In addition to the data on concentrations the GEMS network also classifies each site within a city as either city center, suburban or rural in land type, and we employ these land type categories in our analysis. A list of the cities involved, the years of operation of GEMS stations, and the number of observations from each city is given in Appendix A.
In moving from our theoretical model to its empirical counterpart we need to include variables to reflect scale, technique and composition effects.
As well, we have to include site-specific variables to account for the density of economic activity and meteorological conditions. Our estimations will require the use of data on real GDP per capita, capital to labor ratios, population densities, and various measures of “openness”. The majority of the economic data were obtained from the Penn World Tables 5.6. The remainder was obtained from several sources. A full description of data sources and our methods for collection are provided in Appendix A together with a table of means, standard deviations, and units of measurement for the data.
Linking Theory to the Estimating Equation
To derive an estimating equation, assume measured concentrations at any observation site are a function of the country specific economic determinants of emissions, E; site-specific meteorological and density variables (V); common to world trends in abatement technology and world prices (C); and a site-specific error S that includes other relevant, but unmeasured determinants of pollution, plus an idiosyncratic measurement error reflecting human and machine error. If we take a Taylor series approximation to this general functional form we can then write pollution concentrations at site i, city j, in country k, at time t as
where bE,bV and bC are parameter vectors and Eijkt, Vykt and Ct represent vectors of regressors to be explained below.