Do Atlanta Residents Value MARTA? Selecting an Autoregressive Model to Recover Willingness to Pay

Understanding homeowners’ marginal willingness-to-pay (MWTP) for proximity to public transportation infrastructure is important for planning and policy. Naive estimates of MWTP, however, may be biased as a result of spatial dependence, spatial correlation, and spatially endogenous variables. In this paper the authors discuss a class of spatial autoregressive models that control for these spatial effects, and apply them to sample data collected for the Atlanta, Georgia housing market. The authors provide evidence that a general-to-specific model selection methodology that relies on the generality of the spatial Durbin model (SDM) should be preferred to the classical specific-to-general methodology that begins with an assumption of no spatial effects. The authors show that applying the SDM widens the confidence interval of the estimate of MWTP for transit proximity in Atlanta, relative to ordinary linear regression. This finding has unpredictable consequences for land value capture forecasts and transportation policy decisions.