1. Introduction

Vietnam is in the process of transition from a centrally planned to market-oriented economy. Prior to the reform, wages and labor allocation were heavily regulated by government. The reform has included some measures to liberalize labor market. Wage rates more closely reflect payments for human capital than in the past. For example, the patterns of higher education leading to higher wages are clearer, implying the more efficient operation of the labor market. In this process, changes in wage determinants and discrimination over time are of great concern to economists as it is an important basic labor market issue that deserves regular reassessment.

In the context of Vietnam, the light industry sector (vs. heavy industry) includes 18 industries such as foodstuffs, electronics, textile and garment, printing, etc. (Bales, 2000). During the last decade, the achievement of light industry has been remarkably impressive with the output growth rate of no less than 10% a year. Strikingly, its export represented 34.3% of total country export (GSO, 2000). It is argued that Vietnam export growth has been led by light industry. In terms of employment, it has absorbed the largest share of total industry (72%) (GSO, 2000). These numbers imply that the contribution of light industry to economic growth over the last decades is not deniable. Moreover, economic literature argues that  labor-intensive industry is Vietnam’s comparative advantage. In other words, it is a potential for light industry’s development in the near future under the impact of trade liberalization. However, the question as whether workers gain from its development is not answered clearly. Thus, one can wonder about monetary benefits of workers in the sector who are indeed contributors of light industry’s achivements. And can wage adequately compensate for human capital investments with the minimization of market distortion in such an industry sector? This raises the quest for study on wage determination (in light industry) in general and the extent of labor market discrimination in specific.

Among 18 industries of the sector, textile and garment (T&G) industry sector has been recognized as the key player. Notably, its remarkable growth rate of export, 24.9% on average, has placed it at the second largest foreign exchange earner in the economy and the largest in the sector (Pham, 2000). T&G represented 14.5% of the country export and no less than 44% of light industry export (GSO, 2000). Also, it absorbed the largest employment share of the sector (25%) (Bui, 2001). The remarkably high growth of the T&G sector urges us to investigate its compensation for workers. It is of interest to address whether the achievement of labor-intensive export-oriented industries as T&G are beneficial to workers in terms of wages.

Unfortunately, such contemporary issues involving the light industry sector are not fully understood with convincible findings. Without understanding of human capital compensation in the high growth industry, no one knows the social impacts behind its success. This problem is more exacerbated if one takes promising development perspective of the light industry sector into consideration.

Concerning these issues, this thesis is to provide a comprehensive answer to wage determination and discrimination in light industry. It deals with the central research question: “WHAT ARE WAGE DETERMINANTS AND DISCRMIMINATION IN THE LIGHT INDUSTRY SECTOR?” This thesis will put emphasis on: (i) investigating what are determinants of workers in light industry in general and in the T&G in particular, (ii) investigating whether possible wage discrimination exists, (iii) finding out the source of wage discrimination, (iv) recommending some policy implication.

2. Methodology

As the extension of marginal productivity theory, human capital theory attempts to explore wage determination based on labor productivity. In addition, it provides insights into labor supply. It is considered as the currently “dominant economic theory of wage determination”(Berndt 1991, p152). It has its root in A. Smith’ s economics writing on equalizing differences in “workplace amenities and disamenities” (Berndt 1991, p.152). The modern rendition of human capital theory has blossomed especially from 1970s with the writings of Jacob Mincer, Theodore Schultz and Gary Becker. It extends economic theory with emphasis on empirical phenomenon by offering the notion of human capital and its effect on wages. Specifically, its purpose is to explain why different workers get different wages, even in the competitive market by linking directly human capital investment with wages.

            Human capital is defined as the “the accumulation of prior investments in education, on-the-job training, health, and other factors that increase productivity” (McConnell et al 2000, p.614). Investments in human capital can be considered as any kind of on-the-job training, schooling, other knowledge (such as information), health, morale and migration (Becker, 1993). It is noted that human capital is invested today to raise workers’ productivity, and then wages in the future. Differently, more human capital will increase marginal product of a worker. Among these investments, on-the-job training, and schooling should be emphasized due to their importance and their practical applications in empirical analysis.

Human capital theory with emphasis on empirical phenomenon provides a good quantitative framework for analyzing wage. Then, much of econometric literature has been learned from human capital framework to estimate wage determinants and discrimination.

