2. Background

Utilizing the data from the national labor force survey (LFS) of Thailand from 1985 to 2016, we present the overall picture of income inequality and educational assortative mating in Thailand. The Gini coefficient is the most commonly applied measure of inequality. The Gini coefficient is based on the Lorenz curve, defined as the share of the concentration area compared to the maximum concentration area [Ravallion (Reference Ravallion1988)]. We computed the Gini coefficient by following Šošić (Reference Šošić2004, p. 125):

$$G = 2\mathop \sum \limits_{i = 1}^n iy_i-( {n + 1} ) \mathop \sum \limits_{i = 1}^n y_i/n\mathop \sum \limits_{i = 1}^n y_i$$

where y _i is the income of the first i population units.

Online Appendix Table 1A presents the Gini coefficient in terms of household income and household income per capita in Thailand. The four categories exhibited similar trends in the changes in their Gini coefficients. In 1985, for example, the Gini coefficient for one-couple household income was 0.425, which decreased to 0.389 in 2015 (the trend is illustrated in Online Appendix Figure 1A). Our computed Gini coefficients were consistent with the decreasing trend mentioned by the World Bank (2018).

Online Appendix Figure 2A presents the educational assortative mating trend in Thailand. During the 1980s, the trend was consistent with Smits and Park (Reference Smits and Park2009), indicating that Southeast Asian countries like the Philippines, Indonesia, Malaysia, and Thailand had a relatively high level of educational homogamy compared to other countries in the world. However, after the 1990s, Thailand showed a decreasing trend of educational homogamy, where the proportion of spouses with the same educational attainment has been decreasing over time.

However, when we stratified by education level, we discovered that educational homogamy has only decreased in the group with only a primary education, while educational homogamy among couples who obtained higher education levels (secondary and university) has increased over time (Online Appendix Figure 3A). This pattern of increasing homogamy is consistent with evidence from the US. Eika et al. (Reference Eika, Mogstad and Zafar2019) documented that there was a significant increase in the share of couples who both obtained college degrees between 1962 and 2013, as well as an increase in the Gini coefficient in households during this period.

Even though it seems there exists a relationship between educational assortative mating and income inequality, difficulties were found in the estimation of the educational assortative mating-income inequality nexus. In particular, the documented changes in educational attainment and the increase in the economic return to education have created difficulties in measuring the impact of educational assortative mating on household income inequality.

Eika et al. (Reference Eika, Mogstad and Zafar2019) suggested that it was hard to identify changes in educational homogamy because of changes in educational attainment or educational assortative mating. With higher education, the probability will increase for a university degree holder to marry someone with the same education level, even if assortative mating does not change [Liu and Lu (Reference Liu and Lu2006)]. Moreover, as the return to education has increased over time, it made educational assortative mating more significant to the distribution of household income, with or without changes in mating behavior. To overcome this problem, Eika et al. (Reference Eika, Mogstad and Zafar2019) suggested a two-step procedure to analyze educational assortative mating and its influence on household income inequality, which we will discuss later in the methodology section.

As a result, we could not neglect these two main factors in this case study of Thailand. Online Appendix Figure 4A presents the similar trends for husbands' and wives' educational attainment, which found that the proportion of husbands and wives who attained only primary education decreased, but the proportion with secondary and university degrees increased. This increasing trend is the result of a successful policy of the Thai government to improve education levels, a compulsory education policy.

The changes in women's educational attainment led to changes in their financial contribution to the household income [Oppenheimer (Reference Oppenheimer1994)]. Although the labor force participation rate for married females in Thailand remained relatively stable over time (Online Appendix Figure 5A), the promotion of female education and the rise in their wages led to an increase in their share of household income from 43% to 47% (Online Appendix Figure 6A). An increase in the correlation of men's and women's income within households, accompanied with the increases in assortative mating, the propensity of women to work, and return to education [Reed and Cancian (Reference Reed and Cancian2009)].

Furthermore, based on a simple Mincerian wage regression, Online Appendix Table 2A reports the Ordinary Least Squares (OLS) estimates for a married individual's real log hourly wage based on years of schooling and potential work experience. The main finding is that the returns to education have decreased over the years 1985–2016. On average, one-year increase in education led to an increase in log hourly wage of approximately 0.163, while people with Bachelor's degrees earned a wage premium of around 40%.

