Similarity Of Structural Changes: The Case Of University Enrollment Rates

The paper examines similarity of models with structural changes among heterogeneous panel data units. We propose applying a cosine metric to compare angles between vectors of weighted coefficients as a measure of closeness of economic models. Testing whether the cosine metric value is zero against nonzero, positive, and negative alternatives enriches traditional testing results. The latter merely indicate that models are different since the vectors of coefficients could not be treated as equal. We suggest interpreting nonzero values of a cosine metric as evidence of similarities in the factor structure. This means that similar factors are significant; majority of them affect the dependent variable either in the same direction or opposite. Applying the methodology to study dynamics of university enrollment rates in various countries, the paper provides evidence for the existence of similarities in the factors driving university enrollment rate dynamics. It identifies sustainable cluster divisions and applies the cosine metric to different groups of countries. Notably, evidence is provided that post-communist countries are more similar in the factor structure of the dynamics of university enrollment rates to developed countries than to other developing countries. Increasing access to the internet among population strongly positively contributes to explanation of dynamics of higher education enrollment rates in almost all countries in the 1990s to the beginning of the 2000s.