Tóm tắt Luận án Quantile regression decomposition of the wage gap in Vietnam

MINISTRY OF EDUCATION AND TRAINING UNIVERSITY OF ECONOMICS OF HO CHI MINH CITY -------------- TRẦN THỊ TUẤN ANH QUANTILE REGRESSION DECOMPOSITION OF THE WAGE GAP IN VIETNAM DISSERTATION SUMMARY HO CHI MINH CITY, 2015 ii MINISTRY OF EDUCATION AND TRAINING UNIVERSITY OF ECONOMICS OF HO CHI MINH CITY ------------ TRẦN THỊ TUẤN ANH QUANTILE REGRESSION DECOMPOSITION OF THE WAGE GAP IN VIETNAM Major : Probability and Statistics. Major code : 62.46.01.06 DISSERTATION SUMMARY SUPERVISORS: 1. ASSOC. PROF. PH.D. LÊ VĂN PHI 2. PH.D. BÙI PHÚC TRUNG iii The research is completed at University of Economics Ho Chi Minh City: ................................................................................ Supervisors: 1. Assoc. Prof. Ph.D. Lê Văn Phi 2. Ph.D. Bùi Phúc Trung Examiner 1: ............................................................ Examiner 2: ............................................................ Examiner 3: ............................................................ The dissertation will be defended at dissertation councils, meeting at: University Of Economics Hồ Chí Minh City at ............................................................................. ............................................................................... You can find more information about the dissertation at: National Library or the Library of the University of Economics Ho Chi Minh City. INTRODUCTION 1. The necessary of the topic Wage is one of the most important factors in motivating employees. Because wage depends on a variety of determinants, the existence of the wage gap is inevitable. According to economic theory, the wage gap can be decomposed into two main components. The first component is due to the difference in endowments of the workers. The second one is due to the difference in the coefficients or due to market returns to the endowments. The second component is statistical evidence of discrimination that can lead to inequality in society. Therefore, the main purposes of this study are (1) to estimate the wage regression in Vietnam, (2) to examine the existence of gender and urban/rural wage gap, and (3) to decompose these wage gaps to clarify whether there are wage discrimination in Vietnam throughout the wage distribution. These are the reasons that lead to this doctoral dissertation which is titled as “Quantile regression decomposition of the wage gap in Vietnam” 2. Research objectives This dissertation aim to fulfill the following objectives: 1) Briefly summarizing the background of quantile regression and decomposition method based on quantile regression to analyze the wage gap. 2) Applying advanced quantile regression which account for sample selection bias and the endogeneity of explanatory variables to 2 estimate wage equations for men/women and urban/rural groups in Vietnam across their wage distribution. 3) Determining the gender wage gap in Vietnam and decomposing this gap into the explained and unexplained components during the period from 2002- 2012. 4) Determining the urban/rural wage gap in Vietnam and decomposing this gap into the explained and unexplained components during the period from 2002-2012. 5) Examine the change of wage distribution over the years by comparing quantiles of wage in 2002 with that in 2012. This difference in wage is also decomposition into two components: the one that caused by the change in labor force’s characteristics and the other due to the change in the return of these characteristics. 3. The theoretical and empirical contributions Along with these research objectives this dissertation have some following theoretical and empirical contributions: (a) This dissertation briefly introduces the definition and features of quantile regression method which was first suggested by Koenker & Bassett (1978) and has been used widely around the world but still not popular in Vietnam. There is a few of studies in Vietnam applied quantile regression techniques, especially in the area of wage inequality. None of them cover fully features of quantile regression. 