes students more materialistic and thus,
college students who have worked during high school tend to choose engineering,

Conlputer science and businesse. Greenberg and Steinberg cite this tendency as one of the

negative outcomes of high school employment.

110

During the same period, college freshmen expressed declined interest in the arts,
humanities, and social sciences and increased interest in majors such as engineering, computer
science, and business that are likely to lead to more lucrative careers. (The number of students
entering college who expected to major in business increased by 80 percent between 1966 and
1981. ) In this cultural milieu, it is not surprising that high-school students have flocked to the
workplace: in the learning for its own sake or to careers involving service to others, the lure of the

paycheck must be very powerful(P 34.)

Second, another possible link between high school employment and practical
college major choice might be related to the accumulated knowledge through work
experience. If early working experiences increase the knowledge of the world of work
(D’Amico 1984) or greater “business knowledge” (Greenberger et a1 1982) and thus,
student workers know better about what occupations or jobs have more market value,
they might choose those occupations or jobs which have more market value. By the same
token, if they know better which kinds of skills or education are highly valued in the
market, they might choose more practical courses and might major in fields which are
more rewarding in the labor market such as business, engineering, computer sciences or
health related sciences.

Third, characteristics of jobs students hold during high school are low-skilled and
very casual (so called ‘dead-end’ jobs.) The knowledge of the world of work through such
job experience itself might be too limited to help students in choosing such practical
college majors. However, work experience in such dead-end jobs might help them
understand the hard reality of the world of work and make them not want such kinds of
jobs as adults. Thus, when student workers decide to enroll in college, they might choose

those majors which generally guarantee well-paid and highly skilled jobs.

111

All these possible links are related to one another in the sense that they are based
on the same assumption that high school work experience might affect an individual’s
choice of a college major by influencing his or her perceived firture earnings, occupation,
or job. Thus, it is possible that all these links from work experience to college major
choices work together in their influence upon student workers’ major choice.

However, the relationship between high school work experience and college
major choice might not be causal if students with a certain unobservable characteristic
work more during high school and choose such practical majors in colleges. That is,
individuals who like to work in the labor market might work more during high school and
choose majors that will facilitate their chances for work in the firture. Instead that work
experience makes high school student workers practical, so students who are already
practical work during high school and choose such practical majors in colleges.

There are no previous studies which examine the effects of high school
employment on the choice of college major, or the association between these two, using
empirical data. However, this question is important if high school work experience
influences the choice of college major. (i) As mentioned earlier, the college major can
affect future jobs, occupations, earnings, social status, and even life styles in the long run.
If we look at the effects of high school employment upon only the short term earnings,68
it might be misleading and we might ignore other long-term consequences of high school
employment on students’ future lives. (ii) We can understand one more channel in the

link between high school employment and higher later earnings or higher later

 

6" Most studies on the effects of high school employment on later earnings have focused on earnings gain in
the relatively short term, usually 2-10 years after high school graduation. Even 10 years after high school
graduation is short for the college bound students.

112

occupational status at least for college bound students.69 (iii) Thus, this research has some
implications for government programs involving the school-to-work transition, which
mostly aim to increase early job market stability of the young.

First, using the samples of those students who attended college in the fall semester
after high school graduation (Oct. 1972 and Oct. 1980), we classify college enrollees by
two study areas: vocational field and academic field.70 We tabulate the relationship
between working hours during the senior year and two study areas in Table 2-2-2-1 and 2,
and Table 2-2-2-3 and 4. From the tables, we can note several features of the relationship.
In 1972, many more female students seemed to take more the vocational field than males,
while in 1980 there is no such difference based on gender. Among college enrollees, the
percentage of women who enrolled in the vocational field decreases, while the percentage
of men who enrolled in vocation field increases over-time. Above all, students who work
longer hours are more likely to choose the vocational field and less likely to choose the

academic field.

 

69 Many studies which find positive effects of high school employment later earnings do not provide explicit
explanations for this positive effects. For example, finding the significant positive effects of senior year
working hours on later earnings, Ruhm (1987) argues that his research indicates that “student employment
raises future productivity through the skills, knowledge, work habits, and experience provided on-the-job by
far more than it detracts form educational human capital investments” (p770). However, he does not
provide any evidence for that channel. Two recent studies, Hotz et a1 (1998) and Light (1999) provide
some explanations for these positive effects. Hotz et a1 (1998) point to individual heterogeneity in the labor
market and schooling experience as the reason for the effects while Light suggests the correlation between
high school employment and vocational courses taken by students workers as one reason for the effects
among non-college bound students.

7° The National Center for Education Statistics provides field of study codes as a classification of
instructional programs. They classified three major fields: (1) conventional academic subdivisions of
knowledge training, (2) technological and occupational curriculums leading to associate degrees and other
awards below the baccalaureates, and (3) vocational programs. In NLS-72 samples, only 235 among 2334
male college students and 427 among 2262 female students chose vocational majors. In HS&B samples,
252 among 2010 male students and 299 among 2606 female students chose vocation major and 56 male and
106 female student chose technical majors according to NLS-72 and HS&B survey classification. We

113

Table 2-2-2-1: Working Hours and College Study Field: Male (NLS-72)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

None less 6-10 1 1-1 5 16-20 21 -25 26-30 More Total Total
than 6 hours hours hours hours hours than 30 (%) It
_ .. . - . _ _ u... - __Mwheyrs_..___--_-_-_m-._..._ __ _____- hours
Vocational Field
(%) 8.5 8.7 10.7 12.3 12.7 12.7 13.1 10.5 10.8 287
Academic Field
(%) 91.5 91.3 89.3 87.7 87.3 87.3 86.9 89.5 89.3 2376
Total Number 575 367 346 277 371 291 198 238 2663
Table 2-2-2-2: Working Hours and College Study Field: Female (NLS-72)
None less 6-10 1 1-15 16-20 21-25 26-30 More Total Total
than 6 hours hours hours hours hours than 30 (%) #r
__ m--- ____~_____‘ hours __ hours
Vocational Field
(%) 16.7 16.4 21.8 21.2 21.0 24.6 21.0 23.5 19.5 516
Academic Field
(%) 83.3 83.6 78.2 78.8 79.0 75.4 79.0 76.5 80.5 2132
Total Number 801 414 390 302 348 191 100 102 2684
Table 2-2-2-3: Working Hours and First College Choices: Male (HS&B)
None 1-4 5-14 15-21 22-29 30-34 More Total Total
hours hours hours hours hours than 35 % fl
__ _ __ ______ hours
Vocational Field
(%) 15.8 16.0 13.4 14.3 20.1 23.2 26.4 16.9 308
Academic Field
(%) 84.2 84.0 86.7 85.7 79.9 76.8 73.6 83.1 1520
Total Number 514 89 337 384 244 99 144 1828
Table 2-2-2-4: Working Hours and First College C holces: Female (HS&B)
None 1-4 5-14 1 5-2 1 22-29 30-34 More Total Total
'__ hours hour‘s” hours __hours hours than 35 "/o if
Vocational Field
(%) 19.5 13.6 13.4 15.6 19.2 15.6 23.4 16.9 405
Academic Field
(%) 80.5 86.4 86.6 84.4 80.8 84.4 76.6 83.1 1986
Total Number 738 213 432 569 255 90 94 2391

 

However, the characteristics of each major within the academic field are very

different from one another.71 We further restrict the sample to those who enrolled in the

 

include vocational majors and technical majors in the vocational field because these majors are basically

vocation-oriented majors.
7' Of course, the characteristics of each major within the vocational field are different from one another.
However, the numbers of college students who are in the vocational field in our sample are relatively small

114

academic fields which usually lead to the bachelor’s degree. Students’ first chosen
college majors are classified into seven big categories; business, liberal arts (including
social science), education, engineering (including architecture and computer sciences),
health related, pre-professional (pre-law and pre-medicine), and other majors (military
science, library, ethnic studies and others).

There are some problems with using the first major choices. First, many college
students, especially two-year college students, do not finish their degrees. If college
credits or majors without degrees are not valued by the market, investigating the
relationship between high school employment and the first major choices is not so
meaningfirl. However, Kane and Rouse (1994) show that the “sheepskin” (degree) effect
is small and the returns to credits without degrees, even for two-year college credits, are
high regardless of degrees.72 Second, students change their majors during their college
years. Thus, we examine the relationship between college majors four years afier high
school graduation and working hours, and our results are essentially the same.

There are some advantages to using the first college major in an investigation of
the relationship between high school employment and the choice of college major. First,
we are able to detect the direct effects of high school working experience on the choice of
college major. If we use later college majors, we need to control for many other factors
which could influence students’ major choice during their college years such as,
employment and academic achievement during college. We do not need to worry about

those factors if we use the first major choices. Secondly, since many college students,

 

and a narrower classification within the vocational field is not easy. We did not attempt to classify the
vocational field further.

115

especially those who have worked a lot during high school, do not finish their degrees,
using only major choices with degrees cannot capture the full effects of high school work
experience on the choice of major.

The relationship between first major choices among college students in the
academic field and working hours during the high school senior year are tabulated in the
Table 2-2-2-5 and 6 and Table 2-2-2-7 and 8. First, in both years, regardless of working
hours, male students are more likely than female students to choose more practical and
lucrative majors, such as business and engineering majors. Second, students tend to
choose more such practical majors in 1980 than in 1972 regardless of gender and working
hours.

Third, in general students who work long hours tend to choose more such
practical majors. Table 2-2-2-5 clearly shows that in 1972, male students who work long
hours are more likely to choose business as a major and are less likely to choose liberal
arts than those who work less. While 27.9 percent of students who work more than 30
hours choose business majors, only 14.7 percent of students who do not work at all
choose business majors. Only 13.3 percent of students who work the longest hours choose
liberal arts while 22 percent of students who have not worked at all choose liberal arts.

The female case is strikingly different from the male case. As we see in Table 2-2-
2-6 ,in 1972, only less than 20 percent of female students choose business, engineering
and health sciences while the majority choose education, liberal arts or other majors. It
might be because business women or female engineers were not socially accepted as a

standard female job or an occupation at that time and thus, many jobs in business or

 

72 We also find that credits in courses such as business, engineering, and science have much higher returns

116

engineering areas were not available for females. In the early 19708, teaching may have
been regarded as one typical female college graduate job, and many teaching positions
may have been available for female college graduates. In sum, in the 1972 female case,
there is no distinctive relationship between high school working and the choice of college
major while changing dramatically in the 1980s.

In 1980, a much higher percentage of both male and female students choose
business and engineering majors than in 1972. In the Table 2-2-2-8, male students who
work long hours tend to choose either business, or engineering. About 53 percent, 62
percent, and 59 percent of students in each of the three longest working hour categories
choose either a business or engineering major respectively while 49 percent, 49 percent,
and 42 percent of students in each of the three shortest working hour categories choose
such majors. Also, the last two bottom rows show that students who work longer hours
are less likely to choose liberal arts or other majors. In 1980, female students who work
long hours during high school tend to choose business and engineering majors and are
less likely to choose liberal arts and other majors while these tendencies are less clear
than for males.

In sum, students who work long hours tend to choose more practical majors,
especially business or engineering, and they avoid education, liberal arts, and other
majors, which are less lucrative or less practical.

However, in the health, pre-professional, and science major areas no such clear

relationship between working hours during the high school senior year and major choices

 

than other credits regardless of degrees obtained even after the sample selection is controlled.

117

is shown. However, since the percentages of students who choose these majors are
relatively small, the importance of these majors for the current study is relatively small.

However, working hours during high school, college attendance, and college
major choice are related to academic ability, family backgrounds, and other factors. To
see how these relationships are after controlling for all these factors, we utilize two

econometric methods: the MNL model and the probit model.

 

 

Table 2-2-2-5: Working Hours and Major Choices (%) : Male (NLS-72)
None less 6-10 1 1-1 5 16-20 21-25 26-30 More Total Total
than 6 hours hours hours hours hours than # %
hours 30
hours ___ _. .-
Business 15.2 16.1 19.4 24.7 20.1 23.2 22.7 26.3 473 19.9
Engineering 9.5 8.4 11.0 8.2 8.3 11.8 8.1 8.0 220 9.3
Health 10.1 12.5 11.0 10.3 12.7 11.0 15.1 7.5 265 11.2
Pre- 3.8 2.1 5.2 1.2 3.7 3.5 2.3 4.7 81 3.4
professional
Science 10.8 12.2 11.3 11.9 9.6 9.8 10.5 8.0 253 10.7
Education 13.9 16.1 11.3 10.3 13.2 12.2 11.1 11.3 304 12.8

Liberal Arts 21.7 19.4 16.2 18.9 17.9 13.0 17.4 15.0 428 18.0

Other 15.0 14.6 14.4 14.5 15.4 12.8 19.3 19.3 352 14.8
Major

Total 526 33 5 309 243 324 254 172 213 23 76
Number

 

Notes for Tables 2-2-2-5 : (1) This classification is made for those who are majoring in academic fields which usually
lead to four year college degree. (2) Unit is % (percentage) except the total numbers. (3) Science includes Mathematics.
(4) Engineering includes Architecture and Computer Sciences.

 

 

Table 2-2-2-6: Working Hours and Major Choices (%) : Female (NLS-72)
None less 6-10 1 1-15 16-20 21-25 26-30 More Total Total
than 6 hours hours hours hours hours than # %
30
_._ _-,___--..._...-__ _---.__.- _---.-.-_-_.....-._-~v._._--._._~_.___.. ..-...--._.._.._. -.._--_..-.. W-.- -_..,.-_.__.__- -..-_.-,., ....__‘_...-.__.__.-_.--_,._. hours -.- h.

Business 8.4 7.5 5.9 5.9 9.5 4.9 15.2 12.8 169 7.9
Engineering 1.5 1.7 1.3 1.7 .7 1.4 .0 1.3 487 1.4
Health 8.9 11.9 10.2 10.1 9.5 9.7 8.9 14.1 641 10.0

Pre- 1.2 2.6 1.3 .8 1.5 .7 .0. 1.3 158 10.8

professional

118

 

 

 

 

 

 

 

 

 

Science 8.4 7.5 6.9 10.5 4.7 8.3 3.8 2.6 29 7.4

Education 30.3 30.9 27.5 24.8 33.5 35.4 31.7 26.9 213 30.1

Liberal Arts 22.0 22.5 25.3 25.2 20.4 22.2 22.8 24.4 29 22.8

Other 19.3 15.3 21.6 21.0 20.4 17.4 17.7 16.7 406 19.1

Major

Total 667 346 305 238 275 144 79 78 2132

Number

See notes in Table 2-2-2-5

Table 2-2-2-7: Working Hours and Major Choices (%) : Male (HS&B)
None 1-4 5-14 15-21 22-29 30-34 More Total Total

hours hours hours hours hours than 30 # %

_. __ -... ._......._..4...-..... .» -.....- hours ...- ~.
Business 21.0 19.1 21.6 27.4 28.2 39.5 26.4 374 24.6
Engineering 27.7 30.3 20.6 23.7 30.8 22.4 26.4 391 25.7
Health 3.7 .00 3.8 3.3 3.1 .00 3.8 48 3.2
Pre- .7 .00 1.0 0.3 3.1 2.6 1.9 17 1.1
professional
Science 11.1 9.0 10.3 9.7 3.6 6.6 8.5 139 9.1
Education 6.0 10.1 9.9 7.0 4.1 5.3 7.6 107 7.0
Liberal Arts 16.4 19.1 15.1 15.5 11.3 10.5 12.3 226 14.9
Other Major 13.4 12.4 17.5 13.1 15.9 13.2 13.2 218 14.3
Total 433 89 292 329 195 76 106 1520
Number
See notes in Table 2-2-2-5
Table 2-2-2-8 : Working Hours and Major Choices (%): Female (HS&B)

None 1-4 5-14 15-21 22-29 30-34 More Total Total
hours hours hours hours hours than 30 # %
5 -..--__.__~_ _ hours
Business 30.0 24.5 23.0 32.3 32.5 29.0 33.3 577 29.1

Engineering 8.6 8.2 6.2 8.8 7.3 10.5 15.3 165 8.3

Health 7.7 7.6 10.7 7.5 5.3 2.6 1.4 150 7.6

Prc- 1.9 1.1 2.7 1.7 2.4 1.3 1.4 38 1.9

Professional

Science 7.1 8.7 6.2 5.2 3.9 11.8 9.7 130 6.6

Education 13.0 15.8 13.9 12.3 17.0 13.2 11.1 270 13.6

1 l9

Liberal Arts 16.0 14.1 19.3 16.7 15.5 21.1 13.9 331 16.7

Other Major 15.8 20.1 18.2 15.6 16.0 10.5 13.9 325 16.4
Total 594 184 374 480 206 76 72 1986
Number

 

See notes in Table 2-2-2-5

3. Econometric Models.

We use the MNL model and probit models to see how high school work
experience is related to college major choices after we hold other factors constant. Studies
which investigate the determinants of college major choice indicate that expected
earnings, expected labor force participation and different kinds of cognitive skills are
important in determining major choice (Berger 1988, Polachek 1978, Blakemore and Low
1984, Paglin and Rufolo 1990). However, as mentioned in the foregoing section, early
labor market experience might change an individual’s perception of future earnings and
job amenities or labor force participation, which in turn affects major choice. Using the
MNL model and probit models, we also examine the relationship between high school
work experience and college attendance choice. Mainly for convenience, we mention the
models with respect to only the choice of college major when we describe the models.
Only when it is necessary to mention the models with respect to college attendance

choices, we do so.

120

 

3.1 MNL model.
We assume that individuals choose a major in order to maximize their utility. Let

U ,1 be the level of utility of individual i receives when choosing major j . The level of

utility depends on individual characteristics, X , , like family background, different skills,

tastes, community type or regions where student live, and early working experience.
Attributes of the college majors, such as an individual’s expected future earnings in a
particular major, or expected jobs, could be also important determinants for major choice
as shown in the previous studies. However, we ignore the attributes of a major’s specific
characteristics, such as its expected earnings because we are mainly interested in the
effects of an individual’s early work experience on his or her choice of college major.
Thus, we use a MNL model as a special case of McFadden’s conditional logit model
(CLM) which includes both an individual’s characteristics and choices’ attributes.73

We are particularly interested in whether high school work experience positively
affects the probability of choosing business or engineering majors and negatively affects
the probability of choosing liberal arts and education majors. Previous studies on the
relative earnings associated with different college majors indicate that individuals
choosing engineering or business management majors enjoy significant earnings
premiums over those majoring in education or the liberal arts (Altonji 1993, Berger 1988,

Grogger and Bide 1995, Grubb 1993, Rumberger and Thomas 1993).74 If individuals with

 

’3 There are some studies which use this kind of MNL model. For example, Schmidt and Strauss (1975), in
their investigations on occupational choices, use only individuals’ characteristics such as education,
experience, race, and sex,. Omitting choices’ attributes might cause some unknown problems. However, the
MN L model might be very appropriate in our current investigation because we are mainly concerned with
how an individual’s particular characteristic-high school work experience affects choosing a certain major.
7’ These connections between college majors and later earnings are pretty clear. However, the effects of
expected earnings are not always positive. Even though in general, economic theory predicts the positive

121

high school work experience choose majors with high earnings premiums, the gain in
later earnings of student workers can be explained through such college major choice for
college bound individuals.

