Description
Case Study on Examination of Whether and How Racial and Gender Biases Influence Customer Satisfaction, Customer satisfaction, a term frequently used in marketing, is a measure of how products and services supplied by a company meet or surpass customer expectation. Customer satisfaction is defined as "the number of customers
.
Case Study on Examination of Whether and How Racial and Gender Biases Influence Customer Satisfaction
ABSTRACT We examine whether and how various biases may influence customers' satisfaction evaluations and produce discriminatory judgments for minorities and female service employees. We argue that customer satisfaction evaluations are biased because they are anonymous judgments by untrained raters that usually lack an evaluation standard. In our laboratory and field samples, we find disturbing evidence generally confirming our arguments and suggesting that the presence of nonwhite and women service employees may produce lower aggregated customer satisfaction evaluations which may ultimately hurt individuals and organizations financially. (84 words)
1
Parts of this research were supported by the Business and Economic Development Center, Foster School of Business, University of Washington. We would like to thank the AMJ action editor Peter Bamberger as well as three anonymous reviewers for their help in the review process.
-2-
Customer satisfaction surveys have become a common source of performance feedback for employees and organizations (Hagan, Konopaske, Bernardin, & Tyler, 2006). Mercer Consulting Group reports that in 2006 customer satisfaction surveys were of primary importance for strategic decision making and over two-thirds of organizations used such surveys to determine some aspect of employee compensation (Mercer Consulting, US Policies and Practices Report, 2007). Moreover, customer satisfaction is an important predictor of a wide range of financial measures (see Gupta & Zeithaml, 2006 for a review) so it is not surprising that some companies tie some portion of employee compensation directly to customer satisfaction. For example, a one percent change in customer satisfaction for an average Fortune 500 firm has been shown to lead to a 1.02% change in Tobin's q which equates to a change of $275 million in firm value (Anderson, Fornell & Mazvancheryl, 2004), a $55 million gain or loss in cash flow in the next year (Gruca & Rego, 2005), and a 5.03% change in Return on Investment (Anderson & Mittal, 2000). Satisfying the customer is also increasingly important to organizations as the global economy becomes more service-oriented. Macroeconomic trends indicate 76 percent of U.S. employees work in a service industry, and by 2016 the number of employees working in a service industry is expected to increase by over 17 million (Figueroa & Woods, 2007). The expanding service sector is perhaps one reason why among 681 senior executives surveyed by The Economist Intelligence Unit during October-December 2002, 65% reported customers as their main focus over the next three years compared to only 18% who reported shareholders as their main focus. Many business leaders (Bracken, Church & Timmreck, 2001) and researchers (Salam, Cox & Sims, 1997) have applauded the use of customer satisfaction surveys because they believe that aggregated evaluations are highly reliable measures of employee performance quality. However, a potential disadvantage of using customer surveys, particularly for purposes of making compensation or
-3-
promotion decisions, is that they are ultimately subjective judgments. As a result, they are vulnerable to biases, including those based on the bandwagon effect, confirmation of pre-existing beliefs, education or cognitive ability, as well as stereotypic biases based on the race or gender of the person being rated (Gilovich, Griffin & Kahneman, 2002). Some researchers suspect that biases are unavoidable when gathering subjective evaluations of performance, especially when such judgments come from naïve and inexperienced raters who are not held accountable for the accuracy of their ratings (Pulakos, White, Oppler & Borman, 1989; Wilkinson & Fontaine, 2002; Woehr & Roch, 1996) To date, though, surprisingly little research has examined how and if different biases influence customer judgments about whether an organization's employees, organizational attributes, or services meet or exceed their expectations. The purpose of our research is to examine whether and how customer satisfaction ratings are potentially influenced by gender and racial bias. We extend the existing literature on biases in supervisory ratings of employee performance by focusing on customer satisfaction ratings, which have been mostly excluded from organizational behavior research (see Moshavi, 2004 for a rare exception). We conceptualize these satisfaction ratings as judgments and we examine not only judgments made about individual employees, but also about the organizational context (e.g., perceived cleanliness or appearance) and the organizational unit as a whole. To our knowledge, no research has examined bias in customer judgments of the organizational context or the overall organizational unit. Based on the literature on modern forms of racial and gender bias (Greenwald & Banaji, 1995; Crandall & Eshelman, 2003), we contend that customer satisfaction with organizational contexts or units may be vulnerable to such biases. Finding empirical evidence of racial or gender biases in customer satisfaction would suggest that, from the organization's perspective, there is a financial incentive to
-4-
favor white or male employees. Such a finding may help explain the persistent inequality between demographic groups in the workplace. Two methodological attributes of our research distinguish it from previous studies of bias in customer judgments of employee performance and allow us to conduct a stronger test of the validity of our predictions. First, we take into account employees' objectively-measured performance behaviors when examining customer judgments. The problem with relying solely on customer satisfaction scores to assess customer bias is that such scores can be interpreted as capturing both true performance scores and biases (Latham & Wexley, 1977; Landy, Shankster, & Kohler, 1994). As Rotundo and Sackett (1999: 816) summarize this state of affairs, "There is no definitive way of determining whether the rated criterion used in a validity study is biased. Thus, there is no current method of establishing whether there is bias in performance ratings." Our study design allows us to tease apart differences in satisfaction judgments that are attributable to customer bias arising from employee demographic characteristics from those that are due to objective employee performance (Greenhaus, Parasuraman & Wormley, 1990; Pulakos et al., 1989; Wilkinson & Fontaine, 2002; Woehr & Roch, 1996). Second, we heed recent calls for researchers to engage in "full-cycle" research where initial field-based findings are tested in the laboratory and then re-validated in a different field setting (Chatman & Flynn, 2005; Cialdini, 1995). Full-cycle research allows researchers to compensate for the weaknesses of one context or study design with the strengths of another. It also allows the researcher to investigate a broad initial question in a field setting (e.g., is there bias in customer judgments?) followed by a laboratory study that can utilize more control and enable the researcher to examine more specific questions in detail (e.g., what is a potential cause and consequence of bias in customer judgments?). Finally, the investigator can move back to a field setting to confirm findings
-5-
found in the first two studies. The interplay of field and lab designs prescribed by the full-cycle approach fosters greater theoretical insight as to the causality and generalizability of study findings. Following the full-cycle research model, we first test for bias in customer judgments regarding a sample of professional employees (e.g., doctors). Next, in a carefully-controlled laboratory setting, we test for customer bias again, but this time in a bookstore context. We also identify and measure a specific mechanism that might explain the observed effect. Finally, we test for customer bias in judgments of an organizational unit with a sample consisting of country clubs belonging to a large hospitality company. Since the focal unit rated shifts from the individual in the first sample to the organizational in the third sample, we are able to provide some initial confirmation of the generalizability of our theory. In the following section, we present the theoretical rationale for our predictions regarding the possible effects of customer diversity related biases on customer satisfaction judgments. THEORETICAL BACKGROUND AND HYPOTHESES Customers are often asked to assess an individual service provider (e.g., their salesperson, teller, teacher, or physician; Haas, Cook, Puopolo, Burstin, Cleary & Brennan 2000; Sixma, Spreeuwenberg, & van der Pasch, 1998; Davis & Davis, 1999), the quality of the environment in which they were served (e.g., the merchandise available, the newness or cleanliness of the setting, or the efficiency of the technology; Pellegrin, Stuart, Maree, Frueh, & Ballenger, 2001; Simonet, 2005), or the unit or group providing the service (e.g., the bank, school, country club or law firm; Anderson et al., 2004; Ittner & Larcker, 1998). This last type of judgment is likely to include opinions about both the server(s) and the context in which the economic transaction occurs and is therefore a more global judgment than the other two types previously mentioned. In this study, we investigate the possibility of systematic bias in all three types of satisfaction judgments thereby providing a strong test of the
-6-
potential generality of such biases. An assumption of our paper is that like anyone else who makes a social judgment customers are not immune to information processing biases. One potential source of this bias arises from the demographic characteristics of the service providers who are being rated. Racial and Gender Biases in Customer Satisfaction Judgments U.S. society has made considerable progress in reducing overt expressions of prejudice since the Civil Rights movement during the 1960s (Bobo, 1998). Yet despite these gains, there is abundant social psychological evidence that biases against women and minorities persist in a more covert and non-conscious form. Researchers have used terms like modern racism (McConahay, 1983), aversive racism and sexism (Dovidio & Gaertner, 1981; Gaertner & Dovidio, 1986), or implicit gender and racial stereotypes (Greenwald & Banaji, 1995) to describe these types of biases and many studies have demonstrated their influence on information processing and judgment across a variety of social domains (see Brief, Dietz, Cohen, Pugh & Vaslow, 2000 for an overview). For example, one study showed that when job applicant resume quality was ambiguous, applicants with African-Americansounding names (e.g., Aisha, Rasheed) were much less likely to be called for a job interview than applicants with white-sounding names (e.g., Kristin, Brad) (Bertrand & Mullainathan, 2004). Similarly, when evaluators of orchestral position applicants could see the applicant's gender they were more likely to select men. When the applicant's gender could not be observed, the number of women hired significantly increased (Goldin & Rouse, 2000). In another study, Dovidio and Gaertner (2000) found that while raters were not biased against blacks in a simulated hiring decision when the applicants were clearly qualified or unqualified for a job, raters were biased when the applicant's qualifications were ambiguous. Dovidio and Gaertner (2000) interpreted this finding as supporting an aversive racism framework in which prejudice occurs in a more subtle form in contexts with ambiguity or uncertainty. Based on considerable evidence demonstrating the operation of covert and unconscious
-7-
racial and gender biases across a variety of social domains, there is reason to suspect that such biases can also influence customer satisfaction judgments. Our theoretical arguments supporting the influence of racial and gender biases in customer satisfaction judgments are based on the idea that observers (e.g., customers) have preconceived expectations about others depending on whether the person being observed belongs to a high or lowstatus demographic group (Berger, Conner & Fisek, 1974; Berger, Fisek, Norman & Zelditch, 1977). In the United States, men and whites are considered by most people as members of a high-status social group relative to women and ethnic minorities (see Ridgeway, 1991 for a review). One of the benefits of belonging to a high-status social group is that observers are more likely to make favorable inferences about one's competence, normality, and legitimacy (Aquino & Bommer, 2003; Giannopoulos, Conway & Mendelson, 2005; Sidanius & Pratto, 1999). In contrast, members of lowstatus groups are subject to negative stereotypes and attributions concerning their work-related competencies (Fernandez, 1981; O'Leary & Ickovics, 1992). Based on rating theory (Wherry & Bartlett, 1982), we believe there are at least three reasons to expect customer satisfaction ratings to be susceptible to such stereotypes and racial and gender biases. First, an important difference between performance ratings made by supervisors and judgments made by customers is that usually customers are afforded the luxury of anonymity. Although supervisors' and customers' ratings are both viewed by organizational administrators, in most cases only supervisors are identifiable. Customer anonymity decreases accountability and the desire to engage in the effortful cognitive processing required to conceal or overcome any biases (Richman, Kiesler, Weisband & Drasgow, 1999; Lerner & Tetlock, 1999). Moreover, supervisors, but not customers, know that their ratings are part of the employee record and are used for employee training, feedback, and advancement decisions (Murphy, 1991). Not only are supervisors identifiable, but they must also
-8-
justify their ratings and as such are even more motivated to engage in effortful information processing to help them reduce the influence of racial or gender bias and appear, at least superficially, to be objective. Customer anonymity does not motivate raters to reduce bias, but the customer satisfaction questionnaire instructions and items may actually strengthen the effect of such biases. Supervisors completing a performance appraisal are frequently reminded of the importance of rating accuracy and rate on specific behavioral items (Judge & Ferris, 1993; Kane, Bernhardin & Villanova, 1995). Customers, however, are typically asked only for their "opinions" or "attitudes" about employees or the organization (suggesting they "make a judgment") (Schneider, Ehrhart, Mayer, Saltz, & NilesJolly, 2005). Common customer satisfaction items like "I would recommend this organization to others" and "this organization/employee meets my expectations" do not solicit recall of specific employee or organizational behaviors and so they may provide more information about the state of mind of the rater than the actual performance of the ratee or the organization. Such items are also problematic because customers may have higher expectations for women and non-whites to demonstrate their competence and so will provide lower customer satisfaction judgments for such workers even if their performance is objectively equivalent to that of their male or white counterparts (Biernat & Kobrynowicz, 1997; Yarkin, Town & Wallston, 1982). A variety of techniques have been shown to effectively reduce bias in performance appraisal. For example, Roch and O'Sullivan (2003) found that a combination of frame-of-reference (i.e. having raters establish a prototype of good performance) and behavioral observation training (i.e. focusing raters on specific behaviors) leads to increased accuracy in appraisal. Baltes and Parker (2000) found that halo error training (i.e. knowing what factors should not influence ratings) and structured recall memory intervention (i.e. memory enhancement techniques) reduce bias in performance ratings.
-9-
DeNisi, Robbins, and Cafferty (1989) argued that behavioral diaries aid in the recall and categorization of behavioral events. Customers are not trained or expected to use these techniques when forming satisfaction judgments. In sum, customer satisfaction judgments are likely to be highly susceptible to racial and gender biases because customers are usually anonymous, asked to make summary judgments rather than provide accurate recall of performance-related behaviors, and are untrained in techniques that might help them overcome non-conscious biases. Our arguments lead us to expect that, in general, employees belonging to low-status demographic groups (i.e., women, racial minorities) will receive lower customer satisfaction scores than employees belonging to high-status demographic groups. But even if we found evidence for this difference, it would not demonstrate the operation of bias in customer judgments because it may be that the lower customer satisfaction judgments received by members of lower status demographic groups may in fact indicate lower levels of true performance. Logically, customer satisfaction judgments should be at least partly influenced by employee objective performance (Wherry & Bartlett, 1982). Consequently, more direct evidence of bias would be demonstrated if behaviors performed by a high-status group member are viewed more favorably by customers than the same behaviors performed by someone from a low-status group. Evidence for the plausibility of this hypothesis comes from studies showing that women in leadership roles are rated lower than men in similar roles (Eagly, Makhijani & Klonsky, 1992) and that ethnic minorities and women are rewarded less than whites and males for exhibiting the same advice-giving or ingratiatory behaviors (Westphal & Stern, 2007). There is also evidence that racial minorities and women who achieve equivalent levels of performance as whites and men are judged as having less underlying ability (Biernat & Kobrynowicz, 1997; Yarkin, Town & Wallston, 1982). To examine whether the
-10-
biases in judgments of competence as a function of a target person's group membership found in prior research generalize to customer judgments, we tested the following hypothesis: Hypothesis 1: The relationship between employee objective performance and customer satisfaction judgments will be attenuated for employees belonging to low-status demographic groups compared to employees belonging to high-status demographic groups. Hypothesis 1 deals with customer judgments that ask for evaluations about an individual service provider. But we also believe that racial and gender biases can influence evaluations of the organizational context (e.g., cleanliness, appearance). Our prediction draws from the notion that the positive or negative properties of an item or person can "spill over" on to the nearby context or surrounding targets (Rozin, Millman & Nemeroff, 1986). Our logic is based on the simple idea that evaluations of different aspects of service experience (e.g. the employee, the context) are connected in the rater's non-conscious and conscious belief system (Argo, Dahl, & Morales, 2009; Morales & Fitzsimons, 2007). Theoretical and empirical work on the cognitive structure of attitudes (Anderson, 1981; Wyer & Schrull, 1989) suggests that our evaluation of any person, object, or idea is partly based on our evaluations of other persons, objects or ideas with which the target object is linked. Recent work in marketing has elaborated on this idea (Keller, 2003). Research shows that the evaluation of a product, service or brand is partly related to the evaluation of the persons who are associated with the product, service or brand (e.g., the person in an advertisement, the service provider)(Folkes & Patrick, 2003; Simonin & Ruth, 1998; Morales & Fitzsimons, 2007). Accordingly, we expect that customers' evaluations of organizational contexts will be non-consciously connected to their evaluations of highly visible employees in the organization. Although such a non-conscious connection may be unwarranted, we expect customers' attitudes about low-status employees to be reflected in less favorable evaluations of an organizational context where such employees are highly visible.
-11-
It may even be possible for customers to make a conscious connection between employee status and the level of organizational context quality. (e.g., Rynes, Heneman, & Schwab, 1980; Rynes & Miller, 1983; Spence, 1973). In line with this thinking, job applicants believe that recruiters' competence and thoroughness signals an organization's overall quality (Rynes, Bretz & Gerhart, 1991). Customers may judge an organizational context that employs a low-status employee as signaling an inferior physical and social environment than one employing a high-status employee. Combining our spillover and signaling arguments, we expect the mere presence of members of lowerstatus groups in the organizational environment will lead to less positive customer judgments of the service environment. The following hypothesis tests this prediction: Hypothesis 2: Individuals will report lower customer satisfaction judgments of the environmental context when a highly visible employee in that environment belongs to a low (i.e., women, African-American) rather than high (i.e., male, white) status demographic group. Not every customer will be so susceptible to racial or gender bias that they would evaluate an employee and the environment more negatively simply because they encounter an employee who belongs to a low-status demographic group. We expect the customers who are most prone to making these types of judgments to be those who hold more negative pre-existing attitudes toward females or racial minorities. Individuals with negative pre-existing attitudes toward members of low-status groups quickly associate negative words (e.g. terrible, nasty, evil) with pictures of nonwhite or female faces, and quickly associate positive words (e.g. laughter, glorious, joy) with pictures of white or male faces (Greenwald & Banaji, 1995). We tested this possibility by examining whether the degree to which individuals have non-conscious, negative attitudes toward members of low-status groups would moderate the effect proposed in our first two hypotheses as stated in the following hypotheses:
-12-
Hypothesis 3a: Low-status employees (i.e., women, African-Americans) will receive even lower customer satisfaction judgments than equally- performing high-status employees (i.e., men, whites), when the judges have negative attitudes toward members of low-status groups. Hypothesis 3b: Contexts employing low-status employees will receive even lower customer satisfaction judgments than contexts employing equally- performing high-status employees, when the judges have negative attitudes toward members of low-status groups. Thus far we have argued that customer satisfaction judgments can be influenced by perceiving a single employee belonging to a low-status group within the service environment. But in many cases customers interact with a variety of employees in a given customer service encounter. For example, when doing something as simple as buying groceries, customers observe and interact with deli workers, produce employees, cashiers, and baggers. Employees in each of these positions likely have different demographic characteristics. For this reason, another way to examine the possible influence of bias in the customer judgment process is to see whether the demographic composition of the organizational unit might influence customer judgments of that unit. Extending our theoretical argument to the organizational level, we expect that the degree to which an organizational unit's employees are members of low-status demographic groups will influence customer satisfaction judgments of that organizational unit such that it will be judged as being of worse quality compared to an organizational unit whose employees are mostly from high-status demographic groups. But as we noted when predicting the influence of rater bias on judgments of individual service providers, such a finding might reflect true differences in unit performance. Hence, we make the following, more precise prediction that parallels hypothesis 1:
-13-
Hypothesis 4: The relationship between an organizational unit's objective performance and customer satisfaction will be attenuated as a function of the percent of unit employees belonging to low-status demographic groups (i.e., women and minorities) Our theoretical arguments and hypotheses can be summarized by the conceptual model shown in Figure 1. --------------------------Insert Figure 1 here --------------------------We tested the hypotheses in our model in three studies using different samples and methods. We tested hypothesis 1 by looking at customer judgments of their physician (Medcorp study); we tested hypotheses 1, 2 and 3 by examining customer judgments of a bookstore and one of its employees (Bookcorp study); finally we tested hypothesis 4 by examining customer judgments of their golf club (Golfcorp study). In each study, a large number of customers rated each of the targets. MEDCORP STUDY Our first sample was drawn from all 113 primary-care physicians (i.e., family practitioners) employed by a large health maintenance organization, hereafter referred to as Medcorp (a pseudonym). Medcorp provides coverage and healthcare for about 350,000 people in the Pacific Northwest region of the United States. Within our sample, 38.4% were women, 11.5% were ethnic minorities, and all had a medical degree. The 2006 Diversity Report by the Association of American Medical Colleges reports that 24.5% of practicing physicians are women and 12.1% are nonwhites.2 Measures Medcorp routinely collected patient satisfaction ratings as well as objective behavioral indicators of physician performance that were assumed to have a direct, positive impact on patient
2
However, the Association of American Medical Colleges also reports that these numbers are changing dramatically as 44% of American medical school graduates in 2006 were women and 34% were nonwhite.
-14-
health and well-being. This feature of our data represents a methodological improvement over studies that only measure employee performance with a single subjective rating and are therefore unable to determine if the rating is biased (Rotundo & Sackett, 1999). The dependent variable in our study was patient satisfaction with their physician. The independent variables were physician demographics (race and gender) and three types of patient-centered behaviors. Customer satisfaction. A postcard survey was mailed to a percentage of each physician's patients, following doctor visits, selecting the patients so as to avoid a bias toward those patients with frequent appointments. Patients completed and returned a total of 12,091 surveys for a response rate of 52%, so that each physician was rated by an average of 107 patients. Each patient rated only one physician, so the individual ratings were independent. Patients rated each of the three items targeting their physician on a 5-point Likert scale (1=very poor; 5=excellent) "How would you rate?" (1) the attention the provider paid to you; (2) this provider's thoroughness and competence, and; (3) your opportunity to ask questions of this provider. The three items were highly correlated (average correlation is .93) so the organization combined them to create a composite patient satisfaction variable. These items are more general questions rather than very specific behaviors (e.g. minutes spent with provider, number of questions the doctor asked). The organization did not provide us with access to raw patient-level surveys. Instead, they provided us with data indicating what percentage of each physician's patients rated the physician as "excellent." Thus, the range on this measure for each physician was from 0 to 100 percent. This measure was collected in the same quarter as all other variables. Physician race. Medcorp identified each physician's race, and we coded whites "0" and ethnic minorites as "1." Of the 113 physicians in the sample, 10 were Asian or Pacific Islanders, two were
-15-
black, and one was Native American. The percentage of ethnic minority physicians in our sample is consistent with the national average of 12.1%. Physician gender. We coded males "0" and females "1." Forty-three of the physicians were female, which is slightly higher than the national average of 24.5%. Objectively measured employee performance. With the growing prevalence of health maintenance organizations (HMOs) and the increasing corporatization of medicine (Feinglass & Salmon, 1990), patients are increasingly being viewed by organizational administrators and physicians as customers. Therefore, physicians are increasingly being rewarded for engaging in behaviors that benefit their patients and the organization's customers (Laine & Davidoff, 1996; Stewart et al., 2000). We used the customer-benefitting behaviors identified by Medcorp as our indicators of objective physician performance. Medcorp measures customer-benefitting behaviors along three dimensions. The first is physician productivity, which is the number of health procedures performed and issues discussed in a given time period. The second is the physician's accessibility to customers measured by the number of secure emails that doctors send to customers. The third is the physician's level of quality measured by the standardized prescription rates of particular medications for customers that possess precise disease criteria. All three dimensions reflect behaviors that benefit customers by reducing the amount of time and money customers spend receiving medical care. For all metrics, physicians are shown how they compare to both the organizational goal and the organizational average. More productive physicians are able to treat more customer problems per visit, thereby saving customers' time and trips to the doctor. More accessible physicians provide greater convenience to customers who can simply email any medical questions to their physician. Higher quality physicians are better at preventing costly and deadly health events such as strokes and heart attacks. Physician compensation is tied to each of these
-16-
customer-benefitting behaviors. Physicians who exceed the 40th percentile are given a bonus, while those below the 40th percentile are not given a bonus. Physician productivity. The average number of patients seen, medical issues discussed, and medical procedures performed by each doctor in a standardized 8-hour day was recorded by the organization's scheduling software. Medcorp physicians have a great deal of control over the amount of work that they do in a day as they can control the intensity of each visit (e.g., the number of procedures performed and patient health issues addressed per visit). The number of patients physicians see each day is controlled by organizational administrators. The objective performance assessment we used was the composite of average face-to-face visits and phone visits adjusted by the average intensity of each visit. Intensity was measured by Relative Value Units (RVUs), which are coded by physicians at the end of each visit according to national coding guidelines. RVUs capture the amount of time involved, the required physical and mental effort, the required judgment and technical skill, and the psychological stress entailed (Hsaio, Braun, Becker, & Thomas, 1988; Hsaio, Braun, Dunn, & Becker, 1988). Physicians check one of three RVU boxes after seeing each patient. If the patient appointment is a quick check-back or follow-up appointment, physicians check the first box, which is worth .5 RVUs. If the patient appointment involves at least two patient issues or concerns, but less than four, then the physician checks the middle box, indicating 1.0 RVUs for that visit. If the patient appointment involves five or more patient issues, then the physician checks the third box, which indicates 1.5 RVUs. According to quarterly audits by administrators, Medcorp physicians accurately record RVUs in 90 percent of patient visits. Coding errors resulting from physicians coding too many or too few RVUs are normally and equally distributed. The raw measure of productivity was standardized based on the full-time status of the
-17-
physician and then multiplied by each physician's average visit intensity to obtain the quarterly average RVU-adjusted patient encounters per day. Physician accessibility. The average daily number of secure emails that physicians sent to patients for the quarter was used to measure another type of customer-benefitting behavior. Patients highly value the ability to easily contact their physician. Indeed, a Harris poll has shown that 90% of Americans who are online want the ability to e-mail their physicians, and 37% are even willing to pay for it (Taylor, 2002). Medcorp patients and physicians can communicate electronically regarding health related issues through a secure Internet health portal designed exclusively for patient-doctor communication. To use the system, patients log in to a secure website that provides the patients access to their personal health records, their lab results, and a host of health related information. Patients can send unlimited secure emails through the portal to any physician they visited in the prior two years at no cost and Medcorp physicians are expected to reply to each secure email within 24 hours. Patients are encouraged to contact their physician via the system to ask basic health-related questions, to request prescription refills and to schedule follow-up appointments. Medcorp administrators assign an equal number of patients to each physician (taking into account patient sickness, age and gender) and think that the system saves patients doctor visits, thereby saving patients time and money. In general, physicians do not think that email improves the quality of patient care, but rather that secure email increases convenience for patients (Kleiner, Akers, Burke & Werner, 2002). The Medcorp computer server automatically recorded the number of emails that each medical professional sent to his or her patients. Medical professionals had a great deal of control over how many emails they sent for two reasons: (a) they could try to discourage patients from using the system, and (b) they could choose whether to personally respond to their patients' emails. We calculated the number of emails physicians sent per day, taking into account the number of full working days that
-18-
physicians were in clinic during the time-period of this study. To enhance the normality of the variable, we used an inverse transformation and then reflected these values such that higher values represented greater use (Tabachnick & Fidell, 2003). Physician quality. Every Medcorp primary care physician is responsible for a panel of member-patients. Of the thousands of possible treatments, prescriptions, and procedures that physicians can perform to benefit patients, one of the most important is each physician's prescription rate of statins and angiotensin-converting-enzyme (ACE) inhibitors to patients with cardiovascular disease. Treatment of cardiovascular events such as strokes, clots, and heart attacks is the biggest healthcare cost for patients in the U.S. (Willerson & Cohn, 2000), and these drugs prevent cardiovascular events over patients' lifetimes (Gerstein et al., 2000). 3 According to Medcorp guidelines, all patients with cardiovascular disease should be regularly taking ACE inhibitors and some form of a statin. ACE inhibitors lower blood pressure, and statins lower cholesterol. Nationally, only 50% of all cardiovascular disease patients that should be treated with statins and ACE inhibitors are currently taking such medication (Dubois et al., 2002). These drugs significantly lower the immediate risk of a cardiovascular event (e.g., stroke, heart attack) for all individuals, regardless of sex or previous history of cardiovascular disease (LaRosa, He & Vupputuri, 1999; Yusuf, Sleight, Pogue, Bosch, Davies& Dagenais, 2000). To promote a higher prescription rate, Medcorp administrators send emails to physicians reminding them to prescribe such treatment. While prescription of these medications benefits patients by helping patients avoid death and reduce healthcare expenses, physicians often forget to prescribe these medications (Isles, 2002). This quality variable is the composite of the percent of cardiovascular disease patients 18 years and older who were dispensed the equivalent of a 90-day supply for ACE inhibitors and statins at any
3
We call this variable "quality" because statin and ACE inhibitor prescription rate accuracy are measures of physician quality according to the most influential quality assurance organizations (e.g. HEDIS, NCQA, IHI).
