State and non-state wage differentials in VN

There are a difference between the state sector and non-state sector in Vietnam in wage structure and employment practices.. The state sector has a surplus of labor and has constraints t

THE ECONOMICS UNIVERSITY INSTITUTE OF SOCIAL STUDIES

VIETNAM- THE NETHERLANDS PROJECT ON DEVELOPMENT ECONOMICS

STATE AND NON-STATE WAGE DIFFERENTIALS IN VIETNA.M

Supervisor NGUYEN VAN NGAI

Hochiminh City, December 2003

I also want to show my gratitude to all lecturers for providing me academic knowledge background and to the project staff, Ms Dinh Thi Anh Nguyet and Ms Dang Kim Chi, for their help during the courses

I also want to show my gratitude to all lecturers for providing me academic knowledge background and to the project staff, Ms Dinh Thi Anh Nguyet and Ms Dang Kim Chi, for their help during the courses

ABBREVIATIONS

ABSTRACT

There is no evidence on the extent of wage differentials between state and non-state sector in Vietnam The main objection of this research is to examine the wage differentials between SOE versus DPE and SOE versus FIE, and to find source of the differentials For those purpose, corrected Mincer earnings equations of sectors, which are the Mincer earnings equations corrected selection bias by Heckman Two-Step method, must be compared by using the Oaxaca-Blinder Decomposition method The data sample of VLSS 97-98 is applied for estimation process The results indicated that there is sector discrimination in wage against state sector, and the difference in wage settings the source of the sector discrimination Besides, there was the hesitation of non-state participation and gender discrimination in both sectors

CONTENTS

CER TIFI CATI 0 N Cio···v··· i

A CKNOWLE.DGEMENT oooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooooqooooooooooooooooooooooooo ii ABBREJ!'l"ATIONS oooooooooooooooooooooooooooooooooooooooooooooooooooooooooooo;,ooooooooooooooooooooollooooooetoooooooooooooooooo iii ABS TRA CT oooe OOOitO • • • • • • • • • • • • • • • • • • It • • • • • • • • 0 0 0 0 0 0 0 0 0 • • 0 0 0 0 • • • • • 0 • • • 0 0 • • • • • • • • • • • • • • • ooo 0 • • • • • • • • • • • • • • • 0 0 0 • • • iv CONTENTS oooooooooooooooooooooooooooooooooooooooo•oooooooooooooooooooooooooooooooooooooooooooooooooooooooctooooooooooooo••••••••••• V LIST OF TABLES ••.••.••.• ••.•.• • •• • • • eo •••••••••••••••••••••••••••••••••••••••••••••••••••• vi LIST OF FIGURES ···eo···o···.,··· vi

CHAPTER 1 :INTRODUCTION • • • o ••••••••••• e •••••••••••••••••••• e 1

1.1 Problem statement and the objective of the research ~ 1

1.2 Research question 2

1.3 Methodology and dataset 3

1.4 Structure of the thesis 3

CHAPTER 2: LITERATURE & EMPIRICAL REVIEW • 5

2.1 Definitions 5

2.2 Theoretical Review 7

2.3 Empirical studies 18

2.4 Vietnam labor market overview: 21

CHAPTER 3: MODEL SPECIFICATION AND DATABASE 24

3.1 Corrected Wages Model (CWE) 24

3.2 Likelihood Selection Model (LSE): 27

3.3 Calculation ofthe inverse Mill's ratio (IMR): 31

- · 3.4 Database 31

CHAPTER 4: STATE AND NON-STATE WAGE DIFFERENTIALS IN WETNAM 34

4.1 Descriptive Analysis 34

4.2 Econometric Analysis 57

4.3 Comparison with other researches 79

CHAPTER 5: CONCLUSIONS AND POLICY IMPLICATION 82

5 1 Conclusion 82

5.2 Policy implication ~ 84

5.3 Further researches 85

APPENDIX • , • • • • • 8 7

BIBLIOGRA.PHY •• •• o • • • • • • • • 94

LIST OF TABLES

Table 2.1: Identification of the earnings and sector choice equations 20

Table 3.1 Specification of the corrected wages model 27

Table 3.2 Specification of the Likelihood Selection Model (LSE) 30

Table 4.1 Types of data and levels ofmeasurement 35

Table 4.2: Descriptive statistics 36

Table 4.2: Descriptive statistics (continue) 37

Table 4.3 Statistic Analysis of Wage 39

Table 4.4 Statistic Analysis of Age 40

Table 4.5 Statistic Analysis of Experience 41

Table 4.6 Education and wage rate : 42

Table 4 7 ANO VA results for wage and education relationship 44

Table 4.8 Experience and Wage Rate 44

Table 4.9 ANOVA results for wage and experience relationship 46

Table 4.1 0 Gender and wage rate 4 6 Table 4.11 t-testfor wage difference between gender group 47

Table 4.12 Location and Wage Rate 48

Table 4.13 t-testfor wage difference between residua/location groups 49

Table 4.14 Regions and Wage Rate 4 9 Table 4.15 ANO VA results for wage and regions relationship 51

Table 4.16 Age and Wage Rate 51

Table 4.17 ANOVA results for wage and age relationship 52

Table 4.18.a Pearson correlation coefficients among variables for SOE 54

Table 4.18 b Pearson correlation coefficients among variables for DP E 55

Table 4.18.c Pearson correlation coefficient!/among variables for FIE 56

Table 4.19.a: Heckman model for the logarithm of hourly wage ofSOE, DPE, and FIE in 1998 57

