Abstract:
|
As ever-increasing pressure to do more with less affect probability-based surveys, nonprobability sampling can offer faster results at less cost. However, when population inference is a critical fit-for-purpose criterion, nonprobability surveys alone may not have the statistical rigor required. Even so, a nonprobability sample may serve to augment domains that are underrepresented in a probability survey. For example, if interviewing a required number of subpopulation members is resource prohibitive, data from a targeted nonprobability survey may serve to lower coverage bias exhibited in a probability survey. In this situation, the question is: what is the best method to combine information from both sources?
This research searches for answers to this question through an on-going evaluation of estimation methods that combine probability and nonprobability data. The goal is to identify procedures that maximize strengths of each data source to produce hybrid (combined) estimates with sufficiently low mean square error. Empirical results are presented using data from a national marijuana study.
|