-19-
time within the quarterly reporting period. The component variables approached normality, were standardized, and were added together. The resulting variable was each physician's overall prescription rate of statins and ACE inhibitors for cardiovascular disease patients. The average prescription rate at Medcorp is 50%, similar to the national average. Control Variables We controlled for several variables that were not of direct interest for testing our hypotheses but that could be theoretically related to the dependent variable and might provide plausible alternative explanations for our findings. Average practice busyness. Patients who have to wait long periods of time to see their physician may be less satisfied, so we controlled for the busyness of each physician's practice. At the close of business each day, the Medcorp computer counts how many days into the future each physician's third available appointment is. According to the National Quality Measures Clearinghouse, counting the days until the third next available appointment is the healthcare industry's standard measure of access to care and indicates how long a patient waits to be seen. Doctors who are not very busy typically have three available appointments the next day, whereas busy doctors often do not have three available appointments for several days. The final variable was the quarterly average number of days until each physician's third open appointment slot. Physician full-time status. We included the number of hours a physician worked in our model because patients may be more satisfied if their physician works more hours. Physicians ranged from working 30 to 100 percent of a full-time position. Number of patients in panel. Medcorp assigns physicians to care for a particular group (i.e. panel) of patients. Patients in larger panels may be less satisfied and so we controlled for the total number of patients in each physician's panel standardized by the full-time status of the physician.
-20-
Average patient age. Older patients may have different expectations about doctor demographics, so we included the average patient age for each physician's panel in our model. Average chronic sickness of panel. Sicker patients may be less satisfied, so we controlled for the panel chronic sickness variable calculated by Medcorp (e.g., it captures the percentage of patients with diabetes and cardiovascular disease). Physician age and tenure. Physicians who are older or who have been employed by Medcorp for more years may have more loyal, satisfied patients. Physician tenure by objective performance. Because women and nonwhite physicians tend to be more recently hired than male and white physicians, any influence of physician race and gender on customer satisfaction may be masked by physician tenure. We therefore included the interactions of tenure by objective performance in our models so that we could more clearly determine the interactive influence of physician race by objective performance and physician gender by objective performance on customer satisfaction. Results Table 1 reports the means, standard deviations, and correlation coefficients between the dependent, independent, and control variables. We found no significant differences in our objective measures of performance based on employee race and gender. Our first hypothesis stated that the relationship between employee objective performance and customer satisfaction judgments would be less positive for employees belonging to low-status demographic groups compared to employees belonging to high-status demographic groups. To test this, we examined the interactions of the objective measures of employee performance (i.e. quality, productivity and accessibility) by employee race and gender. We used hierarchical moderated regression models to do so (Aiken & West, 1991). We centered all variables involved in the interaction terms to minimize multicollinearity between the
-21-
interaction terms and their individual components (Aiken & West, 1991). We entered all of the control variables in Model 1. In Model 2 we entered the control variables plus the interactions involving physician gender. In Model 3 we entered all the control variables as well as the interactions involving physician race. Finally, in Model 4, we entered all control variables and all interaction effects. Table 2 presents the results of this analysis. ----------------------------------Insert Tables 1 and 2 about here ----------------------------------The two-way gender X objective performance interactions as a set explained a significant amount of incremental variance in the dependent variable (R2 = .07, p< .01) providing preliminary support for hypothesis 1. Inspection of the individual regression weights showed that the physician accessibility X gender and physician quality X gender interactions were significant (p< .05). We probed the pattern of the interaction by examining the simple slope of the objective performance measures for male and female physicians (Aiken & West, 1991). The results of this analysis are shown graphically in Figure 2. ---------------------------------Insert Figure 2 about here ---------------------------------The figure shows a stronger positive relationship between physician customer-centered behaviors and performance ratings for men than for women. We calculated the significance of the simple slopes for interactions (Aiken & West, 1991). The coefficient of the simple slope of quality behaviors on customer satisfaction was significantly more positive for male physicians (b = .32, p< .01) than female physicians (b = -.01, n.s.). Similarly, the coefficient of the simple slope for accessibility behaviors was significantly more positive for male physicians (b = .13, n.s.) than female physicians (b = -.17, n.s.). Although neither simple slope for accessibility behaviors is significantly different from zero, the simple slopes are significantly different from each other (p< .05). By looking
-22-
at the plots, one can see that the interaction is a cross-over, which shows that the direction of the relationship is the opposite for members of high- versus low-status demographic groups. The two-way race X objective performance interactions as a set explained a significant amount of incremental variance in the dependent variable (R2 = .08, p< .05) providing preliminary support for hypothesis 1. Inspection of the individual regression weights showed that the physician productivity X race and physician quality X race interactions were significant (p< .05). The forms of the interactions are shown graphically in Figure 2. Simple slope analysis reveals that the coefficient of the simple slope of quality on customer satisfaction is significantly more positive for white physicians (b = .29, p < .01) than for nonwhite physicians (b = -.14, n.s.). Likewise, the simple slope of productivity behaviors on customer satisfaction is significantly more positive for white physicians (b = .15, n.s.) than nonwhite physicians (b = -.32, p< .01). Overall, we find support for four of the relationships predicted in hypothesis 1. Discussion Our first study explores whether customers, who in this case were patients of an HMO, express their race- and gender-based biases in customer satisfaction judgments. We found that objectivelymeasured behaviors were only positively related to customer satisfaction for physicians who were white or male. We also found that one type of customer-centered behavior was significantly negatively related to customer satisfaction for women and nonwhite physicians. This second finding was an even stronger result than we anticipated because logically we might expect the relationship between customer-benefiting behaviors and customer satisfaction to be weaker, but still positive, for women and nonwhites compared to men and whites. The observed pattern of relationships indicates that biases against nonwhite and female employees may creep into satisfaction judgments. However, we must also consider this study's
-23-
shortcomings. First, the opposite relationship signs of panel age and age of physician with the DV between the correlation table and regression results suggest that these variables may have somehow influenced our results by suppressing variance in the DV that was irrelevant to prediction of the DV (Tabachnick & Fidell, 2003). However, the correlations are not statistically significant and our results are substantively unchanged regardless of whether these variables are included in the model, so we believe our results are not due to a suppression effect caused by those variables. A more serious limitation is that our Medcorp study only included a small percentage of nonwhites. Moreover, many of the nonwhites were Asians rather than African-Americans. Biases against African-Americans are more negative than those associated with Asians (Song, 2004) and so a study that included AfricanAmericans might be better able to detect the influence of such biases on customer satisfaction judgments. We were also not able to control for employee accents or differences in employee language and communication styles, or whether customers felt certain employees had nonwhite sounding names. It is possible that the biases we observed were due to perhaps some contextual variable such as employee language skill and were not due to pre-existing customer prejudices. Finally, we did not measure whether customer raters had preexisting bias against women and minorities. That is, we had no assessment of the raters' stereotypes or racial/gender biases as potential causes of their ratings. We designed our second study to address the limitations of our Medcorp study. The occupation we chose for our laboratory study was service employees working in a university bookstore and our raters were college students. We also used an experimental design to control for extraneous variables that might have influenced the results of our Medcorp study. BOOKCORP STUDY In our second study (Bookcorp), student raters were asked to observe a video of an employeecustomer interaction in a university bookstore, to evaluate the employee's behavior, and to provide
-24-
satisfaction judgments of the store environment. Our Bookcorp study is different from our Medcorp study in a variety of ways. First, we controlled for the job-related behavior (with a pre-scripted interaction) of the employee and varied only whether the behavior was performed by a male versus a female or a white versus an African-American employee. This aspect of the Bookcorp study's design allowed us to reduce variability in employee behavior thereby providing a better test of whether the same behavior would nevertheless produce different customer satisfaction judgments depending on the employee's gender or race. Second, we assessed how student participants, who were asked to assume the role of customers, not only evaluated the employee (as in our Medcorp study) but also evaluated the organizational context (i.e., the bookstore) in which the employee-customer interaction took place. Third, we assessed each participant's implicit bias towards women and nonwhites to see if these nonconscious attitudes might partly explain gender or racial bias in the ratings. Sample Eighty-six university students from a major northwestern U.S. public university watched two videos of a university bookstore employee interacting with a customer and were asked to evaluate the employee and the bookstore. The bookstore in the video clips was from a large East Coast U.S. university and it is highly unlikely that any of the participants had visited the bookstore previously. The "employees" and "customers" were hired professional actors and the scripted interaction was filmed before the bookstore opened in the morning (although our raters taking the customer perspective were not aware of this). We assigned 33 participants to view the white male, 21 to view the white female employee, and 34 to view the black male employee. Overall, a substantial percentage of our participants were nonwhite (43 percent) or female (38 percent). By having a heterogeneous sample of raters, the subjects are representative of the population of people using the book store. Design
-25-
Our design was a mixed-factorial design with one between subjects factor (Employee Demographic Characteristics) and one within subjects factor (Employee-Customer Interaction). We treated Employee Demographic Characteristics as a between-subjects factor to reduce participant awareness that they were participating in a race or gender related study. We presented all participants with two videos depicting different employee-customer interactions. One video involved the employee ringing up a book and telling the customer that the book's price in the computer was higher than its price on the shelf. The other video involved the same employee trying to help a customer find a book the customer wanted. Each video was about one minute in length. Each participant saw both videos of the same employee. We randomly assigned the ordering of the videos within each condition and found no evidence that the ordering of the videos influenced customer ratings. The customer and employee interaction was from a written script to ensure that their behavior was equivalent across conditions. The store background was also held constant across conditions since the camera was in the same location when filming the different employees for each interaction. Dependent Variables Customer satisfaction with the employee. Our measure of customer satisfaction with the employee asked raters to identify on a 7-point Likert scale (1 = very poor; 7 = excellent) how satisfied they were with (1) speed of service; (2) quality of service; (3) availability of staff for assistance, and; (4) employee responsiveness to customers' issues and concerns. This measure was adapted from an existing customer satisfaction survey we obtained from a large organization (see Appendix 1 for customer satisfaction items used across the three studies). Coefficient alpha for this measure was .74. Customer satisfaction with the context. Our measure of customer satisfaction with the context asked raters to identify on a 7-point Likert scale (1 = very poor, less than expected, definitely would not, or strongly disagree; 7 = excellent, better than expected, definitely would, or strongly agree) how
-26-
satisfied they were with (1) the bookstore's appearance; (2) the degree to which the bookstore was conducive to learning; (3) whether the bookstore had up to date equipment; (4) the degree to which the bookstore's facilities were visually appealing; (5) whether the bookstore's appearance was in keeping with the type of services provided; (6) the bookstore relative to their expectations, and; (7) their likelihood of recommending the bookstore to others. This measure was also adapted from an existing customer satisfaction survey we obtained from a large organization. Coefficient alpha for this measure was .76. Predictor Variables Condition. We had two conditions—one for race and one for sex. The sex condition included participants who viewed the white male employee or white female employee (1 = participants viewed two videos of a white woman employee, 0 = participants viewed two videos of a white man employee). The race condition included participants who viewed either the white male employee or the nonwhite male employee (1 = participants viewed two videos of a nonwhite man employee; 0 = participants viewed two videos of a white man employee). Participants completed survey questions only after watching both videos. Implicit bias. To measure raters' racial and gender prejudices, we administered two Implicit Association Tests (IATs). The IATs were constructed to capture each participant's level of nonconscious bias against nonwhites and women (Greenwald, Nosek and Banaji, 2003). We should note that IAT measurement shortcomings (Blanton & Jaccard, 2006), such as no absolute zero point or equal intervals, makes the interpretation of a respondent's score somewhat unclear. However, we chose to use the IAT as opposed to other types of bias measures (e.g., modern racism scale) because it is more difficult for participants to hide prejudices on the IAT than on explicit measures (Nosek, 2005). The gender IAT was administered after the participants saw the videos and made their customer
-27-
satisfaction judgments, but the race IAT was administered between the videos and the ratings. Prior research indicates no evidence of order effects for the IAT and dependent variables—probably because subjects still respond in socially desirable ways on the explicit measures (Greenwald, Poehlman, Uhlmann & Banaji, 2009). Importantly, implicit attitudes appear to be better predictors of behavior than their explicit counterparts, especially when social sensitivity concerns are high (Greenwald et al., 2009). For instance, implicit (but not explicit) attitudes about African Americans have been shown to predict desire to work with an African American partner on an intellectual task (Ashburn-Nardo, Knowles, & Monteith, 2003), and nonverbal actions (eye contact and other "friendly" behaviors) toward African American interaction partners (McConnell & Leibold, 2001). Though the correlation between implicit and explicit attitudes varies across domains (Nosek, 2005), the predictive validity of each suggests that they represent independent processes that explain unique variance in behavioral outcomes (see Greenwald et al, 2009, for a meta-analysis of the predictive validity of the IAT). Control Variables We controlled for rater race, gender and age to account for rater demographics which might plausibly influence reactions to employee demographics. Results We regressed the customer judgments of the employee and the organizational context on our controls, predictors, and interaction to determine the degree to which customer judgments of the employee and the organizational context reflected race and gender bias. Tables 3 and 4 present the regression models we used to test hypotheses 1, 2 and 3. -------------------------------------Insert Tables 3 and 4 about here --------------------------------------
-28-
Hypothesis 1 states that the relationship between employee objective performance and customer satisfaction judgments will be less positive for employees belonging to low-status demographic groups compared to employees belonging to high-status demographic groups. Because objective performance was held constant due to the employee script, the main effects of employee race and gender on customer satisfaction with the employee were used to test this first hypothesis. Model 2 in Table 3 shows that raters taking the customer perspective were significantly less satisfied with women employees than their equally well-performing white male counterparts (?R2 = .06; b = -.28; p< .05). However, we did not find evidence of bias in customer satisfaction judgments of the nonwhite employee (b = -.02; n.s.). Overall, we found some support for hypothesis 1. Hypothesis 2 states that people would report lower customer satisfaction judgments of the store environment when an employee in that environment belongs to a low (i.e., female, African-American) rather than high (i.e., male, white) status demographic group. Model 2 of Table 4 shows there is a significant main effect of race and gender on judgments of the store environment. Indeed, Model 2 of Table 4 shows a main effect of the female condition (?R2 = .17; b = -.45; p< .01) and the nonwhite condition (?R2 = .15; b = -.44; p< .001), suggesting that raters' biases influence judgments of the organizational context. We found strong support for hypothesis 2. Hypothesis 3a suggests that people would report even lower customer satisfaction judgments of the employee when observing an employee belonging to a low-status demographic group when the customer has negative implicit attitudes toward that group. Model 3 of Table 3 shows that the coefficient for the interaction term IAT Score X Nonwhite Condition is significant and in the expected direction for customer satisfaction with the employee (?R2 = .08; b = -.28; p< .01). To gain more insight into the nature of this effect we plotted the interaction and analyzed the simple slopes (see Figure 3). Individuals with high levels of implicit bias (+1 s.d.) were significantly more likely to report
-29-
lower satisfaction with the nonwhite male's performance than with the white male's (p< .01). However, the coefficient for the interaction term IAT Score X Gender Condition was not significant for customer satisfaction with the employee. As for hypothesis 3b, Model 3 of Table 4 shows that the coefficient for the interaction term IAT Score X Race is significant and in the expected direction for customer satisfaction with the context (?R2 = .04; b = -.18; p< .05). We plotted the interactions and conducted a simple slope analysis (see Figure 3). Customer IAT score (+1 s.d.) was positively related to customer satisfaction with the context when customers were observing a white male employee (b = .33; p< .01) but was negatively related to customer satisfaction with the context when customers were observing a nonwhite male employee (b = -.21; p< .05). The coefficient for the interaction term IAT Score X Gender Condition was significant and in the expected direction for customer satisfaction with the context (?R2 = .04; b = -.23; p< .05). Customer IAT score (+1 s.d.) was positively related to customer satisfaction with the context when customers were observing a white male employee (b = .23; p< .05) but was negatively related to customer satisfaction with the context when customers were observing a white female employee (b = -.21; p< .05). Overall, we found three significant coefficients supporting hypothesis 3. These results and plots suggest that judgments of employees and judgments of the organizational context are vulnerable to non-conscious biases. Discussion We found that raters taking the customer perspective rated the employee and the organizational context as being worse when observing the performance of a low-status employee and this was especially true if the person held implicit biases about that low-status group. The results regarding hypothesis 1 were not as strong as we anticipated suggesting that it may be possible to minimize biases by changing the setting in which the rating takes place. Specifically, the laboratory context was less
-30-
anonymous than a typical customer satisfaction questionnaire setting, which may have weakened the influence of bias on customer satisfaction judgments of employees. Although we told the participants that their responses were anonymous, they may have felt scrutinized because they provided their judgments when an experimenter was present, and wrote their names on a separate sign-in sheet to receive participation credit for a class. Additionally, because the IAT for race was administered prior to the ratings it may have alerted participants that they were in a race and gender study, which also may have weakened the Bookcorp study results. We observed, though, that the effects of bias on judgments of the organizational context were still quite strong, which makes sense to us because participants may have been unaware of and therefore unable to suppress their biases that spilled over onto their judgments of the organizational context. As summarized by hypothesis 4, we expected that the relationship between an organizational unit's objective performance and customer satisfaction would be less positive for organizational units that employ higher percentages of employees belonging to low-status demographic groups (i.e., women and racial minorities) compared to units that employ higher percentages of employees belonging to high-status demographic groups (i.e., men and whites). By returning to the field to test this hypothesis, we complete the full-cycle of research model and assess the generalizability of our theory to a different organization. GOLFCORP STUDY Our sample was drawn from a large country club organization, hereafter referred to as Golfcorp. Golfcorp has 66 country clubs across the United States and roughly 70,000 customermembers and it employs approximately 8,000 people. Our sample consisted of all 66 Golfcorp country clubs. Within our sample, 31.4% of employees were women, 18.1% were Latino, 6.7% were AfricanAmerican, and 1.7% were Asian-American or Native-American.
-31-
Measures Golfcorp routinely collects customer satisfaction ratings as well as objective indicators of facility performance that were assumed to have a direct, positive impact on customers' service experiences. The dependent variable in our study was customer satisfaction with the facility. The independent variables were each club's employee demographics (percentage nonwhite and female employees) and two types of objective club performance. Customer satisfaction with facility. Like many organizations, Golfcorp measures customer satisfaction with a quarterly survey, which is mailed to a percentage of each facility's customers. An average of 63.8 customers rated each facility. The average response rate per facility was 27.3 percent (an average of 234 surveyed customers per facility). The marketing company hired to do the customer survey randomly sampled each facility's customers each quarter until they got either 20 respondents or three percent of the total customer base (whichever was larger). The items used for this measure reflect a focus on the facility context (quality of its clubhouse and golf course) and overall ratings of the facility, similar to what was used in the Bookcorp study. Customers rated each of the items on a 5point Likert scale (1 = very poor; 5 = very good) "How would you rate the following aspects of your club?" (1) Maintenance of grounds/Appearance of clubhouse; (2) Locker rooms and restrooms; (3) Quality of greens; (4) Condition of course; (5) Pace of play; (6) Condition of practice facilities; (7) Ability to obtain desired tee times; (8) Club meets expectations (1 = less than expected, 5 = better than expected), and; (9) Likelihood of recommending club to others (1 = definitely will not, 5 = definitely will). Coefficient alpha for this performance measure is .81. In our analyses, we lagged this measure six months after the independent variables to more conclusively show that employee demographics and objectively-measured performance influence customer ratings, rather than the other way around.
