Survey populations are considered rare if they are difficult to locate or to recruit. Designing a sample to hopefully capture enough members for analysis may produce a prohibitively-expensive survey without guaranteeing the recruitment goals. Consequently, the study may not be fit for the purposes needed.
Dual-frame estimation using specialized sources is one avenue to address the problem. Dual-frame random digit dial surveys supplemented with ethnic surname lists is one such example. Probability samples are not the only avenue for responses from the hard-to-reach populations. Nonprobability surveys such as those conducted via social media members also may be a viable option. However, how might we accurately combine both types of sources when dual-frame estimation was developed for probability surveys?
The goal this on-going research is to identify methodologies that maximize information from probability and nonprobability sources to produce hybrid estimates with sufficiently low mean square error. We evaluate several weighting methodologies including propensity scores and those based on the Bayes method using existing survey data.