Table 4.19.b: Heckman mode/for the logarithm of hourly wage ofSOE, DPE, and FIE in 1998 (cont) 58 Table 4.20 Wald test of functional specification 59

T~ble 4.20.1 The significance and sign of difference in size of the coefficients 59

Table 4.21: The likelihood ratio test for the likelihood selection models 61

Table 4.22: Expected Log Hourly Wages by Sector of Employment, Vietnam 1997-98 67

Table 4.23: Decomposition of state and non-state wage differentials in Vietnam 1997-98 73 Table 4.23: Decomposition of state and non-state wage differentials Vietnam 1997-98 (cant) 74 Table 6.1 Pooling Regression 93

LIST OF FIGURES

Figure 2.1 Employment choice 8

Figure 4.1 Education and Wage rate 44

Figure 4.2 Years of Experience and Log wage rate 46

Figure 4.3 Gender and Wage rate 48

Figure 4.4 Wage rate and Location groups 49

Figure 4.5 Wage differentials between regions 51

Figure 4.6 The relationship between wage and age 53

Figure 4 7 Effects of education attainments on (log) hourly wages 69

Figure 4.8 Gender wage differentials among sectors 71

Figure 4.9 Effects of seniority on (log) hourly wages 71 Figure 4.1 0 Effects of location to expected wages : 7 2 Figure 4.11 Effects of regions to expected wages 7 3

CHAPTER l:INTRODUCTION 1.1 Problem statement and the objective of the research

Since 1986, Vietnam has adopted the doi moi (renovation) policy to transform from the

centrally planned economy to a socialist oriented market economy The reform involved expanding the private sector and opening up the economy These changes have contributed impressive results to economic growth During the period 1993-1997, GDP was increased at a rate of approximately 9% per year, of which the share of the non-state sector in GDP was about 60 per cent (Belser 2000 and GSO 2000)

Matters exist where "over-sized" state-owned enterprises (SOE) have become a significant burden to the government budget SOE showed a decreasing trend contribution to the growth of GDP from 11.4 percent to 9.7 percent in 1996 and 1997 respectively but wages and expenditure increased at the rate of 6.2 percent and 6.9 percent At the same time the number of SOE wage employment increased from 15.48 percent to 16.45 percent in the period of 1992-93 to 1997-98 (World Bank 1999 and

Bales 2000)

There are a difference between the state sector and non-state sector in Vietnam in wage structure and employment practices A non-market process sets wages for SOE A base salary for each education 1 evel i s s et and incremented every three years according to seniority There are differentials according to the position occupied Performance seems irrelevant iii promotions SOE employees have lifetime labor contracts and almost 100 percent are union members; however, they are not allowed collective bargaining or the rjght to strike Conversely in the non-state sector, which comprises domestic private enterprises (DPE) and foreign investment enterprises (FIE), wages are set according to the position occupied and collective bargaining Performance is important in the

consideration of promotion Employees in the non-state sector have an open-ended labor contract, which can be canceled following a notice period

There were a number of studies into the wage differentials between public and private sector (Gunderson 1979, Gyourko and Tracy 1988, Hartog and Oosterbeck 1993, Terrell

1993, Bedi 1998, Tansel 1999, and Zhao 2001) The public sector typically comprises government administration and SOE, and the private sector covers those employees included in social security and health schemes (Tansel 1999 and Zhao 2001) But such study has not been found in Vietnam The goal of such study is to investigate the wage differentials between the state sector and the non-state sector in Vietnam The state sector has a surplus of labor and has constraints to voluntarily lay-off workers and the number of wage employment in the non-state sector showed a decreasing trend These indicators may lead to the conclusion that the wages of workers in SOE are higher than those of DPE and of FIE The results of this research will provide policymakers with a clearer idea of whether or not the state worker earned the premium wage, and where are sources of the wage differentials between state and non-state sector From those findings, suitable policies could be applied to adjust the SOE wage settings and employment practices for increasing SOE performance and decreasing SOE fiscal problems

1.3 Methodology and dataset

The public sector is the state sector with government administration and the private sector is non-state sector Government administrative organizations were excluded from the research because their objective is broader than profit maximization of an enterprise (Borjas 1996) The comparison group for research is based on an enterprise category

In order to answer the first question, the method of comparing sectoral Mincer earnings equations is applied However? the estimates of these Mincer earnings equations are inconsistent because of selection bias so that the Heckman Two-Step Method must be applied to specify the corrected Mincer earnings equations The Oaxaca-Blinder Decomposition method will be used to answer the second question The Oaxaca-Blinder Decomposition method will decompose the net wage differentials into four components included constant terms, endowments, coefficients, and selection bias Econometric analysis helps to find the effects of various determinants on wage Descriptive statistics are also used to clarify the content of the dataset Regression is different to descriptive statistics

The data used for the study were derived from the Vietnam Living Standard Survey in 1997/1998 (VLSS 97-98) This research uses cross section data of interviewees of the survey Because the research focuses on the labor market, the data sample will be

restricted to individuals aged 15-65 ~·

"'",;"'

