Empirical assessment of spatial prediction methods for location cost-adjustment factors

In the feasibility stage of a project, location cost-adjustment factors (LCAFs) are commonly used to perform quick order-of-magnitude estimates. Nowadays, numerous LCAF data sets are available in North America, but they do not include all locations. Hence, LCAFs for unsampled locations need to be inferred through spatial interpolation or prediction methods. Using a commonly used set of LCAFs, this paper aims to test the accuracy of various spatial prediction methods and spatial interpolation methods in estimating LCAF values for unsampled locations. Between the two regression-based prediction models selected for the study, geographically weighted regression analysis (GWR) resulted the most appropriate way to model the city cost index as a function of multiple covariates. As a direct consequence of its spatial nonstationarity, the influence of each single covariate differed from state to state. In addition, this paper includes a first attempt to determine if the observed variability in cost index values could be at least partially explained by independent socioeconomic variables.