Convergence Among the U.S. States: Absolute, Conditional, or Club?

This paper attempts to ascertain which of the convergence hypotheses – absolute, conditional, or club – best describes the economic development of the U.S. states since 1950. We use regression tree analysis to identify convergence clubs among the states and argue that the club characterization of the data dominates the other two. We find three convergence clubs with a state's age and it's initial densities of post offices and telephone cable determining club membership. Abstracting from catch-up effects, those states with higher densities tend to grow faster. 1 1. Introduction In Johnson and Takeyama [2001] we argue that data on per capita personal incomes in the US states over the period 1950 – 1993 are consistent with the club convergence hypothesis. We find evidence of three such clubs, each with a different long-run growth rate – one at the US average, one about ¼ percentage point above that average and one about ¼ percentage point below – implying divergence of state incomes. In that paper we assumed that the US states had a common set of economic fundamentals and differed only in their initial (1950) conditions. Thus, we tested the unconditional or absolute-convergence hypothesis against the club convergence " hypothesis. Our results contrast with those of Barro and Sala-i-Martin [1995] who cannot reject the (absolute)-convergence hypothesis for the U.S. states. This hypothesis is, however, also " 1 rejected in favor of the conditional-convergence by Holtz-Eakin [1993] who finds a role for " physical and human capital variables in explaining cross-state differences in growth rates. Both types of rejections of the absolute-convergence hypothesis – that in favor of the " conditional-convergence hypothesis and that in favor of the club convergence hypothesis – " imply the existence of permanent cross-state differences in per capita income levels although for very different reasons. In the first case, the differences reflect cross-state heterogeneity in variables such as rates of capital accumulation and the appropriate econometric response is the introduction of " control variables " into the growth regression. In the second case,