The main dependent variable is the students’ choice of major for the first fall
semester after their high school graduation (October 1972 for NLS-72 samples and
October 1980 for HS&B samples). As mentioned earlier, we create six college major
categories for college students who choose academic areas: (i) business, (ii) engineering
(including computer science and architecture), (iii) health science, (iv) pre-professional
(pre-law or pre-medicine), (v) science (including mathematics), and (vi) liberal arts
(including other majors).

By using the MN L model, we also investigate the relationship between high
school work experience and college attendance choice. We create three ‘college
attendance’ categories for the full sample: (i) no college, (ii) two-year college, and (iii)
four-year college.

The level of individual utility from choosing major j depends on these observed
characteristics and on unobserved heterogeneity as follows:

U” =6,le +8” (3-1)

The probability that an individual makes choice j is given by:

pr0b(U,j > U”) for all other k ¢j.

 

correlation between major choice and expected earnings, empirical studies show contrasting results. While
Berger (1988) shows the significant and positive effects of expected earnings in each major field on major
choices for males, Eide and Waehrer (1998) show the negative relationship between major choices and
expected wages for females. This contrasting result implies that not only the expected earnings but also how
individuals value the expected earnings are important in the major choice because the general difference of
expected earnings across majors is well known and ability and family backgrounds are controlled for,

122

Let Y, be a random variable indicating the choice made. We assume that the J

disturbances, a are independent and identically distributed from a Weibull distribution

1] 9
(See Green 1997). Accordingly, we can write the probability that individual chooses
major j, as:

4X,
. e
proth. = r) = _‘W'Ze" W)

J
In order to estimate the parameters of the equation, we normalize the parameters

to be equal to zero ( ,60 = 0) for a chosen major, Y = 0. We regard liberal arts and ‘other’

majors together as the comparison group (Y = 0) in the specific major choice for college
students in the academic field and ‘no college’ status as the comparison (Y = 0) in the

choice of college attendance. Then, the probability of choosing major j becomes

m.
prob(Y=j)=——ej—— for j=1,2, ....... j. (33)

The relative probability of choosing j alternative to the probability of

choosing the base (or comparison) major ‘ O’ is

prob(Y=j) / pr0b(Y= 0) = e’B’X’ (3-4)

 

differences in expected earnings for choosing each major across individuals is relatively small or not
important (see Berger 1988).

123

By taking log both sides of the above equation, we can see the relationship of high school
work experience to the odds ratio between a certain choice of college major compared
and the choice of liberal arts and others. In this MNL model, we can only observe the
coefficients of the independent variables for j-l alternatives.

As mentioned in the introduction, the key independent variables of our current
interest are ‘work’, ‘heavy work,’ ‘working hours’ per week during the senior year. In the
HS&B, we also have ‘work junior’ and ‘working hours junior’ per week. We also use the
average working hours per week in the senior and junior years. We expect the signs of
these variables to be negative in the choices of two-year college attendance and four-year
college attendance. Also, we expect the signs of these variables to be positive in the
choices of business or engineering majors.

The coefficients of these variables show the relative effects of working experience
on the choice of a certain major compared to majors in the liberal arts and ‘others’. For
college attendance choice, they Show the relative effects of working experience on the
choice of two-year or four-year colleges. We expect the positive Sign of these work
experience variables in more practical major choices, such as business or engineering and
the negative sign of them in the choice of two—year or four-year colleges.

However, we have several problems with using this MNL model for our
investigation of the effects of high school work experience on the choice of college
major. First, the MNL model has the property known as the independence of irrelevant
alternatives: the odds ratios (3-4) in the MNL model are the same irrespective of the total
number of choices considered (IIA property). This assumption follows from the initial

assumption that the disturbances are independent.

124

Due to the IIA property, using the MN L model might not be appropriate in
investigating the choice of college major. This is because if two majors, for example
business and engineering, can be perceived as too close or too similar to be an
independent alternative by students, this model predicts too high of a joint probability for
the selection for two such majors. This problem might be serious in the investigation of
the choice of college major for those students in academic areas because some of the
seven college majors might be too Similar to be distinguishable.

To address part of this problem, we test the hypothesis that a subset of major

 

choice truly is irrelevant by a Hausman’s specification test developed by Hausman and
McFadden (1984). If the hypothesis is correct, excluding such major choices from the
model will not change the parameter estimates systematically and the parameter estimates
will be still consistent yet inefficient. If it is not correct, the parameter estimates obtained

without these choices will be inconsistent. Test statistics are

2‘ =(/§.—fl} WA.— VA, TWA—4,) <3-5)

where 5 indicates the estimators after the subset major choices are excluded, f indicates

A

the estimator using the full set of major choice, and V and V] are the respective

estimates of the asymptotic covariance matrices. The test statistics are asymptotically
distributed as chi-squared with the degree of independent variables. We report these test

statistics for each major choice when we omit each of college majors.

125

Second, so far we assume that high school work experience influences the choice
of college major or the college attendance decision. However, parameter estimates could
be biased due to the possible endogeneity problems when there is correlation between
error term and high school work experience variables. In the college attendance case,
students who decide not to go to college can work more during high school. In the major
choice case, as mentioned earlier, if individuals who like to work in the labor market
work more in high school and choose majors that will facilitate their chances for work in
the future, then the relationship between high school work experience and college majors
is not causal. Thus, we have an endogeneity problem in our estimate of the effects of high
school work experience on college attendance or the choice of college major. However, in
the MNL model, to control for the endogeneity is no easy task due to its computational

difficulty. Using the probit model, we seriously address this issue.

3-2. Probit models

We classify the academic majors as either ‘practical major’ or ‘non-technical
major.’ We regard business and engineering (including computer science and
architecture) as ‘practical majors’ compared to other majors. We regard liberal arts and
education as ‘the non-technical majors’ compared to other majors. With the probit model,
we first test whether high school work experience is positively related to ‘practical major’
choice and whether it is negatively related to ‘non-technical major.’ For college
attendance choice, we regard two-year college and four-year college attendance the same

as ‘college attendance’ and apply the following probit model.

126

 

3.2.1. Standard Probit model
First, we utilize a standard probit model. Let y,’ the utility of an individual

choosing ‘practical major’ or ‘non-technical major’,

.Yl. = XlA +71Hu/r +ul (3'6)

y. =1tyi' > 0) (37)

ul ~ Normal (0, 1)
where yl is equal to ‘1’ if an individual choose practical major such as business or

engineering at the first college afier high school graduation for ‘practical major’ choice or
for ‘non-technical major’, y] is equal to ‘1’ if an individual choose non-technical major

such as liberal arts, education. These decisions are assumed to be affected by working

experience during high school. HW, represents the variable which indicates high school

working experience as described in the sub-section 3-1.

3.2.2. Probit models with a continuous endogenous variable.

Our main reason to use the probit model is to address the endogeneity issue in the
estimation of the effects of high school work experience on college major choice and/or
college attendance. One way is to utilize a two-stage method in the probit model

suggested by Lee (1981). It is very similar to the two stage least square method (2SLS) in

127

WI. {law

a at £1":

the linear regression. First, we can estimate the following reduced form of working hours

during high school:75
HW, =X,7r+Z,¢+u2 (3-8)

where Z, represents a vector of instrumental variables for the identification of working

hours. (In this case, we also assume that (ul , u2 ) has a zero mean, bivariate normal

distribution, and is independent of X , and Z, . 76)

Next, we run the probit model by plugging in the estimated working hours instead
of the working hours themselves, into the structural equation of college major or
attendance choice (3-7). By this method, we can consistently estimate the coefficient of
‘working hours’ with a scaling factor.77 This scaling factor is not easy to calculate (see
Lee 1981) However, this method has additional drawbacks. The usual standard errors and
other test statistics are not valid. It provides no easy method to test for endogeneity either.

An alternative approach is the method suggested by Rivers and Vuong (1988).
Under the joint normality assumption of ( ul , u2 ), with Var( ul )=1 , we can write

u, = 19,142 + el (3-9)
where el is a random error and is independent of X I , Z, and u,. By plugging (3-9) into
(3-6), we get

yl.=X1fl1+}/1HVVI+61u2+eI (3'10)

 

7’ Since it is more difficult to deal with a dummy endogenous variable than a continuous endogenous
variable, we use only the ‘working hours’ variable here.

76 Under this assumption, we can drive this two stage method by plugging the equation (3-8) into (3-6).
77 The scaling factor is the root of the variance of the error term of the equation (3-6) when the working

hours term HW is replaced by its reduced form.

l

128

From (3-10) and (3-7), we can drive the following two stage procedure.78 First, run the
OLS regression of (3-8) and save the residuals. Second, run the probit model, by putting
the residuals, instead of the error of (3-8), into the structural equation of college major or
the attendance choice equation of (3-10). With this method, we can also consistently
estimate the coefficients with a scaling factor.79 Since the usual probit t-statistic on the

residual is valid, we can test the null hypothesis that the working hours HW is

exogenous.

To find appropriate instruments for the identification of the working hour
equation is not easy because many background variables and personal characteristics
variables seem to be related to both working hours during high school and college
attendance or college major choice.80 Instrument variables should satisfy two conditions.
First of all,
instrumental variables should be sufficiently correlated to working hours during high
school after other variables which can affect working hours during are controlled for. We
can easily test whether instrument variables are sufficiently related to working hours after
other variables are controlled for by running the first stage regression. Second, the

instrument variables should be uncorrelated to the error term (ul ) of the equation (3-6)-

college attendance or practical major choice. However, there is no way to test whether

 

7" For a detailed derivation procedure of this two stage method, see Rivers and Vuong (1988).

79 The scaling factor is the root of the variance of the error term of the equation (3-6) when the error of the
reduced form is added to the equation (3-6).

8° The endogeneity with respect to working hours in college attendance equation might be different from the
endogeneity with respect to working hours in practical major choice equation. The endogeneity in college
attendance equation comes from simultaneous decisions of college attendance and working hours during
high school while the endogeneity comes from a particular omitted variable, a certain tendency to work.

129

this condition is satisfied.8| Thus, we usually assume that instruments are not related to
error term by reasonable arguments. 82

However, among possible candidates as an instrument variable, we do not use
those variables which are significantly correlated with college attendance or practical
major choice.83 If the coefficient of an instrument variable in the college attendance or
practical major choice is significant, it could be so either due to the correlation between
the instrument variable and working hours, or because the instrumental variable affects
college attendance or major choice directly. The latter possibility is high when including
instruments in college attendance or practical major choice equation does not change the
size or significance of the coefficient of working hours in those equations. Even though
such variables do not guarantee that we use correct instruments, it does not harm our
assumptions about the instrumental variables we use.

First, consider the case of college attendance. Four possible candidates are: test
scores squared, the dummy which indicates whether the mother works while a student in
high school (let call it ‘mother work’ dummy), ‘high reservation wage’ dummy and ‘will

not work’ dummy are possible candidates.84

 

8' It is impossible because correlation between the observable and the unobservable can not be examined.
However, In the standard ZSLS method, when we have more than one instruments, we can test whether
additional instruments are not related to the error term by overidentifying restriction test. However, in the
‘two stage probit’ model case, we do not have such method.

’2 When we have more than one instruments with 2SLS (Two Stage Least Squares), we can test whether
instruments except one are correlated to error term by the over-identifying restriction test. However, it is
not known how this test can be done in the case of the probit model.

83 One of meanings of the second condition for instrumental variable is that instrument variables should be
redundant in college attendance or practical major choice. Thus, these variables should not affect college
attendance or practical major choice directly.

8’ In the HS&B survey, students are asked “what is the lowest hourly wage you would be willing to accept
for a job while still in high school” and answered the question as a categorical variable (from below $1.50
to $4.00 or more by every $.50). We make a dummy variable for above $3 categories and a dummy for
those who will not work at any wage during high school, and thus, have no reservation wage. We use ‘high
reservation wage’ dummy as an indicator for demands an individual student face in Chapter 1.

130

While test scores are expected to be linearly related to college attendance, it is
non-linearly related to working hours as we see in Chapter 1. Thus, the second order of
test scores (test scores squared) are not related to college attendance but sufficiently
related to working hours during high school.85

The ‘mother work’ dummy is significantly related to working hours during the
senior year as we see in Chapter 1. The problem is whether it is related to college
attendance. It is possible that the working mother could either encourage or discourage
their children to go to college. A mother’s presence at home might help students to
concentrate on academic activity after school, thus encouraging them to attend college.
However, the working mother also might encourage students to go college through their
experience in the world of work.

As we see in chapter 1, the ‘high reservation wage’ dummy and ‘will not work’
dummy are Significantly related to high school working hours. However, these variables
are also expected to be related to college attendance. Students who are determined to go
to college would demand higher reservation wage or would not work at any wage during
high school. However, as we see in Chapter 1, students who have a higher reservation
seem to work more.

However, if ‘high reservation wage’ dummy indicates the market demand for a

student labor,86 it could affect college attendance directly. Betts and McFarland (1994)

Show that labor market conditions affect two-year college enrollment. However, when we

 

’5 Even though this variable satisfies the conditions of instrument variables, One problem with using this
variable as an instrument is its strong correlation with the test scores. We do not know what problem this
correlation could cause in our estimation of the coefficients of the endogenous variables. In this sense, using
this variable is experimental.

131

 

include ‘high reservation wage’ and ‘will not work’ dummies in college attendance
equation, the coefficients of both variables are not significant at all.87

In the case of ‘practical major’ choice, if we assume that the endogeneity related
to working hours during high school comes from an individual’s unobservable
characteristics, such as tendency to work as mentioned in section 2, instrument variables
should not be correlated with the unobservable characteristic, tendency to work and
should be redundant in the ‘practical’ major choice.88 We can consider local economic
indicators such as unemployment rate and employment growth rate, as instruments. We I:
can also consider ‘high reservation wage’ dummy and ‘will not work’ dummy.

Local economic indicators, unemployment rate and employment growth rate are
sufficiently related to working hours during high school and might not be related to
‘practical major’ choice. In general, the relationship between ‘practical major’ and high
unemployment rate or low employment growth rate could be either positive or negative.
Usually, economic expansion will increase demand for the people who are trained in
business or engineering areas, encouraging college students to enroll in these areas.
However, high unemployment rates could push college students to enroll in these majors

because they might think that majoring in these areas might help them to find jobs later.

 

8" In chapter 1, we claim that students who live in better economic condition have higher reservation wage
and reservation wage represents the market demand or economic conditions an individual student faces.

37 It might be because that we regard two-year and four-year college enrollment together as college
attendance while Betts and McFarland uses only community college enrollment as the dependent variable.
Also, Betts and McFarland examine the effects of cyclical economic conditions (especially unemployment
rate among high school graduates ) on community college enrollment while ‘reservation wage’ indicates
demand individual student faces at a point in time.

38 We can not use test scores squared as an instrument because the test scores and ‘business or engineering’
major choice might be non-linearly related each other. It is because test scores of college students who
choose business or engineering as their majors are expected to be in the middle range of the test scores

distribution.

132

However, we can be sure that these variables are not related to an individual
unobservable characteristics, especially tendency to work. We can easily think that an
individual’s innate tendency to work not to be influenced by local unemployment rate or
other economic conditions.

In the cases of ‘high reservation wage’ dummy and ‘will not work’ dummy, we
can think of two contrasting possibilities. First, these variables, especially the ‘will not
work’ dummy could represent a student’ tendency to work.89 We can regard them as
proxy variables for the omitted variable-tendency to work. Then, ‘high reservation wage’
and ‘will not work’ dummy are expected to be negatively related to ‘practical major’
choice. Thus, we first test whether these variables are related to ‘practical major’ choice
and check if the coefficient of working hours changes when we include these variables in
the practical major choice equation.

The other possibility is that these variables are not related to individual’s tendency
to work. ‘High reservation wage’ dummy can represent economic conditions or market
demand each individual faces and ‘will not work’ dummy can represent just a random
choice by individual student.90 In this case, we can use them as instruments because we
can assume that they are not related to error term of the practical major choice equation.

In controlling for endogeneity related to working hours in the practical major

choice equation, we concentrate on the HS&B male sample because the association

 

89 Even reservation wage can represent individual’s willingness to work or tendency to work.

9° Those students who answered “will not work at any reservation wage during high school’ is less than 10
percent among all students. ‘Will not work’ does not seem to be related to any local economic conditions
such as unemployment rate or employment growth rate or test scores or family socio-economic
backgrounds.

133

 

between high school work experience is the strongest among this group and because we

have these possible instruments.

3-3. Data.

The current study uses two different data sets: NLS-72 and the HS&B Senior
Group data. NLS-72 is a longitudinal survey that originally sampled 22,652 seniors from
the high school class of 1972. It contains extensive family background variables, as well
as a standardized test score (N LS test score). We restrict the sample to those who
responded to the first and second follow-up surveys (1973, July 1974) and took the NLS
test. The full sample sizes are 6,763 for males and 7,013 for females respectively. The
sub-sample sizes of those who were enrolled in the academic area in the first semester
after high school graduation are 2310 for males and 2040 for females.

HS&B is also a longitudinal survey that originally sampled 28,000 seniors and
over 30,000 sophomores from 1,100 high schools in 1980. HS&B contains almost the
same variables as NLS-72. HS&B has better features for studying high school
employment because it contains a labor market indicator file, which showing economic
activities of
areas such as county, SAMA, and states. Also, it contains information on working
experience during the junior year. We use the senior sample and restrict it to those who
responded to the first and second follow-ups and took the NLS test. The sample sizes are
4,063 for males and 4,966 for females. The sub-sample sizes of those who were enrolled
in the academic area in the first semester after high school graduation are 1520 for males

and 1986 for females.

