Journal of Retailing 84 (2, 2008) 195–204
Customer complaining: The role of tie strength and information control Vikas Mittal a,1 , John W. Huppertz b,2 , Adwait Khare c,∗ a
Jones Graduate School of Management, Rice University, P.O. Box 2932, Houston, TX 77005, United States b School of Management, Union Graduate College, Schenectady, NY 12308, United States c School of Business, 1 Faculty Office Building, Quinnipiac University, CT 06518, United States
Abstract We examine the impact of two key constructs, information control and tie strength, on consumers’ likelihood of complaining following service failures. We report convergent results from three types of studies—an experiment, a survey, and secondary data. In the first study, tie strength and information control were systematically varied in an experiment using a restaurant scenario. In a second study, survey data from patients who experienced dissatisfactory service was collected. The third study used field data from 1,470 customers of an HMO. Results from all three studies showed that, following service failure, complaining is more likely when the tendency for information control is stronger and ties are weaker. © 2008 Published by Elsevier Inc on behalf of New York University. Keywords: Customer complaint; Service quality; Customer loyalty; Tie strength; Customer satisfaction; Service recovery
Studies show a high level of variability in the extent to which consumers complain following service failure. A study by USA Today (2000) showed that only 11% of the customers in the health care sector complain after a negative service encounter. TARP (1996) found that the percent of customers who did complain after service failures ranged from 42% for travel and leisure industry to 17% for consumer goods. This suggests that complaining behavior following a service failure is best understood as a contingent behavior. This paper contributes to the literature on service failure and complaining by exploring two specific contingency factors: tie strength and information control. Tie strength is the potency of the bond between the customer and service provider. Information control is a person’s tendency to manage a social exchange by exerting control over the quality and quantity of information exchanged. These constructs are explained in detail later on. We present three studies that examine how these two constructs jointly affect consumer complaining. The first study is an experimental study in a restaurant context with high internal validity. The second study is a survey of primary care patients that balances internal and external validity in a general healthcare setting. The third study is based on secondary data provided
∗ 1 2
Corresponding author. Tel.: +1 203 582 8285. E-mail address:
[email protected] (A. Khare). Tel.: +1 713 432 0703. Tel.: +1 518 388 8738.
by an HMO and it replicates the results of the first two studies but with a behavioral measure of complaining. Consumer complaining: hypotheses Retailers are interested in examining the antecedents of complaint behavior because their investment in service recovery programs can only succeed if consumers voice their complaint to management. The strategic importance of managing customer complaint behavior was established in Fornell and Westbrook (1984) who showed that encouraging complaints today may enable a firm to reduce future complaining activity. It is, therefore, important to understand factors that can systematically affect consumer complaining. Richins (1983, Table 1: 71 found that complaining was related to problem severity and attributions of blame, but not to the effort involved in complaining. Singh (1990) showed that attitudinal and personality variables rather than demographics predicted complaining behavior, with consumers classified as passives, voicers, irates, and activists. Not surprisingly, only 40.9% of the passives complained compared to 83.6% of the voicers and 71.8% of the activists (Singh 1990). More recently, Chebat et al. (2005) showed that consumers’ redress-seeking tendency moderates the extent to which consumers complain. They show that emotions, particularly anger, directly affect complaining behavior but that the magnitude of the impact of these emotions is moderated by the redress-seeking propensity. Specifically, the
0022-4359/$ – see front matter © 2008 Published by Elsevier Inc on behalf of New York University. doi:10.1016/j.jretai.2008.01.006
196
V. Mittal et al. / Journal of Retailing 84 (2, 2008) 195–204
Table 1A Study 1: Correlation Matrix Complaint likelihood
Tie strength (1 = strong, 0 = weak)
Information control
Satisfaction
Need for cognition
−0.15** 0.23* −0.46* 0.12+
– −0.06+ 0.09+ 0.01+
– 0.02+ 0.20*
– −0.03+
–
Mean
2.43
0.50
3.77
3.96
5.13
SD
1.66
0.50
0.95
1.53
1.09
Complaint likelihood Tie strength (1 = strong, 0 = weak) Information control Satisfaction Need for cognition
–
Notes: (+) ns, *p < .01, **p < .05; n = 181.
impact of anger is strongest when consumers have high levels of redress-seeking tendencies. Importantly, they found that consumers high on redress-seeking tendency also complain more. Voorhees and Brady (2005) tested a model of complaint intentions but found no moderating role for the extent to which customers are comfortable lodging a complaint. Building on this research, we examine two antecedents of complaining: tie strength and information control. Tie strength Tie strength is defined as the potency of the bond between members of a network (Granovetter 1973). Operationally, however, tie strength can be measured using a variety of variables such as such as the importance attached to the social relation (Ibarra 1997; Keister 1999), frequency of social contact (cf., Granovetter 1973; Nelson 1989), and the type of social relationship between the members of a dyad (Brown and Reingen 1987)3 . Tie strength, in turn, has been found to affect a number of outcomes such as word-of-mouth propagation (Brown and Reingen 1987), inter-group conflict (Nelson 1989), and career advancement (Ibarra 1997) in organizations. In service contexts, it is reasonable that consumers can develop ties – of varying strengths – with service employees. For example, a patient may have a stronger tie with her regular primary care physician than with a specialist she only has to see occasionally. Even within the same physician’s office, the strength of the tie that a patient has with the physician or the nurse may vary. Similarly, a restaurant patron can have differing tie strengths with different waiters at the same restaurant4 . How does the strength of the tie between the consumer and service employee affect the likelihood that the customer, after experiencing a service failure, will complain to the firm? In most situations, consumers perceive their complaint to the firm as having negative consequences for the service employee, which
3 The goal of this paper is not to develop antecedents of tie strength, but to look at the consequences of tie-strength using established measures and manipulations. 4 Tie-strength differences between a consumer and service employee may also signify the potency and depth of the commercial friendship (Price and Arnould 1999).