-32-
Percent Nonwhite. Golfcorp identified the percentage of white and nonwhite employees in each facility. Across the 66 clubs in the sample, 26.5 percent of employees were nonwhite. According to the U.S. Census Bureau's 2000 census, the percentage of ethnic minority employees in our sample is consistent with the national average of 28 percent of the U.S. population. Percent Female. We also obtained this variable from Golfcorp records. Thirty-one percent of employees in our sample were women, which Golfcorp leaders believe to be consistent with the country club industry average. However, the percentage of women in our sample is lower than the percentage of women across all industries, which is 46%, according to the Bureau of Labor Statistics at the U.S. Department of Labor. Objective facility characteristics. As in our first two studies, we wanted to clearly identify the portion of variance attributable to customer bias versus the portion attributable to better facility performance. We therefore used two attributes as our indicators of objective facility performance— facility productivity and facility attribute quality. Both dimensions reflect facility characteristics that benefit customers. Facilities with more productive employees create more value—both for Golfcorp and for customers. Indeed, Golfcorp executives told us that facilities with higher productivity values charge lower dues to members, are more profitable and are simply better-run facilities. Higher attribute quality benefits customers by allowing customers to enjoy newer and better facilities. Facility managers are shown how they compare to other facilities in terms of quality and productivity. Employee compensation is tied to the productivity measure, but not the quality measure. Employees in facilities that are above average in productivity are given a bonus, while employees at below average clubs are not given a bonus. Facility Productivity. Facility productivity was calculated by Golfcorp's central accounting office for the calendar year ending six months before the dependent variable was collected. This
-33-
variable is each club's annual profits divided by the average number of employees working for the club in that year. The number of employees at each club is centrally controlled such that clubs with more members are allotted proportionally more employees by the central office. Therefore facility productivity is determined by the employees' effectiveness at creating value. Facility quality attributes. Over time, the condition of the golf course and the clubhouse deteriorates and needs to be rebuilt or refurbished. Golfcorp assesses the quality of the course and clubhouse of each club to ensure that customers are receiving a high standard of service. Golf courses are assigned a quality rating based on the percentage of the course that is infiltrated by crab grass and dead spots (1 = more than 40% of course is crab grass or dead spots; 5 = less than 5% of course is crab grass or dead spots). Likewise, clubhouses are assigned a quality rating, indicating how recently they were built or refurbished (1=built or refurbished more than 15 years ago; 5 = built or refurbished 2 years ago or less). The overall facility attribute quality variable is the composite of the clubhouse quality rating and the golf course quality rating. The two component variables approached normality and were added together. The resulting variable was each facility's overall golf course and clubhouse quality. Control Variables Customers may be more satisfied with larger facilities because they offer more amenities, and with facilities that employ a large percentage of young or temporary employees who may be more energetic. Therefore we controlled for facility size, average employee age, and percent temporary employees. Customers may be more satisfied if they have been a member for a long time and have not quit, if they are men or if they are older. Therefore we also controlled for average customer tenure (months), percent male customers, and average customer age. Results
-34-
Table 5 reports the means, standard deviations, and correlation coefficients between the dependent, independent, and control variables. We used hierarchical moderated regression models to examine the hypothesized interaction effects (Aiken & West, 1991). We centered all variables involved in the interaction terms to minimize multicollinearity between the interaction terms and their individual components (Aiken & West, 1991). Table 6 presents the results of our hierarchical moderated regression analysis. We entered all of the control variables in Model 1. In Model 2 we entered the interactions involving sex and the two dimensions of objective performance, In Model 3 we entered the interactions involving race and the two dimensions of objective performance, and in Model 4 we entered all four interactions. ----------------------------Insert Tables 5 and 6 about here ---------------------------Hypothesis 4 stated that the association between an organizational unit's objective performance and customer satisfaction would be attenuated for organizational units that employ higher percentages of employees belonging to low-status demographic groups compared to units that employ higher percentages of employees belonging to high-status demographic groups. The two-way gender X objective performance interactions as a set explained a significant amount of incremental variance in the dependent variable (?R2 = .04, p< .05) providing some further support for this hypothesis. Inspection of the individual regression weights in the full model showed that the facility quality X gender and facility productivity X gender interactions were significant (p< .05). We probed the pattern of the interactions by examining the simple slopes of the objective performance measures for facilities with high and low percentages of female employees (Aiken & West, 1991). Like the Medcorp study plots, we found a stronger positive relationship between objective performance and customer satisfaction for facilities that have a low percentage of female employees
-35-
than for facilities that have a high percentage of female employees. Facility quality was significantly more positively associated with customer satisfaction for facilities with a low percentage of female employees than for facilities with a high percentage of female employees. Likewise, facility productivity was more positively associated with customer satisfaction for facilities with a low percentage of nonwhite employees than for facilities with a high percentage of nonwhite employees. Again, even though the simple slopes were not significantly different from zero, the significant regression coefficient in the full model demonstrates that they are significantly different from each other. By looking at the plots, one can see that the interaction is a cross-over, which shows that the direction of the relationship is the opposite, and statistically significant, for members of high- versus low-status demographic groups. These results support hypothesis 4 for both of the gender X objective performance relationships. The two-way race X objective performance interactions as a set explained a significant amount of incremental variance in the dependent variable (?R2 = .15, p< .05) providing preliminary support for our hypothesis. Inspection of the individual regression weights from the full model showed that the facility attribute quality X race and facility productivity X race interactions were significant (p< .05). Simple slope analysis revealed that the association between facility quality and customer satisfaction was significantly more positive for facilities employing a low percentage of nonwhites than for facilities employing a high percentage of nonwhites. Likewise, the association between facility productivity and customer satisfaction was more positive for facilities employing a low percentage of nonwhites (b= .32, p< .01) than for facilities employing a high percentage of nonwhites (b=-.23, p< .05). These results support hypothesis 4 for both of the race X objective performance relationships. Discussion
-36-
In this Golfcorp study, we found that objectively measured behaviors that benefit customers were positively related to customer satisfaction, but only for facilities with a low percentage of nonwhite and female employees. These results parallel the results of our Medcorp study. GENERAL DISCUSSION We set out to determine if and how customer satisfaction judgments are influenced by racial and gender biases. Across three samples we found converging evidence that customer satisfaction judgments are susceptible to systematic and predictable racial and gender biases. Customers tended to be less satisfied with the services provided by women and nonwhite employees than by men and white employees, even when controlling for objective indicators of performance (Medcorp study). We also found that these biases operated on judgments about the store context when a third-party evaluator observes an employee interacting with a customer (Bookcorp study), particularly if the observer holds negative implicit attitudes toward women or minorities. Finally, we found that evaluations of the organizational unit as a whole are negatively associated with the presence of nonwhite and female employees (Golfcorp study). It is worth noting that we found evidence for the operation of racial biases regardless of whether the nonwhite employees were predominantly Asian (Medcorp study), AfricanAmerican (Bookcorp study), or Latino (Golfcorp study). The consistency of our results across three different samples and methodologies testifies to the robustness and generality of the systematic biases we observed and to the internal validity of our theoretical model. The pattern of these biases may help explain the persistence of demographic inequalities in organizations. To cite just a few examples, women and nonwhites make 25 percent less than their male and white counterparts in equivalent jobs (U.S. Census Bureau, 2006), women and nonwhites are twice as likely as white men to be unemployed and underemployed (NIOSH, 2002), and women and ethnic minorities are not well represented among the ranks of highly paid managers and
-37-
professionals in U.S. corporations and in prestigious occupations like law and medicine (e.g., Baldi & McBrier, 1997; Eagly & Karau, 2002; Wilson, Sakura-Lemessy & West, 1999). Economists have often been perturbed by these demographic inequalities because orthodox economic theory would predict that some of these inequalities should be erased when employers compete for women and nonwhite applicants whose wages are 25 percent less costly than their white and male counterparts (The Economist, 2008). Our results suggest that the evidence available to organizational decision makers is that customers tend to be less satisfied with nonwhite and female employees; however, without the benefit of the research conducted here, decision makers would not be able to determine that these lower satisfaction ratings are attributable at least in part to customer biases. Therefore, decision makers might rationally choose, based on their limited information, to preferentially select white and male employees, as their data is likely to bear out that such personnel are better performers. Evidence from highly publicized lawsuits (e.g., Shoney's Restaurant; Abercrombie & Fitch) suggests that managers are keenly aware of the fact that some customers may prefer white and male employees. Executives in these cases admitted to deliberately favoring white employees in hiring and promotion decisions to enhance customer satisfaction and organizational profitability (Brief et al., 2000). When employees or units are viewed as performing less well by customers as a result of demographic characteristics like race or gender, one could argue that these employees and units should receive fewer rewards, bonuses, and promotional opportunities. But it is important to note that the customer judgments in our studies were inconsistent with other objective indicators of performance. In other words, nonwhites and women may have behaved the same way as their white and male counterparts in trying to provide satisfactory customer service, but if compensation and other organizational benefits are linked to customer satisfaction ratings then they may not be rewarded
-38-
similarly for identical behavior, which would violate the principle of equity which most business enterprises claim to follow. Our results suggest that if customer evaluations become widely and uncritically used to determine pay and promotion opportunities, the job outcomes of women and ethnic minorities could be adversely impacted. For example, consider what would happen if managers notice which employees routinely receive the highest customer satisfaction scores and use this information to make promotion decisions or if university administrators relied heavily on student ratings of teaching effectiveness to influence promotion and tenure decisions. At higher levels of the organization, executives may examine which of their organizational units achieve the highest levels of customer satisfaction and promote those managers further up the organizational hierarchy. Our data suggest that one possible consequence of these decisions is that whites and men will be much more likely than their nonwhite or women counterparts to receive favorable customer satisfaction judgments, which should accelerate their journey up the organizational ladder. Likewise, managers who purposely stock their organizational units with whites and men are likely to have more career success than managers who do not. Our finding that customer biases can spill over onto the surrounding organizational context contributes to the literature on contamination and signaling and also illustrates the subtle operation of racial and gender bias. Marketing researchers have shown how observable customer characteristics like physical attractiveness can influence other customers' desire to purchase a product (Argo et al., 2009). To our knowledge, ours is the first study to show how an observable characteristic of an employee like race or gender can influence customer perceptions of an organization's contextual quality. This finding may provide insight into phenomenon known as "white flight" where whites move out of a neighborhood once a critical mass of nonwhites move in (Gladwell, 2000; Kruse, 2005).
-39-
In an organizational setting, a similar phenomenon may operate in which customers may link conscious or unconscious negative attitudes they have towards members of status groups to employees who belong to these groups. In turn, these associations "contaminate" customer perceptions of the organizational context. This process of contextual spillover may partly explain why managers have often been reluctant to pursue diversity despite the known performance advantages of having a diverse workforce (Joshi, Liao & Jackson, 2006; O'Reilly, Williams, & Barsade, 1997). It may be that managers are aware that diversity has hidden costs because it increases the possibility of "customer flight" to an organization that has fewer employees who belong to low-status demographic groups. Future studies using customer satisfaction as an outcome variable should take into account the demographic make-up of the employees as well as objectively measured organizational characteristics. Our findings cast doubt on the ability of customers to accurately perceive the quality of customer service organizations. The theoretical attributes we suggested as possible causes of customer bias were meant to explain why these biases occur, but like any useful theory they also suggest potential remedies. Based on our theorizing, racial and gender biases in customer satisfaction judgments may be reduced by (1) making customer ratings less anonymous; (2) changing the standards customer use to make their ratings so they emphasize behavior rather than subjective judgments; and (3) introducing customer de-biasing education or training in the evaluation process. Identifiability (i.e. non-anonymity) may provide an especially strong de-biasing effect in service contexts characterized by repeated interaction (e.g. the doctor/patient relationship) because customers would want to make sure their ratings do not jeopardize future received service quality. Behaviorally anchored rating scales may be especially helpful for removing bias in customer expectations (i.e. individuals often hold nonwhites and women to higher standards; Biernat & Kobrynowicz, 1997). Organizations could consider only accepting customer satisfaction surveys from customers who have completed de-biasing training,
-40-
although such a move would be logistically and perhaps scientifically problematic (due to potential selection bias). In addition to addressing factors that cause bias in customer ratings, organizations can take practical steps to minimize the potential adverse impact of customer biases on nonwhite and female employees' careers. For example, organizations might consider only using satisfaction surveys from frequent customers to ensure that raters have sufficient exposure to targets. Organizations could also ask for customer feedback during the customer service encounter so that customers will be most likely to be paying attention and less likely to rely on information subject to memory bias when judging their customer experience. Organizations might also want to let customers know that the data will be used to make career progression decisions so that customers are more motivated to judge responsibly. Organizations could also insert "bias sensitive" questions in customer satisfaction judgments so that responses from potentially biased customers can be given less weight or discarded. However, we urge caution if organizations choose to remove outlier ratings because this tactic may actually decrease rating accuracy if most judges are aware of their biases and therefore tend to over-correct their judgments (Zitzewitz, 2006). Alternatively, organizations may be able to statistically correct for bias when calculating customer satisfaction judgments. Finally, using different survey formats for customer rating scales might also be helpful for circumventing rater biases, like forced-choice, behaviorally anchored rating scales (citing specific valued behaviors), and un-weighted and weighted checklists. Organizations should consider the tradeoffs between these formats and choose the one that is most likely to reduce the effects of customer judgment biases on the career prospects of those who are most vulnerable to being targets of such bias. Limitations
-41-
We believe our findings provide strong, consistent support for our theoretical predictions. However, like all research, ours has its share of limitations. First, role congruence may have been an issue in our Medcorp study as patients may expect their physicians to be white and male and thereby judge non-white or women doctors more harshly. However, there is not much evidence to suggest that they do expect their doctor to be a white male. Indeed, patients prefer their doctor to look like them (women prefer women doctors, and nonwhite patients prefer nonwhite doctors; Chen, Fryer, Phillips, Wilson & Pathman, 2005; Cooper-Patrick, Gallo, & Gonzales, 1999). Likewise, role congruence should not have been an issue in our Bookcorp study, so our consistent results across these two samples provide us with some confidence that role-congruence alone is not responsible for our findings. Still, future research would be well-served to test our hypotheses in a variety of samples. Another potential issue is rater-target congruence. Although including patient gender and race in the analysis slightly strengthened our Medcorp study results, we did not control for these variables in that study because of multi-collinearity (i.e., the respective correlations between customer gender/race and physician gender/race were greater than .90). In our other two studies we found no evidence that rater-target congruence influenced customer satisfaction judgments. Indeed, in post-hoc analyses, the interactions of customer race X employee race and customer gender X employee gender were not significant. To maintain compliance with the rule of thumb that there should be five cases per variable (Tabachnick & Fidell, 2003), we do not report these post-hoc analyses. Another potential limitation is that unobserved variables may be responsible for our results. In our Golfcorp study, employee demographics may have masked the facility's strategy. Facility executives pursuing a low-cost strategy may have hired a large number of women and nonwhite employees, whereas those pursuing a premium-pricing strategy may have hired a large number of whites and men. We ran some post-hoc analyses to test this idea and found no supporting evidence.
-42-
Specifically, we ran the interactions of several variables measuring club strategy (i.e. services offered, turnover rate) by objective performance and our four race and gender interactions remained significant, but none of the additional interactions were significant (we do not report these analyses to maintain an adequate case to variable ratio). Relatedly, we also did not test or report whether nonwhite women face a double jeopardy (Berdahl & Moore, 2006) in terms of customer satisfaction to maintain compliance with the five to one rule of thumb (i.e. we were not able to test the three-way interactions involving employee objective performance, employee race and employee gender), but future research should do so. Finally, we did not test whether the IAT influences ratings of the organizational unit because we found that the IAT influences sub-components of the organizational unit ratings— employee and context ratings. We should also mention that our method of testing for possible bias in performance evaluations was a significant improvement over past studies. First, we used an objective performance standard so that we could compare subjective judgments to this standard and therefore determine whether the customer judgments of performance might be influenced by race and gender. Second, our subjective judgments were based on aggregated judgments from a large number of customers rather than relying on the judgments of a single supervisor. This is important because the large number of raters provides a highly reliable subjective performance rating for each individual, context, or organization. The IAT is new and its predictive validity is relatively untested (Blanton & Jaccard, 2006), so another contribution our paper offers is the first demonstration of the IAT's predictive validity in a management journal. Finally, we controlled for several variables that could provide alternative explanations for our results, such as the average employee age and the average customer tenure with the organization.
-43-
Given these methodological strengths of our research, it is unsettling to find that customers may not respond favorably to organizational characteristics designed to benefit them when these organizations have a high percentage of low-status employees. At Golfcorp, employees at clubs with a high percentage of female and nonwhite employees can in fact be economically harmed by customer satisfaction evaluations because clubs who fail to achieve the target level of customer satisfaction (i.e., below the organizational average) do not receive a salary bonus. The practical implications of our results become more apparent when we when examine the effect sizes in our samples. Across our three studies, the racial and gender bias effects on customer satisfaction judgments explained between 15 and 24 percent of the variance in customer satisfaction. Cohen (1988) provides ballpark descriptors of effect sizes based on R-squared values—"large" (R2 = .25), "medium" (R2 = .09), and "small" (R2 = .01). Therefore, the average observed effect size of racial and gender bias across our three samples is between medium and large. Conclusions In these different samples we demonstrated that customer ratings are biased against women and racial minorities. We had two field studies and one laboratory study utilizing a full cycle research strategy. The effects were demonstrated for three different minority groups and three different contexts involving employee-customer contact. In all three settings we controlled for actual objective behavior or performance along with a series of other controls appropriate for that context. The effects are demonstrated for individual targets as well as the context or organization in which the targets work. In the laboratory sample we showed that the biased ratings are exacerbated by implicit racial or gender bias. In short, these are fairly robust findings across jobs, contexts, raters and ratees. If these results are replicated and generalizable they have significant implications for organizational practice. If managers are serious about the fair treatment of their employees and the
-44-
promotion of diversity they will need to treat customer ratings differently. More specifically, the rating process can be changed by increasing information, responsibility or training for raters and by changing how customer ratings are used. In the latter context, organizations can perhaps measure and discount such biases or use statistical procedures to adjust the ratings to remove the bias. Without such actions, given the increasing dependence on customer ratings, we are likely to not only maintain existing levels of inequitable compensation and advancement for women and minorities, we are likely to increase these inequities. This outcome is unacceptable in a society that is committed legally, morally and socially to fair treatment for all in the workplace.
-45-
References Aiken, L. S. & West, S. G. 1991. Multiple regression: Testing and interpreting interactions. Newbury Park, CA: Sage Publications. Anderson, E. W., Fornell, C., & Mazvancheryl, S. K. 2004. Customer satisfaction and shareholder value. Journal of Marketing, 68: 172-185. Anderson, E. W.,& Mittal, V. 2000. Strengthening the satisfaction-profit chain. Journal of Service Research, 3: 107-120. Anderson, N.H. 1981. Foundations of information integration theory. New York. Academic Press. Argo, J. J., Dahl, D. W., & Morales, A. C. 2009, Positive Consumer Contamination: Responses to Attractive Others in a Retail Context. Forthcoming at the Journal of Marketing Research. Ashburn-Nardo, L., Knowles, M. L., & Monteith, M. J. 2003. Black Americans' implicit racial associations and their implications for intergroup judgment. Social Cognition, 21: 61-87. Aquino, K., & Bommer, W. H. 2003. Preferential mistreatment: How victim status moderates the relationship between organizational citizenship behavior and workplace victimization. Organization Science, 4: 274-285. Baldi, S., & McBrier, D. B. 1997. Do the Determinants of Promotion Differ for Blacks and Whites? Evidence from the U.S. Labor Market. Work and Occupations, 24: 478-497. Baltes, B. B., & Parker, C. P. 2000. Reducing the effects of performance expectations on behavioral ratings. Organizational Behavior and Human Decision Processes, 82: 237-267. Berdahl, J. L. & Moore, C. (2006). Workplace harassment: Double-jeopardy for minority women. Journal of Applied Psychology, 91, 426-436. Berger, J., Conner, T. L., & Fisek, M. H. 1974. Expectation States Theory: A Theoretical Research Program. Cambridge, MA: Winthrop. Berger, J., Fisek, M. H., Norman, R. Z., & Zelditch, M., Jr. 1977. Status Characteristics and Social Interaction: An Expectation States Approach. New York: Elsevier. Bertrand, M., & Mullainathan, S. 2004. Are Emily and Greg More Employable than Lakisha and Jamal? A Field Experiment on Labor Market Discrimination. American Economic Review, 94: 9911013. Biernat, M., & Kobrynowicz, D. 1997. Gender and race-based standards of competence: Lower minimum standards but higher ability standards for devalued groups. Journal of Personality and Social Psychology, 72: 544-557.
-46-
Blanton, H. & Jaccard, J. (2006). Arbitrary Metrics in Psychology. American Psychologist, 61: 27-41. Bobo, L. 1998. Race, interests, and beliefs about affirmative action - Unanswered questions and new directions. American Behavioral Scientist, 41: 985-1003. Bracken, D. W., Church, A. H., & Timmreck, C. W. 2001. Handbook of Multisource Feedback: The Comprehensive Resource for Designing and Implementing MSF Processes. New York, NY: JosseyBass. Brief, A. P., Dietz, J., Cohen, R. R., Pugh, S. D., & Vaslow, J. B. 2000. Just doing business: Modern racism and obedience to authority as explanations for employment discrimination. Organizational Behavior and Human Decision Processes, 81, 72-97. Chatman, J. A., & Flynn, F. J. 2005. Full-Cycle Micro-Organizational Behavior Research. Organization Science, 16: 434-447. Chen, F.M., Fryer, G.E., Jr., Phillips, R. L., Jr., Wilson, E., & Pathman, D.E., (2005). Patients' Beliefs About Racism, Preferences for Physician Race, and Satisfaction With Care. Annals of Family Medicine. 3: 138-143. Cialdini, R. B. 1995. A full-cycle approach to social psychology. In G. G. Brannigan & M. R. Merrens (Eds.), The Social Psychologists: Research Adventures: 53-72. New York: McGraw-Hill. Cohen, J. 1988. Statistical power analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum. Cooper-Patrick, L., Gallo, J.J., & Gonzales, J.J.,(1999). Race, gender, and partnership in the patientphysician relationship. Journal of the American Medical Association. 282: 583-589. Crandall, C. S., & Eshelman, A. 2003. A Justification-Suppression model of the expression and experience of prejudice. Psychological Bulletin, 129: 414-446. Davis, W. E., & Davis, D. R. 1999. The University Presidency: Do Evaluations Make a Difference? Journal of Personnel Evaluation in Education, 13: 119-140. Deci, E. L. & Ryan, R. M. 1985. The general causality orientations scale - Self-determination in personality. Journal of Research in Personality, 19:109-134. DeNisi, A. S., Robbins, T., & Cafferty, T.P. 1989. Organization of information used for performance appraisals: Role of diary keeping. Journal of Applied Psychology, 74: 124-129. Devine, P. G., Plant, E. A., Amodio, D. M., Harmon-Jones, E., & Vance, S. L. 2002. The regulation of explicit and implicit race bias: The role of motivations to respond without prejudice. Journal of Personality and Social Psychology, 82: 835-848. Dovidio, J. F., & Gaertner, S. L. 1981. The effects of race, status, and ability on helping-behavior. Social Psychology Quarterly, 44: 192-203.
-47-
Dovidio, J. F., & Gaertner, S. L. 2000. Aversive racism and selection decisions: 1989 and 1999. Psychological Science, 11: 315-319. Dubois, R. W., Alexander, C. M., Wade, S., Mosso, A., Markson, L., Lu, J. D., Nag, S., & Berger, M. L. 2002. Growth in use of lipid lowering therapies: Are we targeting the right patients? American Journal of Managed Care, 8: 81-86. Eagly, A. H., Makhijani, M. G., & Klonsky, B. G. 1992. Gender and the evaluation of leaders: A metaanalysis. Psychological Bulletin, 111: 3-22. Eagly, A. H. & Karau, S. J. 2002. Role congruity theory of prejudice toward female leaders. Psychological Review, 109: 573-598. Feinglass, J., & Salmon, J. W. 1990. Corporatization of medicine: the use of medical management information systems to increase the clinical productivity of physicians. Journal of Health Services, 20: 233-252. Fernandez, J. P. 1981. Racism and sexism in corporate life. Lexington, MA: Heath. Figueroa, E. B., & Woods, R. A. 2007. Industry output and employment projections to 2016. (2007). Monthly Labor Review, 130: 53-85. Folkes, V.S. & Patrick, V.M. 2003. The positivity effect in perceptions of services: Seen one, seen them all? Journal of Consumer Research, 30: 125-137. Frazer, J. G. 1890/1959. The golden bough: A study in magic and religion. New York: Macmitian. (Reprint of 1922 abridged edition, edited by T. H., Gaster; original work published 1890). Gaertner, S. L., & Dovidio, J. F. 1986. The aversive form of racism. In J. F., Dovidio & S. L., Gaertner (Eds.), Prejudice, Discrimination, and Racism. Orlando, FL: Academic Press. Gerstein, H. C., Yusuf, S., Mann, J. F. E., Hoogwerf, B., Zinman, B., Held, C., Fisher, M., Wolffenbuttel, B., Bosch, J., Richardson, L., Pogue, J., & Halle, J. P. 2000. Effects of ramipril on cardiovascular and microvascular outcomes in people with diabetes mellitus: Results of the HOPE study and MICRO-HOPE substudy. Lancet, 355: 205-212. Giannopoulos, G., Conway, M., & Mendelson, M. 2005. The gender of status: The laypersons' perception of status is gender-typed. Sex Roles, 53: 795-806. Gilovich, T., Griffin, D. W., & Kahneman, D. 2002. Heuristics of judgment: The psychology of intuitive judgment. Cambridge, U. K.: Cambridge University Press. Gladwell, M. 2000. The Tipping Point: How Little Things Can Make a Big Difference. Little Brown Publishing.
