University of Southern California Dissertations and Theses

Essays on labor and development economics

(USC Thesis Other)

Essays on labor and development economics

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

ESSAYS ON LABOR AND DEVELOPMENT ECONOMICS

by

Voraprapa Nakavachara

________________________________________________________________________

A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(ECONOMICS)

December 2007

Copyright 2007 Voraprapa Nakavachara

ii
Acknowledgements

This is the page I have always wanted to write. There are many people whom I
owe my gratitude to. First, I am indebted to John Strauss for giving me invaluable advice,
for patiently making sure that I do my work carefully, for caring about me, and for
believing in me. I am very fortunate to have him as my advisor.
Second, I am thankful for thoughtful suggestions and constructive comments from
my committee members: W. Bentley MacLeod, John Ham, and Gary Painter. I truly
appreciate their guidance.
Third, I am grateful to my parents for giving me this wonderful life, great
opportunities, as well as support. I would not have made it this far without them.
Fourth, I thank Sutham Saengpratoom, Numkrit Jeraputtiruk, Anon Juntavich,
David Autor, and Jean Roth. I barely know these people in real life yet their
overwhelming generosity contributed a great deal to the accomplishment of this work.
Last, I thank my wonderful friends at the University of Southern California. I am
blessed to have met many big-hearted people and to have become friends with them.
Good friendship makes troubles smaller and makes life more meaningful. Nayoung Lee,
Huseyin Gunay, Echu Liu, Heonjae Song, Serkan Ozbeklik, and Brijesh Pinto, I thank
you.

iii
Table of Contents

Acknowledgements ii

List of Tables v

List of Figures vii

Abstract ix

Chapter One: Superior Female Education: 1
Explaining the Gender Earnings Gap Trend in Thailand
1.1 Introduction 1
1.2 Socio-economic Background 8
1.2.1 Growth, Poverty, and Income Inequality 9
1.2.2 Expansion of Education 13
1.2.3 Gender Earnings Inequality in Neighboring Countries: 17
The Literature
1.3 Data 22
1.4 The Thai Labor Market and Gender Inequality 28
1.5 Parametric Decomposition 65
1.5.1 Blinder-Oaxaca (1973): BO 66
1.5.2 Juhn-Murphy-Pierce (1991): JMP 70
1.5.3 BO Results 74
1.5.4 JMP Results 84
1.6 Nonparametric Decomposition 90
1.6.1 DiNardo-Fortin-Lemieux (1996): DFL 91
1.6.2 DFL Results 101
1.7 Conclusion 110

Chapter Two: Wrongful Discharge Laws and the Unexpected Substitution Effect 113
2.1 Introduction 113
2.2 Previous Literature: The Economics of Employment Law 118
2.2.1 Wrongful Discharge Laws (WDLs) 119
2.2.1.1 Implied Contract 119
2.2.1.2 Good Faith 121
2.2.1.3 Public Policy 124
2.2.2 Employment Consequences of Wrongful Discharge Laws 125
2.3 Theoretical Framework for Wrongful Discharge Laws 128
2.4 Data 132
2.5 Empirical Methodology 145

iv
2.6 Results 148
2.6.1 Good Faith 148
2.6.2 Implied Contract 157
2.6.3 Public Policy 165
2.7 Conclusion 170

Bibliography 172

Appendix 178
Appendix Table A.1 178
Appendix Table A.2 179

v
List of Tables

Table 1.1: Ratio of female to male earned income (selected countries) 3

Table 1.2: Average hours worked per week for male and female workers 28
(wage and salary sector)

Table 1.3: Population and labor force structure in Thailand 33

Table 1.4A: Basic summary statistics of wage and salary workers 45

Table 1.4B: Basic summary statistics of wage and salary male workers 46

Table 1.4C: Basic summary statistics of wage and salary female workers 47

Table 1.5A: Earnings equation (without occupations and industries) 75

Table 1.5B: Earnings equation (with occupations and industries) 77

Table 1.6A: Blinder-Oaxaca (1973) results (without occupations and industries) 78

Table 1.6B: Blinder-Oaxaca (1973) results (with occupations and industries) 83

Table 1.7A: Juhn-Murphy-Pierce (1991) results 85
(without occupations and industries)

Table 1.7B: Juhn-Murphy-Pierce (1991) results 89
(with occupations and industries)

Table 2.1: Adoption dates 127

Table 2.2A: Any-training index 136

Table 2.2B: School-training index 138

Table 2.2C: Formal-training index 140

Table 2.2D: Informal-training index 142

Table 2.3A: Good Faith and employment 149

Table 2.3B: Good Faith and wages 158

Table 2.4A: Implied Contract and employment 160

vi

Table 2.4B: Implied Contract and wages 163

Table 2.5A: Public Policy and employment 166

Table 2.5B: Public Policy and wages 168

vii
List of Figures

Figure 1.1: Gender earnings gap in Thailand (1985-2005) 2

Figure 1.2: Labor force participation in Thailand by gender (1985-2005) 30

Figure 1.3: Female labor force participation by country (2005) 31

Figure 1.4A: Female labor force participation and employment by sector 39
(age:15-24)

Figure 1.4B: Female labor force participation and employment by sector 40
(age:25-34)

Figure 1.4C: Female labor force participation and employment by sector 41
(age:35-44)

Figure 1.4D: Female labor force participation and employment by sector 42
(age:45-54)

Figure 1.4E: Female labor force participation and employment by sector 43
(age:55-64)

Figure 1.5: Earnings of male and female workers (1985-2005) 50

Figure 1.6: Hourly wages of male and female workers (1985-2005) 50

Figure 1.7: Gender (hourly) wage gap in Thailand (1985-2005) 52

Figure 1.8: Gender earnings gap by percentile 52

Figure 1.9A: Earnings of male and female workers and gender earnings gap 53
(age: 15-40)

Figure 1.9B: Earnings of male and female workers and gender earnings gap 54
(age: 41-65)

Figure 1.10A: Earnings density estimation for male workers (1985-2005) 57

Figure 1.10B: Earnings density estimation for female workers (1985-2005) 58

Figure 1.11: Earnings density comparison (1985 VS 1995 VS 2005) 59

Figure 1.12: Earnings density comparison (male VS female) 61

viii

Figure 1.13: Hourly wage density comparison (male VS female) 62

Figure 1.14A: Earnings density comparison (male VS female, age: 15-40) 63

Figure 1.14B: Earnings density comparison (male VS female, age: 41-65) 64

Figure 1.15: Relationship between Blinder-Oaxaca (1973) and DiNardo-Fortin- 94
Lemieux (1996) when male wage structure is used as reference wage structure

Figure 1.16: Relationship between Blinder-Oaxaca (1973) and DiNardo-Fortin- 96
Lemieux (1996) when pooled wage structure is used as reference wage structure

Figure 1.17A: Modified DiNardo-Fortin-Lemieux (1996) results 102
(without occupations and industries)

Figure 1.17B: Modified DiNardo-Fortin-Lemieux (1996) results 107
(with occupations and industries)

Figure 2.1: Firing costs across economies (2007) 114

Figure 2.2: Firing costs across OECD countries (2007) 115

Figure 2.3: Number of states adopting Wrongful Discharge Laws 122

Figure 2.4: Pattern of adoption during 1983-1994 123

Figure 2.5: Theoretical framework 130

Figure 2.6: Employment per population for high-skilled and low-skilled labor 153
(categorized using any-training index) in states adopting the Good Faith exception

ix
Abstract

My dissertation consists of two essays on labor and development economics. The
first essay seeks to identify the main factors that contributed to the decline in gender
earnings gap in Thailand’s wage and salary sector from 1985-2005. Two parametric
methodologies, Neumark’s version of the Blinder-Oaxaca method and the Juhn-Murphy-
Pierce method, are implemented in order to decompose gender earnings gap at a point in
time and across time period. I also make a methodological contribution by proposing a
way to modify the DiNardo-Fortin-Lemieux nonparametric decomposition method so
that its results are comparable to those from Neumark’s version of the Blinder-Oaxaca
method. The key findings of this essay are as follows. First, I find that increases in female
education and changes in unobserved factors, which were concurrent with modernization,
were the main sources of the decline in gender earnings gap. Second, over time,
improvements in the education of females in this sector surpassed that of males.
However, the superior education of females did not result in higher female earnings
because of the overwhelming effect of the unexplained factors that supported higher male
earnings. Finally, the nonparametric investigation corroborated the results from the
parametric methodologies.
The second essay investigates how the Wrongful Discharge Laws (WDLs),
imposed during the 1970s and 1980s, affect workers in the United States. Most
economists conjecture that WDLs increase firing costs for firms. In terms of employment,
the literature found a negative or at best zero impact. In terms of wages, most papers
found no impact. Thus the laws seemed to adversely affect an “average” worker. These

x
papers implicitly assumed that labor force was homogeneous. They did not recognize the
fact that labor can be heterogeneous and that firms may treat different types of labor as
different forms of input. My study attempts to overcome this limitation. I treat labor as
heterogeneous (high-skilled and low-skilled) thus allowing the laws to affect firms’
decisions regarding not only the quantity of labor input but also the combination of
different types of labor input. The key finding of this essay is that WDLs are associated
with increases in employment of high-skilled labor, a result unacknowledged in early
studies. WDLs, however, adversely affect employment of low-skilled labor, a result
consistent with the literature.

Chapter One

Superior Female Education:
Explaining the Gender Earnings Gap Trend in Thailand

1.1 Introduction
It is well-known and widely documented that Thailand has experienced a
remarkable increase in real income per capita during the past two decades. Regardless of
some setbacks during the late 1990s, from 1985-2005 the real per capita income more
than doubled.
1
Along with notable income growth, an impressive 35.1 percentage-point
reduction in poverty incidence was observed.
2
However, together with impressive socio-
economic development, many authors also reported an increase in overall income
inequality throughout the mid 1990s (Deolalikar, 2002; Motonishi, 2003; Warr, 2004;
Jeong, 2007). What is not documented in the literature is that, during the same time
period, the gender earnings inequality declined steeply regardless of the concurrent
increase in overall income inequality. Figure 1.1 shows that in 1985 an average male
worker earned 33.96% higher than an average female worker, whereas in 2005 an
average male worker earned only 8.98% higher than an average female worker.
The objective of this essay is to examine the decline in gender earnings inequality
in Thailand’s wage and salary sector from 1985-2005. Specifically, this paper seeks to
identify the main factors that contributed to the closing of this gender earnings gap,

1
Per capita GDP and CPI data from the bank of Thailand.
2
According to the headcount measure from Warr’s (2004) Table 1 (p. 4), in 1986, 44.9% of the population
were considered poor, however, in 2002, only 9.8% of the population were considered poor.
1

Source: author’s calculation from the Thai Labor Force Survey (Quarter 3)

Figure 1.1: Gender earnings gap in Thailand (1985-2005)

exploring whether these contributing factors were related to Thailand’s rapid
modernization and economic development during this time period.
The relationship between gender inequality and economic development has been a
controversial subject of interest in the literature. Xin Meng (1996) showed, using cross-
country data for selected Asian economies, that economic development and female
economic status, such as relative earnings of females compared to males did not have any
significant relationship. Specifically, Meng pointed out how economic inequality
according to gender was worse in richer countries like Japan and Korea than in poorer
countries. She concluded that the problem of gender inequality tended to stem from
social, political, and cultural structures rather than from economic development. Table
1.1 examines the ratio of female to male earnings for a broader range of countries. The
evidence seems to support the notion that economic prosperity cannot explain relative
2
Table 1.1: Ratio of female to male earned income* (selected countries)

Country F/M
Sweden 0.81
Norway 0.75
Cambodia 0.74
Denmark 0.73
Finland 0.71
Vietnam 0.71
Ghana 0.71
Australia 0.70
United Kingdom 0.65
Romania 0.65
France 0.64
Israel 0.64
Hungary 0.64
China 0.64
Canada 0.63
United States 0.62
Switzerland 0.61
Philippines 0.60
Ethiopia 0.60
Portugal 0.59
Poland 0.59
Thailand 0.59
Germany 0.58
Brazil 0.57
Greece 0.55
Argentina 0.53
Ukraine 0.53
Lao People's Dem. Rep. 0.52
Singapore 0.51
Czech Republic 0.51
Spain 0.50
Hong Kong, China (SAR) 0.49
Italy 0.46
Korea, Rep. of 0.46
Bangladesh 0.46
Indonesia 0.45
South Africa 0.45
Japan 0.44
Austria 0.44
Sri Lanka 0.42
Mexico 0.39
Iran, Islamic Rep. of 0.38
Kuwait 0.37
Malaysia 0.36
Turkey 0.35
India 0.31
Jordan 0.30
Pakistan 0.29
United Arab Emirates 0.24
Egypt 0.23
Saudi Arabia 0.15
Source: UNDP Human Development Report 2006

*Estimates are based on data for the most recent year available during 1991-2003
3
earnings of females to males. Although we can see from the table that wealthy
Scandinavian countries rank high on the list and that most Islamic countries rank low on
the list, we cannot draw useful conclusions regarding the rest of the countries. For
example, poorer countries like Cambodia, Vietnam, and Ghana rank very high in terms of
relative earnings of females to males. However, for richer countries like Italy, Korea,
Japan, and Austria, an average woman barely earns half of what an average man does.
Thus, within this context, income per capita and gender inequality do not seem to be
correlated.
However, within the boundaries of a national economy, growth and gender
equality have been seen as somehow positively related. Either growth leads to gender
equality, gender equality leads to growth, or both occur concurrently (United Nations
Development Programme [UNDP], 1995; World Bank, 2005). Growth can bring
prosperity to a country, it can create economic opportunities for women in terms of jobs
and education, and it can lead to women having more bargaining power within and
outside the households. Thus, women are able to raise social awareness about how they
should be treated as equals to men. The literature also argues that gender inequality can
exacerbate social, political, and cultural conflicts and thus obstruct economic
development, depressing the overall well-being of the population. Thus, efforts have been
made on the part of social and political organizations to raise awareness about the
importance of promoting gender equality.
The above evidence suggests that, in order to examine the issue of gender
inequality, one needs to look past the cross-country framework and investigate each
economy in and of itself. Gender status is deeply rooted in the socio-economic, political,
4
and cultural matrices of individual countries. Each country is unique. Thus, the results of
gender analysis are specific to specific countries.
Thailand is the country of interest in this paper. It is a good subject for case study
since it is a developing East Asian country
3
that has recently undergone modernization. In
the past, due to traditional beliefs, Thailand had a male-dominated social structure. The
unequal status of males and females, in terms of access to education and decent job
opportunities, could be observed. A gender gap in school enrollment was evident. A gap
in gender earnings was also apparent. However, many of these inequalities have faded
with the advent of modernization in Thailand. The roles of women and social attitudes
towards them have changed. The gender gap in the schooling of boys and girls has
virtually closed (Knodel, 1997). Also, as mentioned earlier, from 1985-2005, the gender
earnings gap, which was once considerable, has declined significantly. This paper will
analyze Thailand in terms of the relationship between factors of modernization and
gender earnings inequality.
In order to investigate the issue in question, the following methodologies are
implemented. First, Neumark’s (1988) version of the well-known parametric
decomposition proposed by Blinder (1973) and Oaxaca (1973) (hereafter, BO) is applied.
The BO decomposition helps identify, at any point in time, how much of the mean
earnings gap is caused by the differences in the observable characteristics of the two
genders (endowment gap) and by the differences in the pay structures faced by the two

3
According to the World Bank’s website, Thailand is categorized as a country residing in East Asia and
Pacific region. Other countries categorized to be in this region are Cambodia, China, Fiji, Indonesia,
Kiribati, Korea, Lao PDR, Malaysia, Marshall Islands, FS Micronesia, Mongolia, Palau, Papua New
Guinea, the Philippines, Samoa, Solomon Islands, Timor-Leste, Tonga, Vanuatu, and Vietnam.
5
genders (residual gap). Second, the Juhn, Murphy, and Pierce (1991) (hereafter, JMP)
decomposition is utilized to analyze the change in the earnings gap across time
parametrically. Over a period of time, the JMP method can distinguish whether the
increase or the decline in the overall gender earnings gap is due to [1] changes in the gap
of the observable characteristics between genders, [2] changes in the gap of the
unobservable characteristics between genders, [3] changes in the market returns to
observable characteristics, or [4] changes in the market returns to unobservable
characteristics. In this paper, I sometimes refer to the BO method as the time-point
analysis and the JMP method as the across-time analysis.
4
Third, a modified version of
the nonparametric decomposition proposed by DiNardo, Fortin, and Lemieux (1996)
(hereafter, DFL) is implemented. The DFL approach provides a full visualization of how
the differences in the entire earnings distributions of males and females can be
decomposed into two parts. The first part reflects the contributions of the difference in
the observable characteristics, while the second part reflects contributions of the different
pay mechanisms faced by males and females. I modify the standard DFL method to allow
the use of a more general form of the reference wage structure. This modification is
intended to make the DFL analysis comparable to Neumark’s (1988) version of the BO
analysis.
Although the issue of gender inequality has been widely discussed in
industrialized countries, rigorous empirical work using micro-datasets has been only
partially implemented in developing countries. In Thailand, hardly any papers have done

4
The BO method analyzes the earnings gap at a point in time, whereas the JMP method analyzes the
earnings gap across two time points. (Zveglich, Rodgers & Rodgers, 1997 referred to the BO method as the
level analysis and the JMP method as the trend analysis.)
6
rigorous empirical analyses regarding gender issues. This paper attempts to satisfy this
need by utilizing the Thai Labor Force Survey (Thai LFS), a large national micro-dataset
on demographic status and labor earnings of Thai workers, to investigate intensively
gender inequality in Thailand. In addition to applying existing methodologies to Thai
LFS data, I also make a methodological contribution by proposing a way to modify the
DFL method so that the results are comparable to those from Neumark’s (1988) version
of the BO method.
The key findings of the paper are as follows. First, I find that increases in female
education and reductions in the residual gap (difference in the unobservable
characteristics), which were concurrent with modernization, were the main sources of the
decline in male-female gender earnings inequality in Thailand’s wage and salary sector
from 1985-2005. Second, over time, improvements in the education of female workers in
this sector surpassed that of male workers. However, the superior education of females
did not result in higher female earnings because of the overwhelming effect of the
unexplained attributes that supported higher male earnings. Finally, when the analysis
was extended to account for the entire earnings distributions instead of just the mean
earnings, the nonparametric investigation (DFL) corroborated the results from the
parametric methodologies (BO and JMP).
The structure of the paper is organized as follows. Section 1.2 gives background
information regarding growth, poverty, and income inequality in Thailand. It also
discusses how education and gender gap in schooling have evolved during the economic
transition. It then touches upon gender inequality situations in other countries in the
region. Section 1.3 describes the Thai Labor Force Survey (Thai LFS), which is the main
7
dataset used in this study. Section 1.4 examines the Thai labor market, the gender
earnings inequality trends, and the changes in the earnings distributions of males and
females over the period of study (1985-2005). Section 1.5 describes and implements the
parametric methodologies (BO and JMP). Section 1.6 explains the relationship between
the parametric and the nonparametric decomposition methodologies (BO and DFL). The
section introduces my proposed modification to the standard DFL model. The modified
model is implemented and its results are discussed. Section 1.7 concludes the paper.
1.2 Socio-economic Background
In order to explore extensively the topic of gender earnings inequality in
Thailand, it is vital to comprehend the socio-economic background of Thailand and its
relationship to the issue in question. This section provides insights, regarding the Thai
economy. It elaborates the literature on GDP growth, poverty, and income inequality as
they occurred during periods of “miracle” growth, financial crisis, and economic
recovery. It then explores the roles of education and how education has expanded during
Thailand’s modernization. Education for women and the closing of the gender gap in
schooling in Thailand will be discussed. Finally, this section examines the existence and
the evolution of gender earnings inequality in neighboring countries. An elaboration of
Thailand’s labor market and a discussion of gender inequality in Thailand deserve a
separate section. These topics will be investigated in Section 1.4 after the main data
source (Thai Labor Force Survey) is discussed in Section 1.3.
8
1.2.1 Growth, Poverty, and Income Inequality
As recently as a few decades ago, agriculture was the most crucial component of
the Thai economy. A majority of labor was employed in this sector. The share of
agricultural employment was as high as 71% in 1980. However, this phenomenon has
changed dramatically over the past few decades. A majority of the labor force moved
towards manufacturing and service sectors, leaving only 39% employed in the
agricultural sector as of 2005.
5

Such a transition from agricultural employment to non-agricultural employment
was documented as the main source of poverty reduction during the “miracle” era prior to
the Asian financial crisis in 1997 (Jeong, 2007). Manufacturers benefited from the
abundance of cheap and low-skilled labor that had been transferred from the agricultural
sector. This inexpensive labor allowed the manufacturers to produce at low costs, giving
them an advantage in the export market. The growth in exports of labor-intensive
manufactured goods (footwear, textiles and garments) was considered one of the main
sources of overall growth for the Thai economy during that time. GDP growth was
remarkably high, averaging 8.8% per year during 1985-1996.
Many authors reported, using quantitative analyses, a positive relationship
between the rate of GDP growth and the rate of poverty reduction in Thailand during this
miraculous period. According to the headcount index of poverty measure, the proportion
of poor people in Thailand was 35.5% in 1981, declining significantly to 11.4% in 1996
(Warr, 2004). It can also be confirmed by any measure of poverty within the Foster-

5
The National Statistical Office’s calculation based on the Thai Labor Force Survey data.
9
Greer-Thorbecke family that the poverty reduction during this time period was robust
(Deolalikar, 2002; Jeong, 2007).
In spite of this tremendous GDP growth and remarkable poverty reduction,
income inequality among Thai households worsened. It has been pointed out that this
improvement in income and reduction in poverty occurred unevenly across the various
regions within Thailand (Deolalikar, 2002). Not surprisingly, the richest regions, such as
the Bangkok metropolitan area, experienced the highest reduction in poverty, while the
poorest regions, such as the Northeastern area, experienced the lowest reduction in
poverty. The decade of the 1990s also marked the rapid rise of the affluent middle class
population in Bangkok. These middle class people were not necessarily the elites but
were educated individuals from a variety of different socio-economic backgrounds
(Funatsu & Kagoya, 2003). These people gained their class status from the prestige of
their careers, from their growing wealth, and from their political connections. Their
knowledge and abilities allowed them to benefit a great deal from the growing economy.
Thus, they were the people in Bangkok who became wealthier during Thailand’s
industrialization.
This uneven development across regions, although crucial, was, however, not the
main source of Thailand’s increasing overall income inequality. According to Matonishi
(2003), the stimulus that underlay this surge in inequality stemmed from within each
region. Looking deeply into what economic factors actually caused the rise in inequality,
Jeong (2007) argued that the expansion of individuals’ access to credit and the increase in
education levels acquired by household heads were the main sources of this inequality.
10
After a long period of sustained growth, the Thai economy collapsed during mid-
1997. Many factors, such as flawed monetary policies, a lack of appropriate supervision
of financial institutions on the part of the central bank, reckless borrowing and investing
by private investors, and overconfident behavior by other participants in the market,
contributed to Thailand’s susceptibility to the financial crisis (Warr, 1999; Lauridsen,
1998; Tsurumi, 2000; Jansen, 2001). Negative export growth, observed in late 1996, also
contributed to the crisis.
Considering that Thailand’s previous “miracle” growth was largely due to
exports, it is not surprising that the negative export growth rate in 1996, which followed
years of positive and increasing export growth rates, signaled flaws in the economy.
6

Investors began to question the performance of the economy, leading to suspensions of
investment and speculation of a devaluation of the local currency. These speculative
attacks against the local currency in 1997 were often cited as the major cause of the
devastation of the Thai economy.
Starting at the end of 1995, one element that caused the slowdown of exports
appeared to be the appreciation of the Thai baht relative to the Japanese yen. At the time,
the Thai baht was tightly pegged to the US dollar. Thus, when the US dollar appreciated
against the Japanese yen, so did the Thai baht. This appreciation of the local currency
was detrimental to the export industry, since Japan was one of Thailand’s major
importers. Another important factor that caused Thai exports to lose competitiveness was
the concurring increase in the real wage rates of workers. Warr (1999) reported a
significant rise of real wages in the labor-intensive exporting sector during the 1990s. He

6
Warr (1999) argued that the negative export growth did not cause the crisis but triggered it.
11
also argued that Thailand used to maintain competitiveness in this market due to the
abundance of cheap unskilled labor. However, as the labor-intensive exporting sector
grew, cheap labor became scarce.
Thailand was the first country in East Asia to undergo the financial crisis. A large
number of businesses went bankrupt, the stock market crashed, several financial
institutions were closed, and many workers were laid off. The average GDP growth
during 1997-1998 dropped to -6%. The literature reported, however, only a moderate rise
in poverty incidence. Income inequality seemed to be stable regardless of the occurrence
of crisis.
Despite the severity of the financial crisis, Thailand managed to reform its
economy and recover from the financial turmoil. At the time of this writing, a decade has
passed since the crisis. The rate of recovery has been moderate yet steady. The economy
is generally considered to be in good shape. GDP growth has recovered and has remained
quite stable with an average annual growth rate of 4.9% (1999-2005).
7
Poverty incidence,
although slightly increased in the wake of the crisis, has decreased to a level comparable
to before the crisis and has continued to decline as of 2004. Likely, given the direction of
the trends, the level of poverty would have decreased to an even greater degree if there
had been no crisis. The levels of income inequality in Thailand have remained close to
those of the early 1990s (World Bank, 2006b).

7
Data from the Bank of Thailand.
12
1.2.2 Expansion of Education
It is indisputable that economic advancement and educational escalation are
interrelated. Once an economy has reached a certain benchmark of development, the
public, along with the policy makers, naturally turn their attention to the agenda of
promoting higher education. Conversely, expansion of education is also seen as a major
factor stimulating growth and thus generating economic advancement. The economy
benefits from better-educated workers since their superior skills and ability to adopt
newer technology allow them to excel with higher productivity. These educated workers
are also able to contribute more to the accumulation of human capital, leading to the
generation of even more able younger cohorts. The positive impact of education on
national economic development has been widely discussed in the literature. The benefits
of educating females, although not as elaborately examined, are of no less importance.
Besides the typical market benefits, educating females can also yield other positive
externalities such as enhancing the gains that result from educating males. However,
discrepancies in the schooling of boys and girls are still observed across the spectrum of
developing countries (Hill & King, 1991). In this section, I will explore the literature that
touches upon issues of female education and gender disparities in schooling. In East
Asian countries, traditional beliefs regarding gender roles are often to blame for such
disparities. However, this division in gender roles has abraded during modernization of
East Asia. As the country of interest, Thailand will be investigated in these respects. The
topics to be explored include the expansion of education, the closing of the gender gap in
schooling, and the ways in which the traditional views regarding gender roles have
altered during the transition periods of modernization.
13
The literature has emphasized how educating females can be beneficial.
Considering the labor market, Schultz (1991) demonstrated that in many East Asian
developing countries, such as Indonesia, Korea, Taiwan, and Thailand, the monetary
returns to female education were in fact higher than those for male education.
8
Besides
these superior market benefits, Hill and King (1991) demonstrated that educating females
also resulted in reduced fertility, better health, and improved living conditions for the
populace. These non-market benefits, although not measurable in terms of output or
income, have some positive impacts on other participants in the labor markets that allow
these participants to operate in a more efficient manner. Thus, investment in female
education has been empirically shown to be worthwhile, regardless of the women’s
decision to enter the labor market.
However, Hill and King (1991) observed a significant amount of discrepancies in
educational attainments for boys and girls in developing countries during 1960-1988. The
evidence pointed to how each of these countries failed to recognize the importance of
educating females and therefore under-invested in their education. It is interesting to look
into the causes of the underinvestment. Generally, decisions to invest in children’s
education belong to the parents. The literature pointed out how various factors could
hinder parents’ decisions to adequately invest in the education of girls. Most of the costs
of educating children were provided privately by the family in the form of tuition and
forgone labor. These costs could be easily measured in monetary terms. On the other
hand, the benefits of educating girls were mostly in the form of public benefits that might

8
The results from analyzing the Socio-economic Survey dataset (1976, 1981, and 1986) were also shown
to be robust for Thailand when the selection for participation in the labor force has been accurately adjusted
for.
14
be non-pecuniary. The benefit-cost analysis, according to this scenario, would make
investment in female education costly for families (Hill & King, 1991). Future job
opportunities and the expected income of females also play a role. The evidence is clear
(see Table 1.1) that in most economies, men on average earned a higher income than
women in the labor market. There are two ways to explain this fact. One way is to
understand how the foreseen disparities in the labor market, such as females earning less
than males, or females clustering in lower-paying industries, influence parents to invest
more in the education of boys rather than girls. Parents may also be more eager to
persuade boys, rather than girls, to be diligent in their studies and to major in subjects that
will lead them to respectable careers. Thus, future expectations influence parents’
decisions regarding educational investment in their children. A second way to explain
this gender inequality is to understand how traditional beliefs, such as those in East Asian
cultures, push men and women into assuming different roles and responsibilities. Boys
are to become income earners and girls are to become care-takers. When boys become
adults, they earn income to support the family, whereas girls may get married and
become a member of another family. Thus, parents usually prefer sending boys to school
and encourage them to put a significant amount of effort into their studies. These
behaviors result in men having better opportunities and higher income than women when
they enter the labor market. These arguments show how traditional perceptions regarding
roles of males and females in society, particularly in East Asian countries, prevent girls
from receiving adequate investment in their education.

