Cost efficency of nonresponse rate reduction efforts : an evaluation approach

Sammanfattning: Nonresponse, the failure to obtain complete measurement on all sample elements, is an increasing problem in sample surveys. Efforts to reduce the nonresponse rate are expensive and their effect may not be as desired. To allocate the limited resources to where they achieve the most error reduction, it is important to consider the cost-efficiency of efforts. This thesis proposes an approach to evaluate the cost-efficiency of the nonresponse rate reduction efforts for a general survey setup, assuming a direct element sampling design. A theoretical framework, under which both the error and the cost can be evaluated, is introduced. The framework extends the existing notion of response probabilities as fixed quantities by introducing a sequence of response probabilities, taking into account the actions and efforts taken by the survey administration during the data collection period.In addition to the introductory part, the thesis consists of four papers. In the first paper, the basic theoretical framework for evaluation is developed and expressions for the nonresponse bias of some common reweighting estimators are derived under this framework. Furthermore, expressions for the bias difference are given, general expressions as well as expressions under different specific scenarios. In the second paper, expressions of the total variance are derived for the same estimators and under the same framework as in the first paper. In the third paper, the basic theoretical framework in the first two papers is extended to incorporate survey costs, and a cost efficiency evaluation approach is introduced for an ideal situation where all required quantities are known. The approach consists of three steps, involving pairwise comparisons of cost-efficiency for time points during the data collection period and evaluation of point estimator bias and variance. Two different measures of the cost efficiency are proposed and compared, theoretically and through a small numerical study. In the fourth paper, the evaluation approach is extended to situations that may occur in practice. In the more favorable situations considered, unknown quantities are estimated through Monte Carlo simulations.Simulations are also used in the cases where unverifiable assumptions must be made about unknown quantities, as this facilitates comparisons of the effect on conclusions under different assumptions.