can also affect the relationship between the consumer and the service employee. If the strength of the tie between a consumer and service employee is weak, consumers will be less affected by any such consequences and therefore should be less hesitant to complain. In contrast, when the tie is strong, consumers fearing negative consequences for their relationship with the service employee should be less likely to complain. Additionally, strong ties should typically involve stronger and more positive priors about the service employee (Rust et al. 1999). Such positive priors should insulate the service employee from the negative effects of a single service lapse that consumers can attribute to chance (Anderson and Sullivan, 1993). In contrast, when the tie strength is weak the insulating buffer from such positive priors is unlikely to be present, and hence there should be a higher likelihood to complain following a negative service encounter. Based on these two lines of logic we posit: Hypothesis 1. After a service failure, the likelihood to complain to the firm will be higher when the strength of the tie between the consumer and service employee is weak than when it is strong. Information control We define information control as a person’s tendency to manage a social exchange by exerting control over the quality and quantity of information via actions such as asking for information, seeking clarifications, and withholding information relevant to the exchange relationship. This definition explicitly acknowledges that information control is a behavior-based construct manifesting itself in specific behavioral patterns such as increased information elicitation via questioning and clarifications or changing the form and content of information (e.g., using technical language to impart information) in social contexts. The construct of information control emerged in the medical sociology literature examining information exchange between patients and care-providers (Waitzkin 1985). Analyses of verbal protocols of patient–physician interactions show that physicians typically have more control over the content and form of information exchanged during the interaction (Ong et al. 1995). Such control on information may emanate from use of technical vocabulary, more directive behaviors including asking questions and unidirectional actions. As passive recipients of information,
V. Mittal et al. / Journal of Retailing 84 (2, 2008) 195–204
patients have relatively less information control. Their primary mode of exerting information control occurs in the form of asking questions and seeking clarifications (Kettunen et al. 2000). Cartwright (1964) found that more educated and higher income patients sought information more actively than less educated and low-income patients, even though the desire to obtain information was similar across all groups. While it is possible that a mere desire for information acquisition may be positively correlated with information control, they are qualitatively different constructs. Specifically, while information acquisition is an indicator of information control, information control may not be the sole antecedent of information acquisition. When a service failure occurs, it is logical that those having a relatively higher information control tendency will also be more likely to complain. We argue that this will occur for two reasons. First, complaining may reduce uncertainty for high-information control customers who can use it as an opportunity to provide feedback to the firm about the service employee. The act of complaining, in other words, affords the customer an opportunity to reduce uncertainty by telling the organization what happened, seeking information on the cause of service failure, making the appropriate causal attributions about the chain of events leading to service failure (Smith et al. 1999), and ascertaining the redress mechanisms. Naturally, this type of uncertainty reduction would be more appealing to high-information control customers. Second, and simultaneously, the act of complaining may also enable high-information control customers to attenuate the perceived power imbalance between the customer and the service employee. Immediately following the service failure, the consumer may experience a sense of helplessness and the act of complaining can restore a perception of control (Hirschman 1970). In fact, voicing complaint about the service employee can be a potent coping mechanism enabling consumers to restore a sense of equity and balance (Oliver and Swan 1989). Again, correcting the power imbalance should be more important to high-information control customers. Based on both these arguments, we posit our next hypothesis: Hypothesis 2. After a service failure, consumers having a higher information control tendency will be more likely to complain about the service employee to the firm than consumers having a lower information control tendency. Tie strength and information control We also posit that tie strength and information control will jointly influence the likelihood to complain about a service employee. As argued earlier, just as information control attempts act to reduce uncertainty on the consumer’s part, strong ties reduce uncertainty in the buyer–seller exchange relationship (Dwyer et al. 1987). Strong ties between a customer and a service employee inevitably come with strong positive priors and typically reduce uncertainty about the performance of the service employee. Following a service failure, this should mitigate the customer’s likelihood to complain about a service employee with whom strong ties already exist. However, when consumers have a weak tie with the service employee, their motivation to
197
