9.3Regional Income Inequality

Levernier, Rickman and Patridge (1995) studied income inequality across the 48 contiguous U.S. states in 1960, 1970, 1980 and 1990. They regressed states’ Gini coefficients on several economic, demographic, human capital and labor market variables, controlling for fixed regional effects. Their economic regressors include real per capita income and its square, the industrial composition of a state’s workforce and the growth rate of non-agricultural employment. Demographic and labor force variables include urbanization rate, labor force participation rate, racial composition, age characteristics, rate of female-headed households, unionization and the immigration level of the state, as well as the share of government transfer payments in income. Finally, human capital variables are represented by the proportion of the population that has graduated from high school and from college. The authors estimated an OLS regression on their pooled cross section time series data with states’ Gini coefficients as the dependent variable and the variables mentioned above as the regressors and dummies for region and years.

First, they found that dummies for years had an increasing value, meaning that inequality has increased over time between 1960 and 1990. Second, the variables having a negative and significant coefficient (reducing inequality) were: the unionization rate, the proportion of the labor force that has graduated from high school, the level of income, the labor force participation rate and the proportion of workforce in the mining, construction and manufacturing industries. However, the square of the income level and the urbanization level variables were insignificant when the regression is estimated for each year separately. Since the square of the income variable is not significant, the authors conclude that the level of economic development of a state does not appear to significantly affect inequalities, unlike the percentage of the labor force that is in the goods producing sector and that is educated at the high school level. However, the proportion of the workforce that is college educated is not significantly related to the level of income inequality. Hence, even if the greater premium associated with college education has increased U.S. wage inequality, it does not seem to have influenced regional differences in inequality. The racial composition of the state, measured as the percentage of the labor force that is black, surprisingly does not affect income inequality. Finally, the variables found to increase inequalities (having a positive and significant effect on the Gini coefficient) are: the percentage of the state’s population that works on a farm and the share of female-headed households. Mixed effects were reported for the growth rate of the labor force, the level of immigration, the age characteristics, the unionisation rate and the rate of transfer payments of the state, although the latter were positively correlated to inequality for some years.

In discussing their results, Levernier Rickman and Patridge (1995) argued that the Gini coefficient is not a perfect measure of income inequality, and multicollinearity may influence the yearly estimates. Indeed, the Gini coefficient raises concern regarding its construction, as stated by Levernier Rickman and Patridge (1995)

‘The Gini coefficient suffers from the well-known problem that changes in the middle of the distribution have larger influence than changes in the tails of the distribution. If most of the changes in income inequalities are in the tails – i.e. among the lowest or highest income families – the Gini coefficient may be an inadequate measure of income inequality.’

To prevent such difficulties, the authors reestimated their regressions using the variance of the log family income instead of the Gini coefficient, as suggested by Levy and Murnane (1992). Their results are similar to the original ones, and Levernier, Rickman and Patridge concluded that the results are robust to changes in the measurement of income inequality. This analysis also indicated that multicollinearity did not appear to have affected the results, since results using the variance of the log family income would have probably been different from the original ones. Furthermore, the pooled regression results are generally consistent with the yearly regressions.

Levernier, Rickman and Patridge also concluded that there is evidence of convergence in state income inequality over time. Indeed, this type of inequality used to be much more important in the southern states, and over time other states have reached the same level of inequality. Income inequality has increased in New York, California, Illinois and New Jersey and it has decreased in most of the mid Atlantic states including Delaware, Maryland and Virginia.

Langer (1999) also studied income inequality across states for the same years as well as for each year between 1976 and 1995. She measured income inequality with the Gini coefficient, which is computed from two sources: the Bureau of Census decennial census of population, and the Current Population Survey (CPS) for the yearly measure. However, the values she computed for the Gini coefficients are slightly different from the ones used by Levernier, Rickman and Patridge. This might explain some differences in their results. She observed three kinds of general patterns in income inequality over time. First, some states such as New York, California, Louisiana and Delaware exhibit a steady linear increase in their level of inequality since the 1970s. Second, states such as Nebraska have shown a cyclical pattern in inequality over time. Finally, other states such as Virginia have followed a decreasing trend in income inequality, at least until the mid-1980s. Still, Langer further admitted that “the variation in income inequality across states and over time begs theoretical explanation,” but proposed that “the American states are ideal settings to study the forces affecting income inequality.”

Finally, Greenwood (1999) studied the relationships between technology, productivity and income inequality. He first assumes that the development of new technologies such as information technology involves considerable learning costs, and these costs are lower for skilled workers. The demand for skilled workers will then increase, and so will their wages relative to unskilled workers’. Thus income inequality should rise with the development of new technologies. Furthermore, productivity may stall as investment in new technology equipment and knowledge increases. Greenwood (1999) argued that

‘Technological progress is associated with growth in productivity and wage inequality. In the short term, skilled employees earn more than unskilled ones; also, wealthy individuals take advantage of new profit opportunities. However, over time, the level of skill needed to master new technologies declines; also opportunities to make profits are reduced. Therefore, over a long period, everyone gains from technological innovations.’

This chapter has presented a short survey of some of the literature on income inequality, which remained stable in the 1970s, increased sharply in the 1980s, and increased moderately in the 1990s. A new feature of income inequality appeared in the 1980s, with an increase both in between and within groups inequality. Several factors were held responsible for this trend. Candidates include: institutional changes, supply-side and demand-side determinants. Each of these determinants could explain about a third of income inequality. The role of computers and IT was also stressed, with a positive effect on the wage differential. Finally, regional income inequality was discussed, mainly through an empirical analysis of the U.S. states form Levernier, Rickman, and Patridge (1995). The next chapter describes the methodology applied to the analysis of the effects of IT characteristics on income inequality at the state level for the year 1990.