I34

In these regression analyses, as the control variables we include family’s socio-
economic status, race, ability measures, types of high school program, regions of high
school, community types students have lived in during their high school years, and some
other variables for both college attendance and college major choices. Also, the distances
from two-year and four-year colleges are included for college attendance choice. Also,
we included local labor conditions for the HS&B samples in the control variables for
college attendance.

In the NLS-72 and HS&B, a family’s socio-economic status is divided into three
categories: ‘SES-low,’ ‘SES-medium’ and ‘SES-hig.’ We exclude the ‘SES-low’ socio-
economic status as the comparison group in the regression analysis. We also include race
categories such as ‘Black’, ‘Hispanics’, ‘Asian’, ‘Native Indian’ and ‘others’ in the
control variables. Thus, the excluded category is ‘White’. Other family background
variables included are the number of siblings and ‘any dependents (or child)’ which is one
if a student has any dependent (or children). These family background variables might
affect individual’s need or tastes which could affect an individuals’ choices of major. In
most major choice studies, these variables are included as some of the important
determinants of college major choices. (Davies and Guppy 1997).

Four regional categories, ‘North East,’ ‘North Central,’ ‘West’ and ‘South’ are
made and ‘North East’ is excluded. Three community type categories such as ‘Rural
Farming’, ‘Small Town’, ‘Urban’ and ‘Suburban’ for 1972 and three categories such as
‘Rural Farming’, ‘Urban’ and ‘Suburban’ for 1980 are made and ‘Rural Farming’ is
excluded. The regions might be related to demand for different college majors in the

regions, which could affect college major choices. The types of community students

135

 

indicate what adults do for their living in that community and what kinds of role models
high school students have. For example, in an affluent suburban areas, more adults who
are working as professionals or businessmen live there. By seeing the ways of adults live,
students might consider their own future careers and choose appropriate majors following
their role models.

In the college attendance choice and the choice of field (vocational or academic),
we include the averages of mathematics test scores, reading test scores, and vocabulary

test scores. However, in the specific major choice for the students in the academic area,

we include three basic skills separately: vocabulary, reading and math. Paglin and Rufolo
(1990) find that mathematical ability measured in the Graduate Record Exam (GRE) and
the Scholastic Aptitude Test (SAT) math test is an important determinant of field choice
for college students. Math and science majors are highly likely to require more
mathematical thinking or skill while liberal arts might demand more verbal or linguistic
skills. If ‘working hourS’ is related to a certain skill or ability, the positive and significant
coefficient of ‘working hours’ might be superficial and simply represent the relationship
between major choices and those skills.

Another factor we need to consider in the estimation of influence of high school
working experiences on major choices is high school preparation through courses taken
during high school. For those who are well trained in mathematics and science during
high school, in college science (including math) and engineering major (engineering,
architecture, and computer science) might be appropriate. The number of mathematics or
sciences courses taken during high school might be important determinants in choosing

such majors. If high school workers tend to take more vocational courses such as

136

business, or trade or health related courses, and avoid math and science courses, it might
force student workers to choose business or occupational oriented majors. Because
student workers are not prepared for other majors, their major choice can be limited only
to a certain kind of major. Thus, we include the number of high school courses students
have taken in each major subject areas; mathematics, sciences, English, business and
health- related (or other vocational courses). However, the number of courses taken
during high school are coded slightly differently in the NLS-72 and HS&B data sets, so
the coefficients are not comparable. We also include the grade point average (GPA) as
another aspect of high school preparation.

In the HS&B sample for the college attendance, county unemployment rate,
employment growth rate, and log county average income are included as additional
control variables. Betts and McFarland (1994) Show that 1 percent increases in the
unemployment rates are associated with about .4 to .5 percent rises in two-year college
enrollments"

Means and standard deviations of independent variables and instrumental
variables for college students who are in the academic field, are tabulated in the NLS-72
and Table HS&B 3-3-1.92 By comparing these summary statistics between two years, we
notice one big difference across the samples. Blacks and Hispanics are over-represented

in the HS&B sample, and thus the percentage of students within the first quartile of socio-

 

9' The reason to include log county average income is to see how the local income level may have any
effects on students’ college enrollment after his or her own family socio-economic background is controlled
for. Excluding and including this variable does not change the coefficients of high school work experience
variables in the college attendance choice almost at all. However, it is related to college attendance.

92 For the full samples, see in chapter 1. For all college students, we do not report here. Summary statistics
are very similar to college students in the academic field.

137

 

economic status are over-represented. This is because the original sample design includes
much more minorities than the population.

If the relationship between high school employment and college major choice
differs across race or socio-economic status, this sample design might cause some bias of
the estimates of the coefficients. To solve this problem, we use the sample weights

provided in the original data set.93

 

 

Table 2-3-3-1: Summary Statistics for Sub-sample of College Students in Academic Field (NLS-72 and HS&B)

 

 

 

Variable Name Definition Mean and Standard Deviation Mean and Standard Deviation
(NLS-72 Samples) (HS&B Samples)
W W ' ‘ ”Male (2310) Femaie‘l2'0'4‘0) Male (1520) Female
(1986)
WORKING Number of working hours per week 12.4 9.1 13.2 1 1.5
HOURS during high school senior year (10.5) (9.3) (11.6) (10.7)
WORK 0-1 dummy variable that equals one if .78 .69 .72 .70
student work during the senior year (.42) (.46) (.45) (.46)
HEAVY WORK 0-1 dummy variable that equals one if .41 .27 .46 .42
student work more that 15 hours per (.49) (.44) (.50) (.49)
week
WORK JUNIOR 0-1 dummy variable that equals one if .67 .59
student works junior year (.47) (.49)
GRADE Average point of grade during high 2.55 2.89 3.09 3.24
school. (0-4 scale) (.91) (.87) (.65) (.61)
TEST SCORE NLS average test score of six test 54.2 54.8 54.5 52.4
categories (6.02) (6.28) (8.10) (8.15)
SES-MEDIUM‘ 0-1 dummy variable that equals one if .43 .44 .45 .43
student’s SES is in 2"d and 3'“ quartile (50) (.50) (50) (.50)
SES-HIGH 0-1 dummy variable that equals one if .44 .41 .33 .29
student’s SES is within 4": quartile (50) (50) (.47) (45)
BLACK 0-1 dummy variable that equals one if .07 .12 .19 .21
student is black (.26) (.32) (.41) (.41)

 

93 However, the weights are given in the original sample and in the follow-ups might not be appropriate
because the samples we are using are the sub-samples of those. Thus we separate the sample into two

groups--White only and minority and run the same regressions. By doing so, we can see if there are any
differences between Whites and minorities. We do not report the results from separate groups in Tables.

I38

HISPANIC

ASIAN

NATIVE
AMERICAN
SIBLINGS

ANY
DEPENDENT
HIGH SCHOOL
ACADEMIC

HIGH SCHOOL
VOCATIONAL

SMALL TOWN

URBAN

SUBURBAN

NORTH
CENTRAL

SOUTH

WEST

DISTANCE
TWO-YEAR
DISTANCE
FOUR-YEAR
VOCABULARY
SCORE
READING
SCORE

MATH SCORE

MATH COURSE

0-1 dummy variable that equals one if
student is Hispanic

0-1 dummy variable that equals one if
student is asian

0-1 dummy variable that equals one if
student is native amerlcan

Number ofsiblings

0-1 dummy variable that equals one if
student has any dependent

0-1 dummy variable that equals one if
student‘s high school program is
academic

0-1 dummy variable that equals one if
student’s high school program is
vocational

0-1 dummy variable that equals one if
student lives in small town during
senior year

0—1 dummy variable that equals one if
student lives in urban area during senior
year

0-1 dummy variable that equals one if
student lives in suburban during senior
year

0-1 dummy variable that equals one if
student lives in north central during
senior year

0-1 dummy variable that equals one if
student lives in south during senior
year

0-1 dummy variable that equals one if
student lives in west during senior year

Miles from the nearest two-year college

Miles from the nearest four-year college

NLS vocabulary test score

NLS reading test score

NLS math test score

Number of math courses taken during

high school

.03
(.18)
.02
(.14)
.004
(.064)
2.10
(1.77)
.06
(.23)
.73
(.44)

.05
(.23)

.27
(.45)

.25
(.44)

.31
(.46)

.26
(.44)

.31
(.46)

.19
(.39)
16.4
(25.6)
18.2
(25.6)
55.1
(9.53)
55.0
(8.52)
57.1
(8.28)
5.1
(1.75)

139

.03
(.17)
.02
(.13)
.006
(.078)
2.14
(1.76)

(.19)
.72
(.45)

..07
(25)

.29
(.46)

.27
(44)

.28
(.45)

.23
(.42)

.32
(.47)

(.39)
17.8
(26.3)
17.5
(25.1)
55.9
(9.71)
55.7
(8.70)
54.7
(8.69)
4.3
(1.81)

.20
(.40)
.06
(.23)
.011
(.105)
2.86
(2.47)
.001
(.036)

(.48)

.08
(.28)

.27
(.44)

.50
(.50)

.25
(.43)

.33
(.47)

.20
(.40)
14.9
(25.3)
16.0
(22.5)
54.00
(9.54)
53.8
(9.66)
55.69
(8.73)
5.2
(1.56)

.20
(.40)
.04
(.20)
.011
(.105)
2.87
(2.12)
.011
(.105)
.60
(.49)

I '(")J WW7

.15
(.33)

 

.27
(.44)

.50
(.50)

.25
(.43)

.33
(.47)

.20
(.40)
15.7
(26.4)
16.0
(22.6)
52.1
(9.26)
52.6
(9.30)
52.2
(8.43)
4.6
(1.75)

SCIENCE
COURSE
ENGLIGH
COURSE
BUSINESS
COURSE
HEALTH
COURSE
MOM WORK

U RATE

EMP GROWTH

LOG INCOME

HIGH
RESERVATIO-
N WAGE
WILL NOT
WORK

Number of science courses taken during
high school

Number of English courses taken
during high school

Number of business courses taken
during high school

Number of health courses taken during
high school

0—1 dummy variable that equals one if
the mother work during a student‘s
mother work during his high school
years

County unemployment rate

County employment growth rate

Log county average income

0-1 dummy that equals one if
reservation wage is larger than
$3.00

0-1 dummy which is equal to zero
if student will not work at any wage

4.7
(1.90)
6.4
(1.24)
1.9
(1.91)

(.60)
.60
(.49)

3.9
(1.80)
6.4
(1.30)
2.7
(2.8)

(1.38)
.60
(.49)

4.5
(1.82)
6.0
(.92)
1.34
(1.67)
1.01
(1.89)
.68
(.47)

7.33
(2.49)
.89
(3.8)
9.15
(.23)
.78
(.42)

.016
(.127)

3.9
(1.86)
6.1
(.81)
2.55
(2.18)
.95
(1.80)
.72
(.45)

7.48
(2.60) r
1.02
(3.4)
9.14
(.24)
.67

(.47)

 

.014
(.115)

 

We also tabulate the summary statistics for SES categories for the full sample and

the sub-sample of college students in the academic field in Table 2-3-3-2 and 3. The SES

distributions are almost the same over two years for the full samples, while the SES

distribution among college students in the academic field shifts much. This is because

minority students with the same SES have a stronger tendency to go to college if they

have the same conditions. Because of this, using weights for the samples of the college

students in the academic field might not be appropriate.

140

 

Table 2-3-3-2: SES Distribution for the sub-samples of college students in the academic field (weighted)

 

Variable Name Definition

 

SES-MEDIUM‘ 0-1 dummy variable that equals one if
student’s SES is in 2"" and 3" quartile
SES-HIGH 0-1 dummy variable that equals one if

student's SES is within 4'” quartile

Mean and Standard Deviation

Mean and Standard Deviation

 

 

(NLS-72 Samples) (HS&B Samples)

Male (2310) Female (2040) Male (1520) Female (1986)"
.44 .45 .46 .43

(.50) (.50) (.50) (.50)

_45 .42 .41 .37

(.50) (.49) (.49) (.48)

 

Table 2-3-3-3: SES Distribution for full samples (weighted)

 

Variable Name Definition

SES-MEDIUM‘ 0-1 dummy variable that equals one if
student’s SES is in 2'“1 and 3'” quartile
SES-HIGH 0-1 dummy variable that equals one if

student's SES is within 4'h quartile

Mean and Standard Deviation

Mean and Standard Deviation

 

(NLS-72 Samples) (HS&B Samples)
“partitiasrmrgaa; (7013) Male (4064) Female (4966)W
.50 .50 .51 .51

(.50) (.50) (.50) (.50)

.28 .23 .27 .22

(.45) (.42) (.44) (.42)

 

4. Estimation Results

4-1. College Attendance Choice

Our estimation results for college attendance choice not only confirm previous

studies on the relationship between high school work experience and college attendance

but also reveal several other characteristics of the relationship. We can summarize our

findings as follows: (1) In 1972, after controlling for family background, academic ability

and other factors such as region or community types, work experience- especially heavy

work experience during the senior year- is negatively related to college attendance. (2) In

1980, the negative relationship between work experience during the senior year and

college attendance is very weak while the negative relationship between the work

experience during the junior year and college attendance is very strong. (3) These

negative relationships are much stronger in the choice of four-year college attendance

141

 

 

than two-year college attendance. (4) The negative associations between college
attendance and high school work experience is stronger among males than females. (4)
However, controlling for the endogeneity changes these results dramatically. For the 1972
samples, in the probit models for the college attendance, the coefficients of work hours
during the senior year are positive while in 1980, the coefficients are still negative.

However, endogeneity tests Show that there are not significant endogeneity problems in

 

both years.

4-1-1. The MNL model results

 

The MNL model results for college attendance choice are tabulated in Table 2-4-
1-1 and 2, and Table 2-4-1-5 and 6. For convenience, we only report the coefficients of
those variables which represent high school work experience mainly.”4 The coefficients
of work experience variables indicates their effects on the log odds ratio between two—
year college attendance and ‘no college’ or between four-year college and ‘no college.’

To interpret the logit coefficients, we utilize the method used by Rouse (1994).
For the dummy variables such as ‘work’ or ‘heavy work’, afier the MNL regressions are

run, first we calculate the predicted probability of choosing each alternative such as ‘two-

 

9’ The coefficients of other variables have expected signs and significance. For example, the coefficients of
test scores, medium and high SES are positive and significant. The coefficients of the number of siblings,
‘child dummy’ are negative and significant. The coefficient of ‘academic program’ is positive and
significant. The coefficients of unemployment rates do not have always the expected Sign and significance.
Thus, we report the coefficients of county unemployment rate, county employment rates and county log
average income in the following table when the ‘workinLhours’ and other variables are controlled for.

 

Variable Two-year Four-year Two-year Four-year
U rate -.029 -.008 .014 .039
(.021) (.020) (.019) (.017)
Emp Grow .019 -.020 .021 " -.004
(.014) (.012) (.012) (.011)
Log income -.122 .187 .469‘ .652”
(.323) (.279) (.284) (.258)

 

142

year college’ or ‘four-year college’ when the dummy variable is equal to zero and later we
calculate the predicted probability when the dummy variable is equal to one. For
‘working hours’, first we calculate each individual’s predicted probability of choosing
each alternative using their original values for ‘working hours’. Then, we increase
‘working hours’ by five or ten hours and then, calculate a new probability of choosing
each alternative. Lastly, we report the mean differences in the probabilities in Table 2-4-
1-3 and Table 2- 4-1-4.

In 1972, after controlling for other variables, the coefficients of ‘work,’ ‘heavy
work,’ and ‘working hours’ are significantly negative in most cases. The coefficients of
the working hours categories show that particularly working heavily dining the senior
year is negatively associated with college attendance (see Table 2- 4-1-2 and 3). This
result is consistent with the findings of Meyer and Wise (1982) with the NLS-72 data.

This negative association between high school work experience during the senior
year and four-year college attendance is stronger than between the work experience and
two-year college attendance. The Sizes of coefficients and significance of the work
experience variables are much larger in the four-year college choice than two-year college
choice (see Table 2- 4-1-1 and 2). For males, ‘work,’ ‘heavy work,’ a ten ‘working
hours,’ increase are respectively associated with decreases of 3.2 percent, 3.4 percent, and
1.9 percent in the probability of attending a four-year college while there is no such
association with attending a two-year college.

This negative association is stronger among males than females. The sizes of the
coefficients are much larger among the male sample (see Table 2- 4-1-1 and 2). Table 2-

4—1-3 also shows that, ‘heavy work’ and a ten ‘working hours’ increase are respectively

143

 

 

associated with a 3 percent, and a 1.2 percent decrease in the probability of attending a
four-year college.

In 1980, the negative association between work experience during the senior year
and college attendance is much weaker than in 1972. The coefficients of ‘work’ during
the senior year are not negative in some cases, even though they are not significant. The
Sizes and significance of the negative coefficients of ‘heavy work’ and ‘working hours’

decrease greatly (see Table 2- 4-1-5 and 6). 95 Afier ‘working hours’ during the

 

 

 

Table 2-4-1-1: College Attendance Choice: Male 6763 (NLS-72)
Variables Two— F our- Two- Four- Two- Four- Two- Four-
_ __________________ ____~__Year__“__ Year Year. ___u__-_~____Y__e»2_1_r Year Year Year Yearmw
Work -.121 -.330“”
(.108) (.093)
Heavy work -. I 89" -.393"”
(.083) (.073)
Working hours -.010” -.020”‘
(.004) (.003)

1-15 hours -.004 -.127
(.121) (.104)

16-20 hours -. 103 -.293"‘
(.146) (.127)

21-30 hours -.060 -.490*"
(.130) (.116)

More than 30 -.498"" -.647*”

hours (.152) (.136)

-2"‘log-

likelihood 3541.5 3557.1 3564.7 3574.5

Psudo

R square .2797 .2716 .2815 .2823

 

Notes: (1) Standard errors are in parentheses. (2) ""’ means ‘significant between 10% level and 5% level,’ ‘”’ means
‘significant between 5% level and 1% level,’ """" means ‘significant at less than 1% level’. (3) Each two columns
represents one MNL model results. (4) ‘Two-year’ and ‘Four-year’ in the first row represent the log odds ratios: the
log[probability (two-year college)/probability (no college)], and the log[probability (four-year collegeyprobability (no
college)]. (5) The coefficients of other variables included are not reported. Other variables are test scores, GPA,
family’s SES, number of siblings, child dummy, high school program, community types, region, distances from the
nearest two-year college and four-year colleges.

 

95 When the weighted sample is used to accommodate the over-representation of the minorities in the HS&B
samples, the coefficients of these variables are still negative but much smaller and are not significant at 10
percent significance level in most cases. When the samples are separated by two groups, (White only and
minorities). Only for minorities, are the coefficients of these variables negative and significant.

