An Empirical Evaluation of Information Features and the Willingness to be Profiled Online for Personalization Neveen I. Farag University of Michigan Business School Ann Arbor, MI 48105 nfarag@umich.edu M. S. Krishnan Michael R. and Mary Kay Hallman e-Business Fellow Associate Professor and Area Chair of Computer Information Systems University of Michigan Business School Ann Arbor, MI 48105 mskrish@umich.edu March 3, 2003 Please do not cite, quote, or distribute without permission of the authors. Financial support for this study was provided in part by the Michael R. and Mary Kay Hallman fellowship, the Marcy Maguire fellowship, the Information Systems Executive Forum at the University of Michigan Business School, and the Summer Research Opportunity Program at the University of Michigan. An Empirical Evaluation of Information Features and the Willingness to be Profiled Online for Personalization Abstract As customer information becomes a critical success factor in the global networked economy, privacy is of strategic importance. Firms are able to use information to improve customer service and offer personalized features. By understanding the factors affecting consumer willingness to share information, managers can tailor their firm’s e-business offerings to encourage consumers to partake in personalization activities. This study hypothesized that firms can address the risks of online profiling, and therefore encourage consumers to share their information, by making consumers aware of firm procedures. Consumer concern over the information collected and how it will be used (INFO_USE), should affect consumers’ willingness to be profiled online. Customers will allow their activities to be monitored and used to create consumer profiles for business use if they are privy to the profile information collected by the firm. Knowledge of firm procedures should increase consumer perceived control, thereby decreasing consumer perceived risk, and eventually increasing participation in customer relationship activities, including personalized service and advertising. Based on a survey of over 500 online consumers, our hypothesis was supported through our finding that customers who rate knowledge of firm procedures as a “high” concern, are also less willing to allow personal information to be tracked and stored in a profile used for personalized service and advertising. The results of this study should help managers tailor their firm’s online service offerings, and consequently, maximize useful and beneficial information collection across various consumer-segments. 1 Introduction In today’s networked global economy, the ability to collect, analyze, and respond to user information is of growing importance. To survive, companies depend on vast quantities of information to build rapport with existing customers and attract new business (Culnan and Armstrong 1999). As the ease and availability of e-business reduces face-to-face interactions, firms must use consumer information to attempt to offer personalized service that will increase value, and consequently, consumer loyalty. As Peter Weill and Michael R. Vitale state in their book, Place to Space: Migrating to eBusiness Models, “Information technology (IT) infrastructure and the information it contains, particularly customer information, will be a critical success factor for all ebusiness initiatives, thus raising the stakes for the management of the firm's IT investments and assets.” However, implicit in the collection of consumer information lies a concern for consumer privacy; information privacy is one of the most important issues facing management practice (Safire, 2002; Mason, 1986). If managers are not careful, their firms may be the victims of consumer backlash for overstepping the bounds of expected information practices. In order to encourage consumers to participate in online information collection, a firm may therefore implement policies and features that addresses a user’s information privacy concerns. The importance of IT related privacy issues is also seen through the growing global and governmental trend of increasing corporate privacy regulation (Dresner 1996, Franklin 1996, Hendricks 1998). In addition, a pending Congressional privacy bill would require US companies to give customers digital access to personal information that firms possess (Thibodeau, 2002). This bill is an example of a growing trend towards 2 attempting to inform the consumer. Informing the consumer of firm procedures differs from exercising fair information practices. A central element of fair information practices is the ability of individuals to remove their names from mailing lists (Culnan and Armstrong 1999). A central element of informing consumers of firm procedures, however, is the ability of consumers to actually see what of their personal information a company possesses. Thus, informing the consumer deals more directly with allowing consumers’ access to the type of information a firm possesses and how it plans to use it. The objective of this paper is to address the question: How can a firm increase consumer willingness to be profiled online for personalized offerings? More specifically, two research questions guide this study: 1) How does perceived control, achieved through knowledge of information and firm procedures, affect consumer willingness to be profiled online for personalized offerings? and, 2) How does consumer willingness to be profiled online differ across personalized service versus personalized advertising? The major contribution of this research provides empirical evidence that companies can garner increased consumer participation in online information sharing activities by providing consumers with knowledge of and access to information and usage procedures. Through increased consumer participation, firms should then be able to utilize the customer information for competitive advantage. In the next section, we review and discuss prior literature. In section three, we discuss the theoretical model and hypotheses of this paper. In section four, we explain the data and measurement. In section five, we present the analysis and results. In section six, we discuss the results and their managerial implications. We then conclude the paper with directions for future research. 3 2) Prior Literature Public opinion surveys show that many citizens are quite concerned about threats to their information privacy (Equifax 1996; Westin 1997; Harris and Westin, 1998). Several of the expressed privacy concerns centered on how firms collect and use personal data. Previous work has found that how a firm handles buyer-collected personal information offline affects the relationship between the firm and its consumers (Hoffman, Novak, & Peralta, 1999). Privacy issues online are arguably not any different than privacy issues offline. However, due largely to the increased amount of information online, and the fear of identity theft (FTC Report, 1999), consumer comfort with online profiling remains low. In fact, a Business Week (2000) survey showed that 63% of those surveyed were not comfortable with anonymous online profiling; 89% were not comfortable with identified online profiling. While privacy is an issue, there is also a case for how firms can use online information to benefit consumers; we next address this stream of research. Research has shown that companies may be able to use profile information to increase consumer value and move from mass merchandising to personalized service (Pine 1993; Farag and VanAlstyne 2000; Farag and Krishnan, 2002). Blattberg and Deighton (1991) have shown that when detailed consumer-information is collected, firms are able to engage in relationship marketing and target offers more accurately based on a customer’s specific interests. Organizations can therefore gain a competitive advantage from collecting and using transaction data effectively (Farag and Krishnan, 2002; Glazer 4 1991). However, the potential downside of information collection, if implemented improperly, is that it may actually raise a user’s privacy concerns rather than create value (Bloom et. al 1994). Hence, the issue of online information collection has received the attention of researchers, practitioners, and policy makers alike. Several theories on the link between firm information practices and individual behavior are emerging. For example, Milburg, Smith and Burke (2000) developed a conceptual model examining corporate management of personal data, regulatory approaches to information privacy, and consumer reactions across cultures. Culnan (1993) tested an exploratory model that explains consumer attitude toward some direct marketing