Quantitative analysis based on human capital theory will be taken as follows:

2. 1. The Mincerian Wage Equation with Correction for Selection Bias

To estimate wage determinants, we apply the Mincerian wage equation as follows:

                                            (1)

where ln y_i is natural log of wages for the ith individual, X_i is a vector of observed socio-economic characteristics, b₀ is an intercept term, b is a vector of coefficients, and u_i is random disturbance term that reflects unobserved ability characteristics and the inherent randomness of earnings statistics. This equation is sometimes termed as a statistical earnings function.

Furthermore,  an empirical analysis of wage determination should take selection issue into account. Selection bias is considered here as in broad sense that arises from two different sources. Firstly, it is derived from using non-randomly selected sample. Secondly, it might due to self-selection by individuals being investigated (Heckman, 1977). While the non-randomness of sample is not a problem in practice, the self-selection into different sectors needs to be taken into consideration.[1] To account for the potential problem, we employ the Hay’s approach that is a generalization of Heckman Selection Model. Two stages introduced in this approach are as follows:

First, we estimate the parameters of the participation equation by multinomial logit, using pooled sample. In this stage the predicted probability P_jj (probability of  ith individual is in sector j) from multinomial logit model is used for constructing the correction term l_ij

l_{ij = (-1)}J+1

In the second stage, the estimated  l_ij is included among other explanatory variables of the Mincerian equation (1). Then, it is argued that the wage equation augmented by correction term can produce unbiased OLS.

            In conclusion, two-step procedure is as follows. First, we estimate the parameters of the participation in a particular equation by multinomial logit, using pooled sample. We can calculate the selection bias correction term, l_ij. Second, we construct l_ij into (1) and estimate (1) over observed workers by OLS.

2.2. Measuring Wage Discrimination – Decomposition Analysis

A vast body of literature has been developed to address how to decompose variations in workers’ wages into differences in productive endowments and discrimination. There are two separate methods of decomposition. First, some economists such as A. Blinder, R. Oaxaca, M. Ransom, and Neumark emphasize on the index number problem. Second, other economists, for example R. Brown, M. Moon and B. Zoloth develop ways of decomposing along sectoral lines (Appleton et al 1999, p.2). The former is preferred here because this thesis analyses wage determination and discrimination in one industry sector. Moreover, it has become a standard method, widely applying in empirical studies.

            The method for decomposing developed by Oaxaca and Ransom (1994) has based on the assumption that labor marker discrimination not only underpays the wages of the “minority” group (i.e for disadvantaged group) but also overpay for others (i.e. for advantaged group)[2].

To investigate wage differential between two groups, one first regresses each wage equation (1) for each group separately. After obtaining OLS results, the total mean log wage differential will be decomposed into following components:

                        (2)

where  is mean log wage differential, b* is estimated nondiscriminatory wage structure. Neumark(1998) proposed that b* is calculated as:

where W is a weighting matrix. W is given by:

where X is the observation matrix for the pooled sample and X_a is the corresponding matrix for the advantaged only. More details on this decomposition method are presented in Appendix.

In case that wage equation is augmented by correction term and has constant term, the detail decomposition is:

            (3)

where the first term is interpreted as environment gap, which is attributable to nature of the labor market. It is also termed as pure rent or pure premium from being in a particular sector. The second component is attributable to overpayment of advantaged group (also termed as nepotism). The third refers to underpayment of disadvantaged group, also termed as pure discrimination. The positive signs of second as well as the third terms refers to discriminatory behaviors of employers. The fourth reflects productivity differential. A total of the first three terms is termed as the unexplained gap or discrimination.

2.3. Technical Notes

Based on analytical framework, some technical aspects of wage determination analysis should be drawn as follows:

            Firstly, despite the fact that wage rates, earnings and compensation are quite different concepts in the labor economic theory, these differences are ignored in this thesis. The reason is that the analysis on wage determination aims at explaining wage differentials among workers. In this sense, it is not necessary to take into account these differences. Then, in the empirical analysis, the term “wages” obviously do not mean the basic wages of workers but all payment both in cash and in kind workers received from their main jobs.

Secondly, the term “discrimination” used here means the “unexplained portion”. It is meaningful in quantitative dimension rather than its word meaning per se. In fact, it comprises three sub-components: environment gap (or pure rent), the overpayment of the advantage group, and the underpayment of disadvantage group.

            Thirdly, care must be taken about sources of self-selection. Heckman (1977) argued that self-selection derived from an individual’s decision on labor marker decision. Neuman and Oaxaca(2001) proposed that occupational choice could be regarded as a source of selection bias. Meanwhile, Appleton et al. (2001) argued that in the developing countries the sectoral choice became the critical source of selection bias. Which source of selection bias should be corrected for empirical study? The answer depends on both the availability of dataset and aim of study.

