INTRODUCTION
Problem Statement
Education is crucial in today's labor markets, as numerous studies across various countries and time periods have consistently shown that individuals with higher education levels tend to earn more than their less-educated counterparts Pioneering research by Mincer (1974) established an empirical model linking earnings to education and experience, demonstrating that the logarithm of earnings can be expressed in terms of years of schooling, work experience, and the square of work experience.
Since the Vietnam Living Standards Survey (VLSS) was first conducted in 1992-93, numerous studies have utilized its data alongside the Mincerian earnings function to analyze the returns on education in Vietnam, including research by Glewwe.
& Patrinos, 1998; Gallup, 2002; Moock et al., 2003; Liu, 2006; Nguyen Xuan Thanh, 2006; Vu Trong Anh, 2008; Vu Thanh Liem, 2009; Doan & Gibson, 2010; etc The results are also diverse
A significant study by Moock et al (2003) examines the returns to education in Vietnam using the Mincerian earnings function and data from the Vietnam Living Standards Survey (VLSS) 1992–93 The findings reveal that the estimated rates of return to education are relatively low, at just 4.8% Notably, this low return is particularly evident in the average rates associated with primary education.
Research by Psacharopoulos and Patrinos (2002) reveals rate of return estimates for 98 countries over more than 30 years, while Trostel, Walker, and Woodley (2002) provide estimates for 28 countries, and Polachek (2007) covers 42 countries The returns on university education are notably higher, with rates of 13% for females and 11% for males, compared to just 4% and 5% for secondary and vocational education, respectively Furthermore, higher education yields greater returns for females (12%) than for males (10%).
After two decades, it is essential to reassess the returns to education in Vietnam, examining how these returns have evolved over time This analysis also highlights the gender gap in educational returns, determining whether males or females benefit more significantly Additionally, it investigates sectoral disparities among public, private, and foreign sectors to identify any notable differences in returns to education The findings hold critical implications for policymakers, guiding them in shaping effective wage and educational policies.
This study aims to replicate the work of Moock et al (2003) by utilizing the Vietnam Household Living Standard Survey (VHLSS) from 2008, conducted by the General Statistics Office (GSO) It will employ a Mincerian earnings function while introducing a novel regression method, specifically using clustered data at the household level through panel commands, which marks the first application of this technique in this context The findings will provide valuable insights and policy implications based on the results obtained.
Research Objectives
There are 03 main objectives in this study:
This study aims to estimate the private returns to education based on years of schooling and educational levels for both males and females, focusing on recent data from private, public, and foreign sectors.
(2) To assess the variation in returns to education by comparing with the findings from Moock et al (2003);
2 Please refer to the chapter on methodology (Chapter 3) for more details
(3) To propose some policy options.
Research Questions
The research questions are proposed:
Recent studies indicate that the rates of return to education vary significantly based on years of schooling and educational levels, with distinct differences observed between males and females Additionally, these returns differ across sectors, including private, public, and foreign employment, highlighting the importance of educational attainment in determining economic outcomes for both genders.
(2) How are the rates of return to schooling different comparing with 15 years ago? Should the rates increase or decrease?
Research Methodology
In this study, I analyze data from the 2008 Vietnam Household Living Standards Survey (VHLSS) conducted by the General Statistics Office (GSO) Utilizing the Human Capital Model developed by Mincer in 1974, I employ a regression method based on clustered data at the household level, applying panel commands rather than relying on a standard cross-sectional Ordinary Least Squares (OLS) estimator.
Structure of the Thesis
This paper is organized into several key chapters: Chapter 2 reviews existing literature and empirical studies globally and in Vietnam, while Chapter 3 outlines the data samples and research methodology employed Chapter 4 presents the findings derived from descriptive statistics and econometric models Finally, the concluding chapter offers a summary, policy recommendations, acknowledges the study's limitations, and suggests directions for future research.
LITERATURE REVIEW
Definition
Human capital refers to the skills, knowledge, and experience that individuals or populations possess, which are considered valuable assets to organizations or countries.
The rate of return measures the gain or loss on an investment over a specific period, expressed as a percentage of the initial investment cost It encompasses any income generated from the investment, along with realized capital gains.
The return on education can be assessed through various methods depending on the level of analysis At the societal level, it is measured by the investment in education in relation to national wealth At the enterprise level, it reflects the impact of training investments on business performance For individuals, the return is evaluated by comparing years of schooling to lifetime income.
The "individual return to education," often referred to as "private return to education," focuses on the benefits experienced by individuals rather than the societal advantages, which are known as "social return to education." This study specifically examines the individual level of educational returns.
A Standard Model of Human-Capital Investment
Mincer (1974) introduced the standard Human Capital model, which analyzes an individual's earnings through the lens of their years of schooling, labor market experience, and the square of their experience This model is grounded in theoretical principles that emphasize the relationship between education, experience, and income potential.
Mincer argues that potential earnings at time t are influenced by human capital investments made at time t-1 If we denote potential earnings at time t as Et, an individual allocates a portion kt of these earnings to invest in human capital, with a return rate rt in each period Consequently, the potential earnings at time t+1 can be expressed as a function of these investments.
After a number of repeated substitutions, we have:
Assuming that schooling consists of s years of full-time investment, where each year yields a constant rate of return (rs) and the return from post-schooling investment remains steady over time, we can reformulate equation (2.2) accordingly.
ln 0 ln( 1 ) 1 ln( 1 ) ln 0 1 ln t s j j t s j j t E s k E s k
Where, the last approach is for small value of β, λ, and k
To link between potential earnings and experience z from labor market, the post- schooling investment is assumed to be linearly decreased over time
Where T is the last year of working life; z=t-s ≥0; and (0,1)
By subtracting (2.4) from (2.5), we got an equation for net potential earnings:
Equation (2.6) can be rewritten as follows: ln npe t s z z 2 (2.7)
With final assumption that, at any time t, the observed earnings are equal to net potential earning, we have: ln Y t ln npe t (2.8)
Replacing (2.8) into (2.7), we got the standard Mincerian earnings equation: ln Y t s z z 2 (2.9)
Empirical Studies on Estimating Returns to Education
Numerous studies globally utilize the Mincerian earnings function to estimate the returns on education, despite the significant limitation of sample selection bias Ordinary Least Squares (OLS) regression is commonly employed in these analyses Additionally, various supplementary variables, including gender, regional dummies, ethnicity, race, marital status, and union membership, are often included in the estimation process to serve as exogenous factors.