3 (b) Using the advanced quantile regression, this study estimates the wage equations in Vietnam which help examine the determinants of real hourly wage in domestic labor market. The quantile regression techniques applied in this studies was adjusted to account for the problem of sample selection bias and endogeneity that leads to unbiased and consistent estimators. (c) This study constructs the wage equations across the quantiles for each following groups: men, women, urban, and rural. These results are compared in pairs to clarify the difference in their wage structures. (d) This study confirms the existence and estimates the magnitude of gender wage differentials in Vietnam (for the entire sample and for each male/female and urban/rural group of workers). In addition, this study also shows the declined trends of gender wage gap over time in Vietnam. (e) After showing the existence of gender wage differential, this research use Machado – Mata method to decompose this gap into two components: the first component represents difference in average characteristics between men and women; the second component represents differences in returns to these characteristics which may be interpreted as possible gender discrimination. (f) This study demonstrates the urban - rural wage differential and the change of this gap over time by comparing the estimation in the year 2002 with that in the year 2012. 4 (g) This dissertation decomposes the urban/rural wage differential in order to determine the proportion of this disparity which caused by the difference in endowments between urban and rural workers and the proportion of this gap caused by the difference in the market returns to the endowments. (h) After all, this research illustrates in details the change in wage equation over time and shows the decreasing trends in these wage gap over time. CHAPTER 1 THE BACKGROUND OF QUANTILE REGRESSION AND MACHADO – MATA DECOMPOSITION 1.1. Mincerian wage model and some extensions The Mincerian wage equation may be written as 2ln ,tw s z z       where w: real hourly wage; s: years of schoolings, and z: worker’s years of experience. Card (1994) extended the standard Mincer’s wage equation as 2ln ,tw s z z X u          (1.1) where X represents for control variables such as gender, race, region, marriage status, and so on. After Card D. (1994), many studies also augmented the Mincerian wage model by including various explanatory variables into the equation to examine the determinants of compensation and to conduct the ceteris paribus analysis of partial effects on wage. 5 1.2. Quantile regression Quantile regression which was introduced by Koenker & Bassett in 1978 is a method for describing the causality relationship variables at different points in the conditional distribution of the dependent variable. Considering the linear regression model i i iY X u  , the quantile regression estimator for each quantile τ ϵ (0, 1) minimizes the objective function 1 1 ( ) ( n i i i V Y X n               In the other word, 1 1ˆ arg min ( ). k n i i R i Y X n             The quantile regression for quantile τ is written as ˆ( | )i i iQ Y X X  (1.14) 1.3. Sample selection bias correction The problem of sample selection bias correction for linear regression with the pioneering work of Heckman (1979) has been extensively studied in econometrics and in labor economics. Buchinsky (1998a and 2001) was the first to consider the difficult problem of estimating quantile regression in the presence of sample selection and to propose the correction for this bias in the quantile regression. 1.4. Endogeneity and the method of two - stage quantile regression (2SQR) 6 Chevapatrakul et al (2009) suggested the method named 2SQR (two-stage quantile regression) in order to account for the problem of endogeneity in the quantile regression. 1.5. The decomposition method based on quantile regression A decomposition analysis is a standard approach to examine the wage differential between male and female workers. According to Oaxaca - Blinder (1973)’s approach, the mean wage differential is decomposed into one component capturing differences in characteristics and another component referring to different returns using the estimates of male and female wage equations). Analogous to the linear regression case, Machado and Mata (2005) proposed a similar decomposition which combines a quantile regression and a bootstrap approach in order to estimate counterfactual density functions. CHAPTER 2 LITERATURE REVIEW 2.1. Previous studies around the world Some representative studies in investigating the determinants of wage and the wage gap decomposition before the appearance of quantile regression are Edgewort (1922); Becker (1957); Dunlop (1957); Slichter (1950); Cullen (1956); Dalton & Ford (1977); Long & Link (1983); Dickens & Katz (1987); Krueger & Summers (1988); Groshen (1991); Ferber & Green (1982); Lindley, Fish & Jackson (1992); Blackaby et al (2005) 7 Buchinsky (1994) initiated the application of quantile regression in estimating wage regression. This led to a trend of using quantile regressions in order to decompose the gender wage gap at different points of the wage distribution. It can be listed some noticeable studies as Fortin and Lemieux (1998); Ajwad et al (2002); Albrecht et al (2003); Machado & Mata (2005); Melly (2006); Gunawardena (2006); Arulampalam et al (2007); Nestic (2010); Del Río, Gradín & Canto (2011). 