Economics of Education Review 22 (2003) 99–107 www.elsevier.com/locate/econedurev

Vocational education in Thailand: a study of choice and returns Thammarak Moenjak b

a,*

, Christopher Worswick

b

a Monetary Policy Group, Bank of Thailand, 273 Samsen Rd., Pranakorn, Bangkok 10300, Thailand Department of Economics, Carleton University, 1125 Colonel by Drive, Ottawa, Ont. K1S 5B6, Canada

Received 18 February 2000; received in revised form 5 June 2001; accepted 30 August 2001

Abstract This study adds new evidence to the debate on the relative benefits of upper secondary vocational education and of general education at the same level. Using a probit model, the study finds that an individual from a well-to-do family is more likely to undertake vocational education. After correcting for possible self-selection, the study also finds vocational education to give higher earnings returns than general education does. These findings call into question the belief that vocational education has been overvalued and that providing general education to the workforce followed by on-thejob training would provide more benefits. Indeed, the study suggests that an investment to improve the access to vocational education might prove more beneficial.  2001 Elsevier Science Ltd. All rights reserved. JEL classification: I20; I21; I28 Keywords: Rate of return; School choice

1. Introduction This paper aims to identify the factors that influence an individual’s choice between vocational and general education in Thailand as well as the relative returns between these two types of education at the upper secondary level. Since the 1970s there have been various studies that examined an individual’s choice and the relative returns to each type of program. The findings, especially with respect to the relative returns, however, have been inconclusive. Hollenbeck (1993) found the returns to post-secondary vocational education in 1972 in the United States to be significant for women but not for men. Trost and Lee (1984), on the other hand, found that in the United States in 1973, technical education at the upper secondary

* Corresponding author. E-mail address: [email protected] (T. Moenjak).

vocational education level gives higher returns than does upper secondary general education at least for men. (Trost and Lee (1984) did not evaluate the returns to technical education for women.) To investigate the factors that influence an individual’s choice between the two types of education, the paper first estimates a probit model of education choice before proceeding to estimate the relative returns to the two types of education using a self-selection corrected earnings model. The results from the analysis should provide interesting insights with regard to vocational education in an industrializing country. 1.1. Overview of Thailand’s secondary education Throughout Thailand, primary education is compulsory. Individuals are required to enroll in a primary school for at least 6 years from age 7 (Office of the National Education Commission, 1997a). Secondary education, on the other hand, is non-compulsory and is

0272-7757/02/$ - see front matter  2001 Elsevier Science Ltd. All rights reserved. doi:10.1016/S0272-7757(01)00059-0

100

T. Moenjak, C. Worswick / Economics of Education Review 22 (2003) 99–107

divided into two levels. Lower secondary education covers grades 7 to 9. Lower secondary education aims to equip students with literary and numerical skills that would make them competent workers in non-specialized, non-manual jobs, or prepare them for further education levels. Students must complete lower secondary education before continuing on to upper secondary education (grade 10 to grade 12). At the upper secondary education level, students can choose between two streams of education; namely, the general (or academic) stream and the vocational (or technical) education stream. Access to both streams of upper secondary education is rather open, with similar admission criteria.1 The number of places available for general education, however, exceeds that for technical education (Office of the National Education Commission, 1997a). Regardless of their choice, once they finished grade 12, the graduates from both streams are considered equally qualified to sit for national university examinations (Office of the National Education Commission, 1997a). Upper secondary general education, other than preparing students for further education, also aims to equip the students with basic skills that would make them competent workers in general non-manual jobs such as office workers. Upper secondary vocational education, on the other hand, while also preparing students for technical studies in tertiary institutions, aims to equip them with specialised technical skills that are deemed essentials for jobs that require relatively highly specialised skills such as those for mechanics, electricians, as well as business skills such as bookkeeping (Office of the National Education Commission, 1997a).2

