11.1Presentation and Interpretation of the Results

Several OLS specifications based on the model presented in the preceding chapter are reported in Table 11.1. Regression (11.1) shows the results from Levernier, Rickman, and Patridge (1995). Regression 11.2 presents the results from the approximate replication of 11.1, using my own dataset. In regression 11.3, a variable measuring the intensity of IT employment (LITP) is introduced, and several independent variables are dropped from the analysis to avoid multicollinearity. Finally, two variables measuring the density of IT and traditional employment are added to this model in regression (11.4). Because there are exogenous factors that account for differences in states’ Gini coefficients, regional dummy variables must be used to control for fixed regional effects. The omitted dummy is the Census’ North region.

First, the results of regression (11.2) are similar to the ones obtained by LRP as reported in regression (11.1). Indeed, the income variables are insignificant in both cases. This tends to contradict Kuznets’ hypothesis according to which income inequality rises with income until a certain threshold is reached where society becomes more generous towards the poor and inequality starts to decrease. The percent of non-whites (NWHITE) has a positive and significant coefficient in all regressions. Therefore, it seems that income inequality increases for states that have a high proportion of non-white population, perhaps because of segregation or other racial issues. On the other hand, high school (HS), labor participation (LABPART) and goods production employees (PCGOODW) have negative and significant coefficients, which means that an increase in these variables must decrease income inequality. Thus, states with a higher proportion of high school graduates have lower income inequality. Still, the coefficient for the college graduate (COL) is not significant, and the effect of higher education on income inequality is ambiguous, as stated previously. Similarly, states with a high rate of labor force participation or a high percentage of workers in good producing industries have lower income inequality.

Before introducing the IT employment intensity variable (LITP), I had to drop some variables from the model of regression 11.2. There are only 45 states, and adding variables would decrease the degrees of freedom, and eventually increase the risk of multicollinearity. Thus, I dropped the variables that were not significant in regression 11.2 (INC, INC2, COL) except the regional dummies to keep controlling for regional fixed effects. I also dropped the variables high school (HS) and goods employees (PCGOODPW) in order to prevent multicollinearity with the IT employee variable LITP.

Regression (11.3) shows the results of the regression of the Gini coefficient on high school, labor force participation and IT employee variables (NWHITE, LABPART, LITP) and regional dummies. Results still indicate a strong significance of NWHITE and LABPART. The coefficient for LITP is positive and significant at the 0.05 level. Therefore, states with a higher share of their employees working in occupations that are IT intensive have greater income inequality. This result might come from the fact that IT employees usually have greater income than “traditional” employees. The aggregate income gap between these two categories must increase with the share of IT workers (LITP).

Finally, regression (11.4) estimates the effect of the density of employment on income inequality. The effect is different whether the density of IT employment or traditional employment is considered. Although their sizes are similar, these coefficients have a different sign. The density of IT employment has a positive impact on income inequality, which means that as IT workers concentrate in one location, income inequality at the state level is rising. The coefficient for the density of IT employees (ITDENS) is estimated at 0.139 and is significant at the 0.05 level. This means that if the density of IT employees doubles in a county, the level of labor productivity in that county may increase by 13.9%. On the other hand, the coefficient for non-IT employment density is negative and significant at the 0.10 level. Hence, as traditional workers concentrate in one location, the level of income inequality at the state level decreases.

This last result may come from the fact that these two types of employment (IT and traditional) may have different agglomeration and congestion effects. As stated in the second part of this dissertation (chapters 6 through 8), state productivity may increase with the density of IT employment at the county level, and increase less or even decrease with the county density of non-IT employment. Indeed, the literature on regional economics states that the externality effect (or spillover) associated with density is the product of two opposite effects: agglomeration and congestion. First, agglomeration economies arise when workers benefit from being concentrated in space. For instance, workers in research and development departments benefit from physical encounters with co-workers, which allows ideas to spread all over the local area. Since IT workers are dealing with knowledge and information as their main resource, strong agglomeration effects must result from higher employment density in this type of employment. The effect should not be as strong for traditional workers for whom information is not the main resource. In the non-IT density case, the congestion effect might be greater than the one associated with the density of IT employment. This effect might even be greater than the agglomeration effect, resulting in a negative effect of density on productivity. In this case, if productivity is lowered by density then wages and income should also be lowered, reducing income inequality by the same token. Whereas for IT employment density, agglomeration effects should offset congestion effects, resulting in higher productivity, wages and income inequality. This may explain the positive and negative signs obtained for the coefficient of IT and non-IT density (ITDENS, NITDENS) in regression (11.4).