KimBaraszJohn18 be5ba706 b8c3 4ac4 bb48 3cc462bb0e08

Transcript

1 1 Why Am I Seeing This Ad? The Effect of Ad Transparency on Ad Effectiveness TAMI KIM, KATE BARASZ, AND LESLIE K. JOHN *Tami Kim is an assistant professor in the M epartment at the Darden School of arketing D Business ([email protected]). Kate Barasz is an assistant professor in the M arketing Department at IESE Business School ( [email protected]). Leslie K. John is an a ssociate professor of Business Administrati on in the Negotiations, Organizations, and Markets Unit at Harvard Business School ([email protected]). This research is based on the lead author’s dissertation, which was conducted at Harvard Business School. Correspondence: Tami Kim.

2 2 ENT CONTRIBUTION STATEM With growing concerns about privacy in the online space , consumers and regulators are increasingly demanding ad transparency — asking online advertisers to disclose how they gather focuses and use consumers’ personal information to display targeted ads. While existing research un aware of underlying ad practices , there on the effectiveness of online ads when consumers are is a dearth of knowledge on how disclosing ad practices influence s ad performance. We propose and test a conceptual framework that pre dicts how and why ad transparency impacts ad effectiveness. It is premised on the notion that ad transparency undermines ad effectiveness when it exposes marketing practices that violate consumers’ beliefs about “information flows” —how their information to move between parties. Because this territory is largely uncharted, we ought began with an inductive approach to assess how consumers perceive the acceptability of the varied ways in which their personal information is used to generate targeted ads. This approach revealed two dimensions that predict most of the variance in consumers’ perceived acceptability of different ad practices. Then, taking a deductive approach, we invoked theory to explain the relationship between the acceptability of the revealed practice and ad effectiveness: it is driven by consumers’ relative concern for privacy over their interest in personalizat ion. Finally, we demonstrated the moderating role of brand trust: companies can unlock the benefits of revealing when they have already established consumer trust. Our findings acceptable information flows contribute to the literature on digital advertisin g, information norms, and privacy by providing insight into when and why ad effectiveness may be undermined—versus enhanced —by ad transparency .

3 3 ABSTRACT Given the increasingly specific ways marketers can target ads, many consumers and regulators are demanding ad transparency: disclosure of how consumers’ personal information was used to generate ads. We investigate how and why ad transparency impacts ad effectiveness. Drawing on n- sharing literature about offline norms of informatio that ad transparency backfires , we posit when it exposes marketing practices that violate norms about “information flows” —consumers’ beliefs about how their information ought to move between parties. Study 1 inductively shows that consumers deem information flows acceptable (or not) based on whether their personal information was: 1) obtained within versus outside of the website on which the ad appears , and 2) stated by the consumer versus inferred by the firm (the latter of each pair being less acceptable). Studies 2 and 3 show that revealing unacceptable information flows reduces ad effectiveness , which is driven by increasing consumers’ relative concern for their privacy over desire for the personalization that such targeting affords. Study 4 shows the moderating role of platform trust: when consumers trust a platform , revealing acceptable information f lows increases ad effectiveness. Studies 5a and 5b, conducted in the field with a loyalty program website (i.e., a trusted platform ), demonstrate this benefit of transparency.

4 4 -expanding capacity to track online behaviors and display relevant With marketers’ ever ads, consumers have become increasingly wary of the way s firms gather and use their personal information. For example, the revelation that department stores were surreptitiously tracking -store movements using cell phone data sparked controversy in the media and customers’ in (Clifford and Hardy 2013) . Target Inc. made headlines for sending pregnancy -related beyond coupons to a teenage customer based on her recent purchases. Although the inference was accurate—the teenager was, in fact, preg —the targeting elicited controversy in part because nant . More broadly the customer’s family had been unaware of her pregnancy (Hill 2012) and — unbeknownst to many —the sharing of such consumer information amongst firms is extensive: for instance, Facebook purchased data on 70 million U.S. households, enabling the firm to tailor ads based on users’ purchases (Wasserman 2012) . Given the invasiveness of such practices, many consumers and regulators are increasingly —the disclosure of the ways in which firms collect and use consumer demanding ad transparency personal data to generate behaviorally targeted ads —in an effort to empower consumers and better ensure above -board marketing practices (Greiff 2016; Ramirez et al. 2014; Turow et al. 2009) . In recent years, many firms have voluntarily instituted such practices. Facebook introduced a feature allowing users to find out why any given ad is being shown to them; for example, a user may be informed that an ad for Acme Home Theaters appears based on what she does on Facebook, such as the pages she has “liked” and the ads she has clicked on (appendix A) . Relatedly, a growing number of advertisers have been voluntarily displaying the AdChoices icon, a blue symbol indicating that the accompanying ad has been targeted based on the recipient’s characteristics (Alliance 2014) ; similar to Facebook, consumers can find out why the given ad is displayed to them by clicking the icon. In the same spirit, many websites have begun

5 5 e.g to explicitly alert visitors when tracking software ( ., cookies) is in use, to post privacy policies (Culnan 2000) , and to display logos (“privacy seals”) signaling that privacy standards have been met (Farmer 2015) . Although some firms have been forthcoming with this information, others have resisted, -putting practices may undermine ad concerned that drawing attention to potentially off — i.e., the extent to which an ad increases a effectiveness (Edwards 2009; Learmonth 2009) consumer’s interest in a product. Perhaps reflecting this con cern, some firms have implemented weak forms of ad transparency, merely making information available to consumers should they actively seek it out (e.g., by proactively clicking on the “Why am I seeing this ad?” button). How concerned should firms be about making ad practices transparent? How does ad transparency impact ad effectiveness? When might it reduce versus improve ad effectiveness? Might it improve ad effectiveness under certain conditions? In this paper, we expl ore these questions, developing an understanding of 1) how consumers feel about specific advertising practices and 2) how these perceptions change the way consumers engage with the ad. We study these questions by testing the impact of uously revealing actively and conspic information about ad practices to consumers (e.g., by presenting an ad and its ad practice alongside each other). Such conspicuous disclosure is uncommon in today’s marketplace; digital advertisements are not usually accompanied by informa tion on how they were generated, and when they are, this information is typically inconspicuous, merely made available for the motivated consumer to find. However, our experiments are designed to understand how consumers react to ad transparency, conditional on exposure. The topic of when and why consumers seek out such information, including potential individual differences in this propensity, is an interesting and important topic but one that is beyond the scope of the current

6 6 onsumers’ demand for ad transparency in the General Discussion, paper. (We further discuss c .) including what it implies for the generalizability of our findings We situate o ur account in research on offline norms of information sharing and disclosure and —using a combination of induction and deduction—develop a conceptual account of how and why ad transparency affects ad effectiveness. Specifically, our account demonstrates that ad transparency backfires when it exposes marketing practices that violate norms of “information flows” —c ought to flow between parties. Moreover, we onsumer beliefs of how their information address why this occurs: revealing information flows that consumers deem unacceptable reduces ad effectiveness by increasing consumers’ relative concern for their privacy over their interest in the increased personalization that such targeting affords. Finally, we demonstrate that platform trust enhances the upside of ad transparency : when trust in a given platform is high, ads that are displayed on this platform and reveal underlying acceptable information flows outperform non- transparent ones. CONCEPTUAL DEVELOPMENT In contrast to traditional targeted advertising —such as placing ads in specific markets or broadcasting in targeted time slots —online ad platforms allow advertisers to target consumers with greater efficiency, specificity, and accuracy. For example, the modern marketer can gather and integrate individual -level data to facilitate behaviorally targeted ads —ads based on the recipient consumer’s online behavior. Such developments have added many new and effective tools to the marketer’s arsenal, including behavioral retargeting (displaying ads for products consumers recently viewed ; Lambrecht and Tucker 2013) , content -based targeting (displaying

