Growth Regression Revisited: Selection and Conditional Convergence

Baumol (1986)’s growth regression is a starting point in growth empirics. But, as Delong (1988) pointed out, it has serious defect of sample selection and measurement error. These can lead to inconsistency of the estimate of convergence parameter. After their studies, numourous econometric studies have been performed. I incorporate these methods into growth regression framework. In this study, I apply Heckit (1979) to this regression using Peen World Table10 panel data in addition to the same data of Baumol’s. I extend the econometric model also using instrumental variable estimation for solving measurement error problem. As Delong (1988) derived similar way, correcting it with Heckit (1976) and maximum likelihood methods reduce (near) no convergence. His solution seems to be incomplete, in that the selection of countries by Baumol (1986) is not explicitly estimated. In general, the biasedness and inconsistency in growth regression come from more selection rather than correlation of explanatory variables with error such as measurement error or dynamic panel regression.