-48-
Goldin, C., & Rouse, C. 2000. Orchestrating Impartiality: The Impact of 'Blind' Auditions on Female Musicians. American Economic Review, 90: 715-741. Greenhaus, J. H., Parasuraman, S., & Wormley, W. M. 1990. Effects of race on organizational experiences, job performance evaluations, and career outcomes. Academy of Management Journal, 33: 64-86. Greenwald, A. G., & Banaji, M. R. 1995. Implicit social cognition: Attitudes, self-esteem, and stereotypes. Psychological Review, 102: 4-27. Greenwald, A. G., Nosek, B. A., & Banaji, M. R. 2003. Understanding and using the implicit association test: I. An improved scoring algorithm. Journal of Personality and Social Psychology, 85: 197-216. Greenwald, A. G., Poehlman, T. A., Uhlmann, E., & Banaji, M. R. (2009). Understanding and using the Implicit Association Test: III. Meta-analysis of predictive validity. Journal of Personality and Social Psychology. In press. Gruca, T. S., & Rego, L. L. 2005. Customer satisfaction, cash flow, and shareholder value. Journal of Marketing, 69: 115-130. Gupta, S., & Zeithaml, V., 2006. Customer metrics and their impact on financial performance. Marketing Science, 25: 718-739. Haas, J. S., Cook, E. F., Puopolo, A. L., Burstin, H. R., Cleary, P. D., Brennan, T. A. 2000. Is the professional satisfaction of general internists associated with patient satisfaction? Journal of General Internal Medicine,15: 122-128. Hagan, C. M., Konopaske, R., Bernardin, H. J., & Tyler, C. L. 2006. Predicting assessment center performance with 360-degree, top-down, and customer-based competency assessments Human Resource Management, 45: 357-390. Hsaio, W. C., Braun, P., Becker, E. R., & Thomas, S. R. 1988a. The resource-based relative value system. Journal of the American Medical Association, 258: 799-802. Hsaio, W. C., Braun, R., Dunn, D., & Becker, E. R. 1988b. Resource-based relative values: An overview. Journal of the American Medical Association, 260: 2347-2353. Isles, C. G. 2002. Patients with acute coronary syndrome should start a statin while still in hospital. Heart, 88: 5-6. Ittner, C. D., & Larcker, D. F. 1998. Are non-financial measures leading indicators of financial performance? An analysis of customer satisfaction. Journal of Accounting Research, 36: 1-35. Joshi, A., Liao, H., & Jackson, S. E. 2006. Cross-level effects of workplace diversity on sales performance and pay. Academy of Management Journal, 49: 459-481.
-49-
Judge, T. A., & Ferris, G. R. 1993. Social-context of performance evaluation decisions. Academy of Management Journal, 36:80-105. Kane, J. S., Bernardin, H. J., Villanova, P. & Peyrefitte, J. 1995. Stability of Rater Leniency - 3 Studies. Academy of Management Journal, 38: 1036-1051. Keller, K.L. 2003. Strategic brand management. Upper Saddle River, N.J. Prentice Hall. Kleiner, K. D., Akers, R., Burke, B. L., & Werner, E. J. 2002. Parent and physician attitudes regarding electronic communication in pediatric practices. Pediatrics, 109: 740-744. Kruse, K. M. 2005. White Flight: Atlanta and the Making of Modern Conservatism. Princeton, NJ: Princeton University Press. Laine, C., & Davidoff, F. 1996. Patient-centered medicine: A professional evolution. Journal of the American Medical Association, 275: 152-156. Landy, F. J., Shankster, L. J., & Kohler, S. S. 1994. Personnel-selection and placement. Annual Review of Psychology, 45: 261-296. LaRosa, J. C., He, J., & Vupputuri, S. 1999. Effect of statins on risk of coronary disease: a metaanalysis of randomized controlled trials. Journal of the American Medical Association, 282: 23402346. Latham, G. P., & Wexley, K. N. 1977. Behavioral observation scales for performance-appraisal purposes. Personnel Psychology, 30: 255-268. Lerner, J.S. & Tetlock, P.E. (1999). Accounting for the effects of accountability. Psychological Bulletin, 125: 255-275 Mauss, M. 1902/1972. A general theory of magic. New York: W. W. Norton. (R., Brain, Trans.; Original work published 1902). McConahay, J. B. 1983. Modern racism and modern discrimination - the effects of race, racialattitudes, and context on simulated hiring decisions. Personality and Social Psychology Bulletin, 9: 551-558. McConnell, A. R., & Leibold, J. M. 2001. Relations among the Implicit Association Test, discriminatory behavior, and explicit measures of racial attitudes. Journal of Experimental Social Psychology, 37: 435-442. Morales, A. C., & Fitzsimons, G. J. 2007. Product contagion: Changing consumer evaluations through physical contact with "disgusting" products. Journal of Marketing Research, 44:272-283 Moshavi, D. 2004. He said, she said: Gender bias and customer satisfaction with phone-based service encounters. Journal of Applied Social Psychology, 34, 162-176.
-50-
Murphy, K. R. 1991. Criterion issues in performance appraisal research: Behavioral accuracy versus classification accuracy. Organizational Behavior and Human Decision Processes, 50: 45-50 National Institute for Occupational Safety and Health. 2002. The changing organization of work and the safety and health of working people (NIOSH Publication No. 2002-116). Washington, DC: Author. Negro, G., Goodman, S., & Rao, H. 2008. Red Rattinghood: Brokerage, Closure, and Negative Affiliations in Hollywood Blacklisting. Working paper. Nosek, B. 2005. Moderators of the relationship between implicit and explicit evaluation. Journal of Experimental Psychology: General, 134: 565-584. O'Leary, V. E., & Ickovics, J. R. 1992. Cracking the glass ceiling: Overcoming isolation and alienation. In U. Sekaran & F. Leong (Eds.), Woman power: Managing in times of demographic turbulence: 7-30. Thousand Oaks, CA: Sage. O'Reilly, C. A., Williams, K. Y., & Barsade, S. G. 1998. Group demography and innovation: Does diversity help?" In D. Gruenfeld, B. Mannix & M. Neale, (Eds.) Research on Managing on Groups and Teams, pp. 183-207. Stamford, CT: JAI Press, Inc. Pellegrin, K. L., Stuart, G. W., Maree, B., Frueh, B. C., & Ballenger, J. C. 2001. A brief scale for assessing patients' satisfaction with care in outpatient psychiatric services. Psychiatric Services, 52: 816-819. Pulakos, E. D., White, L. A., Oppler, S. H., & Borman, W. C. 1989. Examination of race and sex effects on performance ratings. Journal of Applied Psychology, 74: 770-780. Richman, W. L., Kiesler, S., Weisband, S., & Drasgow, F. 1999. A meta-analytic study of social desirability distortion in computer-administered questionnaires, traditional questionnaires, and interviews. Journal of Applied Psychology, 84: 754-775. Ridgeway, C. L. 1991. The Social Construction of Status Value: Gender and Other Nominal Characteristics. Social Forces, 70: 367-386. Roch, S. G., & O'Sullivan, B. J. 2003. Frame of reference rater training issues: Recall, time and behavior observation training. International Journal of Training and Development, 7: 93-107. Rotundo, M., & Sackett, P. S. 1999. Effect of rater race on conclusions regarding differential prediction in cognitive ability tests. Journal of Applied Psychology, 84: 815-822. Rozin, P., Millman, T., & Nemeroff, C. 1986. Operation of the laws of sympathetic magic in disgust and other domains. Journal of Personality and Social Psychology, 50: 703-712. Rynes, S. L., Bretz, R. D., & Gerhart, B. 1991. The importance of recruitment in job choice - A different way of looking. Personnel Psychology, 44: 487-521.
-51-
Rynes, S. L., Heneman, H. G. III, & Schwab, D. P. 1980. Individual reactions to organizational recruiting: A review. Personnel Psychology, 33: 529-542. Rynes, S. L., & Miller, H. E. 1983. Recruiter and job influences on candidates for employment. Journal of Applied Psychology, 68: 147-154. Salam, S., Cox, J. F., & Sims, H. P., Jr. 1997. In the eye of the beholder: How leadership relates to 360-degree performance ratings. Group and Organization Management, 2: 185-209. Schneider, B., Ehrhart, M. G., Mayer, D. M., Saltz, J. L., & Niles-Jolly, K. 2005. Understanding organization-customer links in service settings. Academy of Management Journal, 48: 10171032. Sidanius, J., & Pratto, F. 1999. Social Dominance: An Intergroup Theory of Social Hierarchy and Oppression. Cambridge, U. K.: Cambridge University Press. Simonet, D. 2005. Patient satisfaction under managed care. International Journal of Health Care Quality Assurance, 18: 424-440. Simonin, B.L. & Ruth, J.A. 1998. Is a company known by the company it keeps? Assessing the spillover effects of brand alliances on consumer brand attitudes. Journal of Marketing Research, 35: 30-42. Sixma H. J., Spreeuwenberg, P. M., & van der Pasch, M. A. 1998. Patient satisfaction with the general practitioner: a two-level analysis. Medical Care, 36: 212-229. Song, M. 2004. Who's at the bottom? Examining claims about racial hierarchy. Ethnic and Racial Studies, 27: 859-877. Spence, M. 1973. Job market signalling. Quarterly Journal of Economics, 85: 355-374. Stewart, M., Brown, J. B., Donner, A., McWhinney, I. R., Oates, J., Weston, W. W., & Jordan, J. 2000. The impact of patient-centered care on outcomes. Journal of Family Practice, 49: 796-804. Tabachnick, B. G., & Fidell, L. S. 2003. Using Multivariate Statistics (Fifth ed.). Boston: Allyn and Bacon. Taylor, H. 2002. Patient/physician online communication: Many patients want it, would pay for it, and it would influence their choice of doctors and health plans. Harris Interactive Health Care News, 2: 13. The Economist. 2008. Black America: Nearer to Overcoming. May 8: 34-36 Westphal, J. D., & Stern, I. 2007. Flattery will get you everywhere (especially if you are a male caucasian): Ingratiation, boardroom behavior, and demographic minority status affect the likelihood of gaining additional board appointments at U.S. companies. Academy Management Journal, 50: 1-22.
-52-
Wherry, R. J. Sr., & Bartlett, C. J. 1982. The control of bias in ratings: A theory of rating. Personnel Psychology, 35: 521-551. Wilkinson, T. J., & Fontaine, S. 2002. Patients' global ratings of student competence. Unreliable contamination or gold standard?. Medical Education, 36: 1117-1121. Willerson, J. T., & Cohn, J. N. (Eds.). 2000. Cardiovascular medicine. Philadelphia: Churchill Livingstone. Wilson, G., Sakura-Lemessy, I., & West, J. P. 1999. Reaching the Top: Racial Differences in Mobility Paths to Upper-Tier Occupations. Work and Occupations, 26: 165-186. Woehr, D., & Roch, S. G. 1996. Context effects in performance evaluation: The impact of ratee gender and performance level. Organizational Behavior and Human Decision Processes, 66: 31-41. Wyer, R.S. Jr. & Schrull, T.K. 1989. Person memory and judgment. Psychological Review, 96: 5883. Yarkin, K. L., Town, J. P., & Wallston, B. S. 1982. Blacks and women must try harder: Stimulus person's race and sex attributions of causality. Personality and Social Psychology Bulletin, 8: 21-24. Yusuf, S., Sleight, P., Pogue, J., Bosch, J., Davies, R., & Dagenais, G. 2000. Effects of an angiotensinconverting-enzyme inhibitor, ramipril, on cardiovascular events in high-risk patients. New England Journal of Medicine, 342: 145-153. Zitzewitz, E. (2006). Nationalism in Winter Sports Judging and Its Lessons for Organizational Decision Making. Journal of Economics & Management Strategy, 15: 67-99
-53-
TABLE 1. Medcorp Sample Means, Standard Deviations, and Correlations between Predictor, Control and Dependent Variables4
M s.d. 1 0.51 0.11 0.66 0.47 -.30 0.80 0.20 -.07 1749.77 550.63 -.10 45.84 4.89 .07 1.04 0.12 .13 14.81 8.51 .20 50.34 6.58 .09 .12 0.32 -.15 .38 0.49 -.06 23.00 1.97 .05 -0.01 1.55 .11 0.16 0.15 .13 2 3 4 5 6 7 8 9 10 11 12
1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13.
Patient Satisfaction Practice Busyness Full time equivalent Number of patients in panel Panel age Chronic sickness of panel Tenure with Medcorp (years) Age (years) Nonwhite Female Productivity Quality Accessibility to Patients
.11 .26 .59 -.07 .05 -.12 -.15 -.14 .14 -.09 .16 .02 .01 -.06 -.63 .12 .22 .03 .07 -.11 -.18
-.03 -.13 .55 .08 .33 .21 .29 -.04 -.14 -.44 -.21 .30 -.06 .08 .21 -.23 .05
-.20 -.05 -.05 .04 .22 .05 -.06
.69 -.12 -.03 -.25 -.31 .12 -.25 -.01 -.04 -.15 .14 .11 .02 .04 -.03 -.04 -.16 -.11 .21 .05 .23
-
4
N = 113; All correlations larger than .15 are significant at p<.05
-54-
TABLE 2. Medcorp Study Analysis Examining Moderating Effects of Physician Race, Gender and Objective Performance on Patient Satisfaction with Physician5
Model 1 Controls Practice Busyness Full time equivalent Number of patients in panel Panel age Chronic sickness of panel Tenure with Medcorp (years) Age (years) Productivity Quality Accessibility Female Nonwhite Interactions Productivity X Tenure Quality X Tenure Accessibility X Tenure Productivity X Female Quality X Female Accessibility X Female Productivity X Nonwhite Quality X Nonwhite Accessibility X Nonwhite Adjusted R2 R2 ?R2 from Model 1 -.23** -.17 -.04 -.23* .23* .34** -.13 .09 .17* .15 -.22* -.12
Model 2 -.18* -.23 -.03 -.27* .25* .35* -.12 .09 .20* .13 -.26* -.09
Model 3 .26** -.20 .01 -.20 .20 .38** -.17 .12 .16 .07 -.17* -.18*
Model 4 -.22* -.24* .01 -.24* .22* .39** -.16 .14 .17* .05 -.20* -.17*
.06 .13 .12 .04
.09 .13 .01 .00 -.18** -.16*
.01 .14 -.16
.03 .14 .06 -.18** -.16* -.21** -.16* -.10 .25 .40 .15**
-.18* -.18* -.13 .17 .25 .21 .32 .07** .23 .33 .08**
5
* p<.05 ** p<.01 N = 113. All participants are medical doctors. The sample consisted of 100 whites, 10 Asian or Pacific Islanders, 2 blacks, and one Native American.
-55-
TABLE 3. Bookcorp Study Effect of Employee Race and Gender on Customer Satisfaction with the Employee6
Customer Satisfaction with the Employee White Male/White Female White Male/Nonwhite Male Condition Condition Model 1 Model 2 Model 3 Model 1 Model 2 Model 3 -.25* -.26 -.27 -.23* -.21 -.20 -.15 -.23 -.21 -.15 -.15 -.16 -.09 -.05 -.05 -.22* -.23* -.21 .05 .15 -.13 -.15 -.14 -.19 -.28* -.26* -.06 -.02 -.01 -.28** .14 .22 .08**
Nonwhite Female Age Implicit Association Test (IAT) score Woman condition (1=white woman employee, 0=white man employee) IAT Score X Woman condition Nonwhite condition (1=nonwhite man employee, 0=white man employee) IAT Score X Nonwhite condition Adjusted R2 R2 ?R2 from previous model
.03 .10
.08 .16 .06*
.08 .16 .00
.08 .14
.08 .14 .00
6
* p<.05 ** p<.01; N = 67 in the white man/nonwhite man condition and N = 54 in the white man/white woman
condition.
-56-
TABLE 4. Bookcorp Study Effect of Employee Race and Gender on Customer Satisfaction with the Organizational Context7
Customer Satisfaction with the Context White Male/White Female White Male/Nonwhite Male Condition Condition Model 1 Model 2 Model 3 Model 1 Model 2 Model 3 -.03 -.03 -.07 -.16 -.08 -.08 .09 -.03 .09 .16 .01 .01 -.26 -.19 -.14 -.17 -.04 -.04 -.12 .03 -.04 .13 .09 .02 -.45** -.38** -.23* -.44*** -.46*** -.18* .20 .26 .04*
Nonwhite Female Age Implicit Association Test (IAT) score Woman condition (1=white woman employee, 0=white man employee) IAT Score X Woman condition Nonwhite condition (1=nonwhite man employee, 0=white man employee) IAT Score X Nonwhite condition Adjusted R2 R2 ?R2 from previous model
.00 .07
.15 .24 .17**
.17 .29 .04**
.00 .07
.18 .22 .15***
7
* p<.05 ** p<.01 ***p< .001; N = 67 in the white man/nonwhite man condition and N = 54 in the white man/white
woman condition.
-57-
TABLE 5.8 Golfcorp Means, Standard Deviations and Correlations between Dependent, Independent and Control Variables. M
1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. Customer Satisfaction with Facility Size (Number of employees) Average Employee Age Percent Temporary Employees Average Customer Tenure (months) Percent Male Customers Average Customer Age (years) Percent Nonwhite Employees Percent Female Employees Facility Quality Facility Productivity (Profit per employee) 3.90 129.05 39.20 .04 60.66 .56 54.23 .26 .31 3.32 14614.26
s.d.
.22 157.51 4.04 .08 22.10 .07 8.40 .18 .12 1.04 7703.06
1
.16 -.18 .15 -.15 .00 -.14 .08 -.07 .08 .13
2
.09 -.07 .14 -.03 .17 -.12 -.21 .16 .14
3
4
5
6
7
8
9
10
.16 .15 -.13 .34 .11 .07 -.14 .12
.01 .06 .01 .20 .05 .16 .16
-.03 .42 -.28 -.07 -.04 .12
-.16 .15 .20 .16 .13
.08 -.06 -.03 .16
-.08 .10 .14
-.08 .09
.15
8
N = 66; all correlations greater than .21 are significant at p < .05
-58-
TABLE 6. Golfcorp Study Regression Results Examining the Interactive Influence of Percentage of Nonwhite Employees, Percentage of Female Employees and Objective Measures of Facility Performance on Customer Satisfaction with Facility9
Customer Satisfaction with Facility Model 1 Model 2 Model 3 Model 4 Controls Size Average Employee Age Percent Temporary Employees Average Customer Tenure Percent Male Customers Average Customer Age Facility Quality Facility Productivity Percent Nonwhite Employees Percent Female Employees Interactions Percent Female X Quality Percent Female X Productivity Nonwhite X Quality Nonwhite X Productivity Adjusted R2 R2 ?R2 from Model 1 .00 .13 .21 -.21 .18 -.12 -.06 -.09 -.04 .17 -.05 .04 .27* -.24* .16 -.06 -.08 -.13 -.01 .01 -.15 -.07 .27* -.24* .16 -.06 -.08 -.13 -.01 .01 -.15 -.07 .30** -.19 .18 -.06 -.11 -.10 -.03 -.05 -.19 -.14
-.25* -.22 -.25* -.34** .03 .17 .04*
-.23* -.31* Percent -.25* Percent -.49** .12 .28 .15** .20 .37 .24***
9
* p<.05 ** p<.01 ***p< .001; N = 66 country clubs.
-59-
FIGURE 1. Conceptual Model of How Bias Influences Customer Satisfaction Ratings10
Attributes of the Rating
Anonymity No Evaluation Standard Lack of Training
Race/Sex of Employee
Customer Racial/Gender Bias
H1 Service Provider Performance H2
H3a H3b
Rating of Employee
Rating of Context H4 Rating of Organizational Unit (Employee and Context)
Organizational Unit Performance
10
Dotted line indicates untested relationships. We expect main effects, but our contribution lies with tests of the interactions.
-60-
FIGURE 2. Medcorp Sample Interactive Effects of Physician Objective Performance and Physician Demographics on Patient Satisfaction with the Physician
High White Physician (**) High Male Physician (**)
Patient Satisfaction with Physician
Low Low High Nonwhite Physician (n.s.)
Patient Satisfaction with Physician
Female Physician (n.s.) Low Low High
Physician Quality
Physician Quality
White Physician (**)
High Male Physician (n.s.)
High
Patient Satisfaction with Physician
Low ** p < .01 * p < .05 n.s. p > .10 Low High
Female Physician (n.s.)
Patient Satisfaction with Physician
Low Low High Nonwhite Physician (**)
Physician Accessibility
Physician Productivity
-61-
FIGURE 3. Bookcorp Sample Interactive Effects of Employee Demographics and Customer Implicit Association Test Score on Customer Satisfaction with the Employee and with the Store Context
High
Satisfaction with Employee
White Male Employee (n.s.)
Low Low High
Nonwhite Male Employee (**)
Implicit Association Test Score
White Male Employee (**)
High
High
White Male Employee (*)
Satisfaction with Context
Satisfaction with Context
White Female Employee (*) Low Low High
** p < .01 * p < .05 Low n.s. p > .10 ** p < .01 * p < .05 n.s. p > .10
Nonwhite Male Employee (*) Low High
Implicit Association Test Score
Implicit Association Test Score
-62-
Appendix 1: Customer Satisfaction Items Used Across the Three Studies Medcorp Study Customer Satisfaction with the Employee: How would you rate the following attributes of your provider (1 = very poor; 5 = excellent): 1. Attention provider paid 2. Thoroughness and competence of provider 3. Ability to ask questions of this provider Bookcorp Study Customer Satisfaction with the Employee: How would you rate the following (1 = very poor; 7 = excellent): 1. Speed of service 2. Quality of service 3. Availability of staff for assistance 4. Employee responsiveness to customers' issues and concerns Bookcorp Study Customer Satisfaction with the Context: How would you rate the following aspects of the bookstore: 1. Appearance of bookstore (1 = very poor; 7 = excellent) 2. Environment of the bookstore was conducive to learning/reading (1 = strongly disagree; 7 = strongly agree) 3. The bookstore has up to date equipment 4. This bookstore's physical facilities are visually appealing 5. The appearance of this bookstore is in keeping with the type of services provided 6. Bookstore meets expectations (1 = less than expected, 7 = better than expected) 7. Likelihood of recommending bookstore to others (1 = definitely would not, 7 = definitely would) Golfcorp Study Customer Satisfaction with the Facility: How would you rate the following aspects of your club (1 = very poor; 5 = very good): 1. Maintenance of grounds/Appearance of clubhouse 2. Locker rooms and Rest Rooms 3. Quality of greens 4. Condition of course 5. Pace of play 6. Condition of practice facilities 7. Ability to obtain desired tee times 8. Club meets expectations (1 = less than expected, 5 = better than expected) 9. Likelihood of recommending club to others (1 = definitely will not, 5 = definitely will)
-63-
David R. Hekman is an Assistant Professor of Management at the Lubar School of Business, University of Wisconsin-Milwaukee. He earned his Ph.D. in Management from the University of Washington. He is interested in improving organizational health by minimizing organizational problems such as weak employee attachment, contagious harmful behaviors, and persistent workplace inequality. Karl Aquino is the Richard Poon Professor of Organizations and Society at the Sauder School of Business, University of British Columbia. His research interests are moral behavior, workplace victimization, revenge and forgiveness, and status and power dynamics in organizations. He received his Ph.D. in Organizational Behavior from Northwestern University. Bradley P. Owens is a Post Doctoral Fellow at the Center for Positive Organizational Scholarship at the University of Michigan. He earned his Ph.D. in Management from the University of Washington. His primary research interests include humility, leadership, race and gender issues, and team dynamics. Terence R. Mitchell received his undergraduate degree from Duke in 1964, and a Masters and Ph.D. from the University of Illinois in organizational psychology. He has been at the University of Washington since 1969 and in 1987 he was appointed the Carlson Professor of Management. He has published journal articles and book chapters on the topics of motivation, leadership, turnover and decision making. Pauline Schilpzand is an Assistant Professor of Character Development at the Army Center of Excellence for the Professional Military Ethic at West Point. She earned her PhD in Management from the Warrington College of Business Administration at the University of Florida. Her primary research interests include behavioral ethics, leadership, and the selection process. Keith Leavitt is an Assistant Professor of Character Development at the Army Center of Excellence for the Professional Military Ethic at West Point. He earned his PhD in Management from the Foster School of Business at the University of Washington. His primary research interests include behavioral ethics, implicit social cognition, research methods, and organizational citizenship.
