A Bayesian approach to nonresponse in sample surveys

Graduation date: 1979A common problem in postal surveys of human populations\ud arises when some of the sampled people fail to respond to the questionnaire.\ud The problem is that the answers provided by the respondents\ud may not be typical of the answers the nonrespondents would\ud provide if these had been solicited more successfully. This creates a\ud "nonresponse bias. " One solution is to subsample the nonrespondents\ud intensively and obtain their answers by some more costly method\ud (e.g., phone calls or personal interviews).\ud This thesis presents a Bayesian approach to the prediction of a\ud finite population total when there is nonresponse. It is assumed that\ud the population is grouped on the basis of one or more known auxiliary\ud variables, that the characteristic of interest is linearly related to the\ud auxiliary variable(s), and that the probability of response is constant\ud within a group but possibly varies among groups