I44

 

 

Table 2-4-1-2: College Attendance Choice: Female 7013

(N LS-72)

 

 

 

 

 

 

 

 

Variables Two- Four- Two- F our- Two- Four- Two- Four-
‘ Year Year Year___ Year Year Year Year Year” _
Work .020 -.049
(.091) (.078)
Heavy work -.136 -.279“"
(.089) (.078)
Working hours -.008" -.012"”
(.004) (.004)
1-15 hours .105 .090
(.101) (.086)
16-20 hours .1 1 I -.O92
(.136) (.119)
21-30 hours -.116 -.302”
(.137) (.121)
More than 30 -.432" -.363"”
hours (.204) (.170)
-2‘log-
likelihood 3456.2 3468.7 3467.3 3478.5
PsudoR square
.2716 .2726 .2725 .2733
Note: See notes in Table 2- 4-1-1.
Table 2-4-1-3: Changes in the Mean Probability of Attending Two- or Four-Year Collge (NLS-72)
-lnamsadcnijariables-..- _. __ Male -.-._ _.-._ female . - ..
Two-Year Four-Year Two-Year Four-year
Work -.2 -3.2 +4
Heavy work -0.7 -3.4 -.4
5 working hours increase -.01 -1.0 -.2
10 working hours increase -.03 -1.9 -.4

 

Note: Units are all %.

junior year is controlled for, the coefficients of ‘working hours’ during the senior year are

not significant at all. Table 2-4-1-3 shows that, only without controlling for the work

experience during the junior year, ‘heavy work’ and a ten ‘working hours’ increase are

respectively associated with around a 2 percent decrease, around a 1 percent in the

probability of attending four-year college. When we control for work experience during

the junior year, the coefficients of the senior year work experience variables are

insignificant.

I45

 

 

However, for both male and females, the work experience during the junior year is
negatively associated with college attendance. The coefficients of ‘work junior’ and
‘working hours junior’ are negative and significant in most cases (see Table 2- 4-1-4, 5,
and 6). In the 1980 male case, after controlling for working hours during the senior year,
‘work junior’ is associated with a 4 percent decrease in the probability of attending a two-
year college while there is almost no association with attending a four-year college. In the
female case, ‘work junior’ is associated with a 1.4 percent decrease in the probability of
attending two-year colleges and a 2 percent decrease in the probability of attending four-
year colleges.

We also use the average working hours per week during the senior and junior
years to see the effects of average work intensity during high school on the college
attendance choice. Also, coefficients of the average working hours are all negative and
significant in both two-year and four-year colleges. A ten hour increase in the average
working hours is associated with around a 1.5 percent decrease in the probability of
attending either two-year or four-year colleges for males, while for females, it is
associated with a 1.7 percent decrease in the probability of attending only four-year
colleges.

The weaker negative relationship between high school work experience during the
senior year and college attendance, especially among females in 1980, could be related to
the relationship between work experience and academic ability or family background. In
1980, students (especially female students) with high academic ability and good family
background hold jobs more ofien and work more hours than in 1972 as we see in Chapter

I.

146

A..—

 

 

The coefficients of ‘working hour’ categories during the senior year Show that in
1972 the relationship between working hours and two-year or four-year colleges is strictly
negative for both males and females, while in 1980 the relationship is non-linear (see the
last columns in Table 2-4-1-1 and 2, and Table 2-4-1-5 and 6). Some can claim the non-
linear relationship as evidence for the positive effects of the moderate work during high
school on educational aspects.96 However, the strict negative relationship between
working hours and college attendance in 1972 and between working hours in the junior

year and college attendance in 1980 makes it difficult to maintain such an argument.

 

Also, in 1980 the coefficients of junior working hour categories show a strict negative
relationship between working hours and two-year or four-year college attendance, though
the results are not reported. We suspect that the non-linear relationship could capture the

fact the students with high academic ability and good family background work moderate

 

 

 

 

hours.

Table 2-4-1-4: Changes in the Mean Probability of Attending Two- or Four-Year Colleges: (HS&B)
_ Independent Variables -.--___-_--._ __m-Male Female

Two-Year Four-Year Two-Year Four-year

Work . . . .

Heavy Work -1.3 -2.3 .8 -2.1

5 working hours increase -.3 -.5 .2 -.5

10 working hours increase -.5 -.9 .3 -1.0

Work Junior‘ -3.7 +.7 -1.4 -2.0

5 average working hours increase -.8 -.7 -.2 -.9

10 average working hours increase -1.5 -l.4 -.4 -l .7

 

“ afier controlling for working hours in the senior year

 

9" ‘Developmental model’ assumes that part-time employment could lead to better performance through the
character building effects of working experience (Marsh 1991).

147

 

Table 2-4-1-5 College Attendance Choice: Male (3941) (HS&B)

 

Variables Two- Four- Two- Four- Two- Four- Two- Four- Two- Four- Two- Four- Two- Four-
Year Year Year Year Year Year Year Year Year Year Year Year Year Year

 

Work -.O42 .024
(.102) (.100)
Heavy -.185‘ -.215
work (.096) "
(.089)
Working -.007‘ - -.005 -.008 -.002 -.006
hours (.004) .009" (.004) "”' (.004) (.004)
(.004) (.004)

Average -.OO9* -.014

working (.005) ""

hours (.004)

Working -.015 -.010

hours it! it

Junior (.005) (.004)

Work -.285 -.061 -.309 -.103

Junior "* (.102) " (.103)

(.108) (.109)

1-14 .218 .297"

hours (.139) (.129)

15-21 .062 .160

hours (.144) (.133)

22-29 -.079 -.311

hours (.156) "'
(.151)

More than -.088 -.142

30 hours (.149) (.144)

-2‘log- 1604.7 1611.4 1611.2 1622.9 1618.4 1628.2 1634.9

erhhood

Psudo R .1995 .2003 .2003 .2017 .2024 .2012 .2032

square

 

Notes: (1) See notes in Table 2-4-1-1. (2) In the HS&B samples, we also include county unemployment
rates, county employment grth rate, and log county average income. (3) The average working hours are
the sum of working hours during the senior and junior years divided by 2.

148

 

 

Table 244-6: College Attendance Choice: Female (4831)

Variables Two-

Four- Two- Four-
WW Year Year Year Year Year Year Year
Work .039

(.092) (.085)

Two-

F our-
Year

Two-

Four-

Two— Four-
Year Year Year Year Year Year

(HS&B)

Two- Four-

Heavy -.119

work (.087) (.080)

Working -.001 -.006" -.001 -.005 -.002 -.006

hours (.004) (.004) (.004) (.004) (.004) (.004)

Average -.009* -.014

working (.005) "

hours (.004)

Working -.012 -.010

hours it. 3!

Junior (.004) (.004)

Work -.187 -.189 -.200 -.220

junior fit #0 fit ##t
(.090) (.082) (.090) (.083)

1-14 .097 .238"

hours (.118) (.106)

15-21 .229" .230“

hours (.120) (.111)

22-29 -.219 -.222"‘

hours (.151) (.137)

more than -.095 -.172

30 hours (.157) (.156)

-2"‘log- 1847.5 1848.2 1855.1 1855.2 1864.1 1876.8

erhhood

Psudo R .1857 .1852 .1858 .1858 .1867 .1880

square

 

 

 

Note: See note

149

 

4-1-2 Probit results

With the probit models, we control for the endogeneity in the relationship
between high school work experience and college attendance, and test whether there is
endogeneity in the relationship. In the NLS-72 samples, we concentrate on the effects of
the ‘working hours’ during the senior year on the college attendance (either two-year or
four-year colleges) because we do not have information on work experience during the
junior year. In the HS&B samples, we concentrate on the effects of the ‘average working
hours’ in the senior and the junior years on college attendance. The results are tabulated
in Table 2- 4-1-7 and 8, and Table 2- 4-1-9 and 10. Also the tables show whether the
instruments are related to college attendance.

Before we use the IV method, we first control for educational aspirations. Three
educational aspiration categories-no college attendance, two-year college attendance, and
four-year college attendance are made and no college attendance category is dropped. In
both 1972 and 1980, when educational aspirations are controlled for, the sizes of the
coefficients of working hours are reduced (see column 2 in the tables). These results
indicate that students with lower educational aspirations are likely to work more.
However, the signs of all the coefficients of working hours are still negative and
significant. Also, just including educational aspiration does not solve the endogeneity
problem because educational aspirations also could be affected by work experience. The
Instrumental variable method might be better in controlling for such endogeneity.

In the 1972 male case, test scores squared is the only instrument to be used. It is

not related to college attendance while ‘mother work’ dummy is significantly related (see

150

 

 

column 5 in Table 2- 4-1-7.) However, test scores squared are sufficiently related to the
working hours as we see in Chapter 1.

Afier the reduced form of the working hour equation is estimated, first the
predicted value of working hours is entered into the college attendance equation (column
3, two stage probit suggested by Lee 1981), and second, the residuals from the reduced
form working hour equation and working hours are entered into the college attendance
equation (column 4, two stage probit suggested by Rivers and Vuong 1988 ).

The results are mixed. We find the coefficient of predicted working hours to be
positive (see column 3). Also, the coefficient of working hours is positive in column 4
and the coefficient of residuals from the reduced form working hour equation is negative
in the college attendance equation (see column 4). Thus, the effects of working hours on
college attendance seem to be positive. However, due to the large standard error of the
coefficient of the residual, we can not claim that there is a strong endogeneity.

In the female case, we use ‘mother work’ as the instrument because it is not
related to college attendance at all (see column 8 in Table 2- 4-1-8). We have the similar
results with the male case. The coefficients of predicted working hours or working hours
are positive. Also, the coefficient of the residual is negative yet, not significant at a
standard significance level (see columns 3 and 4 in Table 2- 4-1-8.)

When we use ‘mother work’ and test scored squared together as instruments, the
sign of the coefficient of predicted working hours is still negative while the sign of the
residual of the working hour equation is negative (see columns 5 and 6 in Table 2- 4-1-8.)
However, using test scores as an instrument might be questionable in the female case

because test scores squared do not seem to be redundant in the college attendance. When

151

 

 

test scores and test scores squared variables are entered together in the college attendance
equation, the significance of test scores disappears (see columns 7 in Table 2-4-1-8.)

For the 1980 case, including the educational aspirations reduces the size of the
coefficient of ‘working hours’ for both males and females (see columns 2 in Table 2- 4-1-
9 and 10). However, the results using the IV method are quite different from those in
1972. Here we use the averaged working hours during the senior and junior years instead
of separating them.97

In the male case, we use the ‘high reservation wage’ dummy, ‘will not work’
dummy, and ‘mother work’ dummy variable as instruments. In column 3 and 4 in Table
2- 4-1-9, we use the first two variables as the instruments, and in columns 5 and 6,we use
all three variables. These variables are not related to college attendance (see column 7 and
8 in Table 2-4-1-9). The coefficients of the predicted working hours are still negative (see
columns 3 and 5 in Table 2-4-1-9). The signs of the coefficients of the residuals are
positive and the sizes of them are very small and standard errors are large. There is no
evidence for the endogeneity.

In the female case, the ‘will not work’ dummy seems to be related to college
attendance somewhat even though the coefficient of this variable is not significant at the
10% significance level (see column 7 in Table 2- 4-1-10). However, the ‘mother work’
dummy is not related to college attendance at all. When we use ‘mother work’ as the
instrument, the results are very similar with the male case. The coefficients of the
predicted working hours are still negative (see column 3). The sign of the coefficient of

the residual is positive, and its size is very small, and standard error is large.

152

 

Even when we use two other variables-the ‘high reservation wage’ dummy and
‘will not work’ dummy as instruments, the coefficients of the predicted working hours are
still negative (see columns 5). Even though the sign of the coefficient of the residual is
negative, its size is very small and its standard error is large. We can claim that there is no
evidence for the endogeneity in 1980 in the female case either.

In sum, even though the endogeneity tests in 1972 and 1980 show that there is no
strong evidence for any endogeneity related to working hours, the signs of coefficients of
working hours, predicted working hours, and residuals from two stage probit models for

the NLS-72 samples are opposite to the signs for HS&B samples. The quite contrasting

 

results between two years seem to be reasonable if we consider the fact that in 1980, more
students with high academic ability and high SES worked than 1972. In 1972, students
who had low academic ability and low SES and thus, might have planed not to go college
seem to hold jobs more and work longer hours. It is very plausible that for high school
student workers who are not able to go to college or plan not to go to college due to lower
academic ability or lower SES, working might help them to go college through savings
or good work habits if there is any effects from high school work experience on college
attendance. Thus, when we control for the endogeneity, the effects of high school work
experience are even positive if any.

However, in 1980, students might have worked regardless of their plans of
attending college. Students with better family backgrounds and higher academic ability
worked more than in 1972. For those students who plan to go college, working might

hinder their academic preparation for colleges.

 

97 As we see in the previous regression results, the significant and negative coefficient of working hours are
mostly related to the working hours in the junior year.

153

 

However, our attempts to control for the endogeneity or to test whether there is

endogeneity are based on the assumption that instrumental variables are not related to the

error term in equation (3-6). Because we do not have methods to test this assumption, we

can not be sure completely that our results are correct. Individuals’ unobservable

characteristics could be correlated to instruments. In that case, our results from two stage

probit models might be wrong. Particularly using the instrument ‘high reservation wage’

dummy might not be correct because in Chapter 1, we show that this variable represents

market demand that an individual student face and is unrelated to test scores. If such

market demand affects anyhow a student’s decision on college attendance (Betts and

McFarland 1994), the results using this variable might not be correct.

 

 

Table 2-4-1-7: Male: Probit for College Attendance (NLS-72)
Variables (2L (3) ZSP (4) ZSP (5) (6)
Working Hours -.007“* .009 .007 -.010*“ -.010"‘"I
(.002) (.027) (.027) (.002) (.002)
Residuals -.017
(.027)
Test scores .032‘" 050*“ 050"" .068" .067"
(.004) (.005) (.005) (.033) (.033)
Test scrores squared -.000 -.000
(.000) (.000)
Mom work -.061*
(.038)
Eud. Aspiration 1152”"
Two-year college (.080)
End. Aspiration 1.723‘"
Four-year college (.066)
3690.4 2802.8 2836.4 2836.4 2839.1
(.4009) (.3044) (.3081) (.3081) (.3084)

 

Notes: (1) Standard errors are in parentheses. (2) “" means ‘signiftcant between 10% level and 5% 1eve1,"“’ means
‘significant between 5% level and 1% level,’ "“"’ means ‘significant at less than 1% level’. (3) In the columns 3 and
4, the results are from two stage probit models with test scores squared as the instrument. In column 3, ‘working
hours’ represents the estimated working hours instead of the actual working hours. (4) ‘Residual’ refers to the residuals
from the reduced form working hours equation. (5) 1n the columns 5 and 6, we include test scores squared and ‘mom
work’ dummy just to show whether these variables are related to college attendance. These variables are not included
in the columns 1, 2, 3, and 4. (6) In columns 1. 2, 3, and 4, we include all the control variables in the MNL model

154

 

 

Table 2-4-1-8: Female: Probit for College Attendance

 

(NLS-72)
____Variab'e$._____ __ m 1'). W12) -_ __-___z -13) 23L“I4)__2_S_1_’_.~.,_.(_5_)_2.$E_-_._L§)..2§_P.m-.-.I?.1____m-l§)-___ ____
Working Hours -.007"’ -.004"“ .008 .009 -.004 -.004 -.007"“ -.010”‘
(.002) (.002) (.023) (.023) (.019) (.019) (.002) (.002)
Residuals -.015 -.003
(.023) (.019)
test scores .042‘” 031“" 042"" 042"" 042“" 042”" .024 .042“
(.003) (.004) (.004) (.004) (.004) (.004) (.032) (.004)
test scrores .000
squared (.000)
Mom work .024
(.037)
Eud. Aspiration 1426“”
Two-year (.071)
college
Bud. Aspiration 1936""
Four-year (.060)
coHege
2822.3 4098.4 2808.9 2822.7 2810.0 2822.8 2809.2 2822.7
(.3000) (.4357) (.2986) (.3001) (.2998) (.3001) .2990 (.3001)

 

Notes: (1) Standard errors are in parentheses. (2) “’ means ‘significant between 10% level and 5% level,’ ""’ means

‘significant between 5% level

and 1% level.’ """’ means ‘significant at less than 1% level’. (3) 1n columns 3 and 4, the results are from two stage

probit models with ‘mom work’

as the instruments. (4) 1n columns 5 and 6, the results are from two stage probit models with test scores squared and

‘mom work’ as the instruments.

(5) ‘Residual’ refers to the residuals from the reduced form working hours equation. (6) In columns 7 and 8, we
include test scores squared and ‘mom work’ dummy just to show whether these variables are related to college
attendance. These variables are not included in the column 1, 2, 3, 4, 5 and 6. (7)1n columns 1, 2, 3, 4, 5 and 6 we
include all the control variables in the MNL model.

155

 

 

Table 2-4-1-9: Males: Probit for College Attendance

 

 

(HS&B)
-Variablss--_ ____.---__, 1.1.). 12.) __,_---_...__I3,1 251’-._z_ 1:11.2SP.___- -151251’. __--~-__.(.,6)_25P (7)23? (81...”

Working hours -.010"'” -.007”“ -.011 -.011 -.014 -.014 -.010"" -.010""‘

(.002) (.003) (.012) (.012) (.011) (.011) (.002) (.002)
Residuals .001 .005

(.013) (.012)
High reservation -.005
wage (.056)
Will not work -.000
(.195)
Mom work .024
(.037)

Educational .729‘”
Aspiration (.074)
Two-year
college
Educational 1.243‘"
Aspiration Four- (.063)
year college

1210.2 1624.8 1192.5 1210.3 1193.3 1210.4 1210.3 1214.0

(.2216) (.2975) (.2183) (.2216) (.2185) (.2216) (.2216) (.2217)

 

Notes: (1) Standard errors are in parentheses. (2) "'" means ‘significant between 10% level and 5% level,’ “" means
‘significant between 5% level and 1% level,’ ‘"" means ‘significant at less than 1% level’. (3) 1n columns 3 and 4,
the results are from two stage probit models with the ‘high reservation wage’ dummy and ‘will not work’ dummy as
the instruments. (4) In columns 5 and 6, the results are also from two stage probit models with the ‘high reservation’
dummy, ‘will not work’ dummy and ‘mom work’ dummy as the instruments. (5) ‘Residual’ refers to the residuals
from the reduced form working hours equation. (6) In the columns 7 and 8, we include the ‘high reservation’ dummy,
‘will not work’ dummy and ‘mom work’ dummy just to show whether these variables are related to college attendance.
These variables are not included in the column 1, 2, 3, 4, 5 and 6. (7) In columns 1, 2, 3, 4, 5, and 6 we include all the
control variables in the MNL model.