“control variables” which may shift the earnings function upward or downward depending on their signs
In their 1997 study, Johnson and Chow analyze the rates of return to schooling in China using OLS regression on 1988 survey data They find that the return to education is 4.02% in rural areas and 3.29% in urban settings Notably, urban females experience a higher rate of return at 4.46%, compared to 2.78% for males Furthermore, urban members of the Communist Party see significantly lower returns to education at 2.42%, in contrast to 3.68% for non-members.
Onphanhdala and Suruga (2007) analyze the returns to education in Laos using data from the Lao Expenditure and Consumption Survey (LECS 3) conducted in 2002/2003, incorporating dummy variables for gender, area, ethnicity, type of business, and region in their regression model Their findings reveal that while the returns to schooling in Laos remain low, they have notably increased from 3.2% in 1997-1998 to 5.2% in 2002-2003, with young workers experiencing higher returns (7.0%) compared to older workers (3.9%) This suggests that the benefits of education will continue to rise as market reforms take full effect Additionally, despite the substantial earnings premiums for highly educated workers, the primary education level is still considered the most profitable investment The study also highlights significant wage differentials, with public sector earnings at 2.2% and private sector earnings at 5.2%.
Heckman (1979) developed a two-step simultaneous model to address sample selection bias from nonrandom data, which has gained widespread use across various research fields In 2008, Siphambe utilized this model to estimate the educational returns in Botswana for the years 2002-2003.
Siphambe (2008) analyzes the 2002-2003 Household Income and Expenditure Survey data to assess the returns to education in Botswana, incorporating variables such as age, education, and marital status into a probit equation to derive the Inverse Mill Ratio for the earnings function The findings reveal an average return to education of 15%, a slight decline from 16% in the 1993-1994 period Notably, upper secondary education experienced a significant drop of 28 percentage points, with returns falling from 36% to 8% Conversely, university education saw an increase in returns, rising to over 50% (24% compared to 11%) Apart from upper secondary education, the return patterns align with Siphambe's earlier findings (2000) Additionally, the study indicates that both females and males experienced similar returns on education (approximately 15%) during 2002-2003, contrasting with Siphambe (2000), where females had higher average returns than males.
Another critical problem when studying educational returns is the endogeneity
In addressing unobserved heterogeneity, Card (1999) identifies three main strategies: first, employing instrumental variables derived from the institutional characteristics of the education system, as demonstrated by Angrist and Krueger (1991); second, utilizing family background as an instrument for educational attainment, as explored by Ashenfelter and Rouse (1998) and Nakamuro.
Research by Inui (2012) and Ashenfelter and Krueger (1994) has primarily concentrated on estimating the average impact of education on earnings, utilizing both Ordinary Least Squares (OLS) and Instrumental Variables (IV) techniques These studies often analyze data from twins to assess the relationship between schooling and income.
Angrist and Krueger (1991) argue that school start age policies and compulsory attendance laws lead to individuals born at the beginning of the year starting school later As a result, these individuals often complete less schooling and are more likely to drop out compared to those born closer to the end of the year.
The estimation draws on a variety of data sets constructed from the Public Use Census Data in 1970 and 1980 The samples focus on males of 16 years old born in the
US to specify the 1920-1929 corhort (in 1970 Census); and 1930-1939 corhort and 1940-1949 corhort (in 1980 Census)
The authors utilize the interaction between quarter-of-birth and year-of-birth as an instrument for education to assess the impact of compulsory schooling laws across different cohorts After adjusting for factors such as age in a quadratic form, race, marital status, and urban residency, their difference-in-difference analysis reveals that the returns on an additional year of schooling are 10% for men born between 1920 and 1929, 6% for those born between 1930 and 1939, and 7.8% for men born between 1940 and 1949.
In their 1994 study, Ashenfelter and Krueger analyzed primary data from the 1991 Annual Twins Days Festival in Twinsburg, Ohio, concluding that workers' abilities and educational attainment are uncorrelated, suggesting that schooling does not directly impact earnings Their final sample consisted of 298 pairs of identical twins, who are presumed to share the same innate abilities yet differ randomly in their educational levels.
3 This empirical study is not included in the review of Card (1999) but in line with the work of Ashenfelter and Rouse (1998), so I add in
Identical twins, or monozygotic twins, originate from a single egg and sperm, making them genetically identical and likely to share similar innate abilities In contrast, fraternal twins, or dizygotic twins, develop from two separate eggs and sperm, resulting in genetic differences and a higher susceptibility to omitted ability bias By utilizing each sibling's report on their twin's educational attainment as an instrumental variable, researchers discovered that an additional year of schooling can increase wages by 12-16%.
Ashenfelter and Rouse (1998) analyzed data from the Annual Twins Days Festival in Twinsburg, Ohio, known as the Princeton Twins Survey, covering three years from 1991 to 1993, which included 340 pairs of identical twins Unlike traditional Mincerian equations that focus on experience, the authors controlled for age and utilized the difference between twin 2's report of twin 1's education and twin 2's own educational report as an instrumental variable Their fixed-effect estimator results indicated that the average annual returns to schooling for identical twins is approximately 9%.
A recent study by Nakamuro and Inui (2012), building on the work of Ashenfelter and Rouse (1998), investigates the causal relationship between education and earnings using a sample of twins in Japan The analysis is based on data from 2,257 identical twin pairs, gathered through a web-based survey, and incorporates adjustments for measurement errors to enhance the accuracy of the findings.
