The U.S. Census Bureau's Small Area Income and Poverty Estimates (SAIPE) program produces model-based estimates for small geographic areas using household survey data, administrative records, postcensal population estimates and decennial census data. This paper proposes and evaluates a method for making year-to-year statistical comparisons of poverty at the county level. The method uses aggregations of regression residuals in order to estimate the underlying serial correlation in SAIPE county-level estimates. Three residual-based estimators for the model error correlation are considered, with alternative weights used for each. The estimators are evaluated using simulations under the assumed error specification, and the effect of a heteroscedastic departure from these assumptions is discussed.