Martin Shanahan and Margaret Correll, The University of South Australia

Trends estimates of the distribution of personal wealth in the United States between 1650 and 1950 have been available for several years. Some authors have cited the long-run trend as evidence of a KuznetsÕ curve. This paper re-examines the studies used to compile the trend estimate. In this preliminary work we argue the evidence for a KuznetsÕ curve in the US wealth data between 1650 and 1950 curve is weak and should be reassessed. The currently accepted story of generally low levels of inequality in the colonial period, followed by sharp rises in wealth inequality between 1820 and 1840, an unclear but probable peak in inequality between 1870 and 1920, followed by decreasing inequality until the mid 1970s is not supported by the evidence. In particular, current colonial wealth estimates are likely to significantly underestimate wealth inequality, while the data available between 1870 and 1920 are particularly weak. Close examination of the original studies used to estimate the distribution of wealth reveal a non-random bias in the method of estimation. Studies of the colonial period in America consistently underestimate the distribution of wealth. Furthermore, the wealth estimates which examine the distribution of wealth between 1870 and the 1920s, by including ÓmissingÓ people and Ómissing wealthÓ provide far better coverage of the entire population than earlier studies. The result is to create an apparent trend in the distribution of wealth arising from different methodologies rather than real differences in inequality. In addition to the non sampling errors resulting from the combination of different data sources over time, there are also sampling errors which are confounded by combining data sets of systematically different sample size. We conclude the combined effect of the non sample and sample errors greatly diminishes any clear trend in the distribution of wealth and may indeed result in no clear ÓKuznetsÕ curveÓ.