Abstract:
|
Probability sampling has been the standard basis for inference from a sample to a target population. In the era of big data and increasing data collection costs, however, there has been growing demand for estimation methods to combine probability and nonprobability samples in order to improve the cost efficiency of survey estimation without loss of statistical accuracy (or perhaps even with improvements in statistical accuracy). An array of methods for combining probability and nonprobability samples are found in the literature, which we have classified into the following methodological groups: calibration, statistical matching, super-population modeling, and propensity-based weighting. In addition, NORC researchers have developed a hybrid calibration method that incorporates "borrowed strength" methods from small area estimation in order to explicitly account for bias associated with the nonprobability sample. We compare and contrast the nonprobability weights and estimates derived from all the methods from food allergies survey data, which were collected via both a probability sample and a nonprobability sample.
|