8.2State Level Analysis: IT and Non-IT Concentration Ratios (Model 1)

This section focuses on the spatial concentration aspect of IT employment and non-IT employment at the state level, and their effects on labor productivity. Table 8.2 shows the estimates of equations based on:

ln ps = lnA + αs lnkns + δ lnCONe,s(8.4)

where state labor productivity (ps) is regressed on state capital-to-labor ratio (kns), state level of education (eds) [regression (8.1)] and the state concentration ratios of IT and non-IT employment [regressions (8.2) and (8.3), respectively], as well as their ratio [regression (8.4)] and the two concentration variables together [regression (8.5)].

The first noticeable result is that all coefficients are statistically significant at the 0.01 level, and the explanatory power of all the models is satisfactory, with adjusted R-square values between 0.66 and 0.71. The control regression (8.1) and all the others indicate that labor productivity varies across states with the capital to labor ratio (elasticity between 0.440 and 0.471) and the level of educational attainment (0.112 to 0.246).

When the concentration ratios of IT and non-IT employment variables are introduced separately [regressions (8.1) and (8.2)], they have the same elasticity value of 0.028. This result means that if in a given state the concentration of IT or non-IT employment doubles, gains in state labor productivity amount to 2.8%. So concentration in general increases productivity. However, when these variables are considered together in the regression (8.5), the employment elasticity of state labor productivity is positive (0.085) with respect to the concentration of IT employment, and negative (-0.067) with respect to non-IT employment concentration. Thus, the concentration of IT employment in a state would have positive external effects on labor productivity maybe through agglomeration economies. On the other hand, the concentration of non-IT employment must be subject to congestion effects greater than agglomeration effects, which would explain its negative contribution to labor productivity once the IT concentration has been accounted for. However, estimates of regression (8.5) have to be considered carefully since multicollinearity between the IT and non-IT employment concentration ratio may arise. Indeed, the correlation coefficient between these two variables is estimated at 0.965. The F-statistic computed for this regression is also lower than the other ones. Hence, this model may not be used to compare the contribution of IT relative to non-IT employment concentration on labor productivity differences. Another model was then introduced, using the ratio of these two variables, which prevents multicollinearity from influencing the results. Indeed, the level of correlation between this ratio and other independent variables is less than 0.11. Results appear in regression (8.4) and indicate a stronger coefficient for this ratio compared to individual concentration variables. Indeed, the elasticity of 0.107 means that the ratio of the concentration of IT relative to non-IT employment concentration may explain 10% of the variation in state labor productivity. These findings suggest that states should favor the concentration of IT employment relative to the concentration of traditional employment in order to increase their level of labor productivity.

The next two sections report results of county level analyses. The location and density quotient are computed for each county, and their effects on county labor productivity is evaluated.