Analyzing Duration Times in Marketing: Evidence for the Effectiveness of Hazard Rate Models

Abstract

Some statistical methods developed recently in the biometrics and econometrics literature show great promise for improving the analysis of duration times in marketing. They incorporate the right censoring that is prevalent in duration times data, and can be used to make a wide variety of useful predictions. Both of these features make these methods preferable to the regression, logit, and discriminant analyses that marketers have typically used in analyzing durations. This paper is intended to fulfill three objectives. First, we demonstrate how decision situations that involve durations differ from other marketing phenomena. Second, we show how standard modeling approaches to handle duration times can break down because of the peculiarities inherent in durations. It has been suggested in recent marketing articles that an alternative to these conventional procedures, i.e., hazard rate models and proportional hazard regression, can more effectively handle duration type data. Third, to investigate whether these proposed benefits are in fact delivered for marketing durations data, we estimate and validate both conventional and hazard rate models for household interpurchase times of saltine crackers. Our findings indicate the superiority of proportional hazard regression methods vis-à-vis common procedures in terms of stability and face validity of the estimates and in predictive accuracy.