-1-
doc_998563967.docx
Case Study on Examination of Whether and How Racial and Gender Biases Influence Customer Satisfaction, Customer satisfaction, a term frequently used in marketing, is a measure of how products and services supplied by a company meet or surpass customer expectation. Customer satisfaction is defined as "the number of customers
.
Case Study on Examination of Whether and How Racial and Gender Biases Influence Customer Satisfaction
ABSTRACT We examine whether and how various biases may influence customers' satisfaction evaluations and produce discriminatory judgments for minorities and female service employees. We argue that customer satisfaction evaluations are biased because they are anonymous judgments by untrained raters that usually lack an evaluation standard. In our laboratory and field samples, we find disturbing evidence generally confirming our arguments and suggesting that the presence of nonwhite and women service employees may produce lower aggregated customer satisfaction evaluations which may ultimately hurt individuals and organizations financially. (84 words)
1
Parts of this research were supported by the Business and Economic Development Center, Foster School of Business, University of Washington. We would like to thank the AMJ action editor Peter Bamberger as well as three anonymous reviewers for their help in the review process.
-2-
Customer satisfaction surveys have become a common source of performance feedback for employees and organizations (Hagan, Konopaske, Bernardin, & Tyler, 2006). Mercer Consulting Group reports that in 2006 customer satisfaction surveys were of primary importance for strategic decision making and over two-thirds of organizations used such surveys to determine some aspect of employee compensation (Mercer Consulting, US Policies and Practices Report, 2007). Moreover, customer satisfaction is an important predictor of a wide range of financial measures (see Gupta & Zeithaml, 2006 for a review) so it is not surprising that some companies tie some portion of employee compensation directly to customer satisfaction. For example, a one percent change in customer satisfaction for an average Fortune 500 firm has been shown to lead to a 1.02% change in Tobin's q which equates to a change of $275 million in firm value (Anderson, Fornell & Mazvancheryl, 2004), a $55 million gain or loss in cash flow in the next year (Gruca & Rego, 2005), and a 5.03% change in Return on Investment (Anderson & Mittal, 2000). Satisfying the customer is also increasingly important to organizations as the global economy becomes more service-oriented. Macroeconomic trends indicate 76 percent of U.S. employees work in a service industry, and by 2016 the number of employees working in a service industry is expected to increase by over 17 million (Figueroa & Woods, 2007). The expanding service sector is perhaps one reason why among 681 senior executives surveyed by The Economist Intelligence Unit during October-December 2002, 65% reported customers as their main focus over the next three years compared to only 18% who reported shareholders as their main focus. Many business leaders (Bracken, Church & Timmreck, 2001) and researchers (Salam, Cox & Sims, 1997) have applauded the use of customer satisfaction surveys because they believe that aggregated evaluations are highly reliable measures of employee performance quality. However, a potential disadvantage of using customer surveys, particularly for purposes of making compensation or
-3-
promotion decisions, is that they are ultimately subjective judgments. As a result, they are vulnerable to biases, including those based on the bandwagon effect, confirmation of pre-existing beliefs, education or cognitive ability, as well as stereotypic biases based on the race or gender of the person being rated (Gilovich, Griffin & Kahneman, 2002). Some researchers suspect that biases are unavoidable when gathering subjective evaluations of performance, especially when such judgments come from naïve and inexperienced raters who are not held accountable for the accuracy of their ratings (Pulakos, White, Oppler & Borman, 1989; Wilkinson & Fontaine, 2002; Woehr & Roch, 1996) To date, though, surprisingly little research has examined how and if different biases influence customer judgments about whether an organization's employees, organizational attributes, or services meet or exceed their expectations. The purpose of our research is to examine whether and how customer satisfaction ratings are potentially influenced by gender and racial bias. We extend the existing literature on biases in supervisory ratings of employee performance by focusing on customer satisfaction ratings, which have been mostly excluded from organizational behavior research (see Moshavi, 2004 for a rare exception). We conceptualize these satisfaction ratings as judgments and we examine not only judgments made about individual employees, but also about the organizational context (e.g., perceived cleanliness or appearance) and the organizational unit as a whole. To our knowledge, no research has examined bias in customer judgments of the organizational context or the overall organizational unit. Based on the literature on modern forms of racial and gender bias (Greenwald & Banaji, 1995; Crandall & Eshelman, 2003), we contend that customer satisfaction with organizational contexts or units may be vulnerable to such biases. Finding empirical evidence of racial or gender biases in customer satisfaction would suggest that, from the organization's perspective, there is a financial incentive to
-4-
favor white or male employees. Such a finding may help explain the persistent inequality between demographic groups in the workplace. Two methodological attributes of our research distinguish it from previous studies of bias in customer judgments of employee performance and allow us to conduct a stronger test of the validity of our predictions. First, we take into account employees' objectively-measured performance behaviors when examining customer judgments. The problem with relying solely on customer satisfaction scores to assess customer bias is that such scores can be interpreted as capturing both true performance scores and biases (Latham & Wexley, 1977; Landy, Shankster, & Kohler, 1994). As Rotundo and Sackett (1999: 816) summarize this state of affairs, "There is no definitive way of determining whether the rated criterion used in a validity study is biased. Thus, there is no current method of establishing whether there is bias in performance ratings." Our study design allows us to tease apart differences in satisfaction judgments that are attributable to customer bias arising from employee demographic characteristics from those that are due to objective employee performance (Greenhaus, Parasuraman & Wormley, 1990; Pulakos et al., 1989; Wilkinson & Fontaine, 2002; Woehr & Roch, 1996). Second, we heed recent calls for researchers to engage in "full-cycle" research where initial field-based findings are tested in the laboratory and then re-validated in a different field setting (Chatman & Flynn, 2005; Cialdini, 1995). Full-cycle research allows researchers to compensate for the weaknesses of one context or study design with the strengths of another. It also allows the researcher to investigate a broad initial question in a field setting (e.g., is there bias in customer judgments?) followed by a laboratory study that can utilize more control and enable the researcher to examine more specific questions in detail (e.g., what is a potential cause and consequence of bias in customer judgments?). Finally, the investigator can move back to a field setting to confirm findings
-5-
found in the first two studies. The interplay of field and lab designs prescribed by the full-cycle approach fosters greater theoretical insight as to the causality and generalizability of study findings. Following the full-cycle research model, we first test for bias in customer judgments regarding a sample of professional employees (e.g., doctors). Next, in a carefully-controlled laboratory setting, we test for customer bias again, but this time in a bookstore context. We also identify and measure a specific mechanism that might explain the observed effect. Finally, we test for customer bias in judgments of an organizational unit with a sample consisting of country clubs belonging to a large hospitality company. Since the focal unit rated shifts from the individual in the first sample to the organizational in the third sample, we are able to provide some initial confirmation of the generalizability of our theory. In the following section, we present the theoretical rationale for our predictions regarding the possible effects of customer diversity related biases on customer satisfaction judgments. THEORETICAL BACKGROUND AND HYPOTHESES Customers are often asked to assess an individual service provider (e.g., their salesperson, teller, teacher, or physician; Haas, Cook, Puopolo, Burstin, Cleary & Brennan 2000; Sixma, Spreeuwenberg, & van der Pasch, 1998; Davis & Davis, 1999), the quality of the environment in which they were served (e.g., the merchandise available, the newness or cleanliness of the setting, or the efficiency of the technology; Pellegrin, Stuart, Maree, Frueh, & Ballenger, 2001; Simonet, 2005), or the unit or group providing the service (e.g., the bank, school, country club or law firm; Anderson et al., 2004; Ittner & Larcker, 1998). This last type of judgment is likely to include opinions about both the server(s) and the context in which the economic transaction occurs and is therefore a more global judgment than the other two types previously mentioned. In this study, we investigate the possibility of systematic bias in all three types of satisfaction judgments thereby providing a strong test of the
-6-
potential generality of such biases. An assumption of our paper is that like anyone else who makes a social judgment customers are not immune to information processing biases. One potential source of this bias arises from the demographic characteristics of the service providers who are being rated. Racial and Gender Biases in Customer Satisfaction Judgments U.S. society has made considerable progress in reducing overt expressions of prejudice since the Civil Rights movement during the 1960s (Bobo, 1998). Yet despite these gains, there is abundant social psychological evidence that biases against women and minorities persist in a more covert and non-conscious form. Researchers have used terms like modern racism (McConahay, 1983), aversive racism and sexism (Dovidio & Gaertner, 1981; Gaertner & Dovidio, 1986), or implicit gender and racial stereotypes (Greenwald & Banaji, 1995) to describe these types of biases and many studies have demonstrated their influence on information processing and judgment across a variety of social domains (see Brief, Dietz, Cohen, Pugh & Vaslow, 2000 for an overview). For example, one study showed that when job applicant resume quality was ambiguous, applicants with African-Americansounding names (e.g., Aisha, Rasheed) were much less likely to be called for a job interview than applicants with white-sounding names (e.g., Kristin, Brad) (Bertrand & Mullainathan, 2004). Similarly, when evaluators of orchestral position applicants could see the applicant's gender they were more likely to select men. When the applicant's gender could not be observed, the number of women hired significantly increased (Goldin & Rouse, 2000). In another study, Dovidio and Gaertner (2000) found that while raters were not biased against blacks in a simulated hiring decision when the applicants were clearly qualified or unqualified for a job, raters were biased when the applicant's qualifications were ambiguous. Dovidio and Gaertner (2000) interpreted this finding as supporting an aversive racism framework in which prejudice occurs in a more subtle form in contexts with ambiguity or uncertainty. Based on considerable evidence demonstrating the operation of covert and unconscious
-7-
racial and gender biases across a variety of social domains, there is reason to suspect that such biases can also influence customer satisfaction judgments. Our theoretical arguments supporting the influence of racial and gender biases in customer satisfaction judgments are based on the idea that observers (e.g., customers) have preconceived expectations about others depending on whether the person being observed belongs to a high or lowstatus demographic group (Berger, Conner & Fisek, 1974; Berger, Fisek, Norman & Zelditch, 1977). In the United States, men and whites are considered by most people as members of a high-status social group relative to women and ethnic minorities (see Ridgeway, 1991 for a review). One of the benefits of belonging to a high-status social group is that observers are more likely to make favorable inferences about one's competence, normality, and legitimacy (Aquino & Bommer, 2003; Giannopoulos, Conway & Mendelson, 2005; Sidanius & Pratto, 1999). In contrast, members of lowstatus groups are subject to negative stereotypes and attributions concerning their work-related competencies (Fernandez, 1981; O'Leary & Ickovics, 1992). Based on rating theory (Wherry & Bartlett, 1982), we believe there are at least three reasons to expect customer satisfaction ratings to be susceptible to such stereotypes and racial and gender biases. First, an important difference between performance ratings made by supervisors and judgments made by customers is that usually customers are afforded the luxury of anonymity. Although supervisors' and customers' ratings are both viewed by organizational administrators, in most cases only supervisors are identifiable. Customer anonymity decreases accountability and the desire to engage in the effortful cognitive processing required to conceal or overcome any biases (Richman, Kiesler, Weisband & Drasgow, 1999; Lerner & Tetlock, 1999). Moreover, supervisors, but not customers, know that their ratings are part of the employee record and are used for employee training, feedback, and advancement decisions (Murphy, 1991). Not only are supervisors identifiable, but they must also
-8-
justify their ratings and as such are even more motivated to engage in effortful information processing to help them reduce the influence of racial or gender bias and appear, at least superficially, to be objective. Customer anonymity does not motivate raters to reduce bias, but the customer satisfaction questionnaire instructions and items may actually strengthen the effect of such biases. Supervisors completing a performance appraisal are frequently reminded of the importance of rating accuracy and rate on specific behavioral items (Judge & Ferris, 1993; Kane, Bernhardin & Villanova, 1995). Customers, however, are typically asked only for their "opinions" or "attitudes" about employees or the organization (suggesting they "make a judgment") (Schneider, Ehrhart, Mayer, Saltz, & NilesJolly, 2005). Common customer satisfaction items like "I would recommend this organization to others" and "this organization/employee meets my expectations" do not solicit recall of specific employee or organizational behaviors and so they may provide more information about the state of mind of the rater than the actual performance of the ratee or the organization. Such items are also problematic because customers may have higher expectations for women and non-whites to demonstrate their competence and so will provide lower customer satisfaction judgments for such workers even if their performance is objectively equivalent to that of their male or white counterparts (Biernat & Kobrynowicz, 1997; Yarkin, Town & Wallston, 1982). A variety of techniques have been shown to effectively reduce bias in performance appraisal. For example, Roch and O'Sullivan (2003) found that a combination of frame-of-reference (i.e. having raters establish a prototype of good performance) and behavioral observation training (i.e. focusing raters on specific behaviors) leads to increased accuracy in appraisal. Baltes and Parker (2000) found that halo error training (i.e. knowing what factors should not influence ratings) and structured recall memory intervention (i.e. memory enhancement techniques) reduce bias in performance ratings.
-9-
DeNisi, Robbins, and Cafferty (1989) argued that behavioral diaries aid in the recall and categorization of behavioral events. Customers are not trained or expected to use these techniques when forming satisfaction judgments. In sum, customer satisfaction judgments are likely to be highly susceptible to racial and gender biases because customers are usually anonymous, asked to make summary judgments rather than provide accurate recall of performance-related behaviors, and are untrained in techniques that might help them overcome non-conscious biases. Our arguments lead us to expect that, in general, employees belonging to low-status demographic groups (i.e., women, racial minorities) will receive lower customer satisfaction scores than employees belonging to high-status demographic groups. But even if we found evidence for this difference, it would not demonstrate the operation of bias in customer judgments because it may be that the lower customer satisfaction judgments received by members of lower status demographic groups may in fact indicate lower levels of true performance. Logically, customer satisfaction judgments should be at least partly influenced by employee objective performance (Wherry & Bartlett, 1982). Consequently, more direct evidence of bias would be demonstrated if behaviors performed by a high-status group member are viewed more favorably by customers than the same behaviors performed by someone from a low-status group. Evidence for the plausibility of this hypothesis comes from studies showing that women in leadership roles are rated lower than men in similar roles (Eagly, Makhijani & Klonsky, 1992) and that ethnic minorities and women are rewarded less than whites and males for exhibiting the same advice-giving or ingratiatory behaviors (Westphal & Stern, 2007). There is also evidence that racial minorities and women who achieve equivalent levels of performance as whites and men are judged as having less underlying ability (Biernat & Kobrynowicz, 1997; Yarkin, Town & Wallston, 1982). To examine whether the
-10-
biases in judgments of competence as a function of a target person's group membership found in prior research generalize to customer judgments, we tested the following hypothesis: Hypothesis 1: The relationship between employee objective performance and customer satisfaction judgments will be attenuated for employees belonging to low-status demographic groups compared to employees belonging to high-status demographic groups. Hypothesis 1 deals with customer judgments that ask for evaluations about an individual service provider. But we also believe that racial and gender biases can influence evaluations of the organizational context (e.g., cleanliness, appearance). Our prediction draws from the notion that the positive or negative properties of an item or person can "spill over" on to the nearby context or surrounding targets (Rozin, Millman & Nemeroff, 1986). Our logic is based on the simple idea that evaluations of different aspects of service experience (e.g. the employee, the context) are connected in the rater's non-conscious and conscious belief system (Argo, Dahl, & Morales, 2009; Morales & Fitzsimons, 2007). Theoretical and empirical work on the cognitive structure of attitudes (Anderson, 1981; Wyer & Schrull, 1989) suggests that our evaluation of any person, object, or idea is partly based on our evaluations of other persons, objects or ideas with which the target object is linked. Recent work in marketing has elaborated on this idea (Keller, 2003). Research shows that the evaluation of a product, service or brand is partly related to the evaluation of the persons who are associated with the product, service or brand (e.g., the person in an advertisement, the service provider)(Folkes & Patrick, 2003; Simonin & Ruth, 1998; Morales & Fitzsimons, 2007). Accordingly, we expect that customers' evaluations of organizational contexts will be non-consciously connected to their evaluations of highly visible employees in the organization. Although such a non-conscious connection may be unwarranted, we expect customers' attitudes about low-status employees to be reflected in less favorable evaluations of an organizational context where such employees are highly visible.
-11-
It may even be possible for customers to make a conscious connection between employee status and the level of organizational context quality. (e.g., Rynes, Heneman, & Schwab, 1980; Rynes & Miller, 1983; Spence, 1973). In line with this thinking, job applicants believe that recruiters' competence and thoroughness signals an organization's overall quality (Rynes, Bretz & Gerhart, 1991). Customers may judge an organizational context that employs a low-status employee as signaling an inferior physical and social environment than one employing a high-status employee. Combining our spillover and signaling arguments, we expect the mere presence of members of lowerstatus groups in the organizational environment will lead to less positive customer judgments of the service environment. The following hypothesis tests this prediction: Hypothesis 2: Individuals will report lower customer satisfaction judgments of the environmental context when a highly visible employee in that environment belongs to a low (i.e., women, African-American) rather than high (i.e., male, white) status demographic group. Not every customer will be so susceptible to racial or gender bias that they would evaluate an employee and the environment more negatively simply because they encounter an employee who belongs to a low-status demographic group. We expect the customers who are most prone to making these types of judgments to be those who hold more negative pre-existing attitudes toward females or racial minorities. Individuals with negative pre-existing attitudes toward members of low-status groups quickly associate negative words (e.g. terrible, nasty, evil) with pictures of nonwhite or female faces, and quickly associate positive words (e.g. laughter, glorious, joy) with pictures of white or male faces (Greenwald & Banaji, 1995). We tested this possibility by examining whether the degree to which individuals have non-conscious, negative attitudes toward members of low-status groups would moderate the effect proposed in our first two hypotheses as stated in the following hypotheses:
-12-
Hypothesis 3a: Low-status employees (i.e., women, African-Americans) will receive even lower customer satisfaction judgments than equally- performing high-status employees (i.e., men, whites), when the judges have negative attitudes toward members of low-status groups. Hypothesis 3b: Contexts employing low-status employees will receive even lower customer satisfaction judgments than contexts employing equally- performing high-status employees, when the judges have negative attitudes toward members of low-status groups. Thus far we have argued that customer satisfaction judgments can be influenced by perceiving a single employee belonging to a low-status group within the service environment. But in many cases customers interact with a variety of employees in a given customer service encounter. For example, when doing something as simple as buying groceries, customers observe and interact with deli workers, produce employees, cashiers, and baggers. Employees in each of these positions likely have different demographic characteristics. For this reason, another way to examine the possible influence of bias in the customer judgment process is to see whether the demographic composition of the organizational unit might influence customer judgments of that unit. Extending our theoretical argument to the organizational level, we expect that the degree to which an organizational unit's employees are members of low-status demographic groups will influence customer satisfaction judgments of that organizational unit such that it will be judged as being of worse quality compared to an organizational unit whose employees are mostly from high-status demographic groups. But as we noted when predicting the influence of rater bias on judgments of individual service providers, such a finding might reflect true differences in unit performance. Hence, we make the following, more precise prediction that parallels hypothesis 1:
-13-
Hypothesis 4: The relationship between an organizational unit's objective performance and customer satisfaction will be attenuated as a function of the percent of unit employees belonging to low-status demographic groups (i.e., women and minorities) Our theoretical arguments and hypotheses can be summarized by the conceptual model shown in Figure 1. --------------------------Insert Figure 1 here --------------------------We tested the hypotheses in our model in three studies using different samples and methods. We tested hypothesis 1 by looking at customer judgments of their physician (Medcorp study); we tested hypotheses 1, 2 and 3 by examining customer judgments of a bookstore and one of its employees (Bookcorp study); finally we tested hypothesis 4 by examining customer judgments of their golf club (Golfcorp study). In each study, a large number of customers rated each of the targets. MEDCORP STUDY Our first sample was drawn from all 113 primary-care physicians (i.e., family practitioners) employed by a large health maintenance organization, hereafter referred to as Medcorp (a pseudonym). Medcorp provides coverage and healthcare for about 350,000 people in the Pacific Northwest region of the United States. Within our sample, 38.4% were women, 11.5% were ethnic minorities, and all had a medical degree. The 2006 Diversity Report by the Association of American Medical Colleges reports that 24.5% of practicing physicians are women and 12.1% are nonwhites.2 Measures Medcorp routinely collected patient satisfaction ratings as well as objective behavioral indicators of physician performance that were assumed to have a direct, positive impact on patient
2
However, the Association of American Medical Colleges also reports that these numbers are changing dramatically as 44% of American medical school graduates in 2006 were women and 34% were nonwhite.
-14-
health and well-being. This feature of our data represents a methodological improvement over studies that only measure employee performance with a single subjective rating and are therefore unable to determine if the rating is biased (Rotundo & Sackett, 1999). The dependent variable in our study was patient satisfaction with their physician. The independent variables were physician demographics (race and gender) and three types of patient-centered behaviors. Customer satisfaction. A postcard survey was mailed to a percentage of each physician's patients, following doctor visits, selecting the patients so as to avoid a bias toward those patients with frequent appointments. Patients completed and returned a total of 12,091 surveys for a response rate of 52%, so that each physician was rated by an average of 107 patients. Each patient rated only one physician, so the individual ratings were independent. Patients rated each of the three items targeting their physician on a 5-point Likert scale (1=very poor; 5=excellent) "How would you rate?" (1) the attention the provider paid to you; (2) this provider's thoroughness and competence, and; (3) your opportunity to ask questions of this provider. The three items were highly correlated (average correlation is .93) so the organization combined them to create a composite patient satisfaction variable. These items are more general questions rather than very specific behaviors (e.g. minutes spent with provider, number of questions the doctor asked). The organization did not provide us with access to raw patient-level surveys. Instead, they provided us with data indicating what percentage of each physician's patients rated the physician as "excellent." Thus, the range on this measure for each physician was from 0 to 100 percent. This measure was collected in the same quarter as all other variables. Physician race. Medcorp identified each physician's race, and we coded whites "0" and ethnic minorites as "1." Of the 113 physicians in the sample, 10 were Asian or Pacific Islanders, two were
-15-
black, and one was Native American. The percentage of ethnic minority physicians in our sample is consistent with the national average of 12.1%. Physician gender. We coded males "0" and females "1." Forty-three of the physicians were female, which is slightly higher than the national average of 24.5%. Objectively measured employee performance. With the growing prevalence of health maintenance organizations (HMOs) and the increasing corporatization of medicine (Feinglass & Salmon, 1990), patients are increasingly being viewed by organizational administrators and physicians as customers. Therefore, physicians are increasingly being rewarded for engaging in behaviors that benefit their patients and the organization's customers (Laine & Davidoff, 1996; Stewart et al., 2000). We used the customer-benefitting behaviors identified by Medcorp as our indicators of objective physician performance. Medcorp measures customer-benefitting behaviors along three dimensions. The first is physician productivity, which is the number of health procedures performed and issues discussed in a given time period. The second is the physician's accessibility to customers measured by the number of secure emails that doctors send to customers. The third is the physician's level of quality measured by the standardized prescription rates of particular medications for customers that possess precise disease criteria. All three dimensions reflect behaviors that benefit customers by reducing the amount of time and money customers spend receiving medical care. For all metrics, physicians are shown how they compare to both the organizational goal and the organizational average. More productive physicians are able to treat more customer problems per visit, thereby saving customers' time and trips to the doctor. More accessible physicians provide greater convenience to customers who can simply email any medical questions to their physician. Higher quality physicians are better at preventing costly and deadly health events such as strokes and heart attacks. Physician compensation is tied to each of these
-16-
customer-benefitting behaviors. Physicians who exceed the 40th percentile are given a bonus, while those below the 40th percentile are not given a bonus. Physician productivity. The average number of patients seen, medical issues discussed, and medical procedures performed by each doctor in a standardized 8-hour day was recorded by the organization's scheduling software. Medcorp physicians have a great deal of control over the amount of work that they do in a day as they can control the intensity of each visit (e.g., the number of procedures performed and patient health issues addressed per visit). The number of patients physicians see each day is controlled by organizational administrators. The objective performance assessment we used was the composite of average face-to-face visits and phone visits adjusted by the average intensity of each visit. Intensity was measured by Relative Value Units (RVUs), which are coded by physicians at the end of each visit according to national coding guidelines. RVUs capture the amount of time involved, the required physical and mental effort, the required judgment and technical skill, and the psychological stress entailed (Hsaio, Braun, Becker, & Thomas, 1988; Hsaio, Braun, Dunn, & Becker, 1988). Physicians check one of three RVU boxes after seeing each patient. If the patient appointment is a quick check-back or follow-up appointment, physicians check the first box, which is worth .5 RVUs. If the patient appointment involves at least two patient issues or concerns, but less than four, then the physician checks the middle box, indicating 1.0 RVUs for that visit. If the patient appointment involves five or more patient issues, then the physician checks the third box, which indicates 1.5 RVUs. According to quarterly audits by administrators, Medcorp physicians accurately record RVUs in 90 percent of patient visits. Coding errors resulting from physicians coding too many or too few RVUs are normally and equally distributed. The raw measure of productivity was standardized based on the full-time status of the
-17-
physician and then multiplied by each physician's average visit intensity to obtain the quarterly average RVU-adjusted patient encounters per day. Physician accessibility. The average daily number of secure emails that physicians sent to patients for the quarter was used to measure another type of customer-benefitting behavior. Patients highly value the ability to easily contact their physician. Indeed, a Harris poll has shown that 90% of Americans who are online want the ability to e-mail their physicians, and 37% are even willing to pay for it (Taylor, 2002). Medcorp patients and physicians can communicate electronically regarding health related issues through a secure Internet health portal designed exclusively for patient-doctor communication. To use the system, patients log in to a secure website that provides the patients access to their personal health records, their lab results, and a host of health related information. Patients can send unlimited secure emails through the portal to any physician they visited in the prior two years at no cost and Medcorp physicians are expected to reply to each secure email within 24 hours. Patients are encouraged to contact their physician via the system to ask basic health-related questions, to request prescription refills and to schedule follow-up appointments. Medcorp administrators assign an equal number of patients to each physician (taking into account patient sickness, age and gender) and think that the system saves patients doctor visits, thereby saving patients time and money. In general, physicians do not think that email improves the quality of patient care, but rather that secure email increases convenience for patients (Kleiner, Akers, Burke & Werner, 2002). The Medcorp computer server automatically recorded the number of emails that each medical professional sent to his or her patients. Medical professionals had a great deal of control over how many emails they sent for two reasons: (a) they could try to discourage patients from using the system, and (b) they could choose whether to personally respond to their patients' emails. We calculated the number of emails physicians sent per day, taking into account the number of full working days that
-18-
physicians were in clinic during the time-period of this study. To enhance the normality of the variable, we used an inverse transformation and then reflected these values such that higher values represented greater use (Tabachnick & Fidell, 2003). Physician quality. Every Medcorp primary care physician is responsible for a panel of member-patients. Of the thousands of possible treatments, prescriptions, and procedures that physicians can perform to benefit patients, one of the most important is each physician's prescription rate of statins and angiotensin-converting-enzyme (ACE) inhibitors to patients with cardiovascular disease. Treatment of cardiovascular events such as strokes, clots, and heart attacks is the biggest healthcare cost for patients in the U.S. (Willerson & Cohn, 2000), and these drugs prevent cardiovascular events over patients' lifetimes (Gerstein et al., 2000). 3 According to Medcorp guidelines, all patients with cardiovascular disease should be regularly taking ACE inhibitors and some form of a statin. ACE inhibitors lower blood pressure, and statins lower cholesterol. Nationally, only 50% of all cardiovascular disease patients that should be treated with statins and ACE inhibitors are currently taking such medication (Dubois et al., 2002). These drugs significantly lower the immediate risk of a cardiovascular event (e.g., stroke, heart attack) for all individuals, regardless of sex or previous history of cardiovascular disease (LaRosa, He & Vupputuri, 1999; Yusuf, Sleight, Pogue, Bosch, Davies& Dagenais, 2000). To promote a higher prescription rate, Medcorp administrators send emails to physicians reminding them to prescribe such treatment. While prescription of these medications benefits patients by helping patients avoid death and reduce healthcare expenses, physicians often forget to prescribe these medications (Isles, 2002). This quality variable is the composite of the percent of cardiovascular disease patients 18 years and older who were dispensed the equivalent of a 90-day supply for ACE inhibitors and statins at any
3
We call this variable "quality" because statin and ACE inhibitor prescription rate accuracy are measures of physician quality according to the most influential quality assurance organizations (e.g. HEDIS, NCQA, IHI).