7 7 on what consumers read , and keyword -based targeting ads based ; Zhang and Katona 2012) (displaying ads based on the terms consumers entered into search engines ; Desai, Shin, and Staelin 2014; Sayedi, Jerath, and Srinivasan 2014; Yang, Lu, and Lu 2013) , all of which have improved marketers’ capacity to reach and persuade consumers. -targeted ads necessarily require access to consumers’ personal data. This However, better presents its own set of challenges with respect to how data are collected , which data are collect ed, and whether consumers would consent to their data being used as such. Most consumers are aware only in the general sense that companies collect and use their data to display ads and are almost never privy to the specifics. For example, fewer than 20% of consumers realize they share their communication history, IP addresses, and web- surfing history when using a standard web browser (Morey, Forbath, and Schoop 2015) , erroneously believing that only explicitly divulged information is harvested. Furthermore , despite the fact that companies routinely sell and package consumer data in aggregated formats, most consumers do not understand how their personal information is consolidated across platforms (Kroft 2013) . Therefore, while improved targeting techniques have been advantageous for firms, the (White 2004) . On the one hand, well -targeted ads are result may be more mixed for consumers objectively more personalized—thus, they should by definition be more relevant and interesting to consumers. They can also fac ilitate the potential discovery of new products uniquely suited to consumers’ needs, wants, and interests (Charts 2014; Tam and Ho 2006) . Transparency, in turn, could make such targeting practices more salient and further increase consumers’ awareness of the degree to which ads are tailored . Indeed , informing consumers that an ad has been targeted based on their behavior can increase the perceived person -product fit, increasing ad effectiveness (Summers, Smith, and Reczek 2016) . However, within the context of consumers’ general lack of

8 8 awareness about how their personal data are used, ad transparency could also make salient something that is not typically top -of-mind: the fact that marketers are collecting and using their personal informat ion, which could raise privacy concerns and potentially decrease ad effectiveness. We reconcile these competing predictions by developing a framework which, at its core, —i.e., ways in which invokes consumer beliefs about which personal information flows We marketers collect and use their personal information—are acceptable versus unacceptable. begin by developing a conceptual account to predict which personal information flows consumers are likely to find acceptable versus unacceptable. In the empirical po rtion of the paper, we first assess whether these predictions emerge in the data ; we then proceed to test how disclosure of these information flows (i.e., ad transparency) affects ad effectiveness. In so doing, we help to account for when and why awareness of behavioral targeting enhances versus detracts from ad effectiveness. A Theory of Offline -to-Online Norm Transference In the offline world, acceptable versus unacceptable flows of information are generally well -established and agreed upon (Grice, Cole, and Morgan 1975; Nissenbaum 2011). A wealth of research shows that people inherently understand, and generally adhere to, implicit rules of communication (Fishbein 1979; Grice et al. 1975) . For instance, people are adept at complying to social rules of disclosure (Altman and Taylor 1973; Caltabiano and Smithson 1983; Derlega and Chaikin 1977; Huang et al. 2017) and penalize non- compliance (Archer and Berg 1978; Rubin 1975). And, with respect to the sharing of personal information in particular, there are norms or “rules” that span across contexts.

9 9 onsider the following two examples of information flow norms in the For instance, c offline world. Example 1: an individual may find it acceptable to share some personal information (e.g., that she is trying to lose weight) with two different friends, but would likely find it unacceptable for one friend to take it upon himself to tell the other. In both cases , the two information; however, the flow pattern of this information is friends ultimately know the same acceptable in the first case but decidedly not in the second case. Example 2: an individual may directly inform a friend about some personal information (e.g., weight loss efforts), but would likely find it unacceptable for that friend to make an overt inference (e.g., speculating about said weight loss efforts), even if the inference is accurate. In both cases the friend ultimately has the same knowledge; however, the first flow pattern tion directly divulged by the source to —informa —is acceptable, while the other the friend —an overt inference (i.e., the lack of a direct link from source to the friend) —is not. In short , in offline contexts, there are strong norms with respect to appropriate flows of personal information—and violations of these norms can feel invasive (Nissenbaum 2004, 2011) . Although offline and online worlds differ greatly in the way personal information is gathered and used, we posit that when determining whether a given ad practice is acceptable, —the ways in which consumers turn to the offline world as a guidepost, where these norms -established and well individuals should behave in sharing personal information—are well - understood. Indeed, philosopher Helen Nissenbaum suggest s that in orienting to the often bewildering online world, consumers look for “the contours of familiar social activities and structures” (Nissenbaum 2011). C onsistent with this theorizing, Moon (2000) empirically demonstrated that the tendency to reciproc ate others ’ disclosures —a well -established norm in offline contexts —also applies in online human- computer interactions.

10 10 These ideas, coupled with the fact that consumers have idiosyncratic, preconceived notions of how firms ought to handle their personal data (Malhotra, Kim, and Agarwal 2004; Milne and Gordon 1993; Smith, Milberg, and Burke 1996), suggest that consumer response to ad transparency will depend on consumers’ perceptions of which information flow patterns should—or should not t consumer —be used to generate the ads. Specifically, we argue tha response to transparency messages will manifest in how they choose to engage with the targeted an ad when it is accompanied by an ad: all else equal, consumers are more likely to engage with -adherent transparency message acceptable, norm (i.e., conveying that the consumer data behind it w ) than an ad with an unacceptable, norm - ere obtained via an acceptable information flow the impact of ad transparency on ad effectiveness violating message. Thus it follows that depends on the perceived acceptability of the information flow it exposes. Specifi cally, we predict: H1 a : A d transparency that reveals unacceptable information flows will reduce ad effectiveness relative to ad transparency that reveals acceptable flows . How will ad transparency fare relative to no ad transparency at all - ? Relative to norm -violating behavior — adherent behavior, norm events more generally —tend to and negative trigger particularly strong reactions ( Baumeister, Bratslavsky, Fink enauer, and Vohs 2001; Fehr and Fischbacher 2004; Fehr and Gächter 2002; Morewedge 2009; Rozin and Royzman 2001 ). Therefore .e., norm -violating , we hypothesize that revealing unacceptable information flows (i activities ) will have a particularly strong effect on consumer behavior: it will reduce ad effectiveness relative to no transparency. Following this logic —t hat “bad is stronger than good” (Baumeister et al. 2001) —revealing acceptable information flows (i.e., norm -adherent activities) may not, on its own, increase effectiveness relative to no transparency . In other words, people

11 11 are likely to punish bad behavior, but may not equivalently reward good behavior; therefore the simple act of disclosing acceptable information may lead to a positive, non- status not necessarily quo response in consumers . Indeed, as we later delineate, additional factors might be necessary -adherent ad transparency to boost ad effectiveness to unlock the capacity for norm relative to no transparency . W e predict: H1b: A d transparency that reveals unacceptable information flows will reduce ad effectiveness relative to no ad transparency. However, one natural question arises: Which information flows do consumers perceive to be acceptable versus unacceptable? The answer to this question, especially as it pertains to behaviorally targeted ads, is not yet known. Therefore as a prerequisite to testing H1 a and H1b, of information flows we use an inductive approach to identify what norms —as they pertain to —consumers believe companies ought to follow . Guided by our theory behaviorally targeted ads -to-online norm transference, we predicted these domains to mirror offline injunctive of offline norms and to be predictive of the effect of ad transparency on ad effectiveness. For instance, as in the aforementioned Example 1, consumers may expect companies to abstain from s haring their personal information with other companies ; and a s in Example 2, consumers may expect companies to only use what consumers have stated about themselves and to refrain f rom making inferences about them . Privacy versus Personalization After determining what constitutes acceptable versus unacceptable information flows, a related second question arises: What underlies the relationship between the accep tability of the information flows revealed by ad transparency and the resultant effectiveness of the ad? From

