                  

a. 2(er WW. Anew... a.»

a . . 7!.
i . 02.2.1»:
. 173.326.

 

 

          

      

 

2:: . u

     
   

             

  

 

 

 

.- 2 .
. Ll!
Let: «8: .
.116. 52.5..” t van
a 133
n. 1...“?
1.... e... n
. i =
2 . . 5,. z... r
‘17.”: «I...

  

.5
rt; . .rlnuuvﬁ.

un-

    

. .n‘)"~n
£3...» a»...
a .
”Hist; y‘h
l!-

                    

 

.1

      

    

9A
a :Ou PE:
{1 nixéx

 

                                   

.1:
s- a.)
1.93:...23...‘ .
1,... 4.7.1.35...“
. 1.15%....
in
.... v .
.1k3hyn

 

 

 

   

. .3: .i.
_..a.mnys.t§2nmi 2..

 

 
 

 

..
1.5.2.... ﬁg
.3. .
algﬁhvihthu i
4..."... ((3.: .i
-1. .1... 2a.. x
. a i g 1:.
. ..

Lainmmpﬁmmﬂzﬁ 5......3n
P225)!!! vaaiﬁﬂm 51%?

 

r: s
. $1
I ..:

    

     

.x..£..t..: .. ‘9

... “gin...“ a.

.ﬂruvl.bnhi 533.1%
A

   

  

33"!!! v1. nth-
...i...‘ . .. .5...va .D.‘ s:
‘53:? ’.qum?.lldnﬂuI.lhbtﬂw-hf .450...
.. . . .595. :11.
0:93. It: 1:31.: a It,
I 3.." $1188.
I) .t. .vtzgahlnntlifi u 7.1.0.525
C9,..9..l12§l9:5 ..\Ojii.xo 380i. ﬁlth
{3:72}... .Vw va'.t’\‘
1:. .339. 33.13.023.133...)
.- 9... .3231. Inn“. «1 . 6..
.. K . . ,
.Nu‘l. .h...x-h1um: 1..
.3595: 3.3 tact-{3);}
‘unvtsc 5.... Agai‘t.‘
. . .ﬁ , cri; 1a..
16.531551. !.l
5.9!,

  
         

        

  

 

 

If: 7.
:1 cﬁ.

 

 

    

. . a if},
urging“. «ﬁﬁﬂﬁﬁm

3
1| .3555 I.
I) f..~r§.¢§. BF: f
1‘}? , .. if...»

 

. .3 if ..
3...ﬂ:z_..._.\ 3;.» HE... 2.»
3.ﬁx..?.fﬂz.m£h}.¥§ _

 

  

                                             

 

gt: .1.

hh’iﬁgﬁt’l- A. ydﬁ’.ﬁqnuu I”. gm”

Vuluﬁlhi gin“...-

ﬁiiftifhﬁuﬁ

. r 55.1.5.3 if?!

fiatltéiiix Z.»
Nifluﬂ if...

      

It. tn
. :5...“ 3.. bu...
.533...

 

a. 2%...

     

15...)..m
. V

.333... r.
z fifﬂﬂﬂg
. .5
.x a 2.. ..f. ..
911...}... H.445! 4.59:...41
z... .Vnﬁiéhsriaaswn. 7:4... “30......
: éf..§...:f...i..as eras...
~ I )’I\ I3‘!‘iIr-:v .n-l.q;’
115:3!!! t2.
\‘va‘a sly-”2"..{IW‘

 

 

  

 

  

. its: :3
...h«..vk......,s
39?, .r

 

 

 

It... 2’1...th v
(it-5‘ ..V 333......-
12. :lrxi..§!l....:tr.oli!
1...? 71.. 3.29:? I ‘3‘!!!)-
f...‘ 91¢? . .2 Qinizitfirt...
911.4 .m»..~l§..st1
25;: LE! [(1513:

It! .7un gt:
1).)...9 if! S. .«C.
A 1.0. f. akxxi...‘
. 1146.3“. t... .

 

.4. PI?!

.r EVEN...
i :73:

    
      

  

 

                   

 

L a, .
I... 31.16: .e
4; iii. 7. .3
L 4...... .\|vrl.

    

          

1.. v I51... ’53.
azvri . ‘11:...1‘ \.
rv L... )t: 395.1 5.31 IvOi

.- . . ;. . .

 

          

 
 

 

l

           

 

 

5.3 A

(:5 n. .

 

 

xi?

1.»!
i.

 

THESiS
Z
2 U 'L)

_

 

LIBRARY
Michigan State a
University

 

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

DATE DUE

DATE DUE

 

@Amomm

 

 

 

 

 

 

 

 

 

 

 

 

 

11/00 6mm.“

 

 

GOVERNMENT POLICY, EDUCATION, AND EARNINGS IN MALAYSIA
By

Hong Peng Ong

A DISSERTATION

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

DOCTOR OF PHILOSOPHY
Department of Economics

2000

ABSTRACT
GOVERNMENT POLICY, EDUCATION, AND EARNINGS IN MALAYSIA
By

Hong Peng Ong

This dissertation attempts to study the effect of government policy and education
on earnings in Malaysia. It is based on the ﬁrst and second waves of the Malaysian
Family Life Survey (MFLS). The ﬁrst chapter concerns the effect of New Economic
Policy (NEP) on earnings differentials and returns to education of Malays relative to
Chinese and Indians. The estimated results show that the earnings differentials of Malays
relative to Chinese and Indians declined from pre-NEP to post-NEP periods. The decline
in relative earnings of Chinese and Indian compared to Malays during the post-NEP
period can be attributed to the faster rate of increase in the level of education and
improvements in the returns to education of Malays, especially at the post-secondary
level. The other reason is the larger shift from the agricultural occupations to the more
urban, higher paying occupations such as professional, clerical, service, and production
related occupation of Malays relative to Chinese. The Malays experience a bigger shift
from agricultural occupations into managerial, clerical, service occupations which offer
relatively better earnings when compared to Indians. Using a linear spline speciﬁcation
on earnings, the key ﬁnding is that the returns to post secondary education of Malays
relative to Chinese and Indians showed the most signiﬁcant improvements from the pre-
NEP to the late post NEP period. These results are compatible with the strategies

employed by the NEP to reduce the racial earnings gap and to improve the returns to

education of Malays through occupational restructuring and improved access to higher
education for Malays.

The second chapter examines the distribution Of schooling and earnings inequality
in Malaysia. The decomposition of earnings inequality indicate that schooling and
occupation are important factors in explaining earnings inequality. However the
unobserved factor remains the biggest factor in explaining the earnings inequality. The
unusually high residual variance in log earnings Of Malays during the pre-NEP period is
Observed even when occupation variables are accounted for. This phenomenon is not due
to the differences in the rate of forgetting between Malays and non-Malays based on tests
of signiﬁcance. But from the view of point estimates, there is substantial difference in the
rate of forgetting between Malays and non-Malay. Even so, the explanation of the effect
of NEP that substantially reduces the residual variance Of Malays during the post-NEP
period is plausible. The comparison Of generalized Lorenz curves Of real earnings Of
Malays and non-Malays suggest that the NEP has played a role in improving the position
of Malays relative to non-Malays but the Malays still lag behind in terms Of social
welfare ranking Of real earnings distribution. These results point to the need to maintain
the policies intended to uplift the economic position of the disadvantaged group.

It is noted that the link between parents’ schooling and children’s schooling
weakened signiﬁcantly for cohorts born after 1960. This suggests that the education
policies and NEP have played a role in providing better educational Opportunities for
offspring of parents with lower education. This augurs well for the future as the weaker
intergenerational transmission of schooling tends to bring about greater equality in

human capital that is associated with greater earnings equality.

This dissertation is dedicated to my mother and the memory of my father.

iv

ACKNOWLEDGMENTS

First and foremost, I wish to express my heartfelt thanks to Professor John Strauss
who is my mentor and advisor. He has been a key source of inspiration at every stage Of
my dissertation. I really appreciate his untiring efforts, his expert guidance and valuable
comments that enabled me to complete my dissertation.

I also wish to express my sincere appreciation to Professor Jeff Biddle who is my
second mentor. He has been actively involved in my research and has offered many
constructive comments and ideas that immensely improved this dissertation.

I am indeed privileged to be under the tutelage of Professor Strauss and Professor
Biddle as I have learnt a great deal from them. I also wish to acknowledge the extra
efforts taken by Professor Strauss and Professor Biddle to read my numerous drafts and
then jointly providing me feedback in such a short time.

I am very grateful to Professor John Giles for his valuable comments and for
providing me with additional relevant literature to enhance my understanding on the
subject of research. He has also given me the support and encouragement to complete my
dissertation. I also wish to thank Professor Jeffrey Wooldridge for his technical guidance
on the test of parameter stability used in this study.

I am extremely fortunate to receive a lot of help from many friends in the doctoral
program. Most notably, I would like to thank Chi-Rok Han, Sang-Hyop Lee and Rehim
Kilic for their signiﬁcant contributions in helping me clear many difﬁcult hurdles that

exist in a doctoral program.

I wish to record my profound gratitude to the Government of Malaysia for
sponsoring my graduate studies here. In return, I hope to serve my country well and at the
same time do justice to the excellent education that I have received at Michigan State
University.

Above all, I wish to thank my wife, Lai Heng and my children, Tze Shen, Gui
Xian, Tze Loon and Tze Han for their sacriﬁces, patience, love and support which made

my graduation possible.

vi

TABLE OF CONTENTS

LIST OF TABLES .......................................................................................... ix
LIST OF FIGURES ....................................................................................... xiii
INTRODUCTION ........................................................................................... 1
CHAPTER 1

THE EFFECT OF NEW ECONOMIC POLICY ON
RETURNS TO EDUCATION AND EARNINGS

IN MALAYSIA ................................................................................................ 4
1. Introduction ................................................................................................ 4
2. Data ............................................................................................................ 7
3. Descriptive Statistics ............... 10
4. Earnings Differential and Earnings Growth ............................................. 12
4.1 Regression Analysis ......................................................................... 12
4.2 Relative Earnings between Races ..................................................... 13
4.3 Attrition of Panel Sample ................................................................. 15
4.4 Earnings Growth ............................................................................... 16
5. Returns to Education ................................................................................ 17
5.1 Overall Sample ................................................................................. 17
5.2 Labor Force Entry before or after NEP ............................................ 19
5.3 Cohort Analysis ................................................................................ 21
6. Cross-section Analyses ............................................................................. 24
7. Conclusion ................................................................................................ 26
References ...................................................................................................... 28
CHAPTER 2
DISTRIBUTION OF SCHOOLING AND
EARNINGS INEQUALITY IN MALAYSIA ............................................... 53
1. Introduction .............................................................................................. 53
2. Literature Review ..................................................................................... 55
3. Theoretical Linkages between Education and
Earnings Distribution ............................................................................... 6O
4. Data .......................................................................................................... 62

vii

 

 

. The Evolution of Schooling Distribution ............................................... 63
5.1 Tests for Parameter Stability .......................................................... 69
. Male Earnings Inequality ....................................................................... 71
6.1 Decomposition Of Earnings Inequality ........................................... 73
6.2 Decomposition Of Earnings Inequality by
Ethnic Group .................................................................................. 74
6.2.1 Measurement Error in Earnings .......................................... 77
6.2.2 Effect of Occupation on Earnings Inequality
and Residual Variance ......................................................... 80
6.3 Social Welfare Ranking of Earnings Distribution .......................... 82
. Intergenerational Transmission of Schooling ......................................... 83
7.1 Regression Speciﬁcation and Variables ......................................... 84
7.2 Empirical Results ........................................................................... 86
7.2.1 Parents’ Schooling ............................................................... 86
7.2.2 Birth Cohorts, Ethnicity and Government Policies ............. 87
7.2.3 School Availability .............................................................. 88
7.2.4 Place of Residence ............................................................... 89
. Conclusion .............................................................................................. 90
References .............................................................................................. 93
Appendix A .......................................................................................... 125
Appendix B .......................................................................................... 135
Appendix C
Analysis Of Measurement Error in Earnings .................................... 136

viii

J—M—“_' “‘5‘..- ‘— .4,--x it.

LIST OF TABLES

CHAPTER 1
Table 1: Descriptive Statistics by Sample and Time of Malays ................. 30
Table 2: Descriptive Statistics by Sample and Time of Chinese ................ 31
Table 3: Descriptive Statistics by Sample and Time of Indians ................. 32
Table 4: Relative Probability of being Employed in Certain

Occupation Groups in 1976 and 1988 ........................................... 33
Table 5: Real Earnings Regression - Simple Speciﬁcation ......................... 35
Table 6: Real Earnings Regression — with Additional Controls ................. 36
Table 7: Real Earnings Regression - Simple Speciﬁcation

(Common panel sample for 3 periods) ....................................... 38
Table 8: Real Earnings Regression by Sample with interaction

between Race and Years of Education ........................................ 39
Table 9: Real Earnings Regression by Sample with interaction

between Race and Years of Education

(Respondents that join labor force before NEP) ......................... 41
Table 10: Real Earnings Regression by Sample with interaction

between Race and Years of Education

(Respondents that join labor force after NEP) ............................ 43
Table 11: Real Earnings Regression by Sample with interaction

between Race and Years of Education

(age cohort 15-34 years old) ...................................................... 45
Table 12: Real Earnings Regression by Sample with interaction

between Race and Years of Education

(age cohort 35-54 years old) ...................................................... 47
Table 13: Real Earnings Regression by Sample in

1976 Cross-section .......................................................

............. 49

Table 14: Real Earnings Regression by Sample
in 1988 Cross-section ............................................................... 51

CHAPTER 2

Table 1: Summary Statistics for Distribution of Schooling
by Five Year Birth Cohort ......................................................... 96

Table 2: Mean Years of Schooling by Schooling Decile
(All persons age 25-29 and 55-59 in 1988) ............................... 97

Table 3: Parameter Instability in Years of Completed
Schooling: Breaks in Mean, Standard Deviation
And Coefﬁcient Of Variation ...................................................... 98

Table 4: Monthly Earnings by Highest Grade Completed
for Males Age 20-54 in 1967-69 ............................................... 99

Table 5: Monthly Earnings by Highest Grade Completed
for Males Age 20-54 in 1976 ................................................... 100

Table 6: Monthly Earnings by Highest Grade Completed
for Males Age 20-54 in 1988 .................................................. 101

Table 7: Log Monthly Earnings Regressions
for Males Age 20-54 in 1967-69, 1976 and 1988 ................... 102

Table 8: Simulated Variance of Log Monthly Earnings
for Males Age 20-54 in 1967-69, 1976 and 1988 ................... 103

Table 9: Log Monthly Earnings Regressions for Malay
Males Age 20-54 in 1967-69, 1976 and 1988 ......................... 104

Table 10: Log Monthly Earnings Regressions for non-Malay
Males Age 20-54 in 1967-69, 1976 and 1988 ........................ 105

Table 11: Simulated Variance Of Log Earnings for Males
Age 20-54 in 1967-69, 1976 and 1988
by Malays and non-Malays ..................................................... 106

Table 12: Log Monthly Earnings Regressions for Malay
Males Age 20-54 in 1967-69, 1976 and 1988
(include occupation variables) ............................................... 107

Table 13:

Table 14:

Table 15:

Table 16:

Table 17:

Table 18:

Table 19:

Log Monthly Earnings Regressions for non-Malay
Males Age 20-54 in 1967-69, 1976 and 1988
(include occupation variables) ..................................................... 108

Simulated Variance of Log Earnings for Males Age 20-54
in 1967-69, 1976 and 1988 by Malays and non-Malays
(include occupation variables) .................................................... 109

Means and Standard Deviation of Schooling Attainment
Equation Variables ...................................................................... 110

Regressions on Years of Schooling Of Male Children
Age 17 and above ...................................................................... 111

Regressions on Years Of Schooling Of Female Children
Age 17 and above ...................................................................... 114

Regressions on Years of Schooling of Male Children
Age 20 and above ...................................................................... 116

Regressions on Years Of Schooling of Female Children
Age 20 and above ...................................................................... 118

APPENDIX A

Table A1:

Table A2:

Table A3:

Table A4:

Table A5:

Means and Standard Deviations Of Variables Used ................ 132
Regression of Simple Difference in Reported Earnings .......... 132
Regression Of Simple Difference in Reported

Earnings of Malays ................................................................. 133

Regression of Simple Difference in Reported

Earnings Of non-Malays .......................................................... 133

Regression of Difference in Reported
Earnings Square ...................................................................... 134

xi

APPENDIX B

Table B1: Log Monthly Earnings Regressions for Malay
Males Age 20-34 in 1967-69, 1976 and 1988
(include occupation variables) ................................................. 135

APPENDIX C
Table C2: Log Monthly Earnings Regressions for Malay

Males Age 35-54 in 1967-69, 1976 and 1988
(include occupation variables) ................................................ 136

xii

 

LIST OF FIGURES

CHAPTER 2
Figure 1: Mean Years of Schooling Of Males

(3-year moving averages) ......................................................... 120
Figure 2: Mean Years of Schooling of Females

(3-year moving averages) ......................................................... 120
Figure 3: Standard Deviation in Years Of Schooling Of Males

(3-year moving averages) ........................................................ 121
Figure 4: Standard Deviation in Years of Schooling Of Females

(3-year moving averages) ......................................................... 121
Figure 5: Coefﬁcient in Variation in Years of Schooling

of Males (3-year moving averages) ......................................... 122
Figure 6: Coefﬁcient in Variation in Years Of Schooling

Of Females (3-year moving averages) ...................................... 122
Figure 7: Lorenz Curves for Schooling by Age Group ............................ 123
Figure 8: Lorenz Curves for Schooling by Age

and Ethnic Group ...................................................................... 123
Figure 9: Generalized Lorenz Curves for Schooling

by Age and Ethnic Group .......................................................... 124

Figure 10: Generalized Lorenz Curves for Monthly Real

Earnings by Ethnic Group in 1967-69 and 1988 ..................... 124

xiii

 

 

 

 

INTRODUCTION

Malaysia is a multi-ethic society and national unity is the over-riding concern of
the Malaysian government. The government has implemented a number of policies to
that strives to achieve a balance between the goals of economic development and the
need to achieve as peace and stability. It is viewed that economic growth with equity is
crucial for national unity which is the pre-requisite for growth and development. The
major policies that are relevant to this study are the educational policy and the New
Economic Policy (NEP). One of the key features of the education policies is the use of
Malay language as the medium of instruction at the secondary and tertiary level of
education to achieve national unity. The other key feature is the emphasis of education as
means for development through the policy of education expansion. The NEP is
considered as a socio-economic engineering program that is intended to redistribute
wealth and to achieve greater equality in schooling and employment opportunities to
reﬂect the racial composition of Malaysia. This dissertation attempts to analyze the effect
of government intervention on schooling attainment and earnings in Malaysia.

Chapter 1 examines the effect of NEP on returns to education and earnings in
Malaysia. To achieve this purpose, the data from the ﬁrst and second waves of the
Malaysian Family Life Survey (MFLS) were used. By dividing the data into pre-NEP,
early post-NEP and late post-NEP periods, the analysis of outcomes before and after the
NEP can be obtained. The ﬁrst research objective is to look at the earnings of Chinese

and Indians relative to Malays before and after the NEP. The estimated results indicate

that the earnings differential of Malays relative to Chinese and Indians have narrowed
over the three periods of analyses. The second research objective was to examine the
returns to education of Malays, Chinese and Indians before and after NEP. The main
result is that the returns to post secondary education of Malays relative to Chinese and
Indians have showed the most signiﬁcant improvements from the pro-NEP to the late
post-NEP period. These results are in line with the objectives of the NEP that are aimed
at improving the relative position Malays who are disadvantaged group.

Chapter 2 examines the distribution of schooling and earnings inequality in
Malaysia. The evolution of schooling distribution have been examined by using
schooling data by birth cohorts. The trend analysis indicate that the levels of schooling
are increasing from the older to younger cohorts for both males and females irrespective
of ethnic group, while schooling inequality has been declining from the older to younger
cohorts. As for earnings inequality, a trend of falling earnings inequality can be observed
for both Malays and non-Malays. A technique of decomposition of earnings inequality
has been applied to examine the extent of the contribution of certain key determinants of
earnings inequality. The simulation results indicate that schooling and occupation are
important factors in explaining earnings inequality. Generalized Lorenz curves are drawn
to provide an idea of the social welfare ranking of real monthly earnings distribution of
Malays and non-Malays. The results show that the social welfare gap of Malays relative
to non-Malays have narrowed after the NEP but the Malays are still lagging behind.
These results suggest that there is still a need to maintain the policies intended to uplift

the economic position of the disadvantaged group.

The Hansen test of parameter stability on the timing of the effect of government
policy on mean schooling and schooling inequality did not reveal a clear pattern of the
timing of government intervention. As such, this study uses birth cohort measures of
schooling to determine whether the timing of changes in schooling decisions by families
in the different ethnic groups coincides with the major changes in government policy. It
is noted that the link between parents’ schooling and children’s schooling weakened
signiﬁcantly for cohorts born after 1960. This suggests that the education policies and
NEP have played a role in providing better educational opportunities for the offspring of
parents with lower education. This augurs well for the future as the weaker
intergenerational transmission of schooling tends to bring about greater equality in
human capital which is associated with greater earnings equality.

To sum up, the Malaysian government has been quite successful in using the
policy tools to reduce the earnings differentials of Malays relative to Chinese and
Indians, to increase the relative returns to education of Malays, to raise overall levels of
education, to lower schooling inequality, and to achieve a more equitable distribution of

earnings.

CHAPTER 1

THE EFFECT OF NEW ECONOMIC POLICY ON RETURNS
TO EDUCATION AND EARNINGS IN MALAYSIA

1. Introduction

One of the most important policy questions in Malaysia is the degree of success
of the New Economic Policy in correcting the economic imbalance among the major
ethnic groups in Malaysia. The great deal of attention and interest in the relative earnings
among the races can be traced to the economic, social, cultural and political conditions in
Malaysia. Prior to the NEP, the occupational scenario in Malaysia was such that Malays
were more likely to work as ﬁshermen or farmers and Chinese were more likely to
engage in business and urban labor market activities while the Indians were
predominantly in the rubber plantations. As a result, the income disparities among the
races were getting wider. This resulted in a lot of racial tensions and culminated in racial
riots in 1969, which was known as the May 13 tragedy. Consequently, the government
realized that intervention was required and launched the New Economic Policy (NEP) in
1970. The NEP was incorporated into the four ﬁve-year development plans implemented
from 1971-19901 (Second to Fifth Malaysia Plan). It was an exercise in social
engineering designed to reduce the socioeconomic imbalances among ethnic groups, to

eradicate poverty, and to restructure the employment patterns of the country.2 It is against

 

' Subsequently, the National Development Policy (NDP) was introduced in 1991, which has similar goals
as the NEP.

2 The objectives of the New Economic Policy were to achieve national integration and unity. The two-
pronged strategy are: (i) to reduce and eventually eradicate poverty by raising income levels and increasing

this background that the topic of earnings and ethnicity is of real concern in Malaysia,
and that provides the motivation for this paper.

In addition, this paper is intended to estimate the returns to education in Malaysia.
This is motivated by the ﬁndings of Smith (1991) who used the ﬁrst Malaysian Family
Life Survey (MFLSl) data. He ﬁnds that education is the single most important variable
in explaining income growth in Malaysia. Hence in the study of relative earnings among
the various races, it is important to examine their respective returns to education. In
addition, it is noted that within the framework of NEP, education is used as a means to
correct the economic disparities among the various races. One strategy is to impose
quotas that reﬂect the racial composition of the country for admission to tertiary
education. It is also intended for employment restructuring by increasing the supply of
qualiﬁed Malays for managerial and professional jobs, which offer relatively higher
earnings.

Furthermore, the development literature has provided ample evidence that
education is an effective form of human capital investment. This is shown in
Psacharopolos (1985) who used data from 61 countries to make cross-country
comparisons on returns to education. The ﬁndings conclude that there are signiﬁcantly
positive returns to education and that returns are highest for primary education, general
curricula, education of women and countries with the lowest per capita income. Heckman
and Hotz (1986) showed that the return to schooling in Panama was 7.5% after
controlling for age and age squared, training, intensity of employment and regional

dummies. As for family background variables, mother’s education was statistically

 

employment opportunities for all Malaysians, irrespective of race; and (ii) accelerate the process of

signiﬁcant and had a larger positive effect than father’s education. These results are
consistent with Smith’s results for Malaysia.

Another study that is similar to this paper is by Gallup (1997) who used the
second wave of Malaysia Family Life Survey (MFLS2). He ﬁnds that male Malay
earnings steadily fell behind male Chinese earnings over the period 1960 to 1988. This
result is not consistent with the ﬁndings of the income cross-section data. He attributed
the inconsistency to the recall bias in the reported work histories. While there is some
overlap in terms of the area of study, this paper intends to examine the issue of the effect
of ethnicity and education on earnings from a different analytical approach. For instance,
Gallup generates annual earnings data for each individual based on the starting and
ending earnings of each job by using interpolation.3 This may not be desirable because
these earnings that are interpolated are unlikely to reﬂect the actual earnings of the
individual respondents.

