| I isolate and identify the source of the bias created by the presence of a spatially distributed endogenous variable in a traditional reduced-form hedonic model. I demonstrate that the presence of a spatially distributed endogenous variable leads to biased results. The direction of the bias depends on the sign of the variable that measures the effect of surrounding property values on the location of the endogenous variable. Moreover, I show that the sign of the impact of the spatially distributed endogenous variable on property values affects the magnitude of the bias, but not the sign. While a simple endogenous variable leads to biased results, I show that the presence of a spatially distributed endogenous variable leads to biased and inefficient estimates. I also show that in the case of a spatially distributed endogenous variable, neither a correction for spatial error autocorrelation, nor a correction for spatial lag autocorrelation will solve the bias issue; rather, the correction must address both issues before consistent estimates emerge.;Finally I study the effect of livestock operations on property values using a generalized spatial two-stage least-squares (GS2SLS) hedonic model in Mercer County, Ohio. Unlike previous studies, I account for a spatially distributed endogenous variable and spatially correlated error terms. I compare the results from a simple hedonic model, a spatial error autocorrelation model and a GS2SLS. While the OLS and the spatial error models produce counter-intuitive model, the GS2SLS and the 2SLS least square generate results that are in line with expectations. The results suggest that failure to correct for these problems leads to biased and inefficient estimates of livestock's impact on property values. |