practices offline. With this model, Culnan showed that public attitudes towards privacy are likely to vary based on dimensions of control. Smith (1994) developed a model to explain corporate approaches to information privacy policymaking. Hoffman et al. (1999) suggested that two larger dimensions govern privacy concerns: environmental control and secondary use of information control. Laufer et al. (1976) illustrated that perceived control over various uses of information results in less consumer concern over privacy invasions. In addition, Stone et. al (1983) showed the more a user values privacy, or rather, the more concerned about privacy, the less control the consumer perceives to have over personal information. Stone and Stone (1990) developed a model for information flows and physical/social structure in work environments based on expectancy theory. The above stream of research is grounded in the basic definition of privacy found in psychology literature. Privacy is defined as the ability of the individual to control the 5 terms under which personal information is acquired and used (Westin 1967). “Information privacy,” then, refers to “the ability of the individual to personally control information about one’s self” (Stone, et al., 1983). Hence, it may be interpreted from this definition that one way to decrease the level of perceived privacy risk for the online consumer is to increase his or her level of control over personal information. Consistent with this line of work, Culnan and Armstrong (1999) identify control as one of the two main information privacy concerns that occur with firm handling of personal information. The other concern being secondary use, where consumer information provided for one purpose may be reused for another purpose without consent (Culnan, 1993; Smith et al., 1996; Godwin, 1991; Foxman and Kilcoyne, 1993). In this paper, we relate the perceived risk of consumers, measured through perceived importance of control, to their willingness to be profiled online. Knowledge has been shown to be a determinant of perceived control (Wortman 1975, Azjen and Driver 1991, Armitage and Conner 1999). In addition, knowledge as a mechanism for control has been examined in prior organizational literature (e.g. Sohn 1994). Previous research has shown that Internet usage is constrained for some adults due to the perceived need for more knowledge and understanding of the medium (Klobas and Clyde 2000). In this paper, we extend the idea of knowledge as a control mechanism towards improving consumer comfort. While consumer comfort with a company is important in any setting, the threat of consumer discomfort is heightened in the Internet setting, where consumers can easily “flame” a company directly by electronic mail or in Internet newsgroups (Bies and Tripp 1996). 6 In this paper, we attempt to test consumer decisions to share information within the framework of expectancy theory. Control is defined as a set of mechanisms designed to motivate individuals to work in such a way that desired objectives are achieved (Kirsch 1996, Jaworksi 1988). In the case of online customer relationship management, a firm’s desired outcome is for consumers to participate in online profiling. A company’s challenge, then, is to mitigate the perceived risks of being profiled online, such that they can provide consumers the desired outcome of personalized service and personalized advertising. A classic conceptual model used to understand how individuals make decisions regarding various behavioral alternatives in the face of perceived risks is grounded in the expectancy theory model of motivation (Vroom, 1964). The central theme of expectancy theory is that outcomes drive individual behavior. Thus, individuals will chose between alternatives by evaluating the outcomes, or the anticipated future consequences of the alternatives. While previous research has suggested that issues of informational control are essential in creating a favorable consumer predisposition toward contributing information to online firms, (Stewart and Segars, 2002), none have examined the issue of control using the expectancy theory framework. The contributions of this paper to the literature are as follows: 1) This paper adds to our understanding of knowledge and perceived control in the online context; 2) This paper examines importance of firm features towards decreasing consumer perceived risk, and increasing consumer perceived control, across two contexts, advertising and service; and, 3) While prior studies have examined likelihood of consumers to partake in online personalization from purely the consumer characteristic standpoint (e.g. Chellappa and Sin, 2002), this paper studies past consumer experiences in the context of privacy 7 concerns, along with consumer characteristics, that affect consumer perceived control; and, therefore willingness to be profiled online. Culnan and Armstrong (1999) have previously examined consumer willingness to share information, however they did so in an offline setting, and focused on the use of procedural fairness, rather than information access and usage awareness. Similarly, Phelps, Nowak, and Ferrell (2000) examine the relationship between consumer purchase decisions and the amount of information control given to consumers in the offline setting. Other studies have examined the likelihood of consumers to partake in online personalization services from purely the consumercharacteristics standpoint. For example, Chellappa and Sin (2002) examine consumer attributes, such as privacy concern and personalization value, and how such attributes affect consumer likelihood of using personalization services. While we also include consumer attributes, the main focus of our study is firm features that increase consumer perceived control online, decrease consumer perceived risk, and thereby encourage consumers to allow online profiling towards personalization. Thus, while control has also been looked at as it relates to Internet privacy concerns (Dinev and Hart, 2002), the association of consumer information features and consumer willingness to be profiled online is yet to be examined. We, therefore, fill this gap, while also examining the effect of such firm features across the contexts of personalized service and personalized advertising. 3) Theoretical Model and Hypotheses Consumer willingness to share information online involves the evaluation of the outcomes of online profiling. Consumers are, therefore, faced with two alternatives, to 8 partake in online profiling, or to decline. A classic conceptual model used to understand how individuals evaluate such alternatives, and their potential outcomes, is the expectancy theory of motivation (Vroom, 1964). Our model is based on the central premise of expectancy theory of how individual behavior is shaped. Expectancy theory states that three elements guide an individual’s evaluation process: 1) Expectancy, or the belief that one’s actions will be followed by a particular outcome; 2) Instrumentality, or the expectation that other outcomes will lead to the expected outcome; and 3) Valence, or the attractiveness of the outcome to the individual. Figure 1 illustrates Expectancy Theory. Perceived control over performance affects expectancy, or the perceived probability that an action will be followed by a given outcome. For expectancy to be high, individuals must believe that they have some degree of control over the expected outcome. When individuals perceive that the outcome is beyond their ability to influence, motivation to participate is low (Vroom 1964). Motivation Force Expectancy Force Directing Specific Actions Perceived Probability Action leads to Outcome = * Instrumentality Valence Perceived Probability Outcomes will lead to Expected Outcome Value of Expected Outcome to Individual * Figure 1: Expectancy Theory Expectancy theory was originally described by Vroom as follows: n Fi = ∑ ( EijV j I j ) j =1 Where: Eij= The strength of the expectancy ( 0 ≤ Eij ≤ 1 ) that act i will be followed by outcome j. Vj=The valence of outcome j Ij= The cognized instrumentality of outcome j This framework has been applied in different settings; while the elements of the theory are widely accepted, researchers in multiple disciplines have also highlighted the 9 challenges of empirical validation. For example, in accounting, the theory has been applied to audit staff performance; however, only weak support was found for the model’s predictive power (Ferris, 1977). Thus, we do not explicitly test expectancy theory, due to the significant problems of conceptualization and methodology associated with specification of each of the