IV method, the authors obtain 9.3% as the average returns to education in Japan
In Vietnam, research on the Mincerian function is limited, with several studies employing Ordinary Least Squares (OLS) regression methods, both one-round and two-round (Glewwe & Patrinos, 1998; Gallup, 2002; Moock et al., 2003; Vu Trong Anh, 2008; Vu Thanh Liem, 2009) Additionally, some researchers have utilized the Heckman two-stage approach to address sample selection bias (Liu, 2006), while others have applied the Heckman one-step model (Doan & Gibson, 2010) or the difference-in-difference method (Nguyen Xuan Thanh, 2006).
Glewwe and Patrinos (1998) analyze the VLSS 1992–93 data to investigate private school attendance in Vietnam, revealing key insights: wealthier families are more likely to enroll their children in private schools over semi-public options; individuals with equivalent educational backgrounds from private schools earn higher wages compared to their public school counterparts; and the return on education in Vietnam during this period is estimated at 1.6%.
Chapter Remarks
The study employed the standard Human Capital model developed by Mincer
(1974) to build up its conceptual framework Under the framework, the logarithm of observed monthly earnings of an individual is explained by years of schooling, years of
(1) Schooling: divided by years of schooling and levels of schooling including primary, secondary, vocational education, bachelor and above
(4) The logarithm of hours work per week
Models are fitted for all, male, and female; private, public, and foreign sectors
Dependent variable: The logarithm of monthly earnings
Random Effects model, clustered on household
Private returns to education for all, male, female; private, public, foreign sectors experience in labor market, squared years of experience, and the logarithm of hours work per week
Many studies analyzing the returns to education in Vietnam primarily utilize OLS regression for their modeling However, this approach has limitations, such as underestimating standard errors within households and overlooking mean variations across different households.
Individuals within the same household often share unobservable characteristics, including cultural and genetic factors, which can influence their earning potential Consequently, the error terms for these individuals are likely to be correlated due to a shared household-level component Failing to account for this correlation may result in significantly underestimated standard errors.
The OLS estimator overlooks the mean variation among households, producing a uniform intercept at the state level that assumes all individuals share the same intercept This assumption is unrealistic, as individuals from diverse households are likely to have different intercepts.
To enhance the analysis, I transformed cross-sectional data into clustered household-level data and employed a random-effects estimator, rather than relying solely on a standard cross-sectional OLS estimator with cluster-robust standard errors This approach accounts for the correlation in the model by incorporating a random effect for the residuals.
RESEARCH METHODOLOGY
Data
The Vietnam Household Living Standard Survey (VHLSS) conducted by the General Statistical Office (GSO) in 2008 provides comprehensive data from 9,189 households across 3,063 communes The sample was weighted according to the 1999 Vietnam Population Census, reflecting that around 70% of Vietnamese households reside in rural areas The communes were randomly selected from nearly 10,000 communes spread over 646 districts and 64 provinces and cities, with an average of three households interviewed in each selected commune.
This research estimates the returns to education for salaried individuals aged 15 to 60 for males and 15 to 55 for females, focusing on their monthly earnings in the labor market (in thousands of VND) The study excludes individuals working for their households to ensure a more accurate analysis of educational returns.
6 VHLSS separates employment into wage employment, farm self-employment, and non-farm self-employment
This study focuses exclusively on wage earners, measuring earnings solely through the salaries and wages received, including any in-kind payments related to their work For further details on the variables and their coding, please refer to Table 3.3.
Years of schooling are determined by the highest grade completed in the general education system For instance, a student in grade 10 is recorded as having completed grade 9, while a student who dropped out in grade 9 would have grade 8 as their highest completed level In higher education, the years of schooling are quantified as follows: 15 years for college, 17 years for a bachelor's degree, 19 years for a master's degree, and 22 years for a PhD.
Data on individual schooling attainments is accessible, but post-school investment information is lacking in the VHLSS Consequently, the Mincerian earning function is utilized to assess variations in post-school investment among employees, measured by the difference in years of experience This is approximated by calculating the employee's age in years and subtracting their years of schooling.
Hours of work per week are affixed as a compensatory instrument (Moock et al.,
Mincer (1974) highlights that variations in work hours throughout an individual's life cycle significantly influence their annual earnings profile In scenarios where individual wealth is perceived as stable, the cost of time increases with experience until peak earning capacity is achieved Consequently, the fluctuations in earning potential are likely to lead to corresponding changes in working hours in the market This suggests that relying solely on observed annual earnings as a dependent variable may overstate the investment in human capital and the rates of return To address this overestimation, weekly hours worked are considered an essential compensatory factor.
After consolidated to remove errors and inconsistencies, the sample data remains 6,956 individuals/employees, living in 4,335 households
7 VHLSS divides education into general education and vocational education.
Research Methodology
Returns to education are estimated based on the Human Capital Model developed by Mincer (1974) which is formulated by an earning function as follows: i i i i i S EXP EXP u
Where, lnYi is the logarithm of the monthly earnings for individual i
The variable Si represents the number of schooling years for individual i, while EXPi denotes the total years of working experience for the same individual Additionally, EXPi 2 signifies the square of individual i's experience Lastly, ui is the error term associated with the model.
The squared experience (EXP²) in equation (3.1) indicates that earnings are expected to rise with increasing years of experience, but at a diminishing rate, with the coefficient (γ2) anticipated to be negative.
To assess the average returns on education across various schooling levels, dummy variables are generated from the continuous variable representing years of schooling The extended earning function is formulated as follows: PRIM, SEC, VOC, UNIV, EXP, and u.
Where, PRIMi, SECi, VOCi, and UNIVi are primary, secondary, vocational training, and university levels of education completed by individual i
The returns on education for each educational level are determined by subtracting the coefficient of the subsequent level from the current level and then dividing by the number of years of schooling at that level For instance, the rate of return for the kth level (rk) is calculated using the formula: rk = (k - k-1) / nk.
