Inequality and Economic Growth: Data Comparisons and Econometric Tests

This paper discusses two issues in the relationship between inequality and economic growth: the data and the econometrics. We first review the inequality data set of Deininger and Squire, which, we argue, fails to provide adequate or accurate longitudinal and cross-country coverage. We then introduce our own measures of the inequality of manufacturing pay, based on the UNIDO Industrial Statistics. In our view, these provide indicators of inequality that are more stable, more reliable, and more comparable across countries than those of Deininger and Squire. Turning to the relationship between inequality and development, we diagnose several common econometric problems in the literature, including measurement error, omitted variable bias, serial correlation in longitudinal data, and the possible persistence of lagged dependent variables. By taking steps to account for these problems, we seek more reliable inferences concerning the relationship between inequality, national income and economic growth. We find evidence that generally supports Kuznets’ specification for industrializing countries: inequality tends to decline as per capita income increases. However, after 1981 two problems emerge. First, per capita GDP growth slows dramatically in most countries. Second, there is a worldwide trend toward rising inequality in our data, independent of GDP or its changes. The timing and geographic pattern of these increases suggest a link to the high real interest rates and global debt crisis of the period beginning in 1982. Date of Present Version: April 5, 2002 I. Data Problems in Research on the Kuznets Hypothesis For students of the relationship between inequality and economic growth, the quality of inequality measures has been a problem ever since Kuznets (1955) formulated his hypothetical inverted “U” relationship between inequality and national income. While some studies have treated this relationship as “stylized fact” (Ahluwalia, 1976) or “economic law” (Robinson, 1976), longitudinal data required to test it rigorously have always been scarce, and measurement controversies persist in available data to this day. In this respect, Deininger and Squire’s effort (hereafter D&S;, 1996) is monumental. D&S; collected many disparate surveys of income inequality, and compiled those meeting certain criteria of process into a single “high-quality” panel, offering 693 country/year observations since 1950. This is now a standard reference, on which dozens of papers have been based. However, the D&S; data do not generate a consistent relationship between income inequality and either income levels or rates of growth. In different papers many different, even contradictory, forms of this relationship have been specified, including both inverted (Barro, 2000) and upright (Ram, 1997) “U” relationships between inequality and income levels, and both downward (Deininger and Squire, 1998) and upward–sloping (Forbes, 2000) relationships between inequality and subsequent economic growth. The differences stem partly from data issues. Despite the apparently large number of observations, the coverage of the D&S; data set remains limited and unbalanced. And serious questions have been raised as to whether the data points are in fact comparable either across countries or through time. As Atkinson and Brandolini (2001) especially argue, the D&S; inequality measures are based on various income definitions, recipient units and processing procedures that cannot be wholly reconciled to each other, even with “high-quality” filtering. shows the various sources of the D&S; Gini coefficients by income definition and recipient units. Table 1. The Distribution of Inequality Measures by Different Definitions in D&S; Data