Evaluation of Using a Model-Assisted Sampling Paradigm Versus a Traditional Sampling Paradigm in a Nationally Representative Establishment Survey
Berzofsky, M. E., Welch, B. L., Williams, R. L., & Biemer, P. P. (2006, August). Evaluation of Using a Model-Assisted Sampling Paradigm Versus a Traditional Sampling Paradigm in a Nationally Representative Establishment Survey. Presented at Joint Statistical Meetings, Seattle, WA.
National inference from a survey is traditionally based upon probability sample selection and survey weighting. For studies requiring estimates for a variety of rare subgroups with variable eligibility and response rates, a model-assisted approach might be considered to better control the subgroup sample sizes. Representative inference under the model-assisted sampling paradigm can be achieved using quotas combined with model-based weighting that does not depended on probability weighting. Using data from a national survey of establishments, we simulated a model-assisted paradigm and evaluated if the estimates were consistent with those under the traditional paradigm. Our findings suggest that the model-assisted approach offers advantages over the traditional sampling approach. For our purposes, a hybrid approach that captures the major advantages of both paradigms proved to be optimal.