Predictive approaches to customer loyalty: the impact of missing data on the predictability of customer loyalty models

Large quantities of observations are utilised to predict and capitalise on customer loyalty. Yet virtually all gathered datasets contain some form of missingness. Incomplete cases may arise for a host of reasons, such as dropouts during an online survey for store card holders, inactive devices or partial WiFi coverage for in-store analytics, missing date of birth or other demographics on loyalty program signup forms, and so forth. The status quo in predictive approaches to customer loyalty is to ignore or discard such incomplete cases. However, this would diminish the external validity of the predictions and may substantially bias the derived results. Data are missing for a reason. Clients who skip certain sections of a survey do so for a reason and might share common traits. Similarly, clusters of incomplete observations in a company’s internal loyalty database could be caused by underlying technical or operational issues. This chapter explores missing data in customer loyalty research in order to proactively assess and handle incomplete observations. Three types of missingness are defined and differentiated. Ad hoc, likelihood, and chained equation approaches are discussed and theoretically as well as empirically compared. Lastly, the chapter provides hands-on techniques to solve missing data problems in customer loyalty research.