elements of the theory, as several researchers have pointed out (Connolly, 1976). Rather, we attempt to validate the effect of knowledge and perceived control on online profiling, by focusing on elements of expectancy theory that consumers can partake in to mitigate the risks associated with being profiled online. In our setting, the outcome is measured in a single dimension related to the experience of the customers after they have shared their personal information with the firm. The outcomes the consumers may experience in return for the action of taking part in online profiling range from useful personalization to information misuse, resulting in email spam, or even worse, identity theft. Thus, the two extremes of the outcome are useful personalization, or information misuse. The force, then, on the action of whether the consumer should partake in online profiling or not is measured as willingness to partake in online profiling. Useful Personalization Misuse of Information The main independent variable we examine in our study, features a firm can implement towards increased perceived control, is associated with Expectancy, the likelihood that an action will be followed by the anticipated outcome. Thus, we are examining factors that increase the likelihood for consumers that their information will be used to produce an outcome of useful personalization. Note that the expectancy theory 10 inherently captures the tradeoff by individuals who surrender a measure of privacy in exchange for some economic or social benefit of personalization. This tradeoff has been directly studied offline as the “privacy calculus,” which measures the usage of personal information versus the potential negative consequences of disseminating personal information (Laufer and Wolfe, 1977; Milne and Gordon, 1993; Stone and Stone, 1990). Our study examines this specific trade-off, which consumers make in the online setting, and what firms can do to decrease consumer perceived risk. Next we discuss the hypotheses that link perceived risk factors related to information features, knowledge about firm privacy policy, consumer privacy concern, and past privacy experience. Individual control affects expectancy, or the probability that the action of sharing information will be followed by a given outcome (Vroom, 1964); the greater such probability, the greater the force to partake in the action. Thus, we believe that knowledge and perceived control online decrease consumer perceived risk of online profiling, and therefore affect consumer willingness to partake in online profiling. Hence, our hypotheses are aligned with the theory of the expectancy model. Hypotheses Information Features Prior studies have identified that consumers are less likely to perceive a risk, and therefore more willing to share information when consumers perceive the ability to control future use of their information (Culnan and Armstrong, 1999; Bies, 1993; and Stone and Stone, 1990). In addition, Expectancy theory postulates that control decreases the perceived risk of partaking in an action by increasing the probability that a given 11 action will be followed by a given outcome. This increased likelihood then leads to an increased force to of action towards online profiling. These findings may indicate that the first step towards encouraging consumers to partake in online profiling is to increase consumer perceived control. Firms may attempt to increase perceived control by allowing consumers to understand the type of information a firm is collecting, and how that information is being used. In this study, we examine “information features,” or features firms implement to inform the consumer of firm procedures. Such features may therefore increase consumer willingness to be profiled online. Firm implemented features that allow consumers to view and control the use of their personal information should effect the expectancy beliefs of consumers, or rather that their action of being profiled online will lead to the desired outcome of useful personalization. Thus, we hypothesize: consumers who rate importance of information features higher will be less willing to partake in online personalization. Hypothesis 1: The more important a consumer deems information-features, the less willing they are to partake in online profiling. Privacy Policy One way firms attempt to address consumer perceived risks of online profiling is by posting their privacy policy online. Privacy policies address the expectancy portion of expectancy theory, or the expectation that an act will be followed by a given outcome. The degree to which firm practices are formalized in written policies has an impact on individuals' perceptions (Vroom, 1964). In addition, announcing a firm's privacy policy has been shown to increase consumer perceived trust by allowing consumers to make informed decisions about disclosing their personal information (Culnan, 1999; Hansen, 12 2000). By posting a privacy policy, firms are essentially telling consumers what outcome to expect when the consumer partakes in the action of online profiling. Thus, by increasing the probability of a positive outcome for the consumer, firms may use privacy policies to increase consumer motivation to participate in online profiling. Consumers that place a high importance on firm privacy policies, likely perceive a lower probability of a desirable outcome, and therefore want written assurance. We then hypothesize that the more importance a consumer places on online privacy policies, the less the consumer will be willing to partake in online profiling. Hypothesis 2: The more importance a consumer places on a privacy policy, the less likely that the consumer will partake in online profiling. Consumer Privacy Concern Consumer concerns are affecting Internet commerce. A 1997 study revealed that purchases via the Internet would receive a $6 billion boost by the year 2000 if consumers believed their privacy wasn’t at stake during such transactions (Greene, 1997). From a theoretical standpoint, personal values, such as privacy concerns, affect the Valence portion of expectancy theory, or the attractiveness of the outcome to the individual. Consumers with a higher level of privacy concern will likely perceive personalization offerings to be of less value than consumers with a lower level of privacy concern. The more a user views online privacy as a concern, the less motivation they will have to partake in the action of online profiling. Prior research has shown that users that express concern over their own privacy perceive little control over the use of their personal information (Stone et. al 1983) and are likely less willing to share such information. We test this finding in our own setting, and specifically in accordance with expectancy 13 theory. Thus, we hypothesize that a consumer’s increased concern for privacy is associated with a lower desire to share their personal information, and therefore a decreased force towards online profiling. Thus, we expect that greater privacy concern is associated with less willingness to be profiled online. Hypothesis 3: The greater a consumer’s general privacy concern level, the less the consumer is willing to be profiled online. Previous Privacy Invasion Personal experiences guide behavior in activities that can be subjectively deemed as privacy-related (Bates 1964). In addition, personal experiences cause a change in privacy concern over an individual’s lifetime (Louis Harris and Associates, Inc. 1991). Previous experiences are accounted for in the Expectancy Model through Instrumentality, the expectation that other outcomes will lead to the expected outcome. Consumers who have previously had their privacy invaded may not believe that sharing information online will lead to the expected outcome of useful personalization. This decreased likelihood of the expected outcome may result in a decreased motivation to partake in the action of online profiling. The role of past experience has been previously analyzed offline in two different formats, with mixed results. In one instance, consumers were asked if they had “experienced a previous invasion of privacy” (Culnan, 1993). In a second instance, consumers were asked if they had previously dealt with the firm. In the former, prior privacy invasion experience was not shown to have a clear association with attitudes toward secondary information use offline, however previous privacy invasion experiences could affect an individual’s concern for privacy (Culnan, 1993). In the latter, 14 prior firm experience was the distinguishing factor from those willing to store information in a customer profile and those who were