In the education system, the years of schooling required at different levels are as follows: primary education necessitates 6 years (nPRIM = 6), secondary education requires 7 years (nSEC = 7), vocational education also demands 6 years (nVOC = 6), and university education needs 4 years (nUNIV = 4), according to Moock et al.
In this paper, I challenge the assumption made by Psacharopoulos (1994) and others regarding primary school graduates, asserting that the notion of foregone earnings spanning the entire study period is inaccurate; instead, I adopt a more precise approach by considering only one year of foregone earnings (nPRIM = 1) Furthermore, I differentiate between lower and upper secondary education, designating four years for lower secondary (nLOWSEC = 4) and three years for upper secondary (nUPPSEC = 3).
New Approach - CLUSTERED DATA APPROACH in Estimating the Returns
Using cross-sectional VHLSS data from 2008, I estimate the rates of return to education, with the data organized by households While the Ordinary Least Squares (OLS) estimator is a popular method for this analysis, it has limitations, including the underestimation of standard errors within households and the neglect of mean variation across different households.
In households with multiple individuals or employees, the residual terms are unlikely to be independent, as these individuals often share unobservable characteristics that influence their earning potential Factors such as household culture and genetics can lead to higher performance, better educational outcomes, and increased earnings in the labor market, even among those with similar levels of schooling or experience Consequently, the error terms for individuals from the same household are correlated due to a common household-level component, and neglecting this correlation can result in significantly underestimated standard errors.
Cameron & Trivedi (2009) highlight a key challenge in clustering, noting that error terms can be correlated within clusters In scenarios where this is the primary concern, employing standard cross-section estimators with cluster-robust standard errors is appropriate This approach is relevant when addressing correlated error terms within households, allowing for the use of OLS estimators However, a limitation of OLS is its tendency to overlook mean variation between households, resulting in a common intercept at the state level This assumption fails to reflect reality, as individuals from different households are likely to have distinct intercepts.
To enhance the analysis, I convert cross-sectional data into clustered household-level data and apply a random-effects estimator instead of using a standard OLS estimator with cluster-robust standard errors This approach accommodates the correlation within the model by incorporating a random effect for the residuals Specifically, the earnings function errors are modeled as comprising a common household heterogeneity component (uj) and an individual idiosyncratic error (eij).
This study analyzes the private returns to education among individuals grouped within households, focusing on the logarithm of monthly earnings in the labor market as the dependent variable Key independent variables include years of schooling, years of experience, squared years of experience, and the logarithm of hours worked per month as a compensatory factor The data spans a 12-month period, and a cluster-effects model is employed to capture the relationships among these variables.
Where, individual i is in household j; yij is dependent variable; xij is a vector of independent variables; uj is unobserved household characteristics; eij is the error term
Transitioning from panel data to clustered data requires a conceptual shift In panel data, each individual is observed multiple times over time, leading to clustering at the individual level In contrast, clustered data involves multiple observations per household, necessitating clustering at the household level Consequently, the household serves as the panel identifier, while individuals within the household represent the time identifier It is crucial to recognize that, unlike time, individuals within a household do not have a natural ordering.
To effectively analyze household data, I will create a unique household identifier (idhh) for each household and assign random integers to each member within those households This approach will generate a dataset where the household identifiers act as panel identifiers, while individual members represent different time points Although this resembles a panel data structure, it is essentially cross-sectional data.
This article illustrates the process of converting cross-sectional data into clustered data For instance, Household A consists of two members identified as A01 and A02, while Household B has four members and Household C has three members, each with their respective identifiers as shown in Table 3.1 This exemplifies the characteristics of cross-sectional data.
Table 3.1: Sample of cross-sectional data idmem idhh age lnearnings yosch exp exp2 lnhrwrk
To convert the data into a clustered format, I created a unique household identifier for each household and randomly assigned integers (1, 2, etc.) to each member within those households The resulting clustered data is detailed in Table 3.2, where "idhh" serves as the panel identifier and "idmem" functions as the time identifier.
Table 3.2: Sample of clustered data idmem idhh age lnearnings yosch exp exp2 lnhrwrk
Empirical Models of the Returns to Education
The Mincerian theory posits that the returns to education are influenced by both employee and firm characteristics In my empirical models, I have chosen not to include variables representing firm characteristics, as the effects of these variables will be absorbed into the error term.
The Mincerian earning regression function has the following form: ij j ij ij ij ij yosch . 1 exp 2 exp 2 lnhrwrk ij u lnearnings (3.5)
In this analysis, we examine the factors influencing monthly earnings within households, represented by individual i in household j The model includes the logarithm of monthly earnings in the labor market, measured in thousands of VND, as well as years of schooling and years of working experience in the current job Additionally, we consider the squared value of years of working experience and the logarithm of hours worked per week to understand their impact on earnings.
The extended Mincerian earning regression function: ij j ij ij ij ij ij ij ij ij
1 lnhrwrk exp exp univ voc uppsec lowsec prim lnearnings
(3.6) Where, prim ij : Primary level lowsec ij: Lower secondary level uppsec ij: Upper secondary level voc ij: Vocational education univ ij: Colleges level and above
No level and illiterates are reference
The private rates of return to various levels of education are calculated as follows:
Variable Coding
Numerous studies have demonstrated a positive correlation between years of schooling and monthly earnings, indicating that each additional year of education leads to increased income Specifically, Johnson and Chow (1997) found a 3.34% rise in earnings per additional year of schooling, while Moock et al (2003) reported a 4.8% increase Siphambe (2000) noted an 8% boost, and Onphanhdala and Suruga (2007) observed increases of 2.17% in the public sector and 5.23% in the private sector Most notably, Siphambe (2008) highlighted a substantial 15% increase in earnings associated with additional years of education.
Numerous studies have empirically demonstrated a positive correlation between experience and the logarithm of monthly earnings For instance, Johnson and Chow (1997) found that each additional year of experience correlates with a 4.2% increase in earnings, while Moock et al (2003) reported a growth of 6.4% Siphambe (2000) identified an 7.9% increase, and Onphanhdala and Suruga (2007) noted a 1.2% rise in the public sector and a 4.1% increase in the private sector Siphambe (2008) observed the highest growth at 8.5% However, it is important to note that the squared years of experience indicate a negative relationship with the logarithm of monthly earnings.