2. The data For the estimation of the probit choice model and the earnings equation, the paper uses data from Thailand’s Labor Force Survey for the years 1989 to 1995 inclusive. The survey is conducted annually by the National Stat-

1 The costs of upper secondary vocational education are on average higher than those of general education at the same level are. In this study, however, we do not have information on the exact costs of study for each of the individuals. The estimation of the returns to education, hence, is performed with the conventional assumption that, on average, student earnings plus scholarships roughly paid for tuition. (See Leibowitz, 1974; Mincer, 1974, for details.) 2 Some vocational schools also offer agricultural studies. The proportions of vocational school students enrolled in agricultural studies, however, have always been very small. In 1994, for example, less than 5% of vocational school students enrolled in agricultural studies (Office of the National Education Commission, 1997b).

istical Office of Thailand. The sample for each Labor Force Survey is drawn randomly from different households throughout the country. Records of the questionnaires available to the public start from the survey year 1989. In the analysis, the sample is pooled from the records of those individuals who were not attending school and were between 15 and 60 years of age at the time of the survey. The sample used for the first-step probit analysis is restricted to those individuals who reported having upper secondary vocational or upper secondary general education as the highest level of education, and whose relevant information on parental education and parental occupation is not missing. For the second-step evaluation of the individuals’ earnings the sample is further restricted to those who reported their earnings and were employed at the time of the survey. Table 1 shows the descriptive statistics of our sample. Note that for our analysis, the sample is restricted to individuals who are sons or daughters of the household heads. Data from Thailand’s Labor Force Survey do not explicitly contain information on parents. Information on parents is important in our analysis as it could reflect the individual’s socioeconomic status, a possible determinant of his/her education choice. To obtain parental information, observations of children are thus matched with those of the individuals from the same household who reported having a parents–children relationship. (The only parents–children relationship reported in the Labor Force Survey is that of the household heads, household head’s spouse and their children.) The resulting sample used in this paper, hence, consists of only those who are the sons or daughters of the household heads. Individuals who did not live with their parents at the time of the surveys are completely excluded from our estimation. Such exclusion suggests that our sample is a selected group, especially if those who live with parents tend to be those who cannot afford to live apart on their earnings. While the possible sample-selection issue is undeniably real, it should not, however, seriously affect the ability of our analysis to address the main questions of the paper. First, since one of our main concern is to evaluate returns to vocational education as compared to general education at the upper secondary level, our sample is large enough to give a good idea of the relative returns between these two types of education at that level, at least for those in the survey who were living with parents. In any case, there is no reason to expect that the relative returns to these types of education would be different for those who do and those who do not live with their parents. Even if we assume that an individual lives with his parents only if he/she can not afford to live on his/her earning, there is no a priori reason to believe why, ceteris paribus, an individual with a particular type of education is less likely to be able to afford

T. Moenjak, C. Worswick / Economics of Education Review 22 (2003) 99–107

Table 1 Descriptive statistics. Sample means and standard deviations (standard deviations in parentheses) Variable

Males

Females

Sample size (N) Education attainment Vocational education

2633

2252

0.435 (0.495)

0.444 (0.497)

5.718 (4.574)

5.268 (4.454)

18.890 (13.052)

17.716 (14.379)

0.165 (0.372) 0.227 (0.419) 0.319 (0.466) 0.123 (0.328)

0.150 (0.358) 0.235 (0.424) 0.324 (0.498) 0.130 (0.337)

0.437 (0.496) 0.298 (0.457)

0.460 (0.498) 0.291 (0.454)

0.222 (0.415) 0.011 (0.104) 0.279 (0.449)

0.265 (0.441) 0.038 (0.192) 0.246 (0.430)

0.366 (0.481)

0.398 (0.489)

0.688 (0.463)

0.666 (0.471)

0.384 (0.486) 0.332 (0.471) 0.068 (0.252)

0.385 (0.486) 0.332 (0.471) 0.088 (0.284)