12 12 ’s perspective, producing effective targeted advertisements entails the careful balance of the firm consumer interest in ad personalization (i.e., how much the ad fits one’s needs and interest s) and concerns (i.e., concern over the safety and control over one’s privacy consumer information personal information; Malhotra et al. 2004; Milne and Gordon 1993; Phelps, Nowak, and Ferrell 2000; Smith et al. 1996) . Previous research suggests that consumers assess the attractiveness of an adverti sed product by looking to its fit with, or personalization to, their needs and wants . Targe (Aguirre et al. 2015; Montgomery and Smith 2009; Tam and Ho 2006) —which ted ads should represent a better fit with and personalization to consumer needs and wants y drive —ma higher ad effectiveness. But marketers can go too far; personalization can backfire. Of particular relevance to the present inquiry, highly personalized ads (compared to less personalized ads) can rouse consumer privacy concerns (Aguirre et al. 2015) . Such concerns can overshadow interest in purchasing the product itself, and more immediately , dampen the likelihood of engaging with advertisements for it—even for a product that personalized targeting would predict be a good person- product fit to (White et al. 2008). Similarly, we propose that ad transparency affects ad effectiveness by shifting consumers’ relative concern s for privacy versus their interest in personalization. Specifically, a message that communicates a norm -violating practice —thus making salient that “acceptable” information flows have been breached —will activate privacy concerns that ultimately eclipse the benefits of personalization ; even the most personalized, perfectly targeted advertisement will flop if the consumer is more focused on the (un)acceptability of how the targeting was done in the first place. Such privacy -related backlash has been documented in other domains, as well . P rivacy concerns prompt behavioral inhibition, akin to how a prevention focus triggers defensive behavior ( Higgins 1997) , making consumers less willing to make

13 13 (Tsai et al. 2011) , open commercial emails (White et al. 2008) purchases , and divulge personal information (Culnan 2000; John, Acquisti, and Loewenstein 2011; Phelps et al. 2000; Singer, . In the realm of behavioral advertising, we suggest that this Hippler, and Schwarz 1992) inhibition manifests as ad avoidance ng from engag ing with the ad or the product it : refraini promotes . The reduction in ad effectiveness resulting from revealing unacceptable information H2: flows will be driven by consumers’ relative concern for privacy over their desire for the personalizatio n that such targeting affords. Unlocking the Upside of Ad Transparency hypotheses thus far have focused on the downside of ad transparency. Specifically, Our we predict that unacceptable ad transparency backfires ( by reducing ad effectiveness), but that acceptable ad transparency may not in and of itself increase ad effectiveness relative to no transparency. However, we also wanted to investigate: when might increase ad ad transparency e suggest that trust in the platform on which the ad is displayed (e.g., Facebook) effectiveness? W moderates the effect of ad t ransparency on ad effectiveness (our predictions so far pertain to the contexts in which consumers’ feelings about the ad platform are relatively neutral —perceived by consumers as neither particularly trustworthy nor untrustworhty). W hen acceptable information flows are revealed on trusted platforms increase ad , we predict that ad transparency may effectiveness . This prediction is predicated on the idea that the trustworthiness of the ad platform has spillover effects on consumers’ reactions to the advertis ements it displays . In support of this claim, consumers’ assessments of trustworthiness in the digital space tend to be generalized (Pavlou, Liang, and Xue 2006; Stewart 2003) ; if a user trusts the Facebook brand, her trust

14 14 —including ads and the applies more expansively to the Facebook website and its features transparency messages that accompany them . Therefore, we expect ad transparency to have a different impact on ad effectiveness as a function of consumers’ individual levels of trust in the . ad platform and generates a High consumer trust is associated with feelings of safety and reliability, willingness to engage (Chaudhuri and Holbrook 2001; Doney and Cannon 1997) . Thus, when -adhering ad transparency practices are displayed, their consumers trust an ad platform and norm —and, in turn, the effect on ad effectiveness response —will be most favorable. However, when -violating pract ices are displayed, their trust is violated; consumers trust an ad platform and norm -violating transparency in this case, consistent with our previous predictions, reactions to norm messages —and the subsequent effect on ad effectiveness —will be negative. Low consumer trust is associated with skepticism and a belief that firms are out to maximize their own self -interest (Bleier and Eisenbeiss 2015; Wang Benbasat 2005) . In and these cases, we suggest that ad transparency : —of any kind—is unlikely to overcome low trust even acceptable targeting practices will be met with the skepticism accompanying dis trust. Consistent with this idea, previous work has shown that suspicion of ulterior motives leads consumers to discount firms’ benevolent acts, such as charitable giving (e.g., Forehand & Grier 2003; L Canli & Schwarz 2006). in-Healy & Small 2013; Newman & Cain 2014; Yoon Gurhan- In sum, with respect to the relationship between trust, ad transparency, and ad effectiveness, w e predict: H3a : When consumers trust the ad platform, a d transparency that adheres to information flow norms will increase ad effectiveness .

15 15 b d transparency will decrease ad H3 : When consumers distrust the ad platform, a effectiveness regardless of whether it adheres to information flow norms. FIGURE 1 OVERVIEW OF STUDIES As a prerequisite to testing hypothese s 1 a-1b, we first generated a list of ways firms obtain and use consumers’ personal information in generating targeted ads, and then measured consumer acceptance of these practices (study 1). A factor analysis identified two key dimensions of consumer acceptability of information flows. Specifically, acceptability varies to whether the personal information used to generate the ad was 1) obtained within according versus outside of the website on which the ad appears ; and 2) explicitly stated by the consumer versus inferred by the firm . Next, we conducted confirmatory experiments. Studies 2 and 3, lab experiments that simulate behavioral targeting, tested wh ether the dimensions identified in study 1 indeed pre dict ad effectiveness (hypotheses 1 a-1b ). Studies 2 and 3 additionally tested the mediating role of consumers’ relative concern for privacy over their interest in personalization (hypothesis 2). The fina l set of studies speaks to the upside of ad transparency (hypothes es 3 a-

16 16 . Specifically, s tudy 4 3b) shows that the effect of ad transparency on ad effectiveness is information flows will be particularly -accepted moderated by trust: disclosures of consumer effective when ad platform . Studies 5a and 5b, conducted within a context in consumers trust the which trust is present (a loyalty program website), demonstrate the upside of transparency in the field. Guided by marketing practice and previous research, we captured our dependent ad effectiveness —in different, but converging, ways across experiments: self measure— -reported purchase interest ( Summers, Smith, and Reczek 2016) in studies 2- 4 and clicking on the ad (Bleier and Eisenbeiss 2015; Tucker 2014) in studies 5a -5b. STUDY 1: IDENTIFYING DIMENSIONS Study 1 was an inductive study. We compiled a list of practices that firms use to generate hem to be. behaviorally targeted ads and measured how acceptable consumers found t Procedure Item generation. We compiled a list of the ways in which the two largest online advertising platforms, Google and Facebook, allow advertisers to employ users’ personal data to generate ads (Facebook 2014; Google 2014) . We supplemented this list by asking a group of Facebook users ( N = 79, 54% male; M to cull the ad transparency text = 22.9, SD = 4.15) age provided in the first ad displayed in their Facebook newsfeed (see appendix B for exact

17 17 were aware , based on industry procedure). Finally, we added additional practices of which we experiences. This effort resulted in 30 items (appendix C). = 35.13, = 149, 52.3% male; Item acceptability. N Participants ( SD = 1 0.14) were M age Facebook generates personaliz told that “ ed advertisements for their users using various methods” and were presented with the 30 ad practices in random order. For each item, participants rated Facebook should show me the extent to which they agreed with the statement: “ advertisements practice] ” (1 = strongly disagree; 7 = completely agree). In this and all other studies, based on [ we report all independent and dependent variables and data exclusions. We recruited participants by foll owing the minimum threshold of 100 participants per cell (thoug h we u sed smaller pre- specified thres holds for pre -tests) . We did not analyze data until we finished collecting the pre - specified number of participants. Stimuli and data (with the exception of the field data collected -b) are available at osf.io/4fjqr in studies 5a . Results . To prepare for factor analyses, we began by performing the Kaiser – Factor Structure Meyer –Olkin test of sampling adequacy and Bartlett’s test of sphericity on the 30 items. The Kaiser –Meyer –Olkin measure of sampling adequacy was wit hin an acceptable range (total matrix sampling adequacy = .9 p < .001), 0), and Bartlett’s test of sphericity was significant ( suggesting that our data were appropriate for factor analyses. We then conducted a principal axis factor analysis with Varimax rotation, suppressing coefficients whose absolute values were below .5. This analysis suggested five factors, accounting for 74.3 % of the total variance. One of these factors included six items that did not reveal a clear pattern, and one item did not fit any of