3. Literature review

Educational assortative mating has been defined many different ways; however, the most common definition in the literature is men and women with similar education levels marrying more or less frequently than random matching [e.g. Schwartz and Mare (Reference Schwartz and Mare2005), Breen and Andersen (Reference Breen and Andersen2012), Olivo-Villabrille (Reference Olivo-Villabrille2017)]. Becker (Reference Becker1973, Reference Becker1974) proposed that people tend to search for a partner to marry in the marriage market like a matching game, while in Greenwood et al. (Reference Greenwood, Guner, Kocharkov and Santos2016), individuals met their potential partners by drawing from the distribution of singles, responding to educational choices, and by choosing whether or not to marry.

Several previous studies have found that educational assortative mating explained household income inequality [Burtless (Reference Burtless1999), Breen and Salazar (Reference Breen and Salazar2010), Schwartz (Reference Schwartz2010)]. Previous studies in the U.S. have found that assortative mating is related to changes in income inequality. Schwartz (Reference Schwartz2010) suggested that men's and women's preferences for mates became more and more symmetric as female labor force participation grew, generating greater inequality across couples. Breen and Salazar (Reference Breen and Salazar2010) and Esping-Andersen (Reference Esping-Andersen2007) found that the increase in women's education and labor force participation, combined with educational homogamy, caused greater household inequality in the US and European countries.

Katz and Autor (Reference Katz, Autor, Ashenfelter and Card1999) suggested that the rise in income inequality across families was mainly caused by the increasing correlation between the earnings of spouses. In Germany, Pestel (Reference Pestel2016) examined the effect of marital sorting on income inequality while considering labor supply choice and found that, where women are more attached to the labor market, this effect is disequalized, irrespective of adjusting for labor supply choices. Eika et al. (Reference Eika, Mogstad and Zafar2019) demonstrated that an increase in educational homogamy contributed an increase in income inequality in the US, Denmark, Germany, and the UK. This evidence is consistent with many empirical studies: e.g., Western et al. (Reference Western, Bloome and Percheski2008) and Atkinson and Morelli (Reference Atkinson and Morelli2014) for the US; and Breen and Salazar (Reference Breen and Salazar2010) for the UK.

Even though several studies have found that assortative mating related to changes in income inequality, some studies showed only small effects on income inequality. Olivo-Villabrille (Reference Olivo-Villabrille2017) suggested assortative mating alone had little effect on household income inequality, while Eika et al. (Reference Eika, Mogstad and Zafar2019) found that assortative mating accounted for a modest but non-negligible part of the inequality. Kremer (Reference Kremer1997), Cancian and Reed (Reference Cancian and Reed1999), and Hryshko et al. (Reference Hryshko, Juhn and McCue2017) found that assortative mating played a minor role in income inequality or had negative effects on inequality, while Schwartz (Reference Schwartz2010) and Greenwood et al. (Reference Greenwood, Guner, Kocharkov and Santos2014) found a positive relationship between the two.

A few related studies were found about developing countries. Torche (Reference Torche2010) explored the isomorphic association between marital sorting and inequality in Brazil, Chile, and Mexico, and found substantial variation in marital sorting between the countries. Meanwhile, Dahan and Gaviria (Reference Dahan and Gaviria2001) presented evidence on 16 Latin American countries. Hu and Qian (Reference Hu and Qian2015) found that increasing educational homogamy for married couples, among those with senior high and tertiary educations in urban China, was associated with an increase in inter-household income inequality and a decrease in intra-household income inequality.

We found one study mentioning educational assortative mating in Thailand, which is the study of Smits and Park (Reference Smits and Park2009), who studied educational assortative mating in 10 Asian countries, including Thailand. They found that the decreasing trend in educational homogamy corresponded to educational expansion. The paper also suggested that countries with higher female employment and more modern societies tended to have lower educational homogamy. Liao and Paweenawat (Reference Liao and Paweenawat2021) examined the educational expansion and high labor force participation rate of married Thai women over time, and found an inverse relationship between married women's labor supply and wages.

On the other hand, several studies on Thailand have focused on the determinants of inequality in the country. Meesook (Reference Meesook1979) found a decrease in inequality as a result of economic growth, but over time, several studies have found the opposite result [Krongkaew (Reference Krongkaew1985), Krongkaew et al. (Reference Krongkaew, Tinakorn and Suphachalasai1992), Israngkura (Reference Israngkura2003)]. Fofack and Zeufack (Reference Fofack and Zeufack1999) using Socio-Economic Survey (SES) data from 1986 to 1996, suggested that income inequality in Thailand was negatively correlated with average education. Paweenawat and McNown (Reference Paweenawat and McNown2014), using SES data from 1992 to 2011, found a positive relationship between years of schooling and income inequality. However, none of the studies took into account educational assortative mating as a main factor associated with income inequality in Thailand.