Work characteristics Experience Hourly earnings Baht Region of residence North Northeast Middle Bangkok Area of residence Municipal area Sanitary district Marital status Married Divorced, separated, widowed Migration status Migrant No. of family members More than five Survey year after 1991 After 1991 Birth cohort B1965–69 B1970–74 B1975 AND AFTER

living out on his/her own. Indeed, our descriptive statistics below show that, for both men and women, vocational school and general school graduates are represented roughly equally in our sample. This relatively even distribution reflects the situation in the total survey population. The reason presented here also applies to the fact that the other main concern of the paper is to evalu-

101

ate the individual’s relative probabilities of attaining one type of education over another. Second, as argued by Bedi and Gaston (1997), given the prevalence of ‘joint’ families in developing countries, the individuals may simply live with their parents out of convenience in location or because they are participating in family enterprises.3 Unlike western countries where nucleus families are more widespread, ‘joint’ families in Thailand remain relatively common. Individuals may live with parents even after they are married and have children of their own. Since the Labor Force Survey do not indicate who are the ‘breadwinners’ of the families, it is premature to assume that all the individuals living with their parents are still subsidized by their parents incomes (and all those who are not living with their parents are not). Cultural norm dictates that many of the individuals in our sample could actually be providing for their parents (who may have long retired, since the maximum individual’s own age in our sample reaches 60). With the two reasons given above, there is no a priori reason to believe that estimation results of the paper would not at least partly reflect the relative probabilities of the individual’s education choice and the relative returns to the different types of education for those of the general population. Admittedly, until large-scale data that include parental information become available for Thailand, this sample selection issue remains real.

3. Theoretical model According to Mincer (1974), earnings determinants could be estimated using the following functional form: ln wi⫽b⬘Xi⫹dVi⫹ei.

(1)

where lnwi is the log of hourly earnings of individual i; Xi is a vector of personal, and background characteristics of individual i, Vi is a vector of the individual i’s educational attainment, and ei is the random disturbance term. Griliches (1977), however, pointed out that the coefficient estimates of the OLS estimation of the classical model could suffer from what is now commonly known as ‘self-selection bias’. When the individual’s background and ability ‘dictate’ his/her educational attainment, the individual is said to be ‘self-selected’ into that particular educational attainment. If educational attainment of an individual is at least partly determined by the individual’s abilities and family backgrounds, estimating the classical earnings equation without taking into

3 An example of a ‘joint family’ is a family where the families of the children of the household head reside together in a household.

102

T. Moenjak, C. Worswick / Economics of Education Review 22 (2003) 99–107

account the possibility that family background and ability might influence educational attainment could give biased results. The significant magnitudes of coefficient estimates of education variables might at least be partly attributable to the individual’s background and ability rather than returns to education per se. For the investigation on the returns to vocational education in Thailand, we correct for the possibility that the estimated relative returns to education might at least be partly attributable to the factors that influenced the individual’s choice of education by using the standard twostage procedure with inverse Mills ratio.

4. Econometric specification 4.1. First stage probit choice model For the first stage probit choice model, our dependent variable takes a value of 1 if the individual’s choice of upper secondary education is a vocational one, and takes a value of 0 if upper secondary general education is chosen. The explanatory variables included in the vector Zi include parental education, parental occupation, region of residence, area of residence, and the number of household members. All of the explanatory variables, except number of household members take the form of zero– one dummy variables. For our estimation, father’s and mother’s educational attainment dummy variables are used as proxies for household socioeconomic status.4 In our context, we would expect higher parental educational attainment to imply higher socioeconomic status. Furthermore, in the case where ability is partially attributable to genetic markings, and that education attainment is partly attributable to individual’s ability, parental education variables also proxy for the individual’s ability. In our model, father’s and mother educational attainment dummy variables are entered separately. The father’s education dummy variable takes a value of one if father’s highest level of education attainment is higher than primary education, and zero otherwise. The same is true for mother’s education dummy variable. In their work, Behrman and Wolfe (1984), Chiswick (1986) and Heckman and Hotz (1986) all argued that mother’s education (rather than father’s) might have more significant 4 Note that the terms ‘father’ and ‘mother’ used in this thesis are actually ‘household head’ and ‘spouse of the household head’. The terms ‘father’ and ‘mother’ are used here because they are less cumbersome and more informative in our context. The sample used in this thesis consists of only the sons and the daughters of the household head. In any case, more than 90 percent of the household heads with spouse present in the data are males (and hence the ‘father’ term is not entirely inappropriate).