18 18 , we excluded these the factors. Following the approach used in past research (e.g., Aaker 1997) items and conducted a principal axis factor analysis again. This revealed a four -factor structure all items on each factor were . iance. The loadings of accounting for 75.2% of the total var 5 or higher (absolute value). or higher (see appendix D for factor Cronbach’s alphas for each factor were .86 loadings). We named the four factors: information obtained within -website, information obtained -website, attributes stated by the consumer, and attributes inferred by the firm. The four cross —whether the personal factors thus represent the four ends of two dimensions information used was 1) versus outside of the website on which the ad appears; to generate the ad obtained within by the consumer versus inferred by the firm and 2) explicitly provided . Participants thought that firms should not use information obtained cross -websites ( M cross - (148) = t = 1.52; SD = 4.58, 80) relative to that obtained within = 1. SD = 3.05, M -website ( within 10.61, p < .001). Participants also thought that firms should not make inferences about their stated on ones that the consumer had = 1.70) relative to relying = 3.11 , SD ( attributes M inferred ( M = 4.08, < .001). = 1.88; t (148) = - 7.30, p SD stated Discussion In sum, study 1 revealed two dimensions that predict most of the variance in how consumers perceive information flow norms —specifically, perceived acceptability varies depending on whether the personal information used to generate the ad was 1) obtained within versus outside of the website on which the ad appears; and 2) explicitly stated by the consumer versus inferred by the firm. These dimensions align with offline norms of information sharing.

19 19 first dimension recollects the information flow norms described in example 1 Indeed, the : generating an ad based on consumer information obtained from a different website is akin to talkin g behind someone’s back. Similarly, the second dimension recollects the information flow making an overt inference about someone can be taboo. norms described in example 2: Furthermore, these inductive insights are broadly consistent with Smith, Milberg and Burke d unauthorized secondary use of personal information as one of the key (1996), which identifie drivers of privacy concerns in organizations. -WEBSITE INFORMATION STUDY 2: REVEALING WITHIN VERSUS CROSS FLOWS Study 2 tested the effectiveness of revealing within versus cross -website information flows . We predicted cross -website information flows would decrease ad that revealing effectiveness relative to the within -website flows and also relative to no transparency (hypothes es 1 a-1b ). Study 2 also assessed whether the effect of ad transparency on ad effectiveness is mediated by the differential activation of consumers’ concerns for privacy over their interest in personalization (hypothesis 2). P articipants were shown an ad for a bookstore. Between -subjects, participants were either given no information on why they were seeing the ad (baseline), or one of two transparency messages (i.e., that the ad had been generated either based on information obtained within -platform versus cross -platform). To incr ease the credibility of this manipulation and to simulate real behavioral targeting, all participants first engaged in an online movie browsing task.

20 20 Procedure = 449, 48% male; M Participants ( = 35.9, SD = 11.7) were recruited from Amazon’s N age and assigned to one of three conditions: a baseline condition or one of two transparency mTurk within conditions ( cross -website ). Participants were informed that the study consisted -website or of two parts. Part 1. Participants were informed that they would first be asked to browse movies. Between -subjects, we manipulated whether they browsed movies within the website on which they would subsequently receive an advertisement, or on a different website. W ithin -website participants encountered everything on a sing le browser window, while cross -website participants encountered two distinct browser windows. Specifically, participants in the within- website [cross -website] condition were told: “You will be browsing movies within [outside of] this survey platform. Plea se click on any movies that catch your eye to learn more about them, and feel free to check out as many movies as you would like. We will give you a minute to do this task, but you can spend more time if you’d like. When you are done browsing, please proce ed to the next page [When you are done browsing, please return back to this page and proceed to the next page].” In the within -website condition, participants saw a list of 10 movies on the same page as the above instructions (e.g., Logan, La La Land; appendix E). When they clicked a title, they saw a short description for the movie. In the cross -website condition, participants clicked on a link that took them to an external site. Specifically, upon clicking the link , a new browser page opened, displayi ng the same web page as that for the within -website condition (the only

21 21 -website participants, this webpage was clearly external to the difference being that for the cross ; appendix E). survey platform as opposed to embedded within the survey To keep partic ipants’ experiences consistent, participants in the baseline condition also engaged in this movie browsing task, with half randomized to browse movies within- website , -website . In our results and half randomized to browse movies cross is , we collapse across th control factor. In addition, to ensure participant engagement in the browsing task (i.e., part 1), all participants were told upfront that they may be asked to answer questions about the browsing experience. Part 2. Participants were informed that part 2 entailed evaluating an ad. Specifically, they read, “We will be showing you an ad that is targeted for you. Targeted advertising is a type of advertising that allows companies to reach consumers based on various traits.” Participants waited for five seco nds, during which we were ostensibly generating an ad for them, and then saw an ad about a fictitious online bookstore, UBooks.com. The transparency conditions included a message next to the ad. Those in the within- platform condition saw “You are seeing this ad based on the products you clicked on while browsing our website (i.e., within this survey ” and those in the cross -website condition saw, “You are seeing this ad based on the platform) products you clicked on while browsing a third- party website (i.e. , outside of this survey platform).” In the baseline condition, there was no message (see appendix F for stimuli). The language used for the two transparency conditions in this and subsequent studies was adapted from the actual text that firms use when lis ting their ad practices (see appendix G for an example). Ad effectiveness. Participants indicated the extent to which they agreed with the statements : “I am interested in visiting the website for UBooks.com” and “I am interested in

22 22 on a 7- point scale Strongly disagree to 7 = S trongly buying products from UBooks.com” (1 = ( We averaged these items to create a composite score of ad effectiveness agree). α = .92). Privacy over personalization. To assess our proposed mediator, we asked participants: “In order to provide more personalized recommendations for you, marketers need to gather more information about you. In other words, when receiving an advertisement, there is a tradeoff ation. Upon between maintaining your privacy and enjoying the benefits of greater personaliz on a seeing the above ad, which factor is more important to you when evaluating a targeted ad?” point scale (1 = Privacy is more important to me to 10 = Personalization is more important to 10- coded this item such that higher sc ores represented higher relative concern for me). We reverse- privacy. Pretest s We conducted two pretests to ensure that ad practices disclosed in the within -website and cross -website conditions 1) were indeed perceived as acceptable and unacceptable respectively and 2) did not differ in perceived accuracy. Pretest one. To examine whether -website and cross - ad practices disclosed in the within conditions were indeed perceived as acceptable and unacceptable, respectively , w e website and = 37 .0, SD = 11.5) N = 92, 35.9% male; M recruited a separate group of participants ( age ertising is a type of advertising that allows companies to reach informed them that “Targeted adv consumers based on various traits.” Participants were randomly as signed to one of two conditions: within -website and cross -website. We asked participants in the within -website condition to suppose that they saw an ad that was “based on the products you clicked on while

23 23 -website condition browsing our website (i.e., within this survey platform)”; those in the cross were asked to suppose that they saw an ad that was “based on the products you clicked on while party website (i.e., outside of this survey platform).” Participants rated how browsing a third- point scale (1 = N ot at all acceptable they found the revealed advertising practice on a 7- website condition rated their acceptable to 7 = V ery acceptable). As intended, those in the within- = -website condition ( M significantly more acceptable than those in the cross ad practice as within SD = 1.73; M = 3.07, SD = 1.70; t (90) = 5.16, 4.92, < .001). p cross To ensure that perceived accuracy did not differ across the within- website Pretest two. -website conditions, we recrui ted participants ( N = 82, 41% male; M = = 36.1, SD and cross age 13.0) ts were and gave them the same information about targeted advertising as above. Participan website, cross -website), and saw the randomly assigned to one of two conditions (i.e., within- same scenario as above. Then, they rated how much they thought the ad would fit their needs and ot at all to 7 = V ery much). There was no stat istically significant difference between wants (1 = N . = .94) p = 4.55, SD = 1.60; M (80) = .07, = 4.53, SD = 1.15; t conditions ( M cross within Results Ad effectiveness. A one -way ANOVA revealed a significant main effect of transparency F (2, 446) = 14.05; p < .001). on ad effectiveness ( As predicted, ad effectiveness was reduced in the cross M -website condition ( = 2.83, SD -website ( M = 3.56, = 1.77) relative to both the within SD = 1.73; t (298) = -3.67, p < .001) and baseline conditions ( M SD = 1.72; t (293) = 5.14, = 3.87, p < .001). The latter two conditions did not differ ( t (301) = 1.54, p = .13). Privacy over personalization . Results of the process measure were similar ( F (2, 446) =

