The American Wage Structure, 1920-1947

This paper uses industrial wage data and a systematic if unconventional selection of methods to examine changes in the inter-industry structure of wages between 1920 and 1947. We first sort among the available data on wage change by industry and occupation for blocs that exhibit common patterns of wage changes over time, reducing the 83 time series available to us into eight distinct groups. Following this, we present a systematic decomposition of the sources of wage variation across groups and through time. The fact that our cluster analysis relies on wage-change observations in percentage form implies that our discriminant analysis produces eigenvectors in time-series format; thus each eigenvector is itself an artificially constructed economic time series. We identify four such forces that together explain 97% of the variance in wage change across groups, and identify variables in the historical record that appear to correspond closely to these forces. This raises a beguiling possibility. It may be that simple explanations account for most of the relative-wage changes during the years under study. In a reversal of the usual notions of micro-to-macro causality, it may be that a small number of macroeconomic variates account for a large proportion of distributional changes. In a final section, we compute an estimate of the evolution of inequality in the wage structure over time. This estimate is independent of our clustering procedures and of our discriminant analysis, and is measure is well-suited to regression analysis. Using it, we test a simple macroeconomic explanation of inequality in the wage structure. The results appear to support the argument that well-known macroeconomic and social developments, including changes in the unemployment rate, in strikes, and in the exchange rate, played the determining roles in the evolution of wage inequality during this time.