1.4 Structure of the thesis

The research is organized as follows

Chapter 1 provides basic information and a background to the research

Chapter 2 will provide the I iterature on i ndividuall abor supply, human capital, labor market discrimination, institutional approach to labor market, the Heckman's Two-Step Method, and the Oaxaca-Blinder Decomposition Method, which theoretically supports the answer to the question in literature section An empirical studies' review follows Lastly the current situation of wage employment and wage differentials among sectors

in Vietnam will be demonstrated

Chapter 3 will specify the model and database used for this study

Chapter 4 will present descriptive statistics oft he dataset, the estimation models and their results The latter section will focus on the results of the analysis in comparison with previous studies

Chapter 5 draws some conclusions and provides for policy implementation

CHAPTER 2: LITERATURE & EMPIRICAL REVIEW

The chapter first provides some definitions related to this study The theoretical review then investigate the choice between two sectors as these choices relate to the wage differentials among employees, and the source of these wage differentials The chapter continues with empirical studies These will involve research of public and private wage differentials in Turkey and China whose methodology approximates this study Finally, the chapter will review the Vietnam labor market and the differences between the state and non- state sectors

2.1 Definitions

The worker's earnings are the sum of wages, bonus, and other compensations in kind collected per working hour There is no distinction between earnings and wages

The state sector is defined as the state-owned enterprises (SOE) and the non-state sector

as the foreign investment enterprises (FIE) and the domestic private enterprises (DPE)

The FIE include 100 percent foreign owned enterprises and joint venture enterprises The DPE include cooperatives, private enterprises, small household enterprises, and mixed enterprises (GSO 2000)

Mincer's Earning Equation implies wage determinants under age-earnmgs profiles while the corrected wage equations (CWE) are considered as the Mincer's earning equation under corrected selection bias

The likelihood selection equation (LSE) is the likelihood equation of multinomial sector choice probit equation, which explains the probability of sector participation of an individual

The Ordinary Least Squares method (OLS) is the method to construct the sample regression function that is basis to estimate the population regression function (Gujarati 1995)

Selection bias in the thesis is t~e problem in OLS estimation of a Mincer earnings equation The problem emanates from two sources Firstly, there is self-selection among individuals when they participate in the labor market and are employed into sectors Sample data collected is not stochastic Secondly, wage data is not available for non-workers or of wage reservation hence the Mincer's earnings equation becomes a censored regression model The bias leads the OLS estimation of the Mincer earnings equation to provide inconsistent co-efficient estimates

Inverse MilT's Ratio (IMR) or selection term, which is calculated from the estimated probability value of the LSE, is used for correcting inconsistent estimates of parameters

in OLS regression of the Mincer earnings equation in the Heckman Two-Step Method The IMR represents the probability of sector participation in the sectoral CWE (Pindyck and Rubinfeld 1998)

There is a distinction between Maximum-Likelihood Approach (MLA) and Maximum Likelihood Estimation (MLE) MLA is the application of the maximum-likelihood approach to estimate linear models and non-linear models (Pindyck and Rubinfeld t"998) MLE is an alternative method of selection bias correction (Brendt 1991 )

2.2 Theoretical Review

The theoretica~ review will investigate the theory of individual labor supply to explain the decision of whether to work or not and in which sector to participate, the theory of traditional ·human capital and labor market discrimination describing why there are wage differentials among workers

2.2.1 The theory of individual labor supply

The theory of individual labor supply explains how an individual selects employment

In essence, the theory of individual labor supply is an application of the theory of consumer behavior It is assumed that an individual who is constrained by time and budget can afford consumer goods An individual should allocate a number of hours available between leisure activities and the labor market, and have available non-labor income (property income, lottery prizes, and remittances) and 1 abor income (equal to hourly wage rate and total working hours) (Berndt 1991, and Borjas 1996) Income can

be increased by sacrificing hours of leisure for additional hours of work The budget line presents the boundary of individual opportunity set between hours of leisure and income (or consumption) EL, EM, and EH respectively in Figure 2.1 illustrate budget lines at each hourly wage rate OV describes non-labor income, OT is available total number of hours, and hours wage rates included W reservation, W low, W high are slopes of budget lines EL, EM, EH respectively It is also possible to choose a combination of consumption and hours of leisure that maximizes utility The indifference curve represents a particular level of individual satisfaction or an indifferent choice between consumption and hours of leisure The UO, U1, U2 curve in Figure 2.1 illustrates levels

of individual utility or satisfaction The U2 curve gives the highest level of individual utility, and U0 the lowest level

Figure 2.1 employment choices

Source: To work or not to work (Borjas 1996 p.31) and Consumer Behavior (Pindyck and

Rubinfeld 1998 p.57-78)

Hours of Leisure

Point E where the indifference curve U0 touches the budget line EL with hours wage

rate of Wreservation is the point at which an individual decides to work or not to work Here the hours wage rate Wreservation is called the reservation wage An individual will decide to enter the labor market when the offered wage is higher than the reservation wage The reservation wage is affected by factors including hours ofleisure, non-labor income, and the prices of consumer goods (Berndt 1991, and Borjas 1996) Similarly,

with the hours wage rate W high higher than W1ow, a worker who has a budget line EH reaches the indifference curve U2 higher than U 1 of another worker who has a budget line EM The worker will decide to participate in a particular sector whenever they perceive the wage in this sector is higher than the wage in the other sectors, ceteris

paribus Ceteris paribus is the notion of exogenous factors However, workers are different so the reservation wage will vary from worker to the worker (Borjas 1996)

