A Comparison of Spatial Statistical Methods in a School Finance Policy Context

Abstract

A shortcoming of the conventional ordinary least squares (OLS) approaches for estimating median voter models of education demand is the inability to more fully explain the spatial relationships between neighboring school districts. Consequently, two school districts that appear to be descriptively similar in terms of conventional measures of taste, tax price, and income—but which are located in different geographical contexts—actually may spend at very different levels. In this article, the issue is addressed by employing a relatively recent spatial analytical method called Geographically Weighted Regression (GWR), and a more common spatial modeling approach, spatial autoregression (SAR). This study uses 2000 and 2004 data from the Missouri Department of Elementary and Secondary Education (DESE) along with the United States Census 2000 School District Special Tabulation, and analyzes each school district's educational spending and underlying wealth, income, and demographic data. Evidence from this study suggests that both GWR and SAR provide useful and different insights. Virtually all of the error related to spatial autocorrelation is eliminated in both SAR and GWR approaches. Further, key parameter estimates vary between the OLS and SAR specifications, and vary regionally in the GWR specification. The article concludes that there may be a place for the use of spatial statistics in education finance policy research.