Abstract
By replicating earlier research in convergence theory, this thesis is seen as complement by applying updated data to different approaches. I will present the results by Baumol (1986), De Long (1988), Barro (1991) and Pritchett (1997) to see if their results are robust to new revisions of the data. Baumol introduced a univariate growth regression and found a pattern of convergence for 16 advanced economies, which provided evidence of growth convergence in a unconditional manner. The updated data contain a larger time and country coverage, and by running the same regression I find no evidence in the data of effects of GDP per capita on growth. Due to issues of selection bias and a concern with measurement error in the GDP estimates in the data, De Long analysed different magnitudes of such measurement error. Using his framework, I found that allowing for errors in the estimates created a positive and significant effect of GDP per capita on growth. Baumol did not account for such error, which created a downward bias in his original results that favoured convergence. If allowing for estimate errors, then there is no evidence of convergence in the new data. This is supported by Pritchett, who introduces a method to construct new income distributions. I find that such an approach provides evidence of increased cross-country income variations in the last 100 years. Assuming a univariate specification, might result in omitting different country-specific or time-variant effects. In a conditional sense, controlling for human capital in a cross-sectional regression provides positive and statistically significant effects of human capital on growth. This coincide with Barro’s findings that convergence is conditional. It also strengthens the idea that Baumol’s regression and findings are unsatisfactory in generalising growth patterns across countries.