2.2.2 The theory of neoclassical human capital

The neoclassical human capital theory best explains wage determinants among workers because the theory explains why there should be wage differences for different workers

in a perfect labor market (Fine 1998) Accordingly, workers vary in ability and acquired skills or human capital Human capital can be gained through education, training, and work experience (Borjas 1996, and Fine 1998) The theory assumes that the individual present value of age-earnings profiles is

The return to schooling

where NPV is Net Present Value of individual age-earnings, n is lifetime and r is the rate of discount A student must pay for books, fees, and so on (C) throughout years of schooling (i) to receive a future wage (W) as the return for schooling throughout a working life (n-i) The model not only explains how an individual decides how much schooling to acquire but also indicates the wage gap among workers because of the education difference Accordingly, a person self-estimates a rate of discount depended

on the probability of future rewards or on the return for schooling, therefore the future wage (W) among workers varies according to educational achievements, ceteris paribus Even if the rate of discount is similar, there are differences in individual abilities hence

the future wage (W) and the wage-schooling locus, which is the salary that employers are willing to pay for every level of schooling, differs from worker to worker Similarly, the theory considers the on-the-job training during working time as human capital

investment The employer when hiring a worker must pay a salary (W) and implicated on-job-training fees (G) so the employer will pay a salary (W2 = W +G) in the next year: for the accumulated experience of the worker

In many empirical studies, the Mincer's earmngs equation 1s considered as the implications of the human capital model for the age-earnings profile to explain wage determinants of workers (Mincer 1974, Berndt 1991, Bmjas 1996) This expressed as follows:

logW = w(EDU, EXP, EXP2, AGE, OTHER) (2.1) where W is earnings, EDU is education, EXP is experience, EXP2 is a quadratic on EXP, AGE is age of employees, OTHER are other factors which affect wage determinants (Borjas 1996) The logarithm of wage is used instead of the actual wage because logarithms can be used to compare positive numbers Consider W1 and W2 as the wage oft he worker 1 and 2 respectively, and when the wage difference between workers is small, the result is:

After EDU years of schooling, incremental earnings are,

(2.5)

If the rate of the return is the same for all levels of schooling, r1 = r2 = r3 = = rx and (1

+ r) approximate eR as well as the multiplicative disturbance term eu appended, the equation (2.5) will become

The on-job-training is measured by the working time devoted to training investment, which is calculated by a multiplication between the proportion of working time saved for on-job-training (k) and years of working experience (EXP) Unfortunately, data on k

is missing data in all data sources Further, tJ;le neoclassical theory of human capital suggested that worker's earnings curve follows parabolic shape over working time because worker acquires more human capital when young and less that when older (Borjas 1996) This led Mincer to use two variables for years of working experience (EXP) and quadratic on EXP (EXP2) for measuring the value of on-job-training human capital The age of worker (AGE) should be added into the wage equation because of the effect of overtaking age (Borjas 1996) One important assumption of the neoclassical paradigm is a perfect competitive market Notwithstanding this assumption the worker generally meets an imperfect labor market As a result Mincer's earnings equation requires additional variables (OTHER)

2.2.3 The theory of labor market discrimination

In the imperfect labor market, the worker is not only differs from other workers in

human capital but also is discriminated by his individual characteristics The theory of labor market discrimination demonstrated that wage differentials among workers emanated from taste discrimination including race and gender (Borjas 1996) For instance, a white employee may not like to work for a black employer and with black

colleagues An employer could take into account the race and the gender of the applicant, and the consumer may discriminate about the gender and race of the seller The employer could be willing to pay a wage (W) for a worker but could vary the wage for a black worker or for a female worker to W 1 x ( 1 +d), where d is the discrimination parameter So the recruitment decision is based on a comparison of W and W 1 x ( 1 +d), employer hire only for black workers or for females workers if W 1 x ( 1 +d) < W and hire only white workers or male workers if W1 x (1 +d)> W The Mincer earnings equation then explains the wage differentials among workers should take into account gender (SEX) and race (RACE) variables as follows (Fine 1998):

2.2.4 Institutional approaches to the labor market

Wage determinants exist among workers, but decision of labor participation and of sector participation respectively based on the level of reservation wage and sectoral wage In answer to the question of what is the source of wage differentials between the state and the non-state sector, institutional approaches to the labor market explicitly explain the differentials as political (Fine 1998) Accordingly, the state sector is suspected to have a higher than competitive wage because the state sector is supported

by government If the non-state company pays more than market wage, they could reduce profits or get the loss These results will encourage the fim1s to no longer overpay While the bad financial situation resulting from this behavior in state firms is simply passed on to taxpayers Unfortunately, taxpayers have little opportunity to look after what their representatives are using the budgets from tax Moreover, the state has a political incentive to overpay employees The vote and the potential political force of state workers are reasons to overpay the state employees for their cooperation and political support The differentials could also emanate from institution arrangements in communist countries as reported by Borjas (Borjas 1996) In the communist countries, the political institution is built on single communist party The communist party gathers

their fundamental force from public worker so that public worker should earn an overpaid wage by political target The worker could perceive the sector difference from wage and employment settings as well as the non-pecuniary aspects (Tansel 1999 and Zhao2001)

2.2.5 Estimation Methods:

In practical terms, the state and the non-state wage differentials can b e c omputed b y

