We start with the discussion that production of physical capital is different from the production of human capital especially in terms of capital intensities employed. We argue that a proper treatment of this aspect warrants no less than a full fledge two-sector model where all three factors of production physical capital, human capital and raw labour are used in both production technologies but in different steady-state ratios.
We then adopt an extremely simplified “two-sector” model, where only one factor of production, raw labour, is linearly converted to human capital. We see that this approach does not produce better point estimates of the structural form than one-sector approach adopted by MRW (1992). Moreover we find that the resulting regression equation, apart from some minor details, looks exactly the same for both approaches.
However, a slight improvement in overall fit of the model, indicated by and higher adj.R2 is seen. This improvement is even greater if we abstract from the cross-country variation in active-life expectancy. In any case the improvement can be associated with the variable we use to estimate the investment in human capital. We use share of labour force (s_H) attending school rather than just secondary enrolment ratios as done in MRW (1992). This result leads to the conclusion that our variable serves as a better proxy for the output invested in human capital than the SCHOOL variable. This holds even if we use (s_H / 1-s_H) instead of (s_H / s_Y) as a proxy, which theoretically are the same variable.
In the wake of these results we confront the criticism by Klenow and Rodriguez-Clare (1997) of the SCHOOL variable. They show that a more comprehensive SCHOOL, that also includes primary enrolment, explains only 11% of the variation in the data whereas TFP ends up explaining 67% of the variation. We reject their criticism in favour of MRW (1992). Firstly because they too only include primary and secondary enrolment rates and not tertiary. And secondly average years of schooling, that we use in our data, is an even more comprehensive measure of schooling. And it supports findings of MRW (1992).
We also discover that estimating the model using exact same samples as used in MRW (1992) but latest available data renders the results meaningless. But using both the updated samples 26 and latest data does not show any such characteristic and produce results almost identical to that of MRW (1992). Sampling is primarily based on oil and non-oil countries.
One possible explanation would be that some countries no longer belong to same sample as they did in 1985. To verify this we run the regression using only those countries that belong to same sample both in 1985 and in the updated sample. This reduces slightly the sample size but to our surprise does not produce any better results than using the original samples of MRW (1992) on the latest data. This is indeed a very interesting finding and it warrants further investigation but that is out of the scope of this study.
And finally we also notice that choosing output per-capita, output per working-age person or output per labour force unit as our dependent variable is insignificant to the results produced in all samples and model variants considered in this study.