complain will be more susceptible to their information control tendencies, and so higher information control tendencies will lead to relatively higher complaint likelihood. This is likely to happen because in the case of weak ties, not only is uncertainty high (because of weaker positive priors) but also customers are motivated to reduce uncertainty by activating information control tendencies. In other words, weak tie strength with the service employee will not inhibit the consumer’s tendency to exert information control via complaining behavior. On the other hand, when the tie between the service provider and customer is relatively strong, the consumer is less likely to act on his information control tendencies. Therefore, we posit that when ties are strong customers are less likely to complain to the firm, irrespective of their information control tendencies. The above argument suggests: Hypothesis 3. For high-information control customers, the likelihood to complain about a service employee following a service failure will be higher for those with weak ties compared to those with strong ties with the service employee. However, among those with low information control complaining behavior will be similarly low irrespective of tie strength. Study 1 Subjects and design A total of 181 undergraduates at a large northeastern university participated in this study for course credit. Each student read a scenario describing a service failure at a restaurant (see Appendix A). There were two versions of the scenario, each corresponding to high and low tie strength with the waiter. After reading the scenario, participants rated their overall satisfaction and likelihood to complain. Then, they filled out a battery of items measuring information control and need for cognition (Cacioppo and Petty 1982). The presentation order of items pertaining to these two scales was randomized. Finally, respondents filled out manipulation checks and demographic information. Dependent variables The key dependent variable in this study is likelihood to complain. We measured it using a 7-point scale (1 = very unlikely, 7 = very likely) in response to the following question: How likely are you to go to the manager and complain about this waiter at Toni’s? Independent variables Tie strength: Tie strength is manipulated within the scenario. The scenario corresponding to strong tie between the service provider and the customer is shown in Appendix A. In the strong tie-strength version respondents were told that: You have been coming back to the same restaurant for many months and the same waiter has been waiting on you. As you enter the restaurant, the waiter immediately recognizes you
198
V. Mittal et al. / Journal of Retailing 84 (2, 2008) 195–204
Table 1B Study 1: regression analysis for likelihood to complain about service employee Independent variable
Dependent variable: likelihood to complain (1 = very unlikely, 7 = very likely)
Intercept Tie strength (1 = strong, 0 = weak) Information control Tie strength × information control Need for cognition Satisfaction
Coefficient
p-Value
3.73 −1.92 0.18 −0.59 0.04 −0.51
<.01 <.05 <.10 <.01 ns <.01
Model significance: R2 = 0.32; F(5, 175) = 16.46, p < .01; n = 181.
and escorts you to a table. After having received an update on your current events, he takes your order. In contrast, in the weak tie-strength version participants were told that: You have been coming back to the same restaurant for many months. You enter the restaurant and a waiter escorts you to a table. Then the waiter takes your order. We administered a 4-item manipulation check of tie strength at the end of the study. The items (Appendix A) were averaged to form an index of tie strength (α = .91). With respect to the service provider, that is, the waiter at the restaurant, those in the strong tie condition had a higher mean than those in the weak tie condition (4.60 vs. 1.87, p < .01). Information control: Information control is a measured construct. We developed items using established scale development processes. The scale development is described in the next section. Here we note that the items had acceptable internal consistency (α = .78) and were therefore averaged. Need for cognition: To establish discriminant validity and assess the nomological net we also administered items that pertained to the need for cognition construct. The items are described in Appendix A and are similar to the short form used in Wood and Swait (2002). The items displayed adequate internal consistency and they were averaged to form a single measure (α = .77). The measure was reverse coded so that high scores indicate higher need for cognition. We note that in later analyses, the results are virtually identical using items that were not reverse-coded. Overall satisfaction: Because of its strong impact on complaint likelihood (Oliver 1997) we included overall satisfaction in the model for the sake of completeness. Two items measured overall satisfaction (Appendix A). They were averaged to form a single measure of satisfaction (r = .57). Results Information control (measure development and purification) We developed a pool of items based on studies reporting verbal protocols of patients in medical settings (e.g., Ong et al. 1995; Waitzkin 1984). We then conducted qualitative interviews with 10 students asking them to describe attitudes and behaviors that may correlate with a tendency to exert information control. They were specifically asked to think of consumption settings such as a class-
room, restaurant, purchasing a product, or consuming a product after purchase. From this we generated ten items that were included in a pretest study administered to 85 students. An exploratory factor analysis enabled us to reduce them to eight items that were included in the final study. In the final study, one item loaded poorly on the factor and was deleted from the analysis leaving us with the seven items shown in Appendix A5 . It could be that those having higher need for cognition may also have a higher tendency for information control. It may also be that information control is not conceptually distinct from need for cognition. Thus, it would be important to demonstrate discriminant validity between these two constructs. Thus, the purified items for information control were included in a factor analysis along with the need for cognition items. Reassuringly, all items loaded on their respective constructs. Discriminant validity was assessed using the two-step approach recommended by Anderson and Gerbing (1988). A confirmatory factor analysis showed that a two-factor solution provides a better fit than a one-factor solution with the incremental chi-square being significant at p < .05. Each item was significantly related to the relevant construct, but not to the other. Factor loadings for items related to both constructs are shown in Appendix A.