Fourthly, although Oaxaca method has been standard, care must be taken about detailed  wage decomposition. Specifically, when trying to estimate the separate contribution of each dummy to the unexplained portion, the wage decomposition will produce the arbitrary results. In other words, the estimates are not independent upon the choice of left-out reference group while the overall decomposition is not invariant.

3. Returns to Human Capital and Wage Discrimination – Empirical Evidence

3.1. Data

            This thesis uses two data sets: one is derived from  the second VLSS was undertaken between December 1997 and November 1998 and the other comes from the surveys of 150 textiles and garments firms in Vietnam carried out in 2001. The first is used to analyze wage determinants and discrimination in the light industry sector while the second is to refer the wage analysis in the T&G.

3.1.1. VLSS 2

To analyze the wage determinants as well as possible wage discrimination in the light industry sector, the sample is set up based on VLSS 2. The sample includes 1493 individuals who all involving the light industry sector of which 730 are wage earners. An individual is defined as a wage earner in light industry if (i) he/she has a main job involving one of 18 industries included in the light industry sector during the past 7 days prior to the interviewed day, (ii) this job is not seasonal work, (iii) he/she receives a salary or wage for this works.

3.1.2. The T&G Firms Survey

Data served for analyzing wage determination in the T&G is derived from the most recent survey of textiles and garments firms in Vietnam implemented by the Institute of Economics in 2001. The interviewed firms were randomly drawn from 8 provinces of Vietnam, which are Ho Chi Minh City, Dong Nai, Binh Duong in the South, and Ha Noi, Hai Phong, Nam Dinh, Thai Binh and Phu Tho in the North. These are the 8 top provinces of terms of employment in the whole industry, representing 78% of total revenue and 76% of total jobs in the industry. The sampling frame was constructed based on the census of manufacturing firms conducted by GSO/UNIDO in 1998. The sampling frame did not consist of small firms with 50 workers and less, due to their perceived poor accounting books.

            The survey also includes information on workers. Eight workers were chosen from each sample firm, and therefore the worker sample size reached 1200 observations.  All variables regarding wages, education, experience and migrant status of workers have been collected.

Because the sampling probability differs across firms and workers, the sample needs to be weighted to infer population characteristics from the sample characteristics. Both firm weight and worker weight have been calculated based on standard technique of sampling and post-stratification weighting. Doing as such, the sample is considered as the representative for the population.

3.2. Variables Specification

First, regarding multinomial logit model, we assume that  individuals face two mutually exclusive choices: working as wage earners or self-employed for light industry workers. For the case of the T&G workers, we assume that  individuals face three mutually exclusive choices: working in SOEs, working in PEs, and working in FIEs.  The sectoral choice depends on the perceived net differentials in the wage and non-wage compensation in each of these sectors (Tansel, 1999).  In other words, individuals decide in which sector to participate by comparing  their utility in each sector. Thus, the sectoral choice depends on workers’ preferences and human capital. In this sense, variables specification for participation in a particular sector equation is ad hoc. Human capital variables are often formal education attainment, experience, and migrant status. Workers’ preferences are shaped by other characteristics such as gender, marital status, location, and non-labor income (e.g. see Tansen, 1999). In some cases, characteristics of households such as dependency ratio, sectoral location of household head, non-labor income of household, etc. might be incorporated into this equation (e.g. see Liu, 2002) (see more details in Appendix)

Second, we estimate Mincerian wage equation augmented by selectivity correction term. Regarding the wage equation, the variables specification is as follows:

The dependent variable in the wage equation is real hourly earnings in natural logarithm. In theory the  “wages” definition is different from earnings and compensation but in practice, this difference is ignored. As in other empirical studies on wage determination, a worker’s wages are calculated by summing both payments in cash and in kind. To do so, we have the indicator that truly reflects price of labor in the market. The real hourly wage of a worker is his wage divided by total actual work hours after adjusting for regional deflator. More details in regional deflator are presented in Appendix 3.

            The other control variables in the wage equation are briefly described as follows:

            In the case of light industry’s wage analysis, years of schooling variable is served as a proxy for education. Differently, for the T&G case, a set of dummies is used to capture the wage effects of educational levels, namely secondary education, technical training, university. If one completes a particular level, it is coded one. It equals zero if otherwise.

