Measuring Temporal Stability in Regression Models of Population Estimation

Abstract

This paper introduces an empirical indicator designed to measure the temporal stability of regression models used to produce subnational population estimates. Analysis of 67 counties in Florida centers on 1970 total population estimates generated from ratio-correlation and difference-correlation models. Comparisons are made between eight different regression specifications and employ a quantitative measure of relative estimate accuracy. The major findings of this study are that (a) variable measurement and type are important determinants of estimate accuracy, and (b) although temporal stability of the coefficients impacts estimation errors, the influence is not as pervasive as is suggested in the literature.