-19-
time within the quarterly reporting period. The component variables approached normality, were standardized, and were added together. The resulting variable was each physician's overall prescription rate of statins and ACE inhibitors for cardiovascular disease patients. The average prescription rate at Medcorp is 50%, similar to the national average. Control Variables We controlled for several variables that were not of direct interest for testing our hypotheses but that could be theoretically related to the dependent variable and might provide plausible alternative explanations for our findings. Average practice busyness. Patients who have to wait long periods of time to see their physician may be less satisfied, so we controlled for the busyness of each physician's practice. At the close of business each day, the Medcorp computer counts how many days into the future each physician's third available appointment is. According to the National Quality Measures Clearinghouse, counting the days until the third next available appointment is the healthcare industry's standard measure of access to care and indicates how long a patient waits to be seen. Doctors who are not very busy typically have three available appointments the next day, whereas busy doctors often do not have three available appointments for several days. The final variable was the quarterly average number of days until each physician's third open appointment slot. Physician full-time status. We included the number of hours a physician worked in our model because patients may be more satisfied if their physician works more hours. Physicians ranged from working 30 to 100 percent of a full-time position. Number of patients in panel. Medcorp assigns physicians to care for a particular group (i.e. panel) of patients. Patients in larger panels may be less satisfied and so we controlled for the total number of patients in each physician's panel standardized by the full-time status of the physician.
-20-
Average patient age. Older patients may have different expectations about doctor demographics, so we included the average patient age for each physician's panel in our model. Average chronic sickness of panel. Sicker patients may be less satisfied, so we controlled for the panel chronic sickness variable calculated by Medcorp (e.g., it captures the percentage of patients with diabetes and cardiovascular disease). Physician age and tenure. Physicians who are older or who have been employed by Medcorp for more years may have more loyal, satisfied patients. Physician tenure by objective performance. Because women and nonwhite physicians tend to be more recently hired than male and white physicians, any influence of physician race and gender on customer satisfaction may be masked by physician tenure. We therefore included the interactions of tenure by objective performance in our models so that we could more clearly determine the interactive influence of physician race by objective performance and physician gender by objective performance on customer satisfaction. Results Table 1 reports the means, standard deviations, and correlation coefficients between the dependent, independent, and control variables. We found no significant differences in our objective measures of performance based on employee race and gender. Our first hypothesis stated that the relationship between employee objective performance and customer satisfaction judgments would be less positive for employees belonging to low-status demographic groups compared to employees belonging to high-status demographic groups. To test this, we examined the interactions of the objective measures of employee performance (i.e. quality, productivity and accessibility) by employee race and gender. We used hierarchical moderated regression models to do so (Aiken & West, 1991). We centered all variables involved in the interaction terms to minimize multicollinearity between the
-21-
interaction terms and their individual components (Aiken & West, 1991). We entered all of the control variables in Model 1. In Model 2 we entered the control variables plus the interactions involving physician gender. In Model 3 we entered all the control variables as well as the interactions involving physician race. Finally, in Model 4, we entered all control variables and all interaction effects. Table 2 presents the results of this analysis. ----------------------------------Insert Tables 1 and 2 about here ----------------------------------The two-way gender X objective performance interactions as a set explained a significant amount of incremental variance in the dependent variable (R2 = .07, p< .01) providing preliminary support for hypothesis 1. Inspection of the individual regression weights showed that the physician accessibility X gender and physician quality X gender interactions were significant (p< .05). We probed the pattern of the interaction by examining the simple slope of the objective performance measures for male and female physicians (Aiken & West, 1991). The results of this analysis are shown graphically in Figure 2. ---------------------------------Insert Figure 2 about here ---------------------------------The figure shows a stronger positive relationship between physician customer-centered behaviors and performance ratings for men than for women. We calculated the significance of the simple slopes for interactions (Aiken & West, 1991). The coefficient of the simple slope of quality behaviors on customer satisfaction was significantly more positive for male physicians (b = .32, p< .01) than female physicians (b = -.01, n.s.). Similarly, the coefficient of the simple slope for accessibility behaviors was significantly more positive for male physicians (b = .13, n.s.) than female physicians (b = -.17, n.s.). Although neither simple slope for accessibility behaviors is significantly different from zero, the simple slopes are significantly different from each other (p< .05). By looking
-22-
at the plots, one can see that the interaction is a cross-over, which shows that the direction of the relationship is the opposite for members of high- versus low-status demographic groups. The two-way race X objective performance interactions as a set explained a significant amount of incremental variance in the dependent variable (R2 = .08, p< .05) providing preliminary support for hypothesis 1. Inspection of the individual regression weights showed that the physician productivity X race and physician quality X race interactions were significant (p< .05). The forms of the interactions are shown graphically in Figure 2. Simple slope analysis reveals that the coefficient of the simple slope of quality on customer satisfaction is significantly more positive for white physicians (b = .29, p < .01) than for nonwhite physicians (b = -.14, n.s.). Likewise, the simple slope of productivity behaviors on customer satisfaction is significantly more positive for white physicians (b = .15, n.s.) than nonwhite physicians (b = -.32, p< .01). Overall, we find support for four of the relationships predicted in hypothesis 1. Discussion Our first study explores whether customers, who in this case were patients of an HMO, express their race- and gender-based biases in customer satisfaction judgments. We found that objectivelymeasured behaviors were only positively related to customer satisfaction for physicians who were white or male. We also found that one type of customer-centered behavior was significantly negatively related to customer satisfaction for women and nonwhite physicians. This second finding was an even stronger result than we anticipated because logically we might expect the relationship between customer-benefiting behaviors and customer satisfaction to be weaker, but still positive, for women and nonwhites compared to men and whites. The observed pattern of relationships indicates that biases against nonwhite and female employees may creep into satisfaction judgments. However, we must also consider this study's
-23-
shortcomings. First, the opposite relationship signs of panel age and age of physician with the DV between the correlation table and regression results suggest that these variables may have somehow influenced our results by suppressing variance in the DV that was irrelevant to prediction of the DV (Tabachnick & Fidell, 2003). However, the correlations are not statistically significant and our results are substantively unchanged regardless of whether these variables are included in the model, so we believe our results are not due to a suppression effect caused by those variables. A more serious limitation is that our Medcorp study only included a small percentage of nonwhites. Moreover, many of the nonwhites were Asians rather than African-Americans. Biases against African-Americans are more negative than those associated with Asians (Song, 2004) and so a study that included AfricanAmericans might be better able to detect the influence of such biases on customer satisfaction judgments. We were also not able to control for employee accents or differences in employee language and communication styles, or whether customers felt certain employees had nonwhite sounding names. It is possible that the biases we observed were due to perhaps some contextual variable such as employee language skill and were not due to pre-existing customer prejudices. Finally, we did not measure whether customer raters had preexisting bias against women and minorities. That is, we had no assessment of the raters' stereotypes or racial/gender biases as potential causes of their ratings. We designed our second study to address the limitations of our Medcorp study. The occupation we chose for our laboratory study was service employees working in a university bookstore and our raters were college students. We also used an experimental design to control for extraneous variables that might have influenced the results of our Medcorp study. BOOKCORP STUDY In our second study (Bookcorp), student raters were asked to observe a video of an employeecustomer interaction in a university bookstore, to evaluate the employee's behavior, and to provide
-24-
satisfaction judgments of the store environment. Our Bookcorp study is different from our Medcorp study in a variety of ways. First, we controlled for the job-related behavior (with a pre-scripted interaction) of the employee and varied only whether the behavior was performed by a male versus a female or a white versus an African-American employee. This aspect of the Bookcorp study's design allowed us to reduce variability in employee behavior thereby providing a better test of whether the same behavior would nevertheless produce different customer satisfaction judgments depending on the employee's gender or race. Second, we assessed how student participants, who were asked to assume the role of customers, not only evaluated the employee (as in our Medcorp study) but also evaluated the organizational context (i.e., the bookstore) in which the employee-customer interaction took place. Third, we assessed each participant's implicit bias towards women and nonwhites to see if these nonconscious attitudes might partly explain gender or racial bias in the ratings. Sample Eighty-six university students from a major northwestern U.S. public university watched two videos of a university bookstore employee interacting with a customer and were asked to evaluate the employee and the bookstore. The bookstore in the video clips was from a large East Coast U.S. university and it is highly unlikely that any of the participants had visited the bookstore previously. The "employees" and "customers" were hired professional actors and the scripted interaction was filmed before the bookstore opened in the morning (although our raters taking the customer perspective were not aware of this). We assigned 33 participants to view the white male, 21 to view the white female employee, and 34 to view the black male employee. Overall, a substantial percentage of our participants were nonwhite (43 percent) or female (38 percent). By having a heterogeneous sample of raters, the subjects are representative of the population of people using the book store. Design
-25-
Our design was a mixed-factorial design with one between subjects factor (Employee Demographic Characteristics) and one within subjects factor (Employee-Customer Interaction). We treated Employee Demographic Characteristics as a between-subjects factor to reduce participant awareness that they were participating in a race or gender related study. We presented all participants with two videos depicting different employee-customer interactions. One video involved the employee ringing up a book and telling the customer that the book's price in the computer was higher than its price on the shelf. The other video involved the same employee trying to help a customer find a book the customer wanted. Each video was about one minute in length. Each participant saw both videos of the same employee. We randomly assigned the ordering of the videos within each condition and found no evidence that the ordering of the videos influenced customer ratings. The customer and employee interaction was from a written script to ensure that their behavior was equivalent across conditions. The store background was also held constant across conditions since the camera was in the same location when filming the different employees for each interaction. Dependent Variables Customer satisfaction with the employee. Our measure of customer satisfaction with the employee asked raters to identify on a 7-point Likert scale (1 = very poor; 7 = excellent) how satisfied they were with (1) speed of service; (2) quality of service; (3) availability of staff for assistance, and; (4) employee responsiveness to customers' issues and concerns. This measure was adapted from an existing customer satisfaction survey we obtained from a large organization (see Appendix 1 for customer satisfaction items used across the three studies). Coefficient alpha for this measure was .74. Customer satisfaction with the context. Our measure of customer satisfaction with the context asked raters to identify on a 7-point Likert scale (1 = very poor, less than expected, definitely would not, or strongly disagree; 7 = excellent, better than expected, definitely would, or strongly agree) how
-26-
satisfied they were with (1) the bookstore's appearance; (2) the degree to which the bookstore was conducive to learning; (3) whether the bookstore had up to date equipment; (4) the degree to which the bookstore's facilities were visually appealing; (5) whether the bookstore's appearance was in keeping with the type of services provided; (6) the bookstore relative to their expectations, and; (7) their likelihood of recommending the bookstore to others. This measure was also adapted from an existing customer satisfaction survey we obtained from a large organization. Coefficient alpha for this measure was .76. Predictor Variables Condition. We had two conditions—one for race and one for sex. The sex condition included participants who viewed the white male employee or white female employee (1 = participants viewed two videos of a white woman employee, 0 = participants viewed two videos of a white man employee). The race condition included participants who viewed either the white male employee or the nonwhite male employee (1 = participants viewed two videos of a nonwhite man employee; 0 = participants viewed two videos of a white man employee). Participants completed survey questions only after watching both videos. Implicit bias. To measure raters' racial and gender prejudices, we administered two Implicit Association Tests (IATs). The IATs were constructed to capture each participant's level of nonconscious bias against nonwhites and women (Greenwald, Nosek and Banaji, 2003). We should note that IAT measurement shortcomings (Blanton & Jaccard, 2006), such as no absolute zero point or equal intervals, makes the interpretation of a respondent's score somewhat unclear. However, we chose to use the IAT as opposed to other types of bias measures (e.g., modern racism scale) because it is more difficult for participants to hide prejudices on the IAT than on explicit measures (Nosek, 2005). The gender IAT was administered after the participants saw the videos and made their customer
-27-
satisfaction judgments, but the race IAT was administered between the videos and the ratings. Prior research indicates no evidence of order effects for the IAT and dependent variables—probably because subjects still respond in socially desirable ways on the explicit measures (Greenwald, Poehlman, Uhlmann & Banaji, 2009). Importantly, implicit attitudes appear to be better predictors of behavior than their explicit counterparts, especially when social sensitivity concerns are high (Greenwald et al., 2009). For instance, implicit (but not explicit) attitudes about African Americans have been shown to predict desire to work with an African American partner on an intellectual task (Ashburn-Nardo, Knowles, & Monteith, 2003), and nonverbal actions (eye contact and other "friendly" behaviors) toward African American interaction partners (McConnell & Leibold, 2001). Though the correlation between implicit and explicit attitudes varies across domains (Nosek, 2005), the predictive validity of each suggests that they represent independent processes that explain unique variance in behavioral outcomes (see Greenwald et al, 2009, for a meta-analysis of the predictive validity of the IAT). Control Variables We controlled for rater race, gender and age to account for rater demographics which might plausibly influence reactions to employee demographics. Results We regressed the customer judgments of the employee and the organizational context on our controls, predictors, and interaction to determine the degree to which customer judgments of the employee and the organizational context reflected race and gender bias. Tables 3 and 4 present the regression models we used to test hypotheses 1, 2 and 3. -------------------------------------Insert Tables 3 and 4 about here --------------------------------------
-28-
Hypothesis 1 states that the relationship between employee objective performance and customer satisfaction judgments will be less positive for employees belonging to low-status demographic groups compared to employees belonging to high-status demographic groups. Because objective performance was held constant due to the employee script, the main effects of employee race and gender on customer satisfaction with the employee were used to test this first hypothesis. Model 2 in Table 3 shows that raters taking the customer perspective were significantly less satisfied with women employees than their equally well-performing white male counterparts (?R2 = .06; b = -.28; p< .05). However, we did not find evidence of bias in customer satisfaction judgments of the nonwhite employee (b = -.02; n.s.). Overall, we found some support for hypothesis 1. Hypothesis 2 states that people would report lower customer satisfaction judgments of the store environment when an employee in that environment belongs to a low (i.e., female, African-American) rather than high (i.e., male, white) status demographic group. Model 2 of Table 4 shows there is a significant main effect of race and gender on judgments of the store environment. Indeed, Model 2 of Table 4 shows a main effect of the female condition (?R2 = .17; b = -.45; p< .01) and the nonwhite condition (?R2 = .15; b = -.44; p< .001), suggesting that raters' biases influence judgments of the organizational context. We found strong support for hypothesis 2. Hypothesis 3a suggests that people would report even lower customer satisfaction judgments of the employee when observing an employee belonging to a low-status demographic group when the customer has negative implicit attitudes toward that group. Model 3 of Table 3 shows that the coefficient for the interaction term IAT Score X Nonwhite Condition is significant and in the expected direction for customer satisfaction with the employee (?R2 = .08; b = -.28; p< .01). To gain more insight into the nature of this effect we plotted the interaction and analyzed the simple slopes (see Figure 3). Individuals with high levels of implicit bias (+1 s.d.) were significantly more likely to report
-29-
lower satisfaction with the nonwhite male's performance than with the white male's (p< .01). However, the coefficient for the interaction term IAT Score X Gender Condition was not significant for customer satisfaction with the employee. As for hypothesis 3b, Model 3 of Table 4 shows that the coefficient for the interaction term IAT Score X Race is significant and in the expected direction for customer satisfaction with the context (?R2 = .04; b = -.18; p< .05). We plotted the interactions and conducted a simple slope analysis (see Figure 3). Customer IAT score (+1 s.d.) was positively related to customer satisfaction with the context when customers were observing a white male employee (b = .33; p< .01) but was negatively related to customer satisfaction with the context when customers were observing a nonwhite male employee (b = -.21; p< .05). The coefficient for the interaction term IAT Score X Gender Condition was significant and in the expected direction for customer satisfaction with the context (?R2 = .04; b = -.23; p< .05). Customer IAT score (+1 s.d.) was positively related to customer satisfaction with the context when customers were observing a white male employee (b = .23; p< .05) but was negatively related to customer satisfaction with the context when customers were observing a white female employee (b = -.21; p< .05). Overall, we found three significant coefficients supporting hypothesis 3. These results and plots suggest that judgments of employees and judgments of the organizational context are vulnerable to non-conscious biases. Discussion We found that raters taking the customer perspective rated the employee and the organizational context as being worse when observing the performance of a low-status employee and this was especially true if the person held implicit biases about that low-status group. The results regarding hypothesis 1 were not as strong as we anticipated suggesting that it may be possible to minimize biases by changing the setting in which the rating takes place. Specifically, the laboratory context was less
-30-
anonymous than a typical customer satisfaction questionnaire setting, which may have weakened the influence of bias on customer satisfaction judgments of employees. Although we told the participants that their responses were anonymous, they may have felt scrutinized because they provided their judgments when an experimenter was present, and wrote their names on a separate sign-in sheet to receive participation credit for a class. Additionally, because the IAT for race was administered prior to the ratings it may have alerted participants that they were in a race and gender study, which also may have weakened the Bookcorp study results. We observed, though, that the effects of bias on judgments of the organizational context were still quite strong, which makes sense to us because participants may have been unaware of and therefore unable to suppress their biases that spilled over onto their judgments of the organizational context. As summarized by hypothesis 4, we expected that the relationship between an organizational unit's objective performance and customer satisfaction would be less positive for organizational units that employ higher percentages of employees belonging to low-status demographic groups (i.e., women and racial minorities) compared to units that employ higher percentages of employees belonging to high-status demographic groups (i.e., men and whites). By returning to the field to test this hypothesis, we complete the full-cycle of research model and assess the generalizability of our theory to a different organization. GOLFCORP STUDY Our sample was drawn from a large country club organization, hereafter referred to as Golfcorp. Golfcorp has 66 country clubs across the United States and roughly 70,000 customermembers and it employs approximately 8,000 people. Our sample consisted of all 66 Golfcorp country clubs. Within our sample, 31.4% of employees were women, 18.1% were Latino, 6.7% were AfricanAmerican, and 1.7% were Asian-American or Native-American.
-31-
Measures Golfcorp routinely collects customer satisfaction ratings as well as objective indicators of facility performance that were assumed to have a direct, positive impact on customers' service experiences. The dependent variable in our study was customer satisfaction with the facility. The independent variables were each club's employee demographics (percentage nonwhite and female employees) and two types of objective club performance. Customer satisfaction with facility. Like many organizations, Golfcorp measures customer satisfaction with a quarterly survey, which is mailed to a percentage of each facility's customers. An average of 63.8 customers rated each facility. The average response rate per facility was 27.3 percent (an average of 234 surveyed customers per facility). The marketing company hired to do the customer survey randomly sampled each facility's customers each quarter until they got either 20 respondents or three percent of the total customer base (whichever was larger). The items used for this measure reflect a focus on the facility context (quality of its clubhouse and golf course) and overall ratings of the facility, similar to what was used in the Bookcorp study. Customers rated each of the items on a 5point Likert scale (1 = very poor; 5 = very good) "How would you rate the following aspects of your club?" (1) Maintenance of grounds/Appearance of clubhouse; (2) Locker rooms and restrooms; (3) Quality of greens; (4) Condition of course; (5) Pace of play; (6) Condition of practice facilities; (7) Ability to obtain desired tee times; (8) Club meets expectations (1 = less than expected, 5 = better than expected), and; (9) Likelihood of recommending club to others (1 = definitely will not, 5 = definitely will). Coefficient alpha for this performance measure is .81. In our analyses, we lagged this measure six months after the independent variables to more conclusively show that employee demographics and objectively-measured performance influence customer ratings, rather than the other way around.
