Journal of Interactive Marketing

A failure to distinguish between the goal of regression-based segmentation modeling as being “predictive” rather than “explanatory” is viewed as the prime catalyst behind the unwarranted model intervention that occurs in the direct response industry today. Moreover, such “model meddling” is not merely an innocuous indulgence, but can result in opportunity costs or real-dollar losses in terms of the utilization of less-than-optimal models. The inability to interpret correctly the meaning of the regression coefficient values is another factor which can lead to the unjustified modification of demonstrably valid segmentation models, as is made evident with an example. Finally, the premature deletion of potentially viable predictors due to certain misconceptions surrounding the harmful effects of “multicollinearity” is also cited as one more example of inappropriate model intervention.

This article develops formulas for panel size for designed experiments. Using a simple three-attribute problem, the author discusses situations for determining panel sizes in the context of the multiple regression and logistic regression formulations. Modeling member net present value and response rate as functions of the attributes of an acquisition program are the two situations considered. The article also has a pedagogical flavor and discusses how the maximum likelihood approach is used in estimating the variance of the partial logistic regression coefficients. Illustrative examples are provided to show the application of the formulas. For marketers, the bottom line result is useful: testing with designed experiments requires far fewer observations than classical tests of hypotheses, where pairwise comparisons are carried out. This is because all the information is used simultaneously to estimate model coefficients. For a non-technical overview, see the end of the article.

The need to maximize advertising effectiveness and minimize communication costs has increased direct marketers’ and packaged-goods advertisers’ reliance on individual-level consumer information. The use of such information, however, has raised a number of questions regarding consumer privacy. This study therefore attempts to increase direct marketers’ understanding of privacy issues by examining how well informed consumers are with respect to information gathering and use practices. The results of a regional survey of 266 adults 18 years old and older suggested: 1) privacy is an important concern; 2) many people are not very knowledgeable about specific direct marketing practices; 3) consumer concern is affected by type of practice and specificity of information; and 4) most favor restrictions on the gathering and use of personal information. Overall, the results suggest consumer ignorance may be a significant contributor to privacy concerns and that a strong committment to consumer education may be necessary to avoid government regulation and legislation.

The “K–S statistic” is a popular (in fact, almost a standard) measure of model strength for credit risk scoring models. This article defines the “K–S statistic” and explains how it is used in the context of testing statistical hypotheses. It also points out a common interpretation error made when using this statistic. This article was written with the credit marketer, who uses risk models in conjunction with his direct mail campaigns, in mind. But since any measure of risk model strength may also be used to measure the strength of a response model, it is hoped that this article is found useful by the rest of the direct marketing world who employ modeling to its advantage.