Spatial variation of residential and employment land consumption rates in a metropolitan region: Atlanta, Chicago, Sacramento, San Antonio

This study attempted to develop an improved spatial model for estimating variations of residential and employment land consumption rates in a metropolitan region. In pursuing this goal, there were two major research streams in this dissertation. The one is the research question of what factors influence the differences of land consumption rates in a region? The related research question is what model specifications do we need to estimate it? In order to answer the research questions, the following research issues are focused in particular. First, we tested alternative accessibility measures under different urban form assumptions: monocentric, polycentric and dispersive. Using the criterion of maximum explanatory power, it is found that a gravity type accessibility measure with power form is the most useful in explaining the variations of residential and employment land consumption rates. Second, with theoretical considerations, relevant factors that matter with spatial variations of residential and employment (commercial and basic) land consumption rates are identified and reviewed. These variables can be broadly grouped by Transportation Accessibility factors and Locational Amenity factors. We also hypothesized the relationship between explanatory variables and land consumption rates, and empirically tested with four metropolitan regions data: Atlanta, Chicago, Sacramento and San Antonio. Rather than accepting a deterministic logic (‘all relevant factors should be consistent in any urban regions’) or an ‘every case is different’ approach, this study assumed a contingency approach, and developed a procedure of assessing and selecting relevant factors for statistical analysis. The resulting model is called as a reduced model, which works better than a full model. Third, we also considered ‘spatial interaction effects’ in the models, and thus specified Spatial Autocorrelation model (SAR) and Spatial Lag model (SL). These spatial models are compared with OLS models. The results suggested that spatial regression models are preferred in 10 of 12 cases