Research indicates a positive correlation between education levels and the number of hours worked per week with monthly earnings, as demonstrated by various studies (Johnson and Chow, 1997; Moock et al., 2003; Siphambe, 2000; Onphanhdala and Suruga, 2007; Siphambe).
From the above findings, in my study, I would expect that the relationship between log of monthly earnings and its determinants will follow expected signs as the following table:
Table 3.3: Description of the Variables and Variable Coding
The monthly earnings in the labor market, measured in thousands of VND, is a key observable variable According to the Vietnam Household Living Standards Survey (VHLSS), respondents are asked about their total earnings over the past 12 months, including both salary and in-kind payments This total is then divided by 12 to calculate the average monthly earnings.
Years of schooling, measured in years, is a continuous variable that positively impacts returns to education, indicating that more years of schooling lead to higher financial returns Data on years of schooling is sourced from the general education system, where individuals typically complete 15 years for college, 17 years for a bachelor's degree, 19 years for a master's degree, and 22 years for a PhD.
Years of experience, measured in years, is a continuous variable that positively impacts earnings, indicating that greater experience correlates with higher returns to education This experience is calculated by subtracting years of schooling from age Conversely, the squared years of experience suggest a negative relationship, indicating that the connection between experience and educational returns is not linear but rather parabolic, reflecting diminishing returns to education.
The primary education level is represented as a dummy variable, where prim=1 indicates individuals who have completed primary education without any vocational training, while a value of 0 applies to others This variable is expected to have a positive sign, suggesting that individuals with primary education earn more compared to those with no education or who are illiterate Thus, the primary education level is associated with higher earnings.
Lower secondary level is a dummy variable; lowsec=1 for graduated lower secondary education and no vocational education; 0 for others Similar to primary level, the expected sign is positive uppsec (+)
Upper secondary level is a dummy variable; uppsec=1 for graduated upper secondary education and no vocational education; 0 for others The expected sign is positive voc (+)
Vocational education is classified as a dummy variable, where a value of 1 indicates completion of graduated vocational education—encompassing vocational primary, vocational intermediate, secondary professional, vocational colleges, and general education that is below the level of colleges Conversely, a value of 0 represents all other educational backgrounds The anticipated outcome suggests a positive correlation with university education.
University level is a dummy variable; univ=1 for graduated college level and above; 0 for others The expected sign is positive lnhrwrk (+)
The log of hours worked per week is a continuous variable that reflects the total hours an individual works weekly This variable serves as a compensatory factor, with an anticipated positive sign, indicating that increased hours worked positively influence the returns on education.
RESEARCH FINDINGS AND DISCUSSION
Descriptive Statistics
4.1.1 Distribution of the Dependent and Explanatory Variables
To effectively analyze sample data, it is essential to visually explore it using histograms, which are particularly useful for assessing normal distribution The subsequent figures illustrate the distribution of both dependent and explanatory variables, with Figure 4.1 showcasing histograms of log earnings categorized by gender.
F re qu en cy log of monthly earnings
Graphs by gender=1 if Male
Source: Author’s calculation using data from VHLSS 2008
Figure 4.1 draws histograms of log of earnings for male, female, and for all sample All of them are following normal distribution
Figure 4.2: Histograms of log of earnings (by sector)
0 5 10 0 5 10 public sector private sector foreign investment Total
F re qu en cy log of monthly earnings
Source: Author’s calculation using data from VHLSS 2008
Figure 4.2 also reveals histograms of log of earnings but separated by sector, specifically by public, private, and foreign sectors All are following normal distribution
Figure 4.3: Histograms of years of schooling and log of hours worked/week
0 5 10 15 20 the number of years schooling
0 1 2 3 4 5 log of hours worked/week
Source: Author’s calculation using data from VHLSS 2008
Figure 4.3 illustrates the distribution of years of schooling alongside the logarithm of hours worked per week, both exhibiting a normal distribution Notably, the histogram for years of schooling reveals significant peaks at 9, 12, and 17 years, which correspond to lower secondary, upper secondary, and university education levels, respectively.
To explore possible relationships/patterns between variables, scatterplots are good instruments to conduct
Figure 4.4: Scatterplots of monthly earnings and years of schooling
0 5 10 15 20 the number of years schooling log of monthly earnings Fitted values
Source: Author’s calculation using data from VHLSS 2008
Figure 4.4 illustrates a positive correlation between the logarithm of monthly earnings and years of schooling, indicating that an increase in educational attainment is associated with higher monthly income.
Figure 4.5 illustrates a positive correlation between monthly earnings and education levels, indicating that individuals with higher education attain greater financial returns.
Figure 4.5: Scatterplots of monthly earnings and education levels
No level PRIM LOWSEC UPPSEC VOC UNIV education levels log of monthly earnings Fitted values
Source: Author’s calculation using data from VHLSS 2008
Figure 4.6: Scatterplots of monthly earnings and years of experience
0 20 40 60 years of experience log of monthly earnings Fitted values
Source: Author’s calculation using data from VHLSS 2008
The relationship between monthly earnings and years of experience is not linear, as indicated by Figure 4.6, which shows a quadratic correlation This suggests that while earnings tend to rise with increased experience, the rate of growth diminishes over time Consequently, this finding is crucial for developing regression equations that incorporate squared years of experience.
4.1.2 Descriptive Statistics of the Dataset
Table 4.1 provides the descriptive statistics of the dataset The mean age of the sample is 34 years, higher than in 1992/1993 sample of Moock et al (2003) which is
31 years The mean age of males is 34.6 years, slightly higher than of females 33 years
The average years of schooling stand at 9.5, compared to 7.9 years reported by Moock et al (2003) Educational attainment is nearly equal for both genders, with males averaging 9.34 years and females slightly higher at 9.77 years.
