Journal of Marketing Research Vol. XLIV (February 2007), 153���163 153 �� 2007, American Marketing Association ISSN: 0022-2437 (print), 1547-7193 (electronic) *Murali Chandrashekaran is Professor of Marketing and Head of the Marketing Cluster (e-mail: muralic@agsm.edu.au), and Kristin Rotte is Assistant Professor of Marketing (e-mail: kristinr@agsm.edu.au), Aus- tralian Graduate School of Management, University of New South Wales. Stephen S. Tax is Associate Professor of Service Management, Faculty of Business, University of Victoria (e-mail: stax@uvic.ca). Rajdeep Grewal is Dean���s Faculty Fellow and Associate Professor of Marketing, Smeal Col- lege of Business Administration, Pennsylvania State University (e-mail: rug2@psu.edu). The first author thanks seminar series participants at the Graduate School of Business at Stanford University and at the Fuqua School of Business at Duke University. The first two authors also acknowl- edge that support for this research was provided by the Faculty Research Grant program at the Australian Graduate School of Management and Uni- versity of New South Wales. MURALI CHANDRASHEKARAN, KRISTIN ROTTE, STEPHEN S. TAX, and RAJDEEP GREWAL* Evidence reveals that many customers who state that they are satisfied with a service provider nevertheless defect. In this article, the authors focus on identifying which customers are vulnerable to defection despite their stated high levels of satisfaction. Building on the recently developed Judgment Uncertainty and Magnitude Parameters (JUMP) model, the authors decompose customers��� stated satisfaction into two related but independent facets���satisfaction level and satisfaction strength���and then examine the role of satisfaction strength in the trans- lation of satisfaction into loyalty. Using data from an ongoing customer satisfaction tracking study being conducted by a large U.S.-based service organization, Study 1 examines the role of satisfaction strength in shaping the satisfaction���loyalty link in a business-to-business setting. Study 2, a conceptual replication in a business-to-consumer context, examines the hypothesized relationships in a service failure/recovery situation. The studies strongly demonstrate that satisfaction strength plays a central role in the translation of stated satisfaction into loyalty. A key finding is that though satisfaction translates into loyalty when satisfaction is strongly held (i.e., low uncertainty), the translation is significantly lowered, on average, by approximately 60% when the same satisfaction is more weakly held (i.e., high uncertainty). The studies also indicate that prior relationship aspects (length of relationship, volume of business, and favorability of prior experiences) result in even greater vulnerability. Satisfaction Strength and Customer Loyalty Even as the literature on customer satisfaction burgeons and practitioners make great strides in dissecting databases to know which customers are less or more profitable, antici- pation of defection remains an elusive goal. The starkness of this reality is underscored by anecdotal and empirical evidence that suggests that many customers who state that they are satisfied with a service provider subsequently switch to a competitor. For example, a study for the U.S. Office of Consumer Affairs (Technical Assistance Research Program 1986) finds that in households with service prob- lems, only 54% would maintain brand loyalty if their prob- lems were satisfactorily resolved. Similarly, Reichheld (1996) notes that 65%���85% of customers who defect report before defection that they were satisfied or very satisfied. In turn, a recent meta-analysis of customer satisfaction research finds that satisfaction explains less than 25% of the variance in repeat purchase (Szymanski and Henard 2001 see also Rust et al. 1999). In this article, we focus on understanding which cus- tomers are vulnerable to defection despite stating high lev- els of satisfaction. We are motivated by research that con- tends that the translation of satisfaction into loyalty depends on other variables. For example, Lemon, White, and Winer (2002) and Rust, Lemon, and Zeithaml (2004) maintain that many nonsatisfaction elements drive loyalty and illustrate that relational elements that increase switching costs are important factors in whether satisfaction has a strong rela- tionship to loyalty. We focus on an aspect of the satisfaction judgment itself that has received little attention in the litera- ture���namely, the strength with which the satisfaction judg- ment is held. Our central thesis is that satisfaction strength is a vital determinant of customer vulnerability because it plays a

154 JOURNAL OF MARKETING RESEARCH, FEBRUARY 2007 crucial role in the translation of stated satisfaction into cus- tomer loyalty. Examining the strength with which customer sentiments are held is important in the context of ongoing customer relationships and after critical events that may serve to destabilize customer relationships (e.g., service failures). Because service variability and service failures can sensitize customers to the potential downside of dealing with a service provider, they increase uncertainties and fears that can threaten the relationship and result in cus- tomers becoming vulnerable to defection. In turn, customer behavior hinges on successful resolution of these uncertain- ties. However, extant satisfaction research focuses almost entirely on the magnitude of judgments (e.g., the level of customer satisfaction) and ignores the contemporaneous conviction (i.e., the strength) with which customer judg- ments are held. As a result, researchers have not secured insights into how ongoing service delivery and service recovery actions crystallize into strongly held customer sen- timents (producing ���secure��� and loyal