Experience: the proxy for the potential labor market experience, or potential on-the-job training. It equals age minus 6(years) minus years of schooling. The experience is included both in linear and quadratic terms.

Gender: the gender dummy is coded as one if a worker is a male, coded as zero if otherwise

Region:  if a worker is living in the south the region dummy equals one, it equals zero if otherwise.

Migrant status: a light industry worker is defined as a migrant if 7 years ago he/she lived in other area. Meanwhile, a T&G worker is considered as a migrant if 3 years ago he/she lived in another province.

Sector: if a worker is working for SOEs, the sector equals one. Sector is coded as 0 if otherwise.

Occupation: to capture wage effects of occupation choice, we introduce two dummies. The first dummy for managers and professionals with skilled laborers as reference group, the second is for skilled workers with non-skilled labor workers as the reference group.

3.3. Evidence from the Light Industry Sector

3.3.1. Returns to Human Capital

            It is of interest to look at determinants of labor force participation in light industry. Notably, all variables constructed are statistically significant at 13% or lower (more detailed in Appendix). The results show that marital status and migration have negative effects on decision to wage sector. For example, migrant status discourages men from labor force participation.  On the other hand, other variables have positive effects. The more educated men are more inclined to be wage earners. Location has strong effects on men’s decision. For example the favor has been done for southern and/or urban residents.

Now it is worth investigating what determine wages. As can be seen in the Table 1, both regression results with and without selection bias correction are reported. The results seem to be similar because of the statistical insignificance of the lambda variable-correction term. In other words, the lambda is not statistically significant at any conventional confidence level (i.e. 10% level or lower). Statistically, it implies that the sample selection has no effects on wage. Nevertheless, our analysis is still based on regression results with selectivity correction.

Practically, light industry workers are compensated for their human capital. Quantitatively, the above table shows that the most important forms of human capital investment statistically have wage effects. However, it is noting that the extent of wage effect varies with respect to each kind of human capital investment. Thus, it urges us to investigate which kind determines workers’ wages in the specific industry, namely light industry.

Table 1: The Result of Wage Equation for Light Industry

(Dependent variable: Log real hourly wage)

*: denotes to statistical significance at the 5% level

Source: Author’s calculation based on VLSS2

Education and Experience 

Education, one of the most important human capital investments, has statistically significant, but small wage effects. Another year of schooling is rewarded about 4.76%. It equals to average return to schooling of the whole economy in the 1992-1993 (Moock et al. 1998). In a recent study, Liu (2002) shows that the return to schooling is about 4% for the private sector worker, and about 7.4% and 5.8% for government sector and SOEs, respectively. Then, it is safe to say that the return to education in light industry is not high as compared with the economy as a whole. The reason may be due to its characteristics. The average years of schooling in light industry is 8.6 while the whole economy is 9.1 (Bales, 2000). In this sense, the low return to education in light industry is partly due to the low average education level of workers. On the other hand, most jobs of many industries in the light industry sector require basic skills such as sewing in the footwear, garment and therefore the sector offers employment opportunities for workers with less educated, e.g. with primary education. However, both educated and less educated workers apply for, and then the more educated are often hired (Belser, 2000). As a result of the over-qualification of workers, education investments are underpaid.

            Differently, the not so high return to education discourages talents from taking part in its labor force. Other industries such as banking and service, which promise higher returns, have advantages over light industry to attract high education level workers. That is a challenge to the development of light industry towards further exploitation of comparative advantages

As compared with education, experience pertains to lower returns. An additional potential experience raises wages by about 3.58%. The lower return of experience suggests that employers regard formal education more important than potential experience. It is entirely consistent with traditional ideology of Vietnamese. Formal education is always appreciated among Vietnamese society.

The signs of coefficients associated with experience and experience squared confirm theoretical arguments. It means wages follow the parabolic shape with respect to experience, peaking somewhere at workers’ mid-working life.  To investigate the experience level at which the log wage is the highest, one first differentiates the wage equation with respect to experience:

¶lny_i /¶exp= 0.0358-0.000652*2*exp

Set this equal to zero, it is found that at the level of  28 potential years, log wages are maximized. Despite that it also depends on workers’ schooling, on average, 50-year-old workers with university completion enjoy the highest return. This result makes us feel confused as according to Labor Code  the retirement age is 60 years in full for males and 55 years in full for females.

