Population and employment densities have been the focus of many studies in recent decades. Because of their complex characteristics, it is difficult to model them with simple functional forms and variables. In the past, the negative exponential model, with distance(s) to major Central Business District(s) (CBDs), as the major explanatory variables(s), have been tested extensively, using a monocentric framework initially and a polycentric one more recently. As an alternative to these models, a new density modeling approach is proposed in this research, based on the integration of concepts related to both multiple population and employment centers and landscape ecology theory.
Regression models are specified to estimate the empirical relationships between population and employment densities, and (1) distances to major CBDs, (2) distances to amenities and disamenities, and (3) landscape indices that characterize the land-use structure. A comprehensive spatial and non-spatial database is built over all Traffic Analysis Zones (TAZ) of the seven counties of Central Ohio. Extensive applications of GIS have been necessary to compute indices and distances to CBDs, amenities, and disamenities, as well as to map urban morphology and distances. Negative exponential models have been estimated at both the metropolitan and county levels, using ordinary least square regression. The best exponential models at the metropolitan level, with distances to the 43 major CBDs within the seven counties as independent variables, explained about 58% and 59% of the variations in population and employment densities, respectively, but several distance coefficients were not significant and of the wrong signs. Estimated at the county level, homogenous exponential models for population and employment densities explain: 39% and 42% of variations in Franklin County; 63% and 73% of variations in Delaware County; 67% and 64% of the variations in Licking County; 84% and 76% of the variations in Fairfield County; 61% and 73% of the variations in Pickaway County; 59% and 68% of variations in Madison County; and 58% and 82% of variations in Union County, respectively. Finally, the best pooled models of population and employment densities, with county dummy variables, second-order terms for the significant landscape indices (MECI: Mean Edge Contrast Index, and DO3: Dominance Index), and five interaction variables, explain about 66% and 73% of the variations in population and employment densities in the Columbus MSA, respectively.
The results provide evidence that distances to major metropolitan and county CBDs and two spatial indices significantly explain the variations of both population and employment densities. MECI and DO3 are consistently significant across the seven counties. In addition, the distances to the MSA and county CBDs play important roles in most models. The results provide further evidence that the Columbus MSA is polycentric for both population and employment densities. Models for rural counties tend to perform better than urban counties, due to their less complicated residential and employment spatial structure. Areas for further research are discussed.