One major difference is that this paper is based on the new sample, panel and
children sample of MFLSZ and Panel Sample of MFLSl while Gallup’s paper is conﬁned
to the New sample of MFLSZ. In view of the recall error confronted by Gallup with
MFLSZ data, this paper utilizes data from the last ﬁve years of survey from 1984 through
1988 only. The additional contribution of this paper is the focus on the effect of the NEP
on relative earnings and returns to education among the ethnic groups.

A recent paper by Schafgans (1998) used the parametric Heckman as well as

semiparametric approach to examine ethnic wage differences by gender. The key results

 

restructuring Malaysian society to correct economic imbalances so as to reduce and eventually eliminate
the identiﬁcation of race with economic function.

3 The individual earnings interpolation were constrained to have the same second derivative of -.5150. The
estimated wage regression is as follows:

are that there are increasing returns to higher education among all ethnic groups for men
and women, and there is no signiﬁcant evidence of ethnic discrimination against Malays
among men and women. Schafgans study is based on the cross-section data set of
MFLS2. The data is pooled from the new, panel and senior sample. The difference
between this study and Schafgans paper is in the estimation method and data set
coverage. This paper attempts to examine the non-linearity on returns to education and
relative earnings among the ethnic groups before and after the NEP.
To sum up, the key research questions in this paper are as follows:
(a) What are the earnings of Chinese and Indians relative to Malays before and after
the NEP; and
(b) What are the returns to education before and after NEP among the Malays,

Chinese and Indians.

Section 2 of this chapter contains a description of the data set. The descriptive
statistics by sample and race over time are presented in Section 3. Empirical results on
the earnings differentials and earnings growth are in Section 4. Section 5 discusses results
on the returns to education. Section 6 compares the cross-sectional and panel data results
as well as results based on earnings derived from main job and all jobs. The conclusions
of the study are in Section 7.

2. Data
The data for this paper is based on the ﬁrst and second wave of MFLS. The ﬁrst

wave was ﬁelded in 1977 while the second was carried out during the period August

 

wage = -384 + 49.6 age - .5150 age2

1988 through January 1989. RAND and the National Population and Family
Development Board of Malaysia conducted this survey. The overall purpose of the MFLS
was to enable the study of household behavior in diverse settings during a period of rapid
demographic and socioeconomic change.

The focus of this paper is on male earnings. This allows comparisons consistent
with the studies by Smith and Gallup, and avoids problems of sample selection, which
are more serious for the female sample. The panel sample from the ﬁrst wave consists of
1262 private households with at least one ever-married woman less than 50 years old at
the time of the initial visit. These households were located in 52 randomly selected
geographic areas to be representative of Peninsula Malaysia. 1047 husbands of the
female primary respondents were interviewed on the work history questionnaire. After
dropping respondents with missing values in work history data the number of male
respondents for MFLS] panel sample is 937. Job history data is collected retrospectively
starting from age 15 or at the time of entry into labor force for those who started working
after age 15. The job information is collected based on every job change or at every 3-
year interval.

The samples in the MFLSZ used for analysis in this study are the panel,
children and new sample. The husbands of Panel respondents that were interviewed in
the second wave are 717 respondents. The children sample that was interviewed for the
male life history data are selected sons and sons-in-laws of Panel women and comprise
833 respondents. The new sample consists of 1513 men who are husbands of new sample

women. After deleting respondents with missing data and inconsistent job history

information, the pooled panel and children sample comprises 1272 respondents and the
new sample consists of 1409 respondents.4

Work history data were collected retrospectively in both the ﬁrst and second wave
of the survey. But the critical difference is that information was collected only when
there were job changes in the second wave. In the ﬁrst wave, job history information was
recorded at regular 3-year intervals. The other problem is the manner which earnings data
were collected. In the ﬁrst wave only total earnings from all jobs were recorded, but for
second wave, earnings from main and secondary jobs were recorded separately. The
problem arises because only starting and ending earnings of every job change were
recorded in the second wave. As there are numerous cases with different starting and
ending dates of main and secondary jobs, we cannot simply aggregate the earnings to
obtain total earnings.5 In order to derive total earnings we need to interpolate the earnings
of both main and secondary jobs. Since the approach of interpolation of earnings is
subjective and inaccurate, the computation of total earnings through interpolation will
aggravate the measurement error problem. As such, this study uses earnings data from all
jobs for the ﬁrst wave. But for the second wave, earnings are from the main job only. The
sensitivity of using earnings from main job and all jobs is tested in Section 6. Another
difference is that earnings in kind were not imputed for the work history data in the
second wave. Therefore, for the sake of comparability, earnings for the ﬁrst and second

wave used in this study refer to monetary earnings (excluding earnings in kind).

 

4 The children sample used for analysis is 692 and panel sample is 580. The number of respondents for new
sample used by Gallup is 1412 respondents.

5 For example if the main job starts in 1984 and end in 1987, the secondary job starts in 1985 until 1988.
Then we need to interpolate the earnings of main job in 1985, 1986 as well as earnings of secondary job in
1986 and 1987 to derive the total earnings for 1985, 1986 and 1987.

The analyses from the ﬁrst wave comprise only the panel sample. It is divided
into two periods, that is, the pre-NEP period (1965-69), and early post-NEP period
(1971-76). The analyses from the second wave are for the late post-NEP period (1984-
88)6 based on the stratiﬁcation into children and panel sample, new sample, and the total

pooled children, panel and new sample.

3. Descriptive Statistics

The descriptive statistics of the Malays, Chinese and Indians over different time
periods are presented in Tables 1-3 respectively. It is interesting to note that the mean
real earnings are increasing over the three time periods, that is the pre-NEP, early post-
NEP and late post-NEP periods.7 The only exception is for the Chinese in the panel and
children sample in 1984-88, which recorded slightly lower mean real earnings. This is
partly due to the higher attrition rate of wealthier and more educated Chinese in the panel
sample and also earnings in 1984-88 covers only main job earnings.

It is also worthwhile to note that the years of education are increasing over time
for all races, but, that Malays recorded the fastest increase in the amount of schooling
after the NEP. The years of education of parents also recorded an upward trend over time
for all races.

In terms of age, the average age of the total panel sample in 1965-69 is 30.8 years
and 36.56 in 1971-76. In 1984-88, the mean age of the total pooled panel, children and

new sample is 34.52 years. It is noted that the panel sample has the highest mean age

 

6 For the second wave, only data for the last ﬁve years (1984-88) was used to avoid the problems of
memory recall in retrospective data which was encountered in Gallup’s study.
7 This is despite the fact that the real earnings for 1984-88 consist of main job earnings only.

10

(48.3 years), followed by the new sample (33.7 years) and the youngest are the children
sample (26.9 years).

There is an interesting development regarding the rate of urbanization over time
among the various races over time. It is observed that the percentage of Malays residing
in urban areas increased over time from 30.6% (pre-NEP period) to 43.7% in late post-
NEP period (total pooled sample). However, the percentage of Indians residing in urban
areas remained stable over time at around 50%. As for the Chinese, the percentage
residing in urban areas increased from 55.6% (pre-NEP period) to 81% (late-NEP
period).

One of the major strategies employed by the NEP policy is employment
restructuring. As such, the question of whether the occupational distribution among the
various races changes over time is a pertinent question. In order to answer this question, a
linear probability model is used to estimate the relative probability of the various races to
be engaged in certain occupations. Table 4 shows the results of the probability of being
employed in certain occupation groups of Chinese and Indians relative to Malays in 1976
and 1988 and the differences between the two periods. It is observed that Malays relative
to Chinese experienced a bigger shift out of the rural agricultural occupations to the more
urban and better paying occupations such as professional, clerical, service and production
related occupations.8 Similarly, there is larger movement out of agricultural occupations

into managerial, clerical, and service occupations by Malays relative to Indians. These

 

8 Agricultural occupations includes farmers operating on their own land, agricultural workers and
ﬁshermen. Production related workers includes miners, food and beverage processors, tailors, carpenters,
bricklayers, painters, blacksmith, plumbers, rubber and plastic product makers, chemical processors.
Service workers includes cooks, waiters, housekeeping and related service workers, cleaners, hairdressers,
protective service workers.

11

results provide some evidence that occupational restructuring between the races have

occurred during the post-NEP period.

4. Earnings Differential and Earnings Growth

4.1 Regression Analysis

Ordinary Least Squares regressions are estimated separately for the relevant
samples based on the pre-NEP (1965-69), early post-NEP (1971-76) and late post-NEP
(1984-88). The basic estimating model is of the form:

Y,. = or + 'y Chinese + 4) Indian + A education + BX ,. + 8 i.

where subscript i and t represents individual and year respectively, Y is the dependent
variable which is real monthly earnings in natural logarithm terms, Chinese is the
dummy variable for Chinese and Indian is a dummy variable for Indians, education is
years of education based on the linear spline speciﬁcation, X is a vector of control
variables such as year dummies, potential experience and its square, family background

characteristics, region dummies, and 8 is the error term.9

 

9 This regression model is applied to the samples in the ﬁrst and second wave of the survey.

12

4.2 Relative Earnings between Races

The simple speciﬁcation of earnings regression with ethnic and year dummies,
experience and its square for the pre and post NEP periods are presented in Table 5. It is
estimated that on average, Chinese earn 99.0 %10 more than Malays during the pre-NEP
period (1965-69) and declined to 91.4% more than Malays during the early post-NEP
(1971-76). In the late post-NEP period (1984-88), the earnings differential between
Chinese and Malays are further reduced. On average Chinese earn more than Malays by
57.9% (panel and children sample) and 86.6% (New sample) and 68.5% (Total pooled
sample). The Indians on average earn 39.4 % more than Malays for pre-NEP period and
25.86% in the early post-NEP period. During the late post-NEP period, the average
earnings of Indian are higher than Malays by 17.7% (Panel and children sample) and
7.0% (New sample) and 13.1% (Total pooled sample). The above results show that there
is a clear trend of a reduction of earnings gap between Chinese and Indians versus the
Malays during the post-NEP period.

One of the reasons for the narrowing of earnings differentials among the races is
the faster rate of increase in the levels of education and better returns to education of
Malays after the NEP compared to Indians and Chinese. The other reason is the larger
shift in occupations from agricultural sector to the higher paying labor market activities
in the service, sales, production, transportation sectors compared to the Chinese. While
the Malays have faster rate of growth in the professional and technical, administrative
and managerial occupations compared to Indians in the same occupation category.

Another possible reason is that the Malays enjoy a fastest rate of earnings growth

 

'° (e .6“- 1)*100% = 99.0%.

13

followed by Chinese and Indians during the post-NEP period. The discussion on the
analysis of annual earnings growth is presented in section 4.3 below.

This simple speciﬁcation is further augmented by adding controls such as years of
education, region dummies, parent’s education, number of jobs, starting and ending job
dummies. The reason for adding number of jobs as a control for 1965-69 and 1971-76
periods is because it covers data from the ﬁrst wave which collected only the aggregate
earnings for all jobs for respondents who have more than one job at the same time.
Starting and ending earning dummies are added for the period 1984-88 period of analysis
because only starting and ending earnings of jobs were collected retrospectively in the
second wave.11 The regression results are shown in Table 6. It is observed that even with
education, parent’s education, region of residence accounted for, Chinese still earn more
than Malays by 68.5% (pre-NEP), 59.0% (early post-NEP) and 54.5% (late post-NEP —
Total pooled sample). Indians on average still earn more than Malays by 14.1% (pre-
NEP), 8.8% (early post-NEP) and 17.2% (late post-NEP — total pooled sample).

For purposes of comparison of relative earnings among the races over time, the
simple speciﬁcation12 is preferred as it allows the coefﬁcients of the ethnic dummies to
reﬂect changes in the level of education, individual and family background
characteristics and region across groups and time. This is because the NEP is aimed at
correcting ethnic economic imbalances by raising the level of education, employment

restructuring to more urban labor market activities which involves the shift from rural to

 

” There are concerns that there may be problems of selectivity as those who change jobs may have
systematic differences with those who do not change jobs. Using starting and ending jobs is a crude way to
control for the type of data collected in the second wave. It is noted that introducing starting and ending job
dummies do not affect the estimated earnings differential or the returns to education.

14

urban areas. Therefore, it would not be appropriate to have education and regional
controls when comparing the relative earnings among races. Although this analysis is not
a direct test of the causal effect of NEP on the ethnic relative earnings, it does provide
some feedback on the outcomes of the relative earnings before and after N EP. The results
from the simple speciﬁcation in Table 5 do indicate that there is some degree of success
in the implementation of the NEP as the earnings differential between races have

narrowed considerably during the post-NEP period.

4.3 Attrition of Panel Sample

There are concerns that the narrowing of differentials between the races in the
panel and children sample in 1984-88 may be due to attrition of wealthier Chinese. In
view of this another set of earnings regression analysis is computed based on the
common panel sample for the pre-NEP and post-NEP periods. The earnings regression
results are shown in Table 7. A similar trend can be observed whereby the earnings
differentials between the Malays and Chinese narrowed from 86.8% (pre-NEP) to 84.4%
(early post-NEP) and 78.4% (late post-NEP). As for the earnings gap between the Malays
and Indians it also narrowed from 63.7% (pre-NEP) to 39.8% (early post-NEP) and

38.5% (late post-NEP).

 

12 Adding number of jobs and starting and ending job dummies do not change the relative earnings
differential. The results also do not change when year dummies are replaced with a time trend variable in
the simple speciﬁcation.

15

4.4 Earnings Growth

Since the data used are longitudinal, the earnings growth may also be attributed to
the changing sample compositions. As such, we cannot separate out the effects of
earnings growth due to productivity factors or due to changing sample composition. This
study attempts to get a cleaner estimate of earnings growth in the post-NEP period by
comparing the current earnings from all jobs for the ﬁrst wave (1976) and the second
wave (1988) of the same individual.13 The dependent variable is the log differences in
real earnings”, which represents an average annual percentage growth in real earnings. A
simple regression is estimated with log differences in real earnings on Chinese and Indian
dummy variables. This has the effect of controlling for individual ﬁxed effects because

the earnings growth is computed for each individual. The estimated results are as follows:

Annual real earnings growth = .268 - .006 Chinese - .013 Indian
(.004) (.005) (.008)

n = 499 R-squared = .006 p-values are in parentheses.
Although the results are not signiﬁcant at the 10% level, the estimated coefﬁcients above

show that Chinese and Indians respectively have a slower annual earnings growth of

- 0.6% and 1.3% than Malays over the period 1976 through 1988.15 However, the slower

 

'3 This is possible by using the panel sample of the ﬁrst and second wave of the survey. There are 580
cases of panel sample that were available in 1988. Out of these, there were 44 cases were unemployed in
1976 and thus no earnings were reported, 7 cases for other races, and 30 cases did not have adequate
information were dropped from the sample. The ﬁnal sample size is 499.

" Annual real earnings growth = (natural log earnings in 1988 - natural log earnings in l976)/12

'5 It is observed that based on earnings growth, the earnings gap (Table 9) between Indians and Malays
should have narrowed by a greater margin than the earnings gap between Chinese and Malays during the
early to late post-NEP periods. This discrepancy is because the earnings growth is based on current

16

earnings growth of Chinese and Indians relative to Malays presents part of the

explanation for the narrowing of earnings differentials among the races.

5. Returns to Education

5.1 Overall Sample

The returns to education with experience, its square, ethnic dummies, year
dummies, parents education, region dummies, number of jobs, and starting and ending
job dummies as controls are indicated in Table 6. A linear spline speciﬁcation is used to
show the non-linearity in returns to education. It is interesting to note that the returns to
education are highly signiﬁcant and convex for the pre-NEP and post-NEP periods. The
other striking trend is that the returns to primary, secondary and post-secondary
education are declining over time. In fact, there is a compression of returns to education
from the pre-NEP to post-NEP periods. This could be due to the rapid increase in supply
relative to the demand of educated workers with post-secondary education during the
post-NEP period. The next pertinent issue is the relative returns to education among the
main ethnic groups in Malaysia.

With monthly real earnings as the dependent variable, a linear spline speciﬁcation
of year of education interacted with race is used to estimate the returns to education. The
estimated results are reported in Table 816. It is observed that the coefﬁcient of returns to

primary education for Chinese relative to the Malays is negative but insigniﬁcant at the

 

‘earnings while the earnings differential is based on retrospective earnings for 1984-87 period which
captures earnings of those who change jobs only. It is observed that those who change jobs in the Indian
sample have relatively higher earnings which accounts for the smaller rate of decrease in earnings gap
between Indians and Malays.

17

10% level during the pre-NEP period. It is also insigniﬁcant during the early post-NEP
period. In the late post-NEP period, it is negative and insigniﬁcant for the pooled panel
and children sample, but, it is signiﬁcantly lower than Malays by 7.2% for the new
sample. The estimated returns to education to primary education for Indians are
signiﬁcantly lower than Malays for all time periods. However, there is a trend towards a
smaller difference in returns to primary education of Indians relative to Malays over time.

At the secondary level of education, the Chinese have signiﬁcantly lower returns
to secondary education compared to Malays for all time periods. The estimated returns to
secondary education of Chinese relative to Malays narrowed slightly by about 2% during
the early post-NEP and by about 4-7% depending on sample for the late post-NEP period.
The return to secondary education of Indians relative to Malays is negative but
insigniﬁcant for the pre-NEP, early post-NEP and late post-NEP (new sample). However,
it is signiﬁcantly lower than Malays for the total pooled sample during the late post-NEP
period.

The estimated returns to post-secondary education produce the most interesting
result. During the pre-NEP period, both Chinese and Indians respectively have
signiﬁcantly higher returns to post-secondary education of 20.1% and 17.8% compared
to the Malays. Although the Chinese and Indians still have higher returns to post-
secondary education during the early and late post-NEP, but, the magnitude of the
difference in the relative returns to education were substantially reduced. In fact, during
the early post-NEP period, the Chinese only have signiﬁcantly higher returns of 7.3%

and Indians 3.1% signiﬁcantly higher than Malays. In the late post-NEP period, the

 

‘6 Replacing year dummies with a time trend variable produce similar estimates.

18

returns to post-secondary education of Chinese and Indians are not signiﬁcantly different
from Malays based on the total sample. These ﬁndings show that Malays gained
substantial ground over the Chinese and Indians in terms of returns to post-secondary
education and may be due to the NEP’s quotas for Malays in positions requiring post-

secondary education.17

5.2 Labor Force Entry Before or After NEP

The next question of interest is to examine the difference in the returns to
education among the races for those who ﬁrst join the labor force before or after the NEP.
The issue of interest here is whether Malays have relatively better returns to education
than Chinese and Indians after the implementation of NEP. The second issue is whether
Malays who join the labor force would be able to beneﬁt more than Malays who enter the
labor force after the NEP. The reason is that Malays who are already working before
NEP would have relatively less labor mobility to beneﬁt from the preferential
employment policies.

Table 9 provides the results of the regression estimates for those respondents who
initially join the labor force before the NEP.18 Overall, the returns to primary education
between Chinese and Malays are not signiﬁcantly different during the pre-NEP, early
post-NEP and late post-NEP (pooled sample). As for the returns to secondary education,
Malays have signiﬁcantly higher returns than Chinese but the magnitude declined over

time. Regarding the returns to post-secondary education, Chinese have relatively higher

 

'7 Racial quotas are effectively enforced in the public sector and govemment-owned companies. But for the
private sector, ﬁscal incentives are utilized to encourage private companies to comply with the employment
quotas by race.

19

return (20.1%) to post-secondary education than Malays during pre-NEP period. But the
mangitude of the returns to post-secondary education were reduced to 14.4% during the
early post-NEP period. During the late post-NEP period, the returns to post-secondary
education of Chinese were not signiﬁcantly different from the Malays.

Comparing the relative returns to education of Malays and Indians at the primary
schooling level, it was found the Malays have signiﬁcantly higher return than Indians
during the pre-NEP and early post-NEP period. But it was not signiﬁcantly different
during the late post-NEP period. At the secondary education level, the returns to
education were not signiﬁcantly different for the three time periods. However, the return
to post-secondary education for Indians were relatively higher (17.8%) than Malays with
post-secondary education during the pre-NEP period. During the early post-NEP period,
the return to post-secondary education was still higher than Malays with similar
education level but it declined to 10.9%. But for the late post-NEP period, there is no
signiﬁcant difference in the returns to post-secondary education between the Malays and
Indians. To sum up, it appears among those who joined the labor force before NEP, the
relative returns to post-secondary education of Malays showed the biggest improvement
after the implementation of NEP.

Table 10 shows the regression results during the late post-NEP period for those
who start work after the NEP. Based on the total pooled sample, the returns to primary
and secondary education of Chinese and Indians are signiﬁcantly lower than Malays who
join the labor force after NEP during the late post-NEP period. The returns to secondary

education of Chinese and Indians who join the labor force after NEP are also

 

‘8 It is noted that there is no change in the estimates in the pre-NEP period in Table 9 and 10 because all

20

signiﬁcantly lower than Malays who join the labor force after NEP. While the returns to
post-secondary education of Chinese and Indians are signiﬁcantly higher than Malays
who join the labor force after NEP (panel and children sample). But there is no
signiﬁcant difference for the new sample and total pooled sample. These results imply
that during the late post-NEP period, Malays that join the labor force after NEP have the
advantage in returns to primary and secondary education. In terms of post-secondary
education, for the total pooled sample Malays have lower returns than Chinese and

Indians but it is not signiﬁcant.

5.3 Cohort Analysis

The next question is how does the NEP affect the young and the old? In order to
answer this question the sample is stratiﬁed into the younger age group who are between
15-34 years old and the older group who are 35-54 years old.19 The implementation of
the NEP created a signiﬁcant increase in the demand for Malays workers with higher
education. At the same time, the NEP also increased the supply of Malay workers with
higher education through the racial quota system for admissions to institutions of higher
learning. But it is expected that the supply of college educated Malays would lag behind
the demand for Malay college workers after the NEP policy was implemented. This is
because college education takes a long time to complete and the intake of students in
universities are limited. In addition, the expansion of Universities require a lot of

resources and time. The greater increase in demand for Malays with post-secondary

 

respondents that have earnings data in 1965-69 would join the labor force before NEP.
'9 The cut-off age of 54 years of completed age is because the ofﬁcial mandatory retirement age in
Malaysia is 55 years old.

21

education relative to its supply is expected to result in a higher premium attached to
Malay workers with post-secondary education after the NEP. Malays in the older cohorts
are expected to complete their education before the NEP and have less labor market
mobility. Malays in the younger cohorts are expected to make schooling decisions that
would take advantage of the favorable policies contained in the NEP. As a consequence
the better educated younger cohorts are expected to be in a better position to gain
employment in higher paying jobs. Therefore, it is expected that the NEP would beneﬁt
the younger Malays more than the older Malays. Overall, the return to post-secondary
education of Malays relative to non-Malays is expected to improve after the NEP.

The estimated results of the younger and older age cohorts are in Table 11 and 12
respectively. Among the younger age cohort, it is observed that there is no signiﬁcant
difference in the returns to primary education of Chinese relative to the Malays during
the pre-NEP and early post-NEP period. However, the return to primary education of
Chinese is signiﬁcantly lower than Malays during the late-post NEP period. As for
Indians, the results show that the return to primary education of Indians is lower than
Malays but the coefﬁcients are insigniﬁcant for all time periods. As for secondary
education, the return to education of Chinese is signiﬁcantly lower than Malays during
the pre-NEP (14.5%) and early post-NEP period (13.1%). However the differential return
to secondary education is reduced to (4.9%) with the Malays still enjoying signiﬁcantly
higher returns. A similar picture is obtained for the differential in returns to secondary
education for Indians and Malays. Although the return to secondary education of Indians
is signiﬁcantly lower than Malays over all time periods, there is a declining trend from

(8.3%) during pre-NEP to 7.1% (early post-NEP) and 3.8% (late post-NEP, total pooled

22

sample). As for the return to post-secondary education, the Chinese have a substantially
higher (24.1%) return than Malays during the pre-NEP period. But the return to post-
secondary education narrowed considerably during the early post-NEP period as the
Chinese only have signiﬁcantly higher (11.0%) return than Malays during the early post-
NEP period. During the late post-NEP period, the relative return to post-secondary
education is further reduced for the panel and children sample (8.4% higher for Chinese),
and insigniﬁcantly different for the new sample. Overall, for total pooled sample the
Chinese have a signiﬁcantly higher return to post-secondary education of 4.4% only
during the late post-NEP period. The younger cohort Indians also has a substantially
higher (22.3%) return to post-secondary education than Malays during the pre-NEP
period. This differential is reduced to 4.7% (not signiﬁcant) during the early NEP and
5.3% during the late post-NEP period (total pooled sample). Indians have a signiﬁcantly
higher return to post-secondary education than Malays by 11.6% during the post-NEP
period (panel and children sample). The above results show that the returns to post-
secondary education of Malays (relative to Chinese and Indians) in the younger age
cohort improved substantially during post-NEP period.