156

 

Table 2-4-1-10: Females: Probit for College Attendance

 

 

 

 

(HS&B)
Variablss _-_ -0). --_____-.(_2)_-__-_ “43) _ (4).. (5.) (6) (7) L8) _---__.
Working hours - -.0069""‘ -.008 -.008 -.005 -.005 -.007"'" -.007”*
0071"" (.003) (.011) (.011) (.044) (.044) (.002) (.002)
(.002)
Residuals .001 -.002
(.012) (.045)

High reservation -.033
wage (.045)
Will not work -.187

(.157)
Mom work .002

(.004)

Educational .705‘"
Aspiration (.060)
Two-year
coHege
Educational 1.236‘”
Aspiration Four- (.055)
year college

1398.8 1915.4 1398.3 1398.8 1388.8 1398.8 1400.5 .1398.8
(.2098) (.2873) (.2084) (.2098) (.2083) (.20998) .2101 .2090

Notes: (1) Standard errors are in parentheses. (2) “’ means ‘significant between 10% level and 5% level,’ "“’ means
‘significant between 5% level and 1% level,’ “”’ means ‘significant at less than 1% level’. (3) In the columns 3 and
4, the results are from two stage probit models with the “high reservation’ dummy and ‘will not work’ dummy as the
instruments. (4) In columns 5 and 6, the results are from two stage probit models with the ‘high reservation wage’
dummy, ‘will not work’ dummy and ‘mom work’ dummy as the instruments. (5) ‘Residual’ refers to the residuals from
the reduced form working hours equation. (6) In the columns 7 and 8, we include the ‘high reservation wage’ dummy,
‘will not work’ dummy and ‘mom work‘ dummy just to show whether these variables are related to college attendance.
These variables are not included in the column 1, 2. 3. 4. (7) In columns 1. 2, 3, 4, we include all the control variables
in the MNL model.

157

 

4-2 College Major Choice.

Our examination of the relationship between high school work experience and the
choice of college major reveals several interesting points: (1) When we first distinguish
college majors by the academic field and the vocational field, and run the probit model
for the choice of the vocational field, there is no relationship between high school work

experience and the choice of study field except for the 1972 female case. (2) When we

TiT

restrict the sample to those who are in the academic area, and run the MNL model for the

six major alternatives mentioned earlier, the association between high school work

 

experience and the choice of college major is very strong. Students who work more }
during the high school senior year tend to choose business or engineering major and avoid

the liberal arts or education majors. (3) This tendency is stronger among the 1980 senior

cohort than among the 1972 senior cohort, among males than females, and among Whites

than minorities. (4) However, in the 1980 female case, work experience during the junior

year has the opposite relationship with college major choice to work experience during

the senior year. (5) There is no evidence which supports that students who have a certain

unobservable characteristic, such as a tendency to work, work more during the high

school senior year and choose engineering or business majors as long as our instruments

are correct. The high school work experience seems to influence student workers’ college

major choices.

4-2-1 Vocational field versus Academic field
We make a dummy variable ‘the vocational field’ and set ‘the vocational field’

equal to one if a student choose the vocational field and otherwise to zero. With this

158

 

variable as the dependent variable, we run the standard probit model. The results are
tabulated in Table 2- 4-2-1 and 2. (The coefficients of other control variables are not
reported).

For males, after controlling for the background variables, academic ability and the
number of courses of each subject area during high school, there is no significant
relationship between high school work experience and ‘vocational major’ choice during
both years. However, there is a big change for females. In 1972, female students who
work during their senior year are much more likely to choose a vocational major. The
coefficients of ‘work’, ‘heavy work,’ and ‘working hours’ are significantly positive.
‘Work’, ‘heavy work,’ and a ten hour increase in ‘working hours’ are respectively related
to increases of 3 percent, 3 percent and .3 percent in choosing the vocational field (but
marginal effects are not reported.)

In 1980, there is no such relationship shown even for females. The change in the
female case might be explained by the two changes in high school employment and
college attendance. First, more able and well-off students participated in the labor force
and work in 1980 than 1972. Second, the percentage of females who attended four-year

colleges rose from 20 percent in 1972 to 36 percent in 1980.

159

 

Table: 2-4-2-1: Probit for Vocational Major Versus Academic Major

(NLS-72)

 

 

 

 

 

 

 

 

 

Male (2585) Female (2531)
._ (I) (2) (3) (I) (2) (3)
Work .098 .120*
(.086) (.066)
Heavy work .1 13 .107"
(.069) (.061)
Working hours .004 .007”
(.003) (.003)
Grade -.045 -.045 -.045 -.093" -.094" -.093"
(.045) (.045) (.045) (.041) (.041) (.041)
Test scores -.008 -.008 -.008 -.021"‘" -.022"" -.022”‘
(.008) (.008) (.008) (.007) (.007) (.007)
Math course .014 .015 .015 .010 .009 .010
(.024) (.024) (.024) (.020) (.020) (.020)
Science course .026 .025 .025 .099‘" 099"" .099‘”
(.022) (.022) (.022) (.019) (.019) (.019)
English course -.033 -.033 -.033 -.059“ -.059” -.059"
(.030) (.030) (.030) (.025) (.025) (.025)
Business course -.029 -.028 -.028 .035" .035” .035”
(.026) (.026) (.026) (.013) (.013) (.013)
Health course .005 .005 .005 -.026 -.025 -.025
(.059) ( .059) (.059) (.039) (.039) (.039)
x square 33.4 33.4 33.4 116.9 116.9 116.9
R Square (.0191) (.0191) (.0191) (.0469) (.0469) (.0469)
Table: 2-4-2-2: Probit for Vocational Major Versus Academic Major (HS&B)
Male Female
__.___ ._ l 1) (21 (3.).-. ___.(,l,,), _- __3_____.,.Q.), , ,_ t3) w
Work .001 -.056
(.091) (.072)
Heavy work .108 .064
(.086) (.067)
working hours .005 .003
(.003) (.003)
work junior -.060 -.087 -.098 -.O74 -. 104 -.106
(.087) (.086) (.086) (.069) (.068) (.068)
Grade -.l72” -.172" -.172" -. 166‘" -.166"* -. 165*"
(.068) (.068) (.068) (.061) (.061) (.061)
Test scores -.027"" -.027”" -.027"'""' -.027 -.028 -.028
(.006) (.006) (.006) (.006) (.006) (.006)
Math course -.009 -.009 -.009 -.014 -.014 -.014
(.026) (.026) (.026) (.002) (.002) (.002)
Science course -.059“‘ -.059"'"I -.059"” .021 .022 .021
(.024) (.024) (.024) (.019) (.019) (.019)
English course -.001 -.001 -.001 -.046 -.045 -.044
(.034) (.034) (.034) (.032) (.032) (.032)
Business course -.087‘" -.087‘" -.087"* -.010 -.012 -.012
(.024) (.024) (.024) (.016) (.016) (.016)
Health course -.017 -.017 -.017 .001 .002 .002
(.020) (.020) (.020) (.019) (.019) (.019)
X square 242.4 244.2 244.4 176.7 177.0 177.0
R Square (.1462) (.1473) (.1474) .0812 .0813 .0814

 

Notes for Table 2- 4-2-1 and 2: (1) Standard errors are in parentheses. (2) ‘*’ means ‘significant between 10% level
and 5% level,’ ‘"’ means ‘significant between 5% level and 1% level,’ "W" means ‘significant at less than 1% level.’
(4) The coefficients of other variables are not reported.

160

 

 

4-2-2 Major choices among college students in the academic field

The MNL results on the choice of college major among college students in the
academic field are tabulated in Table 2- 4-2-3~8 and Table 2- 4-2-9~l 8. The coefficient
of each variable indicates the effects of each variable on the log odds ratio between the
choice of a certain major of the six categories (business, engineering, health, pre-
professional, science, and education), and the choice of ‘liberal arts and others.’ We
report all the coefficients of all the independent variables only once for each sample in the
case where the high school work experience is represented as ‘working hours.’ In other
cases, we report only the coefficients of the high school work experience variables
(‘work,’ ‘heavy work,’ and ‘work junior’) because they are the variables of current
interest and because the coefficients of other variables hardly change when just changing
the work experience variable.

For the interpretation of these coefficients, as we do in the case of college
attendance choice, the changes in the mean probability in each major choice in the
response to changes in the variables of high school work experience are calculated and
tabulated separately (Table 2- 4-4-3 and 5 and Table 2-4-2-13, l6, 9, and 18.)

In the tables where we report the all the coefficients, for each major choice, we
also report Hausman test statistics for the IIA hypothesis The test statistics are calculated
by comparing the estimates of coefficients of the variables from the full samples and the
estimates of the variables when each major choice (and the observation in that major) is
dropped. When we include all the independent variables, most Hausman tests are not
successful because many of coefficients of the independent variables have large standard

errors and are not significant, which does not satisfy the conditions for the test. Thus, we

161

 

 

also use the restricted MNL model where we include only the work experience, test
scores in math and vocabulary, the number of courses in math and sciences during high
school as the independent variables. The reported test statistics are from these restricted
models. Even in this case, Hausman test statistics are often negative. However, there is no
evidence that these majors are too close to be alternatives to one another. In the cases

where the test statistics are valid, the test statistics are very small and almost all the p

 

values are 100 percent (see Hausman test statistics in Table 2- 4-2-3 and 6, and in the

Table 2- 4-2-9 and 14).")8

 

 

The MNL results show several important characteristics of the determinants of the 9
college major choices. First, the effects of many family background variables do not seem
to be very important in many major choices and are not consistent across samples
compared to students’ ability and high school preparation. After ability measures, high
school grades, the number of courses taken in each subject during high school, and high
school programs are controlled for, the coefficients of family SES, race, number of
siblings, and whether students have any children or not are not significant in many major
choices. Also, the coefficients of these variables are different between gender and
between the two years, 1972 and 1980. Also, the effects of regions and community types
seems to be different across samples.

Second, the effects of different cognitive skills are important and consistent across

samples. Not only mathematical ability but also vocabulary and reading ability matter in

 

9’ We also examine whether there are any changes in the coefficients in the MNL model with all
independent variables when a major and the observation in that particular major is dropped from the
sample. Then there are almost no changes in the coefficients.

162

the choice of college major, while Paglin and Rufolo (1990) single out only the
mathematical ability as one important determinant of the choice of college major in terms
of cognitive skills. Students with high vocabulary ability tend to choose ‘liberal arts and
others’ majors rather than business, engineering, or science majors. Both male and female
students with high math ability tend to choose science and engineering. Male students
with high math ability also tend to choose a business major.

Third, high school preparation is very important in the determination of college
majors. Students who have taken more high school mathematics courses tend to choose
business, engineering, and science majors. In the 1972 female case, the number of
mathematics courses is very important in the various major choices such as business,
engineering, and science. The number of science courses is one of the significant positive
factors in the choices of engineering, health, and science major for both male and female
students, while the number of English courses is one of the significant negative factors in
choosing the same such majors. Students who have taken more business courses are more
likely to choose a business major.99

Fourth, most importantly, even after different kinds of abilities and the numbers of
courses taken in different subjects are controlled for, ‘working hours’ in the senior year
turns out to be one of the key determinants for the choice of college majors. Most signs
of the coefficients of ‘working hours’ are positive and significant in the choices of

business and engineering. '00 Also the coefficients of other variables of work experience

during the senior year, ‘work,’ and ‘heavy work’ are almost all positive and significant in

 

99 However, we can not exclude a possibility that high school students who are going to choose an
engineering or a business major might take more courses in mathematics.

163

 

 

the choices of business or engineering. The positive and significant coefficients of these
variables indicate the strong association between the high school work experience and
lucrative (practical) major choices even afier family background, different cognitive
skills, and high school preparation are controlled for. Such an association is more evident
in 1980 than in 1972, and among male students than among female students.

In 1972, for males the coefficients of ‘working hours’ are significant and positive
in the choice of business major (see Table 2- 4-2-3) and also the coefficients of ‘work’
and ‘heavy work’ are significant and positive in the choice of business major (Table 2-4-
2-4). The coefficients of these variables are also positive in the choices of engineering,
health, pre-professional, and science, even though only the coefficient of ‘work’ is
significant for only the choice of health. These coefficients imply a positive association
between high school work experience and more lucrative or practical major choices, and
a negative association between high school work experience and liberal arts or education.

In Table 2- 4-2-5, the changes in the mean probability in each major choice in the
response to changes in the variables of high school work experience are reported for 1972
males. ‘Work’ is on the average associated with increases of around 5 percent and 2
percent in the probability of the choices of business major, and health major respectively
and decreases of around 4.5 percent of the choice of liberal arts, and 2.5 percent in the
choice of education major respectively. ‘Working hours’ and ‘heavy work’ shows similar
associations. The more a male student works during his high school senior year, the more

likely he is to choose a business major rather than ‘liberal arts’.

 

’00 In 1980, without controlling these factors, even for females the coefficient of ‘working hours’ for the
choice of business major is positive and significant.

164

 

 

Table 2-4-2-3: Male(2310): MNL Results for Major Choice

(NLS-72)

 

 

 

Variables Business Engineering Health Pre- Science Education
~ professional ‘
Constant -.106 -3.548 -3.262 -6.409 -S.414 1.890
(.719) (1.027) (.960) (1.476) (1.065) (.761)
Working hours 016‘” .007 .010 .010 .006 -.009
(.006) (.008) (.007) (.012) (.008) (.007)
Vocabulary score -.037" -.036“‘ -.035”‘ -.030‘ —.024" -.023
(.009) (.012) (.011) (.018) (.011) (.010)
Reading score -.040"‘ .003 -.017 -.025 .020 -.017
(.010) (.014) (.013) (.020) (.014) (.011)
Math score .035‘” .032” .015 .007 .026‘ -.014
(.011) (.016) (.015) (.022) (.016) (.012)
Math course .162‘” .442‘” .237‘” .047 .175‘“ .035
(.045) (.068) (.060) (.086) (.063) (.048)
Science course -.040 .169‘” .358‘" -.088 .456‘" .009
(.042) (.054) (.051) (.079) (.053) (.046)
English course -.058 -.282‘" -.230"” .049 -.336”’ .041
(.056) (.079) (.073) (.102) (.077) (.059)
Business course 276'" .031 -.033 .570 .054 .067
(.047) (.084) (.078) (.376) (.090) (.056)
Health course .068 .149 .057 .138 -.090 .096
(.111) (.125) (.136) (.166) (.174) (.111)
Grade .095 .224‘ .397‘" .291 ‘ .385‘" .060
(.085) (.114) (.109) (.172) (.115) (.91)
SES-medium .367 -.358 .700" .319 .293 -.064
(.217) (.268) (.304) (.466) (.299) (.212)
SES-high .545 -.191 1033‘” .333 .395 -.462‘
(.228) (.277) (.315) (.482) (.309) (.235)
Black -.044 -.689 .462 .671 .337 -.508
(.267) (.452) (.338) (.499) (.386) (.292)
Hispanic -.067 -.332 .513 . -.837 -.228
(.370) (.522) (.481) (.788) (.375)
Asian .059 -1.1 l l .207 .575 .333
(.512) (.814) (.558) (.504) (.502)
Number of siblings .013 -.127‘“ -.024 .061 .011 -.027
(.036) (.050) (.046) (.064) (.045) (.040)
Any dependent .151 .163 .077 -.531 .019 .329
(.271) (.445) (.378) (.763) (.447) (.278)
High school -.151 .145 .094 .998" -.077 -.275’
Academic (.160) (.242) (.222) (.407) (.234) (.171)
High school .671” 1454"" .660 .556 -.515 -.138
Vocational (.287) (.417) (.459) (.834) (.774) (.339)
Small town .813‘” .140 .470 .110 .181 .156
(.171) (.232) (.220) (.367) (.221) (.188)
Urban .578‘“ .314 .626‘” .400 .222 -.332
(.180) (.227) (.216) (.444) (.225) (.212)
Suburban .732‘” .340 .690‘” .643‘" .240 .146
(.175) (.220) (.211) (.319) (.215) (.194)
North central -.017 .560” .813‘” .227 .700 .106
(.187) (.242) (.229) (.384) (.230) (.210)
South -.I38 .318 .556" .624 .241 -.173
(.177) (.233) (.222) (.437) (.232) (.203)
West -.l63 .837‘” .495' .481 780"“ .021
(.215) (.265) (.268) (.411) (.256) (.236)
Hausman Test
Statistics I! 1.99 -.67 .29 .72 .50 ~5.77
R Square .1073
_(3; square) (873.3)

 

Notes: (1) ‘Socio-economic status: low' in social economic status is excluded. Excluded race is ‘White’. ‘General program’ category
for high school program. ‘North Eastem' is excluded for the region. Excluded community of residence is the ‘Rural Farming’
category. (2) ‘Native American’ and ‘other’ categories for the ‘race’ are included in the regression and is not reported. (3)

indicates that the coefficient has huge standard errors due to the small number of individuals who chose that major.

165

 

 

Table 2-4-2-4: Male: MNL Results for Major Choice (NLS-72)

 

 

 

 

(+10)

Note: The numbers are percents.

 

 

Variables Business Engineerin Health Pre- Science Education X square
professiona (R Square)
1

Work 435““ .162 .381” .096 .208 -.041 876.1

(.159) (.174) (.186) (.281) (.189) (.166) (.1070)

Work more than .212“ .1 15 .202 .248 .084 -.086 870.5

15 hours (.128) (.168) (.158) (.248) (.166) (.145) (.1063)

Note: see notes in Table 2- 4-2-3

Table 2-4-2-5 : Male Changes in the Mean Probability of Choosing Each Major (NLS-72)

Variables Liberal Business Engineering Health Pre- Science Education

Arts and professional

Work -4.6 +4.8 -.8 +2.1 +2 +3 -2.4

Heavy work -2.3 +3.4 +.2 +1.2 +.6 +.1 -l.9

Working hours(+5) -| .7 +1.1 +.00 +.2 +.l +0 -.8

Working hours -2.4 +2.1 +06 +4 +2 +0 -1.6

In the 1972 female case, the association is much weaker. The coefficient of

‘working hours’ is positive in the choices of business and health majors while it is

negative in the choices of engineering, pre-professional, or science. However, none of the

coefficients of ‘working hours’ are significant ( see Table 2- 4-2-6). The coefficient of

‘work’ is positive in the choices of business, engineering, health, and pre-professional

while is negative in the choices of science and education majors. None of these variables

are statistically significant. The coefficient of ‘heavy work’ is positive in the choices of

business, health while is negative in the choices of engineering, pre-professional and

science. Only for the choice of business, is the coefficient of ‘heavy work’ significant.