unwilling to do so offline (Culnan and Armstrong, 1999). In our research, we look at the former, prior privacy invasion experience, where previous results have not been conclusive. Previous research did not find a significant association between previous privacy invasion experience and attitude toward secondary information use (Culnan and Armstrong, 1999). However, we are attempting to assess the effect of previous privacy invasion experience in a different context, the online context, rather than the direct mail context. Therefore, we expect previous online privacy invasion experience to have a significant effect on willingness to partake in online personalization. Thus, we hypothesize that previous privacy invasions decrease consumer willingness to be profiled online. Hypothesis 4: Consumers who have previously experienced an online privacy invasion are less willing to partake in online profiling. Personalized Service and Personalized Advertising Previous research has shown that firms can improve the perceived value of services offered by mitigating a customer’s perceived risk (Heskett et al. 1990). As discussed earlier, expectancy theory stipulates that the perceived benefit and the perceived risk of an outcome effect consumer Valence perceptions, or the attractiveness of the outcome to the individual. The perceived benefit of an outcome, such as useful personalization, can motivate consumers to partake in online profiling, despite privacy concerns. On the other hand, the perceived risk associated with an outcome can decrease the force of action toward partaking in online personalization. In this study, we examine two separate contexts, personalized advertising and personalized service. We expect 15 consumers to place different values on the two outcome contexts, due to varying levels of perceived benefit and risk of the activities. Such a difference in each outcome’s attractiveness should therefore affect consumer willingness to share information. . Previous research has attempted to understand consumers’ “willingness to tradeoff” personal information for benefits (Westin 1991) across various industries. Such research found that public attitudes toward such tradeoffs were inconsistent across various industry contexts; such inconsistencies suggest a variance in consumer perceived benefits across industries. Industry differences are likely embedded in the service type offered to the customer. Therefore, in this study we examine consumer willingness to be profiled online across two types of service offerings, rather than industries. Due to the negative connotation of advertising (McLaughlin 2002), consumers assess the benefits of personalized service and personalized advertising to be different. However, since most of the independent variables are such the greater concern should translate in to less willingness to share; we expect the hypotheses to hold across the two contexts of advertising and service. The hypothesis we expect to be different across contexts is Hypothesis 4, dealing with previous privacy invasions. Recall that we believe that previous privacy invasions affect Instrumentality, or the expectation that other outcomes will lead to the expected outcome. Because the outcome of personalized advertising is generally viewed as less valuable than personalized service, we expect that a previous privacy invasion will have a significant effect on consumers motivation to participate in an activity with negative connotations, namely online profiling in the advertising context. On the other hand, we believe that personalized service will be viewed as an outcome less 16 associated with previous negative outcomes, and thus prior privacy invasions will not have a significant effect. This leads us to the following hypotheses: Hypothesis 4a: Consumers who have previously experienced an online privacy invasion are less willing to partake in online profiling in the advertising context. Hypothesis 4b: Consumers who have previously experienced an online privacy invasion will be just as likely to partake in online profiling in the personalized service context as those consumers who have not previously experienced an online privacy invasion. Importance informationknowledge features Importance of Privacy Policies Previous Online Privacy Invasion Privacy Concern - Willing to be profiled online for Personalized Service - Willing to be profiled online for Personalized Advertising Figure 2: Research Model 4) Data and Measurement The context for this research is the use of personal information gathered through web sites and user willingness to allow personal information to be collected and used by firms online. The study is based on a fresh analysis of data from a survey done at a large Internet service provider during the summer and fall of 1998. The survey focuses on 17 issues of personal information collection through specific online scenarios, as well as general attitudes and user demographics. The survey was designed to focuses on how people respond to situations where personal information is collected online. In a prestudy, it was found that variance across participants in information sharing habits was best revealed through questions based on specific online scenarios. Thus, specific purchasing scenarios, focused around information goods and financial services, were utilized. The survey also aimed to determine participants' general attitudes and demographics. Attitude and demographic questions were taken from other studies, such as Westin (1998), so the sample could be matched up against the previous studies. For all the constructs of this study, explicit questions were used as a mechanism for deriving information from the participants. The survey was developed and pre-tested on non-technical employees and summer students of the service provider, as well as with two classes at Harvard and the Massachusetts Institute of Technology (MIT). Prospective survey participants were selected from the Digital Research, Inc. (DRI) Family Panel. The DRI Family Panel is a group of random Internet users that participate in product evaluations and survey responses for Family PC magazine. Approximately one-third of the panel members are FamilyPC subscribers, and most of the panel members who are not subscribers joined the panel after visiting the FamilyPC Web site. Invitations to complete the Web-based survey were emailed to 1,500 Family Panel members (selected randomly), resulting in 523 completed surveys in November of 1998, a response rate of 35%. Code numbers were used to ensure that each respondent completed the survey only once, and a sweepstakes was offered to encourage participation. 18 Similar to recent work on information privacy (Stewart and Segars 2002, Harris and Westin 1998), the sample differed from a nationally representative sample in education, Internet usage, and household income. Summary demographic information is shown in Tables 2A and 2B. Variable Use Computer at Home Table 2A: Demographics (Yes/No) Yes No 375 (98.4%) 2 (.5%) Use Computer at Work Send or Receive Email 260 (68.2%) 379 (99.5%) 119 (31.2 %) 2 (.5%) Visit Web sites 379 (99.5%) 2 (.5%) Are you Male or Female Male: 183 (48.0%) Female: 195 (51.2%) Table 2B: Demographics Variable How often do you use the Internet? (1= once a month, 7 = several times a day) In What Year did you first hear about the Internet? What is the highest level of school completed? (1= Less than High School, 5= Post graduate) Total 1997 Household Income (1= $15k or less, 6 = $75k or over) How many people live in your household How many children, ages 8-12, live in your household? Mean 6.44 Std. Dev. 0.86 1994 3.58 3.10 0.96 4.33 1.37 3.31 1.31 0.50 0.99 All of the items selected for use in this study were chosen from the larger questionnaire administered by this large Internet service provider. Most of the items used were single questionnaire items that aligned with the construct being assessed (For example a single question of whether a person would be willing to participate in online profiling for online personalized service was used to measure a consumer’s willingness to be profiled for personalized service.) The item selection was based on the attitude the construct was attempting to assess. There were two constructs where multiple items were selected, namely Privacy Concern and Information Features. For these constructs, all items that theoretically fell into either the category of consumer privacy concern or 19 features a firm could implement towards increase consumer information control, were factor analyzed, as explained below. Dependent Variable Two dependent variables were used to measure consumer willingness to be