15
East Asian countries, although very different from one another in their historical
foundations, share a similar ground in terms of gender roles. The extent of gender
segregation, however, differs across countries. According to Cameron, Dowling and
Worswick (2001), this segregation of gender roles is considered to be moderate in
Thailand compared to other countries in the region. During the “miracle” boom, like
other East Asian countries, Thailand went through modernization and the level of
education within the population moderately increased. In Thailand, achieving higher
education not only yields higher economic gain but also allows the individuals to attain a
higher social status (Knodel, 1997). Thus, education has been a subject matter that has
received public attention. The Thai government and related organizations have made
significant attempts to encourage higher education for the country’s populace.
According to Knodel (1997) and Hawley (2004), the Thai government imposed
compulsory education of 6 school years in 1978. I believe that they might be referring to
the Primary Education Act (No. 5) B.E. 2521 which required children aged 8 years old to
enroll in primary education until they turned 15 or until they completed 6
th
grade
(primary education). In 1999, the National Education Act B.E. 2542 was imposed. The
act mandated children aged 7 years old to enroll in primary and secondary education until
they turned 16 or until they completed 9
th
grade (lower secondary education). According
to Hawley (2004), the average number of years of education achieved by members of the
population has increased rather significantly during Thailand’s modernization. In the late
1980s, the country reached near universal primary education (6
th
grade). Similar to other
East Asian countries, Thailand used to harbor the traditional preference of sending boys
to school. However, as modernization occurred, this traditional view eroded. According
16
to Knodel (1997), more and more Thai parents have expressed indifference as to whether
to send boys or girls to school. These parents revealed that with limited resources in the
family, the most intelligent children should be allowed to pursue further schooling,
regardless of their gender. Asian Development Bank [ADB] (1998) also documented the
attempts of the Thai government and of many other non-profit agencies to launch various
programs to support girls to pursue higher education. The purpose of these programs was
to prevent girls from entering into prostitution when they grew up. Accordingly, the
gender gap in the education of boys and girls, which used to be evident, has been seen to
be narrowing as a result of modernization (ADB, 1998; Schultz, 1991). In fact, Knodel
(1997), using the 1990 Thai Census data, documented the closing of the gender gap in
educational attainment for the primary and secondary levels.
1.2.3 Gender Earnings Inequality in the Neighboring Countries: The
Literature
East Asian countries are economically and culturally interrelated. In order to
understand where Thailand stands in terms of gender inequality issues, it is interesting to
explore the extent of gender inequality in neighboring countries. Unlike in developed
Western countries, the literature regarding gender inequality in East Asia has yet to
emerge. The difficulty of gaining access to rigorous datasets for some East Asian
countries has been the major hindrance for scholars studying the subject matter. Only a
few studies have managed to analyze the gender earnings gap using available datasets.
Most of the analyses were done parametrically. The East Asian countries that will be
discussed here are Indonesia, Japan, Taiwan, Korea, China, and Vietnam.
17
Indonesia’s miracle growth, like that of Thailand, occurred in conjunction with a
massive transition of the workforce from the agricultural sector to the formal wage and
salary sector. An increase in real wages, a reduction in poverty, and an increase in
education levels occurred along with this impressive growth. Smith, Thomas,
Frankenberg, Beegle, and Teruel (2002) documented a doubling of female real hourly
earnings and a 50% increase in male real hourly earnings during 1987-1997. According
to Thomas, Beegle, and Frankenberg (2003), the 1997 crisis did not affect employment so
much as it caused a large drop in real earnings. As for the matter of male-female
inequality in the labor market, Pirmana (2006), using the SAKERNAS data (Survei
Tenaga Kerja Nasional or the Indonesian Labor Force Survey), reported a slight decline
in the gender earnings gap during 1996-2004. More specifically, in 1996, the average
male earned 32.64% higher than the average female. However, in 2004, the average male
earned only 23.34% higher than the average female. The World Bank (2006a)
documented the Indonesian government’s imposition of labor laws to enforce equal pay
for equal work. However, violations of these laws were still reported to be prevalent. The
article also reported a narrowing of the gender gap in school enrollment. Using the
SUSENAS 2002 data (Survei Sosial Ekonomi Nasional or the Indonesian National Socio-
Economic Household Survey), World Bank (2006a) found that at the primary and lower
secondary levels, the number of girls enrolled was greater than the number of boys
enrolled. However, as reported by Pirmana (2006), the gender earnings gap still existed.
Pirmana’s decomposition showed that the majority of the gender earnings gap was
accounted for, not by the observable characteristic differences, but by differences in how
the skills of males and females were compensated.
18
In Japan, labor laws prohibit gender discrimination in the labor market. However,
evidence of males earning more than females was prevalent (Daly, Kawaguchi, Meng &
Mumford, 2006). Kawaguchi (2004) documented a narrowing of the gender earnings gap
during 1990-2000. In 1990, the average male earned approximately 48.6% higher than
the average female. The gap decreased to 39.2% in 2000. The decomposition results
showed that the reduction of the returns to job tenure was one of the main contributors to
the decline in the gender earnings gap. Generally the job tenure of men was longer than
that of women. Thus, the lower returns to job tenure reduced the advantage that men had
and narrowed the gap. The author reasoned that the transition from a seniority-based to a
result-based compensation system in Japan during the 1990s was the main cause of the
decline in the returns to job tenure.
Unlike other countries previously discussed, Taiwan experienced a stagnant
gender earnings gap during 1978-1992. Zveglich, Rogers, and Rogers (1997), using
Taiwan’s Manpower Utilization Survey, reported the total earning gaps to have slightly
fluctuated around 0.43 log-points throughout 1978-1992. That is, the average male earned
43% higher than the average female. According to the time-point decomposition analysis,
the authors reported that about half of the gap was explainable by the observable
characteristic differences between the genders. Estimates of the contribution of the
explained gap (62.6% in 1978 and 42.3% in 1992) were close to the estimates given by
Gannicott (1986). Besides a time-point decomposition analysis, Zveglich, Rogers, and
Rogers (1997) also carried out an across-time decomposition analysis. They observed the
improvement of Taiwanese female characteristics, especially in terms of education and
experience, throughout the period of study. However, the residual gap was also shown to
19
be growing, and its growth rate was substantial enough to offset totally the improvement
in female characteristics. The counteraction of the two opposing forces resulted in a
stagnant male-female earnings gap during the study period.
Similar to Taiwan, Korea also experienced a stagnant gender earnings gap.
However, Rodgers (1998), using Korea’s Occupational Wage Survey (OWS),
documented that the rigidity of Korea’s gender earnings gap lasted only until 1983. After
1983, the gap began to narrow. The average male earned 75% higher than the average
female in 1983, but the average male earned only 55.7% higher than the average female
in 1992. At any given point in time within the study period, the decomposition showed
that the majority of the gap could be explained by the observed differences between
genders (ranging from 66.8% to 83.8%). Across time, females’ observed characteristics
were initially inferior to those of males’, however, they improved significantly
(especially in terms of education) during 1983-1992. These improvements were the major
source of the narrowing of the total earnings gap from 1983-1992. Like Taiwan, Korea
also experienced a growing residual earnings gap during this period. However, the
improvement in observed characteristics of females during 1983-1992 was large enough
to outweigh the increase in the unobserved differences between genders, resulting in an
overall decline in the gap.
Since the late 1970s, China, the most populous country in the world, has been
undergoing a process of major reformation from a centralized economy to a market
economy. According to Liu, Meng, and Zhang (2000), China’s major economic reforms
started in 1978. These reforms allowed most firms to gain some flexibility regarding the
production process in terms of output and technology. It was not until 1992, however,
20
when an updated agenda was ratified, that employers were given more flexibility in
dealing with workers in terms of pay, employment, and promotions. Gustafsson and Li
(2000), using data from the Urban Household Income Survey, documented a slight
increase in the gender earnings gap during 1988-1995. The gender gap, indicating how
much the average male earned compared to the average female, stood at 15.6% in 1988
and 17.5% in 1995. For 1988-1995, the time-point decomposition analysis showed that
slightly more than half of the gender gap was accounted for by differences in the market
returns to skills of males and females. Across time, the growing residual gap was the
major cause of the worsening of the overall gender earnings gap throughout the period.
The authors argued that the rising residual gap could be explained by Chinese economic
reforms. Prior to these reforms, under the central planning regime, Chinese workers, both
male and female, were assigned jobs and earnings by government agencies. The
centralized regime called for equality for both male and female workers. Thus, when the
system became decentralized, workers were hired and compensated according to how
their characteristics were valued by the employers. Within a patriarchal society such as
China, female workers might not have been as equally preferred as male workers. It is
also possible that some characteristics of females that were not observed in the data
prevented females from earning as much as males.
In Vietnam, Liu (2004), using the Vietnam Living Standard Surveys (1992/1993
and 1997/1998), reported a slight decline in the gender earnings gap from 1993-1998.
The size of the gap in terms of log differentials was 0.257 in 1993 and 0.194 in 1998.
That is, the average male earned 25.7% and 19.4% higher than the average female in
1993 and 1998, respectively. The decomposition at each time point showed that the
21
majority of the gap was accounted for by the unequal rates of returns to the workers’
skills. In 1998, the estimated endowment gap was negative, indicating that female
workers had more favorable observable characteristics than male workers. However, the
unexplained portion of the total gap was larger, resulting in males earning more than
females. Across time, the decomposition results showed that the residual gap had
widened. The author pointed out how the growing residual gap could have stemmed from
the erosion of the centralized system, as in China. Once employers faced lower
limitations in their ability to determine wages, their attitudes towards certain
unobservable characteristics became important. It is possible that these preferences may
have worked in a way that was not favorable for females. The World Bank (2006c)
reported the existence in Vietnam of cultural beliefs regarding gender stereotypes. The
study reported how illustrations in school materials put fixed ideas into children as to
how each gender should have different roles in the household and workplace. Regardless
of the growth of the residual gap, the data demonstrated a decline of 6.3% in the overall
gender earnings gap. This is because female observable characteristics were improving so
quickly that they outweighed the opposite effects from the growing residual gap.
1.3 Data
The main dataset used in this paper is from the Thai Labor Force Survey (Thai
LFS). The Thai LFS has been conducted by the Thai National Statistical Office (NSO)
since 1963. Currently, actual raw data are available only from 1985, whereas from 1963
22
to 1984, only aggregate statistics are available.
9
The main purpose of the Thai LFS is to
assess the labor force characteristics of the country. The households covered are private
households and special households (persons living in groups or in quarters within the
compound of a factory). Institutional households (jails, college or school dormitories, and
military bases) are excluded.
Before 1994, the number of households in the sample size was 27,780
(approximately 84,000 persons). From 1994 to 2000, the sample size was 60,492
(approximately 170,000 persons). From 2001 onwards, the sample size increased to
79,560 (approximately 200,000 persons). From 1971-1983, two rounds of surveys were
conducted, namely, Q1 (dry or non-agricultural season) during January-March, and Q3
(wet or agricultural season) during July-September. From 1984-1997, a third round, Q2,
was established. Q2 was conducted during April-June, the period in which a reasonably
large group of new workers enters the labor force right after graduating from school. In
1998, the fourth and final round, Q4, was established. This was conducted from October-
December. Since 2001, the survey has been administered monthly, with the data grouped
into quarters. The data structure was initially constructed as a repeated cross-section, but
since 2001, a rotating panel structure was used.

9
See Appendix Table A.1 for data availability during 1985-2005. In some years, the data were missing for
some quarters due to insufficient resources (staff and budget), since the majority of resources were
allocated for the collection and compilation of census data. This was true for the year 1990, 1993, 1995,
and 1997. As for the reason why currently the raw data are available only from 1985, I have asked quite a
few people at the NSO, but no one really had an answer.
23
A two-stage sampling methodology has been utilized in the survey process.
Thailand is comprised of 76 provinces, and each province is divided into municipal areas
and non-municipal areas.
10
The primary sampling units in the first stage are blocks for
municipal areas and villages for non-municipal areas. The number of blocks and villages
sampled is determined by employing the probability proportional to size scheme where
size is the total number of households in the areas. In the second stage, private
households are the ultimate sampling units. Fifteen households are chosen for each
municipal area and twelve households are chosen for each non-municipal area. Once
sampled, the head of each household is interviewed.
The survey collects information regarding the individuals’ employment situation
and demographic background. Previously, questionnaires came in two types: short form
and long form. The short form included questions relating to demographic, education,
and employment (age, gender, marital status, highest education level, location, type of
workers, occupation, industry, hours worked, etc.). In addition to the questions asked in
the short form, the long form also included income and migration questions
(compensation type, wage rate, previous location, etc.). The long form was used in Q1
and Q3, while the short form was used in Q2 and Q4. Since 1999, the long form has been
used for all quarters. However, the survey questions were not always consistent. In some
quarters, additional questions were added. These questions included information
regarding individual literacy, activities during the non-agricultural season, smoking

10
Initially, Thailand was comprised of 73 provinces. In 1994, 3 new provinces, Nong Bua Lam Phu, Am
Nat Charoen, and Sra Kaeo were added. These new provinces were formerly parts of other existing
provinces.
24
habits, computer usage, etc. Unfortunately, some of the information was removed from
the Thai LFS dataset and assembled separately as other dataset projects.
11

Previously, in regards to Thai LFS documentation, individuals age 13 and older
were classified according to whether they were in the total labor force or not.
12
The total
labor force is composed of the current labor force and of seasonally inactive laborers,
with the current labor force comprised of employed and unemployed individuals. In
1998, the government issued the Labor Protection Act B.E. 2541, which stated that the
legal minimum age for child labor was 15 years of age (previously, the legal age was
13).
13
Thus, in 2001, the NSO re-defined the labor force population as anyone age 15 or
older. An individual is defined as employed if he or she is working for pay; not working
but receiving pay or holding a job that he or she will be returning to; or working without
pay. Employed workers are classified as private employees, government employees,
employers, self-employed workers, family business workers, and state-enterprise
employees.
14,15
Of these categories, private employees, government employees, and state-
enterprise employees are considered wage and salary workers.

11
The original questionnaires are available from the NSO in the Thai language (scanned PDF) for every
quarter of every year. In some quarters or in some years, I found supplemental questions in the
questionnaires, but the actual data fields were left blank. I was informed that some of them had become
parts of other projects.
12
National Statistical Office of Thailand [NSO] (2003a and 2003b).
13
See Jeraputtiruk (2004) and Garen and Jeraputtiruk (2005) for discussions regarding the Labor Protection
Act B.E. 2541 (1998) and analyses regarding the effects of the act on the Thai labor market.
14
The difference between employer and self-employed worker was not explained explicitly in the
questionnaire. Usually for a family that owns either a business or a farm will have the oldest employed
member reported himself as either an employer or a self-employed worker and have the rest of the family
members (who work) reported themselves as family business workers.
15
The co-op worker (cooperative) category was added in 2001.
25
In 2001, there was a major change in the Thai LFS questionnaires. Instead of
quarterly, the survey was conducted monthly. The repeated cross-section structure was
replaced by the rotating panel structure. Industry, occupational, and educational codes
were modified.
16
Actual hours worked were recorded differently. Prior to 2001, if an
individual had a permanent job but was not working during the survey week, the usual
hours worked was recorded. However, from 2001 onwards, zero was recorded for such a
case. The structure of the questions relating to income was also modified. In the surveys
conducted prior to 2001, all wage and salary workers were asked to identify their
compensation types as hourly, daily, weekly, monthly, other, or unknown. If the worker
reported one of the first four categories, then he or she would be asked to report his or her
wage accordingly. If he or she categorized the compensation type as “other”, then he or
she would be asked to report the average daily wage. If he or she answered that the
compensation type was “unknown”, then he or she would be asked to select a salary
range from a set of pre-defined salary ranges. From 2001 onwards, this set of questions
was altered. Specifically, hourly, daily, and weekly workers were asked to report their
wages accordingly. All wage and salary workers, including hourly, daily, and weekly
workers, were also asked to report their average monthly earnings.
In this study, I focus the analyses on full-time
17
wage and salary workers aged
between 15 and 65 years old. The Q3 Thai LFS data from 1985 to 2005 are utilized.
18

16
I attempted to make the educational categories compatible for the data before and after 2001 in order to
estimate statistics relating to the data. The industry codes can be converted with only trivial errors, but the
occupational codes cannot be converted. I consulted this with an NSO officer who was in charge of the
Thai LFS data project.
17
Full-time workers are workers who work 35 hours or more per week.
18
Q3 is the only quarter with income information that is available for every year from 1985-2005.
26
Since most workers are paid monthly and from 2001 onwards every worker has been
asked to report their average monthly earnings, the analyses in this paper are based on
monthly earnings. Other types of reported earnings are converted to monthly earnings
according to Appendix Table A.2.
19
The earnings are deflated using the year 2002 as the
base year.
It is reasonable to argue that hourly wages may constitute a more appropriate
measure of income compared to monthly earnings. Monthly earnings can be affected by
the number of hours each worker decides to work and one may expect male workers to
work more hours than female workers. Table 1.2 compares average hours worked per
week for male and female workers in Thailand’s wage and salary sector.
20
The table
shows that hour differences between the two genders are trivial and thus should not cause
any concern. For most years, less than 1% of the workers reported their hourly wages.
From 1985 to 2000, about 50% of the workers reported their monthly earnings and from
2001 onwards all workers reported their monthly earnings. By using monthly earnings, I
introduce minimum inaccuracies associated with the unit conversion process. I will also
show in the next section that the gender gap in income, in terms of mean gap and density
gap, estimated using monthly earnings and hourly wages are very similar to each other.

19
Very few outliers and top-coded observations were dropped.
20
Note that from 1985-2000, usual hours worked are recorded. From 2001 onwards, actual hours worked
are recorded.
27
Table 1.2: Average hours worked per week for male and female workers (wage and salary sector)

Average hours worked per week
Year
Male Female
1985 50.03 50.24
1986 49.59 50.05
1987 49.35 49.85
1988 50.56 50.21
1989 50.59 50.52
1990 50.34 50.58
1991 50.53 50.20
1992 50.50 50.15
1993 50.27 49.62
1994 50.46 49.70
1995 50.31 49.59
1996 50.85 49.73
1997 49.94 49.13
1998 49.18 48.39
1999 48.99 48.40
2000 49.48 48.65
2001 49.10 48.72
2002 48.95 48.56
2003 49.48 48.78
2004 49.76 48.87
2005 49.54 48.80

1.4 The Thai Labor Market and Gender Inequality
This section describes the Thai labor market and discusses various issues
regarding gender inequality in the market. Statistics and figures presented in this section
(and beyond) are derived from the Thai LFS dataset, unless otherwise noted. This section
is organized as follows. First, the labor force participation of both genders is explored. A
discussion regarding why the participation rate of females in Thailand has been
substantial is conducted. Next, the segregation of the labor market into a formal sector
(wage and salary) and an informal sector (non-wage and salary) is discussed. Since the
main analyses of the paper will focus on workers in the formal sector, the overall
28
characteristics of male and female workers in this sector are thoroughly investigated. I
then explore the situations of female workers and the formal legislation that protects
female workers’ rights. Finally, the average gender earnings gap and the earnings density
gap between male and female workers are carefully explored. I address how these gaps
evolved over time and how they evolved differently for the younger and the older
workers.
The labor force participation rate
21
of Thai workers segregated by gender is
shown in Figure 1.2. The male labor force participation was 89% in 1985. It declined
during the 1990s and has stayed at 85% since 2000. For females, the labor force
participation was about 77% in 1985, increasing slightly during the late 1980s, but
diminishing during the 1990s and stabilizing at around 70% since 2000. For those who
are familiar with the levels of female labor force participation in other countries, it is not
difficult to perceive that the participation rate for Thai females has for many years been
rather large compared to that of other countries. The high participation rate of females
still prevails in Thailand. Figure 1.3 compares the female labor force participation for
selected countries using data from the International Labor Organization.
22
The figure
confirms that Thailand falls into the category of countries with high female labor force
participation rate.

21
Total labor force divided by population.
22
Note that the data collection process and the definition of labor force may not be fully consistent across
countries. However, the figure shows roughly how female labor force participation rates compare across
countries.
29
35%
40%
45%
50%
55%
60%
65%
70%
75%
80%
85%
90%
95%
100%
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
Year
Percentage
Male Female

Figure 1.2: Labor force participation in Thailand by gender (1985-2005)

Typically, age, education, expected earnings, number of children, family wealth,
unearned household income, and the income of husbands influence women’s choices as
to whether they will participate in the labor force or not. Besides these factors, societal
norms and cultural practices also have an impact. According to Phananiramai (1993) and
ADB (1998), in the past, the corvee labor system under Thailand’s traditional monarchy
required peasant males of prime age to serve the monarch and noble families for a period
of time each year. Thus, these men had to be absent from their homes for sporadic
periods of time. The system implicitly compelled women to take full charge of the
household while the men were gone. Besides typical domestic chores, women had to
30
0
10
20
30
40
50
60
70
80
Saudi Arabia
Egypt
Turkey
Pakistan
India
Sri Lanka
Italy
United Arab Emirates
Malaysia
France
Japan
Austria
Korea, Republic of
Singapore
Germany
Indonesia
Bangladesh
Hong Kong, China
Lao People's Dem. Rep.
Philippines
United Kingdom
Portugal
Finland
Sweden
Denmark
United States
Switzerland
Canada
Norway
Thailand
Myanmar
China
Ghana
Viet Nam
Cambodia
Country
Percentage

Source: International Labor Organization LABORSTA

Figure 1.3: Female labor force participation by country (2005) 31
work (mostly in family farming or other household production) in order to provide food
and supplies for their family. This was probably the reason why a high proportion of Thai
women worked during pre-modern times. The tradition of working females persists into
the modern day. Currently, the corvee labor system no longer exists, yet the participation
of females in the labor force remains high.
Like most developing countries, the Thai labor market is segmented according to
employment laws and social policies. The formal sector (wage and salary), comprised of
private, government, and state-enterprise employees, gains the benefit of labor protection
laws (severance pay requirements, minimum wage legislations, etc.) and social insurance
(social security benefits, unemployment insurance, etc.).
23
However, self-employed and
family-business workers, who constitute the informal sector (non-wage and salary) are
not entitled to any employment protections or insurance.
Table 1.3 illustrates the structure of population and labor force in Thailand during
1985-2005. The population grew from 51.4 million to 64.9 million. The share of wage
and salary workers
24
was approximately 26.51% in 1985. The share, which grew to
39.85% in 1996, the year before the Asian financial crisis, dropped modestly during the
crisis, and expanded to 48.45% by 2005. Thus, the share of wage and salary workers
almost doubled over the past 20 years. Table 1.3 also segregates Thai wage and salary
workers according to their compensation types. Most workers (roughly half) were paid
monthly, while the second most pervasive type of payment was daily.

23
Technically speaking, the national employment protection laws and social policies (such as the Labor
Protection Act and the Social Security Act) cover only private employees. However, government and state-
enterprise employees are entitled to other types of protections and insurances.
24
Wage and salary employment divided by overall full-time employment.
32
Table 1.3: Population and labor force structure in Thailand

1985 1986 1987 1988
Total Population 51450741.82 52227697.95 53679806.00 54599384.29
Employed Full-time Workers
(Age:15-65)
22798876.56 23874451.49 24233882.38 25935599.75

Employed Full-time Workers
(Age: 15-65)
Categorized by Work Status
Private Employee 4500786.02 19.74% 4828070.83 20.22% 5201569.48 21.46% 5667103.06 21.85%
Govt Employee 1542301.84 6.76% 1507814.16 6.32% 1559456.65 6.44% 1489165.92 5.74%
Employer 245062.99 1.07% 250030.99 1.05% 333971.89 1.38% 310529.48 1.20%
Self-Employed 6978027.79 30.61% 7191252.18 30.12% 7606140.79 31.39% 7499735.10 28.92%
Family Business 9532697.92 41.81% 9832534.33 41.18% 9227687.52 38.08% 10681983.95 41.19%
State-Enterprise
Employee
N/A 264749.00 1.11% 305056.05 1.26% 287082.24 1.11%
Cooperative N/A N/A N/A N/A

Wage & Salary Workers
(Priv. & Govt. & State-Ent.)
6043087.86 26.51% 6600633.99 27.65% 7066082.18 29.16% 7443351.22 28.70%

Wage Type*
Hourly 27576.83 0.46% 19088.41 0.29% 77001.10 1.09% 28027.72 0.38%
Daily 2110329.00 34.92% 2326934.00 35.25% 2518299.00 35.64% 2899798.00 38.96%
Weekly 21509.94 0.36% 47460.57 0.72% 24848.22 0.35% 28261.09 0.38%
Monthly 2983168.00 49.36% 3279378.00 49.68% 3493074.00 49.43% 3779889.00 50.78%
Other 701468.10 11.61% 673149.80 10.20% 714764.80 10.12% 570793.80 7.67%
Unknown 199036.00 3.29% 254623.50 3.86% 238095.20 3.37% 136581.50 1.83%
Non-Cash N/A N/A N/A N/A

33
Table 1.3, Continued

1989 1990 1991 1992
Total Population 55518320.05 56405036.44 57234550.69 57826735.50
Employed Full-time Workers
(Age:15-65)
27858205.19 26714652.14 27693617.36 29098418.11

Employed Full-time Workers
(Age: 15-65)
Categorized by Work Status
Private Employee 6044307.63 21.70% 6426346.18 24.06% 6962164.03 25.14% 7492848.50 25.75%
Govt Employee 1482639.68 5.32% 1486461.22 5.56% 1545762.20 5.58% 1692953.54 5.82%
Employer 441180.89 1.58% 348115.47 1.30% 602525.63 2.18% 702551.29 2.41%
Self-Employed 7948753.98 28.53% 8013450.70 30.00% 8127838.24 29.35% 8068954.11 27.73%
Family Business 11650947.29 41.82% 10132492.76 37.93% 10239857.06 36.98% 10818025.54 37.18%
State-Enterprise
Employee
290375.72 1.04% 307785.81 1.15% 215470.20 0.78% 323085.13 1.11%
Cooperative N/A N/A N/A N/A

Wage & Salary Workers
(Priv. & Govt. & State-Ent.)
7817323.03 28.06% 8220593.21 30.77% 8723396.43 31.50% 9508887.17 32.68%

Wage Type*
Hourly 49319.49 0.63% 21353.51 0.26% 27732.66 0.32% 20536.18 0.22%
Daily 2946398.00 37.69% 2970406.00 36.13% 3461656.00 39.68% 3715528.00 39.07%
Weekly 54986.98 0.70% 71609.85 0.87% 47696.66 0.55% 40291.60 0.42%
Monthly 3703584.00 47.38% 4161740.00 50.63% 4240458.00 48.61% 4840775.00 50.91%
Other 718638.40 9.19% 768699.70 9.35% 779163.90 8.93% 699936.80 7.36%
Unknown 344395.90 4.41% 226783.80 2.76% 166689.30 1.91% 191820.20 2.02%
Non-Cash N/A N/A N/A N/A

34
Table 1.3, Continued

1993 1994 1995 1996
Total Population 58649852.29 59444067.84 59450877.06 60045315.37
Employed Full-time Workers
(Age:15-65)
28970790.74 28945142.18 29468494.22 29150978.12

Employed Full-time Workers
(Age: 15-65)
Categorized by Work Status
Private Employee 8306163.58 28.67% 8266272.70 28.56% 8674675.25 29.44% 9350607.49 32.08%
Govt Employee 1792132.43 6.19% 1903445.06 6.58% 2023541.11 6.87% 1900954.04 6.52%
Employer 577521.09 1.99% 533954.33 1.84% 872357.84 2.96% 729561.33 2.50%
Self-Employed 8294685.82 28.63% 8285888.65 28.63% 8634690.35 29.30% 8634457.69 29.62%
Family Business 9638829.71 33.27% 9576458.05 33.08% 8906811.31 30.22% 8170505.87 28.03%
State-Enterprise
Employee
361458.11 1.25% 379123.39 1.31% 356418.36 1.21% 364891.70 1.25%
Cooperative N/A N/A N/A N/A

Wage & Salary Workers
(Priv. & Govt. & State-Ent.)
10459754.12 36.10% 10548841.15 36.44% 11054634.72 37.51% 11616453.23 39.85%

Wage Type*
Hourly 48953.64 0.47% 50702.77 0.48% 22111.63 0.20% 16350.97 0.14%
Daily 4106560.00 39.26% 4150118.00 39.34% 4346162.00 39.32% 4699158.00 40.45%
Weekly 64039.07 0.61% 44168.61 0.42% 52260.56 0.47% 49777.87 0.43%
Monthly 5388171.00 51.51% 5529597.00 52.42% 5869380.00 53.09% 6007180.00 51.71%
Other 702573.20 6.72% 632884.20 6.00% 549749.80 4.97% 550482.90 4.74%
Unknown 149457.10 1.43% 141370.30 1.34% 214970.40 1.94% 293503.80 2.53%
Non-Cash N/A N/A N/A N/A

35
Table 1.3, Continued

1997 1998 1999 2000
Total Population 60648992.07 61248436.30 61856729.57 62481449.80
Employed Full-time Workers
(Age:15-65)
27996656.15 28454645.32 28027807.92 29051603.95

Employed Full-time Workers
(Age: 15-65)
Categorized by Work Status
Private Employee 8767328.12 31.32% 8348431.63 29.34% 8659786.16 30.90% 9589063.35 33.01%
Govt Employee 1181709.83 4.22% 2163703.68 7.60% 2336240.18 8.34% 2312093.25 7.96%
Employer 620758.55 2.22% 745169.73 2.62% 826491.92 2.95% 967846.40 3.33%
Self-Employed 8438790.74 30.14% 8553170.62 30.06% 8621258.78 30.76% 8391111.20 28.88%
Family Business 8767023.68 31.31% 8171244.37 28.72% 7235474.71 25.82% 7432879.95 25.59%
State-Enterprise
Employee
221045.23 0.79% 472925.29 1.66% 348556.17 1.24% 358609.80 1.23%
Cooperative N/A N/A N/A N/A

Wage & Salary Workers
(Priv. & Govt. & State-Ent.)
10170083.18 36.33% 10985060.60 38.61% 11344582.51 40.48% 12259766.40 42.20%

Wage Type*
Hourly 18045.04 0.18% 9300.87 0.08% 10866.33 0.10% 15250.64 0.12%
Daily 4546204.00 44.70% 3890213.00 35.41% 4019275.00 35.43% 4876128.00 39.77%
Weekly 41030.44 0.40% 53690.41 0.49% 49599.89 0.44% 34199.40 0.28%
Monthly 4939530.00 48.57% 6340912.00 57.72% 6578901.00 57.99% 6499238.00 53.01%
Other 479194.40 4.71% 533211.50 4.85% 510814.70 4.50% 634345.80 5.17%
Unknown 146079.40 1.44% 157732.90 1.44% 175125.90 1.54% 200604.50 1.64%
Non-Cash N/A N/A N/A N/A
36
Table 1.3, Continued

2001 2002 2003 2004 2005
Total Population 63001139.61 63526907.85 64062601.75 65197160.37 64884044.79
Employed Full-time Workers
(Age:15-65)
27464303.25 28594375.08 28576728.67 29158400.80 28899392.58

Employed Full-time Workers
(Age: 15-65)
Categorized by Work Status
Private Employee 9326450.10 33.96% 9884502.42 34.57% 10215458.13 35.75% 11177799.47 38.33% 11052238.98 38.24%
Govt Employee 2273406.10 8.28% 2244450.67 7.85% 2204779.77 7.72% 2345120.26 8.04% 2607369.09 9.02%
Employer 789441.41 2.87% 938846.97 3.28% 960466.52 3.36% 890333.73 3.05% 906034.00 3.14%
Self-Employed 8094153.12 29.47% 8261501.35 28.89% 8247396.86 28.86% 8254129.02 28.31% 8267253.98 28.61%
Family Business 6625583.68 24.12% 6926547.55 24.22% 6622605.39 23.17% 6119494.25 20.99% 5684011.11 19.67%
State-Enterprise
Employee
341678.69 1.24% 327256.76 1.14% 304568.59 1.07% 326687.37 1.12% 343126.54 1.19%
Cooperative 13590.16 0.05% 11269.35 0.04% 21453.40 0.08% 44836.68 0.15% 39358.89 0.14%

Wage & Salary Workers
(Priv. & Govt. & State-Ent.)
11941534.89 43.48% 12456209.85 43.56% 12724806.49 44.53% 13849607.11 47.50% 14002734.61 48.45%

Wage Type*
Hourly 18591.27 0.16% 28633.99 0.23% 12286.08 0.10% 27531.41 0.20% 13676.01 0.10%
Daily 4354533.00 36.47% 4502654.00 36.15% 4481619.00 35.22% 5031700.00 36.33% 4958169.00 35.41%
Weekly 53086.13 0.44% 30863.37 0.25% 46587.28 0.37% 50067.13 0.36% 26562.89 0.19%
Monthly 6656624.00 55.74% 6978011.00 56.02% 7264706.00 57.09% 7622650.00 55.04% 7845205.00 56.03%
Other 828299.70 6.94% 873857.60 7.02% 892897.90 7.02% 1073341.00 7.75% 1119595.00 8.00%
Unknown 21568.88 0.18% 33696.76 0.27% 22495.74 0.18% 36988.57 0.27% 19442.74 0.14%
Non-Cash 8832.37 0.07% 8492.89 0.07% 4214.59 0.03% 7329.14 0.05% 20083.96 0.14%

37
According to the data presented, the wage and salary sector has been rapidly
expanding and has become an important part of the labor market. This sector contributes
greatly to the national GDP; thus, it has a considerable effect on the country’s economic
and social environments. It is also the sector where employment laws, social policies, and
international negotiations have a direct impact. Understanding what occurs in this market
sector can enhance our perception of how overall market circumstances evolve.
Therefore, I focus my analyses on the wage and salary sector in this paper.
Recall that in Figure 1.2, the labor force participation for both males and females
declined during the 1990s. Panel A of Figure 1.4A through 1.4E breaks down the labor
force participation of both genders into age groups (15-24, 25-34, 35-44, 45-54, and 55-
64). The figures show that the decline in the labor force participation rate was due to a
drop in the participation of the group of the youngest workers (age 15-24). The labor
force participation rates of other age groups were quite constant over time.
25
An
explanation for the reduction in the participation of the youngest workers is that levels of
education increased significantly during the period, with the youngest age group in
school rather than in the workforce. Thus, this group was the only one for which the
participation rate dropped. The gap between the male and the female labor force
participation rates for the oldest group (age 55-64) was the largest compared to other
groups. ADB (1998) documented that in certain industries female workers tended to
retire at an age lower than 60, typically becoming replaced by younger workers.