-32-
Percent Nonwhite. Golfcorp identified the percentage of white and nonwhite employees in each facility. Across the 66 clubs in the sample, 26.5 percent of employees were nonwhite. According to the U.S. Census Bureau's 2000 census, the percentage of ethnic minority employees in our sample is consistent with the national average of 28 percent of the U.S. population. Percent Female. We also obtained this variable from Golfcorp records. Thirty-one percent of employees in our sample were women, which Golfcorp leaders believe to be consistent with the country club industry average. However, the percentage of women in our sample is lower than the percentage of women across all industries, which is 46%, according to the Bureau of Labor Statistics at the U.S. Department of Labor. Objective facility characteristics. As in our first two studies, we wanted to clearly identify the portion of variance attributable to customer bias versus the portion attributable to better facility performance. We therefore used two attributes as our indicators of objective facility performance— facility productivity and facility attribute quality. Both dimensions reflect facility characteristics that benefit customers. Facilities with more productive employees create more value—both for Golfcorp and for customers. Indeed, Golfcorp executives told us that facilities with higher productivity values charge lower dues to members, are more profitable and are simply better-run facilities. Higher attribute quality benefits customers by allowing customers to enjoy newer and better facilities. Facility managers are shown how they compare to other facilities in terms of quality and productivity. Employee compensation is tied to the productivity measure, but not the quality measure. Employees in facilities that are above average in productivity are given a bonus, while employees at below average clubs are not given a bonus. Facility Productivity. Facility productivity was calculated by Golfcorp's central accounting office for the calendar year ending six months before the dependent variable was collected. This
-33-
variable is each club's annual profits divided by the average number of employees working for the club in that year. The number of employees at each club is centrally controlled such that clubs with more members are allotted proportionally more employees by the central office. Therefore facility productivity is determined by the employees' effectiveness at creating value. Facility quality attributes. Over time, the condition of the golf course and the clubhouse deteriorates and needs to be rebuilt or refurbished. Golfcorp assesses the quality of the course and clubhouse of each club to ensure that customers are receiving a high standard of service. Golf courses are assigned a quality rating based on the percentage of the course that is infiltrated by crab grass and dead spots (1 = more than 40% of course is crab grass or dead spots; 5 = less than 5% of course is crab grass or dead spots). Likewise, clubhouses are assigned a quality rating, indicating how recently they were built or refurbished (1=built or refurbished more than 15 years ago; 5 = built or refurbished 2 years ago or less). The overall facility attribute quality variable is the composite of the clubhouse quality rating and the golf course quality rating. The two component variables approached normality and were added together. The resulting variable was each facility's overall golf course and clubhouse quality. Control Variables Customers may be more satisfied with larger facilities because they offer more amenities, and with facilities that employ a large percentage of young or temporary employees who may be more energetic. Therefore we controlled for facility size, average employee age, and percent temporary employees. Customers may be more satisfied if they have been a member for a long time and have not quit, if they are men or if they are older. Therefore we also controlled for average customer tenure (months), percent male customers, and average customer age. Results
-34-
Table 5 reports the means, standard deviations, and correlation coefficients between the dependent, independent, and control variables. We used hierarchical moderated regression models to examine the hypothesized interaction effects (Aiken & West, 1991). We centered all variables involved in the interaction terms to minimize multicollinearity between the interaction terms and their individual components (Aiken & West, 1991). Table 6 presents the results of our hierarchical moderated regression analysis. We entered all of the control variables in Model 1. In Model 2 we entered the interactions involving sex and the two dimensions of objective performance, In Model 3 we entered the interactions involving race and the two dimensions of objective performance, and in Model 4 we entered all four interactions. ----------------------------Insert Tables 5 and 6 about here ---------------------------Hypothesis 4 stated that the association between an organizational unit's objective performance and customer satisfaction would be attenuated for organizational units that employ higher percentages of employees belonging to low-status demographic groups compared to units that employ higher percentages of employees belonging to high-status demographic groups. The two-way gender X objective performance interactions as a set explained a significant amount of incremental variance in the dependent variable (?R2 = .04, p< .05) providing some further support for this hypothesis. Inspection of the individual regression weights in the full model showed that the facility quality X gender and facility productivity X gender interactions were significant (p< .05). We probed the pattern of the interactions by examining the simple slopes of the objective performance measures for facilities with high and low percentages of female employees (Aiken & West, 1991). Like the Medcorp study plots, we found a stronger positive relationship between objective performance and customer satisfaction for facilities that have a low percentage of female employees
-35-
than for facilities that have a high percentage of female employees. Facility quality was significantly more positively associated with customer satisfaction for facilities with a low percentage of female employees than for facilities with a high percentage of female employees. Likewise, facility productivity was more positively associated with customer satisfaction for facilities with a low percentage of nonwhite employees than for facilities with a high percentage of nonwhite employees. Again, even though the simple slopes were not significantly different from zero, the significant regression coefficient in the full model demonstrates that they are significantly different from each other. By looking at the plots, one can see that the interaction is a cross-over, which shows that the direction of the relationship is the opposite, and statistically significant, for members of high- versus low-status demographic groups. These results support hypothesis 4 for both of the gender X objective performance relationships. The two-way race X objective performance interactions as a set explained a significant amount of incremental variance in the dependent variable (?R2 = .15, p< .05) providing preliminary support for our hypothesis. Inspection of the individual regression weights from the full model showed that the facility attribute quality X race and facility productivity X race interactions were significant (p< .05). Simple slope analysis revealed that the association between facility quality and customer satisfaction was significantly more positive for facilities employing a low percentage of nonwhites than for facilities employing a high percentage of nonwhites. Likewise, the association between facility productivity and customer satisfaction was more positive for facilities employing a low percentage of nonwhites (b= .32, p< .01) than for facilities employing a high percentage of nonwhites (b=-.23, p< .05). These results support hypothesis 4 for both of the race X objective performance relationships. Discussion
-36-
In this Golfcorp study, we found that objectively measured behaviors that benefit customers were positively related to customer satisfaction, but only for facilities with a low percentage of nonwhite and female employees. These results parallel the results of our Medcorp study. GENERAL DISCUSSION We set out to determine if and how customer satisfaction judgments are influenced by racial and gender biases. Across three samples we found converging evidence that customer satisfaction judgments are susceptible to systematic and predictable racial and gender biases. Customers tended to be less satisfied with the services provided by women and nonwhite employees than by men and white employees, even when controlling for objective indicators of performance (Medcorp study). We also found that these biases operated on judgments about the store context when a third-party evaluator observes an employee interacting with a customer (Bookcorp study), particularly if the observer holds negative implicit attitudes toward women or minorities. Finally, we found that evaluations of the organizational unit as a whole are negatively associated with the presence of nonwhite and female employees (Golfcorp study). It is worth noting that we found evidence for the operation of racial biases regardless of whether the nonwhite employees were predominantly Asian (Medcorp study), AfricanAmerican (Bookcorp study), or Latino (Golfcorp study). The consistency of our results across three different samples and methodologies testifies to the robustness and generality of the systematic biases we observed and to the internal validity of our theoretical model. The pattern of these biases may help explain the persistence of demographic inequalities in organizations. To cite just a few examples, women and nonwhites make 25 percent less than their male and white counterparts in equivalent jobs (U.S. Census Bureau, 2006), women and nonwhites are twice as likely as white men to be unemployed and underemployed (NIOSH, 2002), and women and ethnic minorities are not well represented among the ranks of highly paid managers and
-37-
professionals in U.S. corporations and in prestigious occupations like law and medicine (e.g., Baldi & McBrier, 1997; Eagly & Karau, 2002; Wilson, Sakura-Lemessy & West, 1999). Economists have often been perturbed by these demographic inequalities because orthodox economic theory would predict that some of these inequalities should be erased when employers compete for women and nonwhite applicants whose wages are 25 percent less costly than their white and male counterparts (The Economist, 2008). Our results suggest that the evidence available to organizational decision makers is that customers tend to be less satisfied with nonwhite and female employees; however, without the benefit of the research conducted here, decision makers would not be able to determine that these lower satisfaction ratings are attributable at least in part to customer biases. Therefore, decision makers might rationally choose, based on their limited information, to preferentially select white and male employees, as their data is likely to bear out that such personnel are better performers. Evidence from highly publicized lawsuits (e.g., Shoney's Restaurant; Abercrombie & Fitch) suggests that managers are keenly aware of the fact that some customers may prefer white and male employees. Executives in these cases admitted to deliberately favoring white employees in hiring and promotion decisions to enhance customer satisfaction and organizational profitability (Brief et al., 2000). When employees or units are viewed as performing less well by customers as a result of demographic characteristics like race or gender, one could argue that these employees and units should receive fewer rewards, bonuses, and promotional opportunities. But it is important to note that the customer judgments in our studies were inconsistent with other objective indicators of performance. In other words, nonwhites and women may have behaved the same way as their white and male counterparts in trying to provide satisfactory customer service, but if compensation and other organizational benefits are linked to customer satisfaction ratings then they may not be rewarded
-38-
similarly for identical behavior, which would violate the principle of equity which most business enterprises claim to follow. Our results suggest that if customer evaluations become widely and uncritically used to determine pay and promotion opportunities, the job outcomes of women and ethnic minorities could be adversely impacted. For example, consider what would happen if managers notice which employees routinely receive the highest customer satisfaction scores and use this information to make promotion decisions or if university administrators relied heavily on student ratings of teaching effectiveness to influence promotion and tenure decisions. At higher levels of the organization, executives may examine which of their organizational units achieve the highest levels of customer satisfaction and promote those managers further up the organizational hierarchy. Our data suggest that one possible consequence of these decisions is that whites and men will be much more likely than their nonwhite or women counterparts to receive favorable customer satisfaction judgments, which should accelerate their journey up the organizational ladder. Likewise, managers who purposely stock their organizational units with whites and men are likely to have more career success than managers who do not. Our finding that customer biases can spill over onto the surrounding organizational context contributes to the literature on contamination and signaling and also illustrates the subtle operation of racial and gender bias. Marketing researchers have shown how observable customer characteristics like physical attractiveness can influence other customers' desire to purchase a product (Argo et al., 2009). To our knowledge, ours is the first study to show how an observable characteristic of an employee like race or gender can influence customer perceptions of an organization's contextual quality. This finding may provide insight into phenomenon known as "white flight" where whites move out of a neighborhood once a critical mass of nonwhites move in (Gladwell, 2000; Kruse, 2005).
-39-
In an organizational setting, a similar phenomenon may operate in which customers may link conscious or unconscious negative attitudes they have towards members of status groups to employees who belong to these groups. In turn, these associations "contaminate" customer perceptions of the organizational context. This process of contextual spillover may partly explain why managers have often been reluctant to pursue diversity despite the known performance advantages of having a diverse workforce (Joshi, Liao & Jackson, 2006; O'Reilly, Williams, & Barsade, 1997). It may be that managers are aware that diversity has hidden costs because it increases the possibility of "customer flight" to an organization that has fewer employees who belong to low-status demographic groups. Future studies using customer satisfaction as an outcome variable should take into account the demographic make-up of the employees as well as objectively measured organizational characteristics. Our findings cast doubt on the ability of customers to accurately perceive the quality of customer service organizations. The theoretical attributes we suggested as possible causes of customer bias were meant to explain why these biases occur, but like any useful theory they also suggest potential remedies. Based on our theorizing, racial and gender biases in customer satisfaction judgments may be reduced by (1) making customer ratings less anonymous; (2) changing the standards customer use to make their ratings so they emphasize behavior rather than subjective judgments; and (3) introducing customer de-biasing education or training in the evaluation process. Identifiability (i.e. non-anonymity) may provide an especially strong de-biasing effect in service contexts characterized by repeated interaction (e.g. the doctor/patient relationship) because customers would want to make sure their ratings do not jeopardize future received service quality. Behaviorally anchored rating scales may be especially helpful for removing bias in customer expectations (i.e. individuals often hold nonwhites and women to higher standards; Biernat & Kobrynowicz, 1997). Organizations could consider only accepting customer satisfaction surveys from customers who have completed de-biasing training,
-40-
although such a move would be logistically and perhaps scientifically problematic (due to potential selection bias). In addition to addressing factors that cause bias in customer ratings, organizations can take practical steps to minimize the potential adverse impact of customer biases on nonwhite and female employees' careers. For example, organizations might consider only using satisfaction surveys from frequent customers to ensure that raters have sufficient exposure to targets. Organizations could also ask for customer feedback during the customer service encounter so that customers will be most likely to be paying attention and less likely to rely on information subject to memory bias when judging their customer experience. Organizations might also want to let customers know that the data will be used to make career progression decisions so that customers are more motivated to judge responsibly. Organizations could also insert "bias sensitive" questions in customer satisfaction judgments so that responses from potentially biased customers can be given less weight or discarded. However, we urge caution if organizations choose to remove outlier ratings because this tactic may actually decrease rating accuracy if most judges are aware of their biases and therefore tend to over-correct their judgments (Zitzewitz, 2006). Alternatively, organizations may be able to statistically correct for bias when calculating customer satisfaction judgments. Finally, using different survey formats for customer rating scales might also be helpful for circumventing rater biases, like forced-choice, behaviorally anchored rating scales (citing specific valued behaviors), and un-weighted and weighted checklists. Organizations should consider the tradeoffs between these formats and choose the one that is most likely to reduce the effects of customer judgment biases on the career prospects of those who are most vulnerable to being targets of such bias. Limitations
-41-
We believe our findings provide strong, consistent support for our theoretical predictions. However, like all research, ours has its share of limitations. First, role congruence may have been an issue in our Medcorp study as patients may expect their physicians to be white and male and thereby judge non-white or women doctors more harshly. However, there is not much evidence to suggest that they do expect their doctor to be a white male. Indeed, patients prefer their doctor to look like them (women prefer women doctors, and nonwhite patients prefer nonwhite doctors; Chen, Fryer, Phillips, Wilson & Pathman, 2005; Cooper-Patrick, Gallo, & Gonzales, 1999). Likewise, role congruence should not have been an issue in our Bookcorp study, so our consistent results across these two samples provide us with some confidence that role-congruence alone is not responsible for our findings. Still, future research would be well-served to test our hypotheses in a variety of samples. Another potential issue is rater-target congruence. Although including patient gender and race in the analysis slightly strengthened our Medcorp study results, we did not control for these variables in that study because of multi-collinearity (i.e., the respective correlations between customer gender/race and physician gender/race were greater than .90). In our other two studies we found no evidence that rater-target congruence influenced customer satisfaction judgments. Indeed, in post-hoc analyses, the interactions of customer race X employee race and customer gender X employee gender were not significant. To maintain compliance with the rule of thumb that there should be five cases per variable (Tabachnick & Fidell, 2003), we do not report these post-hoc analyses. Another potential limitation is that unobserved variables may be responsible for our results. In our Golfcorp study, employee demographics may have masked the facility's strategy. Facility executives pursuing a low-cost strategy may have hired a large number of women and nonwhite employees, whereas those pursuing a premium-pricing strategy may have hired a large number of whites and men. We ran some post-hoc analyses to test this idea and found no supporting evidence.
-42-
Specifically, we ran the interactions of several variables measuring club strategy (i.e. services offered, turnover rate) by objective performance and our four race and gender interactions remained significant, but none of the additional interactions were significant (we do not report these analyses to maintain an adequate case to variable ratio). Relatedly, we also did not test or report whether nonwhite women face a double jeopardy (Berdahl & Moore, 2006) in terms of customer satisfaction to maintain compliance with the five to one rule of thumb (i.e. we were not able to test the three-way interactions involving employee objective performance, employee race and employee gender), but future research should do so. Finally, we did not test whether the IAT influences ratings of the organizational unit because we found that the IAT influences sub-components of the organizational unit ratings— employee and context ratings. We should also mention that our method of testing for possible bias in performance evaluations was a significant improvement over past studies. First, we used an objective performance standard so that we could compare subjective judgments to this standard and therefore determine whether the customer judgments of performance might be influenced by race and gender. Second, our subjective judgments were based on aggregated judgments from a large number of customers rather than relying on the judgments of a single supervisor. This is important because the large number of raters provides a highly reliable subjective performance rating for each individual, context, or organization. The IAT is new and its predictive validity is relatively untested (Blanton & Jaccard, 2006), so another contribution our paper offers is the first demonstration of the IAT's predictive validity in a management journal. Finally, we controlled for several variables that could provide alternative explanations for our results, such as the average employee age and the average customer tenure with the organization.
-43-
Given these methodological strengths of our research, it is unsettling to find that customers may not respond favorably to organizational characteristics designed to benefit them when these organizations have a high percentage of low-status employees. At Golfcorp, employees at clubs with a high percentage of female and nonwhite employees can in fact be economically harmed by customer satisfaction evaluations because clubs who fail to achieve the target level of customer satisfaction (i.e., below the organizational average) do not receive a salary bonus. The practical implications of our results become more apparent when we when examine the effect sizes in our samples. Across our three studies, the racial and gender bias effects on customer satisfaction judgments explained between 15 and 24 percent of the variance in customer satisfaction. Cohen (1988) provides ballpark descriptors of effect sizes based on R-squared values—"large" (R2 = .25), "medium" (R2 = .09), and "small" (R2 = .01). Therefore, the average observed effect size of racial and gender bias across our three samples is between medium and large. Conclusions In these different samples we demonstrated that customer ratings are biased against women and racial minorities. We had two field studies and one laboratory study utilizing a full cycle research strategy. The effects were demonstrated for three different minority groups and three different contexts involving employee-customer contact. In all three settings we controlled for actual objective behavior or performance along with a series of other controls appropriate for that context. The effects are demonstrated for individual targets as well as the context or organization in which the targets work. In the laboratory sample we showed that the biased ratings are exacerbated by implicit racial or gender bias. In short, these are fairly robust findings across jobs, contexts, raters and ratees. If these results are replicated and generalizable they have significant implications for organizational practice. If managers are serious about the fair treatment of their employees and the
-44-
promotion of diversity they will need to treat customer ratings differently. More specifically, the rating process can be changed by increasing information, responsibility or training for raters and by changing how customer ratings are used. In the latter context, organizations can perhaps measure and discount such biases or use statistical procedures to adjust the ratings to remove the bias. Without such actions, given the increasing dependence on customer ratings, we are likely to not only maintain existing levels of inequitable compensation and advancement for women and minorities, we are likely to increase these inequities. This outcome is unacceptable in a society that is committed legally, morally and socially to fair treatment for all in the workplace.
-45-
References Aiken, L. S. & West, S. G. 1991. Multiple regression: Testing and interpreting interactions. Newbury Park, CA: Sage Publications. Anderson, E. W., Fornell, C., & Mazvancheryl, S. K. 2004. Customer satisfaction and shareholder value. Journal of Marketing, 68: 172-185. Anderson, E. W.,& Mittal, V. 2000. Strengthening the satisfaction-profit chain. Journal of Service Research, 3: 107-120. Anderson, N.H. 1981. Foundations of information integration theory. New York. Academic Press. Argo, J. J., Dahl, D. W., & Morales, A. C. 2009, Positive Consumer Contamination: Responses to Attractive Others in a Retail Context. Forthcoming at the Journal of Marketing Research. Ashburn-Nardo, L., Knowles, M. L., & Monteith, M. J. 2003. Black Americans' implicit racial associations and their implications for intergroup judgment. Social Cognition, 21: 61-87. Aquino, K., & Bommer, W. H. 2003. Preferential mistreatment: How victim status moderates the relationship between organizational citizenship behavior and workplace victimization. Organization Science, 4: 274-285. Baldi, S., & McBrier, D. B. 1997. Do the Determinants of Promotion Differ for Blacks and Whites? Evidence from the U.S. Labor Market. Work and Occupations, 24: 478-497. Baltes, B. B., & Parker, C. P. 2000. Reducing the effects of performance expectations on behavioral ratings. Organizational Behavior and Human Decision Processes, 82: 237-267. Berdahl, J. L. & Moore, C. (2006). Workplace harassment: Double-jeopardy for minority women. Journal of Applied Psychology, 91, 426-436. Berger, J., Conner, T. L., & Fisek, M. H. 1974. Expectation States Theory: A Theoretical Research Program. Cambridge, MA: Winthrop. Berger, J., Fisek, M. H., Norman, R. Z., & Zelditch, M., Jr. 1977. Status Characteristics and Social Interaction: An Expectation States Approach. New York: Elsevier. Bertrand, M., & Mullainathan, S. 2004. Are Emily and Greg More Employable than Lakisha and Jamal? A Field Experiment on Labor Market Discrimination. American Economic Review, 94: 9911013. Biernat, M., & Kobrynowicz, D. 1997. Gender and race-based standards of competence: Lower minimum standards but higher ability standards for devalued groups. Journal of Personality and Social Psychology, 72: 544-557.
-46-
Blanton, H. & Jaccard, J. (2006). Arbitrary Metrics in Psychology. American Psychologist, 61: 27-41. Bobo, L. 1998. Race, interests, and beliefs about affirmative action - Unanswered questions and new directions. American Behavioral Scientist, 41: 985-1003. Bracken, D. W., Church, A. H., & Timmreck, C. W. 2001. Handbook of Multisource Feedback: The Comprehensive Resource for Designing and Implementing MSF Processes. New York, NY: JosseyBass. Brief, A. P., Dietz, J., Cohen, R. R., Pugh, S. D., & Vaslow, J. B. 2000. Just doing business: Modern racism and obedience to authority as explanations for employment discrimination. Organizational Behavior and Human Decision Processes, 81, 72-97. Chatman, J. A., & Flynn, F. J. 2005. Full-Cycle Micro-Organizational Behavior Research. Organization Science, 16: 434-447. Chen, F.M., Fryer, G.E., Jr., Phillips, R. L., Jr., Wilson, E., & Pathman, D.E., (2005). Patients' Beliefs About Racism, Preferences for Physician Race, and Satisfaction With Care. Annals of Family Medicine. 3: 138-143. Cialdini, R. B. 1995. A full-cycle approach to social psychology. In G. G. Brannigan & M. R. Merrens (Eds.), The Social Psychologists: Research Adventures: 53-72. New York: McGraw-Hill. Cohen, J. 1988. Statistical power analysis for the behavioral sciences (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum. Cooper-Patrick, L., Gallo, J.J., & Gonzales, J.J.,(1999). Race, gender, and partnership in the patientphysician relationship. Journal of the American Medical Association. 282: 583-589. Crandall, C. S., & Eshelman, A. 2003. A Justification-Suppression model of the expression and experience of prejudice. Psychological Bulletin, 129: 414-446. Davis, W. E., & Davis, D. R. 1999. The University Presidency: Do Evaluations Make a Difference? Journal of Personnel Evaluation in Education, 13: 119-140. Deci, E. L. & Ryan, R. M. 1985. The general causality orientations scale - Self-determination in personality. Journal of Research in Personality, 19:109-134. DeNisi, A. S., Robbins, T., & Cafferty, T.P. 1989. Organization of information used for performance appraisals: Role of diary keeping. Journal of Applied Psychology, 74: 124-129. Devine, P. G., Plant, E. A., Amodio, D. M., Harmon-Jones, E., & Vance, S. L. 2002. The regulation of explicit and implicit race bias: The role of motivations to respond without prejudice. Journal of Personality and Social Psychology, 82: 835-848. Dovidio, J. F., & Gaertner, S. L. 1981. The effects of race, status, and ability on helping-behavior. Social Psychology Quarterly, 44: 192-203.
-47-
Dovidio, J. F., & Gaertner, S. L. 2000. Aversive racism and selection decisions: 1989 and 1999. Psychological Science, 11: 315-319. Dubois, R. W., Alexander, C. M., Wade, S., Mosso, A., Markson, L., Lu, J. D., Nag, S., & Berger, M. L. 2002. Growth in use of lipid lowering therapies: Are we targeting the right patients? American Journal of Managed Care, 8: 81-86. Eagly, A. H., Makhijani, M. G., & Klonsky, B. G. 1992. Gender and the evaluation of leaders: A metaanalysis. Psychological Bulletin, 111: 3-22. Eagly, A. H. & Karau, S. J. 2002. Role congruity theory of prejudice toward female leaders. Psychological Review, 109: 573-598. Feinglass, J., & Salmon, J. W. 1990. Corporatization of medicine: the use of medical management information systems to increase the clinical productivity of physicians. Journal of Health Services, 20: 233-252. Fernandez, J. P. 1981. Racism and sexism in corporate life. Lexington, MA: Heath. Figueroa, E. B., & Woods, R. A. 2007. Industry output and employment projections to 2016. (2007). Monthly Labor Review, 130: 53-85. Folkes, V.S. & Patrick, V.M. 2003. The positivity effect in perceptions of services: Seen one, seen them all? Journal of Consumer Research, 30: 125-137. Frazer, J. G. 1890/1959. The golden bough: A study in magic and religion. New York: Macmitian. (Reprint of 1922 abridged edition, edited by T. H., Gaster; original work published 1890). Gaertner, S. L., & Dovidio, J. F. 1986. The aversive form of racism. In J. F., Dovidio & S. L., Gaertner (Eds.), Prejudice, Discrimination, and Racism. Orlando, FL: Academic Press. Gerstein, H. C., Yusuf, S., Mann, J. F. E., Hoogwerf, B., Zinman, B., Held, C., Fisher, M., Wolffenbuttel, B., Bosch, J., Richardson, L., Pogue, J., & Halle, J. P. 2000. Effects of ramipril on cardiovascular and microvascular outcomes in people with diabetes mellitus: Results of the HOPE study and MICRO-HOPE substudy. Lancet, 355: 205-212. Giannopoulos, G., Conway, M., & Mendelson, M. 2005. The gender of status: The laypersons' perception of status is gender-typed. Sex Roles, 53: 795-806. Gilovich, T., Griffin, D. W., & Kahneman, D. 2002. Heuristics of judgment: The psychology of intuitive judgment. Cambridge, U. K.: Cambridge University Press. Gladwell, M. 2000. The Tipping Point: How Little Things Can Make a Big Difference. Little Brown Publishing.