OLS regression in at least in 2 procedures One procedure uses a dummy sector variable

in a statistical Mincer's earnings function The other procedure aims to separate the

'

sample into state and non-state workers and to estimate separate Mincer earnings equations with both intercept and slope coefficients differing, then the wage differentials will be estimated throughout decomposed components of the wage equations by the Oaxaca-Blinder Decomposition method (Oaxaca 1973, Blinder 1974, Berndt 1991) However, workers who have either a very 1 ow market w age or a v ery high reservation wage will not work Furthermore workers may participate in the state sectorbecausethewage in SOE is higher than the other sources (Borjas· 1996) This means there is a self-selection among workers hence the sample of workers is not a random sample Besides, wage data is not available fort hen on-worker The missing data problem is covered as a dependent variable in Mincer earnings equation So, both

of these errors are called a sample-selection bias or self- selection bias (Heckman 1979) For example, the motivation of Mincer earnings equation (2 7) comes from the fact that

·it observes earnings for individuals who are working but does not observe earnings for individuals who are not working The equation (2.7) can be rewritten as

log W = {JX + u (2.8)

where X is vector of explanatory variables ~ is a vector of unknown parameters of these variables and u is the error term A vector of estimated inconsistent parameters ~ will result if the estimate equation (2.8) includes observations where individuals have a wage However, individuals only participate in work whenever the wage of the work

(W) is higher than reservation wage (Wr)_ (Bmjas 1996) The reservation wage 1S assumed as

where Xr is a vector of variables, which determine the participation decision of individuals as compared to X's vector determination W; y is a vector of parameters of Xr; and u' is an error term As a result of (W > Wr),

(2.10.a) or

j3X- rXr + v > 0 (2.10.b)

where v = (u - u,) The cov(u, u,) will not equal zero in order for the result of ~X with the sample of employees to lead to biased estimates of ~ for whole population Thus, sector choice shoul~ be employed in the research of wage differentials

There are two methods that can be applied to correct the selection bias of the Mincer earnings equation including Heckman's Two-Step Method and Maximum Likelihood

Estim~tion Methpd (MLE) (Pindyck and R1,1bi~f~ld 1998, Yun 1999)

2.2.5.1 The Heckman's Two-Step Estimation Method

In the Heckman's Two-Step Method, an individual decide to work if their expected wage is higher than their reservation wage, and a worker decide to join a sector by perceiving the net c}iffer~ntials in wage and non-wage compensation in each of these sectors The taste ~nd preference of employees, human capital, other characteristics

~ontribute to determine the sector selectivity (Lewis 1996) Thus, the probability (Pj) that an individual choic.e in alternative j sector is assumed as:

(2.11)

wh~re n is a total of sector choices, Z is a vector of explanatory variables affecting sector choice, and a is a vector of unknown parameters of the alternative j Equation (2.11) is a sta~d1:1rd probit equation, and its parameters (a) can be estimated by maximizing the LSE Pj (or its logarithm) using the en~ire sample of workers and non-workers in the first step of the m¥thpd The estimates Pj parameters are then used to compute estimates of the IMR (A.) as follows:

Here, $ is the Standard Normal Density Function, and <I> is the Standard Nonnal Distribution Function In step 2 of the method, the estimated IMR (A.j) will be appended

as an additional regressor (called selection term) to the sectoral Mincer's model to obtain the sectoral CWE Thus, the model is developed as follows:

log~ -logWo = (f3ot- f3oo)+ Bo (:XI-Xo )+ (,81-f3o )XJ + (.n-1 ~ -7Z"oAo )(2.14.b) or logYJ?; -logf¥o =(Poi-f3oo)+0.5(f31 + f3o)('xi -xo)+o.5(f31- Po XXI +Xo)+(7r~~ -1l"oAu)(2.14.c)

where over-bar denotes the mean of variables The mean of log wage differentials is decomposed into the four components in the right of the equation The first component

is the difference in constant terms This is often interpreted as the premium from being

· in a given sector (Terrell 1993) The second component explains the difference in employees' endowments, and the third component is due to the market returns to the

endowments The final component reveals the difference in the selection bias (A= n.A) The first and the third components constitute the total structure differential that answers

to the first research question whether the state earns the premium Take an example of the equation (2.14.c); the total structure differential (Dg) can be re-written as follows:

The term ( Boi + ,81 (XI+ Xo)) is considered as the hypothetical earnings (in logarithm)

of workers in both sectors when they work with their own productivity but earn

according to the CWE of state sector While the term (Poo + flo(X1 + Xo)) is the hypothetical earnings (in logarithm) of all workers according to the CWE of non-state sector So the sector d iscrimiJ.?.ation is the total structure differentials ( Dg), which is computed by the average difference between the hypothetical earnings according to the

CWE of state sector and those of non-state sector

2o2.5.2 The Maximum Likelihood Estimation Method (MLE)

The MLE method aims to set up a Likelihood Tobit Function of sectors for the joint estimation of Mincer's earnings equation (2.7) and selection equation (2.11) as follows

where Pj = 1 if worker and Pj = 0 if not

ej is disturbance of selection equation (2.11)

The Oaxaca-Blinder decomposition method is modified as follows:

logw; -logW0 = (/J'01 - /3'00 )+ /3\ (X1 - X 0 )+ (fJ\-/3'0 )X 0 + (A1 - A0 ~r

logw; -logWo = (f3'oJ-f3'oo)+ f3'o (xl -Xo)+(f3\-f3'o)Xl +(AJ -Ao) (2.17)

