|
Abstract:
|
A review of SIPP-based research suggests many users do not account for the survey's complex sampling design when estimating the variances of point estimates. This practice leads to inappropriate conclusions about the precision of variance estimates. We estimate the variance of several demographic and economic characteristics (e.g., age, race, ethnicity, public and private health insurance rates, employment rates, program participation, and poverty rates). The variances are estimated using four unique estimators: simple random sample approach, generalized variance parameter approach, a Taylor Series estimator, and a replicate weight approach. The results show not accounting for sample design will lead to biased variance estimates. We conclude by offering specific advice to the SIPP data user community about the best way to calculate standard appropriately.
|
