Error Trade-offs in Regression Appraisal Methods

This chapter presents a conceptualization of the problem of how many sales to use in a sales comparison method appraisal. We propose that the answer to this question depends on a tradeoff between different kinds of errors and is data dependent, varying across data sets. In hedonic regression appraisals, the law of large numbers does not necessarily hold. Increasing variance, omitted variables, and measurement errors in the sample may overwhelm the efficiency benefits of larger sample size. Heterogeneity within samples makes for the common problem of large standard errors in regression appraisal price predictions (Lenz & Wang, 1998) and suggests disaggregating data to improve the precision of price estimates. A “law of medium numbers” applies, meaning that heterogeneous data may yield smallest standard errors when subsets of data are used for estimates rather than the whole population of sales.