24 24 p -webs ite condition ( M = 3.12; = .045). Specifically, privacy concerns were higher in the cross SD = 2.43) relative to both the within- website ( M = 6.75, SD = 2.49; t( 298 ) = - 7.47, p = 2.51, .01) M = 6.97, SD = 2.64; t( and baseline conditions ( ) = 1.67, p = .09). The latter two 293 conditions did not differ ( t (301) = - .75, p = .46). Mediation. The d - versus cross- website conditions on ad ifferential impact of the within effectiveness was mediated by participants’ relative concern for their privacy over their interest in personalization: A 5,000 sample bootstrap analysis using PROCESS Model 4 (Hayes 2013) a significant indirect effect (b = - .17, SE = .08; 95% confidence interval: [ -.23, - .02]). indicated Discussion Study 2 showed that ad transparency reduced ad effectiveness when it revea led cross - website tracking — an information flow that consumers deem unacceptable, as identified by our inductive study 1. Consistent with hypothesis 2, ad transparency that revealed unacceptable information flows heightened concern for privacy over interest in personalization, reducing ad effectiveness . STUDY 3: REVEALING STATED VERSUS INFERRED INFORMATION FLOWS Study 3 tested the effectiveness of revealing that an ad is based on stated versus inferred attributes. We predicted that ads based on inferred information would decrease ad effectiveness relative to stated attribute transparency and relative to no transparency (hypothese s 1 a-1b) . Like study 2, s tudy 3 also predicted privacy over personalization concerns to mediate the relationship

25 25 . Participants between ad transparency and ad effectiveness were shown an ad for an online art gallery. Between -subjects, participants were either given no information on why they were seein g the ad (baseline), or one of two transparency messages (i.e., that the ad had been by the consumer versus inferred by the firm generated either based on information stated ). To increase the credibility of this manipulation and to simulate real behavioral targeting, all participants first completed an online shopper profile. Procedure recruited from = 12.45) were SD = 37.66, = 348, 45% male; N Participants ( M age Amazon’s mTurk and shown one of three different versions of an ad varying in transparency: no transparency (baseline condition), or one of two transparent ads —one revealing it had been targeted based on stated attributes , and one revealing it had been targeted based on inferred attributes. First, participants were told: “We are a group of research ers in marketing. Today, we will be showing you targeted advertisements. Targeted advertising is a type of advertising that allows companies to reach consumers based on various traits.” Next, participants were asked to build an online shopper profile by fi lling out a form. This form listed three demographic questions that were randomly chosen from the following: gender, age, highest level of education completed, current relationship status, and student status. On the next screen, participants were further informed: “In order to provide targeted online advertisements for you, marketers can rely on the information that you’ve given them voluntarily (i.e., what you’ve stated about you), or infer

26 26 ernet connection and things other types of information about you based on things like your Int that you click on (i.e., what they infer about you).” After completing an unrelated task, participants were presented with one of three different versions of an ad for an online art gallery store. Manipulation. The transparency conditions included a message next to the ad: “You are seeing this ad based on your information that you stated about you” (stated attribute condition); “You are seeing this ad based on your information that we inferred about you” (inferred attribute condition). In the baseline condition, there was no message (see appendix H for stimuli). Participants indicated the extent to which they agreed with the following Ad effectiveness. statements on a 7 : “I am interested in visiting the website fo r UGallery.com” and “I -point scale am interested in buying products from UGallery.com” (1 = Strongly disagree to 7 = Strongly agree) . We averaged these items to create a composite ad effectiveness score ( α = .93). Privacy over personalization. We measured the relative activation of privacy concerns versus interest in personalization using the same item as in study 2. Pretest s As in study 2, w e conducted two pretests to ensure that ad practices disclosed in the stated attribute and inferred attribute conditions 1) were indeed perceived as acceptable and unacceptable, respectively and 2) did not differ in perceived accuracy. Both pretests used the same design s as in study 2’ s pretests, with the exception that participants read about stated or inferred information flo ws.

27 27 whether ad practices disclosed in the stated attribute and inferred Pretest one. To examine attribute conditions were indeed perceived as acceptable and unacceptable, respectively, we M = 13.3) to one of two = 38.7, SD randomly assigned participants ( N = 70, 45.7% male; age . As intended, those in the stated condition rated their ad practice conditions (i.e., stated, inferred) M = 1.57; = 4.55, SD to be significantly more acceptable than those in the inferred condition ( stated SD = 1.78; t (68) = 2.49, p = .02). = 3.54, M inferred Pretest two. To ensure that perceived accuracy did not differ across the stated and inferred conditions, we randomly assigned participants ( N = 11 4, 40.4% male; M = 34.7, SD = age . There was no statistically significant 11.6) to one of two conditions (i.e., stated, inferred) = 1.45; SD SD = 1.30; M (112) = 1.26, = 4.54, t = 4.86, M difference across conditions ( inferred stated p = .21) . Results Ad effectiveness. -way ANOVA revealed a significant main effect of trans parency A one F (2, 345) = 4.12; p = .02). As predicted, ad effectiveness was reduced in the on ad effectiveness ( = 3.10, = 2.52, SD = 1.35) relative to both the stated attribute ( M condition ( M inferred attribute =1.82; (232) = 2.77, p = .01) and baseline conditions ( M SD t =2.96, SD = 1.62; t (230) = = 2.22, p t (228) .03). The latter two conditions did not differ ( .63, p = .53 ). = - Privacy over personalization . Results were similar for privacy ( F (2, 345) = 2.83; p = .06). Specifically, privacy concerns were higher in the inferred attribute M = 3.46, SD condition ( = 2.26) relative to both the stated attribute ( M = 4. 15, SD =2.70; t( 232 ) = 2.12, p = .04) and t( baseline conditions ( = 4.07, SD = 2.30; M 230 ) = 2.05, p = .04). The latter two conditions did

28 28 t = - .23, p = .82). not differ ( (228) The d ifferential impact of inferred versus stated conditions on ad Mediation. effectiveness was mediated by participants’ relative concern for their privacy over their interest 5,000 sample bootstrap analysis using PROC in personalization. A ESS Model 4 (Hayes 2013) suggested a significant indirect effect (b = - -.35, - .01]) . .17, SE = .08; 95% confidence interval: [ Discussion Study 3 showed that ad transparency reduced ad effectiveness when it revealed that the ad was based on consumer attributes inferred by the firm —an information flow that consumers deem unacceptable, as identified by our inductive study 1. Consistent with hy pothesis 2, ad transparency that revealed unacceptable i s increased co ncern for privacy over nformation flow , thus reducing ad effectiveness. interest in personalization Finally, note that in s tudies 2 and 3, ads that revealed acceptable information flow s performed just as well as those in the baseline condition. T o examine whether and when ad transparency could boost ad effectiveness, above and beyond the baseline control, t he next set of studies were conducted on ad platforms for which consumers have pre (dis)trust . -established Thus , these studies test whether trust can unlock the benefits of disclosing ( versus not disclosing ) acceptable information flows. STUDY 4: THE MODERATING ROLE OF TRUST

29 2 9 Study 4 tested hypothese s 3 a-3b by examining the moderating role of individual trust. To do so, we measured consumer trust in Facebook, and differences in ad platform examined whether the level of trust in the ad platform (i.e., trust in Facebook) interacted with ad transparency .e., the degree to which an ad practice adhered to norms of information flows . In —i addition, we again measured our mediator in study 4, thus providing an additional test of hypothesis 2. Procedure N = 46 2, 52.5% male; M Participants ( = 33.5, SD = 10.2) who indicated that they had a age Facebook account were recruited from Amazon’s mTurk. We first assessed how much participants trusted Facebook by asking them to rate the extent to which they agreed or disagreed with the statements ( Chaudhuri and Holbr ook 2001): “I trust Facebook”, “I rely on Facebook”, ,” each on a 7 -point scale (1 = Strongly “Facebook is an honest brand”, and “Facebook is safe disagree to 7 = Strongly agree). We averaged these items into a composite trust measure (α = .88). Next, all p articipants were asked to log into their Facebook accounts. To ensure that they were logged in, we asked a question that required them to be logged in to know its answer (appendix I ) and participants could not advance in the survey until they correctly ans wered this question. Next e guided participants to locate the first ad in their newsfeed and to click on its , w “Why am I seeing this ad” message (appendix B). Participants were asked to copy and paste the message and the company name. Forty -eight participants failed to accurately complete this step; we excluded them from analysis ( however, the results hold when they are included) .