Hypotheses tests We ran a regression analysis with tie strength as a nominal factor (1 = strong, 0 = weak) and information control as a continuous factor. The model included the main effects of both these factors and their interaction. In addition, we included need for cognition and satisfaction as covariates for the sake of completeness. The dependent variable was likelihood to complain. Results, descriptive statistics, and correlations are shown in Table 1B. Hypothesis 1 posits that those in the high tie-strength condition will have a lower likelihood to complain than those in the low tie-strength condition. The coefficient for tie strength is negative and statistically significant (β = −1.92, p < .05). Thus, Hypothesis 1 is supported. Hypothesis 2 posits that those having a higher information control tendency are more likely to complain. A marginally significant coefficient for information control (β = 0.18, p < .10) is partially supportive of Hypothesis 2. Hypothesis 3 posits an interaction between tie strength and information control: the differential impact of tie strength is greater for those having high information control, but not for those having low information control. A statistically significant and negative coefficient for the tie strength × information control interaction (β = −0.59, p < .01) supports Hypothesis 3. We used a mean split6 on information control to create four subgroups and plotted the mean complaint likelihood to visually illustrate the results (see panel A of Fig. 1). Among those having high information control, complaint likelihood is higher in the weak tie condition compared to the strong tie condition (3.00 vs. 2.27, p < .05). Among those having low information control, the mean score is identical in the strong and weak tie conditions (2.20 vs. 2.21, ns). This pattern of results fully supports Hypothesis 3.
5
We also ran the analyses after including the deleted item in the information control scale. The results were unchanged. 6 The pattern of results is identical when a median split is used.
V. Mittal et al. / Journal of Retailing 84 (2, 2008) 195–204
199
experienced a negative service encounter in a healthcare setting. Replicating these results in a different service setting should increase confidence in our results.
Study 2 Sample A survey was administered to graduate students enrolled in two universities in the Northeast. Participants were asked to fill the survey with respect to a dissatisfactory visit to a doctor. They were asked to reflect about a visit, in the past year, to their doctor where they had been dissatisfied and answer the questions based on that particular encounter. A total of 350 participants got the survey. Among them, 127 indicated having no dissatisfactory experience with their doctor, and thus no further data were collected from them. Among the remaining 223 participants, 33 surveys were unusable due to missing data. Thus, the final analysis is based on 190 surveys of dissatisfactory service encounters between a patient and a doctor. Measures Fig. 1. Panel A: impact of tie strength and information control on complaint likelihood (Study 1) and panel B: impact of tie strength and information control on complaint likelihood (Study 2). Follow-up analysis: To ascertain if need for cognition is similar to the proposed information control construct, we replaced information control with need for cognition and re-ran the model. Neither the main effect nor the interactive effects of need for cognition were significant (p > .10). This provides further evidence of discriminant validity between need for cognition and information control constructs by establishing that both constructs have unique consequences. Whereas complaint intent is systematically affected by information control, it is not affected by need for cognition.
Discussion These results support Hypothesis 1, Hypothesis 2, and Hypothesis 3. Fig. 1 shows that the differential effect of tie strength manifests only among those having a higher level of information control. However, Study 1 – despite high internal validity – lacks external validity. It is based on a convenience sample of students responding to a hypothetical scenario using only a measure of complaint intention. In Study 2 we conducted a survey of consumers who had actually
Likelihood to complain: This was measured on a 7-point scale (1 = very unlikely, 7 = very likely) in response to the question: “How likely are you to complain to the medical director or practice manager about your experience with this doctor?” Tie strength: Tie strength was measured using three items each measured on a 7-point scale: How close do you feel to the doctor? (1 = not close at all, 7 = very close); How strong is your tie to this doctor (1 = very weak, 7 = very strong); How familiar are you with this doctor (1 = not familiar at all, 7 = very familiar)? The items had high reliability and were therefore averaged (α = .85). Information control: Respondents rated their agreement with the following items on a 7-point scale (1 = strongly disagree, 7 = strongly agree): I like to be well informed on issues I am concerned with; In class, I never hesitate to ask questions to my instructors; I feel uncomfortable when my doubts remain unanswered; When I visit my doctor, I often have many questions but feel shy asking him (reverse scored); I generally do what the doctor tells me without asking too many questions (reverse
Table 2A Study 2: correlation matrix
Complaint likelihood Tie strength Information control Satisfaction Relationship duration
Complaint likelihood
Tie strength
Information control
Satisfaction
Relationship duration
– −0.69* 0.53* −0.14+ −0.15**
– −0.33* 0.11+ 0.06+
– −0.29* −0.24*
– 0.00+
–
Mean
2.50
5.22
4.65
2.11
3.00
SD
1.78
1.63
1.25
1.12
4.35
Notes: +ns, *p < .01, **p < .05; n = 190.
200
V. Mittal et al. / Journal of Retailing 84 (2, 2008) 195–204
Table 2B Study 2: OLS regression analysis for complaining likelihood Independent variable
Intercept Tie strength (1 = strong, −1 = weak) Information control (1 = high, −1 = low) Tie strength × information control Satisfaction (1 = low, 5 = high) Relationship duration (years)
Dependent variable: likelihood to complain (1 = very unlikely, 7 = very likely) Coefficient
p-Value
2.45 −0.50 0.50 −0.19 0.02 −0.04
<.01 <.01 <.01 <.01 ns <.05
Model significance: R2 = 0.62; F(5, 184) = 59.37; p < .01; n = 190.