Approximately 13% of the workforce is either uneducated or illiterate, a figure that rises to 22% according to Moock et al (2003) Additionally, 21% have completed primary education (50% in Moock et al (2003)), while 25% have attained lower secondary education Only 14% have completed upper secondary education, contrasting with just 8% in Moock et al (2003) Furthermore, 13% have pursued vocational training (12% in Moock et al (2003)), and 14% have achieved college education or higher, compared to 7% in Moock et al (2003).
Over the past 15 years, Vietnam has experienced a notable shift in labor force distribution, with the private sector now employing 61% of the workforce, up from 58% reported by Moock et al (2003) In contrast, the public sector's share has decreased to 33% from 42% in the same study Additionally, the foreign sector has emerged, employing 6% of the labor force, a significant increase from the previous figure of zero noted by Moock et al (2003).
8 Till Nov 12, 1996 did the Law on Foreign Investment in Vietnam be first adopted to allow foreigners to invest in Vietnamese market
The mean monthly earnings in Vietnam are approximately VND 1,428,000 (USD 68.84), a significant increase from VND 152,000 (USD 14) reported in Moock et al (2003) Notably, male earnings average VND 1,513,460, which is about 17% higher than female earnings at VND 1,293,880 This indicates a narrowing of the gender earnings gap over the past 15 years, as it was 40% higher in the earlier study, reflecting positive trends in Vietnamese economic growth Additionally, the average workweek has slightly increased to 47 hours, compared to 46 hours in 2003.
Source: Author’s calculation using data from VHLSS 2008
Regression Results
Table 4.2 presents the results of estimating simple Mincerian earnings functions based on years of schooling, indicating that the average private rate of return for each additional year of education is approximately 9% This figure stands in contrast to the 5% return reported by Moock et al.
9 Exchange rate year 2008: 1USD = 16,977 VND (source: ADB)
10 Exchange rate year 1993: 1USD = 10,857 VND (source: ADB)
(2003), nearly double during the last 15 years Females enjoy higher returns to school than males (11.47% vs 8.33%) This pattern unchanged when comparing with Moock et al (2003), whereas 6.8% vs 3.4%
Table 4.2: Earnings function by years of schooling
Log of hours worked/week 0.6480 **** 0.6784 **** 0.6043 ****
Number of groups 4,335 3,510 2,337 sigma_u 0.41 0.39 0.37 sigma_e 0.55 0.53 0.57 rho 0.36 0.35 0.29
Note: * significant at 10% level , **significant at 5% level, *** significant at 1% level, **** significant at 0.1% level
Source: Author’s calculation using data from VHLSS 2008
Post-schooling investment yields an average return of 7.31% for each additional year of experience, an increase from the 6.4% reported by Moock et al (2003) Notably, the returns for men and women are quite similar, with men achieving a return of 7.25% and women 7.01%.
The intraclass correlation coefficient, rho = 0.36, indicates the level of correlation among individuals within a household If this correlation were to drop to zero, it would imply that household clustering is insignificant, allowing for the use of simple regression without the need for random effects In this scenario, there would be no distinction between Ordinary Least Squares (OLS) and Random Effects (RE) estimators.
An intraclass correlation close to 1 suggests no individual variation, indicating uniformity among individuals In this case, a rho value of 0.36 shows a correlation among individuals within a household, making the random effects (RE) estimator a more suitable choice.
The standard deviation of unobserved household characteristics, denoted as sigma_u, is 0.41, which represents the intercept's standard deviation at the household level In contrast, sigma_e, with a value of 0.55, signifies the standard deviation of residuals at the individual level The intraclass correlation, rho, can be computed using these two standard deviations, sigma_u and sigma_e.
e sigma u sigma u sigma n correlatio Intraclass
The standard deviation (σ_u = 0.41) indicates that, in addition to a common intercept of Constant = 2.6296 (Table 4.2), there exists a household-level component following a normal distribution with a standard deviation of 0.41 This suggests that individuals from various households exhibit different intercepts, which are normally distributed with a mean of 2.6296 and a standard deviation of 0.41.
Table 4.3: Earnings function by sector of employment
Log of hours worked/week 0.5891 **** 0.7117 **** 0.5898 ****
Number of groups 1,700 2,816 337 sigma_u 0.47 0.42 0.43 sigma_e 0.47 0.55 0.43 rho 0.50 0.37 0.50
Note: * significant at 10% level , **significant at 5% level, *** significant at 1% level, **** significant at 0.1% level
Source: Author’s calculation using data from VHLSS 2008
Employees in the public sector tend to have higher returns on education, with a rate of 9.95%, compared to 5.59% in the private sector However, the foreign sector boasts the highest rate of return, at 11.9% This trend is consistent with previous findings, such as Moock et al (2003), which reported returns of 6.2% and 3.9% for public and private sector workers, respectively, although the foreign sector was not accounted for due to its relatively small size at the time.
Post-schooling investment yields varying rates of return across different sectors, with the foreign sector leading at 8.15%, followed closely by the public sector at 8%, and the private sector trailing at 6.38% Compared to 15 years ago, when returns were 4.6% for the public sector and 7.2% for the private sector, these figures indicate a significant shift in the market Notably, the foreign sector has emerged as the most lucrative option for post-schooling investment, while the public and private sectors have swapped their previous standings.
Table 4.4: Earnings function with schooling levels (for all, males, and females)
Log of hours worked/week 0.6856 **** 0.7178 **** 0.6540 ****
Number of groups 4,335 3,510 2,337 sigma_u 0.39 0.36 0.35 sigma_e 0.55 0.53 0.57 rho 0.33 0.31 0.28
Note: * significant at 10% level , **significant at 5% level, *** significant at 1% level, **** significant at 0.1% level
Source: Author’s calculation using data from VHLSS 2008
Dummy variables derived from years of schooling are used to assess the earnings premium associated with different education levels University graduates experience the highest return on their educational investment, earning 126% more than those with no education In comparison, vocational workers earn 77% more, while upper secondary and lower secondary level workers earn 50% and 26% more, respectively Additionally, primary-level workers receive a 16.21% earnings premium.