Gender and Marital Status

Not surprisingly, gender is statistically significant at 5% level with the high value of the coefficient, implying a substantial wage effect. Being a male raises workers’ wages by 26.2%. It says that the gender wage gap and possible discrimination have been in fact in light industry. It is of interest to further explore what are behind this. Then, in next part, we will focus more details on this issue.

In contrast, marital status variable is not statistically significant. Although this result is contrary to empirical findings in other countries where marital status has wage effects (e.g. Paternostro & Sahn, 2000), it confirms our intuition that this variable has strong effects on participation decision rather than wages.

Migration and Regional Location

Migrant variable is not statistically significant. This result may not follow our expectation. It is interesting to explore what lies behind this. First, it is necessary to look at the government’s policy and regulation on labor mobility. Until in the early 1990s, free geographic labor moves were strongly regulated by central government. The government planned the number of migrants as well as migration process. In this sense, practical admittedly,  migrants cannot be treated as a “disadvantaged” group. Moreover, local governments at large cities raise high barriers to migrants. For example, having a house as well as having a permanent job is must for those who intend to migrate into city. Meeting these requirements means that migrants have certain endowments that are no less than the urban residents’, or even higher. It is well in line with previous studies on migration in Vietnam (e.g. UNDP, 1998). Notably, in the late 1990s with the increasingly widening region gap as well as loosened regulations on geographic mobility, free geographic moves have been accelerated. Hopefully, the empirical findings in the T&G in the next part can capture this trend, and further explore the possible wage discrimination against migrants.

Differently, the insignificance of migrant status is due to sample characteristics. As noted earlier, the VLSS 2 sample is based on either VLSS1 or in 1995 MPHS. If household had moved away, another was chosen randomly. In fact among households in the VLSS 1 sample, 300 households had moved away, of which 19 households were temporarily from commune. Thus, VLSS 2 cannot contain useful information on geographic migration at that time.

The south dummy is included to capture the wage effects of local labor market. A substantial premium, 56.36%, is attached for the southern workers with the same endowment. The high premium is a strong incentive for north-south migration. It is in line with previous studies such as Moock et al. (1998). They showed that workers in the South earned 50 % more than those in the North despite the fact that had a third less schooling. They also argued that it was due to the higher productivity with better machinery equipments in the south. Moreover, it is clear that in the South the marketisation degree may be higher than in the rest. Then, employers in the South have incentive to compensate workers at the wage rates reflecting more truly their human capital than the North.

Ownership

            Sector variable is not statistically significant at any conventional confidence level(i.e. 10% or better) but its coefficient shows the positive sign. Perhaps, quantitatively, sector has no wage effects. The main reason lies behind this is, to a certain extent of marketisation degree, the private sector, which pursuits profit maximization, has not enough incentive to truly compensate for human capital. At the same time, the SOEs aim at not only profit maximization but also employee satisfaction. In this sense, they have no reason to lower workers’ wages and salaries even in the case that they are loss-making. Another study by Bales and Rama (2001) argue that the public sector workers are overpaid based on ‘worker approach” but based on “job approach” the finding is on contrary despite that they use the same dataset, namely VLSS2. In a nutshell, in Vietnam, the public - private wage gap is not clear, especially in the early transition phase. Yet, it is no doubtful that increasing degree of marketisation is widening this gap. Thus, further study on possible public - private wage gap can be done only with more recent dataset.

Occupation

Regarding occupation dummies included, the professional dummy is statistically significant while the skilled is on contrary. Quantitatively, the occupational choice between managerial and professional jobs and other jobs becomes critical to wage determination. On average, managers and professionals earn 46.28 % more than others. Practically, the high premium of being professionals entails the indirect returns to human capital as to be a manager or professional, one obviously must certain productive characteristics. In this way, the high premium of being professional is a monetary benefit that compensates for “white-collar” workers with high educational level.

Constant term

            The negative sign of constant term may suggest that  light industry workers do not enjoy premium. However, it is statistically insignificant at 5% level. Quantitatively, it is difficult to get more information on this.

In conclusion, the estimates of  wage equation in light industry prove that human capital investment are compensated.  The literature argues that the clearer earnings-human capital patterns reflect the labor market efficiency. In this sense, the human investment reward is a positive sign in the labor market for light industry. Yet, the low rates of return suggest that the reward is still modest, relative to other industries. This finding may do not satisfy some economists who wonder about the heterogenousity of light industry. They may assume that there is a difference between the wage structure of two  groups in the sector: industries with export orientation (garment, electronics...) and the rest (tobacco...). In other words, the hypothesis suggests that export oriented industries tend to pay higher for human capital. We try to test this hypothesis by using ttest but in fact the difference is statistically insignificant. Thus the message here is more convincible: the impressive growth of light industry does not bring monetary benefits to workers as expected.