profiled for customized service and customized advertising. These five-point Likertscaled variables are: 1) consumer willingness to be profiled by a familiar site for customized service (CSERV); and 2) consumer willingness to be profiled by a familiar site for customized advertising (CADV). All instrument question details can be found in Appendix A. Independent Variables Previous research has found that, in the context of systems development, a client’s knowledge of a project is instrumental in allowing the client to feel in control (Kirsch 1996). In this study, we aim to extend these findings and the expectancy theory of motivation to the online client-firm relationship; thus, we expect that client concern over knowledge of information and use (INFO_KNOWUSE) will decrease their willingness to share information online. The first independent variable, “knowledge of information and use” (INFO_KNOWUSE), was measured by four three-point Likert-scaled items. The four items were factor analyzed using a varimax rotation. All four items loaded unambiguously on a single factor and were combined to form the “knowledge of information and use” (INFO_KNOWUSE) construct (Cronbach alpha=0.75). The second independent variable, “importance of privacy policy” (PRIV-POL), is aimed at providing a contrast to the “knowledge of information and use” independent 20 variable. It is possible that consumers have no interest in knowing the details of what information is being stored and how it is used. Rather, they may only be interested in knowing that the company has a privacy policy. Thus, we control for the importance of such a privacy policy through the use of a single three-point Likert-scaled item. Prior research has also established that demographic variables are associated with an individual’s privacy concern. For example, Culnan (1995) found that demographics, experience with direct marketing, and privacy concern were significantly associated with individual knowledge regarding information removal procedures. However, prior research also suggests that such demographic differences are captured by both attitudinal and behavioral variables (Azjen and Fishbein 1980). For this reason, we control for the demographic variables privacy concern and previous privacy invasion1 only. An individual’s previous experience can shape their concern in information sharing. As shown in previous work (Culnan 1995), concern for privacy was measured using one demographic variable, namely, whether a participant believed his or her privacy had been previously invaded (PREV-INV). Seventy-three respondents (19.2 percent) reported being victimized by what seemed to be an invasion of their privacy online (compared with 21 percent from Culnan’s 1995 study and 23 percent from the 1991 Equifax survey). Concern for information privacy is a tested, multidimensional construct (Stewart and Segars 2002, Smith et. al 1996). However, due to the limitations of the secondary data used, we elected instead to control for general privacy concerns, as previously done by Culnan (1993). Privacy concern was measured by two four-point Likert-scaled items. The two variables were factor analyzed using a varimax rotation. Both items loaded 1 We did run the model controlling for gender, education, and computer usage. The results of the variables of interest were unchanged, and the variables did not add much explanatory power to the models. 21 unambiguously on single factor and were combined to form a “general privacy concern” (PRIV-CONC) scale (r=0.59, p<0.0000; Cronbach alpha= 0.87) Table 3 contains descriptive statistics for, and correlations between, the dependent and independent variables. Table 3 Descriptive Statistics and Inter-Construct Correlations for Dependent and Scaled Independent Variables Variable Mean S.D. CSERV 2.12 0.79 CADV INFOKNOWUSE PRIV-POL PREV-INV PRIV-CONC 2.53 1.47 1.58 1.81 1.76 Table 4 CSERV CADV 1.02 0.45 0.61 -0.12 -0.17 0.63 0.39 0.69 -0.01 -0.08 -0.21 -0.08 -0.13 -0.21 INFOKNOWUSE PRIVPOL PREVINV 0.44 0.08 0.28 0.02 0.29 0.11 Descriptive Statistics and Inter-Item Correlations for Dependent and Scaled Independent Variables Variable PRIVCONC 2 PRIV-CONC1 PRIVCONC 1 1.0000 PRIV-CONC2 INFO-USE1 INFO-USE2 INFO-USE3 INFO-USE4 PREV-INV PRIV-POL 0.4671 0.4324 0.4732 0.2263 0.1755 0.0736 0.2368 1.0000 0.3985 0.4224 0.2722 0.2453 0.1046 0.2462 INFOUSE1 INFOUSE2 INFOUSE3 INFOUSE4 PREVINV PRIVPOL 1.0000 0.5629 0.0521 0.0678 0.0610 0.2762 1.0000 0.2450 0.2190 0.0257 0.2707 1.0000 0.7823 0.1094 0.2596 1.0000 0.1007 0.2900 1.0000 0.0253 1.0000 Methodology Our dependent variables, willingness to be profiled for customized service (CSERV), and willingness to be profiled for customized advertising (CADV), are both five-point Likert scaled items. Essentially, the dependent variables are rank-ordered ordinal variables; though level two is higher than level one, the difference in willingness to be profiled between levels two and one is not necessarily the same as the difference 22 between level three and two. Hence, use of simple multiple regression in such a case would lead to inefficient ordinary least squares estimates (OLS) (Kmenta 1986). In order to overcome the limitations of OLS in this setting, a multinomial logit or probit choice model can be utilized. Since there is an inherent ordering in the five levels of profilewillingness, we utilize an ordered probit model for estimation (Zavonia and McElvey 1975). We develop the ordered probit formulation below. We define a latent (unobserved) continuous variable that measures the continuous (cardinal) value of profile-willingness. In the standard regression framework, the latent variable is a linear combination of the explanatory variables as provided below: Latent variable= β ' X i + ε i (1) Note that Xi is a vector of explanatory variables for the ith consumer, β ' is the associated vector of parameters, and ε i is the stochastic error term, which is assumed to be normally distributed across observations in a probit model. In addition, δ i is the threshold for the ith level of willingness to share information. The translation of the latent variable to the observed variable is given in Appendix B. To ensure that the probabilities sum to one, the probability of willingness to be profiled at level 5 is the complement of the sum of the probabilities of the remaining four values. In order to ensure that all probabilities are positive, the following conditions must be met: δ 3 > δ 2 > δ1 > 0 (2) Where the cutoffs δ 3 , δ 2 , and δ 1 must be estimated along with the parameter β ' . Details on the cutoffs can be found in Appendix B. The likelihood expression for an observation i, is stated as: 23 1 Li = 2π l j − zi2 ∫ e 2 dzi ∏ j =1 l j −1 5 Dj (3) where the upper limits of the integral are: l0= - ∞ , l1= − β ' X i , l2=δ 1 − β ' X i , l3=δ 2 − β ' X i , l4=δ 3 − β ' X i , and l5= + ∞ . In equation (3), only one of the five integrals applies for a specific consumer response of willingness to be profiled for customized offerings. Thus, with each observation treated as independent, the likelihood expression for all consumers is the product of the likelihood for each individual consumer. n L = ∏ Li . (4) i =1 The vector of explanatory variables X consists of four factors: 1) knowledge of information and use (INFO_KNOWUSE); 2) importance of privacy policy (PRIV_POL); 3) previous privacy invasion (PREV_INV); and 4) general privacy concern (PRIV_CONC). In addition, we include an intercept term; each element of the β vector measures the marginal impact of the associated explanatory factor. Note that due to the non-linearity, we cannot directly compare the parameters by magnitude. The model specification is as follows: β ' X i = α + β1 INFO _ KNOWUSEi + β 2 PRIV _ POLi + β 3 PREV _ INVi + β 4 PRIV_CONCi Although we model the underlying measures of willingness to have personal information used by a familiar site for customized service (CSERV_FAM) or customized advertising (CADV_FAM) as a linear combination of the factors, the probability associated with each rating is inherently nonlinear as indicated in equation (3). 