25
Note that the trends for overall employment rates are very similar to those for the labor force
participation rates. Thus, the overall employment rates are not shown here.
38

Panel A: Labor force participation (age: 15-24)

Panel B: Wage and salary employment (age: 15-24) Panel C: Non-wage and salary employment (age: 15-24)

Figure 1.4A: Labor force participation and employment by sector (age: 15-24)
39

Panel A: Labor force participation (age: 25-34)

Panel B: Wage and salary employment (age: 25-34) Panel C: Non-wage and salary employment (age: 25-34)

Figure 1.4B: Labor force participation and employment by sector (age: 25-34)
40

Panel A: Labor force participation (age: 35-44)

Panel B: Wage and salary employment (age: 35-44) Panel C: Non-wage and salary employment (age: 35-44)

Figure 1.4C: Labor force participation and employment by sector (age: 35-44)
41

Panel A: Labor force participation (age: 45-54)

Panel B: Wage and salary employment (age: 45-54) Panel C: Non-wage and salary employment (age: 45-54)

Figure 1.4D: Labor force participation and employment by sector (age: 45-54)
42

Panel A: Labor force participation (age: 55-64)

Panel B: Wage and salary employment (age: 55-64) Panel C: Non-wage and salary employment (age: 55-64)

Figure 1.4E: Labor force participation and employment by sector (age: 55-64)
43
This observation explains the phenomenon of large numbers of female workers retiring
earlier than male workers, a phenomenon resulting in a substantial gap in the labor force
participation of males and females for the eldest group of workers (age 55-64).
Panel B and Panel C of Figure 1.4A through 1.4E show the employment rates
26
of
male and female workers for each age group segregated by sector (wage and salary sector
and non-wage and salary sector). For both sectors and for all age groups, employment
rates for males were higher than those for females. These employment trends indicate an
obvious increase in employment within the wage and salary sector and an apparent
decline in employment within the non-wage and salary sector for both genders and for all
age groups except for the oldest group. A plausible explanation is that the oldest group
was too old to switch to the other sector where profitability tends to be higher. The
employment levels for both sectors across age groups
27
suggest that the younger the
workers, the more likely they are to be employed in the wage and salary sector and the
less likely they are to be employed in the non-wage and salary sector.
Tables 1.4A, 1.4B, and 1.4C summarize the characteristics of workers in the wage
and salary sector. According to Table 1.4A, in which the characteristics for both male and
female workers are explored, the average real earnings of workers increased 60% from
1985 to 1995. The financial crisis caused this figure to drop approximately by 15%. After
the crisis, the average real earnings went up but as of 2005, it still has not caught up with
pre-crisis levels. The table also illustrates the aging of the worker population.

26
Employment divided by population.
27
The youngest group (age 15-24) is ignored because their employment rates are still affected by
schooling.
44
Table 1.4A: Basic summary statistics of wage and salary workers

Mean Education Group
Year
No. of
Observations
Population %Female
Real
Monthly
Earnings*
Age
%Less than
Primary
%Primary
%Lower
Secondary
%Upper
Secondary
%University
or Higher**
1985 12034 5849066.58 39.02% 4415.46 30.66 54.22% 12.97% 16.09% 4.08% 12.04%
1986 12573 6351505.43 39.79% 4510.28 30.83 48.88% 13.98% 18.37% 4.40% 13.94%
1987 11552 6848846.78 41.02% 4494.52 30.91 46.98% 16.38% 17.32% 4.68% 14.24%
1988 12121 7297854.73 41.37% 4584.49 30.79 45.52% 17.04% 16.35% 4.79% 15.87%
1989 16822 7545307.74 41.40% 4697.56 30.92 43.86% 20.27% 16.21% 4.85% 14.62%
1990 15827 7968151.29 41.23% 5095.91 30.93 40.31% 22.76% 16.60% 4.98% 15.25%
1991 18039 8539637.77 41.39% 5111.04 30.93 42.01% 22.04% 16.14% 5.34% 14.34%
1992 18597 9277205.28 40.83% 6133.63 31.54 41.40% 20.33% 16.58% 5.26% 16.34%
1993 18018 10244564.95 41.29% 6624.86 31.49 37.83% 23.48% 16.80% 5.51% 16.33%
1994 35216 10335707.88 41.64% 6712.64 31.97 37.48% 22.42% 17.79% 5.67% 16.56%
1995 35851 10773146.14 41.10% 7067.06 32.25 36.13% 23.18% 18.46% 5.50% 16.69%
1996 35362 11357127.02 40.60% 7051.50 32.60 36.42% 23.71% 18.70% 5.23% 15.93%
1997 29508 10011235.50 41.31% 6404.31 32.61 37.99% 24.21% 18.77% 6.35% 12.64%
1998 32321 10791387.95 43.28% 7234.58 33.50 29.74% 20.44% 21.35% 6.91% 21.49%
1999 33106 11204099.12 44.06% 7115.91 33.62 29.47% 19.36% 20.48% 7.85% 22.79%
2000 33336 12061730.64 44.13% 6858.15 33.72 28.78% 21.40% 20.03% 8.30% 21.48%
2001 40791 11876302.31 44.31% 6996.02 33.60 25.08% 21.82% 15.73% 14.35% 22.69%
2002 42754 12402649.45 44.09% 6993.73 33.70 24.28% 22.05% 15.71% 14.74% 22.85%
2003 42007 12632106.81 44.50% 7052.11 33.94 22.64% 22.09% 15.89% 15.14% 23.75%
2004 42663 13742087.88 43.89% 6996.62 34.07 22.37% 21.73% 16.09% 15.72% 23.45%
2005 45837 13896957.27 45.39% 7249.27 34.54 21.30% 20.34% 16.41% 15.79% 25.34%

*Real monthly earnings (2002=100)
**Includes diploma-level education

45
Table 1.4B: Basic summary statistics of wage and salary male workers

Mean Education Group
Year
No. of
Observations
Population
Real
Monthly
Earnings*
Age
%Less than
Primary
%Primary
%Lower
Secondary
%Upper
Secondary
%University
or Higher**
1985 6787 3567031.52 4939.84 31.68 54.00% 11.90% 18.20% 4.26% 11.00%
1986 7008 3824149.82 5066.25 31.81 47.36% 13.18% 21.74% 4.72% 12.56%
1987 6367 4039130.73 4997.03 32.05 46.45% 16.11% 19.47% 4.94% 12.61%
1988 6683 4278724.47 5120.87 31.72 45.25% 16.12% 18.60% 5.22% 14.32%
1989 9238 4421575.72 5201.42 31.77 43.93% 19.60% 18.19% 4.82% 13.22%
1990 8839 4682692.71 5602.00 31.94 41.23% 21.74% 18.36% 5.22% 13.34%
1991 9787 5004899.11 5616.75 31.63 41.94% 21.49% 18.19% 5.61% 12.62%
1992 10094 5489067.87 6621.18 32.25 42.45% 18.99% 18.77% 5.58% 14.13%
1993 9826 6014362.81 7202.75 32.29 38.23% 22.71% 18.62% 5.90% 14.50%
1994 19661 6031767.15 7299.18 32.76 38.01% 22.39% 19.17% 6.02% 14.33%
1995 19855 6345294.24 7637.13 33.06 36.62% 23.43% 19.85% 5.80% 14.24%
1996 19813 6746446.30 7554.21 33.21 36.61% 24.58% 19.89% 5.50% 13.40%
1997 16521 5876083.07 6924.31 33.14 37.29% 25.16% 19.98% 6.32% 11.22%
1998 17501 6120730.76 7696.51 34.26 30.34% 21.16% 22.75% 7.24% 18.43%
1999 17635 6267969.24 7658.20 34.42 29.66% 20.24% 22.38% 8.07% 19.60%
2000 17742 6738429.08 7203.21 34.43 28.66% 22.32% 22.09% 8.81% 18.10%
2001 21759 6613538.09 7310.78 34.17 25.38% 22.82% 17.08% 15.32% 19.07%
2002 22991 6934298.37 7411.66 34.27 24.51% 23.16% 17.01% 15.76% 19.25%
2003 22357 7010685.49 7416.60 34.45 22.26% 23.94% 17.50% 15.58% 20.23%
2004 22896 7711284.83 7298.98 34.45 21.99% 23.39% 18.14% 16.23% 19.54%
2005 24203 7589291.74 7501.31 34.94 21.53% 21.97% 18.39% 16.56% 20.80%

*Real monthly earnings (2002=100)
**Includes diploma-level education

46
Table 1.4C: Basic summary statistics of wage and salary female workers

Mean Education Group
Year
No. of
Observations
Population
Real
Monthly
Earnings*
Age
%Less than
Primary
%Primary
%Lower
Secondary
%Upper
Secondary
%University
or Higher**
1985 5247 2282035.06 3595.80 29.07 54.56% 14.64% 12.78% 3.81% 13.66%
1986 5565 2527355.61 3669.04 29.36 51.17% 15.17% 13.28% 3.90% 16.04%
1987 5185 2809716.05 3772.13 29.28 47.74% 16.77% 14.22% 4.31% 16.58%
1988 5438 3019130.26 3824.35 29.48 45.90% 18.34% 13.17% 4.18% 18.07%
1989 7584 3123732.02 3984.36 29.73 43.77% 21.21% 13.41% 4.88% 16.61%
1990 6988 3285458.58 4374.59 29.49 39.00% 24.22% 14.08% 4.64% 17.96%
1991 8252 3534738.66 4394.99 29.94 42.12% 22.83% 13.23% 4.97% 16.77%
1992 8503 3788137.41 5427.18 30.51 39.88% 22.27% 13.41% 4.80% 19.54%
1993 8192 4230202.14 5803.25 30.36 37.26% 24.56% 14.21% 4.95% 18.94%
1994 15555 4303940.73 5890.63 30.87 36.74% 22.45% 15.86% 5.20% 19.68%
1995 15996 4427851.90 6250.14 31.09 35.42% 22.83% 16.47% 5.07% 20.20%
1996 15549 4610680.72 6315.92 31.69 36.14% 22.43% 16.96% 4.84% 19.62%
1997 12987 4135152.43 5665.39 31.86 38.98% 22.85% 17.05% 6.37% 14.67%
1998 14820 4670657.19 6629.23 32.49 28.94% 19.49% 19.52% 6.48% 25.50%
1999 15471 4936129.88 6427.30 32.60 29.23% 18.25% 18.07% 7.57% 26.84%
2000 15594 5323301.56 6421.36 32.83 28.93% 20.24% 17.43% 7.66% 25.75%
2001 19032 5262764.22 6600.48 32.89 24.70% 20.55% 14.04% 13.14% 27.24%
2002 19763 5468351.09 6463.75 32.97 24.00% 20.64% 14.05% 13.46% 27.42%
2003 19650 5621421.32 6597.53 33.31 23.11% 19.78% 13.90% 14.58% 28.13%
2004 19767 6030803.05 6610.01 33.58 22.85% 19.61% 13.47% 15.06% 28.45%
2005 21634 6307665.53 6946.01 34.05 21.03% 18.38% 14.03% 14.86% 30.81%

*Real monthly earnings (2002=100)
**Includes diploma-level education
47
Thailand experienced a “baby boom” period from 1963-1983. The dynamic of aging
observed in the data may have reflected the phenomenon of these baby boomers entering
the labor market. The aging of workers in the labor market can also be explained by the
postponement of entrance into the labor market by the younger generations as more and
more of them attain higher education. Over time, the proportion of female labor has
expanded. Workers have become more educated. The proportion of workers with less
than a primary education (lower than 6
th
grade) has declined, while at the same time, the
proportion of workers with an upper secondary education (12
th
grade) and the proportion
of workers with a university education or higher grew during this period.
Table 1.4B and 1.4C reproduce statistics similar to those in Table 1.4A but they
segregate the data for males (Table 1.4B) and for females (Table 1.4C). The data show
that, on average, male workers earned more than female workers, with the earnings
differences tending to be lower over time. In 1985, male workers were about 2.6 years
older than female workers. However, the age gap between the genders has declined and
by 2005 had almost closed. For many years, there was a higher proportion of male
workers with at least a primary, lower secondary, or upper secondary education,
compared to female workers. However, there was a higher proportion of female workers
with at least a university education, compared to male workers. The proportion of
workers with at least a university education grew substantially for both genders from
1985-2005. In 1985, 11% of male workers and 13.66% of female workers had at least a
university education, but 20.80% of males and 30.81% of females had at least a
university education in 2005.
48
In Thailand, labor protection laws state explicitly that “an employer shall treat
male and female employees equally in employment unless the description or nature of
work prevents such treatment.”
28
These laws prohibit female workers from engaging in
occupations that require them to encounter dangerous or hazardous work conditions.
Employers are forbidden from demanding pregnant women to work from 10PM to 6AM.
Although the law allows female employees in general to work from 12AM to 6AM, a
labor inspector has the right to alter these hours if the work environment is deemed
dangerous for female workers. Also, women are entitled to 90 days maternity leave.
According to ADB (1998), these laws, although intended to protect female workers, may
actually discourage employers from hiring married females.
Figure 1.5 illustrates the average log earnings of male and female workers in the
wage and salary sector during the past two decades (1985-2005). Many features are
readily apparent. First, real earnings for both males and females tended to possess the
same pattern. These earnings increased significantly for both genders during the early
1990s. This surge ceased by the mid-1990s, with average earnings dropping in 1997, the
year of the Thai financial crisis. After the crisis, average earnings improved slightly
remaining rather constant until 2005. Figure 1.6 illustrates the average log hourly wages
of male and female workers.
29
It can be easily seen that the income trends of males and
females estimated using hourly wages are very similar to those estimated using monthly
earnings (Figure 1.5).

28
Labor Protection Act B.E. 2541 (1998) Chapter 1, Section 15.
29
Hourly wages are calculated by Monthly earnings divided by (4.2 * hours worked per week)
49

Figure 1.5: Earnings of male and female workers (1985-2005)

Figure 1.6: Hourly wages of male and female workers (1985-2005)

50
Second, although the earnings trends for men and women looked quite similar, the
gap between the two has actually narrowed over time. This is even more evident when
the log earnings differentials are plotted, as in Figure 1.1. The figure shows a remarkable
drop in the gap from 1985 to 2001. The magnitude of the gap was approximately 0.34
log-points in 1985. It has declined and stabilized at about 0.09 log-points since 2001. The
log hourly wages differentials are plotted in Figure 1.7. It is obvious that the income gap
estimated using hourly wages and monthly earnings are very similar. Figure 1.8 plots log
earnings differential VS earnings percentile. In comparing the years 1985, 1995, and
2005, Figure 1.8 shows a reduction in the gender earnings gap for every percentile group
during the data period. This is illustrated by the 1985 graph at the top, the 1995 graph in
the middle, and the 2005 graph at the bottom. This evidence unambiguously suggests the
narrowing over time of the gender earnings gap in the Thai wage and salary sector.
Figure 1.9A (Panel A and B) and Figure 1.9B (Panel A and B) break down the
gender earnings gap by age group (15-40 and 41-65). It is apparent that the gap was
wider for the older workers. The earnings trend of the older workers tended to be more
volatile than that of the younger workers. It is evident that, for both young and old
workers, the gender earnings gap declined.
To give a more detailed representation of the matter, it is useful to see how the
actual earnings distributions for males and females have evolved over time. I use the
kernel density estimation method to estimate such density functions. The reason for
examining the densities of earnings is because the mean earnings cannot illustrate the
51

Figure 1.7: Gender (hourly) wage gap in Thailand (1985-2005)

Figure 1.8: Gender earnings gap by percentile

52

Panel A: Earnings of male and female workers (age: 15-40)

Panel B: Gender earnings gap (age: 15-40)

Figure 1.9A: Earnings of male and female workers and gender earnings gap (age: 15-40)
53

Panel A: Earnings of male and female workers (age: 41-65)

Panel B: Gender earnings gap (age: 41-65)

Figure 1.9B: Earnings of male and female workers and gender earnings gap (age: 41-65)
54
skewness of income or the division of income into (high-income and low-income)
groups. In the pre-modern time, only a limited number of people, usually the rich and the
noble families, got access to education and thus high-paying careers. One may expect the
densities of earnings for early years to be right-skewed. A bimodal structure for the
earnings densities can also be expected especially for older workers since the society was
more divided (in terms of education and thus income). Expansion of education during the
modernization allowed more people to get access to education. Thus, one may expect the
densities of earnings for later years to be less skewed and less divided. The density of
earnings for male workers ( ) (y f
M
) can be estimated as follows:
∑
=
−
=
M
n
i
i i M
h
y y
K
h
w
h y f
1
) ( ) ; (
ˆ
. (1)
Under this expression,
i
y is the log of real monthly earnings,
M
n is the number of male
observations,
i
w represents the survey weight from the data (normalized to sum to one),
h is the bandwidth, and K is the kernel function.
30
This exact same methodology can also
be implemented to estimate the earnings density of female workers ( ) (y f
F
).

30
The Kernel function used is Epanechinikov:
⎪
⎪
⎩
⎪
⎪
⎨
⎧
≤
−
=
otherwise
z if
z
z K
0
5 / /
5
)
5
1
1 (
4
3
) (
2

The bandwidth used is the “optimal” bandwidth (Silverman, 1986):
5 / 1
)
349 . 1
, var min( 9 . 0
−
⋅ ⋅ =
M
y
y
n
IQR
h
,
where
y
var and
y
IQR are the variance and the interquartile range (the difference between the 3
rd
and the
1
st
quartiles) for the log wage data.
M
n is the number of male observations.
55
The density estimations for male and female workers during 1985-2005 are
illustrated in Figure 1.10A and 1.10B. The male densities were fairly symmetric (with
some trivial right-skewness in a few years). The modes were smooth and the tails were
thick during the 1980s. However, the modes became more spiked and the tails became
narrower in the 1990s. As for females, we can observe some obvious right-skewness of
the densities in almost all years throughout the 1980s and the early 1990s. However, the
right-skewness tended to disappear in later years for females. The bimodal structure is
observed for female earnings densities during the 1980s.
Panel A of Figure 1.11 places the 1985, 1995, and 2005 male earnings densities
on the same axis. The 1985 density is represented by the long-dash line, the 1995 density
is represented by the short-dash line, and the 2005 density is represented by the solid line.
The variance was at its largest in 1985, and then became smaller in 1995 and 2005. The
entire earnings distribution shifted significantly to the right from 1985 to 1995. However,
in 2005, the distribution shifted slightly leftward. Thus, the data indicate that male
workers in the wage and salary sector (no matter where they were located in the earnings
distribution) were receiving higher real incomes in 1995 compared to 1985. However, in
2005, some of these males earned less than what they used to earn in 1995. Panel B of
Figure 1.11 places the 1985, 1995, and 2005 earnings densities of female workers on the
same axis. Similar to the case of male workers, the 1995 density lay significantly to the
right of the 1985 density, suggesting that female workers in 1995 earned a higher income
than their 1985 counterparts. Unlike the case of male workers, the 2005 female earnings
density did not shift to the left; it almost completely overlapped the 1995 density.
56
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1985
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1986
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1987
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1988
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1989
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1990
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1991
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1992
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1993
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1994
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1995
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1996
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1997
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1998
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1999
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
2000
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
2001
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
2002
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
2003
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
2004
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
2005

Figure 1.10A: Earnings density estimation for male workers (1985-2005)
57
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1985
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1986
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1987
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1988
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1989
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1990
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1991
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1992
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1993
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1994
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1995
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1996
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1997
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1998
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
1999
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
2000
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
2001
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
2002
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
2003
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
2004
0 .2 .4 .6 .8 1 1.2
Density
4 5 6 7 8 9 10 11 12
Log of real earnings
2005

Figure 1.10B: Earnings density estimation for female workers (1985-2005)
58

Panel A: Male earnings

Panel B: Female earnings

Figure 1.11: Earnings density comparison (1985 VS 1995 VS 2005)
59
Figure 1.12 contrasts male and female earnings densities for selected years (1985,
1995, and 2005). The long-dash line represents male density and the short-dash line
represents female density. For 1985 (Panel A), the diagram demonstrates clearly the
dissimilarities between the male and female densities. The female density was right-
skewed and lay to the left of the male density. The peaks of the two densities did not
coincide. In 1995 (Panel B), the female density still lay to the left of the male density;
however, the two densities were very close to each other. Both densities had narrower
bases and had steeper modes in 1995, compared to 1985. In 2005 (Panel C), although at
the upper end and the lower end of the density graph, males still earned slightly higher
than females, the two densities coincided almost completely. These observations confirm
that the gender earnings gap for Thai workers not only narrowed at the mean values but
also narrowed for overall density. Figure 1.13 plots male and female densities using
hourly wages. The figure confirms that the density gap between males and females
estimated using hourly wages are quite similar to those estimated using monthly earnings
(Figure 1.12).
Figure 1.14A (Panel A, B, and C) and Figure 1.14B (Panel A, B, and C)
reproduce the earnings density analyses from Figure 1.12 but segregate the workers into a
younger group (Figure 1.14A, age 15-40) and an older group (Figure 1.14B, age 41-65).
The figures confirm that the earnings density gap between male and female workers was
narrower for the younger workers than for the older workers. A bimodal structure was
very apparent for older workers of both genders. This illustrates that the older workers
were distinctly divided into high-skilled (high earnings) and low-skilled (low earnings).
60

Panel A: 1985

Panel B: 1995

Panel C: 2005
Figure 1.12: Earnings density comparison (male VS female)
61

Panel A: 1985

Panel B: 1995

Panel C: 2005
Figure 1.13: Hourly wage density comparison (male VS female)
62

Panel A: 1985

Panel B: 1995

Panel C: 2005
Figure 1.14A: Earnings density comparison (male VS female, age: 15-40)
63

Panel A: 1985

Panel B: 1995

Panel C: 2005
Figure 1.14B: Earnings density comparison (male VS female, age: 41-65)
64
However, for the younger workers their earnings densities had a unimodal shape
reflecting similar levels of skills among themselves. Disparities between male and female
earnings densities can be observed in 1985 and 1995 for both age groups, although those
differences are more apparent for older workers. However, in 2005, for younger workers
(Figure 1.14A Panel C), the earnings of males and females coincided almost completely.
As for the older workers in 2005 (Figure 1.14C Panel C), female workers were still worse
off than male workers at the lower end of the distributions.
1.5 Parametric Decomposition
The decline in Thailand’s gender earnings gap can be further investigated using
decomposition analyses. Both parametric and nonparametric methodologies are
implemented in this paper, the former in this section and the latter in the next (Section
1.6). The purpose of the parametric decomposition is to identify which factors have
contributed to the existence of and changes in the gender earnings gap. Neumark’s (1988)
version of the Blinder (1973) and Oaxaca (1973) technique (BO) is used. This technique
decomposes earnings differences into a residual gap and a gap accounted for by the
differences in the observed characteristics between the two genders. Another parametric
approach that is implemented in this section is the one proposed by Juhn, Murphy, and
Pierce (1991) (JMP). The authors extended the BO technique to document changes in the
gender earnings gap across time periods. The JMP method investigates how changes in
the gender earnings gap can be explained by changes in the (observed and unobserved)
characteristic differences between the two genders and changes in the market returns to
those characteristics. This section is organized as follows. First, the BO technique is
65
elaborated. Then, the JMP method and its relationship to the BO method are discussed.
Finally, both methodologies are implemented and the results are investigated.
1.5.1 Blinder-Oaxaca (1973): BO
The BO technique segregates the gender earnings gap into two parts. The first
part, the endowment gap, is explained by the differences in the observed characteristics
of males and females. The second part, the residual gap (or sometimes called the
unexplained gap), is accounted for by the differences in the pay mechanisms faced by
males and females. This residual gap was originally interpreted by some scholars as
reflecting gender discrimination, since the unequal rates of compensation for the two
genders could suggest that the two genders were being treated differently. In fact, males
and females have some unobserved dissimilarities, and these dissimilarities could
contribute to male and female workers being remunerated at unequal rates. Thus, the
residual gap is not necessarily a result of gender discrimination.
Using the BO technique, I start with the familiar parametric earnings equations.
These earnings equations are written separately for males (superscripted as M) and for
females (superscripted as F):
.
F
i
F F
i
F
i
M
i
M M
i
M
i
e b x y
e b x y
+ =
+ =
(2)
i
y is the log earnings of worker i.
i
x is the vector of observable attributes of the worker.
These attributes include education, age, age-squared, marital status, community
characteristics, and regional information. Some scholars also include occupation and
industry variables in their regressions. Thus, these scholars’ coefficients would reflect the
66
rates of returns to workers’ attributes in which occupation and industry are properly
accounted for. However, males and females have different levels of access to occupations
and industries, partly due to unobserved characteristics.
31
Some authors exclude this
occupation and industry information in their analyses because including it could cause
statistical bias. In this paper, I perform analyses that both exclude and include occupation
and industry variables, with the former comprising a base model and the latter comprising
the extended model.
From (2), the least-squared estimates of
M
b and
F
b can be obtained.
32
These beta
coefficients are sometimes called, respectively, male wage structure and female wage
structure. The mean earnings of male and female workers can be estimated using these
coefficients and the average characteristics of the workers.
33
Subtracting the means of the
two groups yields the following expression:
F F M M F M
b x b x y y y − = − = Δ (3)
The decomposition procedure requires that a reference wage structure be chosen.
This reference wage structure is a pay mechanism that operates in an “ideal society” – a
society where everyone faces the same pay mechanism regardless of their gender.
34
The

31
Some literature claimed that within each occupation and industry, the different levels of access for males
and females were also due to discrimination.
32
Note that
M
b and
F
b also include the intercept terms.
33
Specifically,
M M M
b x y = , and
F F F
b x y = .
34
Early literature referred to this reference wage structure as the nondiscriminatory wage structure. These
authors assumed that males and females were compensated differently (as can be observed from the
different values of
M
b and
F
b ) solely because of discrimination. In other words, both male and female
observable characteristics would be compensated at the same rate, the nondiscriminatory rate, if there were
no discrimination. In fact, the rate of returns to male and female characteristics may differ due to factors
other than discrimination, such as the differences in their unobservable characteristics. Therefore, in this
67
original work by Blinder (1973) and Oaxaca (1973) suggested the use of either male
wage structure (
M
b ) or female wage structure (
F
b ) as the reference. However, this
approach of using either
M
b or
F
b is rather arbitrary.
Extending the original theoretical model of discrimination proposed by Becker
(1957) and Arrow (1972), Neumark (1988) proposed a non-arbitrary estimator of the
reference wage structure.
35
This index can be estimated from the coefficient of the
regression of the entire sample (pooled regression). Neumark’s (1988) method is widely
recognized and is the approach that will be followed in this paper.
With the pooled wage structure
A
b chosen as the reference wage structure
(Neumark, 1988), the terms
A M
b x and
A F
b x are both added to and subtracted from (3):
.
) ( ) ( ) (
F F M M A
F A F A M M A F M F M
b x b x b x
b b x b b x b x x y y y
Δ + Δ + Δ =
− + − + − = − = Δ
(4)
In this representation, the total earnings gap between males and females is
decomposed into the portion explained by the observable differences of males and
females (first term on the right-hand side) and the residual gap that is driven by the
differences in the market returns to the skills of males and females (the remainder of the

paper, I will refrain from using the term “nondiscriminatory wage structure” and will utilize the term
“reference wage structure” throughout.
35
The theoretical model of discrimination explains why the two groups of workers (the majority and the
minority), who are perfect substitutes in the production function (i.e., they have the same productivity), are
not paid equally. The first explanation (Becker, 1957) is that employers are prejudiced and incur some
disutility if they hire a minority. The second explanation (Arrow, 1972) is that employers may have some
prior information that a particular group of workers may have inferior work habits and based on this pay
these workers less. Imposing the assumption that the employer’s utility function is homogeneous of degree
zero for male and female workers within the same skill group (i.e., utility does not change if the numbers of
males and females within a skill group are scaled up or down in the same proportion), Neumark (1988)
showed that the nondiscriminatory wage structure for each skill group (quantified by the marginal product
of labor for that skill group) can be derived from the weighted sum of the wages of males and females
within that group. He showed that this quantity is equivalent to the least square coefficient estimated from
the entire sample.
68
terms on the right-hand side).
36
The residual gap is composed of two terms, the
“advantage of males” (second term on the right-hand side) and the “disadvantage of
females” (third term on the right-hand-side). The positive values of these two terms
indicate that males are being more favorably compensated and that females are being less
favorably compensated compared to the reference pay structure.
37

It is important to point out that expression (4) can also be thought of as a
generalized equation for the BO decomposition that has the ability to accommodate
various types of reference wage structures.
38
For example, if male wage structure is used
as the reference, as in the original work by Blinder (1973) and Oaxaca (1973), then
A
b in
expression (4) is replaced by
M
b . The resulting expression is equivalent to adding and
subtracting the term
M F
b x to and from (3). That is:
.
) ( ) (
b x b x
b b x b x x y y y
F M
F M F M F M F M
Δ + Δ =
− + − = − = Δ
(5)
Similar to (4), the first term and the second term on the right-hand side denote the
endowment gap and the residual gap, respectively. It is important for readers to make
note of expression (5). Although not utilized in this paper, expression (5) is helpful in
understanding the relationship between the BO decomposition and the nonparametric
decomposition that will be discussed in Section 1.6.