3.3.2. Gender Wage Gap

Does the gender wage gap exist?

The above results from wage determinants show that wage disparity is in fact in the light industry sector.[3] It is consistent with previous studies, e.g. Liu (2002), which argued the existence of wage related discrimination on the basis of gender in the economy. The log wage differential in light industry is about 0.25.

However, one can wonder about the validity of this gap. To answer, we calculate the gender wage gap at the mean values of explanatory variables. To put it simply,  it is concerned with wage gaps between female workers and male workers with similar characteristics. It is found that on average, female workers’ wages are about 80% of those of male workers even though they have similar endowments. Then there is no denying the existence of  wage gap on the basis of gender in light industry.

Sources of the Gender Wage Gap

Despite that the comparison of wage determinants is plausible, it fails to explain fully the source of gender wage gap. Thus, it is necessary to analyze the results obtained from decomposition. A closer look at the gender wage gap indicates that the gender wage gap is mostly due to discrimination.  Discrimination accounts for 93.62% of the gap while the productivity gap represents only 6.38%. It confirms that gender discrimination is in fact in light industry.           

Table 2 shows that discrimination causes the log wage differentials of 0.4458. It means that due to discrimination, male workers are overpaid by 56.17%, holding other things equal. Notably, there is no evidence of male advantages as well as female disadvantages. The negative signs of both sub-components are translated that female workers receive a premium while male workers: a discount. In other words, they do not contribute to discrimination in the labor market. Its main reason lies behind the characteristics of the light industry sector that generally favors female workers. For example, many kinds of  jobs offered by electronics or beverage and food processing industries are well suited for females rather than for males.

Table 2: Decomposition of Gender Wage Gap

Source: Author’s Calculation based on VLSS2

            Discrimination is entirely determined by pure premium or environment gap. As can be seen in Table 2, it represents the log wage differential of 0.7232. The message here is clear: the existing condition favors male workers, and therefore pays higher to them. Despite that there is no discriminatory behaviors of employers, men overpayment still persists, causing discrimination in the labor market[4]. Clearly, it has been rooted in the traditional ideology based on the social norm that men are productive. In this sense, it is difficult to tackle this problem.

            Productivity gap is only small with the log wage differential of 0.0299. It is evident that the differential between productive characteristics of males and females is not large. Also, it shows that the access to opportunity for human capital investment is not different on the basis of gender. It may be the result of economic history, especially resource allocation as well as education policy of the government under the centrally planned system.

             In short, discrimination is a major determinant of the gender wage gap in light industry despite that there is no signs of discriminatory behaviors. Environment gap is the sole source of discrimination, placing female workers at a disadvantage. In this sense, discrimination is not solved without great effort emphasized on changing social norms and perception bias against females. 

3.4. Evidence from the T&G

3.4.1. Sectoral Location and Returns to Human Capital

It is of interest to look at sectoral choice equations. SOEs sector is regarded as comparison group because in fact there is a distinction between SOEs  and non-SOEs. To make a decision on participating in which sector, individuals consider expected pecuniary and non-pecuniary benefits offered by different sectors. In this sense, SOEs sector is quite different from others because they pursuit employee satisfaction rather than profit maximization like PEs and FIEs. Also, the government use jobs in SOEs as income transfer (Bales and Rama, 2002).

            The estimates of multinomial logit model produce plausible results. First, it confirms that more educated workers are more inclined to be in SOEs. Much of empirical work in Vietnam agrees with this result (e.g. Moock et al, 1998). Yet, it is important to note that while all variables reflecting education in PEs equation are significant, only university education dummy in FIEs is significant at 5% level.  Second, experience has effect on sectoral choice between SOEs and PEs. Workers with more experience are more inclined to participate in PEs. Not surprisingly, it is due to the recruitment policy of PEs that favors workers with more experience level, which is in turn a result from their aims of training cost minimization. In contrast, experience plays insignificant role in sectoral choice between SOEs and FIEs. Third, male workers are found to prefer to work in SOEs than other sectors. This result also agrees with findings from other studies. For example, MOLISA(1998) reveals that the predominance of male workers in SOEs despite the fact that labor force participation rates do not vary with respect to gender. Fourth, non-migrants are likely to be in SOEs rather than non-SOEs sector. In reality, there are a few observations of migrant in sample SOEs firms. Next, the results from regression show that sectoral choice depends on residence of workers. Workers who live in the south are much more inclined to be in non-SOEs. This trend is clearer in workers’ decision on FDI sector participation. In the same fashion, individuals living in HCMC or Ha Noi are likely to be PEs workers rather than SOEs. Yet, the insignificance of this variable in FIEs participation equation implies that it may have no effect on sectoral choice between SOEs and FIEs. Last, marital status is likely to play no role in determining to participate SOEs or non-SOEs.