30 30 , participants coded the of their t ransparency message with respect to our Next content independent variable of intere st: the degree to which the ad practice adhered to norms of information flows. Specifically, participants indicated “Yes ,” “No ,” or “I’m not sure” to each of the following four statements: This ad was presented to me based on 1) my activities within party websites (outside of Facebook), 3) the information that Facebook, 2) my activities on third- I stated about myself on Facebook, and 4) the information that Facebook inferred (i.e., guessed) about me. This coding generated a “transparency score” for each ad, with higher transparency scores reflecting more acceptable practices. Specifically, we assigned a value of +1 to “Yes” respons es to the first and third statements, which depict acceptable information flows (i.e., -website tracking and stated attributes, respectively) within 1 to “Yes” responses , and a value of - to the second and fourth statements, which depict unacceptable informa tion flows (i.e., cross - website tracking and inferred attributes, respectively). We assigned a neutral value of 0 to all “No” and “I’m not sure” responses ). Following an ( See appendix J for descriptive statistics to create a transparency score approach in psychology literature, we then summed these values for each observation (e.g., Costa and McCrae 1980) . Finally, participants answered a series of questions based on the ad they had received. First, they rated how much they were interested in : 1) visiting the company website, 2) spending their money at the company, 3) clicking the like button for the company’s Facebook page, and 4) spending their money on the advertised pro duct on a 7 -point scale (1= S trongly disagree to 7 = Strongly agree) averaged these items into a composite ad effectiveness measure (α = .93). . We As in studies 2 and 3, participants indicated how much they were concerned about privacy over personalization. We also administered a trait measure of privacy concern ( α = .85; Malh otra, Kim, and Agarwal 2004) , which did not interact with our independent variables .

31 31 Results Ad transparency and ad A linear regression revealed that transparency effectiveness. scores < .001) . In other words, significantly predicted ad effectiveness (b = .31, SE = .09, p participants who received an ad that revealed acceptable practices were more likely to engage relative to those receiving an ad that revealed unacceptable practices. with the ad Privacy over personalization. A linear regression also revealed that transparency scores p significantly predicted privacy concerns (b = .25, SE = .11, .03). That is, participants who = received an ad that revealed acceptable practices were less likely to be concerned about privacy over personalization relative to those receiving an ad that revealed unacceptable practices. Mediation. The impact of ad transparency on ad effectiveness was mediated by participants’ relative concern for their privacy over their interest in personalization . Consistent revious studies, a 5,000 sample bootstrap analysis using PROCESS Model 4 (Hayes 2013) with p b = - .17, SE = .08; 95% confidence interval: [ -.23, - .02]). suggested a significant indirect effect ( Moderation. Next, we examined whether the relationship between transparency scores and ad effectiveness was moderated by platform trust. A linear regression revealed a significant interaction between trust and transparency score (b = .15, SE = .06, p = .02). To decompose this interaction, we followed procedures recommended by Spiller, Neyman technique to identify Fitzsimons, Lynch, and McClelland (2013) and used the Johnson- the range(s) of trust for which the simple effect of revealing acceptable information flows was significant. This analysis revealed a significant effect of ads with acceptable transparency on ad effectiveness for any model with trust greater than 3.56 (BJN =.18, SE = .09, p = .05), but not for

32 32 were more likely any model with trust less than 3.56. In other words, users who trust Facebook than those who dis trust to engage with an ad that revealed acceptable information flows Facebook. See figure 2. We also conducted a moderated mediation analysis to simultaneously test moderation by trust and mediation by privacy concern. A 5,000 sample bootstrap analysis showed that the index of moderated mediation excluded zero (95% CI = [.002, .04]), suggesting a significant indirect effect ( Hayes, 2013). FIGURE 2 FO RM AD TRANSPARENCY AND AD PLAT TRUST 7 6 5 4 3 Ad Effectiveness 2 1 1 2 4 5 6 7 3 Trust BENEFIT OF AD TRANSPARENCY IN THE FIELD STUDIES 5A- 5B: THE (versus not revealing Study 4 suggest ) acceptable ed that the benefit of revealing the effect information flows emerges in contexts in which trust is present . Our final studies tested of ad transparency on ad effectiveness under favorable circumstances: when consumers both -adherent transparency messages. trust the platform and are shown norm

33 33 To do this, we conducted two different field experiments with a trusted platform. Previous research has shown that loyalty programs are a context in which consumer trust is high; members typically have higher trust than non- members ( Ashley et al. 2011; Chaudhuri and Holbrook 2001; García Gómez, Gutiérrez Arranz, and Gutiérrez Cillán 2006) . Thus, studies 5a loyalty program point -redemption and 5b were conducted within the context of two different s, and sought to replicate the beneficial effect of ad transparency in the field. Specifically, site s disclosed that personal these studies sought evidence of a boost for ad transparency when it wa information was obtained within the point -redemption site (i.e., within , study 5a) or -website explicitly by the user (i.e., stated attributes , study 5b). Field Setting Description We collaborated with a provider of a white label online platform to companies that operate rewards programs . For example, if a consumer is enrolled in a hotel membership program and accumulates points, he/she can use those points to pur chase various items on the hotel’s rewards redemption sit e, serviced by the provider with whom we partnered . The provider therefore customizes the platform to the given loyalty program and features that program’s own branding. Studies 5a and 5b involved two companies that use this reward site, one per experimen t. Study 5A Procedure

34 34 rewards redemption site were randomly assigned to one of Users of a given company’s website. The experiment was live for two conditions: baseline (no transparency) versus within- ted of 9,079 unique users who logged into the two weeks in the spring of 2017; the sample consis rewards site during the two -week period. As a user browses products on this site, a sidebar is present, which suggests items for the user to buy (appendix K ). These items are a form of behaviorally targeted advertising; they are tailored based on the given user’s behavior. Our experiment leveraged this sidebar feature by -website condition, the sidebar was titled “Recommended manipulating its title: in the within based on your cli cks on our site;” in the control condition, it was simply titled “Recommended.” In both studies 5a and 5b, the primary outcome was the behavioral measure most proximal to the manipulation: the propensity to click on the recommended items. As exploratory measures, the rewards website also provided data on additional outcomes: the number of seconds spent on the pages of recommended items, and revenue generated from the recommended items . Pretests We conducted two pretests: one to confirm that trust in a given brand is generally higher members ( consistent with Ashley among members of the brand’s loyalty program relative to non- et al. 2011; Chaudhuri and Holbrook 2001; García Gómez, Gutiérrez Arranz, and Gutiérrez Cillán 2006) ; the other to assess whether the sidebar was in fact perceived to be a targeted ad. Pretest one . To assess whether the field setting was indeed a context in which trust is = 37.5, = 12.08) were randomly SD M present, participants on mTurk ( N = 164, 40.2% male; age assigned to one of two conditions: loyalty versus non- loyalty. Those in the loyalty [non- loyalty]

