Abstract

This paper and talk develops the large-sample design-based behavior of survey estimates starting from fixed base-weights, followed by modeled response propensities for use in Inverse-Probability Weights, with final weights obtained by raking Inverse Probability Weights to a subset of population targets known from high-quality censuses or surveys. The theory supplies large-sample distributions and variance estimates in the insufficiently studied case where design weights must move a lot to adjust survey estimates either because of nonresponse or frame undercoverage or both. We illustrate the theory on low-response-rate probability and nonprobability Tracking Surveys of the US Census Bureau collected in 2019-2020 regarding attitudes to the decennial census.