UC Riverside

In this study we examine the regression-based ratio-correlation method and suggest some new tools for assessing the magnitude and impact of coefficient instability on population estimation errors. We use a robust sample of 904 counties from 11 states and find that: (1) coefficient instability is not a universal source of error in regression models for population estimation and its impact is less than commonly assumed; (2) coefficient instability is not related to bias, but it does decrease precision and increase the allocation error of population estimates; and (3) unstable coefficients have the greatest impact on counties under 20,000 in population size. Our findings suggest that information about the conditions that affect coefficient instability and its impact on estimation error might lead to more targeted and efficient approaches for improving population estimates developed from regression models.