On the model selection to represent human capital distribution–an empirical study

The distribution of human capital is a recent addition to the list of fundamental determinants of growth. This paper addresses an important problem of how to utilize recently available data on the highest educational attainment of adults in the labor force to represent the human capital distribution of a country by a smooth probability density function. Using the data of Barro and Lee, over-dispersed Poisson and negative binomial distributions are selected to fit the distribution of the number of years of school. Our empirical investigations suggest that these standard discrete probability distributions are too smooth to account for properties of the data, i.e., schooling is more likely to be terminated at the completion of a level of attainment (e.g., primary, secondary, and higher education) than in the middle of one, an important feature contained in the frequency distribution of highest educational attainment. Future research should focus on more complex models which account for these spikes in the data.