7  Endogenous regressors in the experience and wage regressions

To this point in the analysis, education and training outcomes have been treated as an exogenous set of indicator variables in both the work experience and wage regressions. We have assumed that only the receipt of formal qualifications (ie, School Certificate, 6^th Form or Higher School Certificate, University Bursary, University Diploma or Degree, or Vocational Qualification) were important for the accumulation of work experience and the determination of hourly earnings. Holding these qualifications constant, any additional time spent in study or training would, on average, reduce the accumulation of work experience and thereby indirectly lower the wage rate faced in the labour market.

In this section, we experiment by measuring education and training as a single, continuous variable in terms of the amount of time spent in human capital investment. We also treat education and training as an endogenous variable in the estimation of both the work experience and wage rate regressions. As suggested by a referee in an earlier draft of this report, data on the sex-composition of the siblings in the families of the CHDS children are used as instrumental variables for this purpose. Finally, we also ‘endogenise’ work experience as a determinant of hourly earnings regression.

7.1  Education as an exogenous, continuous determinant of experience

We start by re-estimating the OLS regressions for work experience and hourly earnings using a single variable for the total time spent in education and training to age 21. These are essentially more restricted specifications of the regressions discussed in previous sections. In the work experience regressions, time spent in education was found to have offsetting effects on this accumulation process. On the one hand, more time spent in education meant less time available for work (a negative impact on the accumulation of work experience). On the other hand, the attainment of formal qualifications meant more employment opportunities and stability (a positive impact on the accumulation of work experience). Similar offsetting forces existed in the hourly earnings regression. Time spent in education directly raised wages through formal qualifications, but indirectly lowered wages by reducing work experience.

To simplify this discussion, we report only the results from the regressions using the two-way split in youth ethnicity. The first two columns in Table 18 display the results from re-estimating the OLS regressions reported in Table 10, where the variable ‘Effective Years of Education and Training’ essentially replaces the six previous variables (‘Years Not Enrolled in Education and Training’, ‘School Certificate’, ‘6^th Form or Higher School Certificate’, ‘University Bursary’, ‘University Diploma or Degree’ and ‘Vocational Qualification’). As expected, the estimated coefficients on this new explanatory variable are negative in both the short and long regressions. The net effect of additional time spent in education and training is to reduce, on average, the accumulation of work experience between ages 16 and 21. The estimated coefficients of –0.292 and –0.352 suggest that every year of effective full-time study or training reduces work experience by between 3.5 and 4.2 months (–0.292•12 and –0.352•12). Both effects are significantly different from zero at better than a 1% level.

Collapsing this information on education and training into a single explanatory variable has little impact on the estimated effects of ethnicity on the accumulation of work experience. Recall from Table 10 that the estimated coefficients on this variable were –0.247 and –0.169 in the short and long regressions, respectively. Only the former estimated coefficient was statistically significant (at better than a 2.5% level). The estimated coefficients on the same variable in the short and long regressions, reported in the first two columns of Table 13, are –0.306 and –0.181. Only the former is statistically significant (at better than a 1% level). Although the effects of ethnicity increase slightly in magnitude in this new specification, these differences are not statistically significant.

Overall, the explanatory power of these new regression specifications declines when we collapse the effects of education and training into a single variable. The R² statistics were 0.299 and 0.346 on the short and long regressions in Table 10, respectively. The new R² statistics are 0.194 and 0.273 in similar OLS regressions in Table 13. Moreover, the adjusted R² statistics decline with these new specifications. It appears that the more flexible specification used originally, which included both the time spent in education and training and the qualifications gained from these activities, can be justified on this dimension.