Next, the discussion is focussed on the results of the older age cohort in Table 12.
It is observed that there is signiﬁcantly lower (9.0%) return to primary education of
Chinese compared to the Malays during the pre-NEP period. However, there is no
signiﬁcant difference in the return to primary education between the Chinese and Malays
in the early post-NEP period, late post-NEP period. The results show that the return to
primary education of Indians is signiﬁcantly lower (8.6%) than Malays during the pre-

NEP period. It remained signiﬁcantly lower by 7.8% in the early post-NEP period and

23

5.0% during the late post-NEP period (pooled panel, children and new sample). In terms
of secondary education, the return to education of Chinese is higher than Malays but
insigniﬁcant during the pre-NEP period. But in the early post-NEP period the return to
secondary education relative to Malays is lower but insigniﬁcant. However, the return is
signiﬁcantly lower (4.1%) during the late post-NEP period (total pooled sample). The
return to secondary education of Indians relative to Malays also declined from the pre-
NEP to early post-NEP and late post-NEP periods. Among the older cohorts, the Malays
experienced a greater improvement in the return to post-secondary education than the
Chinese and Indians during the early and late post-NEP periods. However, among the
Malays there is no signiﬁcant difference in the returns to education of younger and older

cohort Malays.

6. Cross-section Analyses

Cross-section results in 1976 are shown in Table 13. The 1976 cross-section
results indicate that the retums to education at the primary, secondary and post-secondary
level are qualitatively similar to the 1971-76 period. The cross-section results based on
current earnings from main job and all jobs in 1988 are shown in Table 14. Besides
providing the opportunity to compare with 1984-88 results, it can also address the
concern over the difference in the use of the deﬁnition of earnings between main job
earnings (second wave) and all job earnings (ﬁrst wave). The main ﬁnding is that the
results from earnings of main job and all jobs are qualitatively similar for the panel, panel
and children, new, as well as total pooled sample. However, the results based on main job

earnings have smaller standard errors and higher R-squared for all the relevant samples.

24

This result is important because it conﬁrms that the difference in the deﬁnition of
earnings in the ﬁrst wave and second wave is not a serious problem.

Next the comparison of return to education for the panel and children sample for
1984-88 (Table 8) is compared with the 1988 cross-section results (Table 14). The return
to primary education interacted with race for panel and children are both insigniﬁcant but
have different signs. On the other hand, the returns to secondary and post-secondary
education have qualitatively similar results.

For the new sample, the comparison of 1988 cross-section and 1984-88 results in
terms of returns to primary, secondary and post-secondary education interacted with race
yield similar results qualitatively. As for the total pooled sample, the return to primary
education interacted with Indians is qualitatively different between the 1988 cross-
section and 1984-88 results. The return to secondary education interacted with race are
qualitatively similar. The returns to post-secondary education interacted with race are
both insigniﬁcant but have the same signs based on cross-section 1988 and 1984-88
period. With a few exceptions, it is observed that the overall cross-section results in 1976

and 1988 are qualitatively similar to the results of 1971-76 and 1984-88.

25

7. Conclusion

The NEP does seem to have played a role in narrowing the earnings differential of
the Malays relative to Indians and Chinese. From the pre-NEP to early post-NEP period,
the earnings gap of Chinese and Malays was reduced by about 7.6%. From the early post-
NEP period to late post-NEP period (total pooled sample) the earnings gap further
decreased by 22.9%. The earnings gap of Indians and Malays also reduced by about
13.5% from the pre-NEP to the early post-NEP period. This gap further declined by
12.8% from the early post-NEP to late post-NEP period (total pooled sample). The
decline in relative earnings can be attributed to the faster rate of increase in the level of
education and improvements in the returns to education (particularly post-secondary
education) of Malays. Another related reason is the larger shift from the agricultural
occupations to the more urban, higher paying occupations of Malays relative to Chinese
and Indians. However, despite the substantial convergence in relative earnings
differential, much work remains to done. This is evident as the Chinese and Indians still
earn more than Malays by 68.5% and 13.1% respectively during the late post-NEP period
(total pooled sample).

The estimated results show that there is increasing returns to education during the
pre-NEP and post-NEP periods. However, there is a convergence in returns to education
over time that may be attributed to the rapid increase in supply of more educated workers
in the labor market during the post-NEP period. When the linear spline education
variable is interacted with race, the most striking result concerns the return to post-

secondary education. Prior to the NEP, the Chinese and Indians had considerably higher

26

returns to post-secondary education than the Malays. This advantage of Chinese and
Indians was substantially reduced during the early post-NEP period. By the late post-
NEP period, this edge enjoyed by Chinese and Indians no longer existed. This result may
be attributed to the NEP, which is targeted to increase the demand for more skilled and
educated Malays in the labor market.

The analyses of the sample who join the labor force before NEP showed that the
returns to post-secondary education of Malays relative to Chinese and Indians have
improved during the early and late post-NEP periods. The analyses by cohorts reveal that
the NEP that both the younger and older Malays experience improvements in returns to
education relative to Chinese and Indians during the post-NEP period. However, there is
no signiﬁcant difference in the returns of post-secondary education between the younger
Malays and older Malays.

The use of cross-sectional and longitudinal data yields qualitatively similar
results, which means that the results obtained are quite robust. It also indicates that the
difference in the deﬁnition in earnings of main jobs and all jobs is not a serious one.

In conclusion, the NEP has made some progress in narrowing the relative
earnings differential as well as improving the returns to post-secondary education of

Malays.

27

References

Blackburn, ML. and Neumark, D. (1995). Are OLS Estimates of the Returns to
Schooling Biased Downward? Another Look. Review of Economics and Statistics, 72(2):
217-230.

DaVanzo, J, et. al. (1993). The Second Malaysian Family Life Survey - Survey
Instruments, RAND, Santa Monica, CA.

Gallup, IL. (1997). Ethnicity and Earnings in Malaysia, Development Discussion Paper,
Harvard Institute for International Development, MA.

Government of Malaysia (1986). Fifth Malaysia Plan - 1986 - 1990, National Printing
Department, K.L.

 

(1988). Economic Report 1988/89, National Printing Department, K.L.

Griliches, Z. (1977). Estimating the Returns to Schooling: Some Econometric Problems,
Econometrica, 45(1):1-22.

Haaga, J, et. al. (1993). The Second Malaysian Family Life Survey - Overview and
Technical Report, RAND, Santa Monica, CA.

Mincer, J. (1974). Schooling, Experience, and Earnings, Columbia University Press,
New York.

Peterson, C, et a1. (1993). The Second Malaysian Family Life Survey - User’s Guide,
RAND, Santa Monica, CA.

Rosenzweig, M. (1995). Why are there returns to schooling?, American Economic
Review, Papers and Proceedings, 85.2: 153-158.

Sahn, D. and H. Alderman (1988). The effects of human capital on wages and the

determinants of labor supply in a developing country, Journal of Development
Economics, 29: 157-183.

Schafgans, M.M.A. (1998). Ethnic wage differences in Malaysia: Parametric and
semiparametric estimation of the Chinese-Malay wage gap, Journal of Applied
Econometrics, 13:481-504.

Schultz, TR, (1988). Education investments and returns, in H. Chenery and TN.

Srinivasan (eds.), Handbook of Development Economics, Vol. 1, Amsterdam: North
Holland Press.

28

Schultz, T.W. (1975). The value of the ability to deal with disequilibria, Journal of
Economic Literature, 13:827-846.

Smith, J., (1991). Labor markets and economic development in Malaysia, in T.P. Schultz
(ed.), Research in Population Economics, Greenwich, CT: JAI Press.

Strauss, J. and D. Thomas, (1995). Human resources: empirical modeling of household
and family decisions, in TN. Srinivasan and J. Behrman (eds.), Handbook of
Development Economics 3A, Amsterdam: North Holland Press.

29

Table 1. Descriptive Statistics by Sample and Time of Malays

 

 

Sample Panel Panel Panel & New Total1
Children Pooled
Year 65-69 71-76 84-88 84-88 84-88
Variables
Log monthly 5.35 5.57 5.77 6.11 5.93
Real earnings (.952) (.856) (.689) (.660) (.696)
age 31.25 36.81 34.30 33.42 33.87
(9.85) (10.6) (12.7) (6.99) (10.3)
experience 20.58 25.91 20.42 18.90 19.68
(11.5) (12.3) (15.4) (8.67) (12.6)
years of 4.67 4.90 7.88 8.52 8.19
education (3.19) (3.33) (4.07) (3.52) (3.82)
mother’s .634 .678 1.52 1.62 1.57
education (1.35) (1.44) (2.22) (2.07) (2.15)
father’s 1.98 2.08 2.57 2.77 2.67
education (2.10) (2.22) (2.33) (2.38) (2.36)
small town .190 .186 .221 .363 .290
large town .053 .054 .070 .137 .103
city .063 .069 .046 .042 .044
no. of obs 756 1232 1441 1356 2797

 

Note: Standard deviations for continuous variables are in parentheses

(1) Total pooled sample includes panel, children and new sample in MFLS2

30

Table 2. Descriptive Statistics by Sample and Time of Chinese

 

 

Sample Panel Panel Panel & New Total1
Children Pooled
Year 65-69 71-76 84-88 84-88 84-88
Variables
Log monthly 6.06 6.24 6.21 6.72 6.45
Real earnings (.826) (.734) (.714) (.641) (.726)
age 30.29 36.50 36.12 35.88 36.01
(8.96) (9.47) (14.2) (6.70) (11.3)
experience 18.40 24.38 22.37 21.72 22.06
(10.5) (11.2) (16.5) (8.05) (13.2)
years of 5.90 6.12 7.75 8.17 7.95
education (3.61) (3.67) (3.84) (3.48) (3.68)
mother’s .847 .913 1.70 1.72 1.71
education (1.80) (1.94) (2.46) (2.43) (2.44)
father’s 2.42 2.49 3.17 3.22 3.20
education (2.57) (2.61) (2.66) (2.84) (2.75)
small town .306 .301 .561 .461 .514
large town .184 .183 .126 .253 .186
city .066 .067 .089 .132 .1 10
no. of Obs. 708 1091 613 553 1166

 

Note: Standard deviations for continuous variables are in parentheses

(1) Total pooled sample includes panel, children and new sample in MFLSZ

31

Table 3. Descriptive Statistics by Sample and Time of Indians

 

 

Sample Panel Panel Panel & New Total1
Children Pooled
Year 65-69 71-76 84-88 84-88 84-88
Variables
Log monthly 5.70 5.82 5.96 6.17 6.09
Real earnings (.754) (.696) (.688) (.550) (.619)
age 31.03 35.84 32.81 32.36 32.54
(9.08) (10.0) (12.6) (7.31) (9.82)
experience 19.26 24.05 18.65 18.57 18.61
(10.7) (11.3) (13.4) (8.02) (10.6)
years of 5.77 5.80 8.16 7.79 7.94
education (3.82) (3.69) (3.58) (3.36) (3.46)
mother’s 1.28 1.13 2.39 1.92 2.12
education (2.23) (2.00) (2.57) (2.24) (2.39)
father’s 2.92 2.72 2.78 2.72 2.74
education (2.75) (2.51) (2.51) (2.63) (2.58)
small town .239 .250 .358 .269 .306
large town .144 .138 .084 .160 .129
city .112 .112 .068 .029 .046
no. of obs. 222 340 380 543 923

 

Note: Standard deviations for continuous variables are in parentheses

(1) Total pooled sample includes panel, children and new sample in MFLS2

32

Table 4. Relative Probability to be Employed in Certain Occupations
in 1976 and 1988

 

 

Occupation Panel Sample New Sample Difference
1976 1988 1988-1976

Agriculture

Chinese - 25.96%** - 14.77%** 11.19%

Indian - 6.98% - 4.04% 2.94%

Professional

Chinese 1.01% - 2.13% -3.14%

Indian - 2.66% 0.19% 2.85%

Manager

Chinese 1.23% 6.76%** 5.53%

Indian - 0.04% - 0.20% -0.16%

Clerical

Chinese 3.85%* - 4.28%** -8.13%

Indian 5.03%* - 2.47% -7.50%

Sales

Chinese 16.78%** 20.45%** 3.67%

Indian - 4.83% 1.03% 5.86%

33

Table 4 (cont’d).

Service

Chinese - 1.43%
Indian 0.49%
Production

Chinese 12.31% *
Indian - 1.94%
Transport

Chinese - 0.52%*
Indian 2.44%
Laborer

Chinese - l.43%**
Indian 049%“

-21.52%**

-12.90%**

7.36%**

2.15%

8.89%**

14.69%**

- 0.76%

1.56%

-20.09%

-13.39%

-4.95%

4.09%

9.41%

17.13%

0.67%

1.07%

 

Note: * = signiﬁcant at 5% level ** = signiﬁcant at 1% level

Relative probabilities are based on linear probability model computed separately

for each occupation category.
Reference group = Malays.

34

Table 5. Real Earnings Regression - Simple Specification

 

 

Sample Panel Panel Panel & New Total1
Children Pooled
Year 65-69 71-76 84-88 84-88 84-88
Chinese .688 .649 .457 .624 .522
(.046) (.032) (.032) (.032) (.024)
Indian .332 .230 .163 .068 .123
(.061) (.045) (.038) (.029) (.023)
experience .041 -.004 .036 .022 .038
(.007) (.006) (.003) (.006) (.003)
experience2 -.001 -0002 -.0007 -.0008 -.0008
(.0001) (.0001) (.00006) (.0001) (.00005)
66/72/85 .128 .131 .118 -004 .049
(.068) (.061) (.057) (.052) (.039)
67/73/86 .072 .1 16 .010 .060 .021
(.072) (.059) (.058) (.053) (.040)
68/74/87 .088 -.017 .024 .091 .019
(.068) (.058) (.059) (.063) (.044)
69/75/88 .231 .046 .205 .313 .237
(.068) (.061) (.049) (.040) (.032)
76 .120
(.049)
constant 4.96 5.72 5.37 5.82 5.50
(.089) (.077) (.050) (.066) (.037)
R2 .1557 .1728 .1448 .1951 .1758
n 1686 2663 2434 2452 4886

 

Note: Robust standard errors are in parentheses. Year dummies are based on the relevant time
period. (1) Total pooled sample includes panel, children and new sample in MFLSZ

35

Table 6. Real Earnings Regression - with Additional Controls

 

 

Sample Panel Panel Panel & New Totarl
Children Pooled
Year 65-69 71-76 84-88 84-88 84-88
Chinese .522 .464 .362 .527 .435
(.042) (.029) (.030) (.029) (.021)
Indian .131 .084 .137 .147 .159
(.046) (.033) (.030) (.025) (.019)
experience .089 .050 .060 .059 .066
(.007) (.006) (.003) (.006) (.003)
experience2 -002 -.0008 -.0009 -.001 1 -001
(.0002) (.0001) (.00006) (.0001) (.00005)
66/72/85 .127 .113 .077 -.028 .015
(.060) (.053) (.051) (.047) (.035)
67/73/86 .061 .070 .0008 .002 -.017
(.062) (.051) (.051) (.046) (.035)
68/74/87 .065 -.084 -.O44 -.019 -.061
(.060) (.050) (.053) (.055) (.038)
69/75/88 172 -.034 -.067 -.020 -.079
(.060) (.050) (.059) (.058) (.042)
76 -.008
(.042)
Education .067 .055 .03 l .03 1 .028
(0-6 years) (.012) (.009) (.009) (.010) (.007)
Education .136 .159 .112 .091 .109
(7- 12 years) (.014) (.010) (.008) (.007) (.005)
Education .290 .200 .172 .187 .177
(13 + years) (.030) (.016) (.015) (.011) (.009)
Mother’s .011 .019 .011 .014 .012
Education (.010) (.007) (.006) (.005) (.004)
Father’s .034 .020 -.008 .004 -.002
Education (.009) (.006) (.006) (.005) (.004)

36

Small town

Large town

City

Number of

jobs

Start earnings

End earnings

constant

.251
(044)

.189
(.056)

.314
(.056)

.072
(.064)

3.72
(.129)

.3574

1686

.207
(.030)

.181
(.039)

.350
(.042)

.022
(.036)

4.39
(.092)

.4140

2663

Table 6 (cont’d).

.221
(.027)

.167
(.045)

.168
(.048)

-.252
(.048)

-.125
(.051)

4.71
(.089)

.4060

2434

.1 10
(.024)

.187
(.033)

.265
(.053)

-.226
(.050)

-.131
(.054)

4.99
(.098)

.4425

2452

.174
(.018)

.219
(.027)

.226
(.037)

-.252
(.036)

-.134
(.038)

4.78
(.065)

.4342

4886

 

Note: Robust standard errors are in parentheses. Year dummies are based on the relevant time

period. (1) Total pooled sample includes panel, children and new sample in MFLS2

37

Table 7. Real Earnings Regression - Simple Speciﬁcation
(Common panel sample for 3 periods)

 

 

Sample Panel Panel Panel
Year 65-69 71-76 84-88
Chinese .625 .612 .579
(.061) (.043) (.055)
Indian .493 .335 .326
(.085) (.062) (.089)
experience .026 -.01 1 -.053
(.011) (.008) (.016)
experience2 -.0006 .000005 .0003
(.0003) (.0001) (.0002)
66/72/85 .123 .120 .183
(.096) (.081) (. 128)
67/73/86 .077 .121 .288
(.092) (.075) (.146)
68/74/87 .072 -.0005 .21 1
(.088) (.070) (. 159)
69f75/88 .204 .005 .480
(.093) (.081) (.112)
76 .084
(.060)
constant 5.07 5.78 6.81
(.126) (.106) (.321)
R2 .1260 .1538 .2543
n 861 1366 668

 

Note: Robust standard errors are in parentheses. Year dummies are based on the relevant time
period. There are more observations for respondents who change jobs more frequently. This
sample excludes those who joined the labor force after NEP (1970).

38

Table 8. Real Earnings Regression by Sample with interaction between
Race and Years of Education

 

 

Sample Panel Panel Panel & New Total
Children Pooled
Year 65-69 71-76 84-88 84-88 84-88
Education .086 .060 .033 .058 .039
(0 - 6 years) (.018) (.012) (.011) (.016) (.009)
Chinese*(edu -.O31 .009 -.009 -.072 -.027
0 - 6 years) (.022) (.015) (.020) (.029) (.017)
Indians*(edu -.058 -.052 -.O27 -.042 -.028
0 - 6 years) (.024) (.017) (.022) (.021) (.014)
Education .202 .204 . 129 . 101 . 124
(6 — 12 years) (.022) (.015) (.010) (.008) (.006)
Chinese*(edu -.104 -.086 -.059 -.027 -.046
6 - 12 years) (.027) (.019) (.017) (.014) (.011)
Indians*(edu -.022 -.003 -.029 -.013 -.023
6 — 12 years) (.032) (.023) (.016) (.013) (.010)
Education .125 .156 .156 .199 .178
(13 + years) (.046) (.022) (.018) (.014) (.012)
Chinese*(edu .201 .073 .054 -.034 .004
13 + years) (.067) (.035) (.037) (.027) (.023)
Indians*(edu . 178 .031 .050 -.028 -.0005
13 + years) (.062) (.034) (.042) (.024) (.026)
Chinese .731 .497 .525 1.01 .688
(.097) (.066) (.099) (.161) (.088)
Indian .360 .297 .339 .416 .366
(.107) (.073) (.109) (.112) (.073)
experience .089 .05 1 .061 .060 .067
(.007) (.006) (.003) (.006) (.003)
experience2 -.002 -.0008 -0009 -.001 1 -.001
(.0002) (.0001) (.00005) (.0001) (.00005)
66f72/85 .111 .111 .081 -.028 .017
(.059) (.053) (.051) (.047) (.035)

39

67/73/86

68/74/87

69/75/88

76

Mother’s

Education

Father’s
Education

Small town

Large town

City

Number of

jobs

Start earnings

End earnings

constant

R2

No of. Obs.

.043
(.062)

.059
(.059)

156
(.059)

.011
(.010)

.034
(.009)

.237
(.045)

.185
(.055)

.289
(.056)

.078
(.065)

3.62
(.143)

.3680

1686

Table 8 (cont’d).

.059
(.050)

-.088
(.050)

-.043
(.050)

-.016
(.042)

.020
(.007)

.018
(006)

.189
(.031)

.188
(.039)

.333
(.043)

.031
(.037)

4.33
(.103)

.4208

2663

-.003
(.051)

-.041
(.053)

-.074
(.059)

.011
(006)

-.008
(.006)

.211
(.027)

.163
(.045)

.186
(.048)

-.258
(.048)

-.132
(.051)

4.66
(.094)

.4116

2434

.006
(.046)

-.029
(.055)

-.029
(.058)

.013
(.005)

.005
(.005)

.102
(.024)

.179
(.033)

.273
(.053)

-.227
(.050)

-. 130
(.054)

4.79
(.118)

.4496

2452

-.016
(.035)

-.063
(.038)

-.O86
(.042)

.012
(004)

-.002
(004)

.163
(.018)

.215
(.027)

.242
(.037)

-.253
(.036)

-.l36
(.038)

4.67
(.073)

.4396

4886

 

Note: Robust standard errors are in parentheses. Year dummies are based on the relevant time

period. Total pooled sample includes panel, children and new sample in MFLSZ.

Table 9. Real Earnings Regression by Sample with interaction between
Race and Years of Education

(Respondents that join labor force before NEP)

 

 

Sample Panel Panel Panel & New Total
Children Pooled
Year 65-69 71-76 84-88 84-88 84-88
Education .086 .057 .03 1 .063 .038
(O — 6 years) (.018) (.013) (.012) (.020) (.010)
Chinese*(edu -.031 .006 .01 1 -.069 -.009
0 - 6 years) (.022) (.015) (.023) (.033) (.018)
Indians*(edu -.058 -.050 -.014 -.054 -.034
0 — 6 years) (.024) (.018) (.034) (.032) (.023)
Education .202 .212 .139 .1 19 .125
(6 — 12 years) (.022) (.016) (.018) (.015) (.011)
Chinese*(edu -.103 -.094 -.060 -.O37 -.039
6 — 12 years) (.027) (.020) (.031) (.022) (.018)
Indians*(edu -.022 -.023 .032 .025 .026
6 — 12 years) (.032) (.024) (.030) (.023) (.018)
Education .125 .122 .183 .146 . 159
(13 + years) (.046) (.027) (.046) (.042) (.031)
Chinese*(edu .201 .144 -.078 .003 -.072
13 + years) (.067) (.037) (.094) (.102) (.073)
Indians*(edu .178 .109 -.109 -.039 -.055
13 + years) (.062) (.043) (.091) (.050) (.065)
Chinese .731 .506 .425 .988 .612
(.097) (.066) (.107) (.180) (.092)
Indian .360 .274 .173 .431 .323
(.107) (.076) (.172) (.174) (.119)
experience .089 .045 .034 .076 .035
(.007) (.006) (.010) (.018) (.007)
experience2 -002 -0007 -0005 -0013 -.0006
(.0002) (.0001) (.0001) (.0003) (.0001)
66/72/85 .111 .118 .267 .044 .153
(.059) (.054) (.090) (.083) (.062)

41

Table 9 (cont’d).

67/73/86 .043 .071 .229 .092 .169
(.062) (.052) (.092) (.081) (.061)
68/74/87 .059 -.079 .229 .036 .133
(.059) (.051) (.102) (.092) (.070)
69/75/88 156 -.038 .151 -.070 .055
(.059) (.051) (.133) (.102) (.083)
76 -.010
(.043)
Mother’s .01 1 .026 .005 .016 .013
Education (.010) (.007) (.013) (.009) (.008)
Father’s .034 .022 -.012 .01 1 .001
Education (.009) (.006) (.010) (.008) (.006)
Small town .237 . 189 .250 .035 .136
(.045) (.031) (.046) (.037) (.028)
Large town .185 .178 .248 .122 .198
(.055) (.041) (.090) (.050) (.043)
City .289 .359 .115 .247 .250
(.056) (.041) (.092) (.083) (.066)
Number of .078 .032
jobs (.065) (.037)
Start earnings -.361 -.387 -.371
(.117) (.089) (.071)
End earnings -.079 -.235 -. 159
(.l 18) (.096) (.074)
constant 3.62 4.42 4.86 4.51 5.02
(. 143) (.113) (.230) (.273) (.159)
R2 .3680 .4194 .4015 .4171 .4162
No of. Obs. 1686 2548 1052 1 143 2195

 

Note: Robust standard errors are in parentheses. Total pooled sample includes panel, children and
new sample in MFLSZ.