scored); and, asking questions gives me a feeling of being in control. The items exhibited satisfactory reliability (α = .82) and were therefore averaged. High values on this measure indicate higher information control. Encounter satisfaction: Respondents were asked: How did you feel about your experience at the doctor’s office on this particular occasion (1 = very dissatisfied, 7 = very satisfied)? Relationship duration: We also measured how long the respondent had been seeing the doctor with whom they had the negative encounter. This was measured in years. The descriptive statistics and correlations are shown in Table 2A. Note also, that in a confirmatory factor analysis, the items for tie strength and information control loaded significantly on their respective constructs. They were also nonsignificant for the other construct. A two-factor model fit the data better than a one or three factor model supporting discriminant validity for these two constructs. Model and results The following regression model7 was estimated using OLS: Likelihood to complain
a mean split for tie strength and information control. The cell means for complaint likelihood are plotted in panel B of Fig. 1. Among those having high information control, complaint likelihood is higher in the weak tie condition compared to the strong tie condition (4.31 vs. 2.25). However, among those having low information control the difference is attenuated (2.22 vs. 1.40). These results fully support Hypothesis 3. Discussion Using survey data from patients who had a dissatisfying encounter with their doctor, we tested our hypotheses. Consistent with Study 1, our hypotheses are supported. However, in both these studies, we measure intent to complain rather than actual complaint behavior. In Study 3 we use actual complaint behavior to test Hypothesis 3, the key, moderating hypothesis. However, due to the secondary nature of the data, we only had access to less than perfect measures of tie strength and information control. In summary, the key objective of Study 3 is to replicate the interaction obtained in Hypothesis 3 in a way that broadens the applicability of these results by using actual complaint behavior as the dependent measure.
= β1 (tie strength) + β2 (information control) + β3 (tie strength × information control) + β4 (satisfaction) + β5 (relationship duration) We mean centered all the variables to reduce multicollinearity. All VIF scores were less than 2 indicating that multicollinearity is not an issue. As reported in Table 2B, the overall model is statistically significant (p < .01). Next we discuss results pertaining to each hypothesis. The statistically significant and negative coefficient for tie strength fully supports Hypothesis 1 (β = −0.50, p < .01). The coefficient for information control is significant and positive (β = 0.50, p < .01). Thus, Hypothesis 2 is fully supported. As predicted in Hypothesis 3, the tie strength × information control interaction is negative (β = −0.19, p < .01). To visually illustrate the hypothesis we divided the data into four cells using 7
We also estimated a model without covariates. The results were virtually identical to the ones reported here.
Study 3 Data and measures The research was conducted among users of an HMO in a large U.S. metropolitan area. Survey data were initially collected from 2,464 customers who participated in a survey about their healthcare quality. After their participation, the HMO tracked their behavior on various aspects, one of them being whether they complained to the HMO about their healthcare encounter. A total of 1,470 customers indicated experiencing a service failure. As discussed later, the information control construct is validated using the sample of 2,464 customers, but the analysis for testing Hypothesis 3 involves the sub-sample of 1,470 customers who experienced a service failure. Measures For this analysis the sub-sample of 1,470 customers experiencing a service failure was examined to test the effect of tie strength and information control on complaint behavior.
V. Mittal et al. / Journal of Retailing 84 (2, 2008) 195–204 Table 3A Study 3: logistic-regression for demographic antecedents of information control in healthcare service Independent variables
Intercept Age Severity Satisfaction with health status Decision involvement Did you get an adequate explanation for your queries? (1 = yes, 0 = no) Did you ask what to expect during your next visit? (1 = yes, 0 = no) How long did you spend discussing issues with your medical provider?
Dependent variable: information control (1 = customer called HMO to make an inquiry, −1 = customer did not call) Coefficient
p-Value
−0.17 0.00 0.06 0.08 0.18 −0.15
.58 .91 .13 .21 <.01 <.01
0.12
<.01
0.10
<.05
Model significance: change in likelihood χ2 = 47.89; p < .01; n = 2464.
Tie strength: Tie strength was coded as strong (coded as 1) if the patients dealt with their primary care physician (PCP) during the focal encounter, and weak (coded as −1) if they dealt with another service provider (e.g., a specialist). Typically, in an HMO setting, patients have multiple encounters with their PCP leading to a stronger tie with the PCP than with other healthcare providers. This coding of tie strength based on encounter frequency is consistent with the one used in previous research (e.g., Brown and Reingen 1987). Information control: Information control was measured by ascertaining if the customer had called the HMO to make inquiries, to obtain information about benefits coverage, or both. Those who had initiated such a contact were coded as having high information control (coded as 1) and those who did not call were coded as having low information control (coded as −1). Customer-initiated contact with the HMO is a behavioral operationalization of information control. In this study, it is important to rule out alternative constructs or explanations for customer-initiated contacts. A logistic-regression was conducted to predict information control. The predictor variables and the results are summarized in Table 3A. The overall model is statistically significant (p < .01). These results enable a test of rival hypotheses regarding the disposition of the behavioral measure of information control. Most notably, age (p = .91), severity of medical condition (p = .13), and satisfactions with health status (p = .21) are all statistically non-significant predictors of information contact. This finding rules out severity or dissatisfaction alone as manifesting in customer-initiated contacts. It also suggests that it is not just the older patients who initiate contact, ruling out the hypothesis that older patients call more frequently because they have more time or more severe medical problems. Reassuringly, those initiating contacts also have higher involvement in medical decisions made regarding their medical therapy (β = 0.18, p < .01), are more likely to ask their
