roiw_365 102..122 MIXTURE MODELS, CONVERGENCE CLUBS, AND POLARIZATION

We argue that modeling the cross-country distribution of per capita income as a mixture distribution provides a natural framework for the detection of convergence clubs. The framework yields tests for the number of component distributions that are likely to be more informative than “bump hunting ” tests and includes a method of assessing the cross-component immobility necessary to imply a correspon-dence between components and convergence clubs. Applying this approach to Penn World Data for the period 1960 to 2000 we find evidence of three component densities. We find little cross-component mobility and so interpret the multiple mixture components as representing convergence clubs. We document a pronounced tendency for the strength of the bonds between countries and clubs to increase and show that the well-known “hollowing out ” of the middle of the distribution is largely attributable to the increased concentration of the rich countries around their component means. 1