Social Policy Targeting and Binary Information Transfer between Surveys

This paper deals with the optimal transfer of binary information (BIT) on group membership between different statistical surveys of an identical population, a need arising frequently in socio-economic surveys. The limited number of questions asked in any one survey may necessitate information transfer between surveys. We design a method for a a BIT between a source-survey originally including the information and a target survey in which it is needed. An efficient BIT depends on (1) efficient estimation of the statistical model explaining group-membership as estimated by the ROC-curve, (2) the choice of a cutoff value for translating the forecasted logistic probability back into a binary variable and (3) a statistically testable quality control of the transfer. We suggest an optimal cutoff point that minimizes the sum of squared errors instead of the well-known Hosmer-Lemeshow method. Our application illustrates how survey data can be enhanced, when repeated interviews are expensive or difficult to implement. We enhance the Household Expenditure Survey (HES) by transferring a binary variable of households' religious group membership from the Social Survey to the HES. This helps identify extremely poor groups for poverty calculations and improved targeting of anti-poverty policy.