I66

 

 

Nonetheless, the average changes in the probability of choosing each major show

that female students with heavy work experience are more likely to avoid the ‘liberal arts

and others’ and more likely to choose business or health majors than other students (see

Table 2- 4-2-8). For example, ‘heavy work’ is associated with around a 3 percent

decrease in the probability of choosing the ‘liberal arts and others’ and around a 2 percent

increase in the probability of choosing either business or health majors.

 

 

167

Table 2-4-2-6 : F emale(2040): MN L Results for Major Choice (NLS-72)
Variables Business Engineering Health Pre- Science Education
- . _ professional
Constant
-.398 3.923 -2.378 -4.017 -7.787 2.626
(1.048) (2.434) (1.002) (2.340) (1.340) (.624)
Working hours .012 -.013 .013 -.038 -.009 .000
(.010) (.024) (.009) (.027) (.011) (.008)
Vocabulary score -.016 -.057 -.01 l .029 -.013 -.028"
(.014) (.030) (.012) (.031) (.015) (.008)
Reading score -.044"‘ .021 .003 .001 .001 -.020"
(.015) (.035) (.014) (.036) (.018) (.009)
Math score .004 .010 -.01 5 .129 .056" -.008
(.016) (.038) (.017) (.147) (.021) (.010)
Math course .188‘” .404’" .228‘" .129 .182" .128‘”
(.067) (.152) (.062) (.147) (.074) (.040)
Science course .048 .364‘" .371‘" .081 .509‘" .072"
(.067) (.126) (.032) (.136) (.061) (.039)
English course -.135‘ -.167 -.411 -.086 -.203” -.018
(.081) (.184) (.077) (.178) (.089) (.046)
Business course 176“" -.005 .020 -.077 -.l 15 .000
(.038) (.012) (.048) (.125) (.075) (.029)
Health course .070 -.830 .006 .145 -.076 -.086
(.099) (.824) (.094) (.181) (.106) (.075)
Grade .109 .268 421"" .304 .491‘"I -.067
(.132) (.302) (.126) (.302) (.159) (.079)
SES-medium .001 -.627 .139 -.418 -.232 -.287’
(.284) (.654) (.278) (.648) (.313) (.177)
SES-high .148 .004 .292 -.831 -.337 -.492"
(.315) (.663) (.293) (.710) (.333) (.193)
Black .167 -.378 .586" -l .092 .902'”‘“I -.789‘”
(.323) (.809) (.313) (.897) (.383) (.223)
Hispanic .034 .387 .721 .546 .858 -.338
(.501) (1.166) (.502) (.905) (.653) (.349)
Asian .220 .450 .736 . .351 .020
(.684) (1.138) (.523) (.714) (.443)
Number of siblings .018 -.031 .005 265*” .033 .022
(.052) (.124) (.048) (.099) (.057) (.032)
Any dependent .507 . .011 . -.912 -.079
(.394) (.518) (1.066) (.299)
High school -.215 -l 207‘” .130 -.396 .190 -.l4l
Academic (.241) (.521) (.242) (.536) (.324) (.144)
High school 1031‘" .885 -.499 -.224 -1.153 -.037
Vocational (.311) (.758) (.578) (1.129) (1.176) (.260)

 

 

 

Small town .270 -.3 16 .227 .424 -.257 .220

(.246) (.647) (.221) (.655) (.284) (.148)
Urban -.240 .561 -.109 1048‘ -.063 -.107
(.267) (.558) (.233) (.597) (.282) (.157)
Suburban .103 .569 -.498 1.118‘ .532" .115
(.275) (.573) (.578) (.623) (.262) (.164)
Nonh central .401 .187 .720‘” .414 .683” .184
(.327) (.657) (.244) (.671) (.276) (.174)
South .735” .233 .214 .835 .377 .291
(.304) (.555) (.230) (.568) (.263) (.156)
West .796" .343 .630” -.222 .349 .143
(326) (.683) (.267) (.743 (.332) (.189)
-.45 .18 11.6 -3.2 -2.01 3.69
R Square .1184
(1 square) (71077)

 

Notes: (1) Standard errors are in parentheses. (2) “’ means ‘significant between 10% level and 5% level,’ “" means ‘significant
between 5% level and 1% level,’ "’”’ means ‘significant at less than 1% level’. (3) ‘Socio-economic status: low' in social economic
status is excluded. Excluded race is ‘White’. ‘General program’ category for high school program, ‘North Eastem’ for the region are
excluded. Excluded community of residence is the ‘Rural Fanning’ category. (4) ‘Native American' and ‘other’ categories for the
‘race’ are included in the regression and is not reported. (5) indicates that the coefficient have huge standard errors due to the small
number of individual 5 who chose that major.

 

 

 

 

Table 2-4-2-7: Female: MNL Results for Major Choice (NLS-72)
Variables Business Engineering Health Pre- Science Education R Square
P’9i.?§.5_191‘31 (1 Square)
Work .033 .055 .246 .060 -.216 -.048 .1186
(.203) (.436) (.184) (.447) (.207) (.121) (712.1)
Heavy work .353‘ -.335 .271 ~628 -. 172 .172 .1178
(.205) (.519) (.189) (.560) (.238) (.125) (.7071)

 

Note: See the notes in Table 2-4—2-6.

 

 

Table 2-4-2-8: Female: Changes in the Mean Probability of Choosing Each Major (NLS-72 )
Variables Liberal Arts and Business Engineering Health Pre-professional Science Education
. __ .. Bums. . . _ _

Work +1 +2 -.8 +3.4 +.9 -l.6 -1.3
Work more than 15 -3.3 +1.9 -.4 +2.0 —.7 -l.6 +2.3
hours
Working hours(+5) -.2 +4 -.1 +.7 -.04 -.3 -.2

-.4 +.7 -.1 +1.3 -.07 -.7 -.3

Working hours (+10)
Note: The numbers are percents.

For the HS&B samples, the results are tabulated in the Table 2- 4-2-9~13 for
males and for females, in the Table 2- 4-2-14~l 8. Here, ‘work,’ ‘heavy work,’ and
‘working hours’ represent the work experience during the senior year and ‘work junior’
and ‘working hours junior’ represent the work experience during the junior year. In

general, the association between high school work experience and college major choice

168

are stronger in 1980 than 1972. In general, the work experience during high school is
positively associated with the choices of business or engineering majors and negatively
associated with liberal arts and education majors. Also, in the male case, differently from
1972, the work experience is negatively associated with the choice of science major.
These associations are much stronger among males than females.

In the 1980 male case, the coefficients of ‘work,’ ‘heavy work,’ or ‘working
hours’ in the senior year are almost always significant and positive in the choices of
business or engineering in preference to ‘liberal arts and others’ '0' (see Table 2- 4-2-9
and 10). Table 2-4-2-1 1 shows that ‘work’ is associated with a 5 percent decrease in the
probability of choosing liberal arts, while associated with about a 5 percent increase in the
probability of choosing business. ‘Heavy work’ is associated with a 5 percent decrease in
the probability of choosing liberal arts, and is associated with increases of about 5 percent
, 3 percent increases in the probability of choosing business, and engineering respectively.
‘Heavy work’ is also associated with a 2 percent decrease in the probability of choosing
science and education. A ten hour increase in working hours per week during the senior
year is associated with a 2.6 percent decrease in the probability of choosing liberal arts
and with a 2.2 percent, 1.3 percent increase in the probability of choosing business and
the probability of choosing the engineering majors.

When we use the weights to accommodate the over-representation of Blacks and
Hispanics, these associations become much stronger. The coefficients of ‘working hours,’
‘work,’ and ‘heavy work’ are more than doubled (see Table 2- 4-2-12). Table 2- 4—2-13

shows that ‘work’ is associated with an 8 percent decrease in the probability of choosing

169

 

 

liberal arts, while associated with about a 7 percent increase and a 9 percent increase in
the probability of choosing business and the probability of choosing engineering
respectively. Also, it is associated with a 5 percent decrease in the probability of choosing
science, and is associated with about a 3 decrease in the probability of choosing
education. ‘Heavy work’ is associated with a 11 percent decrease in the probability of
choosing liberal arts, and is associated with increases of about 8 percent, and 10 percent
in the probability of choosing business, and engineering respectively. ‘Heavy work’ is

also associated with decreases of 3 percent and 4 percent in the probability of choosing

 

science and education respectively. A ten hour increase in working hours per week

 

during the senior year is associated with a 5.6 percent decrease in the probability of
choosing liberal arts and with increases of 2.8 percent, and 4.6 percent in the probability
of choosing business and engineering majors respectively. These results using weighted
data indicates that the association between high school work experience and college
major choice is much stronger among Whites than minorities. '02

However, in all these regressions, the coefficients of ‘work junior’ are negative in
the choices of business or engineering, even though they are not significant after
controlling for the work experience during the senior year. ‘Work junior’ might capture
the different characteristics of those who have worked during the senior year. Without
controlling for the work experience during the senior year, the coefficients of ‘work

junior’ are positive in the choices of business or engineering (not reported).

 

'0' Even though the coefficient of ‘working hours’ is positive, significant and bigger than those in the
choices of business or engineering, the number of students who choose that major is small (just 17) and the
meanings of the coefficient is relatively not important.

'02 When the samples are separated in two groups ( Whites and minorities) and the same regressions are run,
the results are very contrasting (not reported). While there is no significant relationship between high school
work experience and college major choice for only minorities, there are strong associations for Whites.

170

 

Table 2-4-2-9: Male (1520): MN L Results for Major Choice

(HS&B)

 

 

Variables Business Engineering Health Pre- Science Education
professional
Constant -.287 -2.979 -6.857 -5.990 -4.810 1.474
(.809) (.851) (1.924) (2.768) (1.351) (1.130)
Working hours 020“” .015" .007 .062" -.001 -.000
(.007) (.007) (.015) (.024) (.010) (.011)
Work junior -.264 -.200 -.196 .002 -.254 .127
(.169) (.166) (.356) (.638) (.228) (.262)
Grade .037 .190 .632 -.105 276 -.170
(.134) (.137) (.317) (.458) (.199) (.197)
Vocabulary score -.024“ -.030"" -.008 .056 -.001 -.053”"
(.010) (.010) (.024) (.038) (.016) (.014)
Reading score -.007 .006 -.015 -.052"‘ -.005 -.014
(.010) (.011) (.024) (.029) (.016) (.017)
Math score .002 .022” .003 -.01 1 035*” .010
(.008) (.009) (.017) (.027) (.015) (.012)
Math course .105" .379’“ -.007 -.213 .225" .063
(.054) (.063) (.137) (.198) (.091) (.079)
Science course -.077 .137"‘ 507"" .396“ 312"" -.010
(.047) (.049) (.135) (.186) (.077) (.071)
English course .013 -.226‘" -.124 -.102 -.376"" .011
(.080) (.083) (.178) (.257) (.108) (.110)
Business course .312‘" -. 149” -.078 .201 -.095 -.029
(.045) (.055) (. 120) (.155) (.081) (.075)
Health course -.184"" -.045 -.039 -.057 -.040 -.158""l
(.044) (.041) (.096) (.139) (.062) (.062)
SES low-medium -.046 .101 .361 . .041 .055
(.233) (.234) (.512) (.341) (.330)
SES high- -. 148 .029 -1.534"' .225 -.l49 -.064
medium (.223) (.222) (.824) (.636) (.326) (.321)
SES high .099 -.155 .869" -.706 .096 -.299
(.222) (.223) (.479) (.790) (.367) (.339)
Black .089 .259 -.233 1.140 .021 -.393
(.224) (.226) (.610) (.723) (.331) (.346)
Hispanic .066 .230 1.270‘” .338 .010 -.360
(.221) (.223) (.417) (.832) (.321) (.332)
Asian .353 1044"" 1.070 . 1.221 "" .303
(.435) (.371) (.733) (.443) (.540)
Number of .102" .094“ .055 271”" .182“* .048
siblings (.045) (.043) (.104) (. 100) (.056) (.065)
High school -.129 .261 .020 -.934 .082 -.332
Academic (.184) (.196) (.427) (.655) (.286) (.262)
High school -.402 .246 -.209 -.220 .108 -.332
Vocational (.293) (.318) (.856) (.815) (.499) (.407)
Urban .577” .202 .745 .753 .269 -.041
(.228) (.226) (.517) (.831) (.297) (.314)
Suburban .430" . 128 .645 .689 -.173 -.126
(.198) (.194) (.470) (.759) (.267) (.320)
North central .482" 732‘" 1.210" .819 .261 1.133""
(.224) (.218) (.516) (.893) (.310) (.396)
South .576‘" .825‘“ 1458*" .622 .837“ 1.003"
(.218) (.215) (.487) (.845) (.290) (.391)
West .299 .602 .148 .361 .464 1244*"
(.256) (.246) (.617) (.980) (.336) (.419)
Hausman Test -4.67 9.33 1.39 -.89 .75 -22.7
(p value) (.9994)
R Square .1346
()9 square) (662.5)

 

171

 

 

 

 

Notes: (1) Standard errors are in parentheses. (2) ""’ means ‘significant between 10% level and 5% level,’ ‘"’ means
‘significant between 5% level and 1% level,’ ""“’ means ‘significant at less than 1% level.’ (3) ‘Socio-economic
status: low’ in social economic status is excluded. Excluded race is ‘White.’ ‘General program’ category for high
school program is excluded. ‘North Eastern’ is excluded for the region. Excluded community of residence is the ‘Rural
Farming’ category. (4) ‘Native American’ and ‘other’ categories for the ‘race’ are included in the regression and is not
reported. Also, the coefficients of ‘child’ is not reported here. (5) ‘.’ indicates that the coefficient has huge standard
errors due to the small number of individuals who chose that major.

 

Table 2-4-2-10: Male: MNL Results for Major Choice (HS&B)
Variables Business Engineering Health Pre- Science Education x square
____ _. _ - __m- _. _._--__-_.____ - _. ._ ___-.. __-.pr9,f.e.§§ion.al-__-- m-.. -.-_._-_-_.<B_§_‘!E‘§’F1 -
Work .411‘"l .170 .072 .719 .021 .384
(.180) (.173) (.371) (.717) (.236) (.271)
Workjunior -.254 -. 165 -.176 .259 -.270 .023 651.4
(.171) (.169) (.361) (.632) (.231) (.264) (.1324)
Heavy work .450‘" .309" .015 1.206" -.079 -.111
(.157) (.156) (.337) (.594) (.210) (.237)
Work junior -.238 -. 183 -.152 .162 -.240 .149 661.5
(.167) (.164) (.351) (.627) (.224) (.259) (.1344)
Working hours .020‘” .016“I .003 .062 -.240 .149
(.007) (.007) (.015) (.026) (.224) (259)
Working hours -.006 -.009 .007 .002 -.012 .009 663.3
junior (.07) (.007) (.015) (.025) (.010) (.010) (.1348)
Average .014‘ .007 .009 .060 -.012 .007 651.3
working hours (.008) (.008) (.018) (.026) (.012) (.012) (.1324)

 

Note: (1) See notes in the Table 2- 4-2-9.

 

 

Table 2-4-2-1 1: Male: Changes in the Mean Probability of Choosing Each Major (HS&B)
Variables Liberal Business Engineering Health Pre- Science Education
Arts and professional

0ther .______----_._ _.
Work -5.0 +4.6 +2 -.2 +.4 -l.2 +1.3
Heavy work -5.1 +5.6 +3.2 -.5 +1.0 -2.2 -2.0
Working hours -l.3 +1.1 +.7 -.04 +3 -.4 -.3
(5, 10) -2.6 +2.2 +1.3 -.09 +.7 -.9 -.6
Work junior" +3.2 -2.8 -1.3 -.1 +3 -.1 +1.7
Working hours +5 -.3 -.6 +0 -.3 +5 +.l
junior(5, 10) +1.0 -.6 -l.2 +.1 -.6 +9 +2
Average -.8 +8 +3 -.0 +3 -.6 +.l
working hours -1.6 +1.6 +6 -.1 +.7 -1.2 +.1

 

Notes: (1) The numbers are percent. (2) The changes in the probability in response to ‘work junior’ changes are
calculated after the work experience in the senior year is controlled for.

172

T

.' ' .Y.

 

 

Table 2-4-2-12: Male: Weighted MNL Results for Choice of Each Major (HS&B)

 

 

Variables Business Engineering Health Pre- Science Education x square
professional (R Squarelm-
Work .651" .687“ .285 1.742 .267 -.203
(.214) (.255) (.706) (.901) (.326) (.359)
Work junior -.248 -.160 -.300 -.702 .243 .353 68760.9
(.248) (.237) (.572) (.766) (.308) (.378) (.1507)
Heavy work .802‘” .841‘" -.398 3.348” .114 -.237
(.225) (.218) (.574) (. 100) (.309) (.334)
Work junior -.228 -.139 -. 138 -.820 .144 -.237 76041.5
(.243) (.229) (.564) (.685) (.300) (.334) (.1582)
Working hours .036‘” 042“” .016 .153 .005 -.016
(.010) (.010) (.028) (.041) (.015) (.015)
Working hours -.228 -.139 -.138 -.820 .144 .356 65422.40
junior (.243) (.229) (.564) (.685) (.300) (.370) (.1601)

 

Note: (1) See notes in Table 2- 4-292.

 

 

 

Table 2-4-2-13: Male: Chgges in the Mean Probability of Choosing Each Major with Weights (HS&B)

Variables Liberal Business Engineering Health Pre- Science Education
Arts and professional

Work -8.1 +7.0 +8.7 +.1 +1.0 -5.1 -3.3

Heavy Work -1 1.1 +8.0 +9.7 -2.3 +2.4 -2.7 -3.9

Working Hours -2.6 +1.5 +2.4 -.1 +.7 -.7 -l.0

(5, 10) -5.3 +2.8 +4.6 -.2 +1.7 -1.5 -1.8

 

Note: (1) The numbers are percents. (2) ‘Workjunior’ is controlled for.