Table 4.4 presents the earnings premiums associated with various educational levels, using no education as the reference point Meanwhile, Table 4.5 illustrates the private rates of return to schooling across different educational levels, highlighting the advantages of flexible education comparisons The findings indicate that higher educational attainment correlates with significant earnings premiums, resulting in substantial private returns on investment in education (Moock et al.).
2003, p.507) The outcomes in Table 4.5 obtains from equation (3.3) and the earning function results disclosed in Table 4.4
Table 4.5: Private rates of return to schooling by level of education (%)
Educational level All Males Females
Primary (vs less than primary) 16 15 16
Upper secondary (vs lower secondary) 8 8 10
Source: Author’s calculation using data from VHLSS 2008
According to the results in Table 4.5, investing in a university education proves to be the most beneficial choice, yielding a 19% higher return compared to upper secondary education after four years of study Both male and female graduates experience equal returns of 20% at the university level This finding aligns with the conclusions of Moock et al (2003), which also identified university education as the optimal investment during that period Comparisons with Moock et al (2003) are made selectively due to variations in reference systems and educational level classifications.
Investing in primary education yields a notable 16% return, as supported by Moock et al (2003), making it the second most profitable option Both vocational and upper secondary education offer an 8% return, but vocational education requires 6 years to complete, compared to just 3 years for upper secondary Given the similar return rates and shorter duration for upper secondary education, it emerges as a more advantageous investment choice.
Lower secondary education offers a minimal return on investment, with only a 2% overall rate For individuals with primary education, investing an additional four years to achieve lower secondary education results in just a 2% increase in returns However, by dedicating at least three more years to attain upper secondary education, they can potentially gain an additional 8% return on their investment.
Chapter Remarks
The sample profile typically exhibits variables that adhere to a normal distribution Bivariate analysis indicates a positive correlation between the logarithm of monthly earnings and both years of schooling and education levels Additionally, the relationship between the logarithm of monthly earnings and years of experience is characterized by a quadratic pattern.
The average age of the sample is 34 years, with an average of 9.5 years of schooling Approximately 13% of the labor force is illiterate or has no formal education, while 21% have completed primary education, 25% have finished lower secondary, and 14% each have completed upper secondary and higher education The public sector employs 33% of the labor force, compared to 61% in the private sector and only 6% in the foreign sector The average monthly earnings stand at around VND 1,428,000 (USD 84).
Multivariate analysis indicates that each additional year of schooling correlates with an 8.95% increase in the average return on education, with females benefiting more than males (11.47% compared to 8.33%) Workers in the foreign sector experience the highest returns at 11.9%, followed by the public sector at 9.95% and the private sector at 5.59% Investing in university and primary education proves to be advantageous, whereas lower secondary education yields relatively poor returns Additionally, upper secondary and vocational education represent a moderately good investment choice.
CONCLUSION AND POLICY RECOMMENDATION
Conclusion of the Study
Utilizing clustered household-level data derived from the VHLSS 2008 cross-sectional dataset and employing a Random-effects model to account for random household effects, this analysis estimates the rates of return to education in Vietnam through a Mincerian earnings function The findings reveal significant insights into the economic benefits of education within the Vietnamese context.
An additional year of schooling is linked to an 8.95% increase in the average return on education, nearly doubling from the 5% rate reported by Moock et al (2003) 15 years ago Notably, females benefit more from education than males, with returns of 11.47% compared to 8.33% This gender disparity in educational returns has remained consistent over the past 15 years, reflecting a shift from 6.8% for females and 3.4% for males.
Workers in the public sector experience higher returns on education, with a rate of 9.95%, compared to 5.59% in the private sector However, the foreign sector leads with the highest return at 11.9% Over the past 15 years, these rates have significantly increased from 6.2% in the public sector and 3.9% in the private sector, although the overall pattern remains consistent.
Higher levels of education correlate with increased rates of return on educational investment University graduates experience the highest return at 126%, followed by vocational workers at 77%, upper secondary workers at 50%, lower secondary workers at 26%, and primary-level laborers at 16.21% Compared to 15 years ago, these figures have significantly changed, with university returns rising from 43.7%, vocational from 20.7%, secondary from 32.5%, and primary from 13.4%.
Investing in university education yields the highest return, with a 19% increase compared to upper secondary education after four years of study In contrast, primary education offers a 16% return, significantly higher than the no level of education, which stood at 13% in 1992 Both upper secondary and vocational education provide an 8% return, while lower secondary education offers only a 2% return compared to primary education.
Policy Recommendation
The wage disparity between the public and foreign sectors makes it challenging for the public sector to retain and attract skilled workers To address this issue, the government should raise public sector wages to match those in the foreign sector By equalizing these wage rates, the public sector can not only retain talent but also enhance overall work efficiency, as employees will no longer be distracted by higher-paying opportunities elsewhere.
Higher education levels yield greater returns on investment, indicating potential for increased private financing, particularly in university and upper secondary education If the government reallocates some educational costs to individuals and families, it is unlikely to diminish the incentive for investing in these higher levels of education due to their high return rates Therefore, the government should promote the involvement of private financing institutions in education loans and establish favorable conditions for individuals to access preferential loan options.
The return rates for vocational and upper secondary education are both 8%, presenting an appealing alternative for students deciding their paths after lower secondary school This finding is significant for policymakers focused on aligning employment opportunities across various educational levels, as well as for career advisers who guide lower-secondary graduates based on their skills, abilities, and market demand.
Limitations of the Study
A limitation of this study is the potential endogeneity in the empirical models (3.5) and (3.6), where the error term may include factors beyond years of schooling that affect earnings, such as innate ability Individuals with varying levels of innate ability may correlate with differing years of schooling, making it difficult to determine the extent to which increased earnings are attributable to education versus individual ability Consequently, years of schooling is treated as an endogenous variable While including controls for ability, such as IQ tests, could address this endogeneity issue, the necessary data is not available in the current dataset.
