7.1.3Model 3: County Density of IT vs. Non-IT Activity and Labor Productivity

This model is an attempt to evaluate the relationships between county density of employment and state labor productivity. It is largely inspired by the study of Ciccone and Hall (1996), who built an ingenious model that leads them to conclude that when average employment density doubles in the counties of a state, gains in labor productivity at the state level amount to 6%. My first goal is to approximately replicate their results, using my datasets at the state level and county level. The nonlinear equation estimated by the authors is:

lnps = lnA + ε lnedus + lnDs(θ)(7.9)

where edus is a measure of education level for state s and is given by the weighted average years of education. Ds (θ) is a measure of the density of activity at the county level, nonlinear in parameter θ, and is defined as:

Ds(θ) = [ Σc (nc θ.ac 1- θ)] / ns(7.10)

where the density index (Ds) depends on employment in county c (nc), the area of county c (ac), and state employment (ns). Parameter θ is the elasticity of state labor productivity with respect to employment density at the county level. Ciccone and Hall estimated this parameter at a value of 1.052, and 1.06 when the same equation is estimated using instrumental variables to control for the direction of causality. Since CH have already demonstrated that controlling for causality direction does not significantly alter the results, I will simply estimate equation 7.9, without adding instrumental variables. I expect to find a value for parameter θ to lie between 1.05 and 1.06. If I do, then this will indicate the dataset produced in part I, and the county data set built in this part of the dissertation are of quality that is as good as the data used by CH. Based on this fact, will then be able to use my state-county dataset with confidence.

Finally, my goal is to estimate urbanization economies associated with IT and non-IT employment. In order to do so, I need to construct a variable that will allow me to evaluate these economies of scale. Urbanization economies arise when the number of establishments in all industries taken together increases in a county. These agglomeration economies are therefore internal to the urban area. In the literature, Carlino (1985) argues that a good variable capable of gauging the magnitude of urbanization economies is the total number of establishments in a given SMSA divided by the distance from close-by SMSAs. However, this measure, involving the calculation of physical distances between SMSAs, is very difficult to construct, and would be even more difficult at the county level. Instead, I will construct a measure of the density of employment in a given industry as a variable capable of measuring urbanization economies.

The variable reflecting urbanization economies is very similar to the location quotient, except that it relates to the density instead of the intensity of employment. It will be called the density quotient (DQ) and is computed for each industry e in each county c such as:

DQe,c = (ne,c / ac) / (ne,s / as)(7.11)

It is also similar to the concentration ratio, but the density quotient is computed at the county level, not the state level. The higher the value of DQe,c, the higher the density of employment of type e in county c with respect to state density.

This variable will be related to labor productivity as:

ln pe,c = lnA + αc lnkne,c + π lnDQe,c + ΣDs-1(7.12)

Again, since we are interested in the differences between IT and non-IT effects, I will also use the ratio of the IT density quotient to the non-IT density quotient (RDQ), and estimate its effect on labor productivity using the following relationship:

ln pc = lnA + αc lnknc + φ lnRDQc(7.13)

If φ is positive and significant, then the increase in the density of IT employment relative to the density of non-IT employment will increase labor productivity.