Although customer complaints are a well-studied aspect of business, no study has measured the impact of actual complaints and recoveries on subsequent customer purchasing. The authors develop a customer base model to investigate the effectiveness of recovery in preventing customer churn. They calibrate it on panel data that track actual purchases, complaints, and recoveries for 20,000 new customers of an Internet and catalog retailer over 2.5 years. Complaints are associated with a substantial increase in the probability that the customer stops buying, but the size of the increase depends on prior customer experiences: prior purchases mitigate the effect, and their impact is long-lasting, whereas prior complaints exacerbate the effect, but their impact is short-lived. Thus, unless the customer leaves the company after a complaint, or a second failure occurs shortly after the first, the relationship quickly returns to normal. Recovery counters the effect of the complaint but, in almost all cases, does not entirely offset it. The authors use simulation to translate the results to financial impact and discuss implications for researchers and managers.
We derive a general structure that encompasses important coefficients of interrater agreement such as the S-coefficient, Cohen's kappa, Scott's pi, Fleiss' kappa, Krippendorff's alpha, and Gwet's AC1. We show that these coefficients share the same set of assumptions about rater behavior; they only differ in how the unobserved category proportions are estimated. We incorporate Bayesian estimates of the category proportions and propose a new agreement coefficient with uniform prior beliefs. To correct for guessing in the process of item classification, the new coefficient emphasizes equal category probabilities if the observed frequencies are unstable due to a small sample, and the frequencies increasingly shape the coefficient as they become more stable. The proposed coefficient coincides with the S-coefficient for the hypothetical case of zero items; it converges to Scott's pi, Fleiss' kappa, and Krippendorff's alpha as the number of items increases. We use simulation to show that the proposed coefficient is as good as extant coefficients if the category proportions are equal and that it performs better if the category proportions are substantially unequal. (PsycINFO Database Record
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.