For the case of 1980 females, from Tables 2-4-2-14, 15, 16, 17, and 18, we can
observe several characteristics of high school work experience and college major choices.
First, even for females, work experience is a key determinant of female students’ choices
of college major. Second, however, the association is much weaker than in the male case.
Third, the negative association between the work experience during the junior year and
business or engineering majors is much stronger than for males. Fourth, even in the
female case, these associations seem to be much stronger among Whites than minorities.

The coefficient of ‘working hours’ is positive and significant only in the choice of
the engineering major. The coefficients of ‘heavy work’ are positive and significant in the
choices of business and engineering. ‘Heavy work’ is associated with about a 3 percent

decrease in the probability of choosing liberal arts, while is associated with increases of

173

 

about 3 percent, and 2 percent in the probability of choosing business, and engineering
respectively. A ten hour increase in working hours per week during the senior year is
associated with a 1 percent decrease in the probability of choosing liberal arts and with a
1 percent increase in the probability of choosing an engineering major. The coefficients
of ‘work’ in the senior year are not significant in almost all cases.

However, when the weighted data are used, these associations are much stronger.
‘Work’ is associated with a 3.6 percent decrease in the probability of choosing liberal
arts, while is associated with about a 3.4 percent increase in the probability of choosing
engineering. ‘Heavy work’ is associated with a 4 percent decrease in the probability of
choosing liberal arts, while associated with increases of about 3.9 percent , and 4.3
percent increases in the probability of choosing business and engineering respectively.
‘Heavy work’ is also associated with a 2.4 percent decrease in the probability of choosing
science. A ten hour increase in working hours per week during the senior year is
associated with a 1.4 percent decrease in the probability of choosing liberal arts and with
increases of 1.0 percent and 2.1 percent in the probability of choosing business and
engineering majors respectively. Again, these results using weighted data indicate that the
association between high school work experience and the choice of college major is much
stronger among Whites than minorities.’03

However, in all these regressions, regardless of holding the work experience

during the senior year constant, the coefficients ‘work junior’ are negative and significant

in the choices of business or engineering. However, when we include ‘working hours

 

'03 When the sample are separated by two groups (Whites and minorities) and the same regressions are run,
the results are once again very contrasting. While for minorities there is no significant relationship between
high school work experience and college major choice, there is strong association for Whites.

174

 

 

junior’ instead of ‘work junior’, the coefficients of ‘working hours junior’ are not

significant. We suspect that ‘work junior’ could represents a certain characteristic of

students who work during their junior year.

 

175

 

Table 2-4-2-14: Female (1986): Male: MNL Results for Major Choice (HS&B)
Variables Business Engineering Health Pre- Science Education
.___.__-____ _-_.. __ ._ Wm profgsglonal
Constant .668 -1.612 -2.449 -3.216 -6.080 2.059
(.698) (1.206) (1.113) (2.112) (1.283) (.851)
Working Hours .005 .017"I -.005 .006 .006 .004
(.006) (.009) (.010) (.010) (.010) (.007)
Working Junior -.338" -.422‘ -.270 -.176 -.176 -.007
(.132) (.194) (.202) (.213) (.213) (.163)
Grade .130 .192 .270 .338 .338‘ -.150
(.117) (.181) (.184) (.206) (.206) (.138)
Vocabulary -.034‘" -.051”‘ -.041”' -.016 -.016 -.040"‘"
Score (.009) (.013) (.013) (.016) (.016) (.010)
Reading Score -.006 .012 .000 -.009 .009 -.009
(.008) (.013) (.013) (.016) (.016) (.010)
Math Score .006 .041‘” -.001 .002 .043” .004
(.006) (.013) (.009) (.018) (.015) (.008)
Math Course .126‘” .436‘" .136‘ .096 .105 .058
(.042) (.073) (.070) (. 120) (.077) (.051)
Science Course -.071 .018 398"" -.180* .364‘“ -.017
(.038) (.055) (.065) (.090) (.070) (.046)
English Course -.096‘ -.333‘" -.075 -.063 -.198 .013
(.072) (.107) (.121) (.240) (.130) (.091)
Business Course .318 .069 .054 .018 .056 -.002
(.032) (.048) (.052) (.240) (.058) (.039)
Health Course .036 .002 .068 -.I 19 .007 .063
(.036) (.057) (.052) (.119) (.064) (.042)
SES Low- .049 .022 .441 .572 -.357 -.144
Medium (.184) (.267) (.272) (.502) (.317) (.221)
SES High- .131 .101 .111 .176 -.399 -.009
Medium (. 187) (.268) (.296) (.559) (.317) (.223)
SES High .072 -.672“ -.199 .327 -.162 -.442"‘
(.085) (.285) (.294) (.519) (.282) (.223)
Black -. 183 -.090 -.236 .773' .799’" -.985""
(.188) (.280) (.287) (.469) (.306) (.246)
Hispanic .080 -.056 .030 .109 .570‘I -.290
(.183) (.277) (.285) (.508) (.307) (.218)
Asian -.091 -.290 .425 . .341 4.144"
(.343) (.472) (.443) (.458) (.561)
Siblings .035 .014 -.061 .001 -.023 -.038
(.038) (.056) (.060) (.112) (.064) (.051)
High School .091 .059 -.01 1 .640 .061 .109
Academic (.156) (.236) (.242) (.454) (.277) (.182)
High School 650“" .453 .027 -.238 -.923 -.241
Vocational (.210) (.343) (.384) (.826) (.645) (.291)
Urban -.047 -.334 -.002 -.156 .257 -.496“
(.186) (.280) (.276) (.590) (.310) (.217)
Suburban -.032 -.098 -.362 .615 -.001 -.527"‘
(.163) (.235) (.250) (.492) (.282) (.185)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

176

North Central -.271 .001 960"“ -l .175 .040 .239
(.188) (.286) (.313) (.623) (.294) (.234)
South .058 .719” 1038"" .368 .222 568”"
(.181) (.263) (.367) (.453) (.287) (.226)
West -.325 .197 .702“ -.137 .451 .235
(.205) (.316) (.358) (.544) (.318) (.253)
Hauman Test -6.26 1.74 -3.21 -.71 1.95 -48.3
R Square .1 141
(x sipiare) 748.8
Notes: See notes in Table 2-4--2-9.
Table 2-4-2-15: Female: MNL Results for Major Choice (HS&B)
Variables Business Engineering Health Pre- Science Education 1 square
______ ~___ __ professional (R Square)____
Work -.028 .099 .213 .083 -.089 .126
(.142) (210) (.224) (.400) (.230) (.176)
Workjunior -.303" -.360" -.350‘ .126 -.122 -.018 745.7
(.133) (.196) (.204) (.375) (.218) (.163) (.1136)
Heavy work .208‘I .385” -.081 -.094 .018 .088
(. 130) (. 190) (.206) (.360) (.211) (.157)
Workjunior -.353" -.412" -.282 .168 -.152 -.005 750.3
(.132) (.193) (.200) (.369) (.211) (.161) (.1143)
Working hours .005 .019" -.007 .001 .003 .005
(.006) (.009) (.010) (.017) (.010) (.008)
Working hours -.008 -.017 -.006 .004 .006 -.004 743.6
junior (.006) (.010) (.010) (.018) (.010) (.008) (.1133)
Ave Working -.004 .002 -.013 .005 .008 .001 736.4
hours (.007) (.011) (.012) (.020) (.012) (.009) (.1122)
Note: See notes in Table 2- 4-2-10.
Table 2-4-2-16 : Female: Changes in the Mean Probability of Choosing Each Major (HS&B)
Variables Liberal Business Engineering Health Pre- Science Education
Arts and professional
____ ________ ___ Others __..._.. .
Work -.8 -1.4 +5 +1.3 +.l -.7 +1.2
Heavy work -2.8 +2.7 +2.3 -1.2 -.4 -.1 -.1
Working hours -.5 +.2 +5 -.4 -.0 -.1 +0
(5, 10) -l.0 +.4 +1.0 -.8 +6 -.2 +0
Workjunior +4.7 -4.6 -1.9 -.8 +6 +0 +2.0
Working hours +6 -.6 -.5 -.1 +.l +3 +0
junior (5, 10)
+1.2 -1.1 -1.0 -.2 +.2 +.5 +.2
Ave Working +.2 -.4 +.l +.1 +.1 +.3 +.2
hours
+.4 -.9 +.2 +.2 +.l +.6 +.4
Note: See notes in Table 2- 4-2-1 1
Table 2-4-2-17: Female: Weighted MNL Results for Major Choice (HS&B)
Variables Business Engineering Health Pre- Science Education
__ __-,______--_._ , __ 2...... yrotsssisnal -
Work .130 .648" .301 .378 -.267 .040
(.197) (.325) (.326) (.624) (.355) (.254)

 

 

 

 

 

 

Work in Junior -.499" -.480‘ -.063 .227 .000 .094 50946.0
(.193) (.289) (.282) (.480) (.295) (.234) (.1555)

Heavy Work .327‘ .749” -.040 -.208 -.096 .046
(. 189) (.291) (.285) (.493) (.302) (.222)

Work in Junior -.541” -.482" .010 -.385 -.062 .095 45571.1
(.192) (.292) (.276) (.539) (.294) (.234) (.1569)

Working Hours .010 .033” -.004 .004 .004 -.002
(.009) (.014) (.012) (.021) (.016) (.010)

Work junior -.524”" -.500‘ .017 .358 -. 100 .116 47681.0
(.193) (.296) (.277) (.505) (.294) (.236) (.1562)

Note: See notes 2-4-2-10.

Table 2-4-2-18: Female: Changes in the Mean Probability ofChoosing Each Major with Weights (HS&B)

Variables Liberal Business Engineering Health Pre- Science Education
Arts and professional

_---_W-__--_-_.ch¢rs ‘. ,

Work -3.6 +.4 +3.4 +1.5 +.4 -2.4 -.8

Heavy work -3.9 +3.9 +4.3 -1 .2 -.7 -l.4 -1.2

Working hours -.7 .6 +1.1 -.3 -.1 +0 -.5

(5, 10) -1.4 +1.0 +2.1 -.6 -.2 +0 -.9

 

Note: See notes 2- 4-2-1 1.

From these results of the four samples, male and female in both years (1972 and

1980), we can conclude that among college students who are in the academic field,

students who have worked especially long hours during the high school senior year tend

to choose college majors which can lead to high paying jobs, such as business or

engineering, and to avoid liberal arts and education majors.

However, these associations are much stronger in 1980 than 1972, and among

males than females. The stronger association in 1980 than 1972 seems to be related to the

changes in the characteristics of student workers observed in Chapter 1. As we see in

Chapter one, more students with high academic ability held jobs and worked more often

in 1980 than in 1972. In 1972, even if we assume that the high school work experience

might make students more practical, it might be difficult for students with low academic

ability to choose business or engineering because the courses being taught in these majors

177

 

 

could be much more technical and require higher academic ability than courses being
taught in the liberal arts and education.

The weaker association among females might be explained by that for female
students the jobs in the health and education fields might be also attractive because many
females work in these areas. The coefficients of work experience variables are often

significant among the choices of majors such as health and education.

4-2-3 Probit results

By using the probit model, we test whether students with work experience tend to
choose a ‘practical major.’ ‘Practical majors’ include the majors of business or
engineering. Usually, these majors lead to high paying jobs later. However, ofien, other
majors such as health, science, and pre-professional are also very technical or lead to high
paying occupations, thus we also test whether students with work experience avoid ‘non-
technical major.’ ‘Non-technical majors’ includes only liberal arts and education. The
results of standard probit model are tabulated in Table 2- 4-2-19 and Table 2- 4-2-17.

The results are consistent with the MNL results. For males, the work experience
during the senior year is positively associated with ‘practical major’ and is negatively
associated with ‘non-technical major.’ Almost all the coefficients of ‘work,’ ‘heavy
work,’ and ‘working hours’ during the senior year are positive and significant in the
‘practical major choice.’ In 1980, at the mean values of other control variables, students
who worked during the high school senior year have 5 percent higher probability of
choosing ‘practical major’ than those who do not work at all. Students who work more

than 14 hours per week during the high school senior year have a 9 percent higher

178

 

 

probability of choosing ‘practical major’ than those who work 14 hours or less. A
working hour increase per week increases the probability of choosing the ‘practical
major’ by .4 percent (Marginal effects are not reported.)

The coefficients of ‘heavy work’ and ‘working hours’ are significantly negative in
the ‘non-technical major’ choice. In 1980, students who work more than 14 hours per
week during the high school senior year have 7 percent lower probability of choosing
‘non-technical major’ than those who work 14 hours or less. A working hour increase per
week decreases the probability of choosing the ‘practical major’ by .3 percent (Marginal
effects are not reported.) However, in 1980, the signs of coefficients of the variables
which represent the work experience during the junior year are in the opposite direction
even though the coefficients are not significant (Table 2- 4-2-17).

For females, signs of the coefficients for those variables are almost the same as in
the male case. The coefficients of ‘heavy work,’ and ‘working hours’ during the senior
year are positive in the ‘practical major’ and are negative in the ‘non-technical major.’
However, they are rarely significant. Only the coefficient of ‘heavy work’ in 1980 is
positive and significant in the ‘practical major choice.’ Female students who work more
than 14 hours per week during the high school senior year have about a 6 percent higher
probability of choosing ‘practical majors’ than those who work 14 hours or less.
(Marginal effects are not reported.)

Also in 1980, like the male case the signs of coefficients of the variables which
represent the work experience during the junior year, are in the opposite direction of the

signs of the coefficients of the senior work experience variables. Particularly, the

179

 

 

coefficients of ‘work junior’ are significant in all cases.’04 Regardless of controlling for

the work experience during the senior year, female students who work during their junior

year 6 to 7 percent have lower probability of choosing ‘practical major’ and more than 7

percent higher probability of choosing ‘non-technical major.’

Table 2—4-2—19 : Probit models: Practical majors and non-technical majors: Male and female

 

Male

Female

 

Variables

Work
Heavy Work

Working Hours

Practical Major
(business or
engineering)

Male
.156“
(.071)
.091‘
(.058)
007‘"
(.003)

Non-technical
Major(liberal arts
and education)

Practical Major
(business or
engineering)

Non-technical
Majorfliberal arts
and education)

 

-.183‘”
(.066)
-.112‘
(.057)
-.008‘"
(.003)

Female
.032
(.091)
.092
(.094)
.016
(.041)

 

-.042 1.3"
(.068) ’
-.027 F
(.072)

-.002

(.003)

 

Notes: (1) Standard errors are in parentheses. (2) ‘*’ means ‘significant between 10% level and 5% level,’ “*’ means
‘significant between 5% level and 1% level,’ ‘*"’ means ‘significant at less than 1% level.’ (3) Each cell represent a
result from the probit model. (4) Test scores and the number of courses in each subject taken during high school are
also included in the control variables. All other variables are included and the coefficients are not reported. (5) R
Square and 1 square are not reported.

 

Independent

( 1) Work
Work junior
(2) Heavy work

Work junior

(3) Working hours

Work junior

VEEQLCS

 

 

Table 2-4-2-20 : Probit models: Practical majors and non-technical majors: Male and Female (HS&B)
Male (1520) Female (1986)

Practical Majors Non-technical Practical Majors Non-technical
major major

.124“ -.088 -.016 .003

(.077) (.080) (.070) (.068)

-.083 .111 -.160" .179‘"

(.075) (.078) (.065) (.063)

.234‘" -.201"" .151" -.085

(.069) (.072) (.063) (.061)

-.101 .132" -. 194"" .197“”

(.073) (.076) (.064) (.062)

.010‘" -.009”"' .005‘ -.003

(.003) (.003) (.003) (.003)

-.110 .145“ -.188‘" .l95*"

(.074) (.077) (.064) (.063)

 

'0‘ When we put ‘heavy work’ or ‘working hours’ during the junior year instead of ‘work junior’, the
coefficients are not significant at all.

180

(NLS-72) rug

 

(4) Working hours .01 1”" -.009“" .005 -.002
(.003) (.003) (.003) (.203)
Working hours -.004 -.005 -.006‘ .005
junior (.003) (.003) (.003) (.007)
(5) Working hours 009‘" -.008” .003 -.001
(.003) (.003) (.003) (.003)
(6) Heavy work .214‘” -.174"‘I .113” -.O47
(.067) (.070) (.062) (.060)
(7) Work hoursjunior -.001 .002 -.004 .004
(.003) (.003) (.003) (.003)
(8) Heavy workjunior .044 -.015 -.080 .074
(.068) (.071) (.065) (.063)
(9). Workjunior -.046 .085 .180" -.l64”‘
(.071) (.074) (.061) (.062)

 

Notes: See notes in Table 2-4-2-19.

The main purpose of using the probit model is to control for the possible

endogeneity in the relationship between high school work experience and college major

choice. We focus on the 1980 male sample because the association between the work

experience and college major choice is very strong, and only in 1980, do we have some

instruments to identify the working hour equation. Our results indicate that there is no

 

such endogeneity in the relationship.

The results are tabulated in Table 2- 4-2-21. In the first two columns (1) and (1)’,
we use unemployment rate and growth rates as instruments, in the columns (2) and (2)’,
we use ‘high reservation wage’ dummy, and the ‘will not work’ dummy as instruments.
in the columns (3) and (3)’, we use all four together as instruments, in the columns (4)
and (4)’, we use ‘high reservation wage’ dummy, the ‘will not work’ dummy, and
employment growth rate as instruments.

Before we use these variables as instruments, we test (1) whether these variables

are redundant in the probit model for ‘practical major’ and (2) whether they are

181

 

sufficiently correlated with ‘working hours’ during the senior year. '05 These test results
are tabulated in Table 2- 4-2-22 and 23.

First of all, as we mentioned earlier, if the tendency to work cause such
association between high school work experience and ‘practical major’ choice and the
‘will not work’ dummy represents students’ tendency to work, ‘will not work’ dummy
may be negatively related with the ‘practical major’ choice. The coefficient of the ‘will
not work’ dummy in the column 1 in Table 2- 4-2-22 is not significant at all, showing that
there is no such relationship between ‘will not work’ dummy and ‘practical major choice.
Even though the coefficient is negative, the very large standard error makes the existence
of such a correlation very doubtful. Also, including this variable does not change the
coefficient of working hours.

Also, ‘high reservation wage’ dummy could represent individual student’s
willingness to work. However, when we include both ‘will not work’ dummy and ‘high
reservation wage’ dummy together, the coefficients of both variables are not significant
and the coefficient of working hours does not change almost at all. These results
themselves could be indirect counter evidence for the endogeneity in the relationship
between working hours and practical major choice if these variables represent student
workers’ tendency to work.

