Proceedings of the Survey Methods Section CALIBRATION ALLOCATION OF SAMPLE FOR MULTIPLE CHARACTERISTIC SURVEYS UNDER STRATIFIED RANDOM SAMPLING

This paper examines the problem of sample allocation for a stratified random design given multiple characteristics of interest. Existing solutions use non-linear programming to obtain a minimum cost and optimal allocation for a given set of variance constraints. If the resulting cost is not acceptable, the existing solution relaxes all variance constraints uniformly (i.e. allows uniform tolerance) such that an acceptable cost can be met. This is not reasonable, as not all of the constraints have an impact of the same order. An alternative formulation is required where there is a direct control on cost and the objective function is defined in terms of the variance constraints such that the two formulations become equivalent. Such a formulation is proposed using the idea of penalized distance function as in the ridge-calibration of sampling weights in Rao and Singh (1997). The proposed method is termed calibration allocation and allows for suitable, non-uniform tolerances when cost is reduced