35 35 condition were told: “Please name one company that meets the following requirements: 1) you is company at least once over the past six months and 2) you are have made a purchase from th then rated the extent [are not] enrolled in this company’s loyalty rewards program.” Participants point scale (1 = Not at all; 7 = V to which they trusted the company they had listed on a 7- ery hose in the loyalty condition ( M = 5.99, SD = 1.01) trusted the company they listed much). T loyalty condition ( M = 5.34, SD = 1.38, t (162) = 3.41, p significantly more than those in the non- = .001). . To assess whether the sidebar was in fact perceived to be a targeted ad, Pretest two SD 4.6, ere shown a screenshot = 11.1) w = 3 = 153, 50.6% male; N participants on mTurk ( M age of the point redemption site, with the product sidebar highlig hted. Participants were told: argeted advertising is a type of advertising that allows companies to reach consumers based on “T ey agreed with the statement: “T he various traits.” They were asked to rate the extent to which th highlighted portion comes across as a targeted advertisement” (1 = S trongly disagree; 7 = Strong ly agree). Results indicated that the sidebar was indeed perceived as a targeted ad; the mean rating was statistically significantly higher than the scale midpoint ( M = 5.10, SD = 1.40, t (152) = 9.79, p < .001). Study 5A Results We conducted a mixed effects regression analysis to evaluate the effect of revealing acceptable information flows (in this case that the consumer was targeted based on information obtained within- website) on the propensity to click on recommended items. To account for individual -level differences, we included participant ID as a subject variable. As predicted,

36 36 -website condition were more likely to click on recommended items participants in the within than those in the baseline condition ( = .10, SD = .001 versus M = .001; = .09, SD M baseline within -Square = 100.56, p < .001). Participants in the within -website Wald Chi condition also spent more time (seconds) on the recommended products’ pages ( M = 8.52, SD = .14 versus within F M SD = .14; and in turn, more money on the (1, 279,006) = 117.73, p < .001) = 6.35, baseline recommended products ( M (1, 279,006) = = .22, SD = .02 versus M F = .16, SD = .02; baseline within p 4.24, = .04). Study 5B Procedure Study 5b was run concurrently with study 5a with another company that uses the rewards site. Users were randomly assigned to one of two conditions: baseline (no transparency) and stated attribute . The sample consisted of 1,862 participants who logged into the rewards site during the two -week period. As in study 5a, we leveraged the sidebar feature by manipulating the title of the sidebar . In the stated attribute ndition, the sidebar was titled, “Recommended based on what you’ve co shared with us;” in the control condition, it was simply titled “Recommended.” Study 5B Results We conducted a mixed effects regression analysis to evaluate the effect of revealing acceptable information flows (in this case that the consumer is being targeted based on information they explicitly stated about themselves) on the propensity to click on recommended

37 37 -level differences, we included participant ID as a subject items. To account for individual variable. As predicted, p articipants in the stated attribute condition were more likely to click on = .003 versus e condition ( recommended items than those in the baselin SD M M = .12, baseline stated SD = .002; Wald Chi -Square = 76.06, p = .08, < .001). Participants who were exposed to the stated attribute products’ pages condition also spent more time (seconds) on the recommended = 5.98, M = 8.31, SD = .43 versus ( < .001). There p SD = .42; F (1, 28,136) = 14.98, M stated baseline was no impact of condition on revenue ( M = .04; = .11, SD = .04 versus M SD =.17, baseline stated (1, 28,136) = .86, = .35). F p Discussion Studies 5a and 5b, conducted within a loyalty program point redemption website —a field setting in which ad platform trust wa s presumed to be , and which our pilot data suggested was , present can increase ad effectiveness. —suggested that revealing acceptable information flows However, we note that because we did not manipulate trust, alternative explanations are possible. GENERAL DISCUSSION In response to growing pressure from consumers and regulators, the number of firms adopting ad transparency is on the rise. Our research explores how ad transparency affects ad effectiveness, and is based on the premise that it depends on the perceived acceptability of the revealed practice. Because this territory is largely uncharted, we began with an inductive approach to capture how acceptable (or not) consumers perceive various data collection and use

38 38 consumers deem information flows acceptable is driven practices. Study 1 revealed that whether side of by the extent to which the ad is based on 1) consumers’ activity tracked within versus out the website on which the ad appears and 2) attributes explicitly stated by the consumer versus inferred by the firm (the latter of each pair is deemed less acceptable ). Next, we conducted two dimensions identified in study confirmatory experiments. Studies 2 and 3 confirmed that the . These studies also documented that 1 indeed predict consumers’ response to ad transparency consumers’ relative concern for privacy over their interest in personalization mediates the relationship between the acceptabili ty of the revealed practice and the effectiveness of the ad. Finally, we demonstrated the moderating role of platform trust: companies can reap the benefits of ad transparency that reveals acceptable information flows when consumers trust the ad platform. We contribute to the literature on digital advertising. Existing research has focused on the effectiveness of online ads when consumers are unaware of underlying ad practices; however, as recent articles suggest, consumers are becoming increasingl y conscious of and concerned about how marketers may be using their information (Lambrecht and Tucker 2013; Morey et al. 2015) . Recent data breaches involving major companies —from Walmart to Ashley Madison to Equifax —have exacerbated this concern. Thus, it is critical to understand how consumers’ willingness to engage with online ads is affected by their awareness of the data practices used to deliver such ads. In addition, much of the literature on behaviorally targeted ads has been correlational; our work adds to the small but growing body of experimental work on the topic (e.g., Bleier and Eisenbeiss 2015; Schumann, von Wangenheim, and Groene 2014; Summers, Smith, and Reczek 2016) . We also contribute to the literature on privacy. Privacy concerns have been shown to play

39 39 e.g., Phelps et al. 2000; John et al., a role in consumers’ willingness to divulge information ( dvertising is underexplored. 2011; Hofstetter, Rueppel, & John, 2017) , but their role in targeted a —making salient to -targeted Particularly as advertising becomes increasingly specific and well consumers that they are being tracked (e.g., seeing an ad featuring a product you just browsed; ; Paul 2017) — encountering an ad based on sensitive or private web searches you have done privacy concerns may likewise increase, with possible downstream consequences for disclosure, -firm relationships. On the other hand, given research suggestive search behavior, and consumer of consumers’ Loewenstein 2012), capacity to adapt to privacy violations (Acquisti, John, and and the stickiness of those violations, the opposite may occur, with consumers becom ing As we describe below, f desensitized over time. uture research could explore the complex and -evolving relationships between transparency, privacy concerns, and marketing outcomes. ever Key l The following limitations qualify the contribution of our work. First , we imitations. focused on one form of ad transparency —th e type that conspicuously re veals information to consumers . Doing so enabled us to test the relationship between ad transparency and ad n an internally valid way ) effectiveness i ( i.e., independent of consumer demand for transparency act of answering consumer s and regulators’ and as a result, may help marketers predict the imp greater transparency. However, these advantages come with a t least one distinct calls for disadvantage: reduced external validity, as t oday’s standard practice requires consumers to proactively seek out information ( e.g., Facebook users have to click on the “Why am I such seeing this ad?” button to view ad generation practices ). As we discu ss below, our hope is that our findings serve as a stepping stone for researchers to further investigate this topic, incl uding the type of ad transparency that more closely mirrors today’s practices (i.e., the type that relies on consumers to seek out information), and drivers of consumer demand for it.

40 40 Second, because we did not directly manipulate trust in studies 5a and 5b, there may have been other factors that led us to observe a positive relationship between ad transparency and ad alternative explanations are possible, future research could establish this effectiveness. Because relationship more definitely by manipulating trust —a method complementary to the approach taken in study 4, in which we measured individual differences in platform trust to test i ts moderating role. The goal of this manuscript -level: to examine consumer was intentionally broad and high perceptions of the ways firms collect and use personal data, and to identify overarching, widely applicable dimensions that characterize these perc eptions. As such , many questions remain and the topic is ripe for further investigation. Below, we outline what we see as some of the highest priority areas for future research on ad transparency: Moderators of effectiveness. C onsumer response to ad transparency may depend on a variety of yet -untested variables. For example, it may vary as a function of the ad content (e.g., quality, accuracy, or specificity of the ad or product displayed), basis for targeting (e.g., the sensiti vity or confidentiality of the information on which consumer was targeted), language of transparency disclosure (e.g., perceived as trying to inform or trying to persuade; low or high elaboration in describing ad practices ), and impetus for the disclosure transparency disclosure (e.g., voluntary or legally mandated). With respect to the latter, when the disclosure is voluntary, does it matter whether the decision to be transparent is —or is perceived to be —made by the advertiser versus t he ad platform ? Relat edly, w hen we explored the role of trust (studies 4, 5a and 5b), we looked at trust in the platform, as opposed to trust in the advertised brand (and used “neutral” brands that were neither particularly trustworthy nor untrustworthy) . Future research could examine when and why the source of the transparency disclosure matters, and how this