impact on the individual’s education decision, as it was often the mother who provided the learning environment for her child. Other than parental education, household head’s occupation (or ‘father’s’ in our context) can also reflect the socioeconomic status of the household. Here we classify the different occupations reported in the Labor Force Survey into three separate groups. White collar includes professional and managerial occupations. Blue collar and crafts includes clerical, sales, traders and craftsmen. Menial includes farmers, agricultural workers, and service providers. Socioeconomic status is assumed to rank in descending order from white collar, blue collar and crafts, and menial, respectively. Since certain individuals might have migrated between the time of education and the time of the survey, in the case of these individuals, the reported region/area of residence might not be the same as the actual region/area of upbringing. To account for this possibility, if the period of residing in the reported residence was less than the potential experience, we use the reported prior region and area of residence as the region and area of upbringing instead.5 4.2. Second stage earnings equation The dependent variable used is the log hourly earnings. Hourly earnings are used because other measures of earnings (for example, daily or monthly earnings) do not factor out the ‘labour supply’ effect. An individual might report having received higher monthly earnings simply because, ceteris paribus, he/she simply supplied more labour. Hourly earnings, hence, are a better measure for our estimation purpose. For our estimation, the log of hourly earnings is used as the dependent variable instead of hourly earnings because it reduces the effects of earnings outliers. To evaluate the returns to vocational education in Thailand, a dummy variable indicating if the individual has completed vocational education is included in the model. The dummy variable takes a value of one if the individual has completed upper secondary vocational education and a value of zero if the individual is an upper secondary general graduate. Experience variables are included in the model since, ceteris paribus, workers with more years of job experience are likely to earn more. (Higher experience is often associated with higher skills and higher productivity.) A firm is likely to use higher wages to induce experienced workers to stay on in their jobs, as the cost of training new workers could be very expensive. With the higher wages, the firm makes the workers’ costs of leaving the

5 Regions and areas of residence are classified according to those done in the Labor Force Survey.

T. Moenjak, C. Worswick / Economics of Education Review 22 (2003) 99–107

firm high. To capture the relationship between experience and earnings, we include two experience variables, namely potential experience and potential experience squared. The potential experience squared variable is included in addition to the potential experience variable to capture the possibility of a non-linear relationship between experience and earnings. Note here that for our estimation, potential experience is included instead of actual experience because the Labour Force Survey data we use reports neither the workers’ actual experience on their current job nor in their former jobs. To control for the relationship between experience and earnings, hence, we use potential experience variable (measured in years) as a proxy for the actual experience. Note that potential experience is defined here as the age reported at the time of the survey minus the age left school. Other dummy variables included in the model include those proxy for regions and areas of residence, marital status, migration status, birth cohort, and sample year.

Table 2 Probit maximum likelihood estimation of the decision to undertake vocational educationa Variable

Males

Females

Constant

⫺0.348** (0.107)

⫺0.192 (pvalue=0.11) (0.117)

0.253** (0.068)

0.111 (0.076)

0.046 (0.075)

0.011 (0.081)

0.187** (0.060) 0.272** (0.069)

0.182** (0.068) 0.296** (0.071)

0.116 (0.107) ⫺0.052 (0.099) 0.312** (0.085) 0.343** (0.106)

⫺0.269** (0.118) ⫺0.285** (0.106) 0.222** (0.092) 0.507** (0.116)

0.180** (0.072) 0.006 (0.073)

0.085 (0.078) ⫺0.138* (0.078)