customers) or pre- cipitate sentiments that are largely fragile (producing vul- nerable customers who might profess favorable sentiments that may be impotent and laden with uncertainty). It would also be useful to understand whether various aspects of prior relational experience (e.g., length of relationship with the supplier, volume of business, favorability of prior experience) insulate customers from this vulnerability. In the next section, we propose a two-dimensional view of satisfaction that centers simultaneously on the satisfac- tion level and strength. We then assess our theorization in the context of two studies. Table 1 presents an overview of the two studies. The two studies, which cover different busi- ness contexts, different operationalizations of loyalty, and different aspects of prior relational experience, strongly converge to support our theory and illuminate the role of satisfaction strength and prior relational experience in the translation of satisfaction into loyalty. RESEARCH FRAMEWORK We build on the extant satisfaction and judgment forma- tion streams of literature to recognize that a customer���s overtly stated satisfaction is composed of two related but distinct dimensions: the level of satisfaction and the covert strength with which that satisfaction judgment is held. Sev- eral lines of thought support this two-dimensional concep- tualization of revealed satisfaction. For example, scholars in the area of services marketing note that customer expecta- tions are often fuzzy (e.g., Rust et al. 1999), and it is often difficult for customers to estimate precisely the level of service they received (Parasuraman, Zeithaml, and Berry 1985). Thus, it is likely that the resultant satisfaction judg- ments are laden with uncertainty. Consequently, customers are likely to differ in the strength with which they hold their satisfaction. This is also consistent with the psychological view of human judgments that Koehler (1994, p. 461) suc- cinctly expresses: ���Although we believe a great many things, we hold some of our beliefs with greater conviction than others.��� The characterization of satisfaction as a multifaceted ���statistical��� construct stridently resonates with the literature on modeling individual judgments as well. For example, Chandrashekaran and colleagues (2000 see also Chan- drashekaran, Rotte, and Grewal 2005) propose the Judg- ment Uncertainty and Magnitude Parameters (JUMP) model, which focuses on the simultaneous manifestation of uncertainty in the variation surrounding the central ten- dency of stated judgments. Grounded in the notions of Brunswikian uncertainty, which considers human judg- ments in a social context (see Juslin and Olsson 1997), the JUMP model takes an overtly stated measure and estimates the impact of independent variables on the magnitude and uncertainty that is inherent in the overt judgment. Like- wise, noting that the uncertainty in customer expectation judgments is reflected in the variance surrounding simple point expectations, Rust and colleagues (1999) caution researchers that it may be insufficient to measure only the point expectation, as has been standard practice instead, researchers should also measure uncertainty with respect to the level of service that will be received. In summary, therefore, we argue that a constructive way to view satisfaction is that it is based on fuzzy expectations and a fuzzy assessment of experience, which in turn is based on an understanding of the distributions of actions (and outcomes) of others. Similar to Rust and colleagues (1999), we suggest that by viewing satisfaction as a simple point assessment, extant research has immediately assumed that all that is important is the mean of these distributions. In turn, consistent with the JUMP model (Chandrashekaran et al. 2000 Chandrashekaran, Rotte, and Grewal 2005), we suggest that though the mean of this distribution is an index of the level of satisfaction, the variance of the distribution is related to the strength (i.e., conviction or certainty) with which this satisfaction is professed. Satisfaction Model Letting SATi denote the stated satisfaction of the ith cus- tomer, we recognize that SATi is a realization from a distri- bution of possible judgments, such that SLi, the satisfaction level, reflects the mean, and SUi, the satisfaction uncer- tainty, manifests itself in the variance of that distribution: (1) SATi = ��0 + SLi + ��i var(��i) = ��2 + SUi, and (2) SLi = Xi�� �� SUi = Zi��, �� where ��2 denotes the measurement- and model-error vari- ance Xi = [x1i, ..., xpi] and Zi = [z1i, ..., zki] denote row- vectors of variables hypothesized to affect satisfaction level and uncertainty, respectively and �� �� = [��1, ��2, ..., ��p] and �� �� = [��1, ��2, ..., ��k] denote column vectors of the impacts of Xi and Zi, respectively. The specific elements of Xi and Zi come from theory and the specific substantive setting of the particular research study. Consistent with the JUMP model procedure (Chandrashekaran, Rotte, and Grewal 2005), we can straightforwardly estimate the parameters of interest using feasible generalized least squares, in which (1) SAT is regressed with ordinary least squares on X to obtain esti- mates for ��0 and �� �� (2) the squared residual (i.e., e2 = [SAT ��� ��� X��]2), which is an estimate of the individual- level variance with the same asymptotic properties as ��i 2, is regressed on Z to obtain estimates and (3) e2 is regressed with weighted least squares (WLS) on Z with 2( + Zi��)2 as the weight and with the constraint that + Zi�� 0, where and are the WLS estimates of ��2 and �� and (4) SAT is regressed on X with a WLS model with + Zi�� as the weight to obtain unbiased and efficient estimates �� �� �� �� �� 2 �� �� �� �� �� �� 2 �� �� �� �� �� 2 �� �� �� 2 �� �� �� �� 2 �� �� �� 0