Parameter Extrapolation in Marketing Research

Abstract

One of the most important intentions of marketing planning is inferring about the entire customer population from customer samples. In many cases relevant data from recent marketing tests concerning small, representative customer samples are being used. Response bias may occur due to nonresponse or delay in customer responses. This article supposes that it is possible to recognize the above mentioned bias in structure data which are known for the entire customer population. In these cases extrapolation procedures should use the information from structure data to correct classical estimates. This contribution discusses a technique that allows linear weighting of sample objects so that the total sum of squares between parameters in the target population and their estimates based on the sample are being minimized If the sum of weights is 1, one can prove that the weighted estimation is unbiased for independent, identically distributed random variables. This article features two methods of determining weighting vectors and discusses their characteristics.