42

Table 10. Real Earnings Regression by Sample with interaction between

Race and Years of Education
(Respondents that join labor force after NEP)

 

 

Sample Panel & New Total
Children Pooled
Year 84-88 84-88 84-88
Education .073 .056 .064
(0 - 6 years) (.033) (.022) (.019)
Chinese*(edu -.098 -.101 -.104
0 - 6 years) (.060) (.059) (.051)
Indians*(edu -.071 -.031 -.046
O — 6 years) (.039) (.027) (.023)
Education .115 .099 .117
(6 — 12 years) (.013) (.011) (.008)
Chinese*(edu -.O45 -.038 -.048
6 — 12 years) (.023) (.022) (.017)
Indians*(edu -.047 -.034 -.042
6 - 12 years) (.020) (.017) (.013)
Education .159 .212 .186
(13 + years) (.019) (.014) (.012)
Chinese*(edu .089 -.035 .034
13 + years) (.038) (.030) (.024)
Indians*(edu . 105 -.009 .038
13 + years) (.036) (.026) (.022)
Chinese .978 1.22 1.12
(.330) (.332) (.285)
Indian .612 .362 .478
(.208) (. 139) (.123)
experience .108 .069 .1 1 1
(.008) (.011) (.006)
experience2 -0025 -0013 -0025
(.0004) (.0004) (.0003)

43

85

86

87

88

Mother’s

Education

Father’s

Education

Small town

Large town

City

Start earnings

End earnings

constant

R2

No of. Obs.

-.026
(.058)

-.138
(.057)

-.202
(.059)

-.234
(.064)

.016
(006)

-.002
(.007)

.172
(.033)

.110
(.048)

.193
(.056)

-.184
(.052)

-.132
(.056)

4.36
(.205)

.4704

1382

Table 10 (cont’d).

-.071
(.055)

-.056
(.056)

-.070
(064)

-.03 l
(.067)

.01 1
(006)

.0009
(.006)

.149
(.032)

.205
(.042)

.303
(.063)

-.118
(.058)

-.053
(.062)

4.72
(.156)

.5032

1309

-.062
(.040)

-.122
(.040)

-.180
(.043)

-. 190
(.047)

.013
(.004)
(.004)

.168
(.023)

.203
(.032)

.237
(.043)

-. 162
(.040)

-. 100
(.043)

4.36
(.127)

.4991

2691

 

Note: Robust standard errors are in parentheses. Total pooled sample includes panel, children and
new sample in MFLS2.

Table 11. Real Earnings Regression by Sample with interaction between
Race and Years of Education

(age cohort 15-34 years)

 

 

Sample Panel Panel Panel & New Total
Children Pooled
Year 65-69 71-76 84-88 84-88 84-88
Education .055 .074 .043 .041 .039
(0 - 6 years) (.021) (.020) (.025) (.024) (.018)
Chinese*(edu .014 .037 -.102 -.102 -.101
0 - 6 years) (.030) (.027) (.056) (.050) (.041)
Indians*(edu -.044 -.043 -.060 -.023 -.030
0 — 6 years) (.031) (.028) (.031) (.029) (.021)
Education .247 .225 .l 19 .080 .1 1 1
(6 — 12 years) (.020) (.015) (.012) (.011) (.008)
Chinese*(edu -.145 -. 131 -.040 -.029 -.049
6 - 12 years) (.026) (.021) (.023) (.020) (.016)
Indians*(edu -.083 -.071 -.050 -.021 -.O38
6 — 12 years) (.036) (.028) (.020) (.018) (.014)
Education .127 .195 . 127 .221 .170
(13 + years) (.051) (.024) (.021) (.015) (.014)
Chinese*(edu .241 .1 10 .084 -.037 .044
13 + years) (.080) (.043) (.038) (.033) (.026)
Indians*(edu .223 .047 .1 16 -.01 1 .053
13 + years) (.066) (.035) (.044) (.055) (.034)
Chinese .507 .472 .989 1 . 16 1.08
(.140) (.139) (.306) (.282) (.226)
Indian .380 .410 .544 .322 .389
(.142) (.130) (.165) (.150) (.113)
experience . 126 .068 .1 13 .091 .120
(.014) (.015) (.008) (.012) (.006)
experience2 -.003 -0012 -.003 -0025 -003
(.0005) (.0005) (.0004) (.0004) (.0003)
66n2/85 .147 .144 -.029 -.024 -.045
(.068) (.064) (.056) (.049) (.037)

45

67/73/86

68/74/87

69/75/88

76

Mother’s

Education

Father’s
Education

Small town

Large town

City

Number of

jobs

Start earnings

End earnings

constant

R2

No of. Obs.

.042
(.071)

.075
(.066)

.192
(.067)

.015
(.013)

.014
(.011)

.141
(.053)
(.068)

.168
(.069)

.065
(.074)

3.62
(.171)

.3475

1115

Table 1 1 (cont’d).

.080
(.064)

-.086
(.063)

-.008
(.065)

-.017
(.054)

.006
(.009)

.003
(.008)

.133
(.043)

.065
(.052)

.175
(.061)

.072
(.046)

4.09
(.154)

.4805

1 203

-.116
(.057)

-.160
(.058)

-.208
(.063)

.013
(.007)

-.003
(.007)

.149
(.032)

.114
(.049)

.200
(.055)

-.187
(.052)

-.101
(.056)

4.53
(.163)

.3898

1369

.012
(.050)
(060)

.022
(.065)

(006)

.007
(006)

.114
(.031)

.149
(.042)

.254
(.065)

-.141
(060)

-.047
(.062)

4.77
(.170)

.4153

1361

-.089
(.037)

-.l32
(.041)

-.148
(.046)

.010
(.005)

.003
(004)

.137
(.023)

.176
(.032)

.222
(.043)

-.l77
(.041)

-.083
(.043)

4.50
(.119)

.4149

2730

 

Note: Robust standard errors are in parentheses. Total pooled sample includes panel, children
and new sample in MFLSZ.

46

Table 12. Real Earnings Regression by Sample with interaction between
Race and Years of Education
(age cohort 35-54 years)

 

 

Sample Panel Panel Panel & New Total
Children Pooled
Year 65-69 71-76 84-88 84-88 84-88
Education .121 .058 .043 .056 .043
(0 — 6 years) (.035) (.018) (.013) (.021) (.012)
Chinese*(edu -.090 —.029 .031 -.063 -.01 1
O - 6 years) (.038) (.021) (.027) (.036) (.022)
Indians*(edu -.086 -.078 .027 -.078 -.050
0 - 6 years) (.042) (.024) (.043) (.037) (.027)
Education .033 .198 .114 .130 .123
(6 — 12 years) (.076) (.036) (.020) (.016) (.012)
Chinese*(edu .075 -.054 -.076 -.036 -.041
6 — 12 years) (.087) (.041) (.032) (.021) (.018)
Indians*(edu .134 .021 .044 -.002 .012
6 — 12 years) (.078) (.037) (.035) (.022) (.018)
Education . 146 .048 .243 . 183 .198
(13 + years) (.122) (.050) (.036) (.026) (.021)
Chinese*(edu .063 . 134 -.010 -.018 -.038
13 + years) (.127) (.060) (.081) (.047) (.044)
Indians*(edu .071 .130 -.088 -.025 -.O41
13 + years) (.121) (.062) (.065) (.035) (.031)
Chinese .950 .554 .395 .971 .661
(.136) (.077) (.127) (.197) (.113)
Indian .323 .301 -.040 .553 .413
(.168) (.093) (.216) (.204) (.144)
experience -.050 .057 .023 .081 .043
(.054) (.024) (.024) (.029) (.016)
experience.2 .0007 -.001 -0003 -0014 -.0007
(.0008) (.0004) (.0004) (.0005) (.0003)
66/72/85 .040 .028 .299 .012 .150
(. 120) (.087) (.098) (.118) (.079)

47

67/73/86

68/74/87

69/75/88

76

Mother’s

Education

Father’s
Education
Small town
Large town
City

Number of
jobs

Start earnings
End earnings

constant

R2

No of. Obs.

.076
(. 127)

.018
(.122)

.055
(.121)

.002
(.017)
(.016)

.461
(.088)

.348
(.091)

.513
(.090)

.093
(.105)

5.40
(.917)

.4428

551

Table 12 (cont’d).

-.001
(.081)

-.132
(.083)

-.O92
(.081)

-.041
(.069)

.033
(.011)

.027
(.009)

.227
(.046)

.293
(.062)

.459
(.054)

.005
(.050)

4.41
(.413)

.4113

1352

.315
(.102)

.214
(.110)

.100
(.126)

-.005
(.015)

-.002
(.010)

.246
(.048)

.307
(.090)

.126
(.110)

-.422
(.110)

-.336
(.107)

5.03
(.412)

.5140

835

.116
(.109)

.089
(.119)

-.014
(.117)

.022
(.009)

.002
(.008)

.079
(.040)

.178
(.051)

.271
(.084)

-.330
(.086)

-.217
(.092)

4.41
(.441)

.4726

1090

.201
(.077)

.129
(.083)

.058
(.088)

.013
(.008)

-.0007
(.006)

.158
(.030)

.255
(.043)

.269
(.068)

-.354
(.070)

-.248
(.071)

4.88
(.276)

.4924

1925

 

Note: Robust standard errors are in parentheses. Total pooled sample includes panel, children and

new sample in MFLSZ.

Table 13. Real Earnings Regression by Sample in 1976 Cross-section

 

 

Sample Panel
Education .050
(.016)
Chinese*(education .004
0 - 6 years) (4)24)
Indians*(education - .038
0 - 6 years) (.036)
Education .202
(6 — 12 years) (.023)
Chinese*(education - .062
6 — 12 years) (030)
Indians*(education .007
6 — 12 years) (.043)
Education .152
(13 + years) (.060)
Chinese*(education .052
13 + years) (079)
Indians*(edu .006
13 + years) (.099)
Chinese .447
(.108)
Indian .236
(.166)
experience .043
(.008)
experience2 -.0006
(.0001)
Mother’s .013
Education (.014)

49

Father’s

Education

Small town

Large town

City

Number of Jobs

constant

R2

No of. Obs.

Table 13 (cont’d).
.018
(.010)

.216
(.053)

.220
(.070)

.391
(.082)

-.005
(-044)

4.52
(.157)

.4422

937

 

Note: Standard errors are in parentheses.

50

Table 14. Real Earnings Regression by Sample in 1988 Cross-section

 

 

 

 

 

 

Sample Panel Panel & New Total
Children Pooled
Source of Main All Main All Main All Main All
Earnings
Education .024 .013 .029 .021 .039 .035 .028 .022
(O — 6 years) (.018) (.018) (.014) (.014) (.016) (.016) (.010) (.011)
Chinese*(edu .019 .034 .008 .017 -.041 -.036 -.003 .006
0 - 6 years) (.029) (.029) (.024) (.024) (.026) (.026) (.017) (.017)
Indians*(edu .005 .014 .030 .055 -.007 .005 .01 l .028
0 - 6 years) (.060) (.060) (.041) (.041) (.028) (.029) (.023) (.024)
Education .156 .147 .142 .129 .136 .123 .149 .135
(6 — 12 years) (.030) (.030) (.014) (.014) (.010) (.010) (.008) (.008)
Chinese*(edu - .1 16 -.1 18 -.085 -.076 -.050 -.O39 -.068 -.056
6 - 12 years) (.044) (.044) (.023) (.023) (.016) (.016) (.014) (.014)
Indians*(edu .030 .039 - .030 -.O32 -.056 -.046 -.052 -.045
6 — 12 years) (.067) (.066) (.028) (.028) (.018) (.018) (.016) (.016)
Education .123 .115 .173 .174 .198 .206 .178 .183
(13 + years) (.063) (.061) (.023) (.023) (.019) (.019) (.015) (.015)
Chinese*(edu .101 . l 1 1 .080 .068 -.O31 -.043 .019 .007
13 + years) (.094) (.092) (.045) (.046) (.037) (.037) (.029) (.029)
Indians*(edu -.176 -.174 -.015 .002 -.017 -.028 -.010 -.010
13 + years) (.119) (.118) (.054) (.054) (.039) (.039) (.032) (.032)
Chinese .427 .296 .498 .381 .885 .821 .621 .515
(.135) (.135) (.115) (.117) (.141) (.142) (.088) (.089)
Indians -.009 -. 127 .032 -. 138 .302 .202 .201 .061
(.302) (.301) (.213) (.217) (.151) (.152) (.122) (.123)
Experience .026 .023 .072 .072 .076 .079 .077 .078
(.015) (.014) (.004) (.005) (.009) (.009) (.003) (.003)
Experience2 .0005 -0004 -.001 -001 -001 -001 -.001 -.001
(.0002) (.0002) (.00007) (.00007) (.0002) (.0002) (.00006) (.00006)

51

Table 14 (cont’d).

Mother’s -.010 -.010 .006 .008 .010 .010 .009 .009
education (.170) (.169) (.009) (.009) (.006) (.007) (.005) (.005)

Father’s .006 .004 .003 .002 .014 .014 .009 .009
education (.014) (.014) (.008) (.008) (.006) (.006) (.005) (.005)

small town .319 .297 .254 .229 .152 .148 .206 . 187
(.069) (.068) (.041) (.041) (.031) (.032) (.025) (.025)

large town .426 .399 .253 .234 .212 .193 .257 .235
(.117) (.116) (.063) (.064) (.041) (.042) (.035) (.036)

city .221 .188 .277 .251 .319 .280 .317 .283
(. 187) (. 186) (.084) (.086) (.064) (.065) (.052) (.053)

constant 5.23 5.49 4.37 4.54 4.50 4.59 4.39 4.53
(.316) (.300) (.102) (.104) (. 126) (.127) (.075) (.076)

R2 .4468 .3984 .4477 .4124 .4853 .4595 .4781 .4474

No of. Obs. 580 580 1272 1272 1409 1409 2681 2681

 

Note: Standard errors are in parentheses. Total pooled sample includes panel, children and new sample in
MFLS2. Main refers to earnings from main job only. All refers to aggregate earnings from all jobs.

52

CHAPTER 2

DISTRIBUTION OF SCHOOLING AND
EARNINGS INEQUALITY IN MALAYSIA

1. Introduction

Income inequality is an important policy issue because if distributions of income
within and between groups are uneven, then large segments of the population are not
reaping the beneﬁts of economic growth. It is even more important in Malaysia because
of the sensitivity of income inequality among the races. This is reﬂected in the New
Economic Policy (NEP) which has the prime objective to reduce poverty and income
disparities.

A number of strategies are employed by policy makers to reduce income
inequality. These strategies are designed to promote economic growth with equity.l It is
essential to enlarge the economic pie to ensure that in the course of restructuring society
to enable Bumiputra2 to participate in higher income earning activities, other groups in
the Malaysian society do not experience any sense of loss or feel a sense of deprivation.
One of the major policy instruments utilised in raising overall income and reducing
inequality in Malaysia is through the expansion of education.3 The emphasis on
education is evident from the sizeable budget allocation of development expenditure for

social services. In 1988, 1.1 billion ringgit was allocated for education that is 71.2% of

 

' Fifth Malaysia Plan 1986-1990 pp. 21-28 provides a discussion of the strategies for growth and
distribution issues.

2 Bumiputra comprises mainly Malays and a very small fraction of indigenous ethnic group.

3 Mahatir Mohamed (1998) also concurred with the importance of the role of education in raising overall
income and reducing income inequality.

53

total development expenditure for social services.4 Moreover, the importance of
education in the process of economic development is well documented. The importance
of education is evident from the numerous advantages of educational investment and
expansion.5 For instance, education provides direct satisfaction to persons during school
and later in life. It is an important means to provide trained and skilled manpower, which
is essential for the expanding industrial and manufacturing sector in Malaysia. Education
enhances productivity and income, and thus contributes to economic growth.

Education investments in Malaysia are aimed at boosting economic growth,
achieving wider diffusion of economic opportunities and reducing income inequality.
However, it should to be pointed out that the NEP complements the education policies to
reduce income inequality. The NEP is considered as a socio-economic engineering
program that is intended to redistribute wealth and to provide equitable spread of
education and employment opportunities to reﬂect the racial composition of Malaysia.
The NEP, through the strategy of employment restructuring, is designed to enable the
disadvantaged group to beneﬁt more from increases in the quantity and quality of
employment. The signiﬁcance and emphasis on education in Malaysia provided the
motivation to examine the role of education on the earnings distribution in Malaysia.

The main focus of this paper is on the effect of education on earnings inequality.
The key research goals are:

(1) to establish the link between the distribution of education and the

distribution of earnings over time with reference to the education policy

and New Economic Policy; and

 

4 It accounts for 18.3% of total development expenditure for 1988.

54

(2) to examine the intergenerational transmission of schooling over time;

2. Literature Review

The initial studies on the effects of education on income distribution are at the
aggregate level across countries. Most of these studies, Chiswick (1971), Adelman and
Morris (1973), Chenery and Syrquin (1975), Park (1996), De Gregorio and Kim (1999)
found that higher educational attainment and more equal distribution of education have a
signiﬁcant role in making income distribution more equal. However, Ram (1984) with a
sample of 28 countries ﬁnd that higher level of schooling have a mild (insigniﬁcant)
equalizing effect as most studies have suggested. But larger education variance leads to
more equality in income distribution, which is contrary to most ﬁndings. Ram (1989)
with a different data set found that there is no signiﬁcant relationship between mean
schooling on income inequality for the full sample of 27 countries.

In recent years, studies of the linkage between the distribution of schooling and
income distribution at the individual level within a single country have gained
prominence. Knight and Sabot (1990) used establishment survey data of workers in
Kenya and Tanzania in 1980 to investigate the different approaches taken by the
respective governments to compress wages. They found that the educational expansion
adopted by Kenya was more effective in reducing the earnings inequality than Tanzania’s
pay equalization policy which is only effective in the public sector. This is reﬂected by

the pay premium of secondary education in Kenya which is about 20% lower than in

 

5 Schultz (1963), World Bank (1980, pp. 12-15) and Ram (1989) provide a discussion on the beneﬁts of
education.

55

Tanzania. It is also noted that in 1971, the premium of secondary education was
signiﬁcantly higher in Kenya than in Tanzania.

Lam and Levinson (1992) used household survey data from Brazil to study the
relationship of the distribution of schooling to earnings inequality. It was found that mean
level of schooling experienced steady increases and schooling inequality measured by the
coefﬁcient of variation declined signiﬁcantly. The variance in years of schooling
increased from the older cohorts 1925-27 and peaked with the 1949-51 cohort. For
subsequent younger cohorts, the variance in years of schooling declined. The cohort
effect is disequalizing for older age groups and equalizing for younger cohorts. However,
the overall increase in residual variance from 1976 to 1985 for both age groups and birth
cohorts was large enough to overcome the equalizing effects of declining variance in
years of schooling and declining returns to schooling. Brazil’s experience showed that
even substantial improvements in the schooling component of earnings inequality did not
guarantee overall declines in earnings inequality. But the beneﬁcial effect of lower
schooling inequality on overall earnings inequality is expected to be more evident as
post-1950 birth cohort become an increasing proportion of the labor force.

Lam (1999) used large household surveys from South Africa and Brazil to show
the important differences between the two countries in the link between the distribution
of education and the distribution of income. Lam applied a decomposition technique to
identify the contribution of the various factors that determine earnings inequality. An
interesting feature in his study is the use of estimated coefﬁcients of the earnings
regression and the distribution of characteristics to simulate counterfactual distribution of

earnings. The ﬁndings reveal that inequality of schooling is an important determinant of

56

income inequality in both countries, and plays a key role in the transmission of inequality
across generations.

Cameron (1998) used the National Socio-econornic Survey (SUSENAS) at the
household level to analyze the changes in income inequality in Java between 1984 and
1990. It was found that the increase in income inequality between 1984 and 1990 was
mainly due to the movement out of agricultural sector and to increases in mean incomes
in the traditionally better paid industries relative to agriculture. The increase in the
number of people with higher levels of education signiﬁcantly increased inequality but
the increased supply of better-educated individuals resulted in the ﬂattening of the
education-earnings proﬁle. As such, the net effect of increased education level was a
modest increase in inequality.

’ Levy and Mumane (1992) in their survey of studies on earnings inequality in the
United States summarized that earnings inequality was relatively stable in the 1970s but
has increased rapidly in the 1980s. For men, annual earnings inequality moved from
stability or gradual increases in the 19708 to rapid increases in the 1980s. For women,
annual earnings inequality moved from modest decline in the 1970s to increases in the
19803. For both men and women, increased earnings inequality was driven by the
increase in wage variation. The single most important change within the male earnings
distribution is the declining position of young, less educated men. For young males who
are not college graduates, the economy of the 1980s provided a much reduced
opportunity to earn a middle class income.

Juhn, Murphy and Pierce (1993) used data from the March Current Population

Survey and found that wage inequality remained stable or even declined slightly in the

57

1960s and then increased steadily through 1989. The trend toward greater wage
inequality is due to increases in the premia on both unobserved and observed dimensions
of skills such as education. In decomposing the level of wage inequality, it was found
that residual component appeared to be the most important for the overall increase in
inequality. In particular, returns to unobservable skills have shown a steady increase
since 1970.

It is noted that the above studies were in the context of rising income inequality in
both developing and developed countries. Malaysia presents an interesting alternative
case for analysis as earnings inequality has fallen over time. It would be useful for policy
makers to understand how development with equity is achieved.

The second key goal of this paper is to examine the linkage between parents’
schooling and children’s schooling. This is because intergenerational transmission of
schooling provides an important link between schooling inequality and earnings
inequality. The strength of the relationship of parents’ schooling and children’s schooling
would suggest the degree of importance of intergenerational transmission of earnings
inequality.

Dennis deTray (1988) used the ﬁrst wave of the Malaysian Family Life Survey
(MFLS) to study the differentials in school attendance among children who were 6-18
years old in 1976. The main ﬁnding of this study is that Malaysian families respond to
conditions in their household and communities when they decide whether or not to send
their children to school. Another ﬁnding is that government action has substituted
effectively for a shortfall of private resources among Malay families. Lillard and Willis

(1994) studied the relationship between parents’ education and children’s education

58

using the second wave of MFLS. They examined the effects of parental education on the
progress of their children through elementary, secondary, and post-secondary school. The
ﬁnding is that mothers’ and fathers’ education have positive and signiﬁcant effects on
their children’s educational attainment. The introduction of measured time-varying
economic, demographic, and environmental factors weakens the direct effects of parental
schooling, but does not weaken the correlation of unmeasured components between
parents and children. Lee (1998) analyzed the ethnic differences in fertility and child
schooling in Malaysia using data from the second wave of MFLS. He found that the
parents’ education is a less important determinant of child’s education after the NEP.
Previous studies have either used the ﬁrst or second waves of MFLS for analysis.
In contrast, this study is utilizing the data from both the ﬁrst and second waves of MFLS
for analysis. Speciﬁcally, the children sample is obtained by extracting the children from
the panel sample in MFLS] which is then pooled with the children of the New and Senior
Samples of MFLS2. The advantage of using the children of the panel sample from the
ﬁrst wave is that the problem of relatively higher attrition rate of Chinese can be avoided.
In terms of the regression speciﬁcation, the approach by Lee (1998) is the most
appropriate as it can be applied to establish the relationship between parents’ and
children’s schooling and also to test whether this relationship weakened after the
implementation of NEP and changes in educational policies. Another difference is that

this study focus on the differences in schooling between Malays and non-Malays.

59

3. Theoretical Linkages between Education and Earnings Distribution
Based on the human capital model of earnings by Mincer (1974) an earnings

generating function is expressed as follows:

logYs=logY0+Zlog(1+rj)+u, (1)

1:1
where Y, is the earnings of a person with S years of schooling, Yo with zero schooling, r,-
is the rate of return to the jth year of schooling and n represents the omitted variables.
Using the approximation log(1 + r) = r, equation (1) can be written as:

log Y, = log Y0 + rS + u, (2)
where r is the average private rate of return to investment in S years of schooling. Taking

variances on both sides of equation (2), the distribution of earnings can be written as:
Var(log 13) = Var(rS) = 72Var(S) + SzVar(r) + 2?SCov(r, S), (3)

where a bar over a variable indicates its mean. Holding other variables constant, the
model predicts a reduction in earnings inequality if schooling inequality (Var(S)) is
reduced. If the rate of return and schooling level are independent, an increase in the level
of schooling leads to greater earnings inequality. However, if the level of schooling and
rate of return are dependent, the covariance term can take on a negative or positive value.
If the covariance between the return to schooling and level of schooling is negative, then
an increase in schooling attainment shall reduce earnings inequality. As such, the effect
of an increase in schooling attainment on earnings inequality is an empirical question.
However, Lam (1999) pointed out that if schooling inequality is measured by a
mean-invariant measure of inequality such as coefﬁcient of variation, then it is possible

that a decrease in schooling inequality can be associated with an unchanged or even an

increase in earnings inequality.6 For instance, an increase in mean schooling with its
variance constant implies that the coefﬁcient of variation of schooling decreases but the
variance of log earnings remains unchanged. On the other hand, if the variance of
schooling increases by a smaller rate than mean schooling, it will also result in the
decrease in the coefﬁcient of variation of schooling but then earnings inequality
increases. This implies that there is no theoretical reason to expect a more equal
distribution of schooling should lead to a more equal earnings distribution. Therefore, the
effect of schooling inequality (coefﬁcient of variation of schooling) on earnings
inequality is also an empirical question.