201
medical provider what they can expect during their next visit (β = 0.12, p < .01), and spend more time discussing issues with their medical provider (β = 0.10, p < .05). They are also less likely to feel that the explanations given for their queries were adequate (β = −0.15, p < .01). Finally, we note that item 6 (see Appendix A) comprising the information control scale in Study 1 specifically refers to calling the toll free number of the company to seek clarification about the product purchased. These results indicate that in this study, customer-initiated contacts (CICs) are an appropriate operationalization of information control. In this data, CICs are not a manifestation of factors such as age, dissatisfaction, and severity of medical condition. Strategically, CICs as a measure of information control is important because a firm can readily track this behavior in its internal database and use it to segment the customer base (Bowman and Narayandas 2001). Quality of care: Quality of care is a single-item measure based on a 5-point scale. Respondents rated the quality of care they receive on a scale of 1 (low) to 5 (high). Though not directly relevant to the hypothesis testing, it is included for model completion, as it is likely to be a strong antecedent of patient complaining behavior. Complaining behavior: Complaining behavior was coded as 1 (complained) or 0 (did not complain). It is worth noting that the measure of complaint here is a behavioral measure and not a measure of intention. Moreover, the complaining behavior is coded based on the internal database of the HMO, rather than just a self-report measure. Results and discussion Table 3B summarizes results from a logistic-regression in which complaining behavior is modeled as a function of tie strength, information control, and their interaction. In addition, the perceived quality of care is included as a covariate. The overall model is statistically significant (p < .01). As expected, perceived quality of care influences complaint behavior (β = −0.51, p < .01). There is no main effect of tie strength (β = 0.09, ns)8 . Thus, Hypothesis 1 is not supported. Supporting Hypothesis 2 we find that those with high information control are also more likely to complain (β = 0.18, p < .05). The interaction between tie strength and information control is statistically significant (β = −0.33, p < .01). The proportion of patients complaining within each subgroup is shown in Fig. 2. A z-test of proportions shows that among those with high information control the proportion complaining is higher among those with weak versus strong ties (34.8% vs. 13.3%, p < .05). Among those with low information control the pattern is attenuated (17.1% vs. 21.8%, ns). These results fully support Hypothesis 3.
8 This is important as it also rules out condition-severity as an alternative explanation. One could argue that those seeing the specialists may have more severe condition than those seeing their regular PCP alone. If that were the case, then we should see a main effect of our operationalization of tie strength.
202
V. Mittal et al. / Journal of Retailing 84 (2, 2008) 195–204
Table 3B Study 3: logistic-regression analysis for complaining behavior Independent variables
Dependent variable: complaining behavior (1 = complained, 0 = did not complain)
Intercept Tie strength (1 = strong, −1 = weak) Information control (1 = high, −1 = low) Tie strength × information control Quality of care (1 = low, 5 = high)
Coefficient
p-Value
−0.63 0.09 0.18 −0.33 −0.51
<.05 ns <.05 <.01 <.01
Model significance: change in likelihood χ2 = 98.05; p < .01; n = 1470.
To extend the external validity of the key interaction reported in Studies 1 and 2, Study 3 used a sample of real customers facing service failure and actually complained to the firm. Though the measures of information control and tie strength are likely to contain substantial measurement error, we are able to replicate the key moderating hypotheses tested in Studies 1 and 2. It is also worth noting that any alternative interpretation of these behavioral indicators is unable to explain the entire interaction obtained. Thus, these results represent a very conservative and strong test of the moderating hypothesis. Discussion In two separate retail settings, consumer complaints – both behavioral intentions and actual complaining behavior – were affected by tie strength and information control. In examining these results, it is important to note complementarities among the three studies. The dependent variable in Studies 1 and 2 is complaint intention, while Study 3 measures actual complaint behavior. Study 1 is an experiment, Study 2 is based on survey data, and Study 3 combines survey measures with secondary data. Notably, all three studies support the key, moderating hypothesis. Fornell and Wernerfelt (1987) discuss how firms that do not get adequate feedback in the form of customer complaints can be caught in a downward spiral of deteriorating quality and a shrinking customer base. Our study shows that this impact can be systematically different for different customer
Fig. 2. Study 3: impact of tie strength and information control on complaint behavior.