            Despite that participation equations produce convincing results, one still wants to compare wage equation with and without selectivity correction. Table 3 reports results for different sectors with and without selectivity correction. Convincingly, the Mincerian equations with selectivity correction yield more plausible results. Moreover, the correction terms (lambda) in all three equations are significant at 1% level, implying the selection bias in the OLS estimation of the wage function. With such results, we focus on investigating wage determinants in different sectors based on results from wage equation with selectivity correction.

            As can be seen in the Table 3, the statistical significance of coefficients on correction terms confirms the presence of selection bias. Moreover, the negative sign of lambda in SOEs wage equation implies unobserved characteristics of SOEs workers, which determine their sectoral choice, have adverse impact on their earnings. Meanwhile, PEs and FIEs workers experience the positive correlation between unobserved characteristics and earnings.

            Table 3  reveals that all three sectors reward human capital. Yet, return to human capital varies across sectors. Moreover, some variables become critical in determining wage in a particular sector but show insignificant role in others. Then, it is of interest to examine each wage determinant in each sector to exatract more information.

Table 3: Wage Determinants by Sectors with and without Selectivity Correction

***,**,* denote significant at 1%, 5%, 10% level respectively.

(1): Migrant dummy is dropped in SOEs wage equation because lack of observations.

Source: Author’s calculation based on the 2001 T&G firms survey                                                  

Education and Experience

            With respect to wage effect of formal education achievements, both SOEs and PEs reward for any education level introduced while FIEs do not follow the same pattern. In FIEs wage equation; only coefficient on university education is significant at 10% level. Generally, the higher educational level pertains to the higher return for all sectors. Notably, returns to formal education in SOEs are the highest. Is it evidence of stronger marketisation degree of SOEs as compared with non-SOEs? Or is it a proof of the fact that SOEs aim at employee satisfaction rather than profit maximization? To answer, more information is required but the latter is likely to be more convincing.

            More specifically, the low returns to high school completion do not surprise us. As a common perception, in the light of “option value” concept, workers with secondary education are benefited from further study rather than private return to education. Regarding technical training, SOEs workers experience the highest return. Meanwhile, the wage effect of this variable in PEs is significant at 10% and in FIEs is insignificant at even 10% level. This phenomenon may be due to low quality of vocational education system. In addition, return to university education is the highest in the SOEs and lowest in PEs. The university degree doubles SOEs workers’ wages but raises wages by 32.3%, and by 56.8% for PEs and FIEs workers, respectively. Moreover, it is noted that students must complete secondary education before attending universities. In this sense, workers with the university degree may raise   wages by more than 150%, 73% and 58% for workers in SOEs,  PEs and FIEs. Compared with other studies, e.g. Nguyen (2002), the returns to formal education in the T&G are lower than the average rates of the whole economy.[5] Likely, low educational level of T&G workers accounts for this?

            Variables related to experience are found to be significant, except for experience level for SOEs workers. Importantly, the signs of all variables are as expected: positive sign of linear term in experience, and the negative sign of squared experience. It confirms that return to experience follows the parabolic shape. It is observed that the wage experience profile is steeper in SOEs and FIEs, implying higher impact of an additional experience year in SOEs and FIEs. Return to experience peaks at 29 years of experience for SOEs and FIEs workers while it peaks at 40 years of experience in PEs. Again, the low returns to experience makes us feel confused.

Gender

            Gender wage gap is evident in non-SOEs sectors. A male worker in FIEs is overpaid by 11.6%. At the same time, a male in PEs enjoys 9.4 % increase in earnings. Despite that over-representation of females in T&G, the gender wage gap is still a problem. It is noting the gender wage gap is not evident in SOEs sector. Coefficient on this variable in SOEs equation is insignificant at 5% level. Moreover, the sign of this coefficient in the equation without selectivity correction contradicts that in the equation with selectivity correction. Then, it is still question of whether gender wage gap exists in SOEs sector. Moreover, if gender wage is in fact in non-SOEs, what accounts for this gap? The next section will provide an answer.