Knight and Sabot (1990) explained the linkage between educational expansion
and earnings distribution through the composition effect and wage compression effect.
The composition effect tend to increase income inequality due to the increases in the
relative size of the group with more schooling and higher earnings. The wage
compression effect is due to the increase in supply of educated labor relative to demand,
which compresses wages and results in lower income inequality.7

The literature review reveals that the study of distribution of education on the
distribution of earnings has not been attempted on Malaysian setting. As such, this study

shall make a contribution by replicating Lam’s study using the Malaysian data set.

 

° Lam’s derivation is focussed only on relationship with the ﬁrst term on the right hand side of equation

(3).
7 See Psacharopoulos and Woodhall (1985, p.267) for a more detailed illustration of the mechanisms by
which educational invesment can affect income distribution.

61

4. Data
The analysis is based on the ﬁrst and second wave of the Malaysian Family Life
Survey (MFLS) data conducted by RAND and the National Population and Family
Development Board of Malaysia. The ﬁrst wave was carried out in 1977 and the second
wave during the period August 1988 through January 1989. The data on the schooling
distributions is based on the household roster information which contains basic
demographic and education information for each household member of those
interviewed. The sample for schooling distribution is constructed by pooling the Panel
sample in the ﬁrst wave and the New and Senior samples of the second wave. After
deleting those who were born before 1919 and after 1969, and those with missing data on

education, the sample includes 9788 respondents.

The data for the analysis of earnings inequality is obtained by merging the
household roster information with data on employment history in the ﬁrst wave. This is
to obtain earnings data for 1967-69 and 1976 which represent the period before NEP and
early post-NEP period respectively. The period 1967-69 is used so that the earnings data
of respondents can be captured at least once during this period. This is because earnings
data before 1976 is collected retrospectively once every 3 years or whenever there is a
job change. As for the second wave, the household roster information is merged with data
on current income activities which contain data on earnings in 1988 that is late post-NEP

period. After deleting respondents that are out of the speciﬁed age range (20-54)8,

 

8 The choice of the age range 20-54 is due to several reasons. First is the MFLSl panel sample is small and
sample size problems will arise particularly when stratiﬁed by cohorts or race if the age range 30-49 is
used as in Lam’s study. Second, the age range 20-54 is appropriate because it reduces problem of

62

females and those with missing data, the number of respondents for pre-NEP period is
827, early post-NEP period is 892 and late post-NEP is 4018.

The data analysis for intergenerational transmission of schooling is constructed by
pooling the children of Panel sample in the ﬁrst wave and the children of New and Senior
samples in the second wave. The sample of children aged 17 and above at the time of
survey comprises 1217 males and 826 females. For children age 20 and above, 947 are

males and 627 are females.

5. The Evolution of Schooling Distribution

The expansion of education in Malaysia has been impressive. Table 1 shows the
summary statistics for the schooling of Malays and non-Malays9 in Malaysia by ﬁve-year
birth cohorts. It provides an idea of the history of schooling from cohorts born in 1919 to
1968. The number of observations by ethnic group is shown in Columns 1-2. The mean
years of completed schooling are given in Columns 3-4. Overall, the average years of
schooling for Malays (6.06 years) is lower than non-Malays (6.25 years). However, it is
worthwhile to note that for older birth cohorts (1953 and below), the mean schooling for
non-Malays is higher than for Malays, while this trend is reversed for younger birth
cohorts (1954 and above). Consistent with the trends observed in Brazil, South Africa
and Java, education levels are rising over time in Malaysia, for both Malays and non-

Malays. However, the rate of increase in schooling is higher for Malays from a mean of

 

selectivity of schooling decision and the cut-off point of 54 is because the mandatory retirement age in
Malaysia is 55 years.

9 Non-Malays include Chinese, Indians and other races. The decision to stratify by Malays and non-Malays
is mainly because this classiﬁcation is of interest to policy makers who are interested in how the Malays
fare vis-a-vis the non-Malays. The other reason is that this classiﬁcation uses all possible observations in
the survey which alleviates to a certain extent the problem of small sample size in MFLS l.

63

1.74 (birth cohort 1919-1923) to 9.57 years of schooling (birth cohort 1964-68). For non-
Malays the mean years of schooling for those in birth cohort 1919-1923 is 2.97 and 8.82
for those in birth cohort 1964-68. These trends appear to be compatible to two main
factors: (i) the overall expansion of educational opportunities which increases the level of
education of all groups; and (ii) the narrowing of schooling inequalities between Malays
and non-Malays due to the Education Act in 1961 and the New Economic Policy in 1970
which provided conditions which encouraged the more rapid increase in schooling of
Malays relative to other races among younger cohorts.

Columns 5-6 of Table 1, show the standard deviation in years of schooling by
birth cohorts and ethnic group. The standard deviation of schooling of Malays by birth
cohort is observed to be lower than non-Malays. Columns 7-8 of Table 1 shows the
coefﬁcient of variation in years of schooling by birth cohorts and ethnic group. Overall, a
declining trend in the coefﬁcient of variation from the older cohorts to the younger
cohorts can be observed for both Malays and non-Malays. This suggest that schooling
inequality is decreasing over time.

The mean years of schooling for males and females by ethnic groups are plotted
in Figures 1 and 2 respectively. In analysing the trends in schooling in Malaysia it is
important to bear in mind that there are four major policy changes which can affect
schooling decisions among Malays and non-Malays. The ﬁrst three are related to the

Education Act, 196110. First is that the Malayan Secondary School Entrance Examination

 

1° One salient feature of the Education Act, 1961 is the policy of education expansion by providing free
primary education and automatic promotion from Standard One (Grade 1) to Form Three (Grade 9). The
other important feature is the conversion to Malay medium of instruction for secondary schools and higher
education institutions. The prime objective of using Malay as the national language is to achieve national
unity. See Soloman (1988) , Wong and Hong (1975) for a more detailed illustration of education system in
Malaysia.

(MSSEE) was abolished in 1964. It was replaced by the Lower Certiﬁcate of
Examination (LCE) which is public examination taken at Form 3 (Grade 9) to gain
promotion to upper secondary school. This meant that from 1965 onwards, students at
Grade 6 Were automatically promoted to secondary level up to Form 3. This policy
revision has the effect of increasing educational levels of all groups. Second is the criteria
is that a student needs to at least obtain a pass in Malay language to secure a pass in LCE.
A pass in LCE is a pre-requisite to progress to upper secondary school. The requirement
to pass the Malay language is expected to affect the non-Malays more as it is not their
mother tongue and also Malay language was not as widely used during that time. It is
expected to affect those who were born from 1950 (age 15 in 1965) onwards. Third is
the policy decision to convert all secondary school into Malay medium of instruction
which affected those born from 1961 onwards.11 Fourth is the speciﬁcation of racial
quotas for admission to universities which were favorable to the Malays. This decision
affected those born from 1953 onwards as they would have been in Form 5 in 1970. The
NEP, by limiting the number of places to higher education by ethnic groups affected the
student’s perception of their chances to gain admission to Universities. This was revealed
in the study by Wang (1980) who found that the NEP raised Malays’ educational
aspirations and their expectation of being successful in gaining places in local
Universities. As a result, he found that Malays were more likely than Chinese to continue
to Form Six (Grade 12). However, it should also be cautioned that even though the timing

of the above government policy changes are known, it is still hard to pinpoint when the

 

” The conversion to Malay medium of instruction was fully achieved in secondary schools by 1978 and at
the university level by 1983.

65

government policy actually take effect. It is even more difﬁcult to ascertain the timing as
there are implementation lags in policy interventions.

In the light of the education expansion policy coupled with the educational
policies that favor Malays, it is expected that mean schooling levels would increase for
the younger cohorts but the rate of increase in schooling for Malays should be higher
than for non-Malays. Based on Figure 1, which shows mean years of schooling of males
by ethnic group, it is obvious that both Malays and non-Malays have rising levels of
education over time. For cohorts born before 1953 it is noted that the mean level of
schooling of non-Malays is higher than Malays. But the trend is reversed for the birth
cohort after 1953 as the Malays obtain higher mean levels of schooling than non-Malays.

Figure 2 shows the mean years of schooling of females by ethnic group. Just like
the trend for females, both Malays and non-Malays receive rising levels of schooling
from the older cohorts to younger cohorts. For the older birth cohorts (before 1947) it
appears that the mean level of schooling Malays is lower than non-Malays. This trend is
reversed for the younger birth cohorts (1957 onwards) as the mean levels of schooling are
higher for Malays.

For the schooling of males and females, it is unclear where the timing of
structural breaks in the trends in mean schooling level occurs. Therefore, a test of
structural break shall be presented in the subsequent section to determine the timing of
the effect of government policies on schooling distribution among Malays and non-
Malays.

Figure 3 and 4 plots the standard deviation of years Of schooling which is one

measure of schooling inequality for males and females respectively. From Figure 3, it is

66

noted that Malay males in the birth cohort (before 1943 and after 1956) have lower
standard deviation of years schooling than non-Malay males. As for females, Figure 4
shows that for the birth cohort 1953 and below, Malays have lower standard deviation in
years schooling than non-Malays.

Figure 5 and 6 plots the coefﬁcient of variation of years of schooling of males and
females respectively. First it is noted that there is overall decline in schooling inequality
for both males and females in each ethnic group. Second the schooling inequality of male
Malays is slightly higher than for non-Malays males for the birth cohort 1947 and below.
For the birth cohort after 1954, it appears that the non-Malays males have a marginally
higher schooling inequality than Malay males. For Malay females the decline in
schooling inequality for the birth cohort before 1946 is more drastic compared to non-
Malay females. For the both Malay and non-Malay females in birth cohort 1948-1956
have similar level of schooling inequality. For those who were born after 1958, Malay
females have a slightly lower schooling inequality compared to non-Malay females.

Table 2 shows another way of analysing the evolution of schooling distribution
across cohorts. Columns 1 and 2 presents the mean years of schooling by schooling
decile for respondents age 55-59 and 25-29 in 1988 respectively. Column 3 indicates the
increases in mean years of schooling across cohorts by schooling decile. Similar to the
results in Brazil and South Africa, it is worthwhile to note that the increases in mean
level of schooling has been associated with the compression of the schooling distribution.
This is evident as the top deciles have smaller increases in mean schooling than deciles
around the middle of the distribution. The biggest improvements are in the 4‘h to 7tln

deciles of the schooling distribution. The mean, standard deviation, coefﬁcient of

67

variation and Gini coefﬁcient of years of schooling by age groups are presented in the
last four rows of Table 2. The mean years of schooling conﬁrms the rising levels of
education over time, while the coefﬁcient of variation and Gini coefﬁcients conﬁrm the
Observation that the schooling distributions have become more equal.

Figure 7 plots Lorenz curves for the schooling distributions of the age groups 25-
29 and 55-59 of the total population in 1988. The Lorenz curves conﬁrm the observation
above that the schooling distribution of the age group 25-29 is unambiguously more
equal than the age group 55-59. Figure 8 plots Lorenz curves for the schooling
distributions Of the age groups 25-29 and 55-59 in 1988 stratiﬁed by Malays and non-
Malays. It is of interest to note that among the old cohort (age 55-59) Malays appear to
have greater schooling inequality than non-Malays. The Gini coefﬁcients for Malays and
non-Malays (age 55-59) are .660 and .594 respectively. But the trend is reversed among
the young cohort (age 25-29) where non-Malays have greater schooling inequality than
Malays. The Gini coefﬁcients for Malays and non-Malays (age 25-29) are .182 and .228
respectively. On the whole, both Malays and non-Malays have substantial reductions in
schooling inequality over time. As the Lorenz curves are unaffected by the mean of the
distribution, they cannot be used to rank distributions in terms of social welfare.12
Therefore, the generalized Lorenz curves of schooling by age group and ethnic group is
presented in Figure 9. It is noted that among the older age group (55-59) non-Malays are
better off in terms of schooling distribution. But for the younger age group (25-59)

Malays are better off in terms of schooling distribution. Both Malays and non-Malays in

 

'2 See Cowell (1995) and Deaton (1997) which provides an excellent illustration and discussion on the
measures of inequality which includes the Lorenz curve and generalized Lorenz curve.

68

the younger age group have substantially higher social welfare ranking in schooling

compared to the older age group.

5.1. Tests for Parameter Stability

The analysis on the trends of distribution of schooling among Malays and non-
Malays in Section 5 above, do not provide any clear evidence of the effect of policy
changes on levels of schooling and schooling inequality. In order to test for a structural
break in mean schooling levels and schooling inequality of Malays and non-Malays, a
test of parameter stability proposed by Hansen (1999) is utilized to determine the timing
of the effect of government policies on schooling distribution among Malays and non-
Malays. The statistical procedure introduced by Hansen is useful because the test for
parameter stability makes no a priori assumptions about location of a break, or even if
one exists. The main difference is that instead of testing for a break at a particular point
in the time series, the test examines the entire time series for the location of a break point.
However, it is important to caution that even when a break can be ascertained, it can only
be indirectly attributed to the government policies as what has been found is a break in
the parameters of a regression relationship. It has not been proven that this break is
caused by government policy intervention. The Hansen test has the additional advantage
over the Andrews test because it is more general and more appropriate for this analysis.
This is because the Hansen test for structural change allows for non-stationary regressors
and heteroskedastic error process. Similar to the Andrews test, this technique allows for
only one break in the time series.

The regression relationship of interest is as follows:

69

Y = a + ﬂC + e,
Y equals mean years of schooling or measures of schooling inequality by birth cohort,
ethnic group and gender, C is birth cohort e is the error term. The above regression model
is applied separately to Malays and non-Malays by gender to test for constancy in B. The
structural change in null hypothesis is that the parameters of interest do not change

between periods (birth cohorts) are as follows:

ﬂ.={ﬂ’ i<t0
' 6+6, taro.

The parameter t0 6 [t1 , t2] indexes the relative timing of the structural shift, and 0 indexes

the magnitude of the shift. The test of Ho: 6 = 0 against H1 : 6 ¢ 0.

Table 3 reports the results, trimming 15 percent Off each end of the birth
cohorts”. Trimming parameter speciﬁes how far into the sample (as a percentage of the
full sample) one starts looking for a break, with a symmetric fraction of the sample left
after the latest break evaluated. Trimming is necessary to have sufﬁcient observations
before the earliest break and after the latest break to estimate the parameters. An F-test
for the null hypothesis of whether the estimated variances of the regressions in the two

relevant periods are equal is tested to determine whether to use homoskedastic bootstrap

or heteroskedastic bootstrap results. The F statistic = s,2 /522 ~ F n.1,, T24, where sf =

estimated variance of regression 1 and s22 = estimated variance of regression 2. For the

mean level of schooling, the maximum bootstrap value (24.067) of Malay males occurred

in the birth cohort 1954. As for non-Malay males, the maximum bootstrap value (40.32)

 

'3 The results were obtained by using the GAUSS program that implements the empirical techniques
introduced by Hansen. This program was downloaded from the website: www.ssc.wisc.edu/~bhansen.

70

occurred in the birth cohort 1935. As for Malay females, the maximum bootstrap value
(31.95) occurred in birth cohort 1938 and in 1934 non-Malays females (maximum
bootstrap, 40.489). All the above bootstrap values are signiﬁcant at less than 1%
signiﬁcance level with 15% trimming.

In terms of standard deviation of schooling, the break is signiﬁcant and it
occurred in the birth cohort 1947 Malay. For non-Malay males, the break is for birth
cohort 1947 but it is not signiﬁcant. As for females, the break happened in the birth
cohort 1953 for Malay females (signiﬁcant) and in the birth cohort 1939 for non-Malay
females (insigniﬁcant). With regard to the coefﬁcient of variation of schooling, the
structural break occurred in 1954 for Malay males, 1935 for non-Malay males, 1938 for
Malay females and 1934 for non-Malay females. The tests for breaks in the coefﬁcient of
variation by ethnic group and gender are all signiﬁcant. However, the above tests of
structural break provide different timing of breaks for Malays and non-Malays. In some
instances the timing of break do not occur during the policy intervention periods.
However, this does not mean that government policy does not have an effect on
schooling. Some evidence on the effect of government policy on schooling attainment

shall be discussed in Section 7.2.

6. Male Earnings Inequality
The distribution of schooling and monthly earnings among males age 20-54 in the
periods 1967-6914, 1976 and 1988 are shown in Tables 4, 5, and 6 respectively. Columns

1-3 show the frequency distribution of schooling. It is noted that 45.8% of the sample are

71

Malays and 54.2% are non-Malays in 1967-69. In 1976 the sample comprises 45.9%
Malays and 54.1% non-Malays. As for 1988 the sample of Malays and non-Malays are
50.5% and 49.5% respectively. It is also striking that the percentage with zero schooling
has declined over time. It is observed that those with zero schooling are very low for both
Malays (4.7%) and non-Malays (2.9%) in 1988. This is a remarkable achievement as the
percentage of population with no schooling in South Africa is 10.4% and Brazil is 14.9%
in 1995.

Columns 4-6 of Tables 4, 5 and 6 show mean monthly earnings15 that are
normalized to Malays without schooling. The mean do not include men with zero
earnings as the interview for current income was carried out only for those involved in
income earning activities.16 In 1967-69, Malay men with post-secondary education
received about 4.32 times higher than earnings of Malays with no schooling. But non-
Malay men earn much higher (10.8 times) than Malays with no schooling. In 1976,
Malay and non-Malay men with post-secondary education respectively earn about 6.73
and 8.99 times higher than Malays with no schooling. But in 1988, earnings gap for men
with post-secondary education and men without schooling appears to have narrowed to
about 3.98 times (Malay) and 5.69 times (non-Malays). From 1967-69 period to 1976, it
appears that the mean earnings relative to the mean earnings of Malays with zero
schooling declined for most of the schooling levels except for Malays in upper secondary

and post-secondary level and non-Malays in the upper secondary level. From 1976 to

 

“ The period 1967-69 captures only the latest real earnings of the individual if more than one observation
on earnings are available during this period.

15 Earnings here is deﬁned as monetary earnings that include wages and bonus.

'6 Although those who are unpaid family workers are interviewed, the number of cases recorded is only 7
and is thus not included in the computation of mean earnings.

72

1988, mean monthly earnings relative to Malays with zero schooling declined for all

schooling levels for both Malays and non-Malays.

6.1 Decomposition of Earnings Inequality

The regression results of log monthly earnings of all males aged 20-54 on ethnic
group, schooling, age and its square in 1967-69, 1976 and 1988 are presented in Table
7.17 The log variance of earnings which is a measure of earnings inequality is presented
in the third last row of Table 7. Overall, it can be seen that earnings inequality has been
declining over time.

The decomposition of log variance of earnings is based on the regression
estimates in Table 7. The decomposition of variance of log earnings is intended to
determine the extent of the contributions of age, race and schooling to earnings
inequality. The decomposition results of the simulations based on the respective time
period coefﬁcients and the characteristics during the same period are shown Table 8. The
ﬁrst simulation uses the coefﬁcients on age and its square only. It provides a baseline
level of inequality corresponding to the situation in which the coefﬁcients on race and
schooling are assumed to be zero and without residual variance. The second simulation
adds the effect of race. It represents the case when the partial effect of schooling is zero.
The contribution of age and race to earnings inequality would be 11.3% (1967-69), 6.8%
(1976) and 19.6% (1988) if the racial distribution and the effects of race on earnings

remained the same. When the residual variance of the respective period is added, the

 

‘7 The real earnings regression for 1967-69 include year dummies to control for year effects and number of
jobs, and the regressions for 1976 include number of jobs variable. Number of jobs are included because
the earnings data are based on earnings for all jobs which is due to the nature of earnings data collected in
MFLSl, while the earnings data in 1988 covers main job only.

73

simulated variance of log earnings equals 81.3% (1967-69), 69.4% (1976) and 85.2%
(1988).

The fourth simulation examines the effect of age and schooling on inequality
when the partial effect of race goes to zero. Consistent with the results in Brazil and
South Africa, schooling has a substantial effect on earnings inequality. Schooling
inequality explains 17.8% (1967-69), 28.7% (1976) and 28.7% (1988) of actual earnings
inequality. The addition of residual variance in row 5 explains about 87.7% (1967-69),
91.4% (1976) and 94.3% (1988) of the actual earnings inequality. The simulation of the
combined effect of age, race and schooling is shown in row 6. It accounts for about
29.6% (1967-69), 36.9% (1976) and 34.4% (1988) of total inequality. The residual
variance is added in Row 7 and it reproduces the total variance in log earnings for the

respective time periods.

6.2 Decomposition of Earnings Inequality by Ethnic Group

Separate regressions for Malay and non-Malay males (age 20-54 years) over the 3
time periods are presented in Tables 9 and 10 respectively. These estimated coefﬁcients
are used for the decomposition of log variance of earnings. The results of the simulations
for each of the time periods based on four simulations for each combination of
coefﬁcients are presented in Table 11. In the ﬁrst set of simulations, Malay coefﬁcients
(Table 9) are applied to the Malay sample. In the second set of simulations, non-Malay
coefﬁcients (Table 10) are applied to the Malay sample. For the third set of simulations,
non-Malay coefﬁcients are applied on non-Malay sample. The ﬁnal set of simulations

involves the use of Malay coefﬁcients on non-Malay sample. The simulation in row 3 of

74

Panel A, B, and C is of interest as it measures the contribution of schooling to earnings
inequality. For the within Malay simulations, it is observed that the contribution of
schooling increases by a signiﬁcant percentage from 19.1% (1967-69) to 32.3% (1976)
and remained stable at 32.6% in 1988. The within non-Malay simulation shows that the
contribution of schooling also increases by a signiﬁcant percentage from 23.6% (1967-
69) to 31.3% (1976) and declined slightly to 29.4% in 1988. The increase in the
contribution of schooling in explaining earnings inequality between the pre and post-NEP
period is greater for the Malays relative to non-Malays. It is noted that the simulation of
non-Malay schooling coefﬁcients on the Malay sample accounts for a lower contribution
of earnings inequality. This means that if Malays had the same effects of schooling on
earnings as non-Malays, Malay’s earnings inequality would be lower. Other the other
hand, the simulation of Malay schooling coefﬁcients on non-Malay characteristic
explains a higher percentage of the schooling inequality. This implies that if non-Malays
had the same effects of schooling on earnings, non-Malay’s earnings inequality would be
much greater.

It is noteworthy that the earnings inequality (variance of log earnings) of Malays
and non-Malays declined respectively by 29.7% from 1967-69 to 1976. It further
declined by 11.68% from 1976 to 1988. As for non-Malays earnings inequality also
declined but by a slower margin that is from 1967-69 to 1976, it recorded a decline of
4.45%. From 1976 to 1988, the decline in earnings inequality for non-Malays is 9.3%.
The faster rate in decline in earnings inequality for Malays may be attributed to the
relatively higher increase in level of education accompanied by earnings compression at

the secondary and lower levels of education. It is also driven by the faster decline in

75

schooling inequality of Malays over time. These ﬁndings appears to be compatible with
most of the empirical ﬁndings by other studies and is also consistent with the theoretical
linkages between distribution of schooling and earnings inequality.

But the interesting point to highlight is that the unobserved component measured
by the residual variance seems to be the biggest factor (56-78%) in explaining earnings
inequality. Another interesting ﬁnding is that the residual variance is declining by a
substantial margin for Malays from .696 (1967-69) to .417 (1976) and .372 (1988). For
non-Malays the decline in residual variance is small over time from .455 (1967-69) to
.399 (1976) and .365 (1988). The big drop in residual variance for Malays from 1967-69
to 1976 is noteworthy. It could be due to three possible reasons. First is that it may be
attributed to the implementation of NEP in 1970 which brought about the relatively
higher contribution of schooling towards explaining earnings inequality and thus
shrinking the unobserved component of earnings inequality. The second possibility of
higher residual variance is that the distribution of occupations which is a important policy
variable for the NEP may account for a substantial portion of the residual variance. As
such, occupation dummy variables were added to the next set of regressions to determine
the role of occupation in explaining earnings inequality. The third possible reason could
be due to measurement error in the collection of retrospective earnings. This is because
earnings in 1967-69 are based on retrospective data which is subject to recall error. On
the other hand, the earnings in 1976 and 1988 are based on current earnings and are
therefore subject to less measurement error. The relatively higher residual variance for

Malays in 1967-69 could be due to the Malays with less education reporting their

76

retrospective earnings less accurately. Therefore the measurement error in earnings may
be translated into higher residual variance in earnings of Malays.
6.2.1 Measurement Error in Earnings

In order to address the question Of measurement error in earnings, the data from
the panel sample Of MFLSl and MFLSZ are used. By doing so, two sets of earnings data
can be obtained by matching the earnings from the same individual, occupation and year
that were collected during the ﬁrst and second waves Of the survey. For example, a
respondent is interviewed in 1976 (MFLSl) about his earnings and occupation in 1973.
The same respondent is interviewed again in 1988 and information about his earnings and
occupation in 1973 is collected assuming that there is a job change. However, for
comparability and consistency in earnings data between MFLSl and MFLS2, only those
having one job are included in the sample. This is because Of the different nature Of

earnings data collected in the two waves'8 of MFLS. The sample that can be matched are

389 Observations which is about 15% of total possible matches.

The key question is whether Malays have relatively higher measurement error

variance when recalling past earnings compared to non-Malays. The dependent variable
is the difference in reported log earnings square which equals (1n Y"88 — ln 11,76 )2 where

lnY is monthly earnings in natural log, superscript 76 and 88 indicates the survey year
and subscript i and t represents individual i and time t which is the year that the earnings

data are being recalled. The difference in reported log earnings square is regressed with

 

'8 MFLSl earnings data were based aggregated earnings from all jobs if the respondent has more than one
job. While earnings data from main and secondary job were collected separately in MFLSZ. Since, MFLSZ
earnings data were recorded only when there is a job change it is not possible to Obtain total earnings for
all jobs (without interpolation) when the starting or ending date of main and secondary jobs differ.

