Modeling Owner-Occupied Single-Family House Values in the City of Milwaukee: A Geographically Weighted Regression Approach

Abstract

This study investigates the spatial non-stationarity of the relationship between house values and various attributes in the City of Milwaukee. From the 2003 Master Property (MPROP) data file of the City of Milwaukee, a set of owner-occupied single family houses were randomly selected (representing 99% of confidence within a ±2% range of accuracy of the total population) to model how house values are related to various house attributes. Remote sensing information (the fraction of soil and impervious surface that represent degraded neighborhood environmental conditions) is added to fine-tune the relationship. A geographically weighted regression (GWR) approach is used to investigate spatial non-stationarity. The modeling revealed that significant spatial non-stationarity existed between house values and the predictors. Specifically, the study found that those house attributes—including floor size, number of bathrooms, air conditioners, and fire-places—add more value to houses in the more affluent areas (especially on the east side near Lake Michigan and in suburban areas) than in the relatively poor areas. In addition, older houses in the historical area are more expensive, which differs from other areas. Environmental conditions, though expected to have a negative impact on house values in most areas, did not affect house values in the historical area.