Similarly, the total of structure differentials (Dg) explain the sector discrimination among state and non-state sector

2.3 Empirical studies

Since the methodology of public- private wage differentials studies differs in selection bias correction methods, it is important to present previous studies of both methods However, because the implementation of MLE method is complicated in comparison to Heckman's Two-Step Method using standard procedures (probit and OLS), it is best to introduce Tansel's, and Zhao's work as empirical studies applying the Heckman's Two-Step Method

2.3.1 The public-private employment choice wage differentials in Turkey

Tansel' work researched the public- private employment choice and the wage differentials in Turkey using the i ndividuall evel sample data oft he 1 994 Household Expenditure Survey The sample was restricted to individual of 15 to 65 years of age who were not employed in agriculture The public sector comprised the government administration and SOB, and the private sector is the covered private sector, whose worker is covered by the social and medical insurance Tansel assumed five mutually exclusive choices including those not working, the government administration, SOB, the

~overed private enterprises, and other employment Other characteristics of the research are presented in Table 2.1 The central questions ofTansel's study are "Do public sector employees earn a premium?" and "Are woman discriminated against in the public or in the private?" To answer these questions, he estimated the LSE for five choices and

CWE was estimated with the sample of workers separated into gender, and public and private sectors The Oaxaca-Blinder method decomposed and compared the CWE between public-and private sector The results concluded that the public sector workers earned a substantial premium and women were facing discrimination in the private sector

2.3.2 The state and non-state enterprises wage differentials in China

Zhao 2001's work on the other hand explained earnings differentials between state and non-state enterprises in China using the individual level data sample of the 1996 urban household survey The sample was restricted to individuals of 15 to 7 5 years of age State owned enterprises were SOE, and the non-state enterprises included the urban collective enterprises (UCE), the DPE, and the FIE He assumed that an individual faced four employment choices Other characteristics of the research are presented in Table 1 Zhao's central question was 'what incentives or disincentives are motivating or preventing people from voluntarily leaving the state sector for private sector employment?' To answer this question, he estimated the LSE for four choices and the CWE with the sample of workers divided into enterprises The Oaxaca-Blinder method decomposed and compared the CWE among enterprises The results of this estimation concluded that the research question was not fully answered but it is possible that there

is a higher total earnings in the non-state sector

Table 2.1: Identification of the earnings and sector choice equations

Earning ·Sector choice Earning Sector choice equation equation equation equation

!.Variable group included:

2.4 Vietnam labor market overview:

The theoretical and empirical reviews have just shown the fundamental theories, and methods to correct the selection bias The Vietnam labor market should be overviewed before the best method is specified

The Vietnam labor market is characterized by high rates of population, high growth of labor force and participation rates These were 1 7 percent per year on average of population growth between the 1989 and 1999, approximately 2.6 percent of annual labor force growth between 1992-93 and 1997-98 Male and female rates of participation were 83 percent and 80 percent respectively in 1997-98 (Bales 2000)

Poor p erformance and o ver~taffing are prominent characteristics of the Vietnam state sector This is of concern for government and economists At the end of 1997, weakening SOE had debts of about US$4.4 billion and a number of SOE had an average debt equal to nearly twice the value of state-capital (World Bank 1999) There is about

40 percent of SOE that could make profits, whereas 16 percents of SOE make loss (Belser and Rama 2001) The distribution of SOE wage employment has shown an increased trend of 15.48 percent and 16.45 percent in 1992-93 and 1997-98 respectively

Although SOE reform was implemented in 1995 by equitization and privatization, and

by restructure, there have been many constraints 1.7 millions employees in 5,740 owned enterprises which could be laid off during the reform process is not small number (Belser and Rama 2001) This could create a negative social impact, especially

state-~sVietnam has been facing a high rate of unemployment Moreover, SOE employees are part of the labor elite and represent one of the most important political constitutions

of the Vietnam government (Belser and Rama 2001 ) Hence wage setting and employment practice in SOE has resulted in privileges for SOE employees such as labor

contract guaranteed for 1 ife, wage paid according to seniority and subsidized housing (Moock at al 1998) It is difficult to persuade the SOE employees into voluntary retirement In the meantime, job security and other regulations that decrease employment flexibility tend to make the redeployment in SOE more difficult (Belser 2000)

There are differences in wage setting and employment practice between state sector and non-state sector In the state sector, market processes do not set wages A base salary for each education level is offered and incremented every three years according to seniority There are differentials according to position occupied and performance seems irrelevant

in promotions SOE employees have lifetime labor contracts and almost 100 percent unionization; however, workers are not allowed access to collective bargaining or the right to strike Wages in the private sector are set by the importance of the position occupied and collective bargaining Employees in private sector have open-ended labor contract, which can be canceled with a notice period

SOE, DPE, and FIE workers are subjected to 1995 Labor Code Employees are offered

an open-ended labor contract, which can be canceled with a notice period of one month's notice Compensation and a severance payment equivalent to one month's salary for each working year is payable on termination of employment Workers are covered by the Social Insurance Fund and the Health Insurance Fund The normal hours

of work are 48 hours per week The definition of state and non-state sector has been set

to maintain the comparability in the non-pecuniary aspect of these sectors' job, which is · retirement and health benefits (Belser 2000)