77

the number Of retrospective years from 1976, non-Malay dummy variable and its
interaction with number of retrospective years from 1976 in Model 1. From this point
onwards, number Of retrospective year refers tO number Of retrospective years from 1976.
Model 2 is based on the number of retrospective years and its square, non-Malay dummy
variable and the interactions between the quadratic speciﬁcation of number of
retrospective years and non-Malay dummy variable. The regression results of Model 1
and 2 are shown in Table A5 Of Appendix A.

The difference in reported log earnings square is considered as a measure Of the
variance of measurement error in earnings. Based on Model 1, it is interesting to note that
the estimated coefﬁcient for the number of retrospective years is positive and signiﬁcant.
This implies that the variance of measurement error is increasing and signiﬁcant for both
Malays and non-Malays. This ﬁnding is consistent with the hypothesis that the variance
Of measurement error is positively associated with the number Of years Of recall. It is also
interesting to note that the number Of retrospective years interacted with non-Malay
dummy variable is negative but insigniﬁcant. Based on the usual standard of inference, it
is Observed that there is no signiﬁcant difference in the rate of forgetting between Malays
and non-Malays. However, from the view of point estimates, the rate of forgetting of
Malays is faster than non-Malays. It is noted that the magnitude Of the non-Malay and
number of retrospective years interaction term (-0.37) is considered large relative to the
estimated coefﬁcient of number of retrospective years (.042). At the mean number of
retrospective years for non-Malays, the effect Of the non-Malay interaction term

estimates reduces the measurement error variance Of non-Malays by .55519 compared to

 

‘9 Please see Appendix A for details.

78

Malays. This magnitude is substantial bearing in mind that the mean variance Of
measurement error is between .7423 (Malays) and .8865 (non-Malays). However, caution
needs tO be exercised when interpreting the point estimates because the estimated
coefﬁcient is extremely imprecise. The imprecise estimate could be due to the small
sample size which is unable to detect the subtle differences in measurement error
between Malays and non-Malays.

Based on the quadratic specification for the number Of retrospective years in
Model 2, it is noted that the variance of measurement error decreases initially for both
Malays and non-Malays. Subsequently, the variance of measurement error increases after
10 years of recall for Malays and for non-Malays it increases after 18 years recall. This
ﬁnding is counter-intuitive and I do not have a reasonable explanation for this result.

TO sum up, the results based on the linear speciﬁcation Of number Of retrospective
earnings in Model 1 appears to be more reasonable and is preferred over Model 2. Model
1 Shows that the measurement error variance increases with the number of retrospective
years for both Malays and non-Malays. However, with regard to differences in the rate of
increase in variance Of measurement error by number of years of recall (rate Of
forgetting) between Malays and non-Malays, Model 1 offers two contrasting conclusions
depending on which point Of view is being considered. Based on the statistical point of
view, the rate of forgetting between Malays and non-Malays is not a major factor in
explaining the high residual variance in log earnings of Malays during the 1967-69
period. Conversely, if the results are examined from the view Of point estimates, then the

difference in rate of forgetting between Malays and non-Malays is substantial. But the

79

drawback iS that the point estimates are extremely imprecise. The details of the analysis
of measurement error in earnings are elaborated in Appendix A.
6.2.2 Effect of Occupation on Earnings Inequality and Residual Variance

This section attempts to examine the impact of changes in occupation
distributions over time on earnings inequality and also on the residual variance Of log
earnings. The regression estimates of earnings which include occupation variables over
time for Malays and non-Malays are shown in Table 12 and 13 respectively. When
occupation dummy variables are added to the regression the coefﬁcient estimates of
returns to schooling remain signiﬁcant but decline substantially for both Malays and non-
Malays. The reference group for occupation dummies is the agricultural workers, farmers
and ﬁshermen. Among Malays, laborers appear to have the lowest earning capacity in
1967-69 followed by those respondents engaged in the agricultural sector (although the
coefﬁcients are not signiﬁcantly different. In 1988, agricultural occupations have on
average the lowest earnings among the various occupational groups. Among non-Malays,
occupations in the agricultural sector also have the lowest earnings on average in 1967-
69 and 1976 periods. In 1988 non-Malay laborers have lower earnings (not signiﬁcant)
than non-Malay respondents involved in agricultural occupations. For both Malays and
non-Malays, occupations such as managers and professionals, clerical, sales, service,
production related occupations, transport related occupation all have signiﬁcantly higher
earnings than agricultural occupations for all the three time periods. It is also noted that
there is a compression of earnings among occupations over time for both Malays and

non-Malays.

80

The decomposition Of earnings inequality taking into account the occupation
variables are presented in Table 14. With the addition Of occupation dummies, the
contribution of schooling is lower but still remained substantial. For the within Malay
simulations, the contribution of schooling in explaining earnings inequality increased
over time from 5.4% in 1967-69 to 10.6% in 1976 and 15.6% in 1988. A similar trend is
also Observed for non-Malays as the contribution of schooling in explaining earnings
inequality also increased over time from 7.73% in 1967-69, 9.84 in 1976 and 18.2% in
1988. The addition of occupation variable to schooling and age increases the explanatory
power in explaining income inequality by a signiﬁcant margin. For Malays, age,
schooling and occupation accounts for 32.2% of earnings inequality in 1967-69, 44.9% in
1976 and 42.5% in 1988. As for non-Malays, age, schooling and occupation explains
36.7% of earnings inequality in 1967-69, 42.7% in 1976 and 37.7% in 1988. These
results imply that besides schooling, occupation variables are important in accounting for
the variations in earnings inequality. The effect of the occupation factor is driven by the
movement Of agricultural workers to higher paying jobs. In addition, the compression of
earnings by occupation over time also contributed towards the reduction in earnings
inequality.

The application Of non-Malay coefficients on Malay sample results in a lower
contribution towards explaining earnings inequality of Malays. However, when Malay
coefﬁcients are applied to the non-Malay sample the Opposite effect occurs that is it
results in a higher contribution towards the explanation earnings inequality Of non-

Malays. This implies if the effect of schooling and occupation were switched between

81

Malays and non-Malays, we would expect to see lower earnings inequality among
Malays and higher earnings inequality among non-Malays.

It is of interest to note that even with the addition of occupation variables, the
unobserved effect on earnings inequality remained high. However, the residual variance
Of Malays declined by a large margin from .587 (1967-69) to .344 (1976) and .318
(1988). As for non-Malays, the decline in residual variance is more gradual that is from
.377 (1967-69), .332 (1976) and .322 (1988). The addition Of occupation variable is still
unable to explain the exceptionally high residual variance Of Malays during the 1967-69
period. The separate regression analysis on monthly real earnings of younger cohort
Malays (age 20-34) and older cohort (age 35-54) were carried out. The estimated results
are presented in Tables Bl and C1 of the Appendix. It is interesting to note that the
residual variance of log monthly earnings of both the young and old cohort Malays
during the 1967-69 period are also unusually high.

In short, the above ﬁndings seem to suggest that the NEP has a signiﬁcant role in

the reduction Of residual error variance in log earnings Of Malays in 1976.

6.3 Social Welfare Ranking of Earnings Distributions

The above results indicate that earnings inequality have fallen over time which
appears to suggest that the government policies in Malaysia have been effective to some
extent in redistributing the beneﬁts of economic development more equitably. Next, it
would be interesting to rank the earnings distribution by ethnic group over time by using
the generalized Lorenz curve. The horizontal axis for the generalized Lorenz curve is the

cumulative fraction of the population and the vertical axis shows the cumulative Share Of

82

earnings multiplied by mean earnings. The generalized Lorenz curve is useful to compare
different distributions with different means and different aggregates. If the generalized
Lorenz curve in one period (group) lies above another period (group), it implies that for
all p from 0 to 100 percent of population, the poorest p percent of the population in the
higher generalized Lorenz curve will have more resources in total. Therefore, the higher
generalized Lorenz curve will be preferred by any equity respecting social welfare
function. In order to make comparisons over time valid, the real earnings from all jobs
are computed using the Consumer’s Price Index with the base year being 1980. The
generalized Lorenz curves corresponding to Malays and non-Malays over two time
periods, 1967-69 and 1988 are presented in Figure 10. Comparing over the two
respective time periods, it is Observed that the generalized Lorenz curves of non-Malays
dominates those of Malays. Therefore, by any equity-preferring social welfare function,
the distribution of real earnings of non-Malays is preferred over Malays in each Of the
respective time period. But it is noteworthy the gap between the generalized Lorenz
curves between Malays and non-Malays are narrowing over time. This implies that the
government policies have made some progress tO improve the position of Malays relative
to non-Malays. But it also indicates that real earnings of Malays are still lagging behind
non-Malays and that the policies to uplift the economic position Of Malay may still be

needed for years to come.

7. Intergenerational Transmission of Schooling

The results above highlight the importance of schooling in explaining the

distribution of earnings in Malaysia. Since parents’ schooling is an important factor in

83

determining children’s schooling, the intergenerational transmission of schooling
provides an important link between schooling inequality and earnings inequality.
Therefore, the strength of the relationship of parents’ schooling and children’s schooling
over time is interesting because the implementation of educational policies and NEP
would affect schooling decisions. The lower degree of intergenerational transmission Of

schooling inequality has the potential to lower earnings inequality in the long run.

7.1 Regression Specification and Variables
In order to study the relationship of parents’ and children’s schooling, the basic
regression speciﬁcation is as follows:
S=PB+Cy+H5+e
where S is that cumulative years of schooling of child, P is parents’ schooling
variables, C is child characteristics such as birth cohorts and ethnic group, H is a set Of
household and community variables. The summary statistics of variables used in this
analysis are shown in Table 15. The dependent variables are the number Of years
schooling completed Of children up to age 17 and 20.20 The number of years schooling up
to age 17 provide information on whether the child completed lower secondary education
and entered upper secondary education, while number Of years schooling up to age 20

indicates whether the child completed upper secondary education and entered post-

 

20 . . . . . . .
A set of regressrons based on years of schooling per year of age srnce age srx yielded srmrlar
results. Since the number of years schooling is easier to interpret, the regressions using this dependent
variable is preferred and presented in this study.

84

secondary education.21 Ethnicity is classiﬁed into Malay and non-Malay. Household
variables are based place Of residence when the child is at age 12. The variables used are
dummy variables for urban residence and availability Of pipe water which is measured at
the household level. This availability of pipe water attempts to capture the effect of
different household conditions on schooling decisions. Community variables are
measures Of schooling availability which are constructed by matching the migration
history data to the community data set. The migration history data provides information
on the date Of move and location which is identiﬁed by district at each move. The
community data contain data on the history of school availability in 398 Enumeration
Blocks (EBs) and 52 Primary Sampling Units (PSUS).22 Unfortunately, while we only
know the child’s place of residence at the district level, retrospective migration data at
the EB or PSU level is not available.23 With this limitation, the proportion of sampled
EBS and PSUS within a district that reported having a school Of a given type open at a
particular point in time is calculated. The availability of schools measures are obtained
for each child at the relevant time periods. For instance, the availability of primary
schools is measured at the time when the child is at age 6 and availability Of secondary

school is measured when the child is 12 years old.

 

2' The schooling system in Malaysia is similar to the British system. The primary level of education
consists of 6 years of schooling. This is followed by lower secondary level (3 years) and upper secondary
level (2 years). Post-secondary education is 2 years at Form Six and 3-4 years at college or university level.
22 The EBs represent communities from which the New and Senior samples resided in at the time of
survey. The PSUS represent communities from which the Panel sample resided in at time of survey.

23 Sevaral EBs or PSUS would make up one district.

85

7.2 Empirical Results

The regression estimates Of child schooling on parents’ schooling and other
controls for male and female children aged 17 and above are presented in Table 16 and
17 respectively. The schooling regression estimates for male and female children aged 20

and above are shown in Table 18 and 19.

7.2.1 Parents’ Schooling

The regression estimates of years of schooling Of sons and daughters aged 17 and
20 or more indicate that mother and fathers’ schooling have a signiﬁcantly positive effect
on their children’s schooling as shown in column 2 of Tables 16-19. The effect of
parents’ schooling on children’s schooling attainment remained positive and signiﬁcant
even with the inclusion Of controls of place of residence and school availability measures.
This implies that parental schooling plays an important role in the intergenerational
transmission of human capital.

The government has pursued a policy Of general development through the
expansion Of education system. At the same time, in the pursuit of development with
equity, a set of education and employment policies that are favorable to the
disadvantaged group have been implemented. The effect Of the NEP and education
policies on intergenerational educational mobility is of great interest. In order to test for

such an effect, father and mothers’ level of schooling are interacted with children born

86

from 1960 onwards. The reason is that cohorts born from 1960 onwards were exposed to
the educational policy changes as well as the N EP.24

It is Observed that the estimated coefﬁcients Of parents’ schooling interacted with
the 1960 cohort variable are mostly negative and jointly signiﬁcant for the sample of sons
and daughters aged 17 and above. However, in the sample Of sons and daughters aged 20
and above the interaction terms of parents’ schooling and the 1960 cohort variable are not
signiﬁcant although the estimated coefﬁcients are mainly negative. In general, these
results suggest that the effect Of parents’ schooling weakened after the implementation of
the NEP and educational policies. This further implies that there are greater schooling
opportunities for the Offspring Of less educated parents. From the policy perspective, the
reduction in the effect of parents’ schooling on children’s schooling is desired as it has
the potential to promote greater educational equality and reduce earnings inequality in

the long run.

7.2.2 Birth Cohorts, Ethnicity and Government Policies

Birth cohorts are included in the regression speciﬁcation to pick up the effects of
shifts in government policies and changes in the economy. Three-year birth cohort
dummies are included with the left-out category being those born before 1957. The
regression estimates of birth cohort in column 1 Of Table 16-19 are signiﬁcantly positive.
It also shows an increasingly positive trend for younger cohorts which mean that there is
a rising level Of education for both males and females. The inclusion Of parents’

schooling in column 2 and adding Of cohort interaction terms, schooling availability and

 

2" It is noted that the Hansen tests for structural break fail to produce any clear indication of the timing of

87

household background variables do not affect the signiﬁcant positive trend Of the birth
cohort estimates.

The interaction of non-Malay variable with 1960 birth cohort is intended to
capture the differential rate of increase in schooling Of Malays and non-Malays who are
affected by both the NEP and changes in education policies. For all the three regression
speciﬁcations, the estimates of this interaction term are signiﬁcantly negative for both
sons and daughters age 17 and above. However, for sons and daughters aged 20 and
above, this estimated coefﬁcient of this interaction term is signiﬁcantly negative except

for the speciﬁcation that includes community and place of residence variables.

7.2.3 School Availability

School availability is considered an important variable in the schooling attainment
equation because it is related to the costs of schooling. The availability of school in the
local community lowers the costs of schooling and is expected to have a positive effect
on schooling attainment. At the primary school level, parents have the choice to send
their children tO schools with Malay, Chinese or Tamil medium of instruction. Therefore,
at the primary level, the medium Of instruction is an important consideration. Since all
races attend Malay medium primary schools it is interacted with both Malays and non-
Malays. While Chinese are more likely to attend Chinese schools, and Indians more
likely to attend Tamil schools, it is interacted with non-Malay dummy variable. At the
secondary level, the language of instruction is not important especially for the younger

cohorts as Malay language is the only medium Of instruction for national secondary

 

government policies on level of schooling and schooling inequality for Malays and non-Malays.

88

schools. As such, the availability Of secondary school variable is introduced without
interacting with the language Of instruction.

The estimated coefﬁcients Of school availability variables are shown in column 3
of Tables 16-19. On the whole, the estimated coefﬁcients have the expected signs for
both the male and female children but are mostly insigniﬁcant. For instance, coefﬁcients
for Malay interacted with Malay medium primary are positive and non-Malay interacted
with Malay medium schools are negative. Non-Malays interacted with Chinese or Tamil
schools result in positive coefﬁcients. Had these estimated coefﬁcients been statistically
signiﬁcant, it would suggest that Malays have the advantage if the medium of instruction
is Malay. It also suggest that availability of Chinese or Tamil primary schools increases
schooling attainment Of non-Malay children. As for the availability Of secondary schools
in the district, a positive coefﬁcient is Observed for both male and female children.

However it is only signiﬁcant for male children aged 20 and above.

7.2.4 Place of Residence

The place Of residence is intended to capture some features of the broader
environment in which schooling decisions are made. After controlling for school
availability and family background characteristics, urban place of residence has positive
but insigniﬁcant coefﬁcients in both male and female children schooling regression. The
availability of piped water is a proxy measure of household assets and the socio-

economic position of families at the relevant time of the schooling decision.25 The

 

25 Data on household assets are only available at the current time of survey. The use of income of earnings
as a time-varying measure is problematic because the retrospective earnings data for MFLS2 is only
recorded when there are job changes. As a result very few matches of earnings data to the period of
schooling decision can be Obtained.

89

coefﬁcients of availability of piped water at the household level are generally positive
and signiﬁcant except for female children age 17 and above which is positive but not

signiﬁcant.

8. Conclusion

The trend analysis Of mean schooling by birth cohorts Show that the levels of
education are increasing from the Older to younger cohorts for both sexes Of Malays and
non-Malays. Schooling inequality measured by coefﬁcient of variation, Gini coefﬁcient
and Lorenz curves indicate that schooling inequality has been decreasing from the older
to younger cohorts. A trend of declining earnings inequality for both Malays and non-
Malays has been Observed. The decomposition of earnings inequality reveal that
schooling and occupation are important variables in explaining earnings inequality for
both Malays and non-Malays. The contribution Of the schooling factor is due to rising
level of education and the compression of returns to education especially at the secondary
and lower level of schooling. The decrease in schooling inequality also contribute to
lower earnings inequality. The contribution of occupation towards lower earnings
inequality Operates through two major sources. First is the movement out of agriculture to
better paying jobs in the modern sector and second is the compression of earnings by
occupation over time. Although the residual variance in log earnings are declining over
time, the biggest contribution towards explaining earnings inequality are still the
unobserved factors. Of particular interest is the unusually high residual variance in log
earnings Of Malays during the pre-NEP period, and the sharp drop in residual variance

during the early post-NEP period. Even with the addition of occupation variables this

90

phenomenon remains. The difference in the variance of measurement error in earnings by
years of recall into the past between Malays and non-Malays is not a major factor in
explaining the high residual variance. Hence, it is plausible that the effect Of NEP
contributed towards the substantial shrinking Of the residual variance during the post-
NEP period.

The generalized Lorenz curves Of real earnings Of Malays and non-Malays indicate
that non-Malays are better Off than Malays during the pre-NEP and post-NEP periods. It
also suggests that the NEP has played a role in improving the position Of Malays relative
to non—Malays. But the Malays still lag behind in terms Of social welfare ranking of real
earnings distribution. These results point to the need to maintain the policies intended to
uplift the economic position of the disadvantaged group.

The Hansen tests of parameter stability on the timing Of the effect of government
policy on mean schooling and schooling inequality produce different results for Malays
and non-Malays. It is unable to provide a clear picture on the timing of the effect of
government policy on schooling. However, using cohort analysis, the results are
compatible with the notion that education policies and NEP do affect schooling
attainment of Malays and non-Malays. It is observed from the regression analysis that
non-Malay children born after 1960 do have Signiﬁcantly lower schooling attainment
than Malays.

Of great relevance to this study is that the link between parents’ schooling and
children’s schooling weakened signiﬁcantly for cohorts born after 1960. It implies that
the educational policies and NEP policies are quite effective in providing better

educational opportunities for children with parents with lower education. The weaker

91

intergenerational transmission of schooling tends to reduce schooling inequality. The
greater equality Of education has potential beneﬁts in the future as greater equality in

human capital is associated with greater earnings equality.

 

92

References

Adelman and Morris (1973). Economic Growth and Social Equity in Developing
Countries, Sanford University Press, Stanford.

Andrews, D.W.K. (1993). Tests for Parameter Instability and Structural Change with
Unknown Change Point, Econometrica, V0. 61, no.4:821-856.

Cameron, L. (1998). Income Inequality in Java: Relating the Increases to the Changing
Age, Educational and Industrial Structure, University Of Melbourne, mimeo.

Chenery H and Syrquim M. (1975). Patterns of Development, 1950-70, Oxford
University Press, New York.

Chiswick, BR. (1971). Earnings inequality and economic development, Quarterly
Journal of Economics, 85: 21-39.

Cowell, FA. (1995). Measuring Inequality, Prentice Hall, Hertfordshire.

Deaton, A. (1997). The Analysis of Household Surveys: A Microeconometric Approach to
Development Policy, Johns Hopkins Press, Baltimore.

de Tray, D. (1988). Government policy, household behavior and the distribution of
schooling Of Malaysia: a case study of Malaysia, in T.P. Schultz (ed), Research in
Population Economics, Volume 6, JAI Press, Greenwich, CT.

Government Of Malaysia (1979). Mid-Term Review of the Third Malaysia Plan - 1976-
1980, National Printing Department, K.L.

(1986). Fifth Malaysia Plan - 1986 -1990, National Printing Department,

 

K.L.

(1988). Economic Report 1988/89, National Printing Department, K.L.

 

Gregorio JD. and Lee J. W. (1999). Education and income distribution: new evidence
from cross-country data, Development Discussion Paper, No. 714, Harvard Institute for
International Development, MA.

Hansen, BE. (1999). Testing for Structural Change in Conditional Models, Journal of
Econometrics, forthcoming.

Juhn, G, Murphy, K. M. And Pierce, B. (1993). Wage inequality and the rise in returns to
skill, Journal of Political Economy, 101(3):410-442.

93

Knight J. B. and Sabot R. H. (1990). Educational Productivity, and Inequality: The East
African Natural Experiment, Oxford University Press, New York.

Lam, D. and Levinson, D. (1992). Declining inequality in schooling in Brazil and its
effects on inequality in earnings, Journal of Development Economics, 37: 199-255.

Lam, D. (1999). Generating Extreme Inequality: Schooling, Earnings, and
Intergenerational Transmission of Human Capital in South Africa and Brazil, University
Of Michigan, mimeo.

Lee, SH. (1998) Three essays on government inﬂuence on labor markets, Ph.D.
Dissertation, Michigan State University.

Levy, F. and Mumane, R]. (1992). US. earnings levels and earnings inequality: A
review of recent trends and proposed explanations, Journal of Economic Literature,

30(3):1333-138l.

Lillard, L and Willis, R. (1994). Intergenerational educational mobility: effects of family
and state in Malaysia, Journal of Human Resources, 29(4):]126-1166.

Mahatir, Mohamed (1998). The Way Forward, Weidenfeld and Nicholson, London.

Marin, A and Psacharopoulos, G. (1976). Schooling and income distribution, Review of
Economics and Statistics, 58: 332-338.

Mincer, J. (1974). Schooling, Experience, and Earnings, Columbia University Press,
New York.

Park, K. H.(1996). Educational expansion and educational inequality on income
distribution. Economics of Education Review, 15(1): 51- 58.

Peterson, C, et al. (1993). The Second Malaysian Family Life Survey: User’s Guide,
RAND, Santa Monica, CA.

Piehl, A. M. et a1 (1999). Testing for Structural Breaks in the Evaluation Of Programs,
NBER Working Paper No. 7226.

Pong S. L. (1993). Preferential policies and secondary school attainment in Peninsula
Malaysia, Sociology of Education, 66: 245-261.

Psacharopoulos, G. and Woodhall, M. (1985). Education for Development. Oxford
University Press, New York.

Ram, R (1984). Population increase, economic growth, educational inequality, and

income distribution: some recent evidence. Journal of Development Economics, 14: 419-
428.

94

 

(1989). Can educational expansion reduce income inequality in Less-
Developed Countries? Economics of Education Review, 8(2): 185-195.

 

Soloman, J.S. (1988). The Development of Bilingual Education in Malaysia, Selangor,
Pelanduk Publications.

Strauss, J. and D. Thomas, (1995). Human resources: empirical modeling Of household
and family decisions, in TN. Srinivasan and J. Behrman (eds.), Handbook of

Development Economics 3A, Amsterdam: North Holland Press.

Wang, B. L. (1980). Sex and ethnic differences in educational investments in Malaysia,
Comparative Education Review, 24: 140-59.

Wong, H.K. and Hong, ET. (1975). Education in Malaysia, Heinemann Educational
Books, Kuala Lumpur.

