This thesis investigates the causes of the contradictory conclusions of Pritchett (1996) and De la Fuente & Doménech (2002 and 2006) on the role played by growth in human capital in explaining growth in output. While both models are based on a variation of an augmented Solow model, much in accordance with Mankiw, Romer & Weil (1992), Pritchett finds that cross-national data show no association between growth in human capital, measured by growth in educational attainment, and growth in output. Opposite to this finding, De la Fuente & Doménech’s results propose a coefficient for human capital growth of well above 0.50, and suggests that schooling data of poor quality is a likely source to the discouraging results on the contribution of human capital, found by Pritchett and also other researchers. This thesis examines the differences in the educational datasets composed by Barro & Lee (1993), used by Pritchett, and De la Fuente & Doménech (2002) for 21 OECD countries using five-year growth periods from 1960 to 1985, and finds that there are large disparities in both levels and growth rates between the datasets. Barro & Lee’s data is found to contain implausible jumps and breaks, and over 14% of the growth rates are reported to be negative. This seems highly questionable. De la Fuente & Doménech’s dataset projects much smoother growth in educational attainment, and reports no periods of negative growth. However, these large differences in human capital data are not sufficient to explain the contradicting results. Through regressions on several different variations of both models, other important factors contributing to the disparities are identified:
- Differences in the datasets on output per worker- Differences in the datasets on physical capital per worker- Excluding/including time fixed effects in the model- The calculation method for the proxy on human capital
It is through the cumulative effect of all these dissimilarities that the opposing views on human capital are based. The estimated coefficients on growth in human capital are also found to be highly sensitive to seemingly small alterations in the model or any of its inputs. This suggests that further research and larger data samples are needed before any conclusions on the impact of human capital should be made.