groups. While consumers with high information control will voice complaints on their own volition, it is especially important to identify consumers with low information control and then solicit their feedback. As in Study 3, managers would benefit from identifying behavioral correlates of information control and targeting consumers with different levels of information control. Our results also indicate that strong ties with the service employee may deter consumers from complaining to the organization. Clearly, such suppression of feedback may leave retailers vulnerable to other competitors. Thus, ways of soliciting feedback without harming the customer–service employee relationship must be designed. This could be accomplished by designing complaint mechanisms where customers do not have to identify the employee associated with poor service. More generally, the construct of tie strength needs more investigation. Here we specifically focused on the customer’s strength of tie with the service employee, and not the retail service organization. While they may be related, they are clearly different constructs. How would customers’ tie strength with the service organization moderate the results obtained in this research? This is a question that deserves closer scrutiny especially as service firms develop strong brands and identities to get close to their customers. In the same vein, the distinction between tie strength and relationship commitment (Brown et al. 1995; Morgan and Hunt 1994), two constructs that are seemingly related but conceptually distinct, needs to be examined. It seems that strong ties can emanate from casual relationships, while committed relationships have a deeper psychological underpinning including emotional involvement. It could be that strong ties can result from casual relationships that may not have progressed to the extent of being a committed relationship. In fact, in many commercial relationships (including finding a good restaurant, finding a good doctor, etc.) casual acquaintances wield a lot of influence, both directly and indirectly (Granovetter 1973). Our paper applies the notion of social ties to examine complaining. Relationship commitment, to us, implies a much more intense level of social interaction than most retail service contexts afford. Yet, the empirical overlap and conceptual differences between these two constructs should be examined in future research. This article only assessed complaining behavior to the firm. Other forums for voicing dissatisfaction – other consumers,
V. Mittal et al. / Journal of Retailing 84 (2, 2008) 195–204
friends and family, internet forums, and regulatory agencies – need to be examined as well. Specifically, investigating the joint impact of information control and tie strength on negative word-of-mouth is a useful area of research. For instance, how do customers balance the need to maintain a close tie with a service employee and warning a close friend about potential bad service? One could also examine if tie strength and information control affect the forum where customers voice complaint—friend, firm, or regulator. With its root in classical sociology, the construct of information control holds promise for application in several marketing domains, especially in the information age. Key research priorities include: (1) articulating antecedents of this construct, (2) distinguishing it from constructs such as assertiveness and shyness, and (3) developing multi-dimensional measures that tap into the situational as well as trait aspects of this construct. In addition to complaining, outcomes such as regret (Tsiros and Mittal 2000) could be investigated. It could be that those having higher information control are more likely to be aware of the performance on forgone alternatives and therefore, experience higher level of regret. Overall, other antecedents beyond information control that may lead consumers to complain should be investigated to develop more comprehensive models of consumer complaining. As mentioned earlier, the finding that strong ties between customers and frontline service employees can inhibit complaint voicing should concern managers of retail firms. Even if the service failure was not the employee’s fault, the likelihood that the customer will complain is lower when tie strength is high. At a broader level, our findings pose a unique paradox that managers must resolve: on the one hand firms would benefit from stronger ties between customers and service employees. Yet, these strong ties can also inhibit information flow to the firm. The challenge for managers is to design ways to encourage customers to voice complaint despite strong ties to service employees. One solution may be to design alternative voice mechanisms that are either anonymous or perceived as neutral by customers. Another solution is to encourage complaining by guaranteeing that the employee who is being complained about will not be punished. Researching these issues also presents a nice research opportunity for scholars.
203
Information control (1 = strongly disagree, 7 = strongly agree, α = .78)
Factor loading
1
.84
2 3
4
5 6
7
In class, I always ask questions to clarify my doubts In class, I never hesitate to ask questions to my instructors A person should have control on the flow of information in any conversation In a restaurant, I always ask the waiter a lot of questions about the menu Asking questions gives me a feeling of being in control In the past, I have often called 1–800 numbers to seek clarifications about products I have purchased If I don’t understand something in class, I don’t feel shy asking for clarification
.85 .66
.59
.64 .51
.84
Need for cognition (1 = strongly disagree, 7 = strongly agree, α = .77)
Factor loading
1
.74
2
3
4 5
I try to anticipate and avoid situations where there is a likely chance that I’ll have to think in depth about something (reverse coded) The idea of relying on thought so as to get my way to the top does not appeal to me (reverse coded) I would rather do something that requires little thought than something that is sure to challenge my thinking abilities (reverse coded) I only think as hard as I have to (reverse coded) The notion of thinking abstractly is not appealing to me (reverse coded)
.61
.83
.75 .62
Overall satisfaction (r = .57) 1 2
How do you feel about your experience at Toni’s on this particular occasion? (1 = unhappy, 7 = happy) How satisfied are you with Toni’s handling of the delay in service? (1 = dissatisfied, 7 = satisfied)
Appendix A. Study 1 measures A.1. Scenario (high tie-strength version) Manipulation check: tie strength (α = .91)
You and your friend have decided to go out for dinner at an Italian restaurant called Toni’s. You have been coming back to the same restaurant for many months and the same waiter has been waiting on you. As you enter the restaurant, the waiter immediately recognizes you and escorts you to a table. After having received an update on your current events, he takes your order. After 30 min, you are still waiting for your food. Finally, after 45 min the waiter brings your food. The food tastes good and is reasonably priced, but the waiting time is too long.