The chapter provided keywords using in the research, theoretical reviews and estimation procedure and method supporting the research, empirical studies, and Vietnam labor market situation People differ from each other's in human capital and in individual characteristics People are workers when they perceive the received wages higher than

the reservation wages as well as they are SOE worker if the net wage differentials between SOE versus other sectors are positive State sector differs from the non-state sector in wage setting and employment practice The estimation method, empirical studies, and Vietnam labor market situation are overviewed to indicate that the investigation is researchable and valuable There are available two alternative comparative procedures, two alternative methods to correct the selection bias in the Mincer earnings equation, and empirical studies implemented in countries, especially in developing countries (Turkey and China)

CHAPTER 3: MODEL SPECIFICATION AND DATABASE

The purpose of the thesis is to investigate wage differentials among the SOE, DPE, and FIE so that the sector discrimination should be measured The Chapter 2 has just recommended two procedures to measure the sector discrimination The second procedure, which estimate separately sectoral Mincer earnings equations and the Oaxaca-Blinder method will decompose the wage differentials among sectors later, should be applied because the answer to the second research question is found, and the method to correct the selection bias of the Mincer earnings equation is easily applied In practice, the Heckman's Two Step method to correct the selection bias is standard more than theM aximum Likelihood Estimation (MLE) one (Pindyck and Rubinfeld 1998), and the Oaxaca- Blinder decomposition method as equation ( 2.14.c) will average the weight of the sector discrimination index (see the equation 2.15 in Chapter 2) So, the methodology to investigate wage differentials among the SOE, DPE, and FIE should be the separate estimation of the corrected Mincer earnings equation (CWE) of ownership categories, and then the Oaxaca- Blinder Decomposition method will be applied to decompose the wage differentials between SOE versus DPE and FIE The methodology suggests that a LSE and the CWE should be constructed as well as the calculation of Inverse Mill's ratio (IMR) should be specified This chapter will specify the two models, their independent variables, expected sign of the parameters, and database description of variables Next the chapter will specify the calculation process of the

IMR

3.1 Corrected Wages Model (CWE)

The theory review in Chapter 2 did indicate the differentials in wage emanated from the differentials in human capital, individual characteristics among workers In which the human capital is represented by education and experience concepts, and the individual

characteristics are represented by age, gender concepts (race concept is excluded because Vietnamese race is similar) The empirical review in Chapter 2 did show the concern of economists in the residential location effect to wage differentials among workers The residential location effect is represented by location and region concepts The estimation method in Chapter 2 pointed out the sample- selection bias incurring in the Mincer wage equation so that the probability of sector participation concept should

be concerned in all Mincer wage equation Those classifications and concepts affected

to wage differentials basically built up the sectoral CWE and were summarized in Table 3.1

Log wages are explained by human capital characteristics, labor market discriminations, probability of sector participation, and residential location factors as follows (Fine, 1998):

log(wu) = /301 + f311AGEij + /321SEX u + f331EDU u + f341EXPu + /351 (EXPu f

+ f3 6 LOC + /37 .REG + 7r ./L + U

j I) J I} J I) 1./ (3 1) where Wi is wage rate of worker i for sector j

EDU: individual education

EXP: worker's experience

AGE: individual age

SEX: gender

LOC: urban location

REG: regional residence

'A : the selection term

U : disturbance term

p : a vector of unknown parameters with Po as intercept term

1t : an unknown parameter of selection term

J : stands for SOE, FIE, and DPE

As discussed in Chapter 2, log wage is a dependent variable because the difference in the log wage between sectors equals the percentage difference in the actual wage between sectors· (Borjas 1996)

Education attainment (EDU) in present of education concept that is ordinal variable for different education achievements As discussed in the theoretical review - Chapter 2, the concavity of the age-earnings profile leads to experience concept measured by an experience variable and a quadratic experience variable Years of experience (EXP) represents the experience variable and is computed by following equation (Mincer 1974):

Number 6 indicates the age of school entrance, and year of schooling is the number of years of schooling Squared years of experience variable ( EXP)2 imply the quadratic experience variable The expected sign of EDU and EXP is positive because of the positive human capital effect on the expected wages, while the expected sign of EXP2 is negative because age-earnings profiles are concave Those variables and their expected signs were justified in Borjas work (Borjas 1996) and Tansel work (Tansel 1999)

Age indicates seniority effect and is measured by years of age of an individual (AGE) with a positive expected sign Gender explains gender discrimination that i s a b inary variable (SEX), taking 1 for male and 0 for female, and will be a positive sign of parameter Those variables and their expected signs were justified in Borjas work (Borjas 1996) and Tansel work (Tansel1999)

J>:.- binary variable (LOC) indicates whether an individual lives in an urban area and a vector of dummy regional variables (REG) must be included to control the differentials

in the labor market opportunities The expected sign of LOC and REG parameter could

be positive or negative depending on the employment situation at locations and regions

Those variables were justified in Zhao work (Zhao 2000) and Tansel work (Tansel 1999)

The selection term (called IMR when is calculated from the estimation of the LSE) (A.) which computed by equation (3.5) represents the probability that a person is working in

a sector, and expects a positive or negative sign of the parameter of the level of sector participation If the real parameter of the selection term is a positive sign, it indicates that the worker chooses to work in the sector The variable and its expected sign were firstly justified in Heckman work (Heckman 1979)