36
This residual gap was regarded by early literature as the “discrimination factor”.
37
Given that the observable characteristics included in the equations are beneficial to productivity.
38

A
b in expression (4) can be replaced by [1]
M
b or
F
b as proposed by Blinder (1973) and Oaxaca
(1973); [2]
f M
b b 5 . 5 . + as proposed by Reimers (1983); and [3]
F
F
M
M
b f b f + (where
M
f is the fraction
of the male workers, and
F
f is the fraction of the female workers in the population) as proposed by Cotton
(1988).
69
1.5.2 Juhn-Murphy-Pierce (1991): JMP
Juhn, Murphy, and Pierce (1991) proposed a methodology in which changes in
the gender earnings gap across time periods can be analyzed. Unlike the BO method, the
JMP method not only distinguishes whether the sources of gap changes come from
changes in the explained or the unexplained attributes, the JMP method also recognizes
price and quantity components for those attributes.
In order for the JMP analysis to be compatible with Neumark’s (1988) version of
the BO method that was performed in the previous section, the pooled wage structure will
be used as the reference wage structure. The JMP approach rewrites the earnings equation
to include a standardized residual (
it
θ ) (a residual with a mean of zero and a standard
deviation of one) in place of a typical residual (
A
it
ε ). The overall earnings equation
(pooled regression) can be written as:
.
it
A
t
A
t it it
b x y θ σ + = (6)
it
θ is the standardized residual.
A
t
σ is the standard error of the typical residual
A
it
ε
(i.e.,
it
A
t
A
it
θ σ ε = ). One interpretation of
A
t
σ is that it portrays how scattered the
distribution of the typical residual term (
A
it
ε ) is. The distribution graph for
A
it
ε is
sometimes called the residual earnings distribution because it portrays the distribution for
the remaining part of the actual earnings after the observable characteristics have been
accounted for. Thus, the lower value of
A
t
σ indicates a lower level of dispersion for the
distribution of the residual term, implying a lower residual earnings inequality.
70
Another way to interpret
A
t
σ is to look at (6) as a hedonic price function. The left-
hand side term represents the total earnings paid to an individual. The first term on the
right-hand side represents the “prices” associated with the observable characteristics (
A
t
b )
multiplied by the “quantities” of those observable characteristics (
it
x ). The second term
on the right-hand side represents the “prices” associated with the unobservable
characteristics (
A
t
σ ) multiplied by the (normalized) “quantities” of those unobservable
characteristics (
it
θ ). Thus, the lower
A
t
σ can also be interpreted as representing lower
returns to unobservable characteristics.
Using the pooled wage structure as the reference wage structure in the
decomposition, one ought to utilize
A
t
b and
A
t
σ as the common price system for both
males and females. The price system (
A
t
b and
A
t
σ ) can be obtained from a regression of
expression (6) over the entire sample. The male and female earnings equations need to be
written in terms of pooled prices:
M
it
A
t
A
t
M
it
M
it
b x y θ σ + = (7)
F
it
A
t
A
t
F
it
F
it
b x y θ σ + = (8)
where
A
t
A
t
M
it
M
it M
it
b x y
σ
θ
−
= and .
A
t
A
t
F
it
F
it F
it
b x y
σ
θ
−
=
The simplest way to understand (7) and (8) is to look at these expressions as
hedonic price functions of male and female labor, with the prices (
A
t
b and
A
t
σ ) taken as
given. That is, male and female workers face the same rates of returns to both their
observable and unobservable characteristics. Under this construction, the distributions for
71
the quantities of the unobservable characteristics for males and females (the distributions
of
M
it
θ and
F
it
θ ) can be compared to the distribution for the quantities of the unobservable
characteristics of overall workers (the distribution of
it
θ ). Additionally, the distributions
of
M
it
θ and
F
it
θ can be compared with each other. In terms of the residual earnings
distribution, if females normally rank lower than the average, while males normally rank
higher than the average, one would expect the distribution of
F
it
θ to lie to the left of the
distribution of
A
it
θ , and the distribution of
M
it
θ to lie to the right of the distribution of
A
it
θ .
Evaluating at the mean values, by construction, we have 0 =
A
t
θ . Thus,
M
t
θ portrays how
the average male’s residual earnings (earnings after his observable characteristics are
accounted for) compare to the average residual earnings for overall workers. Also,
F
t
θ
portrays how an average female’s residual earnings compare to the average residual
earnings for overall workers.
Subtracting the estimated earnings of the average female from that of the average
male, the equation can be written as:
.
) ( ) (
t
A
t
A
t t
F
t
M
t
A
t
A
t
F
t
M
t
F
t
M
t t
b x
b x x y y y
θ σ
θ θ σ
Δ + Δ =
− + − = − = Δ
(9)
Resembling BO, JMP at this stage decomposes the total gap (at time t) into an
explained part and an unexplained part. The first term on the right-hand side represents
the differences in the observable characteristics of workers evaluated using the pooled
wage structure. The second term on the right-hand side represents the differences in the
unobservable characteristics of workers evaluated using the pooled wage structure.
72
The similarity between (9) and (4) are clear. The first terms on the right-hand side
of both expressions are equivalent, and the sum of the last two terms in (4) equals the last
term of (9). The magnitudes of the estimated total gap, the explained gap, and the
unexplained gap are equivalent for these two methodologies. However, the interpretations
for the unexplained portion somewhat differ. According to BO, the unexplained portion is
a function of the differences in the market returns to the observable characteristics of
males and females. However, under the JMP representation, the unexplained portion is a
function of the residual earnings inequality and the differences between the unobservable
characteristics of the two genders.
As mentioned earlier, the JMP method advances the BO method in a way that
allows the gender earnings gap to be analyzed across time. Also, the JMP method allows
changes in the total gap to be accounted for by quantity and price factors for both the
observed and the unobserved attributes. Using (9), we can illustrate the change in the
gender earnings gap between two time points (t = 0, 1) as follows:
. ) ( ) (
0 0 1 1 0 0 1 1 0 1
θ σ θ σ Δ − Δ + Δ − Δ = Δ − Δ
A A A A
b x b x y y
Adding and subtracting
A
b x
0 1
Δ and
1 0
θ σ Δ
A
to and from the equation, we arrive at
the final decomposition equation for JMP:
. ) ( ) ( ) ( ) (
1 0 1 0 1 0 0 1 1 0 0 1 0 1
θ σ σ θ θ σ Δ − + Δ − Δ + − Δ + Δ − Δ = Δ − Δ
A A A A A A
b b x b x x y y (10)
Changes in the gender earnings gap between two time points can be explained by
changes in the quantities of the observable characteristics differences over time (first term
on the right-hand side), changes in the prices of those observable characteristics over time
(second term on the right-hand side), changes in the quantities of the unobservable
73
characteristics differences over time (third term on the right-hand side), and changes in
the prices of those unobservable characteristics over time (fourth term on the right-hand-
side). The first two terms in combination are recognized as representing changes in the
explained gap, while the last two terms in combination are recognized as representing
changes in the unexplained gap.
1.5.3 BO Results
I will open this section by discussing the results from the application of the
earnings equation (expression (2)) to the Thai LFS data for the years 1985, 1995, and
2005. This regression is implemented for all workers and for male and female workers
separately. The results from the base specification, the specification that excludes
occupation and industry variables, are shown in Table 1.5A. From 1985-1995, for male
workers and for all workers, returns to all levels of education increased significantly.
Compared to the returns to other levels of education, returns to primary education (6
th

grade) increased the most. During the same time period, for female workers, returns to all
levels of education, except for the primary level, decreased. From 1995-2005, for male
workers and for all workers, returns to all levels of education except for the lower
secondary level (9
th
grade) increased. Returns to primary education increased least
compared to returns to other levels of education. During the same time period, for female
workers, returns to education decreased for primary and lower secondary education and
increased for upper secondary (12
th
grade) and university education. In all years, returns
to experience possessed a concavity, namely a positive coefficient for the linear term and
a negative coefficient for the squared term. The magnitude of the returns to experience
74
Table 1.5A: Earnings equation (without occupations and industries)

1985 1995 2005
All Male Female All Male Female All Male Female
Primary
0.23028*** 0.20159*** 0.22911*** 0.32513*** 0.27669*** 0.34410*** 0.33212*** 0.31286*** 0.31617***
(6th grade)
(0.02638) (0.03698) (0.03200) (0.01792) (0.02407) (0.02498) (0.01283) (0.01705) (0.01844)
Lower Secondary
0.56136*** 0.43513*** 0.72725*** 0.58681*** 0.51340*** 0.63913*** 0.53859*** 0.47405*** 0.57664***
(9th grade)
(0.01915) (0.02497) (0.02687) (0.02241) (0.02809) (0.03536) (0.01362) (0.01807) (0.01986)
Upper Secondary
0.58559*** 0.43926*** 0.79704*** 0.61195*** 0.54396*** 0.66084*** 0.72705*** 0.66856*** 0.76411***
(12th grade)
(0.03387) (0.04399) (0.05614) (0.02499) (0.03371) (0.03167) (0.01296) (0.01728) (0.01894)
University
0.98646*** 0.87659*** 1.16642*** 1.12407*** 1.05930*** 1.22263*** 1.29871*** 1.25200*** 1.35882***
(including Diploma)
(0.01636) (0.02148) (0.02331) (0.01850) (0.02812) (0.02130) (0.01180) (0.01711) (0.01610)
Age
0.07179*** 0.07163*** 0.06988*** 0.06183*** 0.05946*** 0.06345*** 0.04106*** 0.04070*** 0.04791***

(0.00471) (0.00646) (0.00624) (0.00339) (0.00464) (0.00508) (0.00255) (0.00335) (0.00373)
Age-squared
-0.00077*** -0.00075*** -0.00082*** -0.00058*** -0.00054*** -0.00066*** -0.00025*** -0.00023*** -0.00038***

(0.00006) (0.00008) (0.00009) (0.00005) (0.00006) (0.00006) (0.00004) (0.00005) (0.00005)
Urban
0.11660*** 0.14435*** 0.06576*** 0.08944*** 0.09230*** 0.08326*** 0.13089*** 0.12380*** 0.14413***

(0.01555) (0.02025) (0.02136) (0.00821) (0.01069) (0.01206) (0.00679) (0.00923) (0.00964)
Married
0.16997*** 0.14500*** 0.10131*** 0.08881*** 0.09183*** 0.02362 0.06860*** 0.09681*** 0.01971*

(0.01818) (0.02632) (0.02285) (0.01417) (0.02269) (0.01794) (0.00810) (0.01215) (0.01057)
Constant
6.50769*** 6.65249*** 6.44464*** 6.97047*** 7.14004*** 6.86138*** 7.02037*** 7.11263*** 6.86515***

(0.07927) (0.11077) (0.10267) (0.05905) (0.07758) (0.09343) (0.04588) (0.06088) (0.06683)
Region
Dummies
Yes Yes Yes Yes Yes Yes Yes Yes Yes
Occupation
Dummies
No No No No No No No No No
Industry
Dummies
No No No No No No No No No
Observations
12034 6787 5247 35851 19855 15996 45837 24203 21634
R-squared
0.50 0.48 0.57 0.52 0.51 0.57 0.55 0.54 0.60

Robust standard errors in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
75
declined during 1985-1995 and 1995-2005. In 1985 and 1995, living in urban areas
benefited men more than women in terms of earnings; in 2005, the reverse was true. The
coefficients of the marriage dummy variable were higher for men than for women. Over
the period of the study (1985-1995 and 1995-2005), these coefficients of the marriage
dummy variable declined for both genders. Table 1.5B illustrates the earnings regression
results, as per the specification including occupation and industry variables. The
magnitude of the returns to education for all levels decreased for all workers, although
more for females than for males. Like Table 1.5A, returns to education increased
significantly from 1985-1995. However, from 1995-2005, only returns to some levels of
education increased. Coefficient estimates for other variables seemed not to differ much
compared to the ones in the base specification.
Table 1.6A demonstrates the results from the BO analysis (expression (4)) for the
years 1985, 1995, and 2005. Panel A of the table summarizes the overall decomposition
results. The total earnings gap or the total difference is calculated by subtracting the
arithmetic mean of female log earnings from the arithmetic mean of male log earnings.
Note that this difference is equivalent to the log of the geometric mean of female earnings
subtracted from the log of the geometric mean of male earnings.
39
A positive gap
indicates higher average male earnings compared to average female earnings, while a
negative gap indicates the opposite. In each year (1985, 1995, 2005), a positive gender

39
Expressed mathematically:
F
F M
F M F M
w
w w
w w w w
~
~ ~
~
ln
~
ln ln ln
−
≈ − = −
, where
∑
=
=
M
M
n
j
n
j
M
M
w
n
w
1
) ln(
1
ln
;
∑
=
=
F
F
n
j
n
j
F
F
w
n
w
1
) ln(
1
ln
;
M
M
n M
n
M M M M
w w w w w
/ 1
3 2 1
) ... (
~
⋅ ⋅ ⋅ ⋅ = ;
F
F
n F
n
F F F F
w w w w w
/ 1
3 2 1
) ... (
~
⋅ ⋅ ⋅ ⋅ = ;
M
n and
F
n are the number of male and female observations.
76
Table 1.5B: Earnings equation (with occupations and industries)

1985 1995 2005
All Male Female All Male Female All Male Female
Primary
0.18300*** 0.15791*** 0.14885*** 0.26118*** 0.21907*** 0.23211*** 0.23068*** 0.22249*** 0.20636***
(6th grade)
(0.02376) (0.03239) (0.02956) (0.01733) (0.02345) (0.02297) (0.01204) (0.01611) (0.01718)
Lower Secondary
0.38875*** 0.29245*** 0.36995*** 0.46246*** 0.39135*** 0.37361*** 0.36016*** 0.31151*** 0.35949***
(9th grade)
(0.02152) (0.02582) (0.03328) (0.02341) (0.03057) (0.03453) (0.01340) (0.01806) (0.01856)
Upper Secondary
0.36643*** 0.27080*** 0.30483*** 0.46330*** 0.40137*** 0.32195*** 0.48187*** 0.44008*** 0.45920***
(12th grade)
(0.03702) (0.04822) (0.05276) (0.02637) (0.03481) (0.03446) (0.01356) (0.01847) (0.01874)
University
0.62032*** 0.55808*** 0.52301*** 0.81866*** 0.77987*** 0.68890*** 0.78108*** 0.76530*** 0.73712***
(including Diploma)
(0.03062) (0.03995) (0.04284) (0.03094) (0.04488) (0.03855) (0.01668) (0.02326) (0.02229)
Age
0.05931*** 0.05564*** 0.05361*** 0.05300*** 0.04959*** 0.05006*** 0.03408*** 0.03414*** 0.03788***

(0.00450) (0.00615) (0.00590) (0.00322) (0.00437) (0.00482) (0.00232) (0.00308) (0.00334)
Age-squared
-0.00062*** -0.00056*** -0.00062*** -0.00049*** -0.00044*** -0.00051*** -0.00022*** -0.00021*** -0.00030***

(0.00006) (0.00008) (0.00008) (0.00004) (0.00006) (0.00006) (0.00003) (0.00004) (0.00005)
Urban
0.03465** 0.03747* 0.08949*** 0.05454*** 0.04981*** 0.07296*** 0.07888*** 0.06672*** 0.08644***

(0.01705) (0.02158) (0.02332) (0.00816) (0.01026) (0.01221) (0.00640) (0.00878) (0.00884)
Married
0.15604*** 0.13743*** 0.08238*** 0.07389*** 0.07790*** 0.01734 0.05645*** 0.08239*** 0.02652***

(0.01681) (0.02326) (0.02196) (0.01343) (0.02118) (0.01626) (0.00728) (0.01083) (0.00946)
Constant
7.22752*** 7.39659*** 7.34572*** 7.31408*** 7.50270*** 7.65173*** 7.66760*** 7.59290*** 7.80495***

(0.09732) (0.13024) (0.14688) (0.07785) (0.09943) (0.10465) (0.05834) (0.07587) (0.08243)
Region
Dummies
Yes Yes Yes Yes Yes Yes Yes Yes Yes
Occupation
Dummies
Yes Yes Yes Yes Yes Yes Yes Yes Yes
Industry
Dummies
Yes Yes Yes Yes Yes Yes Yes Yes Yes
Observations
12025 6781 5244 35851 19855 15996 45837 24203 21634
R-squared
0.57 0.55 0.63 0.57 0.56 0.63 0.63 0.61 0.68

Robust standard errors in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
77
Table 1.6A: Blinder-Oaxaca (1973) results (without occupations and industries)

Panel A:
1985 1995 2005
Total Difference (T) 0.33961 100.00% 0.20870 100.00% 0.08978 100.00%
Explained (E) 0.09042 26.62% -0.00142 -0.68% -0.08492 -94.59%
Unexplained (U) 0.24919 73.38% 0.21012 100.68% 0.17469 194.59%

Panel B:
Explained (Details)
1985 1995 2005
Education 0.00043 0.13% -0.04069 -19.50% -0.08227 -91.64%
Primary
-0.00632 0.00194 0.01192
Lower Secondary
0.03039 0.01981 0.02348
Upper Secondary
0.00262 0.00448 0.01234
University
-0.02626 -0.06691 -0.13001
Age&Age-Squared
0.06169 18.17% 0.04381 20.99% 0.01763 19.64%
Age
0.18722 0.12201 0.03630
Age-Squared
-0.12553 -0.07820 -0.01867
Urban -0.00175 -0.52% -0.00338 -1.62% -0.00605 -6.74%
Married 0.02885 8.49% 0.01000 4.79% 0.00331 3.69%
All Regions 0.00120 0.35% -0.01116 -5.35% -0.01755 -19.55%
Explained (E) 0.09042 26.62% -0.00142 -0.68% -0.08492 -94.59%

Panel C:
Unexplained (Details)
1985 1995 2005
Advantage of Men 0.09722 28.63% 0.08636 41.38% 0.07929 88.32%
Disadvantage of Women 0.15197 44.75% 0.12376 59.30% 0.09540 106.27%
Unexplained (U) 0.24919 73.38% 0.21012 100.68% 0.17469 194.59%
78
earnings gap between male and female is observed but this gap has been declining over
time. The magnitude of the gap in terms of log-points was 0.33961 in 1985, 0.20870 in
1995, and 0.08978 in 2005. In other words, the average male in Thailand earned
approximately 33.96% higher than the average female in 1985. The average male earned
20.87% higher than the average female in 1995, while the average male earned 8.98%
higher than the average female in 2005.
An interesting feature of these results is that the portion of the total gap that was
accounted for by the explained gap declined significantly and even became negative over
time. The magnitude of the explained gap in terms of log-points was 0.09042 in 1985,
-0.00142 in 1995, and -0.08492 in 2005. The negative value of the explained gap
revealed that females on average had more favorable observed characteristics. In other
words, if observed characteristics were the only factors used in determining earnings, and
these characteristics were compensated for at the same rate for both genders, then on
average we should observe higher earnings for females compared to males. However, we
do not observe higher average earnings for females compared to males in the data. The
reason for this is that the unexplained portion of the gap, although declining over time
(0.24919 in 1985, 0.21012 in 1995, and 0.17469 in 2005), was still larger than the
explained gap. Thus, the unexplained gap outweighed the opposing impact of the
explained gap. In the end, we observe a positive but declining total earnings gap.
Panel B of Table 1.6A separates the explained portion of the gap into several
segments that account for each of the observable characteristics. It is apparent that the
negative explained gaps, observed in 1995 and in 2005, were heavily influenced by the
education factor. In 1985, the education factor was positive and very small (0.00043),
79
slightly contributing to superior male earnings. However by 1995, the education factor
had become negative (-0.04069). By 2005, the magnitude of the education factor had
doubled (-0.08227). This negative contribution of the education factor suggested that if
education were the only factor determining male and female earnings, then on average,
females would earn higher than males. These results are consistent with those of
Phananiramai (1993). Phananiramai performed the BO analysis for the gender earnings
gap in Thailand using Thai LFS data for the years 1980 and 1989. The author observed
the narrowing of this gap and documented improvements in female observed
characteristics, especially in terms of education, during the period of study. In her
findings, the contribution of the education factor was negative, indicating superior
schooling for females in the wage and salary sector.
The results in Panel B of Table 1.6A further show how observable characteristics
other than education contributed to reductions in the total gender earnings gap. However,
the impact of these other characteristics seemed minimal compared to that of education.
In each year (1985, 1995, 2005), age and marriage factors were positive but declining.
Thus, both factors contributed to males earning more than females, but their contributions
became smaller over time. Panel C of Table 1.6A separates the residual gap into a portion
reflecting the “advantage of males” and a portion reflecting the “disadvantage of
females”.
40
The results show that, in all years (1985, 1995, 2005), the disadvantage of
females accounted more for the residual gap than the advantage of males.
It can be stimulating ask why the education of females in the wage and salary
sector became far superior to that of males. As illustrated by the BO decomposition

40
These terms correspond to the second and last terms on the right-hand side of expression (4).
80
results for 1985, average female characteristics, especially terms of education, were still
inferior to those of males. However, this phenomenon of inferior female characteristics
was no longer true by 1995 and 2005.
One possible reason why females outstripped males in terms of education is that
during the period of Thailand’s modernization, traditional beliefs that privileged sending
boys rather than girls to school had abraded. The modernization period also marked an
increase in the public recognition of the benefits of educating girls. Various attempts had
been made on the part of the Thai government and other non-profit organizations to
promote education for girls.
41
As a result, the education of female workers in the wage
and salary sector, along with the education of Thai females as a whole, increased.
Another possible reason for this increase in female education from 1985-2005 is
that the majority of the women employed came from different parts of the overall
(observable) skills distribution. Previously, women from any part of the distribution had
been employed. However, as time progressed, more and more women from the upper-end
of the (observable) skills distribution were employed, while fewer women from the
lower-end of the (observable) skills distribution were employed. Thus, the characteristics
of female workers in this sector improved. For the period 1995-2005, the financial crisis
might also have played a role. Inferior-characteristic females might have been
disproportionately eliminated from the market due to the crisis. However, whether the
crisis had weeded out these inferior-characteristic females, the trend reflecting improving
female characteristics in Thailand’s wage and salary sector had already been ongoing
prior to the crisis.

41
See details in Section 1.2.2.
81
Table 1.6B carries out the BO analysis as per the specification that includes
occupation and industry variables. The narrowing of the explained gap was observed,
with the gap becoming negative by 2005. Like the results in Table 1.6A, the unexplained
gap was larger than the explained gap, with the unexplained gap declining over the time
period.
For each year (1985, 1995, 2005), the explained gap as per the extended
specification including occupation and industry variables (Table 1.6B) was “more
positive” compared to the explained gap as per the base specification excluding
occupation and industry variables (Table 1.6A). In other words, males are “more
superior” to females in the extended model than in the base model. This can be easily
explained for the year 1995. Under the extended model, the explained gap was positive
(0.02637) in 1995, indicating that females had inferior characteristics compared to males.
However, the explained gap in the base model for the year 1995 was negative (-0.00142),
indicating that females had superior characteristics compared to males. These results
suggest that including occupation and industry variables into the analyses ends up
negatively impacting female characteristics relative to those of males. This evidence
supports the notion that females were, in fact, being allocated to lower-paying
occupations and industries compared to males.
42

42
This discrepancy between the base and extended models can also be explained for the years 1985 and
2005. Specifically, in 1985, the explained gap under the extended specification was 0.14169, which was
larger than the one under the base specification (0.0942). The positive explained gaps indicate superior
male characteristics compared to those of females. The larger the gap, the more superior the characteristics
of the males were compared to those of females. In 2005, the explained gaps under both models were
negative, indicating superior characteristics of females compared to those of males. The gap under the
extended model (-0.05799) was “more positive” than the gap under the base model (-0.08492), indicating
that females were less superior to males in the extended model compared to in the base model.
82
Table 1.6B: Blinder-Oaxaca (1973) results (with occupations and industries)

Panel A:
1985 1995 2005
Total Difference (T) 0.33959 100.00% 0.20870 100.00% 0.08978 100.00%
Explained (E)
0.14169
41.72%
0.02637
12.64%
-0.05799
-64.60%
Unexplained (U) 0.19790 58.28% 0.18233 87.36% 0.14777 164.60%

Panel B:
Explained (Details)
1985 1995 2005
Education 0.00124 0.36% -0.02818 -13.50% -0.04603 -51.27%
Primary
-0.00500 0.00156 0.00828
Lower Secondary
0.02115 0.01561 0.01570
Upper Secondary
0.00164 0.00339 0.00818
University
-0.01656 -0.04873 -0.07819
Age&Age-Squared
0.05386 15.86% 0.03914 18.75% 0.01368 15.23%
Age
0.15465 0.10459 0.03013
Age-Squared
-0.10079 -0.06545 -0.01645
Urban -0.00053 -0.15% -0.00206 -0.99% -0.00365 -4.06%
Married 0.02649 7.80% 0.00832 3.99% 0.00273 3.04%
All Regions 0.00199 0.59% -0.00967 -4.63% -0.01709 -19.04%
All Occupations 0.01796 5.29% -0.01021 -4.89% -0.02723 -30.34%
All Industries 0.04066 11.97% 0.02902 13.91% 0.01960 21.83%
Explained (E)
0.14169
41.72%
0.02637
12.64%
-0.05799
-64.60%

Panel C:

Unexplained (Details)
1985 1995 2005
Advantage of Men 0.07722 22.74% 0.07494 35.91% 0.06707 74.71%
Disadvantage of Women 0.12068 35.54% 0.10739 51.46% 0.08070 89.89%
Unexplained (U) 0.19790 58.28% 0.18233 87.36% 0.14777 164.60%

83
Observing the contributions of occupation and industry factors under the extended
model over time (Table 1.6B), we can see that these contributions to the total gap became
smaller. These declining contributions of the occupation and industry factors indicated a
shift toward the allocation of female labor into higher-paid careers over time.
1.5.4 JMP Results
Table 1.7A illustrates the JMP decomposition results (expression (10)) under the
base specification. The data is divided into two sub-periods, 1985-1995 and 1995-2005.
Panel A of the table summarizes the findings. The closing of the gap over time is
represented by a negative number while the widening of the gap over time is represented
by a positive number. As confirmed by the BO analysis, for both sub-periods (1985-1995
and 1995-2005) the total earnings gap declined. The gap declined by 13.09% from 1985-
1995 and by 11.89% from 1995-2005. The majority of the reduction in the total gap was
accounted for by the decline in the explained gap. From 1985-1995, the reduction in the
explained gap accounted for 9.8% of the overall reduction of 13.09%. From 1995-2005,
the reduction in the explained gap accounted for 8.35% of the overall reduction of
11.89%.
Panel B of Table 1.7A shows how each of the observable characteristics plays a
part in the reduction of the explained gap. According to the JMP methodology, each of
the observable characteristics contributes to the reduction of the explained gap in two
ways, by way of changes in quantities (first term on the right-hand side of equation (10)),
and by way of changes in prices (second term on the right-hand side of equation (10)).
From 1985-1995 and from 1995-2005, the quantity factor was the major source of the
84
Table 1.7A: Juhn-Murphy-Pierce (1991) results (without occupations and industries)

Panel A:
1985-1995 1995-2005
Change in Total Difference (dT) -0.13091 -0.11892
Change in the Explained Part (dE) -0.09184 -0.08350
Change in the Unexplained Part (dU) -0.03907 -0.03542

Panel B:
Change in the Explained Part (Details)
1985-1995 1995-2005
Q P Q+P Q P Q+P
Education -0.03455 -0.00657 -0.04112 -0.02420 -0.01738 -0.04158
Primary
0.00769 0.00057 0.00973 0.00025
Lower Secondary
-0.01144 0.00086 0.00577 -0.00210
Upper Secondary
0.00166 0.00019 0.00591 0.00195
University
-0.03246 -0.00819 -0.04561 -0.01748
Age&Age-Squared
-0.02382 0.00594 -0.01788 -0.03326 0.00709 -0.02618
Age
-0.04557 -0.01965 -0.06735 -0.01836
Age-Squared
0.02174 0.02559 0.03409 0.02545
Urban -0.00265 0.00103 -0.00162 -0.00076 -0.00192 -0.00267
Married -0.00970 -0.00914 -0.01884 -0.00571 -0.00098 -0.00669
All Regions -0.01070 -0.00167 -0.01237 -0.00107 -0.00531 -0.00639
Change in Explained Part (dE) -0.08142 -0.01042 -0.09184 -0.06500 -0.01850 -0.08350

Panel C:
Change in Unexplained Part (Details)
1985-1995 1995-2005
Q P Q+P Q P Q+P
Change in Unexplained Part (dU) -0.01962 -0.01945 -0.03907 -0.04880 0.01337 -0.03542
85
gap reduction. From 1985-1995, the improvement in overall observed characteristics of
females accounted for 62.20% (0.08142/0.13091) of the overall convergence of male and
female earnings. The education-quantity factor accounted for 26.39% (0.03455/0.13091),
and the age-quantity factor accounted for 18.20% (0.02382/0.13091) of the total decline
during the time period. From 1995-2005, the overall observed characteristics accounted
for 54.66% (0.06500/0.11892) of the converging gap, with the education-quantity factor
accounting for 20.35% (0.02420/0.11892), and the age-quantity factor accounting for
27.97% (0.03326/0.11892), of the total decline. These results confirmed the findings
from the BO analysis that indicated significant improvements of female observable
characteristics during the time period. Unlike the (observable) quantity factors, the
(observable) price factors were trivial and did not contribute much to the convergence.
Panel C of Table 1.7A investigates the closing of the unexplained component of
the total gender earnings gap, which according to the JMP methodology is composed of
the quantity differences of the unobservable characteristics (residual quantity gap) and
the prices of those unobservable characteristics. From 1985-1995 and from 1995-2005,
the results indicate a decline in the residual quantity gap. From 1985-2005, the decline in
the residual quantity gap accounted for 14.99% (0.01962/0.13091) of the reduction in the
total gap, while from 1995-2005 the decline in the residual quantity gap accounted for
41% (0.04880/0.11892) of the reduction in the total gap. The 41% contribution of the
residual quantity gap to the overall decline during 1995-2005 was the largest contribution
compared to other factors during the period.
86
There are at least four plausible hypotheses that can explain this closing of the
residual quantity gap between the two genders.
43
First, modernization brought in new
traditions regarding family institutions and gender roles, thus allowing women to
participate more in the labor market. In many Thai families, unlike in the past, household
chores are also being allocated to male family members. Modern couples have fewer
children than in the past. According to UNICEF statistics, there has been a remarkable
decline in total fertility rate in Thailand during the period. The total fertility rate was 5.5
in 1970, but it dropped to 2.2 and 1.9 in 1990 and 2005, respectively.
44
The modern
economy also provides various amenities that facilitate household tasks. For example,
children can be put in day-care centers and domestic workers can be hired. All of these
new developments in society have allowed women to dedicate more time and effort
towards career advancement.
A second hypothesis is that women may have accumulated skills that are useful in
the workforce but are not measurable in the data. Unlike in the past, women are now
engaged in various social activities outside the household. This exposure to the public
domain allows women potentially to develop their analytical, communication, and
leadership skills, all of which are essential in career progressions.
Third, it can be argued that during the time period, women who were employed in
the wage and salary sector did not come from the same part of the overall (unobserved)
characteristics distribution. Initially, the market might have allowed women with lower
(unobserved) skills to be employed. However, in later periods, the market

