Successive Sample Selection and Its Relevance for Management Decisions



Frequently organizations collect data characterized by successive selections of consumers based on a sequence of consumer decision outcomes. For example, a database may contain prospective customers who were targeted through direct mail, prospective customers who have responded favorably to the direct mail, asking for additional information, customers who already purchased, and customers who already purchased repeatedly. Another example are surveys that map respondents onto one of a set of ordered stages such as brand awareness, brand consideration, brand purchase and repeat purchase. Those data are collected because they are expected to help management answer the following questions: a) who are the customers on different stages and how likely are they to progress to the next stage?, b) how to select customers likely to reach later stages of the process when the decision maker cares both about probabilities and the covariate pattern targeted?, and c) what actions are likely to move customers to the next stage of the process? Popular approaches to answering these questions analyze stages independently, or collapse the multi staged process to one stage. We develop a model for the joint analysis of multiple stages allowing for dependence between successive selections through correlated errors. Our model, in contrast to independent analyses, controls for successive selection across multiple stages and, in contrast to collapsing multiple stages, correctly reflects the non-linearity resulting from intermediate selection stages. We show that what seem to be similar models at the outset translate into radically different answers to questions b) and c) above. We also show that answers to question a) are relatively more robust across different models. We use simulated data as well as a large scale survey investigating brand awareness, consideration, purchase and repeat purchase for fashion brands for illustration.
Contact information:
Dr. G. Liberali