24 5) Results and Analysis We estimated the probit model for the two independent variables, customized service (CSERV) and customized advertising (CADV), independently. The maximum likelihood estimates of the parameters of these models are presented in Table 4. Table 4: Maximum Likelihood Ordered Probit Estimators Variable INFO_KNOWUSE PRIV_POL PREV_INV PRIV_CONC Log Likelihood Value Prob. > χ 2 CSERV (p-value) -0.2861** (0.040) 0.1622 (0.102) -0.0799 ( 0.583) -0.3364** (0.000) CADV (p-value) -0.2900** (0.031) 0.04595 (0.641) -0.2513* (0.080) -0.2996** (0.001) -441.9066 0.0002 -536.1750 0.0000 As predicted in our hypotheses, the main independent variable of interest, concern over knowledge of information and use (INFO_KNOWUSE), is negatively significantly associated with willingness to be profiled online for both models, customized service, and customized advertising. In addition, the demographic control variable of general concern for privacy online (PRIV-CONC) is negative and significant across both models. However, the demographic variable of previous privacy invasion (PREV_INV) is not significant in the case of CSERV; it is, however, positive and significant in the case of CADV. The overall results for the ordered probit model for CSERV were also significant (Log Likelihood= -441.90658, Prob > χ 2 =0.0002). All the significant parameter estimates are negative. For negative parameters, a larger negative magnitude suggests a lower probability of willingness to be profiled for customized offerings. 25 However, the relationship is intrinsically nonlinear; we cannot proportionately compare across the negative parameter estimates. The third column of Table 4 summarizes the results for the second model, where the dependent variable is willingness to be profiled online for customized advertising (CADV). Here, demographic control variables, previous privacy invasion (PREV_INV), and general concern for online privacy (PRIV-CONC) are significant. By contrast, previous privacy invasion (PREV_INV) was not significant in the CSERV model. The overall results for the ordered probit model for CADV signify a robust model; the model explains a significant amount of variance (Log Likelihood= -536.1750, Prob > χ 2 =0.0000). All the significant parameter estimates are negative in the case of model 2, the CADV model. These results provide support for the main hypothesis of the study, namely, that the value of perceived control, assessed through concern over information-knowledge features, would be negatively associated with consumer willingness to be profiled online. Consumers that are concerned about having access to their information within company databases are less willing to share information. The same finding holds true for consumers who belabor the length of time in which personal information remains in a firm database. These results suggest that managers may need to focus on such concerns to increase the likelihood of customer willingness to be profiled by their firm. By addressing these concerns, firms are likely to increase the number of consumers willing to be profiled online for customized service or customized advertising. The concern for a privacy policy, on the other hand, was neither significant in the case of customized service, nor in the case of customized advertising. Thus, consumer 26 rated importance of a privacy policy was not found to be associated with a willingness to be profiled online for customized service or advertising. Lastly, we hypothesized that demographic variables, such as general privacy concern and previous privacy invasions, would be associated with a consumer’s willingness to be profiled for customized offerings. In both ordered probit models, the general privacy concern is significant, providing support for the hypothesis that general privacy concern would be negatively associated with such profile-willingness across contexts. Hence, users more concerned with privacy are statistically less likely to share information online. However, prior privacy invasion experience was significant only in the context of online advertising, not in the case of online service. Therefore, users with previous privacy invasion experience have a lower probability to be willing to be profiled online for customized advertising. However, such a result does not hold true with regard to online service. As noted earlier in association with expectancy theory, this may indicate a difference in Valence perceptions, or the attractiveness of the outcome to the individual, across the two contexts. Assuming consumers perceive the benefits of online service to be greater than online advertising, they will be more motivated to share information for online service. This increased motivation to share information will make previous privacy invasion experience insignificant in the online service case, but significant in the online advertising case, as our results suggest. 6) Discussion Effective use of customer information is a critical success factor for firms online. Through profiling consumers and presenting customized service and advertising, firms 27 can increase consumer value as well as firm revenue. The challenge for firms, then, becomes collecting and using information in such a way that consumer privacy concerns are addressed. This study examined the role of perceived control, implemented through information-knowledge features. Such features can be used to address a consumer’s willingness to be profiled online for customized offerings. The results have significant managerial implications, as they suggest that firms can garner increased customer participation in customized offerings through the implementation of informationknowledge features. Such increased consumer participation may then translate into a competitive advantage for companies. The findings of this study are in accordance with Expectancy theory, and also with previous findings that indicate that perceived control over events and uses of information, result in a consumer’s decreased concern for privacy invasion (Laufer at al. 1976). In this case, the importance of perceived control reflects an increased consumerconcern about sharing their information with online firms. While previous findings have also shown that users with a greater privacy concern perceive little control over the use of personal information (Stone et. al 1983), ours is the first study to show that perceived control would increase a consumer’s tendency to be profiled online for customized offerings. At the outset, we expected that the importance of a privacy policy would be significantly associated with a willingness to be profiled online. The lack of such a significant association was initially surprising. However, upon reflection, one possible reason for such a result became clear: privacy policies largely go unread by consumers. In fact, according to Forrester Research, less than 1 percent of the visitors to six major 28 online travel sites during April 2001 actually read privacy policies (Regan 2001). Thus, while consumers may rate a privacy policy as important, few of them actually take note of the policy when using a site. Examining the willingness to be profiled online in the specific contexts of online service and online advertising highlights environmental differences. The most distinct difference between online service and online advertising is the effect of importance of previous privacy invasion. Previous work did not reveal a clear association between previous privacy invasion experience and attitudes toward secondary information use (Culnan 1993); therefore, this significant result in the case of online advertising is quite interesting. The differing result across contexts regarding previous privacy invasion suggests that people consider customized service more beneficial than customized advertising and are therefore more willing to be profiled online, despite previous online privacy invasions. Online advertising, on the other hand, is largely perceived as less beneficial (McLaughlin 2002). Previous privacy invasion is, therefore, associated with a lower willingness to be profiled online in the case of customized advertising. Limitations and Future Research Like other empirical research, this research also has several limitations, and the results need to be read with caution. As described above, the study was based on secondary data analysis of a survey designed to measure opinions toward privacy and information sharing online. Concern for information privacy is a tested, multidimensional construct (Stewart and Segars 2002, Smith et. al 1996). However, due to the limitations of the secondary data, we instead controlled for general privacy 29 concerns, as previously done by Culnan (1993). Individual questionnaire items were designed to be unbiased. However, several items, such as previous privacy invasion, were measured using single questionnaire items. In addition, the sample has a slight bias in favor of more educated, affluent, and Internet savvy individuals. Therefore, the results should be viewed with some caution. The strength of the research is that the data sample is consistent with other recent work regarding information privacy (e.g. Stewart and Segars, 2002), and the