Investigating the post-complaint period by means of survival analysis
Introduction
According to Keaveney (1995), the two major reasons why customers switch service providers are: (1) core service failures and (2) unfavorable service encounters with the company's personnel. When customers face a problem they may respond by exiting (=switching to another provider), loyalty (=staying with the supplier anticipating that ‘things will get better’) or voicing (=complaining to the firm or word-of-mouth to third-parties) (Levesque and McDougall, 1996, Tax et al., 1998). Unfortunately, it is only the tip of the iceberg that complains to the company (Stephens & Gwinner, 1998) since dissatisfied customers tend to remain passive when experiencing a problem (Bougie, Pieters, & Zeelenberg, 2003).
Customers who do not complain to the firm when dissatisfied are of special concern to management for several reasons. First, the company loses the opportunity to rectify the problem (Fornell and Wernerfelt, 1988, Levesque and McDougall, 1996) and to restore the customer's satisfaction level (Smith, Bolton, & Wagner, 1999). Second, the firm's reputation can be damaged due to the negative word-of-mouth to friends, family or other people external to the customer's social circle, e.g. via newspapers (Bougie et al., 2003; Singh, 1988) which might result in the loss of prospects as well as current customers (Stephens & Gwinner, 1998). Third, the firm is deprived of valuable information about its products and services (Fornell & Wernerfelt, 1987) that is likely to improve the bottom-line performance and to prevent similar problems in the near future.
On the other hand, customers who complain and receive a proper response to their service failures are more likely to stay (e.g. Conlon & Murray, 1996), to buy new products (e.g. Maxham III & Netemeyer, 2003), to pay price premiums (e.g. Zeithaml, Berry, & Parasuraman, 1996), to engage in favorable word-of-mouth and to recommend the company's services to others (e.g. Maxham, 2001, Maxham and Netemeyer, 2002). Furthermore, they show higher share-of-wallet behavior (e.g. Bowman & Narayandas, 2001) as well as higher commitment and trust towards the company (e.g. Tax et al., 1998). Finally, they are less vulnerable to switch (e.g. Bougie et al., 2003) and less likely to spread negative word-of-mouth to friends (e.g. Blodgett, Granbois, & Walters, 1993), or third-parties, such as other customers (e.g. Zeithaml et al., 1996). In sum, there is overwhelming evidence from previous research that successful complaint handling results in favorable customer outcomes. Additionally, in their study Fornell and Wernerfelt (1988) state that the return on investment in complaint management is likely to reach a 400 percent level.
When considering the consumer complaint behavior (CCB) literature, Stephens and Gwinner (1998) argue that much of the research is dominated by studies trying to understand why customers complain. In their paper, they provide an exhaustive list of investigated antecedents, including individual characteristics, attitudes, situational factors, the cost of complaining, etc. It is only since the last decade that literature has caught up by investigating the consequences of complaint handling (cf. previous paragraph). However, current knowledge is limited in providing insights regarding behavioral intentions or self-reported actual behavior measures resulting from critical incident technique studies in which the respondents are requested to think about their latest service switch (e.g. Keaveney, 1995). As is well known, data on actual behavior are often unavailable. Generally, the only data available are the self-reported intentions of the individuals who completed a post-complaint questionnaire. Nevertheless, many authors argue that intentions are not always translated into subsequent behavior, since respondents typically do not have perfect information about changes that may occur in the future that may affect their behavior (Young, DeSarbo, & Morwitz, 1998).
Unlike previous research, we investigate the impact of complaint handling on customers' actual behavior instead of intended behavior (=perceptual information). As a consequence, our research setting implies the need to link complaint data with complainants' behavioral information that is stored in transactional databases.
In this study, we decided to investigate the complainants ‘next-buy’ decision. We believe that an effective purchase reflects actual retention behavior (Larivière & Van den Poel, 2004). In contrast to the studies that have investigated intended repeat purchasing behavior by questioning items such as ‘In the near future, I intend to buy new products’, we consider an actual product opening as a real and executed consequence of such an intention. The variable ‘next-buy’ expresses whether the customer has bought a new product during the observed period of analysis. The variable is operationalized as a time-varying dependent variable, in which the right-censoring situation is taken into account; that is, customers who have not bought a new product by the end of the observed period of analysis might do so in the future (that is, right-censoring).
