|
Abstract:
|
We demonstrate use of bivariate area-level models to improve small area estimates from one survey by borrowing strength from related estimates from a larger survey. In particular, we demonstrate the potential for borrowing strength from estimates from the American Community Survey (ACS), the largest U.S. household survey, to improve estimates from smaller U.S. surveys. For illustration we use, in conjunction with data from ACS, data from the National Health Interview Survey, the Survey of Income and Program Participation, and the Current Population Survey. To borrow information we propose use of a simple bivariate Gaussian model and also, for proportions, a bivariate binomial logit normal model. Simple theoretical calculations and the results from the examples show that substantial reductions in variances may be achieved by borrowing strength from the ACS via the bivariate models even without using regression covariates obtained from auxiliary sources. Theoretical calculations show how the extent of variance reduction depends on the characteristics of the underlying data.
|