43
In other words, I look for possible reasons why female unobservable characteristics improved during the
period.
44
http://www.unicef.org/infobycountry/Thailand_statistics.html
87
disproportionately allowed only women with high ability to enter. Resembling the
argument made earlier for the case of observed characteristics, the financial crisis might
also have played a role by weeding out more of the lower-ability females in the market
compared to other type of workers. A final hypothesis is that gender discrimination in
earnings, if previously present, must have declined from 1985-2005.
Table 1.7B demonstrates the JMP results under the extended model in which
occupation and industry variables are included. Note that only the analysis from 1985-
1995 can be performed since occupational and industry codes changed in 2001 and
industry codes cannot be converted. From 1985-1995, a reduction of 13.1% in the overall
earnings gap was accounted for by both the reduction in the explained gap (11.5%) and
the reduction in the unexplained gap (1.6%). The major contributors to this reduction in
the total gender earnings gap were the education-quantity and age-quantity factors. The
overall findings under this model were similar to those for the base model.
The results indicating a decline in the explained gap due to improved female
observed characteristics are consistent with studies from some other countries in the
region, namely Taiwan, Korea, and Vietnam.
45
In Taiwan, Zveglich, Rogers, and Rogers
(1997) documented the improvement of Taiwanese female characteristics, such as
education and experience, during 1978-1992. Unlike the case of Thailand, Taiwan’s
residual gap grew significantly enough to cause a fixed male-female earnings gap during
their study period. In Korea, Rodgers (1998) showed that female observed characteristics
improved significantly during 1983-1992, resulting in a narrowing of the gender gap

45
Detailed discussions regarding gender earnings inequality in Thailand’s neighboring countries are given
in Section 1.2.3.
88
Table 1.7B: Juhn-Murphy-Pierce (1991) results (with occupations and industries)

Panel A:
1985-1995
Change in Total Difference (dT) -0.13089
Change in the Explained Part (dE) -0.11532
Change in the Unexplained Part (dU) -0.01557

Panel B:
Change in the Explained Part (Details)
1985-1995
Q P Q+P
Education -0.02127 -0.00814 -0.02941
Primary
0.00609 0.00047
Lower Secondary
-0.00803 0.00249
Upper Secondary
0.00104 0.00071
University
-0.02037 -0.01181
Age&Age-Squared
-0.02016 0.00543 -0.01472
Age
-0.03761 -0.01244
Age-Squared
0.01746 0.01788
Urban -0.00078 -0.00075 -0.00153
Married -0.00892 -0.00925 -0.01817
All Regions -0.01037 -0.00130 -0.01167
All Occupations -0.02658 -0.00159 -0.02817
All Industries 0.00175 -0.01339 -0.01164
Change in Explained Part (dE) -0.08632 -0.02899 -0.11532

Panel C:
Change in Unexplained Part (Details)
1985-1995
Q P Q+P
Change in Unexplained Part (dU) -0.00141 -0.01417 -0.01557
89
which followed a decade of stasis in the earnings gap. In Vietnam, Liu (2004) reported an
improvement in female observable characteristics during 1993-1998. In fact, in 1998, Liu
observed a negative explained gap indicating superior characteristics of females in the
Vietnamese wage and salary sector.
1.6 Nonparametric Decomposition
The BO analysis presented in the previous section illustrates how the gender
earnings gap at a time point can be decomposed at their mean values. Although the BO
method is useful, some apparent limitations regarding the use of this method have been
pointed out in the literature. First, the BO method cannot decompose the entire earnings
distribution. Second, due to its parametric nature, this method assumes a particular form
of relationship between earnings and the observable characteristics of workers.
DiNardo, Fortin, and Lemieux (1996) (DFL) proposed a nonparametric approach
that overcomes these insufficiencies in the BO methodology. Their method allows an
entire distribution of earnings to be decomposed into an unexplained portion and a
portion explained by the differences in the distributions the observable characteristics of
males and females. Moreover, no assumptions regarding the relationship between
earnings and workers’ attributes are needed in this method.
This section is organized as follows. First, I provide a clear explanation of how
the standard DFL methodology can be linked to the BO methodology (where male wage
structure is used as the reference). I then extend the DFL model to a case where the
pooled wage structure is used as the reference. The purpose of introducing this modified
version of the DFL method is to make the analysis compatible with Neumark’s (1988)
90
version of the BO methodology. Finally, the modified DFL model is implemented and
the results are discussed.
1.6.1 DiNardo-Fortin-Lemieux (1996): DFL
The idea behind the DFL methodology is to explain why the distributions of male
and female earnings are different. First, the distribution for the observable characteristics
of male and female workers are different, and second, the wage structures or market
returns to the observable characteristics of male and female workers are different. As in
the BO analysis, the differences in the wage structures for males and females do not
necessarily reflect gender discrimination. These differences, however, reflect
discrepancies regarding the distributions for the unobservable characteristics of males and
females.
According to standard DFL methodology, in order to decompose the earnings
density gap, a counterfactual earnings density needs to be constructed. This
counterfactual earnings density is a hypothetical earnings density that would prevail if
females were to be compensated according to the male wage structure. Since the only
differences between this counterfactual earnings density and the male earnings density
are the dissimilarities in the distributions of the observed characteristics of males and
females, the gap between these two densities is called the explained gap. The gap
between this counterfactual earnings density and the female earnings density captures
everything else other than the differences in the observable attributes; thus this gap is
called the unexplained gap or the residual gap.
91
The relationship between the standard DFL method and the BO method (where
male wage structure is used as the reference) has been elaborated by only a few authors.
Specifically, Barsky, Bound, Charles, and Lupton (2002) and Black, Haviland, Sanders,
and Taylor (2006) demonstrated the linkage between the two methodologies by relaxing
the parametric assumption. Once the assumption regarding the form of conditional
expectation of earnings is relaxed, the BO method (where male wage structure is used as
the reference) can be shown to be a special case of the standard DFL method. Another
way to understand the relationship between the two methods is to consider the methods’
utilization of the reweighing procedure. This similarity between BO and DFL in terms of
reweighing their parameters was discussed by DiNardo (2002) and Lemieux (2002).
The relationship between the standard DFL method and the BO method (where
male wage structure is used as the reference) can be explained as follows. To start, I
would like the reader to take another look at the BO decomposition in expression (5). The
total gap is segregated into an explained gap (first term on the right-hand side) and an
unexplained gap (second term on the right-hand side). Examining closely, one should be
able to see that the explained gap is merely the gap between the actual pay for males
(
M M
b x ) and an extra term (
M F
b x ). The unexplained gap is the gap between this extra
term (
M F
b x ) and the actual pay for females (
F F
b x ). This extra term can be interpreted
as the hypothetical pay received by females when they are compensated according to the
reference wage structure, which is the male wage structure in this case. Thus, the
explained gap illustrates how much the quantity differences in observed characteristics
contribute to the total gap, with the price held constant. The unexplained gap illustrates
92
how wage structure differences contribute to the gap, with the characteristics held
constant. This rationale is illustrated in Panel A of Figure 1.15.
Now, instead of just looking at the mean earnings for males and females, one can
look at the entire earnings distributions for males and females. I denote the male earnings
density function using
M P
M C
f
=
=
. The notation emphasizes that the male earnings density
function is the density function in which male characteristics (subscript C=M) are being
compensated for according to the male wage structure (superscript P=M). Similarly, the
female earnings density function can be denoted by
F P
F C
f
=
=
. Normally, males receive
higher earnings than females, and the density function of males should lie to the right of
that of females. This is true for Thailand, as previously shown in Figure 1.12 and 1.13.
Following standard DFL methodology, a counterfactual earnings density (a
hypothetical earnings density for females when their characteristics are compensated for
according to the male wage structure) is constructed. This counterfactual earnings density
is denoted by
M P
F C
f
=
=
. The three earnings density graphs are drawn in Panel B of Figure
1.15. Analogous to the case of BO decomposition, the explained gap is determined by the
distance between the male earnings density (
M P
M C
f
=
=
) and the counterfactual earnings
density (
M P
F C
f
=
=
). This gap measures the differences in the distributions for male and
female characteristics, with the price held fixed. The unexplained gap is assessed by the
distance between the counterfactual earnings density (
M P
F C
f
=
=
) and the female earnings
density (
F P
F C
f
=
=
). This gap measures the differences in male and female rates of returns to
observable characteristics.
93

Panel A: Blinder-Oaxaca (1973)

Panel B: DiNardo-Fortin-Lemieux (1996)

Figure 1.15: Relationship between Blinder-Oaxaca (1973) and DiNardo-Fortin-Lemieux (1996)
when male wage structure is used as the reference wage structure
94
Recall that Neumark (1988) proposed using the pooled wage structure in place of
the male wage structure as the reference. The formula for Neumark’s (1988) version of
the BO analysis is shown in (4). The BO analysis implemented in Section 1.5 follows this
formula. In expression (4), the explained gap (first term on the right-hand side) represents
the differences between what men would be paid according to the pooled wage structure
(
A M
b x ) and what women would be paid according to the pooled wage structure (
A M
b x ).
Note that these two terms are, in fact, hypothetical since male characteristics are actually
rewarded according to the male wage structure (i.e., they are paid
M M
b x ). Similarly,
women are actually paid
F F
b x . Under this representation, the pooled wage structure is
chosen as the reference, and two hypothetical pay variables are included in the equation.
The unexplained gap is composed of the last two terms on the right-hand side. The first
term, namely the “advantage of males,” is the gap between the actual pay for males
(
M M
b x ) and the hypothetical pay for males (
A M
b x ). This term captures how much
higher males are paid compared to what they would be paid in the “ideal society”.
Similarly, the second term or the “disadvantage of females” is captured by the distance
between the hypothetical pay for females (
A F
b x ) and the actual pay for females (
F F
b x ).
Panel A of Figure 1.16 provides a visual illustration of these concepts.
To incorporate a Neumark (1988)-like index into the standard DFL model, I
propose the construction of two counterfactual earnings densities. The first density is a
hypothetical earnings density if male characteristics (C=M) are compensated for
according to the pooled wage structure (P=A). This hypothetical earnings density
function is denoted by
A P
M C
f
=
=
. The second density is a hypothetical earnings density if
95

Panel A: Blinder-Oaxaca (1973)

Panel B: DiNardo-Fortin-Lemieux (1996)

Figure 1.16: Relationship between Blinder-Oaxaca (1973) and DiNardo-Fortin-Lemieux (1996)
when pooled wage structure is used as the reference wage structure
96
female attributes (C=F) are paid according to the pooled wage structure (P=A). This
hypothetical earnings density is denoted by
A P
F C
f
=
=
. Note that this modified DFL method is
analogous to Neumark’s (1988) version of the BO method in the way that the pooled
wage structure is used as the reference. The gap between the actual earnings density of
males (
M P
M C
f
=
=
) and the counterfactual earnings density of males (
A P
M C
f
=
=
) corresponds to
the “advantage of males” portion of the BO model. Similarly, the gap between the
counterfactual earnings density of females (
A P
F C
f
=
=
) and the actual earnings density of
females (
F P
F C
f
=
=
) corresponds to the “disadvantage of females” portion of the BO model.
The unexplained portion of this modified DFL decomposition is comprised of these two
gaps. Apparently, the explained portion of this modified DFL decomposition can be
assessed from the distance between the two hypothetical earnings densities (
A P
M C
f
=
=
and
A P
F C
f
=
=
). This distance measures the differences in the observable characteristics between
males and females, with the price (pooled price) held constant. Panel B of Figure 1.16
illustrates these ideas.
The counterfactual earnings densities in this modified DFL methodology can be
implemented in ways similar to how the counterfactual earnings density in the standard
DFL method is implemented. One difference is that, in the modified model, the reference
wage structure is derived from the earnings density of all workers, whereas in the
standard model the reference wage structure is derived from the earnings density of only
male workers. In the following expressions, y represents log earnings and x is a vector of
observable attributes. As in BO and JMP, these attributes include education, age, age-
squared, marital status, community characteristics, and regional information. Two
97
versions of the analysis are implemented, one excluding occupation and industry
variables (base specification), and the other including occupation and industry variables
(extended specification).
Similar to the standard DFL, the marginal density of earnings of all workers
( ) (y f ) can be written in terms of the joint density between workers’ earnings and
workers’ observed attributes ( ) , ( x y f ) integrated over the domain of all possible vectors
of the attributes (
x
Ω ). The joint density can then be replaced by the conditional density
of earnings ( ) | ( x y f ) multiplied by the marginal density of the attributes ( ) (x g ). The
implementation can be shown as follows:
∫
∫
Ω ∈
Ω ∈
=
=
= = =
x
x
x
x
dx x g x y f
dx x y f
A C A P y f y f
. ) ( ) | (
) , (
) ; ; ( ) (
(11)
To create the male counterfactual earnings density (the density that depicts the
earnings men would receive if their observable characteristics were compensated for
according to the pooled wage structure) expression (11) needs to be modified.
Specifically, the set of attributes that needs to be integrated over is that of males. Thus,
) (x g , the marginal density of the attributes of all workers, is replaced by ) (x g
M
, the
marginal density of the attributes of male workers. DFL pointed out the complication of
integrating the resulting formula due to x being a vector. Thus, they proposed a
reweighing procedure to simplify the implementation. Following DFL, I reweigh the
function, using
M
ω as the reweighing parameter. The procedure is shown below:
98
. ) ( ) | (
) (
) (
) (
) | (
) ( ) | (
) ; ; ( ) (
∫
∫
∫
Ω ∈
Ω ∈
Ω ∈
=
=
⋅ ⋅ =
=
=
= = =
x
x
x
x
M
x
M
x
M
A P
M C
dx x g x y f
dx x g
x g
x g
x y f
dx x g x y f
M C A P y f y f
ω
(12)
The male reweighing parameter,
M
ω , can be simplified using Bayes’s rule:
.
) (
) | (
) (
) | (
) (
) (
M C P
x M C P
x g
M C x g
x g
x g
M M
=
=
=
=
= = ω
(13)
The construction of the female counterfactual earnings density (the hypothetical
density that depicts the earnings females would receive if their observable characteristics
were compensated for according to the pooled wage structure) can be carried out in a
similar fashion. The procedure is illustrated below:
. ) ( ) | (
) (
) (
) (
) | (
) ( ) | (
) ; ; ( ) (
∫
∫
∫
Ω ∈
Ω ∈
Ω ∈
=
=
⋅ ⋅ =
=
=
= = =
x
x
x
x
F
x
F
x
F
A P
F C
dx x g x y f
dx x g
x g
x g
x y f
dx x g x y f
F C A P y f y f
ω
(14)
) (x g
F
is the marginal density of the attributes of female workers. The female
reweighing parameter,
F
ω , can also be simplified accordingly:
99
.
) (
) | ( 1
) (
) | (
) (
) | (
) (
) (
F C P
x M C P
F C P
x F C P
x g
F C x g
x g
x g
F F
=
= −
=
=
=
=
=
= = ω
(15)
Like the standard DFL case, the probability term, ) | ( x M C P = , can be estimated
from a logit regression.
46
The terms ) ( M C P = and ) ( F C P = can be obtained from the
data by calculating the proportion of workers who are males and the proportion of
workers who are females.
Once we get estimates of the reweighing parameters (
M
i
ω ˆ and
F
i
ω ˆ ), the standard
kernel density estimation method needs to be modified. The formulas for the
counterfactual earnings density functions of males and females are illustrated below:
∑
=
=
=
−
=
n
i
i
M
i i A P
M C
h
y y
K
h
w
h y f
1
) (
ˆ
) ; (
ˆ
ω
(16)
. ) (
ˆ
) ; (
ˆ
1
∑
=
=
=
−
=
n
i
i
F
i i A P
F C
h
y y
K
h
w
h y f
ω
(17)
i
y is log earnings,
M
n is the number of male observations,
F
n is the number of
female observations, and
i
w is the survey weight from the data. The bandwidth (h) and
the kernel function (K) are the same as in (1). Expression (16) gives the counterfactual
earnings density for males if their observed characteristics were to be compensated for
according to the pooled wage structure. Similarly, expression (17) gives the

46
Specifically, the logit model can be written as η θ + = x p . Here, p is 1 if the observation is male and 0
otherwise. x is the vector of the observable attributes. η is the residual. The value of θ
ˆ
x will provide an
estimate for the probability term ) | ( x M C P = .
100
counterfactual earnings density for females if their observed characteristics were to be
compensated for according to the pooled wage structure.
1.6.2 DFL Results
Figure 1.17A (Panel A, B, and C) illustrates the results from the modified DFL
analyses for the years 1985, 1995, and 2005 under the base specification (without
occupation and industry variables). In Figure 1.17A, I plot the male counterfactual
earnings density (male characteristics compensated for according to the pooled wage
structure) and the female counterfactual earnings density (female characteristics
compensated for according to the pooled wage structure) in addition to the actual
earnings densities of males and females. The male earnings density is represented by the
long-dash line, the female earnings density is represented by the short-dash line, the male
counterfactual earnings density is represented by the solid line, and the female
counterfactual earnings density is represented by the dash-dot line.
In Panel A of Figure 1.17A (the decomposition results for the year 1985), the
female earnings density lay significantly to the left of the male earnings density,
illustrating that in general females earned lower than males. Both of the counterfactual
earnings densities lay between the male and female earnings densities. For males, this
allocation implies that if their characteristics were to be compensated for using the pooled
wage structure, then they would earn less than what they actually do. However, they
would still earn more than females. For females, this implies that if their characteristics
were to be compensated for using the pooled wage structure, they would earn more than
what they actually do. However, they would still earn less than males. Recall that in the
101

Panel A: 1985

Figure 1.17A: Modified DiNardo-Fortin-Lemieux (1996) results (without occupations and industries)
102

Panel B: 1995

Figure 1.17A, Continued
103

Panel C: 2005

Figure 1.17A, Continued
104
modified DFL analysis, the distance between the two counterfactual earnings densities is
the explained gap or the endowment gap. The figure shows that the male counterfactual
earnings density lay to the right of the female counterfactual earnings density indicating
more favorable observed characteristics for males compared to females. These results
from the modified DFL density decomposition are consistent with the results from
Neumark’s (1988) version of the BO decomposition (Table 1.6A). According to the BO
results in 1985, the explained gap was positive, indicating that the average male had
superior characteristics compared to the average female. Looking at the gap between
male earnings density and the counterfactual male earnings density (“advantage of
males”) and the gap between the counterfactual female earnings density and female
earnings density (“disadvantage of females”), one can see that the two gaps, considered
in combination (the unexplained gap), constituted the majority of the earnings
discrepancies between males and females. These results are also consistent with the
results from the BO analysis that the majority of the gap evaluated at mean was
accounted for by the unexplained gap.
As for Panel B of Figure 1.17A (decomposition results for 1995), female earnings
density still lay slightly to the left of the male earnings density, but it was closer to the
male earnings density compared to 1985. Both of the counterfactual earnings densities
lay between the actual earnings densities of males and females; the counterfactual
densities almost lay on top of each other. Recall that the distance between the two
counterfactual densities represents the explained portion of the earnings differentials.
This observation indicates a significant improvement in the attributes of females relative
to males during 1985-1995. Women would earn almost exactly the same as their male
105
counterparts if both males and females were paid according to the pooled wage structure.
Looking back at the BO analysis which evaluates the earnings differentials at the mean
values, the average female in 1995 had similar (in fact, slightly superior) overall
characteristics compared to the average male. This is illustrated by the very small
negative endowment gap. Regarding the contribution of the unexplained gap, most of the
mean earnings gap in BO was accounted for by the unexplained gap. Corroborating the
results from the BO model, the results from this modified DFL model also show that
most of the earnings density gap between male and female earnings was accounted for by
the unexplained portion.
Panel C of Figure 1.17A shows the decomposition for the 2005 data.
Unfortunately, the male and female earnings densities almost coincided completely.
Thus, the resulting decomposition cannot be clearly interpreted.
Figure 1.17B (Panel A, B, and C) displays the results from the extended model
where occupation and industry variables are included. Similar to that for the base model,
the decomposition results for the year 1985 (Panel A) indicates superior observed
characteristics of males compared to those of females. The explained gap, the gap
between the two counterfactual earnings densities, was, however, larger under this model
compared to the base model, corroborating the BO decomposition results that including
occupation and industry variables degraded females’ observed characteristics in
comparison to males.
106

Panel A: 1985

Figure 1.17B: Modified DiNardo-Fortin-Lemieux (1996) results (with occupations and industries)
107

Panel B: 1995

Figure 1.17B, Continued
108

Panel C: 2005

Figure 1.17B, Continued
109
For the year 1995 (Panel B), most of the male counterfactual earnings density lay
to the right of the female counterfactual earnings density. However, this was not true for
upper-end high-income females. Comparing these results to the results from the base
model (Panel B of Figure 1.17A), we can see that the allocation of workers into
occupations and industries put most females in lower-paying jobs. However, very few
females, especially the ones with exceptional skills, ended up in higher-paying careers.
The corresponding BO analysis (Table 1.6B) showed that, on average, when occupations
and industries were taken into accounted, female observable characteristics were inferior
in comparison to males. Unlike the modified DFL model, the BO model failed to capture
the phenomenon of the very few exceptional females ending up in positions which were
superior to those of most males.
Panel C of Figure 1.17B displays the results from the extended model for the year
2005. Like the 2005 results under the base specification, the densities lay very close to
one another; therefore the results cannot be easily interpreted.
1.7 Conclusion
Along with the well-documented phenomena of Thailand’s remarkably increasing
real income per capita, significantly declining poverty, and moderately rising income
inequality during the past two decades, gender earnings inequality in Thailand improved
at an impressive rate. This paper examines the social and economic factors that
contributed to the narrowing of the gender earnings gap in Thailand’s wage and salary
sector from 1985-2005.
110
Two parametric methodologies, in addition to a nonparametric methodology, are
implemented in order to decompose the gender earnings gap. Neumark’s (1988) version
of the Blinder (1973) and Oaxaca (1973) (BO) method is applied to examine
parametrically the gender earnings gap for 1985, 1995, and 2005. The Juhn, Murphy, and
Pierce (1991) (JMP) method is utilized to examine parametrically the gender earnings
gap from 1985-1995 and 1995-2005. A modified version of the nonparametric
decomposition proposed by DiNardo, Fortin, and Lemieux (1996) (DFL) is also
implemented. The DFL approach allows an entire earnings distribution to be
decomposed, unlike the parametric approaches where only mean earnings can be
investigated. The DFL approach allows flexibility regarding the form of the relationship
between the earnings and the observable characteristics of workers, unlike the parametric
approaches where some assumptions regarding the relationship need to be made. A
methodological contribution that I make in this paper is my modification of the standard
DFL method to allow for the use of a more general form of the reference wage structure.
This modification is intended to make the DFL analysis comparable to Neumark’s (1988)
version of the BO analysis.
The results show that a substantial increase in the education of females was the
major source of the narrowing of the gender earnings gap in Thailand from 1985-2005. In
fact, in the wage and salary sector, improvements in the education of female workers
surpassed that of male workers. In later years, females have possessed superior
observable characteristics compared to males, but did not earn higher income than males
due to unobservable factors. However, these unobservable factors have been changing in
ways that have improved the position of females in the Thai labor market.
111
Modernization was likely the key factor that caused the increase in the education
of females and the improved status of females in the labor market. Over time, traditional
views of gender roles have eroded, and the importance of investing in the education of
females has been brought to the public’s attention. Although, at the moment, the gender
earnings gap still exists, the results of this study suggest that females are in far better
positions than they used to be two decades ago. The social and economic conditions in
Thailand have indeed been moving in a direction that favors women more than in the
past. It will be interesting to see how this trend evolves in the coming years. It will also
be interesting to see whether other developing countries are also moving towards a
similar trend of gender equality. As discussed in the literature (UNDP, 1995; World Bank
2005), gender inequality can exacerbate social, political, and cultural conflicts. Thus, a
world in which gender equality prevails should be the desirable goal of all.

112

Chapter Two

Wrongful Discharge Laws
and the Unexpected Substitution Effect

2.1 Introduction
The issue of employment protection has been quite controversial in the literature.
There is no agreement as to whether a country should provide such protection, and if so,
at what level. Different countries have different views regarding the optimal levels of
employment protection. Figure 2.1 illustrates the World Bank’s estimates for firing costs,
for each economy, in terms of weekly wages.
1
Firing costs include the costs of legally
mandated periods of advance notices of dismissals, severance payments for workers, and
possible fines. Across economies, firing costs are highest in Sub-Saharan African and
South Asian countries, and are lowest in OECD countries.
Figure 2.2 illustrates the estimated firing costs for OECD countries. Even among
countries in this group, there is very little consensus as to what the level of employment
protection should be. The cost of dismissal, in terms of weekly wages, ranges from 95
weeks in Portugal, 69 weeks in Germany, 28 weeks in Canada, to 0 weeks in the United
States.
As shown in Figure 2.2, the US is a country where the labor force is barely
protected. The US followed the common law doctrine of employment at will which was

1
This methodology for estimating firing costs was first proposed by Botero, Djankov, La Porta, Lopez-de-
Silanes, and Shleifer (2004).
113

37.8
26.1
56.1
55.6
25.7
66
68.3
0
10
20
30
40
50
60
70
80
90
100
East Asia &
Pacific
Eastern
Europe &
Central Asia
Latin America
& Caribbean
Middle East &
North Africa
OECD South Asia Sub-Saharan
Africa
Economy
Firing costs (weeks of wages)

Source: World Bank’s Doing Business Report
Figure 2.1: Firing costs across economies (2007)
114

4
2
16
28
0
26
32
69
24
13
24
2
4
91
39
17
0
13
95
56
26
13
22
0
0
10
20
30
40
50
60
70
80
90
100
Australia
Austria
Belgium
Canada
Denmark
Finland
France
Germany
Greece
Iceland
Ireland
Italy
Japan
Korea
Luxembourg
Netherlands
New Zealand
Norway
Portugal
Spain
Sweden
Switzerland
United Kingdom
United States
Country
Firing costs (weeks of wages)

Source: World Bank’s Doing Business Report
Figure 2.2: Firing costs across OECD countries (2007)
115

established throughout the country by 1953. This doctrine maintains that an employment
relationship is voluntary and indefinite, and thus can be unilaterally terminated by each
party at any time for any or no reason and without any liability. One obvious benefit of
the at will regime is that it allows firms not only to efficiently adjust their workforces in
response to demand shocks, but also to shed their least productive workers.
During the 1970s and 1980s, state courts imposed limitations on the employment
at will convention by allowing three classes of exceptions. These exceptions are often
referred to as Wrongful Discharge Laws (hereafter, WDLs). These laws limit employers’
flexibility in dismissing workers and allow workers to litigate against wrongful discharge.
These laws apply to all sizes of firms and all full-time workers. The three categories of
these laws – namely, Implied Contract, Good Faith, and Public Policy – will be discussed
in detail in Section 2.2.
Most economists conjecture that WDLs increase firing costs for firms. Thus, these
laws may affect the way firms manage their workforces. In terms of employment, while
Miles (2000) found no statistically significant effects of WDLs on either employment or
unemployment, Dertouzos and Karoly (1992) and Autor, Donohue, and Schwab (2006)
found that WDLs reduced employment. In conclusion, the literature found a negative or
at best zero impact of WDLs on employment. In terms of wages, although Dertouzos and
Karoly (1992) found a negative impact of WDLs on wages, Autor, Donohue, and Schwab
(2004) pointed out that Dertouzos and Karoly’s (1992) findings could have been biased
due to the use of problematic instrumental variables. Recent work by Miles (2000) and
Autor, Donohue, and Schwab (2006) found that WDLs did not impact wages. A possible
explanation is that the wages received by workers are downwardly rigid at, least in the
116

short run. Thus, the burden of firing costs cannot be placed directly on workers by a
reduction in their wages.
Overall, the literature suggested that WDLs, in terms of firing costs, negatively
affected the “average” worker in society. The literature analyzing the impact of WDLs
implicitly assumed that the overall labor force was homogeneous. It did not recognize the
fact that labor can be heterogeneous and that firms may treat different types of labor as
different forms of input. Thus, the imposition of WDLs may influence the decisions of
firms regarding not only the quantity of labor input but also the combination of different
types of labor input. This study attempts to overcome the limitations of the existing
model in which the heterogeneity of labor input is not recognized. In this paper, I treat
labor as heterogeneous (high-skilled and low-skilled labor). Thus, I allow for the
possibility of WDLs having different impacts on different types of labor.
The key finding of this paper is that WDLs, particularly in the case of the Good
Faith law, are associated with increases in employment of high-skilled labor, a result that
may have been unacknowledged in early studies. WDLs, however, adversely affect the
levels of employment for low-skilled workers. This negative impact of WDLs on
employment is consistent with the literature.
This paper is structured as follows. Section 2.2 discusses the literature relating to
employment protection in general. It then elaborates the three categories of WDLs
currently recognized in the US and examines the literature assessing the impact of these
WDLs. Section 2.3 outlines a theoretical framework explaining how WDLs can
potentially affect different types of labor (high-skilled and low-skilled labor) differently.
Section 2.4 describes the Current Population Survey (CPS), the main dataset used in this
117

study. Section 2.5 explains the empirical methodology used in assessing the impact of
WDLs on the labor market. Section 2.6 implements the methodology and discusses the
results. Section 2.7 concludes the paper.
2.2 Previous Literature: The Economics of Employment Law
A significant portion of the literature addressed the issue regarding the effects of
employment protection on various measures of labor market performance. In an empirical
cross-country study, Botero, Djankov, La Porta, Lopez-de Silanes, and Shleifer (2004)
showed that employment protection had a negative effect on labor market outcomes in
terms of lower labor force participation and higher unemployment. A study of the Indian
labor market by Besley and Burgess (2004) showed that Indian states that had amended
the Industrial Disputes Act in a way that gave workers more rights and protections (pro-
worker direction) incurred lower output and employment in their formal manufacturing
sector. In the Latin American labor markets, Heckman and Pages’s (2000) cross-country
analyses provided some evidence that employment protection reduced total employment.
Later work also confirmed such findings (see Heckman and Pages (2004), especially the
chapters by Mondino & Montoya, and Saavedra & Torero). Grubb and Wells (1993)
provided some evidence that labor protection in the European labor market tended to
reduce employment for the protected groups and tended to increase employment in other
groups to which the protection did not apply. Edward Lazear (1990) observed,
theoretically that in a Coasean world, workers would pay for employment protection by
accepting lower wages and there could be no employment effects. However, he
118

empirically observed, using European data, that in the real world, employment protection,
in terms of severance pay requirements, reduced employment.
In the United States, employment protection can be provided at both the Federal
and state levels. Federal statutes which apply to the entire country usually offer protection
for certain classes of workers. Oyer and Schaefer (2000) showed that employers could
find ways to circumvent the Civil Rights Acts (1991) that protects workers based on race
or gender. They could do this by instituting mass layoffs of targeted minorities during
downturns instead of individually dismissing them. Recent work by DeLeire (2000),
Acemoglu and Angrist (2001), and Jolls and Prescott (2004) showed that the Americans
with Disabilities Act (1990) worsened the labor market situation for disabled individuals.
Apart from Federal acts, states adopted three classes of exceptions to employment
at will during the 1970s and 1980s. These exceptions indicate situations under which the
default employment relationship is not at will and workers should not be dismissed
“unfairly”. These exceptions are often referred to as Wrongful Discharge Laws (WDLs).
They will be discussed in detail below.
2.2.1 Wrongful Discharge Laws (WDLs)
2.2.1.1 Implied Contract
When a worker can verify that a permanent employment relationship is promised
by his employer, such employment can no longer be regarded as at will and can be
terminated only under just cause.
2
Just causes usually include employee misconduct such
as theft, insider information trading, or inadequate performance. The implicit promise

2
Toussaint v. Blue Cross & Blue Shield 292 N,W.2nd. 880 (Michigan 1980) and Woolley v. Hoffmann-La
Roch, Inc., 499 A.2d 515 (N.J. 1985).
119

which constitutes an implied contract needs not to be written or even oral. Certain
patterns of employment behavior, such as promotions and salary increases or a history of
positive reviews, can constitute such a contract. Courts can also consider statements in
employee handbooks as a binding contract, as Judge C. J. Wilentz stated in the case of
Woolley v. Hoffmann-La Roche, “it would be unfair to allow an employer to distribute a
policy manual that makes the workforce believe that certain promises have been made
and then to allow the employer to renege on these promises.” So far, courts also have
held an employee’s length of service, a company’s internal policies and past practices,
and typical industry practices as grounds for determining the existence of an implied
contract.
An example of this exception is the case of Pugh v. See’s Candies. The case
established the principle that lengthy employment with regular promotion establishes a
long term contract.
3
Pugh started his employment at See’s Candies as a dish washer.
After almost 30 years of employment with successive promotions, he became a vice
president in charge of production. In 1973, the 32nd year of his service, he was
discharged. The reason indicated was that his service was “no longer required by See’s
Candies.” Interestingly, the actual reason for Pugh’s dismissal stemmed from his
disagreement with the firm’s policy towards its union. It was ruled in court that Pugh’s
length of good service was sufficient to establish an implied contract, and that he could
not be dismissed without just cause.