However, if the ‘will not work’ dummy represents an individual student’s random
choice of working during high school and if ‘high reservation wage’ dummy represents
just market demand, these variables are good instruments to identify the working hours

equation. The following appendix table 1 shows such possibilities that among college

 

'05 In the usual two stage least regression, over-identifying restriction test is generally used. However, in the

182

 

 

students in the academic field, the ‘high reservation wage’ dummy represents market
demand and has nothing to do with individual ability, and ‘will not work’ dummy is just
random choices by high school students, being unrelated to either market demand and
individual ability.106 If reservation wage represents an individual’s ability, students with
higher reservation wages should have on average higher test scores. However, there is no
difference in average test scores across. However, the average economic conditions are
much better for student with the highest reservation wages than for student group with the
lowest reservation wages. However, both the average test scores and economic
conditions for students who ‘will not work’ at any wage during high school are very
similar to the average test scores and economic conditions for the whole college students

. . 7
1n the academic field.10

Appendix Table 2: Reservation Wages, Test Scores and County Economic Conditions among White
College Students in Academic Field.

 

 

 

Reservation Reservation Reservation Reservation Will not work
wage equal to wage above Wage between Wage above
$2.00 or less $2.00 and less $3.00 and $3.50

_--_~__-__-- than $3.00 $3.50

Test Scores 56.7 56.1 56.4 56.5 57.3
(7.64) (7.20) (6.76) (7.09) (7.41)

County

Unemployment Rate 8.01 7.92 7.06 6.92 7.03
(2.86) (2.71) (2.50) (2.33) (2.43)

Employment Growth

rates .59 .99 .94 1.23 1.02
(3.57) (3.13) (2.72) (3.26) (3.73)

Percentage of

People employed 42.2 42.3 44.3 45.2 43.7
(6.61) (5.75) (5.66) (5.00) (4.64)

Log County Per

Capita Income 8.42 8.79 10.56 1 1.25 10.85
(4.92) (4.27) (3.75) (3.58) (4.42)

 

Note: Standard errors are in parentheses.

 

two stage probit, we are not sure whether such a method works.
'06 A similar results for the full samples are reported in appendix table in Ch. 1.

183

 

‘High reservation wage’ dummy and the ‘will not work’ dummy are not related to
the choice of college major (see columns 1 and 3 in Table 2-4-2-22.) and can be assumed
not to be related to tendency to work, which is the omitted variable that causes the
endogeneity.

County unemployment rate and employment growth rate are not related to the
choice of college majors (see columns 2 in Table 2-4-2-22). All the coefficients are very
insignificant. The problem is whether these variables are sufficiently related to ‘working
hours’ during the senior year. From Chapter 1, we see that, for the full samples,
which include all high school graduates, these variables are sufficiently related to
‘working hours.’ Now we use only college students in the academic field, we need to test
whether their relationships are different. Table 2-4-2-23 shows that ‘will not work’
dummy and ‘high reservation wage’ dummy are sufficiently related to working hours
while county unemployment rate and county employment growth rate are not significantly
related to working hours.

Finally, the two stage probit results for practical major choice in Table 2- 4-2-

21 show that there is no evidence for the endogeneity we expect. Except for column (1),
the coefficients of predicted working hours are still positive (see columns 2, 3, 4) even
though we can not tell the significance because of invalid standard errors and the scaling
factor of them. Except for the column (1)’, the coefficient of residuals from the first stage
regression of working hours are all negative and insignificant (see columns 2’, 3’, 4’.)

Thus, if there are any relationship between practical major choice and unobservable

 

'07 However, we can not claim that appendix table 1 proves that these instruments are not related to the error

term in the equation (3-6).

184

 

 

characteristics, the relationship is negative. In that case, our estimates could be downward
bias instead of upward bias.

In the cases of column (1) and (l)’, since the correlation between ‘working hours’
and the two instruments; unemployment rate and employment are not significant, such
results might occur.

However, once again our forgoing results are based on the assumption that
instruments- especially ‘high reservation wage’ dummy and ‘will not work’ dummy- are
not related to the error term in the equation (3-6). Even if these instruments are not
related to a particular omitted variable, tendency to work, we can not exclude the
possibilities that these instrumental variables are related to the rest of error term. In the
case where the instruments are related to the error term, our attempts to control for the

endogeneity or test the endogeneity should be wrong.

 

 

 

Table 2-4-2-21: Male practical major cases: Endogeneity 5 issue: (HS&B)
U) 10’ 12) (2f (3) f3)' (4) I4)’

Working -.022 .027 .022 .023

hours (.019) (.019) (.019) (.019)

Predict -.022 .027 .023 .023

working (.060) (.019) (.018) (.019)

hours

Residual .032 -.018 -.013 -.013

from 1“ reg (.060) (.019) (.019) (.019)
99.6 109.4 101.4 109.9 100.9 109.6 100.9 109.6

(.0472) (.0519) (.0481) (.0522) (.0479) (.0520_ (.0479) (.0520__

 

Notes: (1) Standard errors are in the parentheses. (2) "" means ‘significant between 10% level and 5% level,’ “w
means ‘significant between 5% level and 1% level,’ """’ means ‘significant at less than 1% level.’ (3) Test scores and
the number of courses in each subject taken during high school is also included in the control variables. All other
variables in the previous regressions are included and the coefficients are not reported. (4) In columns 1 and 1’, county
unemployment rate and employment growth rates are used as instruments. In the columns 2 and 2’, ‘high reservation
wage’ dummy and ‘not work’ dummy are used as instruments . In columns 3 and 3’, all these variables are used as
instruments. In columns 4 and 4’, ‘high reservation wage’ dummy . ‘will not work’ dummy and county employment
growth rate are used as instruments.

185

 

 

Table 2-4-2-22: Theprobit for thepractical choice: Instruments and practical major choice.

(HS&B)

 

 

 

_ ___.._.__ (I) __-_----__.-12_)- __-_m._-__-__m_l.3,) (4) (5) -_
Working hours 010“" .010‘” .009‘“I 009“” 009*”
(.003) (.003) (.003) (.003) (.003)
Working hours in junior -.004 -.004 -.005 -.005 -.005
(.003) (.003) (.003) (.003) (.003)
Will not work dummy -.050 .011 .012 .012
(.261) (.267) (.268) (.268)
Unemployment rates -.001 .000
(.014) (.014)
Employment growth -.006 -.006 -.006
rate (.009) (.009) (.009)
High reservation wage .081 .081 .081
dummy (.086) (.085) (.085)
109.2 109.5 110.1 110.5 110.5
(.0518) (.0520) (.0522) (.0524) (.0524)

 

Notes: (1) Standard errors are the parentheses. (2) “" means ‘significant between 10% level and 5% level,’ """’
means ‘significant between 5% level and 1% level,’ “‘”’ means ‘significant at less than 1% level’. (3) Test scores and
the number of courses in each subject taken during high school is also included in the control variables. All other

variables including ‘are included in the previous regressions and the coefficients are not reported.

Table 2-4-2-23: OLS for Working Hours During the Senior Year: Instruments and workinghours. (HS&B)

 

 

. __ ...---__....... -_ .. ..-._-_--.._-_(J) 12)-- (3) (4)
will not work dummy -4.271"' ~4.263‘ -4.290"'
(2.201) (2.200) (2.199)
unemployment rates -.121 -.059
(.1 17) (.116)
employment growth rate .122 .126‘ .130‘
(.076) (.075) (.075)
High reservation wage 3.696‘" 3.668"""I 3.694‘"
dummy (.686) (.687) (.687)
R square .1837 .2031 .2049 .2047

 

Notes: (1) Standard errors are in parentheses. (2) “’ means ‘significant between 10% level and 5% level,’ """ means
‘significant between 5% level and 1% level,’ """’ means ‘significant at less than 1% level.’.(3) All other variables
such as test scores. test scores squared, GPA, high school program, race, region, community type, ‘mother work’

dummy, family SES are included.

5. Conclusion.

Previous studies on the effects of high school employment on post-secondary

education find that the more students work during high school, the less likely they are to

attend college and the less likely they are to finish their degrees if they go to college.

Using NLS-72 and HS&B, first we reexamine the relationship between high school

186

 

 

employment and college attendance, and second, we explore the relationship between
high school work experience on the choice of college major among college attendees.

In the relationship between high school work experience and college attendance,
we not only confirm the findings of previous studies but also observe several other
characteristics: (1) In general, work experience, especially heavy work experience during

the senior year is negatively related to college attendance. However, this relationship has

 

changed over time. In 1972, the work experience during the senior year is negatively
related to college attendance. In 1980, the negative relationship between work experience

during the senior year and college attendance is weakened, while the negative relationship

 

between the work experience during the junior year and college attendance is very strong.
(2) These negative relationships are much stronger in the choice of four-year college
attendance rather than two-year college attendance. (3) The negative relationship is
stronger among males than females. (4) After controlling for endogeneity, in 1972, the
working hours during the senior year affects college attendance positively, while in 1980
the average working hours during high school affects it negatively. However, endogeneity
tests do not provide evidences that there are significant endogeneity in the relationship
between high school work experience and college attendance. Also, since we can not
prove that the instruments we use in two stage probit models are completely valid, we can
not completely be sure that these results from two stage probit model are always correct.
Also, we find that high school work experience has strong associations with the
choice of college major. Among college students in the academic field, students with
much work experience in their high school senior year chose majors associated with

higher paying jobs. The longer students work during their senior year, the more they are

187

 

likely to choose practical majors such as engineering or business, and the more they are
likely to avoid non-technical majors such as liberal arts and education. These tendencies
are more apparent among males than females. These tendencies are much stronger in
1982 than 1980. These results are consistent with Light’s finding (1999) that among non-
college bound male students, those who worked during high school took more vocational
courses. Students with work experience tend to choose skill oriented or technical oriented
education.

When we control for a possible endogeneity related to the working hours in the
probit framework, our results indicate that the work experience seem to influence
students’ choices of college majors rather than a certain unobservable characteristic (a
tendency to work) causes such relationship between high school work experience and the
choice of college major. Also, endogeneity test results show no evidence for such
endogeneity in the relationship between high school work experience and the choice of
college major.

However, as mentioned earlier, we need to admit that our results are based on the
assumption that instruments are valid. Since there is no way to test these instruments
meet all the requirements as a good instrument, we can not be sure our results from the
two stage probit models are completely valid ones.

If these instruments are valid, with our results we can explain, at least partially,
the fact that students who worked during high school, especially during their senior year,
earn more later or obtain higher occupational status later even though they have generally
lower post-secondary educational attainment. If high school work experience leads to

higher later earnings and higher later occupational status, it might not be because of high

188

 

school work experience itself, but because of the choices of education or training that
high school student workers make afterwards.

Students with work experience during high school are more practical in later
choices of their education later. High school student workers seem to choose education
with higher economic returns and avoid the education with lower economic returns.
According to a standard human capital theory, generally individuals will choose the major
offering the highest present value of life-time earnings. However, individuals might have
different tastes for their future earning streams or different information on the firture
earnings. Early work experience change either students’ perception of the future earnings
or information on future earnings. '08

However, after controlling for work experience in the senior year, students who
ever worked during their junior year are more likely to choose liberal arts and education
and to avoid business or engineering majors than those students who did not work at all
during their junior year. Among females, even without controlling for work experience
during in the senior year, those who work during their junior year tend to choose liberal
arts or education and avoid the business or engineering major.109 If early work experience
generally makes students more practical in choosing their education, why do working
experiences in two different years have different effects? Even though work experience

during the senior year is the most common work experience among high school students,

 

'08 With respect to this issue, Eide and Waehrer’s results (1996) provide a very interesting point. They find
that the expected earning is negatively related to females’ choices of college major. We do not expect
females to dislike higher earnings.

’09 However, the coefficients of ‘heavy work junior,’ or ‘working hours junior’ in the MNL models for the
choices of college major or the probit models for practical major or non-technical major are not significant
at all for both males and females, meaning that work intensity during the junior year does not matter.

189

.0....- .v . ‘_

 

 

why do work experiences in two different years have the opposite effects? This question
is should be addressed in further research.1 '0

However, also further studies should resolve others issues remaining unsolved in
this study. First, we do not understand why high school work experience has such effects
on the choice of college major. We do not have any information to find out why student
workers choose more practical majors and avoid non-technical majors. Is it because work
experience makes them know and understand better about the reality of the world of
work, and/or which majors or skills are valuable in the market ? Or is it because, as
Greenberg and Steinberg claim, work experience (or money making or spending
experience) makes them more materialistic or selfish ? Further research is required to
understand the effects of high school employment more accurately.

Third, in this study we have not studied the links between major choices and later
earnings. Even though studies which use NLS-72 and HS&B data sets show that higher
wage premiums for the ‘practical major’ and lower wage premiums for the ‘non-technical
major,’ all studies classify major field based upon four-year college degree.1 H For the full
explanation of higher later earnings and lower post-secondary educational attainment of
high school student workers, we need to investigate the links from the major choices to

later earnings.

 

”0 To address this issue, we need to use a MLE method because we have a dummy endogeneous variable in
the binary dependent framework.

”' To see how the working experience affects major field choice in the senior year in college, we use the
MNL method to major field choices. The results do not have much differences.

190

REFERENCES

191

 

References

Alexander, Karl L., Holupka, Scott and Pallas, Aaron M. “Social Background and Academic Determinants
of Two-year versus Four-year College Attendance: Evidence from Two Cohorts a Decade Apart” 1987,
American Journal of Education, November 1987.

Altonji, J. G. 1993 “The demand for and return to education when education outcomes are uncertain.”
Journal of Labor Economics Vol. 1, pp 48-43.

Berger, Mark C. 1988 “Predicted Future Earnings and Choice of College Major” Industry and Labor
Relations Review. Vol. 41, No. 3 (April 1988), pp418-429

Betts, Julian R., and McFarland, Laurel L., 1994, “Safe Port in a Storm: The Impact of Labor Market
Conditions on Community College Enrollments” The Journal of Human Resources, Vol. 30, No 4,

Blakemore, A. E. and Low, S. A. 1984 “Sex Differences in Occupational Selection: the Case of College
Majors” Review of Economics and Statistics February, pp 157-163

Carr, Rhoda V., Wright, James D., and Brody, Charles J. 1996, “Effects of High School Work Experience a
Decade Later: Evidence from the National Longitudinal Survey” Sociology of Education 1996., Vol.69
(January):66-81.

D’amico, Ronald. 1984. “Does Employment During high School Impair Academic Progress?” Sociology of
Education Vol. 57; 152-64

Davies, Scott., and Guppy, Neil., 1997 “Fields of Study, College Selectivity, and Student Inequalities in
Higher Education” Social Forces, June 1977, Vol. 75., No. 4: 1417-38.

Eide, Eric., and Waehrer, Geetha., 1996 “The Role of Option Value of College Attendance in College
Major Choice” Economics of Education Review Vol. 17, No. 1, pp. 73-82

Freeman, Richard B. “A Cobweb Model of the Supply and Starting Salary of New Engineers.” Industrial
and Labor Relations Review, Vol. 30, No. 2, pp 236-48.

Greenberger, Ellen and Steinberg, Laurence. 1986 When Teen_agers Work: The Psychological aand Social
costs of Adolescent Errprlomient. New York: Basic Books.

 

Greenberger, Ellen; Steinberg, Laurence; and Ruggiero, Mary. “ A Job Is a Job... or Is It?” Work and
Occupations Vol. 9, 1982. Pp. 79—96.

Gritz, R. Mark, and MaCurdy, Thomas 1992 “Participation in Low-Wage Labor Markets by Young Men.”
NLSQiscussion Paper 93-16, U. S. Department of Labor.

Grogger, Jeff, and Bide, Eric. 1994 “Changes in College Skills and the Rise in the College Premiun”
Journal of Human Resources, Vol. 30., No. 2, pp 280-310

Grubb, N. W. 1993 “The Varied Economic Returns to Postsecondary Education: New Evidence from The
Class of 1972.” Joumpl of Humfl Resources, Spring, pp. 365-382

Hotz, V. J ., Xu, L., Tienda, M., & Ahituv, A. 1998, “The Returns to Early Working Experience in the

Transition from School to Work for Young Men in the US: an Analysis of the 1980s. Working paper,
University of Califomip Los Angeles. Los Angeles.

192

Kane and Rouse(1995), “Labor-Market Returns to Two- and Four -Year College”, The American Economic
Review Vol. 85 No. 3, June 1995.

 

Lee, L.-F. 1981, “Simultaneous Equations Models with Discrete and Censored Dependent Variables,” in C.
Manski and D. McFadden (eds), Structural Analysis of Discrete Data with Economic Applications.
Cambridge: MIT Press.

Light, Audrey. 1999, “High School Employment, High School Curriculum, and Post-school Wages”,
Economics of Education Review Vol. 18, Pp.291-309

Marsh, Herbert W. 1991 “Employment During High School: Character Building or a Subversion of
Academic Goals?” Sociology of Education 64, 172-189

Meyer, Robert H. and David A. Wise. 1982. “High School Preparation and Early Labor Force Experience.”
Pp. 277-347 in The Youth Labor M_arket Problem: Its Nature, cysespnd Consequences, edited by Richard
B. Freemand and David A. Wise. Chicago: University of Chicago Press.

Paglin, M. and Rufolo, A. M. 1990, “Heterogenous Human Capital, Occupational Choice, and Male Female
Earnings Differences” Joum_al of Labor Economics, Vol. 8, pp 123-144.

Polachek, S. W. (1978) “Sex Differences in College Major” Industrial and Labor Rel_a,tions Review, July.
pp. 498-508

Rouse, Cecilia E (1994) “What to Do after High School: The Two-year versus F our-year College
Enrollment Decision” Choices and Consequences edited by Ehrenberg, Ronald G.

Rivers, D. and OH. Vuong (1988), “Limited Information Estimators and Exogeneity Tests for
Simultaneous Probit Models,” Jourml of Econometrics Vol. 39, pp. 347-366.

Ruhm, Christopher J ., 1997, “ Is High School Employment Consumption or Investment?” Journal of Labor
Economics 15, No. 4.

Rumberger, Russell W. and Thomas, Scott L. 1993, “The Economic Returns to College Major, Quality and
Performance: A Multilevel Analysis of Recent Graduates” Economics of Ed_thation Review Vol. 12, No. 1.

Steel, L. (1991): “Early work experience among white and non-white youths: implications for subsequent
enrolhnent and employment” Youth and Societv Vol. 22, 419-447

193