Migration

            The wage differential between migrants and non-migrants is observed for non-SOEs sector. 23.4% and 18.5% increases in earnings are for FIEs and PEs non-migrant workers. Likely, it suggests that there might be wage discrimination on the basis of migrant status. It is noting that migration is considered as human capital investment in the sense that workers sacrifice the present benefits for the higher expected benefits. At the first stage, migrants often face wage discrimination but for the long term the expected benefits will be greater regardless whether they know wage discrimination at destination. Moreover, their wages at destination are far much higher than those in the their countryside. Then, it is the determinant of geographic migration. Calculation on the T&G dataset indicates that the difference between wages before and after migration is significant at 1% level with the mean monthly wage differential is about 450 thousand dongs. Such the high wage differential is strong incentive for geographic labor moves despite that wage discrimination might exist.

Regional Location

            In addition, the wage effects of local market are impressive in non-SOEs sector. This result is well in line with previous studies, which confirm that offered wages in the south rank the top in the whole economy. More specifically, the gap between two regions in the PEs and FIEs is large. 89.6% and  27.1% are for whom work for PEs and FIEs in the south, respectively.  The more favorable business environment and/or more efficiency at the firm level are the main reasons for such huge gaps? Is firm size wage effect valid in such a case? Bear this caveat in mind. In appendix, we have conducted some simple test on this concern. In practice, on average, the firm size in the south is larger. According to SMEs definition of the government (official letter No 681/CP), 93.1% of the sample in the south is considered as large firms while the north: 76.1%. Firstly, we use ttest to examine the difference in mean hourly wage between SMEs and large firms. Accordingly, the difference in mean hourly wage of workers in the large firm and in the SMEs is about 2.43 thousand dongs per hour. This test is significant at 1% level. This result confirms our hypothesis that there is a positive relation between firm size and wages. Secondly, we try to examine the correlation between workers’ wages and firm size. It is observed that there is positive correlation between workers’ wages and firms’ revenue. Likely, there is a sign of rent sharing (More details in Appendix).

Occupation

            Coefficients on occupational dummies are statistically significant for all three sectors at 1% level. Not surprisingly, the wage differential across occupation for FIEs is the clearest FIEs have both capability and managerial autonomy to pay higher wages to managers and professionals. As such, it ensures that wages of managers are commensurate with their authority and responsibility. PEs follow the same pattern but their rewards for higher level are still low.

Constant term

            It is also noted about constant terms. The constant terms in both  SOEs and FIEs  are significant and have the high value. It is evident that SOEs and FIEs offer high premium for their workers.[6] Likely, the notion of profit sharing accounts for this. It should be noted that most of FIEs and SOEs are large ones. Specifically, 100% of FIEs and 92.5% of SOEs are considered as large firms. As above mentioned, the positive correlation between workers’ wages and firm size has been reported is evident. Moreover, Table 4 shows that SOEs and FIEs achieve higher return to capital as well as higher labor productivity. In this sense, it is safe to say that both SOEs and FIEs have strong incentive to share their profit to employees.

Table 4: Return to Capital and Labor Productivity of T&G Firms by Ownership

Source: Author’s calculation based on the 2001 T&G firm survey                            

In conclusion, the closer examination of wage determination in the all three sector sends a clear message: all three sectors reward for the most relevant human capital investment of workers. That is encouraging sign for the labor market regarding the T&G. However, the analysis also indicates that there might be unexplained wage gap, especially gender wage gap in non-SOEs sector. The following section will aim to explore these gaps.

3.4.2. Gender Wage Gap in non-SOEs

            The results from Mincerian equations indicate that gender plays a significant role in determining wages of non-SOEs workers. The Chow test also confirms that it is appropriate to estimate wage equation for men and women in each sector, namely PEs and FIEs separately.[7] In other words, wage structures of PEs male workers are different from those of PEs females, and so do FIEs workers.

            To examine whether these gaps exist in reality, we calculate gaps between male and female workers at the mean values of explanatory variables. In other word, we examine the gaps between male and female workers who have similar endowments. For both sectors, FIEs and PEs, female workers’ wages are just over 90% of those of males despite the fact that they have similar endowment. Thus, it is clear that gender wage gap is in fact in non-SOEs sector.

            Interestingly, the wage gaps in two sectors take the same value with the log wage differential of 0.14. However, the sources of gender wage gap vary with respect to sector.

Table 5: Decomposition of Gender Wage Gap