3
Pugh v. See’s Candies, 116 Cal. App. 3d 311, 171 Cal. Rptr. 917 (1981).
120

The time patterns for the adoption of the Implied Contract exception are
illustrated in Figure 2.3. See also Figure 2.4 which details the geographical extent of
changes in the law. I have also plotted the evolution of the other two exceptions, namely,
the Good Faith and the Public Policy exceptions, which will be discussed below. It is not
completely clear what motivated these changes. Krueger (1991) presented evidence that
these changes in the laws were in response to the uncertainty of the courts in applying
common law exceptions to at will employment. On the other hand, Dertouzos and Karoly
(1993) argued that certain circumstances related to supply and demand for legal doctrines
in each state urge the adoption of Wrongful Discharge Laws.
2.2.1.2 Good Faith
The Implied Contract law requires firms to provide just cause when dismissing
employees who are deemed to be on a long-term contract. Under the Good Faith
exception, workers do not need to have the implicit promise of long-term employment in
order to avoid being dismissed unfairly. The actual interpretation of this exception varies.
The broadest interpretation suggests that all dismissals must be for a reason. Currently,
courts typically implement a rather narrow application of this rule mostly focusing on the
timing of dismissal and employee’s deserved compensation. The law is illustrated in the
case of Mitford v. Lasala.
4
In this case, Mitford was an accountant fired from a job in
which there was a profit-sharing agreement. It was ruled that termination arose to ensure
that Mitford would not be able to receive a share of these profits. The courts ruled that

4
Mitford v. Lasala, 666 P.sd 1000 (Alaska 1963).
121

0
5
10
15
20
25
30
35
40
45
50
1960 1965 1970 1975 1980 1985 1990 1995 2000
Implied Contract Public Policy Good Faith

Source: Illustrated from Autor, Donohue, and Schwab’s (2006) Legal Appendix
Figure 2.3: Number of states adopting Wrongful Discharge Laws

“good faith and fair dealing ... would prohibit firing [an employee] for the purpose of
preventing him from sharing in future profits.” Typical examples of wrongful
terminations that fit under this class are a salesman being fired right before his
commissions should be paid to him and an employee being dismissed so that the
employer can avoid paying him retirement benefits.
As one can see from Figures 2.3 and 2.4, there are far fewer states adopting the
Good Faith law than the Implied Contract law. The Good Faith exception, at its broadest,
protects all workers from being fired for an unfair reason. Muhl (2001) argued that this
122

Implied Contract Exception
February 1983
Implied Contract Exception
December 1994

Good Faith Exception
February 1983
Good Faith Exception
December 1994

Public Policy Exception
February 1983
Public Policy Exception
December 1994

Figure 2.4: Pattern of adoption during 1983-1994
123

exception “represents the most significant departure from the traditional employment-at-
will doctrine” since it “reads a covenant of good faith and fair dealing into every
employment relationship.”
5
It has not been widely documented why only 12 states have
adopted the exception.
6
Perhaps the complexities involved in its broadest interpretation
make the courts reluctant to adopt the law.
2.2.1.3 Public Policy
Under the Public Policy exception, a termination is wrongful if it is a response to
employee conduct which is protected by law but not favored by the employer. The Public
Policy exception covers cases in which an employee is dismissed for refusing to violate a
well-established public policy. Miles (2000) summarizes the four circumstances of
wrongful termination that fit under this class of exception.
7
These are [1] “an employee’s
refusal to commit an illegal act, such as perjury or price-fixing”; [2] “an employee’s
missing work to perform a legal duty, such as jury duty or military service”; [3] “an
employee’s exercise of a legal right, such as filing a workman’s compensation claim”;
and [4] “an employee’s ‘blowing the whistle,’ or disclosing wrongdoing by the employer
or fellow employees.”
An example of this exception is the 1985 case of Tameny v. Atlantic Richfield
Co.
8
Tameny, the dismissed employee, challenged the company’s decision in court,
claiming that the discharge was based on his refusal to perform an unlawful price-fixing

5
Page 10.
6
The total number of states that currently recognize the good faith exception is 11 since New Hampshire
(adopted in 1974) ended its recognition of the rule in 1980.
7
Page 78.
8
Tameny v. Atlantic Richfield Co., 27 Cal.3rd 167 (California 1980).
124

scheme. Atlantic Richfield argued that since there was no employment contract,
Tameny’s employment was at will and could be terminated at any time. The California
Supreme Court ruled in favor of Tameny, stating that an employer cannot discharge an
employee who has obeyed the law and refused to perform an illegal act.
This exception does not, by itself, constitute a new legal regime; rather, it
oversees the behavior of firms, ensuring that these firms do not violate other existing
state and federal laws. Therefore, it would not be surprising if the Public Policy exception
has minimal effects on the US labor market.
2.2.2 Employment Consequences of Wrongful Discharge Laws
With the increasing costs of discharge, firms have a strong incentive to screen
workers more carefully. Kugler and Saint-Paul (2004) found that WDLs made it harder
for unemployed workers to get new jobs relative to currently employed workers. This
was because employers tended to assume that currently employed workers were less
likely to be lemons. Autor (2003) documented the possibility that employers avoid this
cost of discharge by using temporary help agencies, which allows them to refrain from
the liability associated with regular full-time workers. His study showed that employment
of temporary help increased in association with the adoption of the Implied Contract
exception. Schanzenbach (2003) found that, in association with WDLs, full-time workers
tended to have longer tenures. However, he found limited evidence that WDLs helped
increase returns to tenure. In a more recent study, Autor, Kerr, and Kugler (2007)
assessed the productivity implications of WDLs. They found that WDLs significantly
reduced employment flows and were associated with a reduction in total factor
125

productivity. While most studies found no statistical significant effects of WDLs on
wages, the empirical evidence for the effects of WDLs on overall employment is mostly
negative. An early study by Dertouzos and Karoly (1992) reported large negative effects
of WDLs on employment. More recent investigations by Miles (2000) and Autor,
Donohue, and Schwab(2006) reported much smaller effects. In fact, Miles (2000) found
no significant effects while Autor, Donohue, and Schwab(2006) reported consistently
negative effects, particularly for workers with marginal attachment to the workforce.
They found that the negative effect of WDLs on employment was significantly smaller
than that estimated by Dertouzos and Karoly (1992). Autor, Donohue, and Schwab
(2004) explained that the differences between their results and those of Dertouzos and
Karoly (1992) were due to the problematic instruments used by the latter. They also
argued that their results differed from those of Miles (2000) because they used a different
classification of case laws in identifying the adoption dates. They argued that, with their
classification, they “attempt to locate the first case in a state that might trigger a client
letter from attorneys warning about a change in law” and therefore “maximize the chance
of detecting economic effects of changes to the common law.”
9
In this paper, I use the
adoption dates classification developed by Autor, Donohue, and Schwab (2006) who
kindly provided me with the data from their work. The adoption dates of WDLs are
summarized in Table 2.1.

9
Autor, Donohue, and Schwab (2004) page 7.
126

Table 2.1: Adoption dates
State Implied Contract Public Policy Good Faith
ALABAMA AL 7/1987
ALASKA AK 5/1983 2/1986 5/1983
ARIZONA AZ 6/1983 6/1985 6/1985
ARKANSAS AR 6/1984 3/1980
CALIFORNIA CA 3/1972 9/1959 10/1980
COLORADO CO 10/1983 9/1985
CONNECTICUT CT 10/1985 1/1980 6/1980
DELAWARE DE 3/1992 4/1992
FLORIDA FL
GEORGIA GA
HAWAII HI 8/1986 10/1982
IDAHO ID 4/1977 4/1977 8/1989
ILLINOIS IL 12/1974 12/1978
INDIANA IN 8/1987 5/1973
IOWA IA 11/1987 7/1985
KANSAS KS 8/1984 6/1981
KENTUCKY KY 8/1983 11/1983
LOUISIANA LA 1/1998
MAINE ME 11/1977
MARYLAND MD 1/1985 7/1981
MASSACHUSETTS MA 5/1988 5/1980 7/1977
MICHIGAN MI 6/1980 6/1976
MINNESOTA MN 4/1983 11/1986
MISSISSIPPI MS 6/1992 7/1987
MISSOURI MO 1/1983 – 2/1998 11/1985
MONTANA MT 6/1987 1/1980 1/1982
NEBRASKA NE 11/1983 11/1987
NEVADA NV 8/1983 1/1984 2/1987
NEW HAMPSHIRE NH 8/1988 2/1974 2/1974 – 5/1980
NEW JERSEY NJ 5/1985 7/1980
NEW MEXICO NM 2/1980 7/1983
NEW YORK NY 11/1982
NORTH CAROLINA NC 5/1985
NORTH DAKOTA ND 2/1984 11/1987
OHIO OH 4/1982 3/1990
OKLAHOMA OK 12/1976 2/1989 5/1985 – 2/1989
OREGON OR 3/1978 6/1975
PENNSYLVANIA PA 3/1974
RHODE ISLAND RI
SOUTH CAROLINA SC 6/1987 11/1985
SOUTH DAKOTA SD 4/1983 12/1988
TENNESSEE TN 11/1981 8/1984
TEXAS TX 4/1985 6/1984
UTAH UT 5/1986 3/1989
VERMONT VT 8/1985 9/1986
VIRGINIA VA 9/1983 6/1985
WASHINGTON WA 8/1977 7/1984
WEST VIRGINIA WV 4/1986 7/1978
WISCONSIN WI 6/1985 1/1980
WYOMING WY 8/1985 7/1989 1/1994
Source: Summarized from Autor, Donohue, and Schwab’s (2006) Legal Appendix
127

2.3 Theoretical Framework for Wrongful Discharge Laws
A typical way in which Wrongful Discharge Laws (WDLs) affect the labor
market is how these laws make it more difficult for employers to dismiss employees. The
imposition of these laws deprives firms of flexibility in adjusting their workforces. Thus,
these laws act as an additional cost, a firing cost, imposed on an employment
relationship.
In a frictionless world, workers enjoy the benefits of employment protection.
Thus, workers are willing to pay for protection by accepting lower wages. In an actual
world, wages received by workers cannot be downwardly adjusted, at least in the short
run. Thus, firms bear the costs of WDLs by accepting higher input prices, in this case
higher wages. The assumption of downward wage rigidity, held in the model illustrated
below, has been shown to be appropriate. Miles (2000) and Autor, Donohue, and Schwab
(2006) reported no significant impact of WDLs on wages. I will also verify this
assumption in Section 2.5 and 2.6.
The literature reported negative impacts (some reported zero impacts) of WDLs
on overall employment. These studies considered labor to be homogeneous. They failed
to recognize that labor is, in fact, heterogeneous. Thus, the imposition of WDLs not only
causes firms to scale back their production (due to higher input prices) but also causes
firms to substitute between different types of labor. The idea can be illustrated using a
simple model. Let the economy be consisted of identical firms. A firm’s production
function, ) , ( L H f , acquires two types of input, high-skilled labor (H) and low-skilled
labor (L). Although the model can be generalized to accommodate more types of labor
128

input, this simple model utilizing two types of labor input can provide an adequate
framework without loss of generality. Let the production function, ) , ( L H f , possess a
strict concavity. Also, let the production function be an increasing function of both types
of input. The markets for both output and input are competitive (i.e., firms accept output
prices and wages as given). Let the output price be normalized to one and let the market
wages for high-skilled labor be higher than the market wages for low-skilled labor.
Prior to the imposition of WDLs, the profit function of a firm can be written as
follows:
L w H w L H f
L H
⋅ − ⋅ − =
0 0
) , ( π .
0
H
w and
0
L
w are wages that employers pay to high-skilled and low-skilled labor,
respectively. Note that before WDLs are imposed, wages paid by employers and wages
received by employees are equal. The first order conditions for a firm’s maximization
problem are:
0
H H
w f = and
0
L L
w f = .
H
f and
L
f are the marginal products of labor for high-skilled and low-skilled workers,
respectively. The resulting equilibrium (E0) is shown in Figure 2.5. At E0, the
equilibrium prior to the imposition of WDLs, each firm produces q* unit of output and
employs H0 and L0 units of high-skilled and low-skilled labor. The iso-cost line
corresponding to this equilibrium is AA'.
129

Figure 2.5: Theoretical framework

Once WDLs are imposed, a firing cost F > 0 is added to the costs of employing
labor. As mentioned earlier, wages received by employees are assumed to be downwardly
rigid. Thus, both high-skilled and low-skilled workers need to be paid an amount equal to
what they used to be paid prior to the imposition of WDLs (
0
H
w and
0
L
w ). The cost F > 0
is incurred by the firm. Specifically, wages paid by the firm after the imposition of
WDLs,
1
H
w and
1
L
w , are:
F w w
H H
+ =
0 1

F w w
L L
+ =
0 1
.
130

The above expressions imply that the firing costs incurred by firms are equal for
high-skilled and low-skilled labor. This is because WDLs are costly to firms in the way in
which they prevent firms from flexibly adjusting their workforce regardless of worker
type.
10
Since
0 0
H L
w w < (by assumption), I can derive
F w
F w
w
w
H
L
H
L
+
+
<
0
0
0
0
. Thus, the iso-cost
line becomes steeper. This is shown in Figure 2.5. The iso-cost line shifts from AA' to
BB'. The new equilibrium (E1) reflects the outcome of the substitution effect due to
changes in labor costs. The amount of high-skilled labor input increases (from H0 to H1)
and the amount of low-skilled labor input decreases (from L0 to L1). However, with
higher costs for both types of labor input, q* is no longer the profit-maximizing output
quantity. Firms will thus cut back their production (from q* to q**). The iso-cost line
shifts inward (from BB' to CC') due to the scale effect. At the new equilibrium (E2), the
amount of low-skilled labor input is lower than that of the original equilibrium (L2 < L0).
The amount of high-skilled labor input for this new equilibrium can either be higher than,
lower than, or equal to that of the original equilibrium. This is an empirical question
which will be explored in Section 2.5 and 2.6.

10
It can also be argued that the firing costs for high-skilled and low-skilled labor are different. I denote the
firing cost for high-skilled labor by
H
F and the firing cost for low-skilled labor by
L
F . One possible
hypothesis is
H L
F F > . Workers with lower specific-investment (usually low-skilled) generally have
higher employment variation than workers with higher specific investment (usually high-skilled) (Rosen,
1968). Thus, it is possible that the restrictions on the flexibility to adjust the workforce are more binding
and thus more costly in the case of low-skilled labor employment. The other hypothesis is
H L
F F < . It is
possible that high-skilled workers are more likely to litigate against firms in the case of wrongful discharge.
Thus, under this hypothesis, the firing cost for high-skilled labor is larger since it is associated with the
higher probability of firms having to pay litigation costs and penalty payments. The assumption of
H L
F F F = = in the model can be relaxed with the following proviso. In the case of
H L
F F < , an
additional assumption needs to be made in order for the model to be valid. The assumption is that
H
F must
not be “too much larger” than
L
F . Specifically, I need to assume that
H
L
H
L
F
F
w
w
<
0
0
.
131

2.4 Data
The main data source for this study is the Current Population Survey (CPS). The
CPS is the monthly labor force survey conducted by the US Bureau of Labor Statistics.
The purpose of the survey is to measure labor force participation and employment and to
produce estimates of labor force characteristics for the US civilian non-institutional
population aged 16 and older. About 60,000 households (approximately 100,000 adults)
are interviewed each month. The CPS has a 4-8-4 rotation group structure. Households
are interviewed consecutively for four months and are left out of the sample for eight
months. The households are again interviewed for another four consecutive months and
are then left out of the sample permanently. The earning questions are asked to only one-
fourth of the workers in the survey each month. These are the workers in their fourth and
eighth months of the interviews (i.e., they are in the outgoing rotation groups).
The CPS is composed of the Basic Monthly Surveys and the Supplements. The
Basic Monthly Surveys ask questions about labor force status and basic demographic
information. In addition to the Basic Monthly Surveys, the CPS occasionally includes
supplemental questions on subjects of interest to federal and state agencies, private
foundations, and other organizations. Questions in the CPS supplements vary. Existing
supplements include topics such as job training, job tenure, contingent employment,
worker displacement, veteran status, school enrollment, immigration, fertility, voting,
smoking, computer usage, health, and employee benefits. Questions in the Basic Monthly
Surveys and in the Job Training Supplements are of interest and will be utilized in this
paper.
132

I use the CPS basic monthly files from 1983 to 1994 to construct the employment
and wage data series for the regression analyses. The employment data series are
intended for the main empirical analyses, while the wage data series are intended for
verifying the downward wage rigidity assumption. There are two reasons why I start the
data series in 1983 and not earlier. First, the 2-digit detailed occupation codes that I need
to use in this study changed over the period. More specifically, before 1983, the CPS
followed the 1970 census for the detailed occupation codes, but from 1983 until 2002 the
CPS followed those of the 1980 census. These codes cannot be directly converted without
introducing some inaccuracies due to the imputation.
11

Second, I use the CPS Job Training Supplement questions conducted in January
1983 to categorize the skill levels of different occupations.
12
In this paper, I define high-
skilled workers to be workers employed in high-skilled occupations. High-skilled
occupations are occupations that involve high training (i.e., a large proportion of workers
receive training). By starting the data series after January 1983, I have training variables
defined before the period in which I study the law changes, and hence these training
variables are not affected by these law changes. The CPS Job Training Supplements
gather detailed information about the training that workers received while at their jobs. I
use these training questions to calculate the proportion of workers who have received
training (training indices) in each occupation and to classify occupations into three

11
The 2-digit detailed occupation codes are grouping of the 3-digit codes. There is no one-to-one
relationship between the 1970 census occupation codes and the 1980 census occupation codes. The 1980
census 3-digit codes can be imputed from the 1970 ones and vice versa (see US Bureau of the Census
Technical Paper 59). However, imputation will inevitably introduce some inaccuracies. Thus, I decided not
to do it here.
12
The CPS Job Training Supplement questions were also asked in January 1984 and in January 1991.
133

groups, namely, low-, medium-, and high-skilled occupation groups. Although one may
argue that these training indices may change for some occupations once the WDLs are
introduced, an assumption I make here is that each occupation will not be re-categorized
into a different skill category, once the laws are introduced.
13

The specific training questions that I utilize in this paper are:
Q1. Since you obtained your present job, did you receive any training to improve your
skills? (YES, NO, N/A)
Q2. (If YES to the previous question): Did you receive training in:
a. A school? (YES, N/A)
b. A formal company training program? (YES, N/A)
c. Informal on-the-job training? (YES, N/A)
With the above information, I can calculate, for each occupation, the following
training indices:
1. Fraction of workers who received any kind of training (any-training index)
2. Fraction of workers who received school training (school-training index)
3. Fraction of workers who received formal company training (formal-training index)
4. Fraction of workers who received informal on-the-job training (informal-training
index)
It is worth noting that the universe of the Job Training Supplements contained
employed workers (both at work and not at work) and unemployed workers who had
worked in the past. Question Q1 was asked only to employed workers who were at work.

13
I use the January 1991 Job Training Supplement to verify this and find that the categorization of the
occupations into skill groups barely changed.
134

To calculate the fraction of workers who received any kind of training, I count the
number of workers who answered YES divided by the number of workers who responded
YES or NO
14
(i.e., I exclude the non-responses). For questions Q2-a, Q2-b, and Q2-c, I
can only identify whether the respondents answered YES to the questions. I cannot
distinguish between NO and non-response, so I treat both as NO. The fractions are the
number of workers who answered YES to the question divided by the number of workers
who responded to question Q1.
15
I believe that the training indices calculated are
appropriate approximates of the average amount of training generally acquired by
employees in each occupation.
The rankings of occupations by each of the above indices are illustrated in Table
2.2 (2.2A, 2.2B, 2.2C, and 2.2D). According to the rankings, occupations are classified
into high-, medium-, and low-skilled groups. In Table 2.2A, occupations are ranked using
the any-training index. Observe that there is a great deal of variation in the intensity of
training, ranging from more than 70% in the case of high school teachers and health
diagnostic technicians to 5% for private household service workers. In Tables 2.2B and
2.2C, occupations are ranked by the school-training and by formal-training indices
respectively. With only a few exceptions, the rankings by these three indices (any-
training, school-training, and formal-training) are quite similar. Examples of occupations

14
I use the January supplement weight (adjusted for supplement non-interviewed) in calculating the
fractions.
15
I do this so that the workers taken into account when calculating the school-training, formal-training, and
informal-training indices are consistent with the workers taken into account when calculating the any-
training index. Note that the workers who answer NO to question Q1 are not asked questions Q2-a, Q2-b,
and Q2-c, but they would have answered NO to these questions.
135

Table 2.2A: Any-training index
Occupation Code Occupation Fraction of workers received any kind of training Group
27 Private Household Service Occupations 0.0544938
38 Motor Vehicle Operators 0.1389347
31 Cleaning and Building Service Occupations 0.1481309
41 Freight, Stock and Material Handlers 0.1481502
45 Forestry and Fishing Occupations 0.1483199
44 Farm Workers and Related Occupations 0.1560583
29 Food Service Occupations 0.1584234
42 Other Handlers, Equipment Cleaners, and Laborers 0.1727826
40 Construction Laborers 0.1735819
43 Farm Operators and Managers 0.2040806
36 Machine Operators and Tenders, Except Precision 0.2259195
19 Sales Workers, Retail and Personal Services 0.2275931
39 Other Transportation Occupations and Material Moving 0.2478085
37 Fabricators, Assemblers, Inspectors, and Samplers 0.2557604
34 Construction Trades 0.2847712
Low

25 Mail and Message Distributing 0.3050183
23 Secretaries, Stenographers, and Typists 0.307571
24 Financial Records, Processing Occupations 0.311928
16 Supervisors and Proprietors, Sales Occupations 0.3471562
32 Personal Service Occupations 0.3598349
35 Other Precision Production Occupations 0.375325
26 Other Administrative Support Occupations, Including Clerical 0.3822806
18 Sales Representatives Commodities, Except Retail 0.4198851
30 Health Service Occupations 0.4331955
2 Other Executive, Administrators, and Managers 0.4601972
33 Mechanics and Repairers 0.4694868
22 Computer Equipment Operators 0.4787066
21 Supervisors-Administrative Support 0.5052693
14 Engineering and Science Technicians 0.5078194
9 Teachers, College and University 0.5151968
Medium

136

Table 2.2A, Continued
Occupation Code Occupation Fraction of workers received any kind of training Group
20 Sales Related Occupations 0.5169604
13 Health Technologists and Technicians 0.5297871
3 Management Related Occupations 0.5339963
12 Other Professional Specialty Occupations 0.542066
11 Lawyers and Judges 0.5813953
15 Technicians, Except Health Engineering, and Science 0.5824444
4 Engineers 0.5877689
17 Sales Representatives, Finance, and Business Service 0.6136144
6 Natural Scientists 0.6139696
28 Protective Service Occupations 0.6315429
5 Mathematical and Computer Scientists 0.6743543
8 Health Assessment and Treating Occupations 0.6800746
1 Administrators and Officials, Public Administration 0.716915
7 Health Diagnosing Occupations 0.7292728
10 Teachers, Except College and University 0.7653595
High

137

Table 2.2B: School-training index
Occupation Code Occupation Fraction of workers received school training Group
27 Private Household Service Occupations 0.0019364
25 Mail and Message Distributing 0.0022626
41 Freight, Stock and Material Handlers 0.0134333
38 Motor Vehicle Operators 0.0170355
42 Other Handlers, Equipment Cleaners, and Laborers 0.0218251
39 Other Transportation Occupations and Material Moving 0.0235982
36 Machine Operators and Tenders, Except Precision 0.0254028
31 Cleaning and Building Service Occupations 0.0278159
40 Construction Laborers 0.0296847
44 Farm Workers and Related Occupations 0.034731
19 Sales Workers, Retail and Personal Services 0.035734
29 Food Service Occupations 0.0369253
45 Forestry and Fishing Occupations 0.0385198
37 Fabricators, Assemblers, Inspectors, and Samplers 0.0493016
33 Mechanics and Repairers 0.0708368
Low

16 Supervisors and Proprietors, Sales Occupations 0.0773439
34 Construction Trades 0.0822866
43 Farm Operators and Managers 0.0851407
35 Other Precision Production Occupations 0.0859407
26 Other Administrative Support Occupations, Including Clerical 0.0878191
30 Health Service Occupations 0.0906034
18 Sales Representatives Commodities, Except Retail 0.0946986
32 Personal Service Occupations 0.106825
23 Secretaries, Stenographers, and Typists 0.1204789
24 Financial Records, Processing Occupations 0.1235875
22 Computer Equipment Operators 0.1390122
11 Lawyers and Judges 0.1495271
21 Supervisors-Administrative Support 0.1553749
13 Health Technologists and Technicians 0.1664891
2 Other Executive, Administrators, and Managers 0.1721545
Medium

138

Table 2.2B, Continued
Occupation Code Occupation Fraction of workers received school training Group
17 Sales Representatives, Finance, and Business Service 0.1794171
14 Engineering and Science Technicians 0.2005907
15 Technicians, Except Health Engineering, and Science 0.2021808
3 Management Related Occupations 0.2043212
5 Mathematical and Computer Scientists 0.2180333
8 Health Assessment and Treating Occupations 0.2365209
28 Protective Service Occupations 0.2375705
4 Engineers 0.2412252
12 Other Professional Specialty Occupations 0.2504251
20 Sales Related Occupations 0.2630452
1 Administrators and Officials, Public Administration 0.276863
6 Natural Scientists 0.3063128
7 Health Diagnosing Occupations 0.3380346
9 Teachers, College and University 0.3724549
10 Teachers, Except College and University 0.6409703
High

139

Table 2.2C: Formal-training index
Occupation Code Occupation Fraction of workers received formal training Group
43 Farm Operators and Managers 0.0175515
29 Food Service Occupations 0.0220365
27 Private Household Service Occupations 0.0226557
44 Farm Workers and Related Occupations 0.0227975
41 Freight, Stock and Material Handlers 0.0260338
31 Cleaning and Building Service Occupations 0.0296611
40 Construction Laborers 0.0303548
42 Other Handlers, Equipment Cleaners, and Laborers 0.0337012
36 Machine Operators and Tenders, Except Precision 0.0362584
9 Teachers, College and University 0.0415534
38 Motor Vehicle Operators 0.0458973
45 Forestry and Fishing Occupations 0.0475126
37 Fabricators, Assemblers, Inspectors, and Samplers 0.0548318
19 Sales Workers, Retail and Personal Services 0.0653655
24 Financial Records, Processing Occupations 0.070279
Low

39 Other Transportation Occupations and Material Moving 0.075374
34 Construction Trades 0.0762882
7 Health Diagnosing Occupations 0.0822099
23 Secretaries, Stenographers, and Typists 0.0826247
25 Mail and Message Distributing 0.0828437
10 Teachers, Except College and University 0.0961728
32 Personal Service Occupations 0.1006098
11 Lawyers and Judges 0.106687
30 Health Service Occupations 0.1215534
12 Other Professional Specialty Occupations 0.1334349
26 Other Administrative Support Occupations, Including Clerical 0.1346527
35 Other Precision Production Occupations 0.1388943
16 Supervisors and Proprietors, Sales Occupations 0.1399342
13 Health Technologists and Technicians 0.1562996
2 Other Executive, Administrators, and Managers 0.1639372
Medium

140

Table 2.2C, Continued
Occupation Code Occupation Fraction of workers received formal training Group
14 Engineering and Science Technicians 0.1878971
22 Computer Equipment Operators 0.1916446
3 Management Related Occupations 0.2095282
18 Sales Representatives Commodities, Except Retail 0.2211782
20 Sales Related Occupations 0.2230621
21 Supervisors-Administrative Support 0.2420617
33 Mechanics and Repairers 0.2428236
15 Technicians, Except Health Engineering, and Science 0.2572868
8 Health Assessment and Treating Occupations 0.2588073
6 Natural Scientists 0.2752914
4 Engineers 0.2920008
17 Sales Representatives, Finance, and Business Service 0.2950807
28 Protective Service Occupations 0.3279191
1 Administrators and Officials, Public Administration 0.3502091
5 Mathematical and Computer Scientists 0.3817046
High