Wooldridge, J.M. (1999). Introductory Econometrics: A modern approach, South-
Westem College, Cincinnati.

95

Table 1. Summary Statistics for Distribution of Schooling
by Five Year Birth Cohort

 

 

 

 

 

 

Number of Years of Completed Schooling
Observations
Mean Standard Coefﬁcient of
Birth Deviation Variation
Cohort Malays Non- Malays Non- Malays Non- Malays Non-
Malays Malays Malays Malays
(1 ) (2) (3) (4) (5) (6) (7) (8)
19-23 165 203 1.74 2.97 2.55 3.75 1.47 1.26
24-28 266 277 2.22 3.50 2.54 3.62 1.14 1.03
29-33 367 464 2.32 3.21 3.10 3.64 1.33 1.14
34-38 460 490 2.94 4.29 3.07 3.89 1.04 0.91
39-43 443 556 4.58 5.23 3.52 4.18 0.77 0.80
44-48 487 650 5.56 6.38 3.64 3.86 0.65 0.61
49-53 655 804 6.93 7.12 3.41 3.84 0.49 0.54
54-58 576 655 8.10 7.60 3.67 3.70 0.45 0.49
59-63 593 590 9.37 8.28 3.31 3.45 0.35 0.42
64-68 496 591 9.57 8.82 2.93 3.25 0.31 0.37
Total 4508 5280 6.06 6.25 4.26 4.19 0.70 0.67

 

 

 

 

 

 

 

 

 

96

 

Table 2. Mean Years of Schooling by Schooling Decile
(All persons age 25-29 and 55-59 in 1988)

 

 

 

 

Age 55-59 Age 25-29 Increase
Schooling Decile (1) (2) (3)

l 0 1.93 1.93

2 0 5.77 5.77

3 0 6.03 6.03

4 0 8.40 8.40

5 0.66 9.00 8.34

6 2.39 9.53 7.14

7 3.66 11.00 7.34

8 4.97 11.00 6.03

9 6.01 11.05 5.04

10 10.39 14.45 4.06
Total:

No. of. obs 831 1183 2014

Mean 2.82 8.82 6.00

Std. Dev. 3.44 3.43 -0.01

C.V. 1.22 0.34 -0.88

Gini 0.62 0.21 -0.41

 

 

 

 

97

 

Table 3. Parameter Instability in Years of Completed Schooling:
Breaks in Mean, Standard Deviation and Coefﬁcient of Variation

 

 

 

 

Years of schooling Malay Non-Malay
Max. Test Birth Cohort Max. Test Birth Cohort
Statistic Of Max. Statistic Of Max.
A. Males
Mean 24.067 1954 40.32 1935
p-value: test in breaks (.0001) (.0000)
p-value; H0 3 0'12 : 0'; .2187 .3129
Standard Deviation 23.294H 1947 7.076 H 1944
p-value: test in breaks2 ((3)803; (€370)
p-value: H0 : 0'I = 0'2
Coefﬁcient of Variation 38.007 1954 80.71 1935
p-value: test in breaks (.0150) H (.0000) H
p-value: I-l0 : 0',2 = 0‘,2 .0000 .0000
B. Females
Mean 31.950 1945 40.489 1934
p-value: test in breaks (0000) (0000)
p-value: H0 : 0',2 = 022 2539 .3779
Sfjgfjgf‘tgsfvifgfehs 41.768 1953 14.055 1939
p ' 2 2 (.0000) (.1090) "
p-value: H0 I 0'1 = 02 2722 0289
Coefﬁcient Of Variation 35.809 1938 32,975 1934
p-value: test in breaks (.0220) H (.0030) H
p-value: H 0 : 0',2 = 0',2 .0000 .0001

 

 

 

 

 

Notes: The above results are computed based on Hansen test of parameter stability
(Hansen, 1999). N = 46 Single-year birth cohorts, 1923-1968. An F-test for the null
hypothesis Of whether the estimated variances of the regressions in the 2 relevant periods

are equal is tested. The F statistic = s,2 /s§ ~ F TH, T2-.. where sf = estimated variance

of regression 1 and 3% = estimated variance of regression 2 to determine whether to use

homoskedastic or heteroskedatic tests. The relevant homoskedastic or heteroskedastic
bootstrap p-values are shown in parentheses. Superscript H indicates heteroskedstic p—
values. Bootstrap replications = 1000 and trimming = 15%.

98

 

 

Table 4. Monthly Earnings by Highest Grade Completed for Males

Age 20-54 in 1967-69

 

 

Percentage in

Mean Monthly Earnings
(relative to Malays with

 

 

SchoolinLGroup zero schooling)
Schooling Malay Non- Total Malay Non- Total
Malay Malay
(l) (2) (3) (4) (5) (6)

None 16.1% 9.2% 12.3% 1.00 2.16 1.47
Grade 1-3 18.7% 20.3% 19.6% 1.36 1.99 1.71
Grade 4-5 26.1% 17.6% 21.5% 1.55 2.46 1.96
Grade 6 25.1% 21.0% 22.9% 1.54 3.14 2.33
Grade 7-9 5.5% 15.4% 10.9% 2.43 3.16 3.00
Grade 10-12 6.6% 12.3% 9.7% 3.55 4.35 4.10
Upper 6 & Tertiary 1.9% 4.2% 3.2% 4.32 10.8 9.05
Total 100% 100% 100% 1.66 3.17 2.48
No. Of Obs. 379 448 827
Percentage 45.8% 54.2% 100%

 

 

 

 

 

 

 

99

 

 

Table 5. Monthly Earnings by Highest Grade Completed for Males
Age 20-54 in 1976

 

 

 

Mean Monthly Earnings
Percentage in (relative to Malays with
Schoolirgg Group zero schooling)
Schooling Malay N on- Total Malay Non- Total
Malay Malay
(1) (2) (3) (4) (5) (6)
None 12.7% 7.3% 9.8% 1.00 1.71 1.28
Grade 1-3 16.1% 19.1% 17.7% 1.21 1.69 1.49
Grade 4-5 24.2% 15.9% 19.7% 1.53 2.07 1.76
Grade 6 30.1% 21.7% 25.6% 1.38 2.36 1.83
Grade 7-9 7.3% 18.4% 13.3% 2.27 2.97 2.79
Grade 10-12 7.3% 12.4% 10.1% 3.90 4.91 4.57
Upper 6 & Tertiary 2.2% 5.2% 3.8% 6.73 8.99 8.40
Total 100% 100% 100% 1.71 2.90 2.35
No. of obs. 409 483 892
Percentage 45 .9% 54.1% 100%

 

 

 

 

 

 

 

100

 

Table 6. Monthly Earnings by Highest Grade Completed for Males
Age 20-54 in 1988

 

Percentage in

Mean Monthly Earnings
(relative to Malays with

 

 

 

Schooling Group zero schooling)
Schooling Malay Non- Total Malay N on- Total
Malay Malay
(1) (2) (3) (4) (5) (6)

None 4.7% 2.9% 3.8% 1.00 1.39 1.15
Grade 1-3 6.7% 7.3% 6.9% 1.07 1.68 1.39
Grade 4-5 7.1% 7.0% 7.0% 1.24 2.25 1.73
Grade 6 25.2% 22.9% 24.1% 1.25 2.09 1.65
Grade 7-9 20.2% 30.6% 25.3% 1.43 2.02 1.79
Grade 10-12 26.9% 20.7% 23.8% 1.98 2.85 2.36
Upper 6 & Tertiary 9.2% 8.8% 9.0% 3.98 5.69 4.81
Total 100% 100% 100% 1.71 2.51 2.10
NO. of Obs. 2031 1987 4018
Percentage 50.5% 49.5% 100%

 

 

 

 

 

 

 

101

 

 

Table 7. Log Monthly Earnings for Males
Age 20-54 in 1967-69, 1976 and 1988

 

 

 

 

1967-69 1976 1988

Variable Coef. SE Coef. SE Coef. SE
Non-Malay .518 .056 .330 .046 .387 .019
Schooﬁng:
1-3 years .213 .100 .086 .088 .140 .062
4-5 years .360 .097 .321 .086 .307 .061
6 years .448 .099 .316 .085 .336 .054
7 — 9 years .753 .117 .759 .097 .533 .055
10 -12 years 1.08 .122 1.20 .102 .842 .055
>=13 years 1.86 .174 1.85 .135 1.39 .060
Age .149 .026 .093 .023 . 163 .008
Age Squared -.0020 .0004 -.0011 .0003 -.0019 .0001
1968 -.040 .070 -.049 .051
1969 .093 .067
number of jobs .058 .078
Constant 2.35 .445 3.42 .426 2.19 . 152
N 827 892 4018
R-squared .300 .374 .344
Variance (Log Y) .839 .661 .567
Explained Variance .252 .247 .195
Residual Variance .587 .414 .372

 

 

 

 

102

 

Table 8. Simulated Variance of Log Earnings for Males 20-54
in 1967-69, 1976 and 1988

 

Simulated Variance of

 

 

 

 

 

 

 

 

 

Based on Coefficients Log Earnings
for Variables 1967-69 1976 1988
Var % Var % Var %
1. Age .030 3.58 .016 2.4 .077 13.6
2. Age + race .095 11.3 .045 6.8 .111 19.6
3. Age + race + residual .682 81.3 .459 69.4 .483 85.2
4. Age + schooling .149 17.8 .190 28.7 .163 28.7
5. Age + schooling + residual .736 87.7 .604 91.4 .535 94.3
6. Age + race + schooling .248 29.6 .244 36.9 .195 34.4
7. All * + residual .839 100 .661 100 .567 100

 

 

Notes: Simulations are based on distribution Of age, race, and schooling and coefﬁcients

from Regressions in Table 7.

Simulations for each group use coefﬁcients for variables shown, with all other

coefﬁcients set to zero.

* All includes coefﬁcients for year dummies and number of jobs for 1967-69 and

coefﬁcients for number of jobs for 1976.

103

 

Table 9. Log Monthly Earnings Regressions for Malay Males

Age 20-54 in 1967-69, 1976 and 1988

 

 

 

 

1967-69

Variable Coef. SE Coef. SE Coef. SE
Schooﬁng:
1-3 years .275 .151 .070 .122 .099 .082
4-5 years .468 .140 .405 .112 .246 .081
6 years .468 .146 .320 .115 .320 .070
7 — 9 years 1.06 .225 .934 .157 .600 .074
10 —12 years 1.44 .211 1.40 .155 .956 .073
>=13 years 1.78 .347 2.10 .241 1.50 .081
Age .187 .044 .147 .033 .170 .012
Age Squared -.0027 .0006 -.0018 .0004 -.0020 .0002
1968 -.142 .111
1969 .129 .108
number of jobs —.062 .106 -.089 .064
Constant 1.92 .757 2.45 .615 1.99 .218
N 379 409 2031
R-squared .217 .333 .326
Variance (Log Y) .889 .625 .553
Explained Variance .193 .208 .180
Residual Variance .696 .417 .372

 

 

 

 

104

 

Table 10. Log Monthly Earnings Regressions for non-Malay Males
Age 20-54 in 1967-69, 1976 and 1988

 

 

 

 

1967-69 1988

Variable Coef. SE Coef. SE Coef. SE
Schoohng:
1-3 years .102 .131 .058 .128 .163 .095
4-5 years .197 .133 .205 .130 .359 .095
6 years .356 .132 .261 .127 .354 .085
7 — 9 years .538 .139 .623 .131 .482 .085
10 —12 years .843 .148 1.04 .139 .723 .086
>=13 years 1.75 .194 1.69 .169 1.30 .093
Age .124 .030 .038 .033 .156 .012
Age Squared -.0015 .0004 -.0003 .0004 -.0018 .0002
1968 .093 .087
1969 .080 .082
number Of jobs .356 .123 .036 .084
Constant 2.95 .525 4.71 .640 2.73 .213
N 448 483 1987
R-squared .249 .31 1 .294
Variance (Log Y) .606 .579 .517
Explained Variance .151 .180 .152
Residual Variance .455 .399 .365

 

 

 

 

105

 

Table 11. Simulated Variance of Log Earnings for Males

Age 20-54 in 1967-69, 1976 and 1988 by Malays and non-Malays

 

 

 

 

 

 

Simulated Variance of Log Earnings
Malay with Malay with Non-Malay Non-Malay
Malay non-Malay with with
Based on Coefﬁcients coefficients coefficients non-Malay Malay
for Variables coefficients coefficients

Var % “ Var %“l Var % b Var % r
A. 1967-69
1. Age .036 4.0 .032 3.6 .037 6.1 .043 7.1
2. Age + residual .732 82.3 .728 81.9 .492 81.2 .498 82.2
3. Age + schooling .170 19.1 .085 9.56 .143 23.6 .249 41.1
4. Age + schooling + residual .866 97.4 .781 87.8 .598 98.7 .704 116
5. All c + residual .889 100 .796 89.5 .606 100 .716 118
B. 1976
1. Age .029 4.6 .013 2.1 .012 2.1 .023 4.0
2. Age + residual .446 71.4 .430 68.8 .411 70.1 .422 72.9
3. Age + schooling .202 32.3 .106 17.0 .180 31.1 .315 54.4
4. Age + schooling + residual .619 99.0 .523 83.7 .579 100 .714 123
5. All d + residual .625 100 .523 83.7 .579 100 .716 124
C. 1988
1. Age .084 15.2 .070 12.7 .079 15.3 .094 18.2
2. Age + residual .456 82.6 .442 80.1 .444 85.9 .459 88.8
3. Age + schooling .180 32.6 .125 22.6 .152 29.4 .279 54.0
4. Age + schooling + residual .552 100 .497 90.0 .517 100 .644 125

 

 

 

 

 

 

 

 

 

 

 

 

Notes: Simulations are based on distribution of age, race, and schooling and coefﬁcients
from Regressions for 1967-69, 1976 and 1988 in Tables 9 and 10.

Simulations for each group use coefﬁcients for variables shown, with all other
coefﬁcients set to zero. Residual variance is from same ethnic group as coefﬁcients.

a. % calculated based on total variance of log earnings Of Malays.

b. % calculated based on total variance Of log earnings of non-Malays.

c. All includes coefﬁcients for year dummies and number of jobs.

(1. All includes number of jobs for 1976.

106

 

Table 12. Log Monthly Earnings Regressions for Malay Males
Age 20-54 in 1967-69, 1976 and 1988 (include occupation variables)

 

 

 

 

1967-69 1988

Variable Coef. SE Coef. SE Coef. SE
Schoohng:
1-3 years . 146 .141 -.073 .1 13 .099 .076
4-5 years .334 .132 .282 .104 .191 .075
6 years .231 .141 .117 .109 .225 .065
7 — 9 years .575 .224 .522 .158 .361 .070
10 —12 years .559 .251 .642 . 183 .552 .072
>=13 years .698 .383 1.06 .283 .924 .083
Age .130 .042 .112 .030 .152 .011
Age Squared -.0019 .0006 -.0014 .0004 -.0018 .0002
1968 -. 129 .104
1969 .122 .101
number of jobs .058 .103 .001 .060
Managers 1.17 .220 1.06 .174 .862 .058
Clerical .834 .258 .703 . 187 .628 .055
Sales .850 .189 .611 .130 .331 .053
Service .738 . 181 .647 .156 .603 .040
Production .574 .201 .483 .135 .386 .051
Transport .575 .124 .504 .094 .429 .043
Laborer -. 166 .166 -.142 .205 ' .228 .059
Constant 2.74 .719 3.03 .569 2.24 .206
N 379 409 2031
R-squared .340 .450 .425
Variance (Log Y) .889 .625 .553
Explained Variance .302 .281 .235
Residual Variance .5 87 .344 .318

 

 

 

 

107

 

 

Table 13. Log Monthly Earnings Regressions for non-Malay Males
Age 20-54 in 1967-69, 1976 and 1988 (include occupation variables)

 

 

 

 

1967-69 1988

Variable Coef. SE Coef. SE Coef. SE
Schooﬁng:
1-3 years .069 .120 .028 .118 .133 .090
4-5 years .155 .122 .213 .120 .342 .090
6 years .236 .122 .200 .117 .291 .081
7 — 9 years .213 .133 .405 .123 .354 .080
10 —12 years .275 .160 .543 .146 .493 .084
>=13 years 1.02 .213 1.01 .182 .867 .094
Age .099 .028 .020 .031 .146 .01 1
Age Squared -.0013 .0004 -.0014 .0004 -.0017 .0002
1968 .l 12 .081
1969 .101 .076
number Of jobs .414 .1 14 .088 .079
Managers 1.17 .146 1.11 .131 .797 .058
Clerical .707 .144 .727 .129 .328 .063
Sales .459 .091 .436 .082 .501 .046
Service .588 . 173 .477 .172 .296 .062
Production .566 .097 .470 .088 .369 .046
Transport .509 .104 .437 .089 .309 .043
Laborer .216 . 155 .095 .133 -.062 .066
Constant 3.13 .492 4.85 .592 2.75 .204
N 447 483 1987
R-squared .380 .426 .377
Variance (Log Y) .608 .579 .517
Explained Variance .231 .247 .195
Residual Variance .377 .332 .322

 

 

 

 

108

 

 

Table 14. Simulated Variance of Log Earnings for Males

Age 20-54 in 1967-69, 1976 and 1988 by Malays and non-Malays

(include occupation variables)

 

 

 

 

 

 

 

Simulated Variance of Log Earnings
Malay with Malay with Non-Malay Non-Malay
Malay non-Malay with with
coefﬁcients coefﬁcients non-Malay Malay
Based on Coefﬁcients coefﬁcients coefﬁcients
for Variables Var % " Var %‘il Var % B Var % F
A. 1967-69
1. Age .019 2.14 .011 2.02 .021 3.45 .021 3.45
2. Age + residual .606 68.2 .345 68.1 .398 65.5 .398 65.5
3. Age + schooling .048 5.40 .029 3.26 .047 7.73 .059 9.70
4. Age + schooling + residual .635 71.4 .616 69.3 .424 69.7 .436 71.7
5. Age + school + occupation .286 32.2 .213 24.0 .223 36.7 .310 51.0
6. Age + school + occupation
+ residual .873 98.2 .800 89.9 .600 98.7 .687 113
7. All c + residual .889 100 .801 90.1 .608 100 .696 114
B. 1976
1. Age .014 2.24 .005 0.80 .005 0.86 .011 1.89
2. Age + residual .358 57.3 .349 55.8 .337 58.2 .343 59.2
3. Age 4» schooling .066 10.6 .034 5.44 .057 9.84 .097 16.8
4. Age + schooling + residual .410 65.6 .378 60.5 .389 67.2 .429 74.1
5. Age + school + occupation .281 44.9 .222 35.5 .247 42.7 .313 54.1
6. Age + school + occupation
+ residual .625 100 .566 90.6 .579 100 .645 111
7. All d + residual .625 100 .561 89.8 .579 100 .645 111
C. 1988
1. Age .063 l 1.4 .059 10.7 .066 12.8 .071 13.7
2. Age + residual .381 68.9 .377 68.2 .388 75.0 .393 76.0
3. Age + schooling .086 15.6 .076 13.7 .094 18.2 .109 21.1
4. Age + schooling + residual .404 73.1 .394 71.2 .416 80.5 .431 83.4
5. Age + school + occupation .235 42.5 .181 32.7 .195 37.7 .225 43.5
6. Age + school + occupation
+ residual .553 100 .499 90.2 .517 100 .547 106

 

 

 

 

 

 

 

 

 

Note: Simulations are based on distribution of age, race, and schooling and coefﬁcients
from Regressions for 1967-69, 1976 and 1988 in Tables 12 and 13. Simulations for each group
use coefﬁcients for variables shown, with all other coefﬁcients set to zero. Residual variance is
from same ethnic group as coefﬁcients.

a.

b.
c.
d

 

 

Calculated based on total variance of log earnings of Malays.

% calculated based on total variance of log earnings of non-Malays.
All includes coefﬁcients for year dummies and number Of jobs.

All includes number of jobs for 1976.

109

 

Table 15. Means and Standard Deviations of Selected Variables

 

Children sample at age 17

Children sample at age 20

 

 

 

Variables Son Daughter Son Daughter
Child’s schooling:
Malays 8.83 8.40 8.72 8.17
(2.52) (3.12) (2.97) (3.60)
Non-Malay 8.27 7.96 8.37 7.97
(2.61) (2.88) (3.04) (3.34)
Mother’s schooling:
Malays 3.08 3.22 2.54 2.73
(2.95) (3.19) (2.72) (2.88)
Non-Malays 3.34 3.33 3.08 2.85
(3.46) (3.38) (3.33) (3.32)
Father’s schooling:
Malays 3.21 3.35 3.01 3.11
(2.93) (3.17) (2.72) (2.95)
Non-Malays 3.35 3.78 3.38 3.77
(3.33) (3.43) (3.33) (3.49)
Child’s birth cohort:
Below 1957 .237 .239 .305 .314
1957—1959 . 186 .226 .239 .298
1960-1962 .081 .076 .105 .101
1963-1965 .120 .095 .154 .124
1966-1968 .164 .128 .197 .163
1969-1971 .212 .236 - -
SMK .166 .164 .126 .122
(.248) (.257) (.225) (.236)
Malay SRK .398 .410 .378 .406
(.340) (.340) (.353) (.362)
Chinese SRK .106 .104 .098 .084
(.218) (.209) (.212) (.191)
Tamil SRK .061 .069 .060 .065
(.147) (.161) (.157) (.173)
Urban .148 .143 .118 .105

 

 

 

 

 

110

 

Table 15 (cont’d.)

 

 

Piped Water .487 .501 .450 .469
NO. of observations:
Malays 549 333 397 248
Non-Malays 668 493 550 379
Total 1217 826 947 627

 

 

 

 

 

 

Notes: Means are reported with standard deviations in parentheses. SRK represents
national primary schools, SMK represents national secondary schools. The means are
computed for non-missing values, therefore the number Of observations decreases slightly
for those variables.

 

lll

 

Table 16. Regressions on Years of Schooling of Male Children Age 17 and above

 

 

 

 

p-value

 

 

 

(.032)

(2) (3)
Variable Coef. SE Coef. SE Coef. SE
Child’s birth cohort:
1957-1959 .184 .230 .121 .224 .154 .227
1960-1962 1.18 .364 1.08 .357 1.17 .472
1963-1965 1.69 .318 1.54 .314 1.58 .441
1966-1968 1.94 .300 1.67 .297 1.64 .440
1969-1971 1.94 .283 1.64 .283 1.56 .432
Non-Malay .133 .236 .055 .230 -.245 .242
Non-Malay*cohort(21960) -.799 .308 -.739 .298 -.457 .336
Mother’s schooling:
Primary .668 .178 .957 .250
Secondary 1.58 .332 2.43 .671
Father’s schooling:
Primary .375 . 163 .079 .264
Lower Secondary 1.11 .356 1.19 .640
Upper Secondary 2.43 .448 2.65 .707
Mother’s schooling*birth
cohort21960:
Primary*cohort21960
Secondary* cohort21960 '-708 363
-1.35 .786
Father’s schooling*birth
‘ cohort21960:
Primary*cohort_>_1960
Lower Secondary* '482 '338
cohort_>_l960
Upper Secondary* “152 '774
cohort21960 4.03 .916
F-test for parent’s
Schooling* cohort21960
F-statistlc 2.45

 

112

 

 

 

 

Table 16(cont’d.)
Malay SRK*Malay .170 .499 ‘
Malay SRK*Non-Malay -.396 .537
Chinese/Tamil SRK*Non-
Malay .290 .575
SMK .546 .438
Urban .261 .241
Piped Water .411 .167
R-squared .073 . 148 . 173
No. Of observations 1217 1217 1217

 

 

 

 

Notes: Missing values for parent’s schooling and community variables on are changed to
zero, and a dummy variable was included for these variables to account for the missing
values.

113

 

 

Table 17. Regressions on Years of Schooling of Female Children Age 17 and above

 

 

 

 

(2) (3)
Variable Coef. SE Coef. SE Coef. SE
Child’s birth cohort:
1957-1959 .429 .291 .387 .282 .382 .282
1960-1962 2.92 .479 2.72 .466 3.21 .629
1963-1965 3.01 .454 2.86 .443 3.26 .610
1966-1968 3.66 .418 3.33 .413 3.64 .603
1969-1971 3.09 .376 2.72 .373 3.04 .580
Non-Malay .236 .299 .054 .294 . 146 .331
Non-Malay*cohort(21960) - l .04 .406 -.889 .394 -.645 .434
Mother’s schooling:
Primary .446 .238 .485 .312
Secondary 1.40 .531 2.20 .864
Father’s schooling:
Primary .797 .228 1.31 .355
Lower Secondary 1.97 .438 2.68 .645
Upper Secondary 2.42 .518 3.25 .830
Mother’s schooling*birth
cohort21960:
Primary*cohort21960 -- 100 .489
Secondary*cohort21960 "981 1-02
Father’s schooling*birth
cohort21960:
Primary*cohort21960 "95 1 '463
Lower Secondary* 1 67 886
cohort.>_1960 ' ' '
Upper Secondary*
cohort21960 -2'23 1'07
F-test for parent’s
Schooling* cohort21960
F-statrstlc 2'24
p-value

 

 

 

(.048)

 

114

 

 

 

 

 

Table 17 (cont’d.)
Malay SRK*Malay .600 .568
Malay SRK*Non-Malay -.698 .590
Chinese/'1‘ amil SRK*Non-
Malay .849 .705
SMK .137 .630
Urban .380 .340
Piped Water .242 .221
R-squared . 162 .229 .260
NO. of observations 826 826 826

 

 

 

 

 

Notes: Missing values for parent’s schooling and community variables on are changed to
zero, and a dummy variable was included for these variables to account for the missing
values.

