Surveys Combining Probability and Quota Methods of Sampling

Abstract

The problem of choosing between full probability and quota sampling is analyzed in terms of a Bayesian model of biased measurement. The model differs from previous models in its attempt to approximate the effects of the multistage cluster sampling used in actual national surveys. Allocation rules are developed that may involve devotion of all resources to quota sampling, to probability sampling, or to designs that combine both types of sampling. The sensitivity of the optimal allocation to the specification of the prior distribution and the cost parameters is investigated.