⫺0.071 (0.053)

0.023 (0.056)

⫺0.003 (0.058)

⫺0.025 (0.061)

⫺0.294** (0.074) ⫺0.303** (0.080) ⫺0.559**

⫺0.218** (0.083) ⫺0.274** (0.088) ⫺0.193 (pvalue=0.112) (0.122) ⫺1452.148 189.829**

Father’s education More than primary Mother’s education More than primary Father’s occupation White collar Blue collar Region of upbringing North Northeast

5. Empirical results 5.1. First stage probit equation Table 2 presents maximum likelihood estimation results showing the relationship between background characteristics and the decision, for men and women, to undertake upper secondary vocational education versus general education at the same level. From Table 2, we can see that for both men and women, socioeconomic status as measured by father’s occupation, plays a significant role in influencing the individual’s choice of upper secondary education. Ceteris paribus, males and females from a higher economic status are more likely to undertake vocational education. For both men and women, as compared to having father in a menial occupation, having father in a white collar or a blue collar occupation significantly increases the probability of undertaking vocational education. With regard to father’s education, for men, father’s educational attainment being higher than primary level significantly increases the probability of undertaking vocational education. Such a result further confirms that, at least for men, ceteris paribus, an individual from a higher socioeconomic status is more likely to undertake upper secondary vocational education (as opposed to general education at the same level). For both men and women, being raised in a more prosperous region of the country (the middle region or Bangkok) significantly increases the probability of undertaking vocational education. For women, on the other hand, being raised in a poorer region (eg. the north or the north or the northeast) also decreases that probability.

103

Middle Bangkok Area of upbringing Municipal area Sanitary district No. of family’s members More than five Survey year After 1991 Birth cohort B1965–69 B1970–74 B1975 AND AFTER

Log likelihood c2 (d.f.=15) a

(0.124) ⫺1726.048 154.250**

**Significant at 5% level; *significant at 10% level.

Looking at the cohort effects, it appears that vocational education has become less popular for individuals from the more recent cohorts. For both men and women, being born after 1969 significantly decreases the probability of undertaking vocational education, ceteris paribus. The coefficient estimates of the dummy vari-

104

T. Moenjak, C. Worswick / Economics of Education Review 22 (2003) 99–107

ables for 1970–1974 and 1975 and after are all negative and significant for both men and women. At the first glance, these results for men and women seem to suggest that, while vocational education has been advocated by the government as a crucial provider of skills for the industrialisation of the country, vocational education itself has become increasingly less popular among the new cohorts of population. On the other hand, our estimates here might have neglected the fact that increased access to higher education could have led more vocational school graduates to undertake tertiary education. Since the information on those vocational school graduates who continued on to higher education is not available in the Labour Force Surveys (and hence our sample), we cannot further investigate this point. 5.2. Second stage earnings equation From Table 3, we can see that, for both men and women, upper secondary vocational education gives significantly higher earning returns than does upper secondary general education. Indeed, as compared to general education at the same level, upper secondary vocational education gives higher earnings returns by 63.9 percent for men and 49.4 percent for women. Possible explanations for such results include compensating wage differentials as well as increased demand for the skills provided by upper secondary vocational education due to the industrialisation process. According to compensating wage differential story, the nature of the jobs normally taken by upper secondary vocational education graduates might differ from those normally taken by their upper secondary general education counterparts in such a way that the employers are willing to pay the vocational school graduates more. (For example, jobs normally taken by vocational school graduates such as, say, mechanics, may be more physically demanding, more hazardous, and more specialised that the employers are willing to pay wage premiums to their employees.) On the other hand, with the rapid industrialisation process going on in the economy, employers in the industries may be willing to pay wage premiums to retain the workers with the technical skills provided by vocational education. For both men and women, the concavity of the relationship between potential experience and earnings is not confirmed. For both men and women, while the parameter estimates of the potential experience dummy variables are positive and significant at the one percent level, the potential experience squared variables, though negative, are statistically insignificant. Our results here differ from the results found by earlier studies on the returns to education (see, for examples, Bellew and Moock, 1990; Tansel, 1994; Gaston and Sturm, 1991). One possible explanation for the non-concavity in our