Table 3.1 Specification of the corrected wages model

Classification Concepts Variables Justification Expected Sign Human Capital Education Education attainment Borjar 1996, Positive

expenence

characteristics Gender Dummy (male= I) Tansel1999 Positive

Residential location Location Dummy (urban= I) Tansel 1999, Negative

1, other regions = 0) positive Sample selection Probability of IMR (called selection Heckman Negative

participation

3.2 Likelihood Selection Model (LSE)

The LSE must be specified to compute the IMR (called selection term when appended into the sectoral SWE) so that the LSE is assumed a pro bit model for the probabilities

or

or

or

that individual choose alternative sectors (Tansel 1999) The individual participation decision depends upon identified differentials between labor wage and reservation wage

or on the perceived net differentials in the wage and non-pecuniary in each of these sectors The worker characteristics, non-labor income as well as human capital and other characteristics will determine the sector choice (Berndt 1991 ) Similar to the CWE, education and experience concepts will represent to human capital, age and gender will represent to individual characteristics, and residential location will be presented by location and region concepts Besides, individual non-wage income and household income concepts that represent the reservation wage to reveal the comparison between labor wage and non-wage sector, and among sectors (Tansel 1999) Those classification and concepts were summarized in Table 3.2

The LSE is built to indicate individuals who participated into sectors such as SOE, OPE, FIE, or not-wage sector If the assumption is made, individuals face four alternative exclusive choices: non-working or other employment (j=O), SOE employment (j= 1 ), DPE employment (j=2), and FIE employment (j=3), then the probability distribution is the equation (2.11) in Chapter 2 or the sectoral LSE will be as follows (Pindyck and Rubinfeld 1998):

E IJ =a0 } +IY-"'1] .AGP .l fj +a.SEY +IY-} "'"~} ""'3] .EDU I) +a4 J .EXE I) +a5 } .(EXE)I) 2 +a6 } .HHr "!J

+ a7 .UJC + a8 .LAN + a9 .LOC + a10 .REG + V: (3.3)

where j stands for sector choices (non-working or other employment, SOE, OPE, and FIE)

F : sector choice of individual i for sector j

EDU: individual education

EXP: individual experience

EXP2 squared experience

AGE: age of an individual i

SEX : gender of individual i

HHI : annually household income

UIC :·annually non-wage income

LAN: land property of household

LOC: urban location

REG: regional residence

a : a vector of unknown coefficients

Sector choice (Fij), which represents unobserved probability of sector choice of individual i, is a vector of dummy variables, taking 1 for a sector choice (SOE or DPE,

or FIE) and 0 for others choice as follows:

Fil = 1 for SOE choice

of individual characteristics and human capital parameters in LSE is the same as that in the CWE Those variables and expected signs were justified in Berndt work (Berndt

1991) and Tansel work (Tansel1999)

The annually non-wage income (UIC) variable will explain the individual non-wage income, and the household income will be measured by the areas of land (LAN) and

annually household income (HHI) They are expected to reduce the probability of the

labor force participation by raising the shadow value of a person's time in non-market activities and in self-employment (Schultz 1990) so that the expected sign of the parameters is negative Those variables and expected signs were justified in Schultz work (Schultz 1990) and Tansel work (Tansel 1999)

Similar to the CWE, the location and region concepts are also explained by LOC and REG variables but they represent the difference in prices of consumer goods in

· residential regions and locations affecting to the decision of participation and sector choice An ambiguous sign of the parameters is expected as a result Those variables and expected signs were justified in Tansel work (Tanse1 1999)

Table 3.2 Specification of the Likelihood Selection Model (LSE)

Classification Concepts Variables Justification Expected Sign Human Capital Education Education attainment Berndt 1991' Positive

effect Experience Years of experience Tansel 1999 Positive

expenence

characteristics Gender Dummy (male= 1) Tansel1999 Positive

Location and Location Dummy (urban= 1) Berndt 1991, Negative or

1, other regions= 0) positive Reservation wage Individual Annually unearned Schultz 1990, Negative

income Household Areas of land Schultz 1990, Negative income Annually household Tansel1999 Negative

income

3.3 Calculation of the inverse Mill's ratio (IMR):

Inverse Mill's ratio (A.) that used to correct the selection bias of Mincer earnings equations is the selection term variable whenever it is appended into the sectoral CWE (Lewis 1986) It presents the probability of sector participation and is calculated by the equation (2.12) as follows:

Au = ¢(H iJ )j<P(H iJ) where H iJ = cp-l (P!i) (3.5)

where j stands for SOE, DPE, and FIE sector

A is IMR (or selection term) of individual i in sector j

¢is the Standard Normal Density Function

<P is the Standard Normal Distribution Function

P is the probability of sector j participation of individual i, in relation to the

A

computed variable F ( F 1 ) based on estimated parameters of model (3 3) is identified as

follows:

Probability of DPE choice

Probability of FIE choice

These variables were used for empirical studies of the wage differentials between public and private sector, specifically in the work ofTansel's and Zhao's

3.4 Database

The data used in the equation was extracted from the Vietnam Living Standard Survey

in 199711998 (VLSS 97-98) The sample size ofVLSS 97-98 was set at 6000 household selected randomly from regions, cities and provinces throughout the country This equation uses cross section data of the interviewees of the survey Because the research focuses on the labor market, the data sample should be restricted to individuals aged 15-