Combining Probability and Non-Probability Sampling Methods: Model-Aided Sampling and the O*NET Data Collection Program

This paper presents a brief synopsis of the historical development of hybrid sampling designs that combine traditional probability based sampling techniques with non-probability based quota designs to create model-aided sampling (MAS) designs. The MAS approach is illustrated for an application to a national business establishment survey called the Occupational Information Network (O*NET) Data Collection Program. Through simulation, we provide evidence that the estimates for this survey were not substantively biased by the MAS approach while data collection costs were substantially reduced. For reference in this paper, a model-based sample design uses a model to create quotas in various categories thought to be related to the study variables of interest and then uses a non-random sampling mechanism to obtain the requisite number of respondents in each quota cell. Inference is based solely on the model. By contrast, a model-aided sample design (1) starts with a probability based sample of units combined with quotas to ensure minimum and maximum respondent sample sizes. A combination of the probability based design and the model are used for inference.