case could be the inclusion of the birth cohort variables in our estimation. The way we define them, birth cohort variables are highly correlated with potential experience and hence could take out some of the explanatory variables of the potential experience squared variables. Also, since our sample includes many of the relatively young individuals (those still living with parents) the diminishing effects of the returns to experience might still be relatively weak. With regard to self-selection effect, for both men and women, the parameter estimates of the self-selection correction terms (hi) are negative. For men, the parameter estimate is significant at the 10 percent level. For women, the p-value of the estimate is at 10.6 percent level. The negative signs of the dummy variables suggest that if the general education graduates had indeed chosen vocational education, they are not likely to earn less than what the vocational education graduates earn, ceteris paribus. Table 3 also shows the results from the OLS estimation of Mincer’s classical earnings model. The main difference between the results from the OLS estimation and the self-selection correction model lie in the differences in the estimates of the relative returns to upper secondary vocational education as compared to general education at the same level. According to the OLS estimates of the Mincer earnings model, as compared to general education at the same level, upper secondary vocational education gives higher earnings returns by 23.8 percent for men and 20.7 percent for women (see Table 3). Both of the OLS estimates of the vocational education dummy variables are significant at the one percent level. After taking account of the possible self-selection, however, the returns to vocational education are found to be higher than those for general education by 63.9 percent for men and 49.4 percent for women respectively. The coefficient estimates of the vocational education dummy variables in the self-selection corrected model estimation are significant at the one percent level for both men and women. While the magnitudes of the estimated returns rise dramatically after the self-selection correction procedure, it must be kept in mind that our rather selected sample could play a role for such a rise. Until a more complete set of data becomes available, the dramatic rise in the magnitudes of the estimated relative returns to vocational education remains in question. However, with the high statistical significance of the vocational education dummy variables and the self-selection variables, it is still very reasonable to conclude that, for our analysis, after correcting for self-selection, returns to vocational education remain higher than those to general education. Other results from the OLS and the self-selection corrected estimations appear to agree in general. The signs and the magnitudes of all the other variables appear similar, even though some of significance levels of the para-

T. Moenjak, C. Worswick / Economics of Education Review 22 (2003) 99–107

105

Table 3 Estimated earnings equations (corrected standard errors in parentheses for self-selection corrected results, and normal standard errors for OLS results)a Variable

Males (self-selection corrected)

Males (OLS)

Females (self-selection corrected)

Females (OLS)

Constant

1.830** (0.098)

2.003** (0.102)

1.578** (0.124)

1.707** (0.108)

0.062** (0.009) ⫺0.0003 (0.0003)

0.609** (0.013) ⫺0.0003 (0.0004)

0.070** (0.012) ⫺0.0004 (0.0005)

0.068** (0.014) ⫺0.0003 (0.0005)

0.639** (0.146)

0.238** (0.028)

0.494** (0.180)

0.207** (0.031)

⫺0.017 (0.038) ⫺0.089** (0.039) 0.157** (0.036) 0.366** (0.043)

⫺0.005 (0.051) ⫺0.090* (0.053) 0.192** (0.045) 0.404** (0.054)

0.155** (0.053) 0.283** (0.052) 0.384** (0.045) 0.460** (0.059)

0.131** (0.057) ⫺0.262** (0.057) 0.403** (0.049) 0.508** (0.058)

⫺0.092** (0.032) 0.031 (0.031)

⫺0.055 (0.040) 0.038 (0.041)

⫺0.094** (0.039) ⫺0.007 (0.038)

⫺0.074* (0.042) 0.011 (0.043)

0.043* (0.026) ⫺0.244** (0.098)

0.036 (0.035) ⫺0.249* (0.137)