2.2. Previous studies in Vietnam Very few studies in Vietnam applied quantile regression to investigate wage differentials as well as decompose these wage differentials into explained and unexplained parts. The typical studies can be listed are Hung et al (2007a) and Hung Ho et al (2007b). However, these studies which used the VHLSS 2002 did not account for the problem of endogeneity. CHAPTER 3 DATA AND METHODOLOGY 3.1. Data This study uses the VHLSS 2002 and 2012 to estimate the wage equation in Vietnam labor market and conduct an empirical investigation of wage differentials between the male and female workers as well as the urban and rural areas. In order to dispose of the wage change due to inflation, the data was deflate to obtain the comparable real wages. 8 By comparing the kernel density estimation of wage distribution between male and female worker as well as urban and rural areas, the results demonstrate that the wage distributions in 2002 and 2012 had both location shift and shape shift. This provided evidence that quantile regression is appropriate for the usage of quantile regression- based method in examining wage differentials in Vietnam economy. 3.2. METHODOLOGY This study estimates the following regression: 1 2 3 4 5 6 6 5 1 1 3 5 1 1 lnWage + + _ + Region i i i i i i i i i i i i Married Male Urban Experience Experience sq Degree Occupation Type u                           .i Schooling is defined into seven categories: no schooling (base category), completed primary, completed secondary, completed high school, vocational, college and postgraduate. Dummy variables for occupations, marital status, regions, and ethnicity are also included as control variables. First, this equation was estimated throughout the wage distribution using all observations in the sample to obtain the overall wage regression. After that, it was estimated again over male/female and urban/rural groups. In order to acquire the unbiased and consistent 9 estimators, this study applied the two stage quantile regression in combination with sample selection bias correction. In addition, this study decomposes the wage differentials between male/female, urban/rural and 2002/2012 by using the method of Machado - Mata (2005). CHAPTER 4 RESULTS AND DISCUSSION 4.1. The estimated wage equations in Vietnam The estimated wage equations across the 0.1 – 0.25 – 0.5 – 0.75 – 0.9 quantiles in Vietnam are briefly reported in Table B.2 and Table B.4 along with 2SLS estimation. As we can see, most of the coefficient estimates are statistically significant. The estimates of return to education are positive and increasing along with the qualification levels. This indicates generally that workers with higher qualifications would receive higher real hourly wage. Skilled workers who complete undergraduate or postgraduate course are expected to have substantially higher wage in comparison with the others. Men and women’s wage equations This study conducts the analysis separately for men and women in the year of 2002 and 2012. An intuition of the results in 2012 can be seen from Table B.2, which demonstrates the differences in pattern of wage for the two groups of workers. In 2012, for the lower qualifications (such as primary, secondary, and high school) the 10 returns to women’s education are higher than men’s regression. However, for higher qualifications, the situation is quite opposite. Urban and rural wage equations The trend that higher qualifications higher returns still be stable in both urban and rural wage equations. The education returns in the urban area are higher than the rural area, especially at the bottom of the wage distribution. With workers who complete primary, secondary and high school in rural areas, the returns to education seem to decrease as quantiles increase. In contrary, in urban areas workers with higher qualifications have higher education returns at higher quantiles. On the other hand, there is no clear pattern in the estimation for other cases. Extraction of Table B.2: Wage equations for men and women on 2012 Men’s wage equation in 2012 Women’s wage equation in 2012 Independent variables 2SLS 2SQR 2SLS 2SQR 10% 25% 50% 75% 90% 10% 25% 50% 75% 90% Primary 0.0788*** 0.126** 0.0780** 0.0797*** 0.0116 0.0273 0.138*** 0.0948 0.141*** 0.166*** 0.155*** 0.0524 [2.690] [2.385] [1.963] [2.672] [0.338] [0.572] [3.631] [1.102] [3.128] [4.568] [3.871] [0.869] Secondary 0.121*** 0.169*** 0.132*** 0.107*** 0.0475 0.0845* 0.179*** 0.169* 0.194*** 0.183*** 0.175*** 0.122* [4.013] [3.099] [3.238] [3.488] [1.349] [1.719] [4.497] [1.878] [4.110] [4.800] [4.174] [1.925] High school 0.212*** 0.233*** 0.199*** 0.172*** 0.148*** 0.203*** 0.294*** 0.198* 0.242*** 0.268*** 0.259*** 