An alternative method for addressing endogeneity is the Instrument Variable (IV) approach; however, this research does not utilize IV due to the difficulty in obtaining an appropriate instrument variable from the Vietnam Household Living Standards Survey (VHLSS) Instead, the study focuses on household-level effects rather than individual-level effects, employing a random-effect estimator to assign these household-level impacts to individuals across different households Furthermore, there is concern that the coefficient for years of experience may be overestimated, particularly for women who frequently leave the workforce temporarily for reasons such as pregnancy or child-rearing, as the VHLSS survey does not adequately capture the duration of time spent out of the labor force.
A significant limitation of the study is its focus solely on wage earners, neglecting the fact that over 80% of the Vietnamese labor force is engaged in farm and non-farm self-employment Including this substantial segment could lead to markedly different findings and insights.
Suggestion for further Studies 48 REFERENCE
To address the limitations identified in this study, I recommend utilizing the Instrumental Variables (IV) approach in future research This method can help mitigate unobserved heterogeneity by identifying instrument variables that are correlated with years of schooling yet uncorrelated with the error term, ensuring they have no direct impact on earnings.
For further research, it is recommended to utilize multi-level mixed-effect models that are appropriate for VHLSS data, which is organized into multiple categories such as households, villages, communes, and provinces These models enable the examination of effects that differ across groups, as they randomly assign variations to both intercepts and slopes, allowing for distinct intercepts and slopes for different groups.
Angrist, J.D and Krueger, A.B (1991) Does compulsory school attendance affect schooling and earnings? Quarterly Journal of Economics 106 (4): 979–1014
Ashenfelter, O and Krueger, A.B (1994) Estimates of the economic return to schooling from a new sample of twins American Economic Review 84 (5): 1157–1173
Ashenfelter, O and Rouse, C (1998) Income, schooling and ability: Evidence from a new sample of identical twins Quarterly Journal of Economics 113 (1): 253-284
Cameron, C & Trivedi, P (2009) Microeconometrics Using Stata College Station TX: Stata Press
Card, D (1999) The causal effect of education on earnings, on Orley Ashenfelter and David Card (eds.) Handbook of Labor Economics Amsterdam: Elsevier Press, pp 1801-1863
Glick, P., & Sahn, D E (2000) Schooling of girls and boys in a West African country: the effects of parental education, income and household structure Educations of Education review 19 (2000) 63-87
Johnson, E.N and Chow, G.C (1997) Rates of return to schooling in China Pacific Economic Review, 2:2, 101-113
Doan, T & Gibson, J (2010) Return to schooling in Vietnam during economic transition: Does the return reach its peak? MPRA Paper 24984, University Library of Munich, Germany
Hayden Martin (2005) The Legislative and Regulatory Environment of Higher Education in Vietnam Washington, DC: The World Bank
Heckman, J (1979) Sample selection bias as a specification error Econometrica 47:153-61
General Statistic Office (GSO), (2008) Result of the Vietnam household living standards survey 2008 Retrieved Jun 20, 2011 from http://www.gso.gov.vn/default_en.aspx?tabidQ5&idmid=5&ItemID47
Glewwe, P & Patrinos, H.A (1998) The Role of the Private Sector in Education in Vietnam: Evidence from the Vietnam Living standards survey World Bank - Living Standards Measurement Papers No 132
Gallup, J (2002) The wage labor market and inequality in Vietnam in the 1990s World Bank Policy Research Working paper No 2896
Le Anh Khang (2012) Determinants of secondary school dropout in Vietnam: A panel data edidence Unpublished Master Thesis University of Economics, HCMC Vietnam-Netherlands programme for M.A in development economics
Le Thi Nhat Phuong (2008) Determinants of dropping out of school: the case of Vietnam Unpublished Master thesis Kansas State University Manhattan, Kansas
Liu, A.Y (2006) Changing wage structure and education in Vietnam, 1992–98 The roles of demand Economics of Transition vol 14, issue 4, p 681-706
McCarty, A (1999) Vietnam's Labour market in transition Presented at "Law and Labour market regulation in Asia" conference, University of the Phillippines
Mincer, J (1974) Schooling, experience and earnings New York: National Bureau of Economic Research
Moock, P.R., Patrinos, H.A., and Venkataraman, M (2003) Education and earnings in a transition economy: the case of Vietnam Economics of Education Review 22, 503–510
Nakamuro, M and Inui, T (2012), Estimating the returns to education using a sample of twins – The case of Japan RIETI Discussion paper series 12-E-076
Nguyen Xuan Thanh (2006) Estimating the rate of return to schooling in VietNam: A difference-in-difference approach Fulbright Economics Teaching Program Research paper
Onphanhdala, P and Suruga, T (2007) Education and Earnings in Transition: The Case of Lao Asian Economic Journal, Vol 21 No 4, 405–424
Phan, D.T (2000) Sectoral job choice and rewards in the Vietnamese labour market PhD dissertation, Economics of Development, National Center for Development Studies, Australian National University, Canberra
Princeton University Multilevel Analysis (Ver 1.0) Retrieved Dec 05, 2012 from http://dss.princeton.edu/training/
Psacharopolous, G (1994) Returns to investment in education: A global update World Development, 22(9), 1325-1343
Psacharopoulos, G and Partrinos, H.A (2002) Returns to investment in education: A further update World Bank Policy Research Paper 2881 World Bank, Washington
Siphambe, H.K (2000) Rates of return to education in Botswana Economics of Education Review 19 (2000) 291–300
Siphambe, H.K (2008) Rates of return to education in Botswana: Results from the 2002/2003 household income and expenditure survey data set South African Journal of Economics Vol 76:4 December 2008
Polacheck, S (2007) Earnings over the Lifecycle: The Mincer Earnings Function and its Application IZA Discussion Paper No 3181
Trostel, P., Walker, I., and Woodley, P (2002) Estimates of the Economic Return to Schooling for 28 Countries Labour Economic, 9(1): 1-16.