4. Data and methodology

4.2 Methodology

Following Eika et al. (Reference Eika, Mogstad and Zafar2019), this study adapted a two-step procedure to analyze educational assortative mating and its influence on household income inequality. First, we calculated the assortative mating parameter to examine its magnitude and how it changed over time. Then, we employed the semiparametric decomposition method of DiNardo et al. (Reference DiNardo, Fortin and Lemieux1996) to test whether the increase in educational attainment, the increase in the return to education, and the changes in assortative mating could be the variables for the changes in household income inequality in Thailand over three decades.

We first needed to compute assortative mating by following the method proposed by Eika et al. (Reference Eika, Mogstad and Zafar2019). The marital sorting parameters were the observed probability of couples that had the same education level, relative to the probability when spouses were matched randomly in terms of education attainment:

(1)$$s( {e_w, \;e_h} ) = P( {E_w = e_w, \;E_h = e_h} ) /P( {E_w = e_w} ) P( {E_h = e_h} ) \;$$

where e _w, e _h were the observed education levels for wives and husbands, and E _w, E _h denoted the education levels for wives and husbands. If the parameter is larger than 1, men and women tend to marry individuals with the same level of education more frequently than random chance, indicating positive educational assortative mating.

In addition, we also applied an alternative method proposed by Greenwood et al. (Reference Greenwood, Guner, Kocharkov and Santos2016) and Eika et al. (Reference Eika, Mogstad and Zafar2019) to compute educational assortative mating. We applied a regression model for wives' and husbands' years of schooling to estimate educational assortative mating [equation (2)]. We ran the regression on wives'(husbands') years of schooling on an intercept and husbands' (wives') years of schooling:

(2)$$Edu_y^w = \alpha + \beta Edu_y^h + {\rm Yea}{\rm r}_{ty} + \varepsilon $$

where $Edu_y^w$, $Edu_y^h$ denote the years of schooling for wives (w) and husbands (h) in year y.

The variable Year _ty is a time dummy, where Year _ty = 1 if t = y, and Year _ty = 0 if t ≠ y; t ∈ T ≡ {1985, 1986, 1987, 1988, …, 2016}.

The coefficient (β) measures the changes in assortative mating over time.

Besides, we also used the SES panel in 2005 to 2012, conducted by NSO of Thailand to analyze the assortative mating in recent years. We applied probit model estimating the effect of husband's education level on wife's education level.

$$\Pr ( y_{it} = 1\vert z_{it}, \;x_{it}) = \phi ( {z_{it}\theta + x_{it}\gamma } ) $$

where y _it is an indicator that equals 1 if wife has the education (either primary, secondary or university) in the contingency table. z _it is a dummy variable, indicating the husband obtained either primary, secondary or university education if equals 1, 0 otherwise; x _it is the control variables including age, age squares. θ represents the probability of choosing a husband with certain education level for wife with the certain education levels in the contingency table.

To measure the contribution of educational assortative mating, educational attainment, and return to education to changes in household inequality, we applied the DFL decomposition method [DiNardo et al. (Reference DiNardo, Fortin and Lemieux1996)], which provides income distributions under several counterfactual scenarios. Similar to the counterfactual income distribution applied by Eika et al. (Reference Eika, Mogstad and Zafar2019), we fixed the distribution of one of the three factors at a base year while others varied over time, and compared it with the actual specification. Through a comparison of the counterfactual scenario to actual income distribution in the base year, we could observe how household income inequality has been affected over time.

The distribution of income at time t is:

(3)$$F_I( i\vert t) = \mathop \int \nolimits^ F_{I\vert X}( i\vert x, \;t) dF_X( x\vert t) \;$$

where i is household income and x is educational attainment. F _I|X(i|x, t) is the conditional distribution of income for married couples with characteristic x (educational assortative mating, educational attainment, or return to education) in year t, and F _X(x|t) is the joint distribution of spouses' education at time t.

We let t _i be the year when the factors are measured, t _x be the year when education is measured, and t _p be the year when the educational marital sorting parameter is measured. The income distribution under the counterfactual scenario is:

(4)$$\tilde{F}_I( i\vert t_i, \;t_x, \;t_p) = \mathop \int \nolimits^ F_{I\vert X}( i\vert x, \;t_i) \psi _x( x\vert t_i, \;t_x, \;t_p) dF_X( x\vert t_i) \;$$

(5)$${\rm where}\;\psi _x( x\vert t_i, \;t_x, \;t_p) = \displaystyle{{d{\tilde{F}}_X( x\vert t_x, \;t_p) } \over {dF_X( x\vert t_i) }}$$

where $d\tilde{F}_X( x\vert t_x, \;t_p)$ is the joint distribution of education measured at time t _x, and marital sorting parameter measured at time t _p.