Abstract

We will show how calibration weighting can be employed to combine probability and nonprobability samples in a statistically-defensible manner where the population from which the nonprobability sample has been drawns drawn is a subset of the full population. To this end, we assume the probability of a population element self-selecting into the nonprobability sample can be modeled as a bounded logistic function of variables collected on the nonprobability sample having either known or probability-sample-estimated population totals. This may or may not assume self-selection into the subpopulation covered by the nonprobability sample. Estimating self-selection probabilities with a calibration equation is the key to creating weights for the nonprobability sample and then for the blended sample. Estimates generated by the probability, nonprobability and blended samples will computed and their standard errors measured using the WTADJX procedure in SAS-callable SUDAAN. The results are then compared to the totals they are estimating.