1 2 3 4
How close, would you say, you are to this waiter at Toni’s (1 = not close at all, 7 = very close) How strong, would you say, is your tie to this waiter at Toni’s (1 = very weak, 7 = very strong) How familiar do you feel with this waiter at Toni’s (1 = not familiar at all, 7 = very familiar) “This waiter at Toni’s gives me special treatment.” (1 = disagree, 7 = agree)
204
V. Mittal et al. / Journal of Retailing 84 (2, 2008) 195–204
References Anderson, Eugene W. and Mary W. Sullivan (1993). “The Antecedents and Consequences of Customer Satisfaction for Firms,” Marketing Science, 12 (2) 125–143. Anderson, James C. and David C. Gerbing (1988). “Structural Equation Modeling in Practice: A Review and Recommended Two-step Approach,” Psychological Bulletin, 103 (3) 411–423. Bowman, Douglas and Das Narayandas (2001). “Managing Customer-initiated Contact with Manufacturers: The Impact on Share of Category Requirements and Word-of-Mouth Behavior,” Journal of Marketing Research, 38 (3) 281–297. Brown, Jacqueline J. and Peter H. Reingen (1987). “Social Ties and Word-ofMouth Referral Behavior,” Journal of Consumer Research, 14 (3) 350–362. Brown, James R., Robert F. Lusch and Carolyn Y. Nicholson (1995). “Power and Relationship Commitment: Their Impact on Marketing Channel Member Performance,” Journal of Retailing, 71 (4) 363–392. Cacioppo, John and Richard Petty (1982). “The Need for Cognition,” Journal of Personality and Social Psychology, 42 (1) 116–131. Cartwright, Ann (1964). Human Relations and Hospital Care, London: Routledge and Kegan Paul. Chebat, Jean-Charles, Davidow Moshe and Codjovi Isabelle (2005). “Silent Voices: Why Some Dissatisfied Consumers Fail to Complain,” Journal of Service Research, 7 (4) 328–342. Dwyer, Robert F., Paul H. Schurr and Sejo Oh (1987). “Developing Buyer–Seller Relationships,” Journal of Marketing, 51 (2) 11–27. Fornell, Claes and Birger Wernerfelt (1987). “Defensive Marketing Strategy by Customer Complaint Management: A Theoretical Analysis,” Journal of Marketing Research, 24 (4) 337–346. Fornell, Claes and Robert A. Westbrook (1984). “The Vicious Circle of Consumer Complaints,” Journal of Marketing, 48 (3) 68–78. Granovetter, Mark S. (1973). “The Strength of Weak Ties,” American Journal of Sociology, 78 (6) 1360–1380. Hirschman, Albert O. (1970). Exit Voice, and Loyalty: Responses to Decline in Firms, Organizations and States, Cambridge, MA: Harvard University Press. Ibarra, Herminia (1997). “Paving an Alternative Route,” Social Psychology Quarterly, 60 (1) 91–102. Keister, Lisa A. (1999). “Where Do Strong Ties Come From? A Dyad Analysis of the Strength of Interfirm Exchange Relations During China’s Economic Transition,” The International Journal of Organizational Analysis, 7 (1) 5–24. Kettunen, Tarja, Poskiparta Marita and Liimatainen Leena (2000). “Communicator Styles of Hospital Patients During Nurse–Patient Counseling,” Patient Education and Counseling, 41 (2) 161–180.
Morgan, Robert M. and Shelby D. Hunt (1994). “The Commitment–Trust Theory of Relationship Marketing,” Journal of Marketing, 58 (3) 20–38. Nelson, Reed E. (1989). “The Strength of Strong Ties: Social Networks and Intergroup Conflict in Organizations,” Academy of Management Journal, 32 (2) 377–401. Oliver, Richard L. (1997). Satisfaction: A Behavioral Perspective on the Consumer, New York: Irwin/McGraw-Hill. Oliver, Richard L. and John E. Swan (1989). “Equity and Disconfirmation Perceptions as Influences on Merchant and Product Satisfaction,” Journal of Consumer Research, 16 (3) 372–383. Ong, L.M.L., J.C.J.M. De Haes, A.M. Hoos and F.B. Lammes (1995). “Doctor–Patient Communication: A Review of the Literature,” Social Science and Medicine, 40 (7) 903–918. Price, Linda L. and Eric J. Arnould (1999). “Commercial Friendships: Service Provider–Client Relationships in Context,” Journal of Marketing, 63 (4) 38–56. Richins, Marsha L. (1983). “Negative Word-of-Mouth by Dissatisfied Consumers: A Pilot Study,” Journal of the Marketing, 47 (1) 68–78. Rust, Roland T., J. Jeffrey Inman, Jia Jianmin and Zahorik Anthony (1999). “What You Don’t Know About Customer–Perceived Quality: The Role of Customer Expectations Distributions,” Marketing Science, 18 (1) 77– 92. Singh, Jagdip (1990). “A Typology of Consumer Dissatisfaction Response Styles,” Journal of Retailing, 66 (1) 57–99. Smith, Amy K., Ruth N. Bolton and Janet Wagner (1999). “A Model of Customer Satisfaction with Service Encounters Involving Failure and Recovery,” Journal of Marketing Research, 36 (3) 356–372. TARP (1996), “TARP’s Approach to Customer Driven Quality: Moving from Measuring to Managing Customer Satisfaction,” Washington DC: White House Office of Consumer Affairs. Tsiros, Michael and Vikas Mittal (2000). “Regret: A Model of its Antecedents and Consequences in Consumer Decision Making,” Journal of Consumer Research, 26 (4) 401–417. USA Today 2000 “Poll: 51% Had Problems with Health Plan,” June 8. Voorhees, Clay M. and Michael K. Brady (2005). “A Service Perspective on the Drivers of Complaint Intentions,” Journal of Service Research, 8 (2) 192–204. Waitzkin, Howard (1984). “Doctor–Patient Communication: Clinical Implications of Social Scientific Research,” Journal of the American Medical Association, 252 (17) 2441–2446. Waitzkin, Howard (1985). “Information Giving in Medical Care,” Journal of Health and Social Behavior, 26 (2) 81–101. Wood, Stacy L. and Joffre Swait (2002). “Psychological Indicators of Innovation Adoption: Cross-Classification Based on Needs for Cognition and Need for Change,” Journal of Consumer Psychology, 12 (1) 1–13.