Furthermore, we explicitly test whether the impact of complaint handling varies over time by means of survival forests, meaning that we allow for changes in the impact of complaint handling components on the customers' next-buy decision. In the context of complaints, it is plausible to assume that some effects, such as receiving compensation, fade out after a while. As such, we cannot use conventional ‘proportional’ models that assume stationary effects of the covariates throughout the observed window of observation.
In sum, we contribute to the existing CCB literature by presenting a framework of actual customer behavior in which we account for the right-censoring situation, and we explicitly test for the time-varying impact of explanatory variables by questioning the proportionality assumption.
The rest of this paper is organized as follows. In Section 2 we elucidate both the methodological underpinnings of the proportionality assumption and the survival forests technique. In Section 3, we present the data set and the explanatory variables under investigation. The study results and its implications are reported in Section 4. Section 5 concludes the paper and outlines some directions for further research.
Section snippets
Methodology
In this study we apply survival analysis techniques because our dependent variable is characterized by both a binary classification (‘buy’ or ‘no buy’) and a duration indicator for that purchasing (or censoring) event. First, we present the methodological underpinnings related to the proportionality assumption. Next, we elaborate on the survival forests technique that produces time-varying covariate estimates.
Empirical study
A Belgian financial services provider delivered the data for this study. Its customers have the opportunity to express their complaints to the company whenever they perceive a problem with respect to the products and services they possess. When a customer experiences a (product or) service failure, he has the possibility to formulate his complaint either directly towards the company's complaint department or to his banking agent who will pass it on to the complaint department. In the latter
Findings
The next paragraphs contain the findings of the study. First, we report survival estimates for satisfied versus dissatisfied complainants, and we benchmark them against a random sample of non-complainants. Then, we present the findings of the proportionality investigation and the time-varying parameter estimates resulting from the survival forests. Finally, we suggest how one can approximate non-proportionality in conventional Cox regression.
Conclusion and further research directions
This study investigates the post-complaint period of the customers of a financial services provider. Unlike previous CCB studies that mainly focused on behavioral intentions, we analyze the impact of complaint handling on actual purchase behavior. Moreover, the dependent variable (that is, the complainant's next-buy decision) is conceptualized longitudinally by means of a status and a duration indicator, such that we are able to account for right-censored observations. The primary purpose of
Acknowledgements
The authors would like to thank the anonymous company and its complaint handling department that supplied the data to perform this research study. Moreover, we are grateful to Leo Breiman for the public availability of the survival forests software.
References (35)
- et al.
The effects of perceived justice on complainants' negative word-of-mouth behavior and repatronage intentions
Journal of Retailing
(1993) - et al.
The shape of the hazard function for cancer incidence
Mathematical and Computer Modelling
(2001) - et al.
Maintenance scheduling under age replacement policy using proportional hazards model and TTT-plotting
European Journal of Operational Research
(1997) - et al.
Investigating the role of product features in preventing customer churn, by using survival analysis and choice modeling: The case of financial services
Expert Systems with Applications
(2004) Service recovery's influence on consumer satisfaction, positive word-of-mouth, and purchase intentions
Journal of Business Research
(2001)- et al.
Modeling customer perceptions of complaint handling over time: The effects of perceived justice on satisfaction and intent
Journal of Retailing
(2002) - et al.
BJ: An S-plus program to fit linear regression models to censored data using the Buckley–James method
Computer Methods and Programs in Biomedicine
(2001) - et al.
Customer attrition analysis for financial services using proportional hazard models
European Journal of Operational Research
(2004) Logistic regression using the SAS® System: Theory and application
(1999)- et al.
Survival probabilities (The Kaplan–Meier method)
British Medical Journal
(1998)