141

Table 2.2D: Informal-training index
Occupation Code Occupation Fraction of workers received informal training Group
20 Sales Related Occupations 0
27 Private Household Service Occupations 0.0336552
43 Farm Operators and Managers 0.054813
45 Forestry and Fishing Occupations 0.0628462
38 Motor Vehicle Operators 0.0734377
32 Personal Service Occupations 0.0769121
9 Teachers, College and University 0.0786915
7 Health Diagnosing Occupations 0.0829758
31 Cleaning and Building Service Occupations 0.0922231
44 Farm Workers and Related Occupations 0.0937638
10 Teachers, Except College and University 0.0949497
29 Food Service Occupations 0.0999658
11 Lawyers and Judges 0.1074059
41 Freight, Stock and Material Handlers 0.1123306
23 Secretaries, Stenographers, and Typists 0.1165614
Low

42 Other Handlers, Equipment Cleaners, and Laborers 0.1187727
40 Construction Laborers 0.1232443
34 Construction Trades 0.1340891
24 Financial Records, Processing Occupations 0.1342847
19 Sales Workers, Retail and Personal Services 0.1357762
16 Supervisors and Proprietors, Sales Occupations 0.1369977
2 Other Executive, Administrators, and Managers 0.1494641
39 Other Transportation Occupations and Material Moving 0.1505784
6 Natural Scientists 0.1520941
37 Fabricators, Assemblers, Inspectors, and Samplers 0.1591837
36 Machine Operators and Tenders, Except Precision 0.1677804
21 Supervisors-Administrative Support 0.1708288
12 Other Professional Specialty Occupations 0.1747022
3 Management Related Occupations 0.1767491
18 Sales Representatives Commodities, Except Retail 0.1795304
Medium

142

Table 2.2D, Continued
Occupation Code Occupation Fraction of workers received informal training Group
35 Other Precision Production Occupations 0.1818477
33 Mechanics and Repairers 0.1840206
4 Engineers 0.1893195
14 Engineering and Science Technicians 0.1927129
8 Health Assessment and Treating Occupations 0.1975016
26 Other Administrative Support Occupations, Including Clerical 0.1978101
13 Health Technologists and Technicians 0.1988621
17 Sales Representatives, Finance, and Business Service 0.2034024
25 Mail and Message Distributing 0.2188927
30 Health Service Occupations 0.2285811
15 Technicians, Except Health Engineering, and Science 0.2380981
28 Protective Service Occupations 0.2415931
5 Mathematical and Computer Scientists 0.2467641
1 Administrators and Officials, Public Administration 0.2600389
22 Computer Equipment Operators 0.2660764
High

143

classified as low-skilled are private household service workers, motor vehicle operators,
equipment cleaners, and construction laborers. On the other hand, examples of high-
skilled occupations are engineers, mathematicians, computer scientists, and natural
scientists. Examples of medium-skilled occupations are secretaries, health service
workers, and personal service workers.
In Table 2.2D, occupations are ranked using the informal-training index. The
rankings of occupations under this criterion are very different from that of the first three
criteria. The informal training questions differ significantly from other survey questions
since there is no clear interpretation of what constitutes “informal training.” Most
respondents can accurately identify if they have received school training or formal
company training. However, respondents may not be able to consistently identify whether
they have received “informal training” or not. For example, some respondents may
consider learning-by-doing to be “informal training,” while others may not share the
same belief.
In preparing the monthly employment data, I calculate each occupation group’s
employment in each state divided by the state population. The occupation questions were
asked to people in the labor force and to people who were not in the labor force but who
had worked prior to the time of the interview. Thus, a number of observations were
missing information regarding occupations due to the fact that these people were not
asked about their occupations. These people are still, however, considered to belong to
the population of the states, so I include them in the denominator along with the
unemployed and with people not in the labor force when I calculate each occupation
group’s employment per state population. The monthly employment data series run from
144

February 1983 until December 1994, with the January 1983 training information kept
exogenous from the regression. In preparing data for the wage regressions, I calculate the
average real wage for each occupation group by state. The monthly wage data series also
begin in February 1983 and end in December 1994.
2.5 Empirical Methodology
As Krueger (1991) and Dertouzos and Karoly (1992) have pointed out, the
adoption of WDLs may not be completely exogenous. Specific situations that occurred in
each state, as well as each state’s characteristics, may have driven the adoption of the
laws. However, Autor, Donohue and Schwab (2004) observe that unless one can find
some valid instruments to address the problem, the instrumental variable estimation will
bias the estimates. In this paper, I address the possibility of endogenous adoption
decisions by including state-fixed effects and state-specific time trends, along with time-
fixed effects, in the regressions. These variables capture the unobserved state
characteristics that change over time and may be correlated with the state’s decision to
adopt WDLs. In my analysis, I use the case law classification of WDLs (identification of
the adoption dates for WDLs) developed by Autor, Donohue and Schwab (2006).
16

To study the effects of WDLs on employment for each occupation group, I
employ the following model:
jst s s t
j j j j j st
j st j st jst jst
t
t High t Low High Low High Adopt
Med Adopt Low Adopt x y
ε π π δ
β
β β η α
+ × + + +
× + × + + + × ⋅ +
× ⋅ + × ⋅ + ′ + =
3
2 1
) ln(

16
As mentioned earlier, this information is summarized in Table 2.1.
145

where
jst
y is occupation group j’s employment per state s population at time t (month-
year).
jst
x is the vector of observable characteristics for each occupation group. These
characteristics include the percentage of male workers, the percentage of black workers,
the percentage of workers in each age group (18-35, and 36-55), the percentage of
married workers, the percentage of unionized workers, and the percentage of workers in
each education group (high school, some college, and college education or higher).
st
Adopt is the dummy indicating whether the state is currently adopting WDLs. This
dummy is set to 1 starting with the month right after the initial adoption.
j
Low ,
j
Med ,
and
j
High are the dummies denoting whether the observation belongs to the low-,
medium-, or high-skilled occupations, respectively.
In the model, the effect of WDLs on employment for low-skilled labor is shown
by the coefficient of the
j st
Low Adopt × variable (
1
β ), whereas the effects of the laws on
employment of the medium-skilled and the high-skilled occupations are denoted by
2
β
and
3
β , respectively. One can argue that the high-skilled sector may have been
expanding and that the low-skilled sector may have been contracting during the period of
the study. If that is the case, then the sector-specific time trends ( t Low
j
× and t High
j
× )
should capture such phenomena. One should expect to see negative and significant
coefficients for low-skilled employment trends and positive and significant coefficients
for high-skilled employment trends.
t
δ and
s
π are time-fixed effects and state-fixed effects, while t
s
× π reflects
state-specific time trends. I conduct the employment regression analysis for each type of
146

WDL per each training index. Two versions of the model are implemented: one
excluding occupational characteristic controls
jst
x (base model) and one including
occupational characteristic controls (extended model).
To verify the assumption of downward wage rigidity, the following wage
regression model is executed:
jst s s t
j j j j j st
j st j st jst jst
t
t High t Low High Low High Adopt
Med Adopt Low Adopt x w
ε π π δ
β
β β η α
+ × + + +
× + × + + + × ⋅ +
× ⋅ + × ⋅ + ′ + =
3
2 1
) ln(

where
jst
w is the average real wage for occupation group j in state s at time t (month-
year). Other variables included in this model are analogous to the ones included in the
employment model. Similarly, the effects of WDLs on the average wage for the low-
skilled group are captured by
1
β . The effects of WDLs on the average wage for medium-
skilled and high-skilled labor are captured by
2
β and
3
β respectively. If WDLs, in fact,
do not have an impact on the wages that workers receive, then one should expect these
coefficients to be statistically insignificant. Thus, the assumption of downward wage
rigidity following the imposition of WDLs would be appropriate. As performed with the
employment regression, the wage regression is performed for each type of WDL per each
index of training, with and without the occupational characteristic controls
jst
x .
As Bertrand, Duflo and Mullainathan (2004) have pointed out, one cannot reject
the possibility that the error terms for the data series (both employment and wage) are
serially correlated. The serial correlations will bias the standard errors calculated by the
usual method towards zero. Therefore, I cluster the standard errors by state (Huber-White
147

robust standard errors) to allow possible correlation of these error terms over time and
within each state. I assume that the error terms in different states are independent of one
another.
2.6 Results
2.6.1 Good Faith
Using each of the four training indices to categorize occupations into skill groups,
Table 2.3A reports the effect of the Good Faith exception on employment. For each
training index, two versions of the model are executed: one excluding occupational
characteristic controls (base model) and one including occupational characteristic
controls (extended model).
The first two columns are the results when the any-training index is used. The
effect of the Good Faith law on employment for low-skilled labor is illustrated by
Low Adopt × . The estimates for these effects are negative and significant. The magnitude
of the effect is 10.6% under the base model (column 1), while the magnitude reduces to
7.5% under the extended model (column 2). The effect of the Good Faith law on
employment for medium-skilled labor ( Med Adopt × ) is positive and significant at a
magnitude of 4.8% under the base model and 3.1% under the extended model. For high-
skilled labor, the Good Faith law also has an overall positive and significant effect on
employment. The magnitude is 12.7% under the base model and 11.4% under the
extended model.
148

Table 2.3A: Good Faith and employment
Any Training School Training Formal Training Informal Training
(1) (2) (3) (4) (5) (6) (7) (8)
Adopt(GF)xLow -0.10554*** -0.07495** -0.09688*** -0.07924*** -0.10168*** -0.07068* -0.04824 -0.04859
(0.03689) (0.03612) (0.02803) (0.02673) (0.03604) (0.03631) (0.04846) (0.04795)
Adopt(GF)xMed 0.04769** 0.03084* 0.02860** 0.01998 0.04180** 0.03442** 0.00851 0.01909
(0.01812) (0.01581) (0.01113) (0.01258) (0.01594) (0.01555) (0.02085) (0.02087)
Adopt(GF)xHigh 0.12701*** 0.11445** 0.13059*** 0.12236*** 0.10450*** 0.06530* 0.04713 0.04485
(0.04437) (0.04415) (0.04218) (0.03915) (0.03428) (0.03299) (0.02982) (0.02997)
Low -0.03842 0.08549 -0.11132*** -0.11220** -0.19032*** 0.11703* -0.36121*** -0.32638***
(0.02912) (0.07331) (0.02550) (0.04560) (0.02971) (0.06843) (0.02916) (0.06327)
High -0.69975*** -1.24093*** -0.73359*** -0.98219*** -0.81565*** -0.91238*** -0.41121*** -0.29929***
(0.01846) (0.10815) (0.02138) (0.07681) (0.01281) (0.05239) (0.01655) (0.03324)
LowXt -0.00098*** -0.00082*** -0.00091*** -0.00086*** -0.00129*** -0.00125*** -0.00057*** -0.00081***
(0.00013) (0.00021) (0.00011) (0.00017) (0.00012) (0.00021) (0.00012) (0.00017)
HighXt 0.00086*** 0.00099*** 0.00079*** 0.00054*** -0.00013** -0.00036*** 0.00040*** 0.00055***
(0.00009) (0.00015) (0.00009) (0.00014) (0.00006) (0.00010) (0.00009) (0.00010)
%male 0.02499 -0.04905 -0.02577 0.69998**
(0.15258) (0.19437) (0.21355) (0.26910)
%black 0.31604 0.30468 0.06945 -0.29700
(0.24743) (0.25284) (0.25575) (0.24551)
%age18-35 -0.55691* 0.13680 -0.48610 -0.58359*
(0.28046) (0.14362) (0.29911) (0.29105)
%age36-55 -0.50272* 0.09852 -0.45837 -0.86699***
(0.29184) (0.11583) (0.27493) (0.28767)
%married -0.22578 -0.13077 -0.13742 0.01101
(0.22192) (0.19306) (0.15272) (0.11209)
149

Table 2.3A, Continued
Any Training School Training Formal Training Informal Training
(1) (2) (3) (4) (5) (6) (7) (8)
%union -0.66118 0.27130 -0.56566 -0.13489
(0.52392) (0.49952) (0.44965) (0.31116)
%high school education 0.25193 0.00433 0.50149* -0.53577***
(0.23998) (0.17102) (0.25161) (0.16443)
%some college education -0.19957 -0.63973*** 0.20401 -0.89671***
(0.32763) (0.22989) (0.30101) (0.28438)
%college education & higher 1.60344*** 0.64245** 1.90197*** -0.64401**
(0.36676) (0.25438) (0.36112) (0.28790)
Constant -1.37015*** -1.11489*** -1.33846*** -1.28518*** -1.26966*** -1.49169*** -1.34302*** -0.59853**
(0.00941) (0.38346) (0.00809) (0.15505) (0.01052) (0.31060) (0.00837) (0.26305)
Observations 21450 21450 21450 21450 21450 21450 21450 21450
R-squared 0.84 0.87 0.88 0.88 0.90 0.92 0.83 0.85
F test 3.16675 2.48986 4.78851 3.77166 3.11686 1.96399 1.11318 0.89963
Prob > F 0.03251 0.07121 0.00528 0.01632 0.03443 0.13166 0.35280 0.44816

Robust standard errors in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
150

Columns 3 and 4 display the regression results when the school-training index is
used to categorize occupations into skill groups. The overall effects of the law on the
employment of low-skilled labor are negative at a magnitude of 9.7% in the model
without controls (base model) and 8.0% in the model with controls (extended model). On
the other hand, the Good Faith law is associated with increases in employment for
medium-skilled and high-skilled labor. For medium-skilled labor, the magnitude is 2.9%
under the base model. The effect of the law is no longer significant under the extended
model. For high-skilled labor, the magnitude of the law’s effect is 13.1% and 12.2%
under the base and extended models, respectively.
The results under the formal-training index are illustrated in columns 5 and 6.
Here, I again observe the positive effects of the Good Faith law on the employment of
high-skilled and medium-skilled labor, and I observe the negative effects of the Good
Faith law on the employment of low-skilled labor. The Good Faith law reduces low-
skilled labor employment by about 10.2% under the base model and 7.1% under the
extended model. On the other hand, Good Faith increases medium-skilled labor
employment by about 4.2% under the base model and 3.4% under the extended model. It
increases high-skilled labor employment by about 10.5% under the base model and 6.5%
under the extended model.
Using the informal-training index (columns 7 and 8), I find no significant effects
of the Good Faith law on employment of any type of labor. This is probably because the
respondents’ answers to the informal training questions are spurious. As pointed out
earlier, compared to other questions regarding training, the informal training questions
are vague in the sense that “informal training” is never clearly defined. Therefore, the
151

classification of occupations into groups under this criterion may not be a reliable method
to quantify skill levels.
For most specifications, I observe negative and significant coefficients for low-
skilled labor time trends ( t Low × ) and positive and significant coefficients for high-
skilled labor time trends ( t High × ). This supports the argument that the low-skilled
sector may have been contracting and that the high-skilled sector may have been
expanding over the data period.
Looking at the joint test of significance for all of the adoption variables
(0 = × = × = × High Adopt Med Adopt Low Adopt ), one can reject the null hypothesis that
these variables are not significant at 13.2% for the models using the first three indices of
training. However, under the informal-training index, the F statistics are quite small and
one cannot reject the null.
Figure 2.6 provides a visual illustration of employment per population for low-
skilled and high-skilled labor in states that adopted the Good Faith law during the data
period. The illustration utilizes the any-training index in categorizing occupations into
high-skilled and low-skilled sectors. Across all seven states (Alaska, Arizona, Delaware,
Idaho, Nevada, Oklahoma, and Wyoming), I observe that low-skilled labor constitutes a
larger sector compared to high-skilled labor. In some states, I observe small negative
trends for employment per population of low-skilled labor and small positive trends for
that of high-skilled labor. As mentioned earlier, these trends are captured by the variables
t Low × and t High × in the model. From the results, I still observe very large and very
significant effects of the Good Faith law, even though these trends are already accurately
accounted for.
152

Low-skilled employment High-skilled employment
0 .05 .1 .15 .2 .25 .3 .35
Emp/Pop
1980 1982 1984 1986 1988 1990 1992 1994 1996
Year
GF: 5/1983
Alaska (AK)

0 .05 .1 .15 .2 .25 .3 .35
Emp/Pop
1980 1982 1984 1986 1988 1990 1992 1994 1996
Year
GF: 5/1983
Alaska (AK)

0 .05 .1 .15 .2 .25 .3 .35
Emp/Pop
1980 1982 1984 1986 1988 1990 1992 1994 1996
Year
GF: 6/1985
Arizona (AZ)

0 .05 .1 .15 .2 .25 .3 .35
Emp/Pop
1980 1982 1984 1986 1988 1990 1992 1994 1996
Year
GF: 6/1985
Arizona (AZ)

Figure 2.6: Employment per population for high-skilled and low-skilled labor (categorized using any-training index)
in states adopting the Good Faith exception
153

Low-skilled employment High-skilled employment
0 .05 .1 .15 .2 .25 .3 .35
Emp/Pop
1980 1982 1984 1986 1988 1990 1992 1994 1996
Year
GF: 4/1992
Delaware (DE)

0 .05 .1 .15 .2 .25 .3 .35
Emp/Pop
1980 1982 1984 1986 1988 1990 1992 1994 1996
Year
GF: 4/1992
Delaware (DE)

0 .05 .1 .15 .2 .25 .3 .35
Emp/Pop
1980 1982 1984 1986 1988 1990 1992 1994 1996
Year
GF: 8/1989
Idaho (ID)

0 .05 .1 .15 .2 .25 .3 .35
Emp/Pop
1980 1982 1984 1986 1988 1990 1992 1994 1996
Year
GF: 8/1989
Idaho (ID)

Figure 2.6, Continued
154

Low-skilled employment High-skilled employment
0 .05 .1 .15 .2 .25 .3 .35
Emp/Pop
1980 1982 1984 1986 1988 1990 1992 1994 1996
Year
GF: 2/1987
Nevada (NV)

0 .05 .1 .15 .2 .25 .3 .35
Emp/Pop
1980 1982 1984 1986 1988 1990 1992 1994 1996
Year
GF: 2/1987
Nevada (NV)

0 .05 .1 .15 .2 .25 .3 .35
Emp/Pop
1980 1982 1984 1986 1988 1990 1992 1994 1996
Year
GF: 9/1985 - 2/1989
Oklahoma (OK)

0 .05 .1 .15 .2 .25 .3 .35
Emp/Pop
1980 1982 1984 1986 1988 1990 1992 1994 1996
Year
GF: 9/1985 - 2/1989
Oklahoma (OK)

Figure 2.6, Continued
155

Low-skilled employment High-skilled employment
0 .05 .1 .15 .2 .25 .3 .35
Emp/Pop
1980 1982 1984 1986 1988 1990 1992 1994 1996
Year
GF: 1/1994
Wyoming (WY)

0 .05 .1 .15 .2 .25 .3 .35
Emp/Pop
1980 1982 1984 1986 1988 1990 1992 1994 1996
Year
GF: 1/1994
Wyoming (WY)

Figure 2.6, Continued
156

Table 2.3B reports the effects of the Good Faith law on wages. Across all indices
of training, I do not observe any significant effects of the law on wages. For most
specifications, the joint F-tests of the adoption variables illustrate that one should accept
the null hypothesis that these adoption variables are insignificant. These results support
the notion that the Good Faith law has no impact on wages. Thus, the downward wage
rigidity assumption is verified.
2.6.2 Implied Contract
Table 2.4A reports the effect of the Implied Contract exception on employment
for each of the four training indices. As in the case of Good Faith, two versions of the
model are executed: one excluding occupational characteristic controls (base model) and
one including occupational characteristic controls (extended model).
Columns 1 and 2 show the results when the any-training index is used. The effect
of the Implied Contract law on employment for low-skilled labor ( Low Adopt × ) is
negative and significant at a magnitude of 5.9% under the base model. The effect is no
longer significant once characteristic controls are included. The coefficients reflecting the
impact of the law on medium-skilled labor ( Med Adopt × ) are negative but not
significant. The coefficients reflecting the impact of the law on high-skilled labor
( High Adopt × ) are positive but not significant.
Columns 3 and 4 display the regression results when the school-training index is
used to categorize occupations into skill groups. The results under this criterion show a
157

Table 2.3B: Good Faith and wages
Any Training School Training Formal Training Informal Training
(1) (2) (3) (4) (5) (6) (7) (8)
Adopt(GF)xLow -0.00226 0.00189 -0.01083 -0.00587 -0.01103 -0.00221 0.00900 0.01011
(0.01815) (0.01564) (0.01741) (0.01488) (0.01549) (0.01384) (0.01105) (0.01253)
Adopt(GF)xMed 0.00968 0.00937 0.01970** 0.01733 0.01866* 0.01660 0.01041 0.01060
(0.00962) (0.01140) (0.00937) (0.01103) (0.00999) (0.01212) (0.00907) (0.01135)
Adopt(GF)xHigh 0.02107 0.01555 0.01670 0.01657 0.01363 0.00479 0.00480 0.00028
(0.02008) (0.01865) (0.01829) (0.01766) (0.01170) (0.01185) (0.01101) (0.01171)
Low -0.29351*** -0.18735*** -0.27561*** -0.16539*** -0.36524*** -0.19511*** -0.30303*** -0.20981***
(0.00863) (0.01022) (0.00841) (0.01019) (0.00802) (0.00837) (0.00824) (0.00979)
High 0.19694*** -0.00314 0.21654*** 0.00081 0.18847*** 0.09605*** 0.03057*** 0.02129***
(0.00700) (0.01096) (0.00586) (0.01204) (0.00466) (0.00726) (0.00714) (0.00432)
LowXt -0.00037*** -0.00040*** -0.00034*** -0.00039*** -0.00025*** -0.00024*** 0.00054*** 0.00044***
(0.00006) (0.00006) (0.00005) (0.00005) (0.00006) (0.00005) (0.00004) (0.00004)
HighXt 0.00046*** 0.00061*** 0.00039*** 0.00053*** 0.00016*** 0.00015*** 0.00005 0.00006*
(0.00004) (0.00003) (0.00004) (0.00004) (0.00004) (0.00004) (0.00005) (0.00004)
%male 0.32344*** 0.29926*** 0.21983*** 0.25069***
(0.01783) (0.02444) (0.01979) (0.02010)
%black -0.11003*** -0.08898** -0.06340** -0.10110**
(0.02888) (0.03486) (0.03085) (0.04568)
%age18-35 -0.15793*** -0.17000*** -0.16947*** -0.11930***
(0.03057) (0.03090) (0.02727) (0.02173)
%age36-55 0.04615* 0.04658 0.05504** 0.09975***
(0.02700) (0.03012) (0.02248) (0.02638)
%married 0.12987*** 0.10390*** 0.13578*** 0.17806***
(0.02088) (0.02401) (0.01794) (0.01293)
158

Table 2.3B, Continued
Any Training School Training Formal Training Informal Training
(1) (2) (3) (4) (5) (6) (7) (8)
%union 0.28657*** 0.22547*** 0.29199*** 0.27403***
(0.04219) (0.04770) (0.04008) (0.03364)
%high school education 0.33703*** 0.35287*** 0.26953*** 0.21695***
(0.05112) (0.04466) (0.03357) (0.02234)
%some college education 0.39264*** 0.40641*** 0.34450*** 0.37805***
(0.05084) (0.04675) (0.03672) (0.03260)
%college education & higher 0.68443*** 0.71098*** 0.69451*** 0.80976***
(0.05056) (0.04394) (0.03870) (0.02955)
Constant 2.02341*** 1.41997*** 2.01585*** 1.44179*** 2.04735*** 1.48917*** 2.01185*** 1.39934***
(0.00365) (0.03441) (0.00352) (0.03643) (0.00333) (0.03207) (0.00309) (0.03732)
Observations 21450 21450 21450 21450 21450 21450 21450 21450
R-squared 0.92 0.93 0.92 0.93 0.93 0.94 0.84 0.87
F test 0.45617 0.31209 2.79676 1.91653 1.43035 1.78385 0.48602 1.06149
Prob > F 0.71414 0.81654 0.04984 0.13918 0.24523 0.16253 0.69356 0.37403

Robust standard errors in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
159

Table 2.4A: Implied Contract and employment
Any Training School Training Formal Training Informal Training
(1) (2) (3) (4) (5) (6) (7) (8)
Adopt(IC)xLow -0.05854* -0.03909 -0.06569** -0.04969* -0.03957 -0.02179 0.01662 0.01136
(0.03261) (0.02544) (0.03066) (0.02596) (0.03350) (0.02575) (0.02672) (0.02140)
Adopt(IC)xMed -0.00295 -0.01155 0.00236 -0.00714 -0.00713 -0.01206 -0.04058** -0.03035*
(0.01395) (0.01084) (0.01462) (0.01299) (0.01535) (0.01507) (0.01735) (0.01606)
Adopt(IC)xHigh 0.05005 0.02480 0.04904 0.02814 0.02200 -0.00261 -0.00485 -0.00128
(0.03437) (0.03129) (0.03140) (0.02841) (0.02717) (0.01901) (0.01853) (0.01680)
Low -0.03211 0.10258* -0.09419*** -0.09551** -0.19454*** 0.14090** -0.39951*** -0.36558***
(0.02866) (0.05746) (0.02969) (0.04585) (0.03170) (0.06488) (0.02836) (0.06462)
High -0.71351*** -1.25428*** -0.73976*** -0.97369*** -0.81982*** -0.92097*** -0.42324*** -0.31365***
(0.01929) (0.10956) (0.01885) (0.07885) (0.01261) (0.04963) (0.02258) (0.03910)
LowXt -0.00089*** -0.00084*** -0.00077*** -0.00088*** -0.00126*** -0.00128*** -0.00075*** -0.00091***
(0.00020) (0.00024) (0.00017) (0.00021) (0.00019) (0.00026) (0.00019) (0.00022)
HighXt 0.00074*** 0.00091*** 0.00069*** 0.00047*** -0.00019** -0.00037*** 0.00031*** 0.00047***
(0.00010) (0.00018) (0.00010) (0.00016) (0.00008) (0.00010) (0.00011) (0.00012)
%male 0.07756 -0.00195 0.00194 0.69256**
(0.17246) (0.19933) (0.20391) (0.27021)
%black 0.50558** 0.45147* 0.25186 -0.11713
(0.22294) (0.25149) (0.21592) (0.21842)
%age18-35 -0.75000** 0.00202 -0.63548** -0.70980**
(0.28027) (0.18326) (0.28351) (0.34182)
%age36-55 -0.58514** 0.06080 -0.52633** -0.97427***
(0.25804) (0.11819) (0.22991) (0.31092)
%married -0.23657 -0.19262 -0.18605 -0.01105
(0.19119) (0.16661) (0.12982) (0.11124)
160

Table 2.4A, Continued
Any Training School Training Formal Training Informal Training
(1) (2) (3) (4) (5) (6) (7) (8)
%union -0.47778 0.48761 -0.46207 -0.23430
(0.52171) (0.50169) (0.46501) (0.28919)
%high school education 0.46294** 0.25170 0.69749*** -0.47278**
(0.20913) (0.20680) (0.18860) (0.19578)
%some college education -0.09640 -0.53742** 0.36651 -0.80562**
(0.27957) (0.25159) (0.23029) (0.30135)
%college education & higher 1.78126*** 0.78965*** 2.14787*** -0.46920
(0.32537) (0.25741) (0.29435) (0.32734)
Constant -1.35466*** -1.16261*** -1.33021*** -1.35362*** -1.25271*** -1.54813*** -1.30876*** -0.53079**
(0.01253) (0.36909) (0.01336) (0.14725) (0.01365) (0.30342) (0.01580) (0.24200)
Observations 21450 21450 21450 21450 21450 21450 21450 21450
R-squared 0.83 0.86 0.87 0.88 0.90 0.92 0.82 0.84
F test 2.96051 2.06605 2.23396 2.10266 3.45580 1.26337 2.14307 1.57212
Prob > F 0.04123 0.11684 0.09601 0.11194 0.02335 0.29726 0.10677 0.20803

Robust standard errors in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
161

negative impact of the law on employment for low-skilled labor at a magnitude of 6.6%
under the base model and 5.0% under the extended model. The coefficients reflecting the
impact of the law on high-skilled labor are positive but not significant.
Columns 5 and 6 display the regression results using the formal-training index.
The results show no significant effects of the law on employment for any type of labor.
Columns 7 and 8 display the regression results when the informal-training index
is used to categorize occupations into skill groups. The results show negative impacts of
the law on employment for medium-skilled labor.
For most of the specifications, I observe negative and significant coefficients for
low-skilled labor time trends ( t Low × ) and positive and significant coefficients for high-
skilled labor time trends ( t High × ). Similar to the case of the Good Faith law, these
findings support the argument regarding the contraction of the low-skilled sector and the
expansion of the high-skilled sector during the data period.
Looking at the joint test for significance of all of the adoption variables
(0 = × = × = × High Adopt Med Adopt Low Adopt ), one can reject the null hypothesis that
these variables are not significant at 11.7% for most of the specifications.
Table 2.4B reports the effects of the Implied Contract law on wages. There are no
significant effects of the law on wages for high-skilled and low-skilled labor, although
there is limited evidence of positive wage effects for medium-skilled labor. Thus, the
downward wage rigidity assumption can at least be partially verified in this case.
162

Table 2.4B: Implied Contract and wages
Any Training School Training Formal Training Informal Training
(1) (2) (3) (4) (5) (6) (7) (8)
Adopt(IC)xLow 0.01512 0.00026 0.00965 -0.00114 0.00478 -0.00253 -0.00591 -0.00575
(0.00984) (0.00696) (0.00905) (0.00780) (0.00832) (0.00648) (0.00856) (0.00595)
Adopt(IC)xMed -0.00064 0.00864* 0.00855 0.01302*** 0.01032* 0.01293** 0.02488*** 0.01846***
(0.00697) (0.00482) (0.00589) (0.00470) (0.00529) (0.00535) (0.00676) (0.00470)
Adopt(IC)xHigh 0.00884 0.00583 0.00529 0.00546 0.00466 0.00299 -0.00061 0.00054
(0.01021) (0.00822) (0.00900) (0.00799) (0.00803) (0.00636) (0.00943) (0.00644)
Low -0.30343*** -0.18393*** -0.28083*** -0.16011*** -0.36687*** -0.18819*** -0.28723*** -0.19772***
(0.00983) (0.01077) (0.00796) (0.01102) (0.00746) (0.00917) (0.00865) (0.00873)
High 0.19396*** -0.00126 0.21789*** 0.00122 0.19071*** 0.09673*** 0.04296*** 0.02968***
(0.00608) (0.01207) (0.00495) (0.01288) (0.00342) (0.00707) (0.00725) (0.00420)
LowXt -0.00041*** -0.00038*** -0.00035*** -0.00036*** -0.00025*** -0.00021*** 0.00062*** 0.00051***
(0.00005) (0.00006) (0.00004) (0.00005) (0.00005) (0.00006) (0.00006) (0.00005)
HighXt 0.00044*** 0.00062*** 0.00040*** 0.00056*** 0.00018*** 0.00017*** 0.00011*** 0.00011***
(0.00005) (0.00004) (0.00004) (0.00004) (0.00004) (0.00004) (0.00004) (0.00004)
%male 0.32792*** 0.30158*** 0.23004*** 0.24827***
(0.01697) (0.02109) (0.01780) (0.02050)
%black -0.11062*** -0.08300** -0.06581** -0.11911***
(0.03234) (0.03863) (0.03072) (0.03652)
%age18-35 -0.16236*** -0.17401*** -0.17128*** -0.11423***
(0.03816) (0.04392) (0.03202) (0.02036)
%age36-55 0.04421 0.04647 0.05362** 0.10371***
(0.02881) (0.03409) (0.02505) (0.02580)
%married 0.12845*** 0.09817*** 0.13334*** 0.17884***
(0.02185) (0.02654) (0.02021) (0.01165)
163