⫺0.003 (0.035) ⫺0.251** (0.074)

⫺0.003 (0.040) ⫺0.250** (0.086)

⫺0.111** (0.032)

⫺0.087** (0.042)

⫺0.057 (0.046)

⫺0.029 (0.049)

0.228** (0.032)

0.229** (0.043)

0.212** (0.041)

0.208** (0.046)

0.069 (0.045) 0.153** (0.065) 0.221** (0.096)

0.031 (0.057) 0.115 (0.087) 0.140 (0.126)

0.105* (0.061) 0.260** (0.087) 0.485** (0.116)

0.090 (0.068) 0.237** (0.098) 0.468** (0.132)

0.305

⫺0.178 (0.110) 0.294

0.292

Work characteristics Experience Experience squared Education attainment Vocational education Region of residence North Northeast Middle Bangkok Area of residence Municipal area Sanitary district Marital status Married Divorced, separated, widowed Migration status Migrant Survey year After 1991 Birth cohort B1965–69 B1970–74 B1975 AND AFTER Self-selection term hI Adjusted R-squared a

⫺0.251** (0.091) 0.307

**Significant at 5% level; *significant at 10% level.

106

T. Moenjak, C. Worswick / Economics of Education Review 22 (2003) 99–107

meter estimates of the marital status, and the birth cohort dummy variables change after we correct for the possible self-selection (see Table 3).

presented in Section 3). The unrestricted model (see Eq. (A2)), on the other hand, has an intercept dummy variable for the sons or daughters of household heads as well as its interactive terms with the other explanatory variables included in addition.

6. Conclusion

ln wi⫽bXi⫹ei.

(A1)

ln wi⫽bXi⫹ddi⫹g(di∗Xi)⫹ei

(A2)

The estimation results from the analysis above question the belief that vocational education has been overvalued and that providing general education to the workforce followed by on-the-job training would be more beneficial. After correcting for the possibility of selfselection and other factors such as experience, marital and migration status, the paper finds that, in Thailand, upper secondary vocational education gives statistically higher returns than general education at the same level. Such results imply vocational education underinvestment in Thailand. Furthermore, the findings from the probit estimation that an individual from a well-to-do family is more likely to undertake vocational education suggest inadequate access. More investment to improve the access to vocational education could thus prove beneficial. Further research into the issue, once a more complete set of data become available, should also provide more insights.

Acknowledgements The paper was written while the authors were at the University of Melbourne, Australia. The authors wish to thank Dr Robert Dixon, Dr Lisa Cameron, Dr Lata Ganghadaran, and Quy Tran for their helpful comments, suggestions, and assistance. Any error in this paper, however, is not theirs.

Appendix A. Sensitivity analysis determining whether we can make inferences about the general population As mentioned in the analysis above, our sample exclusion means caution is required when making inferences to the general population. Although, as mentioned in Section 2, there are theoretical reasons to believe that our results with respect to the relative returns to vocational education are at least partly applicable to the general population, for completeness a sensitivity analysis needs done to ascertain such applicability. For our sensitivity analysis, the restricted and unrestricted earnings models are estimated on the general population sample. The explanatory powers of the corresponding restricted and unrestricted models are then compared. In our case, a restricted model (see Eq. (A1)) has an identical econometric specification to that of the simple Mincer’s earnings model (the ‘naı¨ve OLS’ model

where ln wi is log hourly earnings, Xi is a vector of personal characteristics of the individual i, di is a dummy variable indicating whether the individual reported being a son, or a daughter of the household head (di=1) or not (di=0), b, d, and g are the vector of parameters pertaining to the vectors of explanatory variables to be estimated. To check whether the sons and daughters of household heads have different earnings structures from those of the general population, we check if the dummy variable di and its interactive terms are statistically significant as a group. If they are, the sample of the sons and daughters of the household heads used in the analytical chapters above has earnings structures that are statistically different from those of the general population. To check the significance of the dummy variable di and its interactive terms, an F-test could be performed to test the following restriction: d⫽g⫽0.

(A3)

The sensitivity analysis results suggest that our sample of the sons and daughters of household heads have earnings structures that are statistically different from those of the general population. For both men and women, the Fstatistics are larger than the critical value Fcrit.6 Any inference of the results from the earnings models above to the general population thus needs to be done cautiously. In any case, although the sensitivity analysis results are not ‘satisfactory’, the paper’s conclusions with respect to the relative returns between upper secondary vocational and upper secondary general education cannot be completely disregarded. The results of our analysis indeed may remain applicable to the general population for two reasons. First, for both men and women, the parameter estimates of the vocational education dummy variables are found to be positive and significant in both restricted and unrestricted models.7 For men, the estimate of the returns to education in the restricted model in our sensitivity analysis is extremely close to that found in the ‘naı¨ve’ OLS estimation of the earnings model of Section 3 above. (See Table 3 for the ‘naı¨ve’ OLS estimate.)

6 7

Fcrit=苲1.976, Male’s F-stat=2.65, Females; F-stat=3.046. Please contact the authors for the complete results.

T. Moenjak, C. Worswick / Economics of Education Review 22 (2003) 99–107

Second, for the unrestricted model, for men, the coefficient estimate of the interactive dummy variable indicating if an observation is that of a household head’s child with upper secondary vocational education is statistically insignificant. This statistical insignificance means that the relative returns between upper secondary vocational and upper secondary general education for a child of the household head are not statistically different from those for the general population. In the case of women, the coefficient estimate of the aforementioned interactive dummy variable is statistically significant, but with a negative sign. This means the returns to vocational education for the household head’s daughters understate the returns to vocational education for the general women population. With the two reasons cited above, it is arguable that the paper’s conclusion with respect to the relative returns to vocational education remains applicable to the general population. Until a more complete set of data becomes available for further research, however, the inference of the coefficient estimates of the from the self-selection corrected earnings models of this paper needs to be done cautiously.

References Bedi, A., & Gaston, N. (1997). Returns to endogeneous education: the case of Honduras. Applied Economics, 29, 519–528. Behrman, J. R., & Wolfe, B. L. (1984). The socioeconomic

107

impact of schooling in a developing country. Review of Economics and Statistics, 66, 296–303. Bellew, R., & Moock, P. (1990). Vocational and technical education in Peru. Economics of Education Review, 9 (4), 365–375. Chiswick, B. R. (1986). Comment on Hauser and Sewell. Journal of Labor Economics, 4 (3 (Part 2)), S116–S120. Gaston, N., & Sturm, R. (1991). Educational attainment and the returns to education for Australian youth: evidence of selfselection? Economic Analysis and Policy, 21 (1), 29–45. Griliches, Z. (1977). Estimating the returns to schooling: some econometric problems. Econometrica, 45, 1–22. Heckman, J. J., & Hotz, V. J. (1986). An investigation of the labor market earnings of Panamanian males: evaluating the sources of inequality. Journal of Human Resources, 21, 507–542. Hollenbeck, K. (1993). Postsecondary education as triage: returns to academic and technical programs. Economics of Education Review, 213–232. Leibowitz, A. (1974). Home investments in children. Journal of Political Economy, 82 (2), S111–S131. Mincer, J. (1974). Schooling, experience and earnings. New York: National Bureau of Economic Research. Office of the National Education Commission (1997a). Education in Thailand 1997. Bangkok, Thailand. Bureau of Educational System Development and Macro Planning. Office of the National Education Commission (1997b). The Eighth National Education Development Plan: 1997–2002. Bangkok, Thailand: Office of the Prime Minister. Tansel, A. (1994). Wage employment, earnings and returns to schooling for men and women in Turkey. Economics of Education Review, 13, 305–320. Trost, R. P., & Lee, L. F. (1984). Technical training and earnings: a polychotomous choice model with selectivity. The Review of Economics and Statistics, 66, 151–156.