7.2  Education as an endogenous determinant of experience

Our concern is that the time spent in education and training should be considered an endogenous variable in determination of subsequent work experience. Unmeasured factors may influence both human capital investment processes. These correlated disturbance terms would bias the estimated effects of time spent in education and training on the accumulation of work experience. The key is to find at least one instrumental variable, which influences education and training, but is directly unrelated to the accumulation of work experience.

One possibility is to use detailed information on the siblings of the youth in the CHDS. Butcher and Case (1994) used this family background information in the PSID to estimate rates of return to education. In particular, they claimed that the sex composition of siblings served as a valid instrument for the educational attainment of females. Women who were raised with sisters tended to acquire less education. Yet, the number and sex composition of siblings had little impact on the education of males.

The CHDS also contains excellent data on siblings. We know their number, gender and birth order. A regression specification that is even more general than the one used by Butcher and Case is used in this study. We regress the number of effective years of education and training of CHDS youth by age 21 against five variables that measure whether or not they were ‘only children’, and the numbers of younger and older male and female siblings. All five regressors are also interacted with the gender of the CHDS youth to capture any systematic differences in the influence of siblings on human capital investment between males and females.

The results of this ‘first stage’ estimation are disappointing. The R² statistic on this regression is 0.030, indicating that only 3% of the variation in effective years of education and training in our sample can be explained by these regressors. In particular, the number of siblings had a negative and significant effect on educational attainment, but the differences between their gender and birth order relative to the CHDS youth were unimportant. Unlike Butcher and Case, there is no statistical evidence in this regression that the sex composition of siblings serves as a better instrument among females. All of the estimated coefficients on the explanatory variables interacted with the gender of the CHDS youth were not statistically different from zero.

Despite these disappointing results from the first-stage estimation, the second-stage estimates from this Two-Stage, Least-Squares (2SLS) procedure are reported in the last two columns of Table 18. The predicted values replace the actual observations on ‘Effective Years of Education and Training’ in these regressions, and the appropriate standard errors are reported. The estimated coefficients on Education and Training declined in absolute magnitude relative to the earlier OLS results. Only the estimated parameter in the short regression is statistically different from zero.

** Significantly different from zero at 1% level.

* Significantly different from zero at 10% level.

Notes: These data are taken from the 973 respondents in the CHDS who provided valid information for the purposes of this study. The dependent variable is the effective years of work experience accumulated by the individual between the ages of 16 and 21. Youth are defined as ‘Maori’ in these regressions if they identify Maori as at least one of their ethnicities at age 21, and had at least one parental figure claiming Maori ethnicity by age 14 of the CHDS child. All other youth are considered to be ‘non-Maori’. Standard errors are in parentheses. The variable ‘Effective Years of Education and Training’ is treated as an endogenous variable under Two-Stage Least-Squares (2SLS) estimation in the last two columns. The instrumental variables include an indicator for the CHDS child being an only child, and the number of older and younger male and female siblings in the family. These five variables are also interacted with the gender of the CHDS child to allow for the impact of sex composition of siblings on educational attainment to vary by the gender of the CHDS child. See Butcher and Case (1994) for the use of a similar approach with panel data in the U.S. Using this flexible specifications, the R2 statistic from the regression of ‘Effective Years of Education and Training’ against these instrumental variables is only 0.030.

Little weight should probably be placed on these 2SLS results for a number of reasons. First, the instruments capture only a small proportion of the variation in this human capital investment. Second, due to the close relationship in time between this formal human capital investment and early work experience, these same family background factors might directly influence the accumulation of work experience by age 21.[24] Finally, endogenising years of education and training has little impact on the measured effects of ethnicity on the accumulation of work experience. The absolute values of the estimated coefficients on Maori under 2SLS are slightly smaller in the short regression, but slightly larger in the long regression compared to earlier OLS results. Once we control for measured family background differences there are no significant differences in the accumulation of work experience between Maori and non-Maori under both estimation procedures.