AN OPTIMAL METHOD OF BINARY INFORMATION TRANSFER (BIT) BETWEEN SURVEYS OF AN IDENTICAL POPULATION

This paper deals with the optimal transfer of information on group identification between different data sets of an identical population. Such a need might arise frequently in the analysis of socio-economic surveys and in the implementation of social and economic policy. Due to the limited number of questions asked in any given survey, analysts of one survey may need some binary information available in another survey of the same population. We suggest an efficient method for transferring such binary information between the survey in which the information is available (the "source-survey") and the survey in which the information is needed (the "target-survey"). We show that an optimal method for transferring the information depends crucially on two aspects of the process: (1) an efficiently estimated statistical model of of group-membership (the goodness of fit of the ROC-curve) and (2) the determination of the optimal cutoff value (the value chosen for turning the logistic probability forecast back into a binary variable). The proposed method can be useful in the social sciences, in medical research and in any field that requires probabilistic binary data enrichment between data sets drawn from a common population. It is an efficient method for ex-post enhancement of given data sets, when such enhancement by use of other methods is either expensive or impossible. In this paper we perform a BIT on membership in the Jewish Ultra-orthodox community in Israel, known as an extremely impoverished group. Since the information on religious group membership exists only in the "Social Survey" and information on poverty exists only in the Income Survey or the Expenditure Survey, there arises a need for optimal BIT. We present the results of such an optimization and its importance for social policy.