Table 2.4B, Continued
Any Training School Training Formal Training Informal Training
(1) (2) (3) (4) (5) (6) (7) (8)
%union 0.29281*** 0.23759*** 0.29400*** 0.27558***
(0.04364) (0.05007) (0.04210) (0.02829)
%high school education 0.34336*** 0.36603*** 0.27876*** 0.20675***
(0.05483) (0.05564) (0.03960) (0.02254)
%some college education 0.39568*** 0.41834*** 0.35511*** 0.36491***
(0.05087) (0.05472) (0.04420) (0.03162)
%college education & higher 0.68978*** 0.72761*** 0.70434*** 0.79309***
(0.05058) (0.05064) (0.04476) (0.03067)
Constant 2.02198*** 1.40901*** 2.00966*** 1.42414*** 2.04010*** 1.46935*** 1.99188*** 1.39202***
(0.00648) (0.03399) (0.00489) (0.03344) (0.00458) (0.02981) (0.00606) (0.03634)
Observations 21450 21450 21450 21450 21450 21450 21450 21450
R-squared 0.92 0.93 0.92 0.93 0.93 0.94 0.84 0.87
F test 2.35212 1.46140 1.62206 3.52819 2.15874 2.43973 4.77427 9.18536
Prob > F 0.08362 0.23657 0.19629 0.02151 0.10483 0.07550 0.00536 0.00006

Robust standard errors in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
164

2.6.3 Public Policy
The results from the Good Faith and Implied Contract exceptions have suggested,
at some level of precision, that the employment of high-skilled labor is positively affected
by these laws and that the employment of low-skilled labor is negatively affected by
these laws. I have already argued that the sector trend variables help to capture episodes
of contraction for the low-skilled sector and of expansion for the high-skilled sector
during the data period. Thus, the coefficients of the adoption variables should be
appropriate measures of the effects of these laws, rather than merely illustrating the
employment trends of high-skilled and low-skilled labor.
As discussed in Section 2.2.1.3, the Public Policy exception does not constitute a
new legal regime, but rather it restricts the actions of employers in order to protect
existing public policy. As shown in Figures 2.3 and 2.4, many states adopted this type of
exception during the data period. The group of states that adopted the Public Policy law is
similar to the group that adopted the Implied Contract law. Table 2.5A shows that the
Public Policy law has virtually no effect on the employment of any type of labor, which
is true across all training indices. I obtain similar results for wages, as shown in Table
2.5B.
The employment results from the Public Policy exception act as a control for the
employment results from the Good Faith and Implied Contract exceptions. Basically, if
the results from the Good Faith and Implied Contract exceptions are merely the
illustration of employment trends for low-skilled and high-skilled labor, then I should
165

Table 2.5A: Public Policy and employment
Any Training School Training Formal Training Informal Training
(1) (2) (3) (4) (5) (6) (7) (8)
Adopt(PP)xLow -0.00959 0.00236 -0.00095 0.00904 -0.00047 0.00577 -0.03270 -0.03531
(0.04547) (0.03436) (0.03774) (0.03114) (0.04232) (0.03179) (0.02949) (0.02685)
Adopt(PP)xMed -0.00631 -0.01586 -0.01922 -0.02509* -0.02746 -0.03371* 0.00186 0.00829
(0.02028) (0.01444) (0.01373) (0.01298) (0.02296) (0.01740) (0.01239) (0.01263)
Adopt(PP)xHigh -0.02086 -0.04489 -0.01195 -0.02697 0.01097 -0.01851 -0.00118 -0.00425
(0.06022) (0.05013) (0.05684) (0.05031) (0.03777) (0.02810) (0.02878) (0.02582)
Low -0.05914 0.07601 -0.13855*** -0.12720** -0.22457*** 0.12202* -0.35291*** -0.32709***
(0.04366) (0.07021) (0.03270) (0.04877) (0.04357) (0.07123) (0.03240) (0.06506)
High -0.67726*** -1.23363*** -0.71808*** -0.96855*** -0.82345*** -0.92423*** -0.40312*** -0.28810***
(0.02910) (0.10074) (0.03465) (0.07050) (0.01243) (0.05231) (0.02396) (0.03891)
LowXt -0.00103*** -0.00096*** -0.00101*** -0.00108*** -0.00143*** -0.00142*** -0.00048*** -0.00067***
(0.00027) (0.00026) (0.00021) (0.00020) (0.00027) (0.00027) (0.00016) (0.00020)
HighXt 0.00092*** 0.00109*** 0.00080*** 0.00056*** -0.00023*** -0.00040*** 0.00042*** 0.00059***
(0.00016) (0.00019) (0.00017) (0.00020) (0.00008) (0.00009) (0.00013) (0.00014)
%male 0.11097 -0.00087 0.00103 0.67289**
(0.16147) (0.18614) (0.20706) (0.26775)
%black 0.63782** 0.61404** 0.32000 -0.25701
(0.25233) (0.27208) (0.24741) (0.25535)
%age18-35 -0.76758*** -0.03160 -0.64143** -0.71441**
(0.28487) (0.19331) (0.28449) (0.34576)
%age36-55 -0.54024** 0.09880 -0.50323** -0.99223***
(0.21795) (0.12676) (0.20934) (0.31620)
%married -0.24671 -0.19691 -0.18206 -0.03093
(0.18052) (0.16664) (0.12078) (0.11962)
166

Table 2.5A, Continued
Any Training School Training Formal Training Informal Training
(1) (2) (3) (4) (5) (6) (7) (8)
%union -0.60541 0.34902 -0.55707 -0.22821
(0.48821) (0.48239) (0.43767) (0.28510)
%high school education 0.45442** 0.25915 0.70454*** -0.51587**
(0.20994) (0.21957) (0.18398) (0.20324)
%some college education -0.12362 -0.52821* 0.38391* -0.86472**
(0.28849) (0.26657) (0.22819) (0.32653)
%college education & higher 1.81018*** 0.84222*** 2.17192*** -0.55286
(0.32440) (0.26277) (0.29073) (0.34199)
Constant -1.36499*** -1.18537*** -1.32772*** -1.36596*** -1.25036*** -1.56819*** -1.34962*** -0.48591*
(0.01699) (0.32463) (0.01054) (0.13244) (0.01949) (0.29074) (0.01048) (0.26720)
Observations 21450 21450 21450 21450 21450 21450 21450 21450
R-squared 0.83 0.86 0.86 0.88 0.90 0.92 0.82 0.84
F test 0.97087 1.16994 1.29600 1.32761 4.66885 2.72773 1.05771 1.30546
Prob > F 0.41407 0.33076 0.28633 0.27611 0.00601 0.05399 0.37563 0.28323

Robust standard errors in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
167

Table 2.5B: Public Policy and wages
Any Training School Training Formal Training Informal Training
(1) (2) (3) (4) (5) (6) (7) (8)
Adopt(PP)xLow -0.00151 -0.00547 -0.00380 -0.00674 -0.00366 -0.00654 -0.02179 -0.01825
(0.00917) (0.00722) (0.00962) (0.00933) (0.00889) (0.00684) (0.01478) (0.01096)
Adopt(PP)xMed -0.00768 -0.00897 -0.00376 -0.00684 -0.00602 -0.00745 0.00192 -0.00053
(0.00702) (0.00610) (0.00706) (0.00749) (0.00827) (0.00814) (0.00666) (0.00554)
Adopt(PP)xHigh -0.01401 -0.01430 -0.01236 -0.01354 -0.00201 -0.00419 0.00617 -0.00191
(0.01376) (0.01098) (0.01251) (0.01119) (0.01001) (0.00819) (0.01588) (0.01074)
Low -0.29825*** -0.18970*** -0.28023*** -0.16664*** -0.37092*** -0.19735*** -0.29176*** -0.20146***
(0.00860) (0.01188) (0.00869) (0.01275) (0.00880) (0.01122) (0.01115) (0.01071)
High 0.20221*** -0.00026 0.22047*** 0.00078 0.18569*** 0.09153*** 0.02744** 0.02024**
(0.00983) (0.01128) (0.00705) (0.01239) (0.00815) (0.00944) (0.01215) (0.00787)
LowXt -0.00039*** -0.00041*** -0.00035*** -0.00039*** -0.00027*** -0.00025*** 0.00061*** 0.00050***
(0.00006) (0.00005) (0.00006) (0.00005) (0.00007) (0.00006) (0.00005) (0.00005)
HighXt 0.00048*** 0.00063*** 0.00041*** 0.00056*** 0.00015** 0.00014** 0.00003 0.00007
(0.00005) (0.00004) (0.00004) (0.00004) (0.00006) (0.00006) (0.00009) (0.00006)
%male 0.32715*** 0.29813*** 0.22375*** 0.24672***
(0.01708) (0.02186) (0.01943) (0.01836)
%black -0.09594*** -0.06470 -0.05263 -0.11307**
(0.03550) (0.04467) (0.03689) (0.04361)
%age18-35 -0.16297*** -0.17896*** -0.17622*** -0.11790***
(0.04115) (0.04668) (0.03436) (0.01967)
%age36-55 0.04693 0.04565 0.05149* 0.10073***
(0.03027) (0.03555) (0.02661) (0.02302)
%married 0.12932*** 0.09994*** 0.13272*** 0.17900***
(0.02274) (0.02724) (0.02093) (0.01188)
168

Table 2.5B, Continued
Any Training School Training Formal Training Informal Training
(1) (2) (3) (4) (5) (6) (7) (8)
%union 0.28358*** 0.22734*** 0.29296*** 0.26719***
(0.04388) (0.04912) (0.04170) (0.02959)
%high school education 0.34283*** 0.36673*** 0.27857*** 0.21685***
(0.05629) (0.05637) (0.04043) (0.02226)
%some college education 0.39453*** 0.41821*** 0.35348*** 0.37829***
(0.05129) (0.05461) (0.04523) (0.03250)
%college education & higher 0.69130*** 0.73049*** 0.70415*** 0.80379***
(0.05100) (0.05113) (0.04579) (0.02879)
Constant 2.02469*** 1.41684*** 2.01694*** 1.43696*** 2.05068*** 1.48768*** 2.00850*** 1.39861***
(0.00454) (0.03256) (0.00365) (0.03362) (0.00562) (0.03237) (0.00372) (0.02671)
Observations 21450 21450 21450 21450 21450 21450 21450 21450
R-squared 0.92 0.93 0.92 0.93 0.93 0.94 0.84 0.87
F test 0.73320 1.17268 0.99351 1.44161 0.32881 0.67888 1.62417 2.26831
Prob > F 0.53722 0.32973 0.40373 0.24205 0.80452 0.56918 0.19580 0.09223

Robust standard errors in parentheses
* significant at 10%; ** significant at 5%; *** significant at 1%
169

observe similar patterns here. Since I do not, this lends additional support to the
hypothesis that the Good Faith and Implied Contract exceptions are associated with
increases in employment of high-skilled labor and are associated with decreases in
employment of low-skilled labor.
2.7 Conclusion
The common law doctrine of employment at will was established throughout the
United States by 1953. The doctrine requires that a default employment relationship be at
will, and thus can be unilaterally terminated by each party at any time for any or no
reason and without any liability. This at will doctrine although allows firms to effectively
adjust their workforces, provides workers with no employment security.
During the 1970s and 1980s, state courts imposed limitations on this employment
at will doctrine by allowing three classes of exceptions. These exceptions are often
referred to as Wrongful Discharge Laws (WDLs). These laws limit employers’ flexibility
in dismissing workers and allow workers to litigate against wrongful discharge. There are
three categories of these laws – namely, Implied Contract, Good Faith, and Public Policy.
The literature reported negative (or at best zero) impacts of WDLs on overall
employment. These studies implicitly assumed that the overall labor force was
homogeneous. They failed to recognize that labor can be heterogeneous and firms may
treat different types of labor as different forms of input. Thus, the imposition of WDLs
may influence the decisions of firms regarding not only the quantity of labor input but
also the combination of different types of labor input. In this paper, I utilize a simple
model that acknowledges the heterogeneity of labor input. Specifically, in the model,
170

firms acquire two types of labor input, namely low-skilled and high-skilled labor. The
model allows for firms to alter their production and to substitute between different types
of labor input once WDLs are imposed.
The key finding of this paper is that WDLs, particularly in the case of the Good
Faith law, are associated with increases in employment of high-skilled labor, a result that
may have been unacknowledged in early studies. WDLs, however, adversely affect the
levels of employment for low-skilled workers. This negative impact of WDLs on
employment is consistent with the literature.
These results indicate that whether intended or not, WDLs can affect different
groups of workers differently. It will be interesting to see whether other types of labor
laws also have different impacts on low-skilled and high-skilled workers. It will also be
interesting to investigate the subject matter in other countries that experienced similar
changes in legal regimes.
171
Bibliography

Acemoglu, D. & Angrist, J. D. (2001). Consequences of Employment Protection? The
Case of the Americans with Disabilities Act. Journal of Political Economy, 109(5),
915-957.

Arrow, K. (1972). The Theory of Discrimination. In O. Ashenfelter & A. Rees (Eds.),
Discrimination in Labor Markets. Princeton, NJ: Princeton University Press.

Asian Development Bank. (1998). Women in Thailand. Manila, the Philippines.

Autor, D. H., Kerr, W. R. & Kugler, A. D. (2007). Does Employment Protection Reduce
Productivity? Evidence from US States. The Economic Journal, 117(521), 189-217.

Autor, D. H. (2003). Outsourcing at Will. Journal of Labor Economics, 21(1), 1-42.

Autor, D. H., Donohue, J. J. & Schwab, S. J. (2004). The Employment Consequences of
Wrongful-Discharge-Laws: Large, Small, or None at all?. American Economic
Review Papers and Proceedings, 94(2), 440-446.

Autor, D. H., Donohue, J. J. & Schwab, S. J. (2006). The Costs of Wrongful Discharge
Laws. Review of Economics and Statistics, 88(2), 211-231.

Barsky, R., Bound, J., Charles, K. K. & Lupton, J. P. (2002). Accounting for the Black-
White Wealth Gap: A Nonparametric Approach. Journal of the American Statistical
Association, 97(459), 663-673.

Becker, G. (1957). The Economics of Discrimination. Chicago, IL: University of Chicago
Press.

Bertrand, M., Duflo, E. & Mullainathan, S. (2004). How Much Should We Trust
Differences-in-Differences Estimates?. Quarterly Journal of Economics, 119(1), 249-
275.

Besley, T. & Burgess, R. (2004). Can Labor Regulation Hinder Economic Performance?
Evidence from India. Quarterly Journal of Economics, 119(1), 91-134.

Black, D., Haviland, A., Sanders, S. & Taylor, L. (2006). Why Do Minority Men Earn
Less? A Study of Wage Differentials among the Highly Educated. The Review of
Economics and Statistics, 88(1), 300-313.

Blinder, A. S. (1973). Wage Discrimination: Reduced Form and Structural Estimates. The
Journal of Human Resources, 8(4), 436-455.

172
Botero, J. C., Djankov, S., La Porta, R., Lopez-de Silanes, F. & Shleifer, A. (2004). The
Regulation of Labor. Quarterly Journal of Economics, 119(4), 1339-1382.

Cameron, L. A., Dowling, J. M. & Worswick, C. (2001). Education and Labor Market
Participation of Women in Asia: Evidence from Five Countries. Economic
Development and Cultural Change, 49(3), 459-477.

Cotton, J. (1988). On the Decomposition of Wage Differentials. The Review of
Economics and Statistics, 70(2), 236-243.

Daly, A., Kawaguchi, A., Meng, X. & Mumford, K. (2006). The Gender Wage Gap in
Four Countries. The Economic Record, 82(257), 165-176.

DeLeire, T. (2000). The Wage and Employment Effects of the Americans with
Disabilities Act. Journal of Human Resources, 35(4), 693-715.

Deolalikar, A. B. (2002). Poverty, Growth, and Inequality in Thailand. ERD Working
Paper Series, No. 8, Asian Development Bank, Manila, the Philippines.

Dertouzos, J. N. & Karoly, L. A. (1992). Labor Market Responses to Employer Liability.
Santa Monica, CA: Rand, Institute for Civil Justice.

DiNardo, J. (2002). Propensity Score Reweighting and Changes in Wage Distributions.
Meimo.

DiNardo, J., Fortin, N. M. & Lemieux , T. (1996). Labor Market Institutions and the
Distribution of Wages, 1973-1992: A Semiparametric Approach. Econometrica,
64(5), 1001-1044.

Funatsu, T. & Kagoya, K. (2003). The Middle Classes in Thailand: The Rise of the Urban
Intellectual Elite and Their Social Consciousness. The Developing Economies, 41(2),
243-263.

Gannicott, K. (1986). Women, Wages, and Discrimination: Some Evidence from Taiwan.
Economic Development and Cultural Change, 34(4), 721-730.

Garen, J. & Jeraputtiruk, N. (2005). Mandated Fringe Benefits in an Economy with
Covered and Uncovered Sectors: The Case of Labor Market Regulation in Thailand.
Meimo.

Grubb, D. & Wells, W. (1993). Employment Regulation and Patterns of Work in EC
Countries. OECD Economic Studies, 21, 7-58.

Gustafsson, B. & Li, S. (2000). Economic Transformation and the Gender Earnings Gap
in Urban China. Journal of Population Economics, 13, 305-329.
173

Hawley, J. D. (2004). Changing Returns to Education in Times of Prosperity and Crisis,
Thailand 1985-1998. Economics of Education Review, 23, 273-286.

Heckman, J. J. & Pages, C. (2000). The Cost of Job Security Regulation : Evidence from
Latin American Labor Markets. NBER Working Papers, No. 7773, National Bureau
of Economic Research.

Heckman, J. J. & Pages, C. (2000). Law and Employment: Lessons from Latin America
and the Caribbean. Chicago, IL: University of Chicago Press.

Hill, M. A. & King, E. M. (1991). Women’s Education in Developing Countries: An
Overview. In M. A. Hill & E. M. King (Eds.), Women’s Education in Developing
Countries: Barriers, Benefits and Policies (pp. 1-50). Washington, DC: World Bank.

Jansen, K. (2001). Thailand, Financial Crisis and Monetary Policy. Journal of the Asia
Pacific Economy, 6(1), 124-152.

Jeong, H. (2007). Assessment of Relationship between Growth and Inequality: Micro
Evidence from Thailand. Macroeconomic Dynamics, Forthcoming.

Jeraputtiruk, N. (2004). The Effects of Labor Regulation in Developing Countries: The
Evidence on Hourly Earnings Compensation from Thailand. Meimo.

Jolls, C. & Prescott, J. J. (2004). Disaggregating Employment Protection: The Case of
Disability Discrimination. NBER Working Papers, No. 10740, National Bureau of
Economic Research.

Juhn, C., Murphy, K. M. & Pierce, B. (1991). Accounting for the Slowdown in Black-
White Wage Convergence. In M. Kosters (Ed.), Workers and Their Wages.
Washington, DC: AEI Press.

Kawaguchi, A. (2004). Changes in the Japanese Gender Wage Gap in the 1990s. paper
presented to the 2004 Conference on Feminist Economics, Oxford, August.

Knodel, J. (1997). The Closing of the Gender Gap in Schooling: The Case of Thailand.
Comparative education, 33(1), 61-86.

Krueger, A. B. (1991). The Evolution of Unjust-Dismissal Legislation in the United
States. Industrial and Labor Relations Review, 44(4), 644-660.

Kugler, A. D. & Saint-Paul, G. (2004). How Do Firing Costs Affect Worker Flows in a
World with Adverse Selection?. Journal of Labor Economics, 22(3), 553-583.

174
Lauridsen, L. S. (1998). The Financial Crisis in Thailand: Causes, Conduct and
Consequences?. World Development, 26(8), 1575-1591.

Lazear, E. P. (1990). Job Security Provisions and Employment. Quarterly Journal of
Economics, 105(3), 699-726.

Lemieux, T. (2002). Decomposing Changes in Wage Distributions: A Unified Approach.
The Canadian Journal of Economics, 35(4), 646-688.

Liu, A. Y. C. (2004). Gender Wage Gap in Vietnam: 1993 to 1998. Journal of
Comparative Economics, 32, 586-596.

Liu, P. W., Meng, X. & Zhang, J. (2000). Sectoral Gender Wage Differentials and
Discrimination in the Transitional Chinese Economy. Journal of Population
Economics, 13, 331-352.

Meng, X. (1996). The Economic Position of Women in Asia. Asian-Pacific Economic
Literature, 10(1), 23-41.

Miles, T. J. (2000). Common Law Exceptions to Employment at Will and US Labor
Markets. Journal of Law, Economics & Organization, 16(1), 74-101.

Motonishi, T. (2003). Why Has Income Inequality in Thailand Increased? – An Analysis
Using 1975-1998 Surveys. ERD Working Paper Series, No. 43, Asian Development
Bank, Manila, the Philippines.

Muhl, C. J. (2001). The Employment-at-Will Doctrine: Three Major Exceptions. Monthly
Labor Review, 124(1), 3-11.

National Statistical Office of Thailand. (2003a). Report of the Labor Force Survey,
Whole Kingdom, Quarter 2 (April – June 2003). Bangkok, Thailand.

National Statistical Office of Thailand. (2003b). Report of the Labor Force Survey,
Whole Kingdom, Quarter 4 (October – December 2003). Bangkok, Thailand.

Neumark, D. (1988). Employers’ Discriminatory Behavior and the Estimation of Wage
Discrimination. The Journal of Human Resources, 23(3), 279-295.

Oaxaca, R. L. (1973). Male-Female Wage Differentials in Urban Labor Markets.
International Economic Review, 14(3), 693-709.

Oyer, P. & Schaefer, S. (2000). Layoffs and Litigation. Rand Journal of Economics,
31(2), 345-358.

175
Phananiramai, M. (1993). Women’s Economic Roles in Thailand. TDRI Working Paper,
Thailand Development Research Institute, Bangkok, Thailand.

Pirmana, V. (2006). Earnings Differential between Male-Female in Indonesia: Evidence
from Sakernas Data. Working Paper in Economics and Development Studies, No.
200608, Department of Economics, Padjadjaran University, Bandung, Indonesia.

Reimers, C. W. (1983). Labor Market Discrimination Against Hispanic and Black Men.
The Review of Economics and Statistics, 65(4), 570-579.

Rodgers, Y. M. (1998). A Reversal of Fortune for Korean Women: Explaining the 1983
Upward Turn in Relative Earnings. Economic Development and Cultural Change,
46(4), 727-748.

Rosen, S. (1968). Short-Run Employment Variation on Class-I Railroads in the US,
1947-1963. Econometrica, 36, 511-529.

Schanzenbach, M. (2003). Exceptions to Employment at Will: Raising Firing Costs or
Enforcing Life-Cycle Contracts?. American Law and Economics Review, 5(2), 470-
504.

Schultz, T. P. (1991). Returns to Women’s Education. In M. A. Hill & E. M. King (Eds.),
Women’s Education in Developing Countries: Barriers, Benefits and Policies (pp. 51-
99). Washington, DC: The World Bank.

Tiliak, J. B. G. (1991). East Asia. In M. A. Hill & E. M. King (Eds.), Women’s Education
in Developing Countries: Barriers, Benefits and Policies (pp. 247-284). Washington,
DC: The World Bank.

Schafgans, M. M. A. (2000). Gender wage differences in Malaysia: parametric and
semiparametric estimation. Journal of Development Economics, 63, 351-378,

Silverman, B. W. (1986). Density Estimation for Statistics and Data Analysis. London,
England: Chapman & Hall.

Smith, J. P., Thomas, D., Frankenberg, E., Beegle, K. & Teruel, G. (2002). Wages,
Employment and Economic Shocks: Evidence from Indonesia. Journal of Population
Economics, 15, 161-193.

Thomas, D., Beegle, K. & Frankenberg, E. (2003). Labor Market Transitions of Men and
Women During an Economic Crisis: Evidence from Indonesia, In B. Garcia, R. Anker
& A. Pinnelli (Eds.), Women in the Labour Market in Changing Economies:
Demographic Issues (pp. 37-58). Oxford: Oxford University Press.

176
Tsurumi, H. (2000). Asian Financial Crisis. Prologue and the Case of Thailand. Asia-
Pacific Financial Markets, 7, 1-9.

United Nations Development Programme. (1995). Human Development Report. New
York, NY: Oxford University Press.

United Nations Development Programme. (2006). Human Development Report. New
York, NY: Oxford University Press.

US Bureau of the Census. (1989). The Relationship between the 1970 and 1980 Industry
and Occupation Classification Systems. Technical Paper 59. Washington, DC: US
Government Printing Office.

Warr, P. (1999). What Happened to Thailand? The World Economy, 22, 631-650.

Warr, P. (2004). Globalization, Growth, and Poverty Reduction in Thailand. ASEAN
Economic Bulletin, 21(1), 1-18.

World Bank. (2005). East Asia Update, Special Focus: Gender Equality in East Asia.
Washington, DC.

World Bank. (2006a). Indonesia Gender Assessment. Washington, DC.

World Bank. (2006b). Thailand Economic Monitor. Washington, DC.

World Bank. (2006c). Vietnam Gender Assessment. Washington, DC.

Zveglich, J. E., Rodgers, Y. M. & Rodgers, W. M. (1997). The Persistence of Gender
Earnings Inequality in Taiwan, 1978-1992. Industrial and Labor Relations Review,
50(4), 594-609.
177
Appendix

Appendix Table A.1: Actual raw data availability of the Thai LFS from 1985-2005

Q1: Jan-Mar
(Feb round)
Q2: Apr-Jun
(May round)
Q3: Jul-Sep
(Aug round)
Q4: Oct-Dec
(Nov round)
1985 (2528) Y Y Y
1986 (2529) Y Y
1987 (2530) Y Y Y
1988 (2531) Y Y
1989 (2532) Y Y Y
1990 (2533) Y
1991 (2534) Y Y Y
1992 (2535) Y Y Y
1993 (2536) Y Y
1994 (2537) Y Y Y
1995 (2538) Y Y
1996 (2539) Y Y Y
1997 (2540) Y Y
1998 (2541) Y Y Y Y
1999 (2542) Y Y Y Y
2000 (2543) Y Y Y Y
2001 (2544) Y Y Y Y
2002 (2545) Y Y Y Y
2003 (2546) Y Y Y Y
2004 (2547) Y Y Y Y
2005 (2548) Y Y Y Y

Y: available
178
Appendix Table A.2: Conversion Methodology for Monthly Wages

Wage
Type
1985-2000 2001-2005
Hourly monthly earnings = (reported) hourly wage * hours of work per week * 4.2
monthly earnings = (reported) average
monthly earnings
Daily monthly earnings = (reported) daily wage * days of work per week * 4.2
monthly earnings = (reported) average
monthly earnings
Weekly monthly earnings = (reported) weekly wage * 4.2
monthly earnings = (reported) average
monthly earnings
Monthly monthly earnings = (reported) monthly earnings
monthly earnings = (reported) average
monthly earnings
Other monthly earnings = (reported) average daily wage * days of work per week * 4.2
monthly earnings = (reported) average
monthly earnings
Unknown
monthly earnings = mid point of chosen salary range
(highest and lowest salary ranges were dropped)
monthly earnings = (reported) average
monthly earnings
Non-
Cash
N/A
monthly earnings = (reported) average
monthly earnings

179

Abstract (if available)

Abstract The dissertation consists of two essays on labor and development economics. The first essay seeks to identify the main factors that contributed to the decline in gender earnings gap in Thailand's wage and salary sector from 1985-2005. Two parametric methodologies, Neumark's version of the Blinder-Oaxaca method and the Juhn-Murphy-Pierce method, are implemented in order to decompose gender earnings gap at a point in time and across time period. I also make a methodological contribution by proposing a way to modify the DiNardo-Fortin-Lemieux nonparametric decomposition method so that its results are comparable to those from Neumark's version of the Blinder-Oaxaca method. The key findings of this essay are as follows. First, I find that increases in female education and changes in unobserved factors, which were concurrent with modernization, were the main sources of the decline in gender earnings gap. Second, over time, improvements in the education of females in this sector surpassed that of males. However, the superior education of females did not result in higher female earnings because of the overwhelming effect of the unexplained factors that supported higher male earnings. Finally, the nonparametric investigation corroborated the results from the parametric methodologies.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Essays on family and labor economics

PDF

The EITC, labor supply, and child development

PDF

Essays in development economics

PDF

Essays on development and health economics: social media and education policy

PDF

Empirical analysis of factors driving stock options grants and firms volatility

PDF

Three essays on economics of early life health in developing countries

PDF

Essays on development and health economics

PDF

Essays in the economics of education and conflict

PDF

Three essays on cooperation, social interactions, and religion

PDF

Essays on economics of education and private tutoring

PDF

Essays in labor economics: demographic determinants of labor supply

PDF

Three essays on human capital and family economics

PDF

Essays in development and experimental economics

PDF

Essays on health economics

PDF

Essays on development economics and adolescent behavior

PDF

Selected papers on the evaluation of healthcare costs of prematurity and necrotizing enterocolitis using large retrospective databases

PDF

Improvements in women's status, decision-making and child nutrition: evidence from Bangladesh and Indonesia

PDF

Essays on narrative economics, Climate macrofinance and migration

PDF

Essays on environmental economics: education, employment, and experiments

PDF

Long-term impacts of childhood adversity on health and human capital

Asset Metadata

Creator Nakavachara, Voraprapa (author)

Core Title Essays on labor and development economics

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Economics

Publication Date 10/22/2007

Defense Date 10/10/2007

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Blinder Oaxaca,DiNardo Fortin Lemieux,gender inequality,Juhn Murphy Pierce,OAI-PMH Harvest,Thailand,Wrongful Discharge Law

Place Name Thailand (countries), USA (countries)

Language English

Advisor Strauss, John A. (committee chair), Ham, John C. (committee member), MacLeod, W. Bentley (committee member), Painter, Gary D. (committee member)

Creator Email nakavach@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m879

Unique identifier UC1322825

Identifier etd-Nakavachara-20071022 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-581751 (legacy record id),usctheses-m879 (legacy record id)

Legacy Identifier etd-Nakavachara-20071022.pdf

Dmrecord 581751

Document Type Dissertation

Rights Nakavachara, Voraprapa

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu