Publications  Income Distribution 
9. The Kuznets hypothesis (1955) that as income per capita rises over several decades income distribution would first become less equal and then more equal has been supported by a large array of empirical studies [e.g., Bacha (1977), Ahluwalia (1976a, b), Chenery et al. (1974), Adelman and Taft Morris (1973), Cline (1975), Paukert (1973)]. Testing for the Kuznets curve in standard fashion by using crosscountry data, on the assumption that crosscountry and intertemporal Kuznets curves are identical, the Kuznets hypothesis is confirmed, the coefficients remain relatively stable and do not lose much of their significance even after social, economic and regional factors are introduced in regressions. However, the Kuznets curve itself explains only a relatively small part of the total variation in income inequality. 
The sequence of nested regressions allowed us to use F tests for the joint explanatory power of the Kuznets curve and of the groups of variables added to the equations, as is common in the analysis of covariance. The hierarchical approach favors the factors which enter early into analysis, unless the factors are completely orthogonal. In the hierarchical test implied by the sequence of regressions in table 1 the joint explanatory power of the two Kuznets curve variables was quite significant (computed Fstatistics were 13.9 for GINI and 5.8 for SHARE). The joint significance of the social factors was even higher, but economic factors were insignificant. Regions were significant for GINI (F of 5.95) even when entered last but not for SHARE. 
The 'regression' approach may underestimate the significance of all factors. When the Kuznets curve entered after the other factors, equivalent to a 'regression' approach, it remained significant for GINI when entered after the social factors but lost its significance when entered after the economic and regional factors. For SHARE the Kuznets curve lost its significance as soon as education entered the regression.^{footnote4} 