-48-
Goldin, C., & Rouse, C. 2000. Orchestrating Impartiality: The Impact of 'Blind' Auditions on Female Musicians. American Economic Review, 90: 715-741. Greenhaus, J. H., Parasuraman, S., & Wormley, W. M. 1990. Effects of race on organizational experiences, job performance evaluations, and career outcomes. Academy of Management Journal, 33: 64-86. Greenwald, A. G., & Banaji, M. R. 1995. Implicit social cognition: Attitudes, self-esteem, and stereotypes. Psychological Review, 102: 4-27. Greenwald, A. G., Nosek, B. A., & Banaji, M. R. 2003. Understanding and using the implicit association test: I. An improved scoring algorithm. Journal of Personality and Social Psychology, 85: 197-216. Greenwald, A. G., Poehlman, T. A., Uhlmann, E., & Banaji, M. R. (2009). Understanding and using the Implicit Association Test: III. Meta-analysis of predictive validity. Journal of Personality and Social Psychology. In press. Gruca, T. S., & Rego, L. L. 2005. Customer satisfaction, cash flow, and shareholder value. Journal of Marketing, 69: 115-130. Gupta, S., & Zeithaml, V., 2006. Customer metrics and their impact on financial performance. Marketing Science, 25: 718-739. Haas, J. S., Cook, E. F., Puopolo, A. L., Burstin, H. R., Cleary, P. D., Brennan, T. A. 2000. Is the professional satisfaction of general internists associated with patient satisfaction? Journal of General Internal Medicine,15: 122-128. Hagan, C. M., Konopaske, R., Bernardin, H. J., & Tyler, C. L. 2006. Predicting assessment center performance with 360-degree, top-down, and customer-based competency assessments Human Resource Management, 45: 357-390. Hsaio, W. C., Braun, P., Becker, E. R., & Thomas, S. R. 1988a. The resource-based relative value system. Journal of the American Medical Association, 258: 799-802. Hsaio, W. C., Braun, R., Dunn, D., & Becker, E. R. 1988b. Resource-based relative values: An overview. Journal of the American Medical Association, 260: 2347-2353. Isles, C. G. 2002. Patients with acute coronary syndrome should start a statin while still in hospital. Heart, 88: 5-6. Ittner, C. D., & Larcker, D. F. 1998. Are non-financial measures leading indicators of financial performance? An analysis of customer satisfaction. Journal of Accounting Research, 36: 1-35. Joshi, A., Liao, H., & Jackson, S. E. 2006. Cross-level effects of workplace diversity on sales performance and pay. Academy of Management Journal, 49: 459-481.
-49-
Judge, T. A., & Ferris, G. R. 1993. Social-context of performance evaluation decisions. Academy of Management Journal, 36:80-105. Kane, J. S., Bernardin, H. J., Villanova, P. & Peyrefitte, J. 1995. Stability of Rater Leniency - 3 Studies. Academy of Management Journal, 38: 1036-1051. Keller, K.L. 2003. Strategic brand management. Upper Saddle River, N.J. Prentice Hall. Kleiner, K. D., Akers, R., Burke, B. L., & Werner, E. J. 2002. Parent and physician attitudes regarding electronic communication in pediatric practices. Pediatrics, 109: 740-744. Kruse, K. M. 2005. White Flight: Atlanta and the Making of Modern Conservatism. Princeton, NJ: Princeton University Press. Laine, C., & Davidoff, F. 1996. Patient-centered medicine: A professional evolution. Journal of the American Medical Association, 275: 152-156. Landy, F. J., Shankster, L. J., & Kohler, S. S. 1994. Personnel-selection and placement. Annual Review of Psychology, 45: 261-296. LaRosa, J. C., He, J., & Vupputuri, S. 1999. Effect of statins on risk of coronary disease: a metaanalysis of randomized controlled trials. Journal of the American Medical Association, 282: 23402346. Latham, G. P., & Wexley, K. N. 1977. Behavioral observation scales for performance-appraisal purposes. Personnel Psychology, 30: 255-268. Lerner, J.S. & Tetlock, P.E. (1999). Accounting for the effects of accountability. Psychological Bulletin, 125: 255-275 Mauss, M. 1902/1972. A general theory of magic. New York: W. W. Norton. (R., Brain, Trans.; Original work published 1902). McConahay, J. B. 1983. Modern racism and modern discrimination - the effects of race, racialattitudes, and context on simulated hiring decisions. Personality and Social Psychology Bulletin, 9: 551-558. McConnell, A. R., & Leibold, J. M. 2001. Relations among the Implicit Association Test, discriminatory behavior, and explicit measures of racial attitudes. Journal of Experimental Social Psychology, 37: 435-442. Morales, A. C., & Fitzsimons, G. J. 2007. Product contagion: Changing consumer evaluations through physical contact with "disgusting" products. Journal of Marketing Research, 44:272-283 Moshavi, D. 2004. He said, she said: Gender bias and customer satisfaction with phone-based service encounters. Journal of Applied Social Psychology, 34, 162-176.
-50-
Murphy, K. R. 1991. Criterion issues in performance appraisal research: Behavioral accuracy versus classification accuracy. Organizational Behavior and Human Decision Processes, 50: 45-50 National Institute for Occupational Safety and Health. 2002. The changing organization of work and the safety and health of working people (NIOSH Publication No. 2002-116). Washington, DC: Author. Negro, G., Goodman, S., & Rao, H. 2008. Red Rattinghood: Brokerage, Closure, and Negative Affiliations in Hollywood Blacklisting. Working paper. Nosek, B. 2005. Moderators of the relationship between implicit and explicit evaluation. Journal of Experimental Psychology: General, 134: 565-584. O'Leary, V. E., & Ickovics, J. R. 1992. Cracking the glass ceiling: Overcoming isolation and alienation. In U. Sekaran & F. Leong (Eds.), Woman power: Managing in times of demographic turbulence: 7-30. Thousand Oaks, CA: Sage. O'Reilly, C. A., Williams, K. Y., & Barsade, S. G. 1998. Group demography and innovation: Does diversity help?" In D. Gruenfeld, B. Mannix & M. Neale, (Eds.) Research on Managing on Groups and Teams, pp. 183-207. Stamford, CT: JAI Press, Inc. Pellegrin, K. L., Stuart, G. W., Maree, B., Frueh, B. C., & Ballenger, J. C. 2001. A brief scale for assessing patients' satisfaction with care in outpatient psychiatric services. Psychiatric Services, 52: 816-819. Pulakos, E. D., White, L. A., Oppler, S. H., & Borman, W. C. 1989. Examination of race and sex effects on performance ratings. Journal of Applied Psychology, 74: 770-780. Richman, W. L., Kiesler, S., Weisband, S., & Drasgow, F. 1999. A meta-analytic study of social desirability distortion in computer-administered questionnaires, traditional questionnaires, and interviews. Journal of Applied Psychology, 84: 754-775. Ridgeway, C. L. 1991. The Social Construction of Status Value: Gender and Other Nominal Characteristics. Social Forces, 70: 367-386. Roch, S. G., & O'Sullivan, B. J. 2003. Frame of reference rater training issues: Recall, time and behavior observation training. International Journal of Training and Development, 7: 93-107. Rotundo, M., & Sackett, P. S. 1999. Effect of rater race on conclusions regarding differential prediction in cognitive ability tests. Journal of Applied Psychology, 84: 815-822. Rozin, P., Millman, T., & Nemeroff, C. 1986. Operation of the laws of sympathetic magic in disgust and other domains. Journal of Personality and Social Psychology, 50: 703-712. Rynes, S. L., Bretz, R. D., & Gerhart, B. 1991. The importance of recruitment in job choice - A different way of looking. Personnel Psychology, 44: 487-521.
-51-
Rynes, S. L., Heneman, H. G. III, & Schwab, D. P. 1980. Individual reactions to organizational recruiting: A review. Personnel Psychology, 33: 529-542. Rynes, S. L., & Miller, H. E. 1983. Recruiter and job influences on candidates for employment. Journal of Applied Psychology, 68: 147-154. Salam, S., Cox, J. F., & Sims, H. P., Jr. 1997. In the eye of the beholder: How leadership relates to 360-degree performance ratings. Group and Organization Management, 2: 185-209. Schneider, B., Ehrhart, M. G., Mayer, D. M., Saltz, J. L., & Niles-Jolly, K. 2005. Understanding organization-customer links in service settings. Academy of Management Journal, 48: 10171032. Sidanius, J., & Pratto, F. 1999. Social Dominance: An Intergroup Theory of Social Hierarchy and Oppression. Cambridge, U. K.: Cambridge University Press. Simonet, D. 2005. Patient satisfaction under managed care. International Journal of Health Care Quality Assurance, 18: 424-440. Simonin, B.L. & Ruth, J.A. 1998. Is a company known by the company it keeps? Assessing the spillover effects of brand alliances on consumer brand attitudes. Journal of Marketing Research, 35: 30-42. Sixma H. J., Spreeuwenberg, P. M., & van der Pasch, M. A. 1998. Patient satisfaction with the general practitioner: a two-level analysis. Medical Care, 36: 212-229. Song, M. 2004. Who's at the bottom? Examining claims about racial hierarchy. Ethnic and Racial Studies, 27: 859-877. Spence, M. 1973. Job market signalling. Quarterly Journal of Economics, 85: 355-374. Stewart, M., Brown, J. B., Donner, A., McWhinney, I. R., Oates, J., Weston, W. W., & Jordan, J. 2000. The impact of patient-centered care on outcomes. Journal of Family Practice, 49: 796-804. Tabachnick, B. G., & Fidell, L. S. 2003. Using Multivariate Statistics (Fifth ed.). Boston: Allyn and Bacon. Taylor, H. 2002. Patient/physician online communication: Many patients want it, would pay for it, and it would influence their choice of doctors and health plans. Harris Interactive Health Care News, 2: 13. The Economist. 2008. Black America: Nearer to Overcoming. May 8: 34-36 Westphal, J. D., & Stern, I. 2007. Flattery will get you everywhere (especially if you are a male caucasian): Ingratiation, boardroom behavior, and demographic minority status affect the likelihood of gaining additional board appointments at U.S. companies. Academy Management Journal, 50: 1-22.
-52-
Wherry, R. J. Sr., & Bartlett, C. J. 1982. The control of bias in ratings: A theory of rating. Personnel Psychology, 35: 521-551. Wilkinson, T. J., & Fontaine, S. 2002. Patients' global ratings of student competence. Unreliable contamination or gold standard?. Medical Education, 36: 1117-1121. Willerson, J. T., & Cohn, J. N. (Eds.). 2000. Cardiovascular medicine. Philadelphia: Churchill Livingstone. Wilson, G., Sakura-Lemessy, I., & West, J. P. 1999. Reaching the Top: Racial Differences in Mobility Paths to Upper-Tier Occupations. Work and Occupations, 26: 165-186. Woehr, D., & Roch, S. G. 1996. Context effects in performance evaluation: The impact of ratee gender and performance level. Organizational Behavior and Human Decision Processes, 66: 31-41. Wyer, R.S. Jr. & Schrull, T.K. 1989. Person memory and judgment. Psychological Review, 96: 5883. Yarkin, K. L., Town, J. P., & Wallston, B. S. 1982. Blacks and women must try harder: Stimulus person's race and sex attributions of causality. Personality and Social Psychology Bulletin, 8: 21-24. Yusuf, S., Sleight, P., Pogue, J., Bosch, J., Davies, R., & Dagenais, G. 2000. Effects of an angiotensinconverting-enzyme inhibitor, ramipril, on cardiovascular events in high-risk patients. New England Journal of Medicine, 342: 145-153. Zitzewitz, E. (2006). Nationalism in Winter Sports Judging and Its Lessons for Organizational Decision Making. Journal of Economics & Management Strategy, 15: 67-99
-53-
TABLE 1. Medcorp Sample Means, Standard Deviations, and Correlations between Predictor, Control and Dependent Variables4
M s.d. 1 0.51 0.11 0.66 0.47 -.30 0.80 0.20 -.07 1749.77 550.63 -.10 45.84 4.89 .07 1.04 0.12 .13 14.81 8.51 .20 50.34 6.58 .09 .12 0.32 -.15 .38 0.49 -.06 23.00 1.97 .05 -0.01 1.55 .11 0.16 0.15 .13 2 3 4 5 6 7 8 9 10 11 12
1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13.
Patient Satisfaction Practice Busyness Full time equivalent Number of patients in panel Panel age Chronic sickness of panel Tenure with Medcorp (years) Age (years) Nonwhite Female Productivity Quality Accessibility to Patients
.11 .26 .59 -.07 .05 -.12 -.15 -.14 .14 -.09 .16 .02 .01 -.06 -.63 .12 .22 .03 .07 -.11 -.18
-.03 -.13 .55 .08 .33 .21 .29 -.04 -.14 -.44 -.21 .30 -.06 .08 .21 -.23 .05
-.20 -.05 -.05 .04 .22 .05 -.06
.69 -.12 -.03 -.25 -.31 .12 -.25 -.01 -.04 -.15 .14 .11 .02 .04 -.03 -.04 -.16 -.11 .21 .05 .23
-
4
N = 113; All correlations larger than .15 are significant at p<.05
-54-
TABLE 2. Medcorp Study Analysis Examining Moderating Effects of Physician Race, Gender and Objective Performance on Patient Satisfaction with Physician5
Model 1 Controls Practice Busyness Full time equivalent Number of patients in panel Panel age Chronic sickness of panel Tenure with Medcorp (years) Age (years) Productivity Quality Accessibility Female Nonwhite Interactions Productivity X Tenure Quality X Tenure Accessibility X Tenure Productivity X Female Quality X Female Accessibility X Female Productivity X Nonwhite Quality X Nonwhite Accessibility X Nonwhite Adjusted R2 R2 ?R2 from Model 1 -.23** -.17 -.04 -.23* .23* .34** -.13 .09 .17* .15 -.22* -.12
Model 2 -.18* -.23 -.03 -.27* .25* .35* -.12 .09 .20* .13 -.26* -.09
Model 3 .26** -.20 .01 -.20 .20 .38** -.17 .12 .16 .07 -.17* -.18*
Model 4 -.22* -.24* .01 -.24* .22* .39** -.16 .14 .17* .05 -.20* -.17*
.06 .13 .12 .04
.09 .13 .01 .00 -.18** -.16*
.01 .14 -.16
.03 .14 .06 -.18** -.16* -.21** -.16* -.10 .25 .40 .15**
-.18* -.18* -.13 .17 .25 .21 .32 .07** .23 .33 .08**
5
* p<.05 ** p<.01 N = 113. All participants are medical doctors. The sample consisted of 100 whites, 10 Asian or Pacific Islanders, 2 blacks, and one Native American.
-55-
TABLE 3. Bookcorp Study Effect of Employee Race and Gender on Customer Satisfaction with the Employee6
Customer Satisfaction with the Employee White Male/White Female White Male/Nonwhite Male Condition Condition Model 1 Model 2 Model 3 Model 1 Model 2 Model 3 -.25* -.26 -.27 -.23* -.21 -.20 -.15 -.23 -.21 -.15 -.15 -.16 -.09 -.05 -.05 -.22* -.23* -.21 .05 .15 -.13 -.15 -.14 -.19 -.28* -.26* -.06 -.02 -.01 -.28** .14 .22 .08**
Nonwhite Female Age Implicit Association Test (IAT) score Woman condition (1=white woman employee, 0=white man employee) IAT Score X Woman condition Nonwhite condition (1=nonwhite man employee, 0=white man employee) IAT Score X Nonwhite condition Adjusted R2 R2 ?R2 from previous model
.03 .10
.08 .16 .06*
.08 .16 .00
.08 .14
.08 .14 .00
6
* p<.05 ** p<.01; N = 67 in the white man/nonwhite man condition and N = 54 in the white man/white woman
condition.
-56-
TABLE 4. Bookcorp Study Effect of Employee Race and Gender on Customer Satisfaction with the Organizational Context7
Customer Satisfaction with the Context White Male/White Female White Male/Nonwhite Male Condition Condition Model 1 Model 2 Model 3 Model 1 Model 2 Model 3 -.03 -.03 -.07 -.16 -.08 -.08 .09 -.03 .09 .16 .01 .01 -.26 -.19 -.14 -.17 -.04 -.04 -.12 .03 -.04 .13 .09 .02 -.45** -.38** -.23* -.44*** -.46*** -.18* .20 .26 .04*
Nonwhite Female Age Implicit Association Test (IAT) score Woman condition (1=white woman employee, 0=white man employee) IAT Score X Woman condition Nonwhite condition (1=nonwhite man employee, 0=white man employee) IAT Score X Nonwhite condition Adjusted R2 R2 ?R2 from previous model
.00 .07
.15 .24 .17**
.17 .29 .04**
.00 .07
.18 .22 .15***
7
* p<.05 ** p<.01 ***p< .001; N = 67 in the white man/nonwhite man condition and N = 54 in the white man/white
woman condition.
-57-
TABLE 5.8 Golfcorp Means, Standard Deviations and Correlations between Dependent, Independent and Control Variables. M
1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. Customer Satisfaction with Facility Size (Number of employees) Average Employee Age Percent Temporary Employees Average Customer Tenure (months) Percent Male Customers Average Customer Age (years) Percent Nonwhite Employees Percent Female Employees Facility Quality Facility Productivity (Profit per employee) 3.90 129.05 39.20 .04 60.66 .56 54.23 .26 .31 3.32 14614.26
s.d.
.22 157.51 4.04 .08 22.10 .07 8.40 .18 .12 1.04 7703.06
1
.16 -.18 .15 -.15 .00 -.14 .08 -.07 .08 .13
2
.09 -.07 .14 -.03 .17 -.12 -.21 .16 .14
3
4
5
6
7
8
9
10
.16 .15 -.13 .34 .11 .07 -.14 .12
.01 .06 .01 .20 .05 .16 .16
-.03 .42 -.28 -.07 -.04 .12
-.16 .15 .20 .16 .13
.08 -.06 -.03 .16
-.08 .10 .14
-.08 .09
.15
8
N = 66; all correlations greater than .21 are significant at p < .05
-58-
TABLE 6. Golfcorp Study Regression Results Examining the Interactive Influence of Percentage of Nonwhite Employees, Percentage of Female Employees and Objective Measures of Facility Performance on Customer Satisfaction with Facility9
Customer Satisfaction with Facility Model 1 Model 2 Model 3 Model 4 Controls Size Average Employee Age Percent Temporary Employees Average Customer Tenure Percent Male Customers Average Customer Age Facility Quality Facility Productivity Percent Nonwhite Employees Percent Female Employees Interactions Percent Female X Quality Percent Female X Productivity Nonwhite X Quality Nonwhite X Productivity Adjusted R2 R2 ?R2 from Model 1 .00 .13 .21 -.21 .18 -.12 -.06 -.09 -.04 .17 -.05 .04 .27* -.24* .16 -.06 -.08 -.13 -.01 .01 -.15 -.07 .27* -.24* .16 -.06 -.08 -.13 -.01 .01 -.15 -.07 .30** -.19 .18 -.06 -.11 -.10 -.03 -.05 -.19 -.14
-.25* -.22 -.25* -.34** .03 .17 .04*
-.23* -.31* Percent -.25* Percent -.49** .12 .28 .15** .20 .37 .24***
9
* p<.05 ** p<.01 ***p< .001; N = 66 country clubs.
-59-
FIGURE 1. Conceptual Model of How Bias Influences Customer Satisfaction Ratings10
Attributes of the Rating
Anonymity No Evaluation Standard Lack of Training
Race/Sex of Employee
Customer Racial/Gender Bias
H1 Service Provider Performance H2
H3a H3b
Rating of Employee
Rating of Context H4 Rating of Organizational Unit (Employee and Context)
Organizational Unit Performance
10
Dotted line indicates untested relationships. We expect main effects, but our contribution lies with tests of the interactions.
-60-
FIGURE 2. Medcorp Sample Interactive Effects of Physician Objective Performance and Physician Demographics on Patient Satisfaction with the Physician
High White Physician (**) High Male Physician (**)
Patient Satisfaction with Physician
Low Low High Nonwhite Physician (n.s.)
Patient Satisfaction with Physician
Female Physician (n.s.) Low Low High
Physician Quality
Physician Quality
White Physician (**)
High Male Physician (n.s.)
High
Patient Satisfaction with Physician
Low ** p < .01 * p < .05 n.s. p > .10 Low High
Female Physician (n.s.)
Patient Satisfaction with Physician
Low Low High Nonwhite Physician (**)
Physician Accessibility
Physician Productivity
-61-
FIGURE 3. Bookcorp Sample Interactive Effects of Employee Demographics and Customer Implicit Association Test Score on Customer Satisfaction with the Employee and with the Store Context
High
Satisfaction with Employee
White Male Employee (n.s.)
Low Low High
Nonwhite Male Employee (**)
Implicit Association Test Score
White Male Employee (**)
High
High
White Male Employee (*)
Satisfaction with Context
Satisfaction with Context
White Female Employee (*) Low Low High
** p < .01 * p < .05 Low n.s. p > .10 ** p < .01 * p < .05 n.s. p > .10
Nonwhite Male Employee (*) Low High
Implicit Association Test Score
Implicit Association Test Score
-62-
Appendix 1: Customer Satisfaction Items Used Across the Three Studies Medcorp Study Customer Satisfaction with the Employee: How would you rate the following attributes of your provider (1 = very poor; 5 = excellent): 1. Attention provider paid 2. Thoroughness and competence of provider 3. Ability to ask questions of this provider Bookcorp Study Customer Satisfaction with the Employee: How would you rate the following (1 = very poor; 7 = excellent): 1. Speed of service 2. Quality of service 3. Availability of staff for assistance 4. Employee responsiveness to customers' issues and concerns Bookcorp Study Customer Satisfaction with the Context: How would you rate the following aspects of the bookstore: 1. Appearance of bookstore (1 = very poor; 7 = excellent) 2. Environment of the bookstore was conducive to learning/reading (1 = strongly disagree; 7 = strongly agree) 3. The bookstore has up to date equipment 4. This bookstore's physical facilities are visually appealing 5. The appearance of this bookstore is in keeping with the type of services provided 6. Bookstore meets expectations (1 = less than expected, 7 = better than expected) 7. Likelihood of recommending bookstore to others (1 = definitely would not, 7 = definitely would) Golfcorp Study Customer Satisfaction with the Facility: How would you rate the following aspects of your club (1 = very poor; 5 = very good): 1. Maintenance of grounds/Appearance of clubhouse 2. Locker rooms and Rest Rooms 3. Quality of greens 4. Condition of course 5. Pace of play 6. Condition of practice facilities 7. Ability to obtain desired tee times 8. Club meets expectations (1 = less than expected, 5 = better than expected) 9. Likelihood of recommending club to others (1 = definitely will not, 5 = definitely will)
-63-
David R. Hekman is an Assistant Professor of Management at the Lubar School of Business, University of Wisconsin-Milwaukee. He earned his Ph.D. in Management from the University of Washington. He is interested in improving organizational health by minimizing organizational problems such as weak employee attachment, contagious harmful behaviors, and persistent workplace inequality. Karl Aquino is the Richard Poon Professor of Organizations and Society at the Sauder School of Business, University of British Columbia. His research interests are moral behavior, workplace victimization, revenge and forgiveness, and status and power dynamics in organizations. He received his Ph.D. in Organizational Behavior from Northwestern University. Bradley P. Owens is a Post Doctoral Fellow at the Center for Positive Organizational Scholarship at the University of Michigan. He earned his Ph.D. in Management from the University of Washington. His primary research interests include humility, leadership, race and gender issues, and team dynamics. Terence R. Mitchell received his undergraduate degree from Duke in 1964, and a Masters and Ph.D. from the University of Illinois in organizational psychology. He has been at the University of Washington since 1969 and in 1987 he was appointed the Carlson Professor of Management. He has published journal articles and book chapters on the topics of motivation, leadership, turnover and decision making. Pauline Schilpzand is an Assistant Professor of Character Development at the Army Center of Excellence for the Professional Military Ethic at West Point. She earned her PhD in Management from the Warrington College of Business Administration at the University of Florida. Her primary research interests include behavioral ethics, leadership, and the selection process. Keith Leavitt is an Assistant Professor of Character Development at the Army Center of Excellence for the Professional Military Ethic at West Point. He earned his PhD in Management from the Foster School of Business at the University of Washington. His primary research interests include behavioral ethics, implicit social cognition, research methods, and organizational citizenship.
-1-
doc_998563967.docx