results are consistent with theory. Conclusion Personalized service is increasingly becoming a source of value for both consumers and firms (Farag and Krishnan, 2002). However, investments in personalization may come at the cost of consumer privacy. Privacy has, therefore, become an issue of strategic importance for companies operating in the informationcentric, networked global economy. In order to provide consumer-driven customized service, firms must compel consumers to provide them information. Through the use of information-knowledge features, we examined the effect of perceived control on a consumer’s willingness to be profiled online for customized service and advertising. We found that concern over knowledge of information and usage is associated with a decreased consumer willingness to be profiled online across contexts. However, we also found that the perceived benefit of customization affects the importance of previous privacy invasion on that very willingness. In the case of customized service, where benefit is more apparent to consumers, previous privacy invasions are not significant; the potential benefit of the service outweighs the potential risk of a privacy invasion. In the 30 case of customized advertising, on the other hand, the benefit is less apparent, and the risk of an intrusion, (i.e. email spam) is more apparent. In such a case, previous privacy invasion is significant. Thus, companies must focus on reducing such perceived risk through implementing information-knowledge features. For managers to be successful in encouraging consumers to partake in online customized service, their firm’s entire information practices should be given unfettered accessibility. Such accessibility may even be provided in a limited format. Future research may therefore examine, over time, the effectiveness of varying levels of accessibility on consumer willingness to be profiled online. In this study, we have provided results managers can utilize to encourage consumer participation in online profiling for customized service and advertising. Our findings should therefore help firms tailor their online service offerings and maximize information collection across various consumer segments, as a result. 31 Appendix A: Survey Instrument Details Dependent Variables Willingness to have personal information used by a familiar site for customized service (CSERV) was measured by a five-point Likert-scaled item ranging from “Definitely Not” to “Definitely Would”: • Some Web sites assign visitors special user identification numbers. Web browsers can send these numbers back to the site automatically on a return visit. This allows Web sites to recognize return visitors and provide customized service based on previous activities. If a site that you frequented asked you whether it could assign you an identification number so that it could provide you with customized service, would you agree? (mean=2.12, s.d.=0.79). Willingness to have personal information used for customized advertising by a site the user was familiar with (CADV) was also measured by a single five-point Likertscaled item ranging from “Definitely Not” to “Definitely Would”: • Some Web sites use special identification numbers not only to customize site content, but also to customize advertising that appears on the site and make sure that visitors are not repeatedly shown the same advertisements. If a site that you frequented asked you whether it could assign you an identification number so that it could provide you with customized advertising, would you agree? (mean=2.53, s. d. =1.02) These questionnaire items were given in the order presented. 32 Independent Variables The first independent variable, “knowledge of information and use” (INFO_KNOW), was measured by four three-point Likert-scaled items ranging from “Very important” to “Not important”: • Importance of whether a company will allow me to find out what information about me they keep in their databases (mean=1.40, s.d.=0.58) • Importance of whether a site tells me how long they will retain information they collect from me (mean=1.87, s.d.=0.72) • Importance of the purpose for which the site wants to collect info from me (mean=1.29, s.d.=0.52) • Importance of whether a site is going to use the information they collect from me in a way that will identify me (mean=1.30, s.d.=0.54) The second independent variable, importance of a privacy policy (PRIV_POL), was assessed through the use of a single three-point Likert-scaled items ranging from “Very important” to “Not important”: • Importance of whether or not the site posts a privacy policy (mean=1.58, s.d.=0.63) The third independent variable, whether a participant believed his or her privacy had been previously invaded (PREV-INV), was assessed with a single yes/no item. The item used was: • Have you ever personally been the victim of what you felt was an invasion of your privacy when using the Internet? (mean= 1.81, s.d. = 0.3896) 33 The fourth independent variable, privacy concern (PRIV-CONC), was measured by two four-point Likert-scaled items, ranging from “Very concerned” to “Not concerned at all”: • How concerned are you about threats to your personal privacy in America today? (mean= 1.77, s.d.= 0 .75 ) • How concerned are you about threats to your personal privacy when using the Internet? (mean= 1.75, s.d.= 0 .71 ) Appendix B: Mapping of the Latent Variables to the Observed Levels Underlying the indexing in ordered models is a latent but continuous descriptor of the response. In an ordered probit model, the random error associated with the continuous descriptor is assumed to follow a normal distribution. The unobserved willingness to share score is mapped to the observed willingness to share levels as shown below, with ε i distributed as a standard normal. The probability of each value of the score is then beside it. Level 1 if β ' X i + ε i ≤ 0 Prob[Level 1] = Φ (− β ' X i ) Level 2 if 0 ≤ β ' X i + ε i ≤ δ 1 , Prob[Level 2] = Φ (δ 1 − β ' X i ) − Φ (− β ' X i ), Level 3 if δ 1 ≤ β ' X i + ε i ≤ δ 2 , Prob[Level 3] = Φ (δ 2 − β ' X i ) − Φ (δ 1 − β ' X i ), Level 4 if δ 2 ≤ β ' X i + ε i ≤ δ 3 , Prob[Level 4] = Φ (δ 3 − β ' X i ) − Φ (δ 2 − β ' X i ), Level 5 if δ 3 ≤ β ' X i + ε i . Prob[Level 5] = 1 − Φ (δ 3 − β ' X i ), Appendix C: Random Parameters Ordered-probit Model 34 In this model, we allow the effect of the parameters to vary across consumers according to a known multivariate distribution. Generically, we can rewrite the latent variable equation (1) as: Latent variable= Γi' X i + ε i (B1) where Γi = β + γ i . The vector β represents the mean effect of the parameters across all customers, whereas γ i represents the vector of random deviations from the mean vector β for customer i. Therefore, for the sake of parsimony, we allow for variation of the parameters across individual customers. Since γ i follows a multivariate distribution, Γi also follows a multivariate distribution. Since we include the mean vector in Equation (B1), the mean vector of γ i other than a zero vector is not identified. References Ajzen, I. and Driver, B.L. (1991). “Prediction of leisure participation from behavioral, normative, and control beliefs: An application of the theory of planned behavior.” Leisure Sciences 13: 185-204. Azjen, Izek and Fishbein, Martin (1980). Understanding Attitudes and Predicting Social Behavior. Englewood Cliffs, NJ, Prentice Hall. Applegate, L.M. and Wishart, N. (1990). Frito-Lay, Inc: A Strategic Transition (C). Boston, MA, Harvard Business School. Case 9-190-071. Armitage, C.J. and Conner, M. (1999). “The theory of planned behavior: Assessment of predictive validity and "perceived control".” British Journal of Social Psychology 38: 3554. Baker, J.A. (1991). Personal Information and Privacy. Proceedings of the First Conference on Computers, Freedoms, and Privacy, Los Alamitos, CA, IEEE Computer Society Press. Bates, Allen P. (1964). “Privacy, A Useful Concept?” Social Forces 42(4): 430-434. 