A geographically weighted regression approach for explaining spatial variation among school districts in a median voter model of education demand.

摘要

The median voter model of education demand postulates that for each local public school district, a measure of education demand, such as per pupil spending or teacher salaries, can be estimated as a function of the tax rate of the local district, the income levels of the district, and the desire the local district has to provide funds for education. While the median voter model might do well at explaining how much local public school districts spend on education, one may find that when analyzing the results at a more micro level, two school districts that appear descriptively similar, but which are located in different geographical contexts, actually spend at very different levels. Such a case can lead to unexplained variation due to spatial proximity if nearby school districts do indeed influence other school districts' spending rates, and if there are no controls for this influence in a model. This unexplained variation due to spatial proximity is referred to as 'spatial variability'. In this dissertation, a way to more fully explain the spatial variability of the median voter model by applying a new spatial analytical method called geographically weighted regression (GWR) is explored, and the education policy implications of this information are discussed. Evidence from this dissertation suggests that GWR more fully explains spatial variability in a median voter model when compared to ordinary least squares (OLS) and spatial autoregressive (SAR) approaches. It is shown that GWR allows for the mapping of spatially varying parameters in a median voter model, and the dissertation concludes that GWR is a useful method for helping to understand complex geo-spatial school finance policy issues. Further, there is evidence to suggest that poor school districts that are surrounded by relatively wealthier school districts spend at a greater rate than otherwise would be expected compared to similarly poor districts surrounded by other similarly poor districts. Policy options that are designed to address this issue are discussed.