11.2Robustness of Results

According LRP, the methodology used in this analysis raises concern about the robustness of the data, due to mainly two factors. On one hand, as mentioned previously, the Gini coefficient presents some limitations in reporting income inequality, due to its construction procedure. First, it is necessary to assume that income for each family or household equals the midpoint of its income class. Second, for the highest income class, the mean of family income is estimated by subtracting the sum of the income of the other classes from total income. These approximations may result in approximate measures of income inequality. On the other hand, the small number of observations (45 states) compared to the number of variables raises concern about multicollinearity issues. Following LRP, the way to deal with these two problems is to consider a different measure of income inequality, and see if results differ significantly from the ones obtained using the Gini coefficient. Table 11.2 shows the results.

Estimates of regressions with Gini and with the log of variance as a measure of inequality are similar. At least, they are similar regarding the sign and significance of the coefficients for IT employment and density variables (LITP, ITDENS) and non-IT density variable (NITDENS). They are also similar for the non-white variable (NWHITE), but are not significant for the constant and labor force participation variable (LABPART), which leads to suspicion about the true effect of the rate of labor force participation on income inequality. Furthermore, the R-squared is lower when using variance of logarithms, which adds some suspicion to the robustness of the model. Nevertheless, results using each type of measure of inequality tend to be fairly similar, especially regarding the IT variables, which are of most interest in this study.

Table 11.2Comparison of Regressions Using Gini and Variance of Log of Income as Measures of Income Inequality
Regression (11.3) (11.4) (11.5) (11.6)
Dependent variable GINI GINI Variance of log of median family income Variance of log of median family income
Constant 0.587*
(17.85) 0.590*
(16.08) -3.249E-02
(0.82) 1.981E-02
(0.40)
Non-whites (NWHITE) 0.137*
(6.11) 0.132*
(5.84) 5.270E-02*
(1.874) 5.520E-02
(2.02)
Participation rate
(LABPART) -0.346*
(7.30) -0.340*
(7.24) 1.947E-02
(0.34) 3.540E-02
(0.65)
IT employees
(LITP) 0.068
(1.97) - 0.108
(2.435) -
IT density
(ITDENS) - 0.139
(2.25) - 0.253***
(3.35)
Non-IT density
(NITDENS) - -0.128*
(1.70) - -0.271*
(3.00)
Midwest
(REG1) -2.61E-3
(0.53) -4.92E-3
(1.01) 1.261E-02
(2.03) 6.440E-03
(1.06)
South
(REG2) -1.31E-3
(0.22) -2.57E-5
(0.01) 1.924E-02
(2.518) 1.420E-02
(2.02)
West
(REG3) -3.94E-3
(0.74) -3.44E-3
(0.61) 1.769E-02
(2.634) 1.160E-02
(1.64)
R2 0.80 0.82 0.42 0.48
Durbin Watson 1.79 1.96 2.19 2.07
* Significant at the 0.01 level ** Significant at the 0.05 level * Significant at the 0.10 level