35 Berry, Leonard L. (1995). On Great Service: A Framework for Action. New York, The Free Press. Bies, Robert (1993). “Privacy and Procedural Justice in Organizations.” Social Justice Research 6(1): 69-86. Bies, Robert J. and Tripp, Thomas M. (1996). Beyond Distrust: 'Getting Even' and the Need for Revenge. Trust in Organizations: Frontiers of Theory and Research. R. M. Kramer and T. R. Tyler. Thousand Oaks, CA, Sage: 246-260. Blattberg, Robert C. and Deighton, John (1991). “Interactive Marketing: Exploiting the Age of Addressability.” Sloan Management Review 33(1): 5-14. Bloom, Paul N., Milne, George R., and Adler, Robert (1994). “Avoiding Misuse of Information Technologies: Legal and Societal Considerations.” Journal of Marketing 58(1): 98-110. Breslow, R. A., Sorkin, J. D., Frey, C. M. & Kessler, L. G. (1997). Americans' knowledge of cancer risk and survival. Preventive Medicine, 26, 170-177. Bruns, W.J. and McFarland, F.W. (1987). “Information Technology Puts Power in Control Systems.” Harvard Business Review 65(5): 89-94. Chellappa, Ramnath K., and Sin, Raymond (2002). Personalization versus Privacy: An Empirical Examination of the Online Consumer's Dilemma. International Conference on Information Systems, New Orleans, LA. Clarke, R.A. (1988). “Information Technology and Dataveillance.” Communications of the ACM 31(5): 498-512. Culnan, M.J. (1989). Designing Information Systems to Support Customer Feedback: An Organizational Message System Perspective. International Conference on Information Systems, Association for Computing Machinery, New York, NY. Culnan, M.J. (1993). “How did they get my name? An exploratory investigation of consumer attitudes toward secondary information use.” MIS Quarterly 17(3): 341-363. Culnan, M.J. (1999). Georgetown Internet privacy policy study: Privacy online in 1999: A report to the Federal Trade Commission. Washington, DC: Georgetown University Culnan, Mary J., and, Armstrong, Pamela K. (1999). “Information Privacy Concerns, Procedural Fairness, and Impersonal Trust: An Empirical Investigation.” Organization Science 10(1): 104-115. Daganzo, C. (1979). Multinomial Probit: The Theory and Its Application to Demand Forecasting. New York, NY, Academic Press. 36 Dinev, T., and, Hart, P., (2002). “Internet Privacy Concerns and Trade-Off FactorsEmpirical Study and Business Implications.” Working Paper, Florida Atlantic University. Direct Marketing Association (1994). Fair Information Practices Manual. New York, Direct Marketing Association. Dowling, Melissa (1993). When You Know Too Much. Catalog Age. October: 73-75. Dresner, Stewart (1996). “Data protection roundup.” Privacy Laws Business (U.K.) 33(January): 2-8. Equifax (1990, 1991, 1992, 1993, 1996). “Harris-Equifax Consumer Privacy Survey.” Equifax, Inc, Atlanta. Farag, Neveen I., and Krishnan. M.S. “Price Dispersion and Online Information: Quality or Quantity? ” UMBS Working Paper, 2002. Farag, Neveen I., and Van Alstyne, Marshall W. “Information Technology: A Source of Friction? An Analytical Model of How Firms Combat Price Competition Online.” Proceedings of the 2nd ACM Conference on Electronic Commerce, October 2000, Minneapolis, MN. Ferris, Kenneth R. (1977) “A Test of the Expectancy Theory of Motivation in an Accounting Environment.” The Accounting Review 52(3): 605-615. Franklin, Charles E. H. (1996). Business Guide to Privacy and Data Protection Legislation. Paris, ICC Publishing S. A. FTC (1999). “Self-Regulation and Privacy Online.” Report to Congress. Glazer, Rashi (1991). “Marketing in an Information-Intensive Environment: Strategic Implications of Knowledge as an Asset.” Journal of Marketing 55(4): 1-19. Goodwin, Cathy (1991). “Privacy: Recognition of a consumer right.” Journal of Public Policy Management 10(1): 149-166. Greene, M. (1997, October). Who's zoomin' who on the web. Black Enterprise, 28 (3), 40-42. Hanson, W. (2000). Principles of online marketing. Cincinnati, OH: South-Western College Publishing Louis Harris and Associates, Inc. (1991). Harris-Equifax Consumer Privacy Survey, 1992. Atlanta, GA, Equifax, Inc. 37 Harris, Louis and Westin, Alan F. (1998). E-commerce and Privacy: What Net Users Want. Hackensack, NJ, Privacy and American Business. Hauser, J.R. (1978). “Testing the Accuracy, Usefulness, and Significance of Probabilistic Choice Models: An Information Theoretic Approach.” Operations Research 26: 406-421. Hendricks, Evan (1998). “Capital Insights.” Privacy Times 18(15): 1. Heskett, J. L., Sasser, W. E., and Hart, C.W.L. (1990). Service Break-throughs: Changing the Rules of the Game. New York, The Free Press. Hoffman, D., Novak, T., and Peralta (1999). “Information privacy in the market-space: Implications for the commercial use of anonymity on the web.” Information Society 15(4): 129-139. Jaworski, B.J. (1988). “Toward a theory of marketing control: Environmental context, control types, and consequences.” Journal of Marketing 52: 23-39. Kirsch, L.J. (1996). “The management of complex tasks in organizations: Controlling the systems development process.” Organization Science 7: 1-21. Klobas, Jane E. and Clyde, Laurel A. (2000). “Adults Learning to Use the Internet: A longitudinal study of attitudes and other factors associated with intended Internet use.” Library and Information Science Research 22(1): 5-34. Kmenta, J. (1986). Elements of Econometrics. New York, NY, Macmillan Publishing Company. Laufer, R.S., Proshansky, H.M. and Wolfe, M (1976). Some Analytic Dimensions of Privacy. Environmental Psychology: People and Their Physical Settings, 2nd ed. H. M. Proshansky, Ittelson, W.G., and Rivlin, L.G. New York, NY, Holt, Rinehart, and Winston: 206-217. Laufer, R.S. and Wolfe, M. (1977). “Privacy as a Concept and a Social Issue: A Multidimensional Development Theory.” Journal of Social Issues 33(3): 22-42. Mason, R.O. (1986). “Four ethical issues of the information age.” MIS Quarterly 10(1): 4-12. McLaughlin, Laurianne (2002). The Straight Story on Search Engines. PC World. July. Milberg, Sandra J., Smith, H. Jeff, and Burke, Sandra, J. (2000). “Information Privacy: Corporate Management and National Regulation.” Organization Science 11(1): 35-57. 38 Milne, George R. and Gordon, Mary Ellen (1993). “Direct Mail Privacy-Efficiency Trade-offs Within an Implied Social Contract Framework.” Journal of Public Policy and Marketing 12(2): 206-215. Porter, M.E. and Millar, V.C. (1985). “How Information gives you Competitive Advantage.” Harvard Business Review 63(4): 149-160. Regan, Keith (2001). Does Anyone Read Online Privacy Policies? www.EcommerceTimes.com. Safire, William. “The Intrusion Explosion.” The New York Times: May 2, 2002. Schneider, Benjamin, and Bowen, David E. (1995). Winning the Service Game. Boston, MA, Harvard Business School Press. Smith, H. Jeff (1994). Managing Privacy: Information Technology and Corporate America. Chapel Hill, NC, University of North Carolina Press. Smith, H. Jeff and Milberg, Sandra J., and Burke, Sandra J. (1996). “Information Privacy: Measuring Individuals' Concerns about Organizational Practices.” MIS Quarterly 20(2): 167-196. Sohn, Jung Hoon Derick (1994). “Social Knowledge as a Control System: A Proposition and Evidence from the Japanese FDI Behavior.” Journal of International Business Studies 25(2): 295-324. Speier, C., Harvey, M., & Palmer, J. (1998). Virtual management of global marketing relationships. Journal of World Business, 33(3), 263-276. Stewart, Kathy A. and Segards, Albert H. (2002). “An Empirical Examination of the Concern for Information Privacy Instrument.” Information Systems Research 13(1): 3649. Stone, Eugene F. , Gardner, D. G., Gueutal, H.G., and McClure, S. (1983). “A Field Experiment Comparing Information-Privacy Values, Beliefs, and Attitudes Across Several Types of Organizations.” Journal of Applied Psychology 68(3): 459-468. Stone, Eugene F. and Dianne L. Stone (1990). Privacy in Organizations: Theoretical Issues, Research Findings, and Protection Mechanisms. Research Findings in Personnel and Human Resources Management. K. M. Rowland and G. R. Ferris. Greenwich, CT, JAI Press. 8. Thiobodeau, Patrick (2002). Privacy bill carries high IT price. ComputerWorld. May 27. Vroom, Victor H. (1964) Work and motivation. New York: Wiley. 39 Westin, Alan F. (1967) Privacy and Freedom. New York: Athenaeum. Westin, Alan F. (1991). Domestic and International Data Protection Issues. How the American Public Vies Consumer Privacy Issues in the Early 90s -- and Why. Testimony before the Subcommittee on Government Information, and Agriculture, Committee on Government Relations, U.S. House of Representatives. Washington, D.C., U.S. Government Printing Office: 54-68. Westin, Alan F. (1997). Commerce, communication, and privacy online. Hackensack, NJ, Center for Social and Legal Research. Womeodu, R. J. & Bailey, J. E. (1996). Barriers to cancer screening. Medical Clinics of North America, 80, 115-133. Woodman, R.W., Ganster, D.C., Adams, J., McCuddy, M.K., Tolchinsky, P.D., and Fromkin, H. (1982). “A Survey of Employee Perceptions of Information Privacy in Organizations.” Academy of Management Journal 25(3): 647-663. Wortman, C. 1975: Some Determinants of Perceived Control. Journal of Personality and Social Psychology, 31, 282-294. Zavonia, T., and McElvey, W. (1975). “A Statistical Model for the Analysis of Ordinal Level Dependent Variables.” Journal of Mathematical Sociology Summer: 103-120. 40