0.310*** [5.884] [3.588] [4.072] [4.678] [3.519] [3.461] [6.285] [1.869] [4.373] [5.971] [5.257] [4.167] Vocational 0.306*** 0.275*** 0.233*** 0.251*** 0.283*** 0.404*** 0.288*** 0.218* 0.274*** 0.305*** 0.340*** 0.296*** [9.123] [4.533] [5.106] [7.340] [7.213] [7.375] [5.843] [1.949] [4.690] [6.449] [6.564] [3.782] Colleges 0.636*** 0.580*** 0.542*** 0.530*** 0.562*** 0.700*** 0.532*** 0.476*** 0.537*** 0.511*** 0.547*** 0.576*** [15.590] [7.862] [9.785] [12.748] [11.776] [10.513] [9.823] [3.878] [8.365] [9.836] [9.593] [6.680] Postgraduate 1.047*** 0.934*** 0.969*** 0.925*** 1.066*** 1.193*** 0.778*** 0.888*** 0.816*** 0.757*** 0.735*** 0.649*** [12.302] [6.074] [8.384] [10.661] [10.705] [8.589] [7.424] [3.733] [6.564] [7.519] [6.663] [3.889] Control variables Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes t-stat in brackets *, **, *** : significant at 10%, 5%, 1% Source : Author’s calculations 12 Extraction of Table B.4: Wage equations in the urban and rural areas in 2012 Urban wage equation in 2012 Rural wage equation in 2012 Components 2SLS 2SQR 2SLS 2SQR 10% 25% 50% 75% 90% 10% 25% 50% 75% 90% Primary 0.000577 -0.0693 0.0173 0.0761 0.000257 -0.0479 0.148*** 0.176*** 0.183*** 0.143*** 0.102*** 0.0604 [0.012] [-0.773] [0.280] [1.431] [0.004] [-0.506] [5.585] [3.113] [5.242] [5.535] [3.558] [1.544] Secondary 0.0556 0.0402 0.0755 0.116** 0.0509 0.0308 0.190*** 0.294*** 0.227*** 0.153*** 0.148*** 0.101** [1.125] [0.440] [1.196] [2.132] [0.805] [0.318] [6.878] [4.972] [6.236] [5.665] [4.953] [2.467] High school 0.176*** 0.0512 0.159** 0.218*** 0.137** 0.235** 0.301*** 0.343*** 0.290*** 0.217*** 0.245*** 0.279*** [3.317] [0.521] [2.345] [3.735] [2.017] [2.264] [8.732] [4.656] [6.394] [6.462] [6.568] [5.495] Vocational 0.242*** 0.0799 0.153** 0.269*** 0.328*** 0.394*** 0.345*** 0.331*** 0.326*** 0.282*** 0.313*** 0.355*** [4.636] [0.826] [2.297] [4.680] [4.908] [3.856] [10.249] [4.606] [7.347] [8.597] [8.605] [7.149] College 0.484*** 0.349*** 0.380*** 0.431*** 0.518*** 0.765*** 0.591*** 0.577*** 0.569*** 0.479*** 0.530*** 0.574*** [8.419] [3.278] [5.162] [6.831] [7.033] [6.795] [14.053] [6.416] [10.271] [11.687] [11.650] [9.257] Postgraduate 0.766*** 0.736*** 0.656*** 0.686*** 0.851*** 0.994*** [8.911] [4.622] [5.974] [7.265] [7.729] [5.909] Control variables yes yes yes yes yes yes yes yes yes yes yes yes t-stat in brackets; *, **, *** : significant at 10%, 5%, 1% Source : Author’s calculations 4.2. Decomposition results Now we turn to the Machado – Mata technique to decompose the urban/rural wage gap across quantiles into two components – one due to urban – rural differences in the distributions of covariates and the other due to urban-rural differences in the distributions of returns to those covariates. The decomposition of the gender wage gap based on Machado – Mata method is reported in Table C.1. As we can see from Table C.1, real hourly wages seem to be always greater for men than for women at all considered quantiles. This wage gap is smaller at higher wage. The largest gap is found at quantile 0.1. The gender wage differential declines over the time. However, in each year, using the male wage structure as a reference, the gender wage gap is totally due to the differences in returns, which are traditionally interpreted as discrimination. Table C.1 Decomposition of gender wage differential Components All sample By areas 2002 2012 In urban In rural 2002 2012 2002 2012 Quantile 0.1 Raw differential 0.2947*** 0.2173*** 0.1760*** 0.1516*** 0.3941*** 0.2854*** [18.04] [17.48] [8.87] [7.45] [22.98] [13.44] Due to endowments -0.0858** -0.070*** -0.0348 -0.0503** -0.071*** -0.061** [-3.16] [-2.92] [-1.46] [-1.52] [-2.88] [-1.16] Due to returns 0.3805*** 0.287*** 0.2109*** 0.2046*** 0.4655*** 0.3465*** [14.22] [21.81] [8.83] [8.61] [20.82] [12.77] Quantile 0.25 Raw differential 0.2306*** 0.1690*** 0.1595*** 0.1589*** 0.3312*** 0.2071*** [30.73] [19.89] [11.65] [9.29] [29.21] [18.20] Due to endowments -0.075*** -0.076*** -0.046*** -0.051*** -0.064*** -0.068*** [-5.51] [-5.11] [-2.77] [-2.44] [-4.91] [-2.75] Due to returns 0.3059*** 0.2453*** 0.2055*** 0.2101*** 0.3957*** 0.2755*** 14 [23.49] [33.98] [11.65] [12.19] [29.70] [16.96] Quantile 0.5 Raw differential 0.1569*** 0.121*** 0.1565*** 0.1477*** 0.2167*** 0.1471*** [30.37] [15.70] [14.18] [8.10] [35.87] [19.48] Due to endowments -0.084*** -0.085*** -0.073*** -0.033* -0.053*** -0.063*** [-8.000] [-5.81] [-4.47] [-1.70] [-6.27] [-4.01] Due to returns 0.2416*** 0.207*** 0.2295*** 0.1813*** 0.2702*** 0.2106*** [22.81] [23.65] [15.58] [9.57] [35.91] [14.34] Quantile 0.75 Raw differential 0.0912*** 0.086*** 0.1590*** 0.1413*** 0.1287*** 0.1076*** [17.30] [9.10] [11.30] [7.05] [16.75] [9.46] Due to endowments -0.119*** -0.098*** -0.071*** -0.004 -0.067*** -0.095*** [-9.36] [-5.73] [-4.07] [-0.23] [-6.32] [-4.28] D