 

 

115

 

Table 18. Regressions on Years of Schooling of Male Children Age 20 and above

 

 

 

 

(2) (3)
Variable Coef. SE Coef. SE Coef. SE
Child’s birth cohort:
1957-1959 .072 .262 .028 .252 .064 .252
1960-1962 1.16 .431 1.11 .417 .539 .565
1963-1965 1.70 .372 1.57 .362 .988 .533
1966-1968 1.94 .357 1.64 .349 .940 .540
Non-Malay . 188 .268 -.098 .260 -. 168 .273
Non-Malay*cohort(21960) -.801 .394 -.799 .375 -.241 .416
Mother’s schooling:
Primary .790 .221 1.05 .279
Secondary 2.07 .446 2.77 .747
Father’s schooling:
Primary .645 .209 .074 .294
Lower Secondary 1.35 .470 1.16 .712
Upper Secondary 3.73 .585 3.08 .789
Mother’s schooling*birth
cohort21960:
Primary*cohort21960 --726 .463
Secondary*cohort21960 '1-29 954'
Father’s schooling*birth
cohort21960:
Primary*cohort.>_1960
Lower Secondary* 1'03 '418
cohort21960
Upper Secondary* '357 '948
cohort21960 -.256 1.18
F-test for parent’s
Schooling* cohort_>_1960
F-statlstrc 2.09

p-value

 

 

 

(.064)

 

116

 

 

 

Table 18 (cont’d.)

 

 

Malay SRK*Malay .362 .706
Malay SRK*Non-Malay -1.30 .715
Chinese/Tamil SRK*Non-

Malay .132 .760
SMK 1.85 .603
Urban .335 .331
Piped Water .383 .211
R-squared .051 .156 . 197

N o. of observations 947 947 947

 

 

 

 

Notes: Missing values for parent’s schooling and community variables on are changed to
zero, and a dummy variable was included for these variables to account for the missing
values.

117

 

 

Table 19. Regressions on Years of Schooling of Female Children Age 20 and above

 

 

 

 

(1) (2) (3)
Variable Coef. SE Coef. SE Coef. SE
Child’s birth cohort:
1957-1959 .363 .327 .314 .313 .360 .313
1960-1962 3.06 .563 2.86 .542 3.17 .774
1963-1965 3.16 .537 3.02 .520 3.27 .754
1966-1968 3.88 .499 3.51 .492 3.55 .765
Non-Malay .333 .336 -.021 .328 .208 .377
Non-Malay*cohort(21960) -1.08 .535 -.953 .515 -.540 .565
Mother’s schooling:
Primary .382 .293 .456 .347
Secondary 1.81 .590 2.24 .961
Father’s schooling:
Primary 1.07 .290 1.27 .395
Lower Secondary 2.37 .557 2.77 .718
Upper Secondary 3.57 .676 3.57 .923
Mother’s schooling*birth
cohort21960:
Primary*cohort_>_l960 "-325 -663
Secondary*cohort21960 ‘-901 1-26
Father’s schooling*birth
cohort_>_l960:
Primary*cohort21960 "629 '594
Lower Secondary*
cohort21960 '1'35 1‘16
Upper Secondary*
cohort21960 “1'38 1'37
F-test for parent’s
Schooling* cohort21960
F-statrstrc 0.77

p-value

 

 

 

(.574)

 

118

 

 

 

Table 19 (cont’d.)

 

 

 

Malay SRK*Malay . .482 .732
Malay SRK*Non-Malay -1.05 .773
Chinese/Tamil SRK*Non-

Malay .620 1.02
SMK 1.07 .909
Urban .378 .491
Piped Water .562 .278
R-squared . 149 .238 .273

N O. of Observations 627 627 627

 

 

 

 

 

Notes: Missing values for parent’s schooling and community variables on are changed to
zero, and a dummy variable was included for these variables to account for the missing
values.

 

119

Years of completed schooling

Years of completed schooling

Figure 1. Mean Years of Schooling of Males
(3-year moving averages)

11141 #141111 11 l lAlgl 11L

 

11
10“
94
8T
72
6‘
5a
44
32
2-

1—

 

 
   

non-Malays

e . AM

' E/ Malays
#M

 

 

11‘
10"
92
8“
7_

6—«

T I l E l T I I l l I I I I I I I I I I I l I
24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62 64 66 68
birth cohort

Figure 2. Mean Years of Schooling of Females
(3-year moving averages)

 

    
 

non-Malays

 

 

 

Illlllliﬁlllrﬁlllﬁlllrl
24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62 64 66 68

birth cohort

120

standard deviation

standard deviation

Figure 3. Standard Deviation in Years of Schooling of Males
(3-year moving averages)

141141 lld#1111111 14411 1

 

non-Malays

   

 

 

 

I I I I I I I I I I I I I I I I I I T l I I I
24 26 28 30 32 34 36 38 4O 42 44 46 48 50 52 54 56 58 60 62 64 66 68
birth cohort

Figure 4. Standard Deviation in Years of Schooling of Females
(3-year moving averages)

 

 

 

 

 

2 _ a
1 .5 _ ._
1 ._.. l—
T l I I I I I H I I I I I I I I I I I I I I I
24 26 28 3O 32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62 64 66 68
birth cohort

121

coefficient of variation

coefficient of variation

Figure 5. Coefﬁcient of Variation in Years of Schooling of Males
(3-year moving averages)

1111131111111113E114m1
1.4“ 4

 

 
   

non-Malays --«-.~..,;;.._

- n. .
‘.-.A...-

 

 

 

I I I I I I I I I I I I I I I I I I I I I I I
24 26 28 3O 32 34 36 38 4O 42 44 46 48 50 52 54 56 58 60 62 64 66 68
birth cohort

Figure 6. Coefficient of Variation in Years of Schooling of Females
(3-year moving averages)

 

1.4T r

1.3‘4 L

12* _

ml t
Malays

1 T / T

.9“ T

   
 

non-Malays

 

 

 

I I I I I T I I I I I I I f I I j V I I I I I
24 26 28 30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62 64 66 68
birth cohort

122

I.‘ '
fl

Cumulative percent of schooling

Cumulative percent of schooling

Figure 7. Lorenz Curves for Schooling by Age Group

 

 

 

 

 

M l 1 l g l
100

80 n

so 6

40 e

20 a age 55-59

0 .

I I I I I I
o 20 40 so 80 100

Cumulative percent of population

Figure 8. Lorenz Curves for Schooling by Age and Ethnic Group

 

100‘”

 
   
 
   

(I)
O
l

Malays (25-29)
non-Malays (25-29)

0)
O
l

A
O
l

' Malays (55-59)

N
O
l

‘ non-Malays (55-59)

 

 

 

I I I I I I
O 20 4O 60 80 1 00
Cumulative percent of population

123

Cum. °/o of schooling * mean of schooling

Cum. % of eamings‘mean earnings

Figure 9. Generalized Lorenz Curves for Schooling

by Age and Ethnic Group

 

 

  
 

Malays (25-29)

non-Malays (25-29)\

 

 

 

 

I I I I
20 4O 60 80
Cumulative percent of population

Figure 10. Generalized Lorenz Curves for Real Monthly Earnings

by Ethnic Group in 1967-69 and 1988

l l l l

 

800 a
700 J
600 ~
500 ~
400 e
300 —
200 e

100“

 

non-Malays 1 988

non-Malays 1967-69

Malays 1 988

Malays 1 967-69 . "i ,-

 

 

I I I I

20 4O 60 80
Cumulative percent of population

124

 

APPENDIX A

Measurement Error in Earnings

It is Observed that the residual variance Of log earnings regression of Malays were
substantially higher than non-Malays for the period 1967-69. One possible reason is that
there is a positive association between the length Of recall period and measurement error
which is greater for Malays than non-Malays. This could be due to the fact that Malays
were less educated than non-Malays during that period and therefore have a relatively
higher rate of forgetting. In order to examine this relationship, a model to test whether

measurement error is associated with the length Of recall period is written down below.

In 13(1) — 1n Y; = m.,(r) = a0 + a,(r—r)+ (a, + ﬂ,(r—t))e,,., (1)

where suscripts i, t, 1 indicate individual i, time t and report period in year 't' (1976), Y is
the observed earnings, Y* is the true earnings, m is the measurement error, on measures
the systematic over-reporting or under-reporting Of retrospective earnings, [30 and B]
estimates a linear trend effect of recall on measurement error in earnings that is random, t

is the number retrospective years Of earnings being recalled, and 8 is the error term.

an,(r+k)—ln Y; = m,,(r+k) = a0 +a,(r+k—t)+()80 + ,8,(r+k —t))e,,.,,, (2)

where k = 12, since MFLS2 was surveyed in 1988.

Equation (2) - (1),

1n Y,,(r+k)-ln Y},(r) = m,,(r+ k) —m,.,(r) = a,k + (,60 + ﬂ,(r+k))-e
- (:60 + ﬂit!» ' £11.: + (—ﬂl )Eit.r+k 't- (—ﬂl )gitJ 't

ir.r+k

(3)

ﬁx

Assume that 8m ~ (0, of) for all I, 8,131), ~ (0, of) for all T+k, and E(ei(,I - 81ml.) = 0,

Therefore, E[m,., (7' + k) - mi, (1)] = a,k , (4)

where or) may or may not depend on independent variables, X, such as race and
education.

Let,
60=ﬂu+ﬂl(‘l'+k), 6(i=.60+761(7)i an(161 z-ﬂl'
Then equation (3) can be rewritten as,

In Yi,(t‘+k)-—ln 111(1) = m,,(r+k)—mi,(t') = alk +60 -8

it.r+k

, (5)
_ 60 ' 8171 + 61811.r+k . t — 61811.1 't
Assuming (11 = 0, then
E[(m,., (r + k) — m, (1))2] = ago: + (9520: + 2606,03 -z+ 2656,63 . t (6)
+ 6120: - 12 + 6120': -t2 = (602 + 652 )0,2 + 219l (610 + 6,303 ~t+ 28,20,2 -t2
Normalizing of =1, Equation (6) can be rewritten as,
E[(m,, (r + k) - m, (1))2] = (602 + 6752) + 26,090 + 65) - r + 261,2 42 (7)

Empirical Results

A measure Of measurement error on earnings is Obtained by matching the
retrospective earnings data from the same individual, occupation and year from the panel
sample Of MFLS 1 and 2. Due to the different nature Of earnings data collected in the two
waves, only those having one job are included in the sample. The sample that can be

matched are 389 Observations which is about 15% of total possible matches.1

 

1 There is a concern that the sample used may not be representative. This is because the matching procedure
restricts the sample to those who report consistent information for both waves of the survey in terms of
occupation, one job only and time of recall period. Imposing such restrictions may generate a sample of
respondents that is more reliable than the general population.

126

 

"5

Ils-
.1.
I

 

Descriptive Statistics
It is useful to provide some descriptive statistics on the dependent and explanatory
variables used in the regression analyses Of measurement error. The deﬁnitions of the

relevant variables are as follows:

Simple difference in reported log earnings = ln Yif’s — ln Y"76

Difference in reported log earnings square = (In Y1,88 — ln 1376f

where lnY is monthly earnings in natural log, superscript 76 and 88 indicates the survey
year and subscript i and t represents individual i and time t which is the year that the
earnings data are being recalled. The descriptive statistics of the relevant variables are
shown in Table A1.

The simple difference in reported log earnings is a measure of under-reporting or
over-reporting Of earnings in 1988 assuming that the data in 1976 is more accurate
because it has a shorter recall period of 12 years. It is noted that on average there is over-
reporting of earnings for both Malays and non-Malays, but it is slightly higher for
Malays. When computing the mean of the simple difference in reported earnings, the
under-reporting and over-reporting cancels out when averaged over all respondents
within each group. An indicator Of measurement error is the difference in reported log
earnings square. It is Observed that the mean difference in reported log earnings square is
higher for Malays than non-Malays. It is also Observed that Malays have relatively lower
mean years of education than non-Malays. The mean number Of retrospective years of
recall from 1976 is 17.25 years for Malays and 15.25 years for non-Malays. The earliest

year of recall for earnings is 1925.

127

 

 

Zero mean in measurement error

The speciﬁcation in Equation (4) is applied to examine whether measurement
error in earnings have a zero mean using simple difference in reported earnings as the
dependent variable. The dependent variable of simple difference in reported earnings
indicates over-reporting and under-reporting Of earnings. The estimates for the constant
term are of interest. The results for the whole sample in Table A2 indicate that the
constant term is not signiﬁcant in Model 1 when the number of retrospective years2 is the
only regressor. The constant term is also not signiﬁcant in Model 2 when ethnicity is
added. In Model 3 years of education is included as an additional control, the constant
term remains insigniﬁcant. For further analysis, the separate regression results Of Malays
and non-Malays are presented in Tables A3 and A4 respectively. It is interesting to note
that the constant term for Malays are signiﬁcant at the 1% level but insigniﬁcant for non-
Malays in both Models 1 and 2. This suggests that the mean measurement error of
Malays is non-zero while it is zero for non-Malays.

Another question Of interest is whether the under-reporting or over-reporting of
earnings is associated with ethnicity, length Of recall period and education. The estimates
based on the whole sample shown indicates that ethnic dummy, years Of education,
number Of recall years are not signiﬁcant in explaining over-reporting or under-reporting
of earnings. However, number of retrospective years is negative and signiﬁcant at the 5%
level for Malays, but highly insigniﬁcant for non-Malays. The negative coefﬁcient for the
number of retrospective years implies that the longer the period of recall the larger is the

under-reporting of earnings for Malays. The estimated coefﬁcient for years of education

128

.1

me_" '_ ‘ l

 

 

is negative and insigniﬁcant for both Malays and non-Malays. However, the negative
coefﬁcient means that the higher the level Of education the lower is the under-reporting Of

earnings.

Variance in measurement error

The examination Of the variance in measurement error in earnings is of particular
relevance to this study. This is because it enables us to shed some light on whether the
high residual variance in log earnings for Malays during the 1967-69 period is due to
differences in measurement error between Malays and non-Malays. Therefore, it is of
interest to ﬁnd out whether the variance in measurement error in the recall of reported
earnings increases faster for Malays than non-Malays. In order to do so, the regression
based on the speciﬁcation in Equation (7) is carried out3. Regression results with the
difference in reported log earnings square as the dependent variable are shown in Table
A5. Model 1 presents the linear speciﬁcation for number of retrospective years and
Model 2 presents the quadratic Speciﬁcation for the number of retrospective years.

The difference in reported log earnings square is a measure of the variance of
measurement error in earnings. Based on Model 1, it is interesting to note that the
estimated coefﬁcient for the number Of retrospective years is positive and signiﬁcant.
This implies that the variance of measurement error is increasing and signiﬁcant for both
Malays and non-Malays. This ﬁnding is consistent with the hypothesis that the variance
Of measurement error is positively associated with the number Of years of recall. It is also

interesting to note that the number of retrospective years interacted with non-Malay

 

2 For ease of illustration, number of retrospective years is referred as number of retrospective years from
1976.

129

Ii

1'

 

 

dummy variable is negative but insigniﬁcant. Based on the usual standard of inference, it
is Observed that there is no signiﬁcant difference in the rate of forgetting between Malays
and non-Malays. However, from the point of view of point estimates, the rate Of
forgetting Of Malays is faster than non-Malays. The magnitude of the estimated
coefﬁcient of the non-Malay interaction with the number of retrospective years (-.037) is
considered large compared to the estimated coefﬁcient of number Of retrospective years
(.042). Based on the mean retrospective years Of non-Malays, it is estimated that the non-
Malay interaction term reduces the measurement error variance Of non-Malays relative to
Malays by .555.4 However, caution needs to be exercised when interpreting the point
estimates because the estimated coefﬁcient is extremely imprecise. The imprecise
estimate could be due tO the small sample size which is unable to detect the subtle
differences in measurement error between Malays and non-Malays. However, it is
noteworthy that the joint F-test for number of years of retrospective recall and its
interaction with non-Malay dummy variable is jointly signiﬁcant in Model 1.

Based on the quadratic speciﬁcation for the number Of retrospective years in
Model 2, it is noted that the variance of measurement error decreases initially for both
Malays and non-Malays. Subsequently, the variance of measurement error increases after
10 years of recall for Malays and for non-Malays it increases after 18 years recall. This
ﬁnding is puzzling and is counter-intuitive. I am not able to Offer a reasonable

explanation for this result. However, this may be an area for further research and it is

 

3 By pooling the data of Malays and non-Malays, the regression model included the non-Malay dummy
variable and its interaction with number Of retrospective years.

4 Based on the mean retrospective years of non-Malays and the estimated coefﬁcient of number of
retrospective years interacted with non-Malay dummy variable in Model 1, its estimated effect on
measurement error variance equals (15*—.037) = -.555. This is quite a substantial difference between
Malays and non-Malays compared to the mean difference in reported log earnings square which is .7423
(Malays) and .8865 (non-Malays).

130

 

 

suggested that a linear spline speciﬁcation for the number of retrospective years may
yield better results.

The joint F test of the number of years Of recall and its square interacted with the
non-Malay dummy variable in Model 2 is also not signiﬁcant. This means that the
variance of measurement error in log earnings by years Of retrospective recall are not

signiﬁcantly different between Malays and non-Malays.

Conclusion

The unique data set available from the two waves of MFLS has provided a good
opportunity to examine the issue of measurement error in retrospective earnings by
providing two sets of earnings data which can be matched to the same individual, job and
time period. The conclusions that can be derived with regard to the mean Of measurement
error is that it is not signiﬁcantly different from zero for non-Malays. But it is
signiﬁcantly different from zero for Malays. The linear speciﬁcation of number of
retrospective years of recall indicates that the variance Of measurement error in log
earnings is signiﬁcant and positive. But the quadratic speciﬁcation Of number of years of
recall and its square reveals the pattern that the measurement error variance in log
earnings initial decreases and then increase after a number Of years into the past is
contrary to expectation. Following the statistical point Of view, the variance of
measurement error in log earnings do not vary signiﬁcantly by years Of retrospective
recall between Malays and non-Malays. However, the point estimates of the number of
retrospective years (linear speciﬁcation) interacted with non-Malays suggest that the rate

Of forgetting of non-Malays are substantially lower than Malays.

131

 

 

 

Table A1. Means and Standard Deviation of Variables Used

 

 

 

 

 

 

Variables Malays Non- Total
Malays
Simple difference in .0535 .0073 .0324
reported log earnings (.8619) (.9442) (.8996)
Difference in .7423 .8865 .8083
reported log earnings square (2.06) (2.59) (2.31)
Years of education 5.84 6.67 6.22
(3.76) (4.20) (3.98)
NO. of Retrospective Years 17.25 15.25 16.33
(from MFLSl in 1976) (9.02) (10.31) (9.67)
No. Ofobs. 211 178 389 '

 

 

Note: standard deviations are in parentheses

Table A2. Regression of simple difference in reported log earnings

 

 

 

 

Model 1 Model 2 Model 3
Variable Coef. SE Coef. SE Coef. SE
NO. of retrospective
years -.0046 .0057 -.0559 .0912 -.0055 .0059
N on-Malay -.0049 .0057 -.0533 .0921
Years of education -.0047 .0114
Constant .1074 .0955 .1379 .0985 .1758 .1496
R-Squared .0024 .0034 .0038
N 389 389 389

 

 

 

 

 

Note: * signiﬁcant at 10% level, ** signiﬁcant at 5% level

SE = robust standard errors

132

YET—”Til.

 

Table A3. Regression of simple difference in reported log earnings of Malays

 

 

 

 

Model 1 Model 2
Variable Coef. SE Coef. SE
No. of retrospective
years -.0154** .0089 -.0172** .0068
Years Of education -.0106 .0151
Constant .3200 *** .1492 .4132*** .2393
R-squared .0261 .0279
N 211 211

 

 

 

 

Note: * signiﬁcant at 10% level, ** signiﬁcant at 5% level, *** signiﬁcant at 1% level

SE = robust standard errors

Table A4. Regression of simple difference in reported log earnings of non-Malays

 

 

 

 

Model 1 Model 2
Variable Coef. SE Coef. SE
NO. Of retrospective
years .0047 .0089 .0043 .0093
Years Of education -.0043 .0158
Constant -.0641 .1491 -.0291 .2393
R-squared .0026 .0030
N 178 178

 

 

 

 

SE = robust standard errors

133

 

 

Table A5. Regression of difference in reported log earnings square

 

 

 

 

Variable Model 1 Model 2

Coef. SE Coef. SE
Non-Malay .798* .458 .752 .619
NO. Of retrospective years .042** .017 -.062 .050
NO. Of retrospective
years*non-Malay -.037 .029 -.031 .077
No. of retrospective
years square .003* .0018
NO. Of retrospective
years square*non-Malay -.0004 .0022
Constant .020* .256 .640** .268
Joint F-test: no. Of
retrospective years and non-
Malay interaction terms
F statistic 3.01 3.16
p-value (.050) (.014)
Joint F-test: non-Malay
interaction terms only
F statistic 1.12
p-value (.327)
R-square .0156 .0338
N 389 389

 

 

 

Notes: * signiﬁcant at 10% level, ** signiﬁcant at 5% level,*** signiﬁcant at 1% level

SE = robust standard errors

134

 

APPENDIX B

Table Bl. Log Monthly Earnings Regressions of Malay Males
Age 20-34 in 1967-69, 1976 and 1988 (include occupation variables)

 

 

 

 

1967-69 1976 1988

Variable Coef. SE Coef. SE Coef. SE
Schoohng:
1-3 years - .388 .199 -.192 .247 .098 .173
4-5 years - .393 .191 .151 .221 .299 .176
6 years — .138 .180 .120 .212 .157 .149
7 — 9 years .100 .258 .420 .262 .238 .147
10 —12 years .344 .315 .680 .289 .366 .147
>=13 years .368 .426 1.29 .390 .599 .155
Age -.034 . 170 -.039 . 186 . 179 .054
Age Squared .0013 .0031 .0013 .0033 -.0022 .0009
1968 -.316 .124
1969 .039 .114
number ofjobs .012 .124 .1 19 .088
Managers .835 .284 .828 .273 .822 .073
Clerical .748 .305 .762 .237 .597 .067
Sales .418 .243 .367 .180 .434 .070
Service .484 .213 .499 .293 .638 .052
Production .434 .220 .393 .188 .371 .060
Transport .475 . 147 .501 .172 .476 .058
Laborer .289 .355 .288 .315 .256 .073
Constant 5.45 2.30 5.04 2.59 1.94 .741
N 220 160 1068
R-squared .393 .555 .438
Variance (Log Y) .704 .553 .466
Explained Variance .277 .307 .204
Residual Variance .427 .246 .262

 

 

 

 

135

 

APPENDIX C

Table C1. Log Monthly Earnings Regressions of Malay Males
Age 35-54 in 1967-69, 1976 and 1988 (include occupation variables)

 

 

 

 

1967-69 1988

Variable Coef. SE Coef. SE Coef. SE
Schoohng:
1-3 years .378 .206 -.065 .134 .098 .088
4-5 years .802 .196 .312 .127 . 174 .086
6 years .252 .257 .044 .141 .241 .076
7 - 9 years .103 .503 .713 .235 .343 .090
10 -12 years .481 .463 .563 .267 .718 .100
>=13 years 1.31 .975 .746 .526 1.44 .125
Age -.160 .280 .167 .128 .086 .060
Age Squared .0014 .0032 .0020 .0015 -.0009 .0007
1968 -.049 .179
1969 . 169 .183
number of jobs .024 .177 -.062 .082
Managers 1.22 .370 1.20 .239 .787 .092
Clerical .692 .459 .572 .307 .710 .097
Sales .955 .309 .761 .184 .210 .079
Service .936 .306 .745 . 195 .55 3 .063
Production .563 .414 .593 .202 .478 .089
Transport .699 .207 .521 .1 17 .377 .062
Laborer -.615 .428 -.331 .273 . 174 .095
Constant 8.54 6.02 1.94 2.78 3.49 1.29
N 158 249 963
R-squared .405 .424 .459
Variance (Log Y) 1.14 .673 .647
Explained Variance .462 .286 .297
Residual Variance .678 .387 .350

 

 

 

 

136

 

 

IIIIIIIIIIIIIIIIIIIIIIIIIIIIII

IIIIIIII IIIIIIIIIIIIOIIIIOIIIBI IjIIIIIIIIIII IIII