41 41 For instance, if the platform is trusted but the brand in the might interact with trust. , how would consumers react to ad transparency? advertisement is not Future research could also examine ways of counteract ing the negative impact of (without having to change the ways firms generate targeted ads) — for unacceptable transparency . However, example, by enhancing consumers’ perceived control over their personal information d—and not necessarily meaningful — control might affect ad transparency the notion that perceive . More broadly and as we discuss effectiveness points to the potential for consumers to be misled below, more research is needed to understand the consumer welfare implications of firms’ use of consumers’ personal information. Mediators of effectiveness. Our research suggests that the effectiveness of targeted advertising was driven in part by consumers’ relative concern for their privacy ove r their interest relationship between ad transparency and ad effectiveness is in personalization. However, the likely to be multiply -determined and may differ by context. F urthermore, there may be instances in which one’s desire for privacy and for personalization are not in conflict. Thus, w e leave open the possibility that other factors may also mediate the path from transparency to effectiveness. For example, ad transparency may affect the perceived credibility of targeted ads: when told that an ad was based on certain types of information, consumers may be more convinced of the product -consumer fit relative to other types of ad transparency, which could in turn increase consumer interest. On the other hand, learning about certain types of ad pract ices may heighten the degree to which consumers experience identity threat (Branscombe, Schmitt, & Harvey, 1999) . For instance, consider a consumer who learns that she received an ad for Planned Parenthood because of her particular demographic profile . In this case, ad transparency could make her feel reduced to a single membership category, which could in turn decrease her interest

42 42 -targeted ads when they threaten the in the ad. Relatedly, consumers may resist even well expression of valued identities (Bhattacharjee, Berger & Menon 2014). Ad transparency may affect a range of consumer Additional implications for the firm. perceptions and behaviors, beyond the propensity to click on a given ad. Transparency, writ large, has been shown to have positive effects for the firm (Buell, Kim, and Tsay 2016; Buell and Norton 2011) ; ad transparency could have positive downstream effects. For example, given that discl , and that trust plays a role in consumers’ osure breeds trust (John, Barasz, and Norton 2016) acceptance of targeted ads (Aguirre et al. 2015; Bleier and Eisenbeiss 2015) , it is worth investigating how ad transparency might affect consumers’ holistic view of the firm . Relatedly, there may be a positive feedback loop between privacy concerns and trust perceptions, such that enhancing trust may reduce privacy concerns, in turn shifting consumers toward valuing the increased personalization afforded by behaviorally targeted advertising . Another open question is whether encountering ad transparency encourages consumers to seek out more information from In addition, as consumers become more sophisticated in their knowledge of targeted companies. advertisin g, future res earch could investigate other secondary consequences of ad transparency for each of the parties involved . Implications for consumer welfare. Our results also contribute to the debate surrounding the regulation of online privacy (Goldfarb and Tucker 2011) . Proponents of industry self - regulation have been supportive of voluntary firm implementation of ad transparency (Alliance 2014) . Indeed, the AdChoices icon is a collaborative, self -regulatory initiative led by the Digital Advertising Alliance and backed by several leading marketing and advertising associations. These entities tout the benefits of ad transparency in improving consumer welfare (Dienel 2015) , with the presumption that arming consumers with more information will yield better (or a t least,

43 43 -informed) decisions. However, it is conceivable that transparency could also have better based approach to unforeseen adverse effects for consumers. For instance, the information- nsumers’ privacy consumer empowerment stands in contrast to emerging evidence that co -normative factors, making them prone to disclosing information concerns can be affected by non in precisely the contexts in which it may be relatively dangerous to do so (John et al. 2011) . — Furthermore, if transparency (rightly or wrongly) increases consumer trust and complacency become less diligent about protecting their privacy or inadvertently such that consumers policies could be divulging more than they otherwise would—the net benefits of transparency mitigated. Future research should examine the effects of ad transparency on consumer welfare. Ad transparency that is c . T his manuscript onspicuous versus merely available investigated conspicuous ad transparency : after advertisers or ad platforms directly revealed transparency information to consumers, how did consumers respond? Future research might also If left to their own devices, investigate the antecedents of consumer demand for ad transparency. how likely are consumers to seek out this information when it is made available to them ? This propensity could depend on a number of factors, including whether consumers are expecting it to be pleasant versus unpleasant information (Dana , Weber , and Kuang 2007; Loewenstein, Golman , and Hagmann 2017). Relatedly, a re certain types of consumers more likely to seek such information than others? How might such individual differences interact with the impact of ad transparency on ad effectiveness? While some industry surveys have suggested that consumers support ad transparency initiatives and believe it to be important in the online sphere (Dienel 2015) , it is less clear when and why consumers demand such information. Future research c ould elucidate these causal mechanisms. For instance, are consumers more likely t o demand transparency when they see a generic or highly personalized ad? An ad with neutral or sensitive

44 44 content? An ad that they have encountered once or multiple times? And how (if at all) does consumer behavior change after seeking out (versus conspicuously provided with) being transparency information? Future research may better establish the drivers of consumer demand for better transparency. Indeed , this dynamic in supply versus demand for transparency between consumers and one . For example, in food labeling, in response to the concern du jour of firms is a familiar s of consumers (e.g., concerns about subset high fructose corn syrup), firms often respond with e corn syrup” labels). In turn, consumers who did not initially transparency (e.g., “no high fructos to such labels and even update their seek such information may begin to pay attention preferences . Similarly , in the case of ad transparency, many consumers may not currently actively seek this information but start to demand it as transparency p ractices become more common. , If consumer advocates succeed in requiring transparency messages from advertisers we foresee this topic as becoming even more relevant to the current discourse on consumer protection. Trade -off between privacy and personalization. Our studies demonstrated that consumer privacy concerns over interest in personalization mediate the relationship between ad transparency and ad effectiveness, highlighting the importance —to marketers —of understanding how to directly influence these attitudes, and thus influence ad effectiveness. However, the -off has broader significance. Targeted advertising has the privacy versus personalization trade potential to enhance consumers’ online experiences: faced with the choice b etween viewing a website covered in entirely irrelevant ads or highly applicable and interesting ones, consumers would likely prefer the latter. But at what cost? How much personal information —be it demographic, stated preference, or behavioral — are consumers willing to divulge in exchange for

45 45 better personalization? This question lies at the heart of the challenge facing modern advertisers and marketers. For firms, the benefits of personalization might ideally render privacy concerns umers to freely share their own data; however, for consumers, this moot and encourage cons reality still seems far off (Panel 2011; Rainie and Duggan 2016) . Future research examining the relationship between privacy and personalization is therefore highly relevant, not only in the domain of ad transparency and effectiveness, but also when looking at the more holistic relationship between consumer and firm. Dynamic attitudes and practices. Consumer attitudes about the collection and use of personal information have changed over ti me (Affairs 2015; Greenblatt 2013) , and will inevitably continue to change into the future. As devices and data rapidly enable firms to target consumers in increasingly specific ways and new targeting methods are developed, the invasive and unsavory practi ces of today may come to be seen as benign and a cceptable. Moreover, if firms increasingly adopt ad transparency, thereby making targeting practices salient, consumers may adapt to the idea that they are being targeted, perhaps increasing the overall benefit of transparency (and possibly also consumer demand for it). While our overall framework offers flexibility necessary to mold into ever -morphing attitudes and practices, empirical research must keep pace. Conclusion. We opened by noting a growing trend of transparency in online advertising. While many advertisers may be slow to embrace ad transparency, our findings indicate that by consider ing norms of information flows , advertisers can mitigate the effects of exposing practices that consumers deem unsavory. For example, had Target understood and adhered to these norms, it could have avoided the now -infamous case of sending targeted, pregnancy -related coupons based on inferred information. We suggest that in addition to continually refining their

46 46 targeting practices, firms might benefit by also being sensitive to consumers’ attitudes toward the process of generating ads. A perfectly targeted ad can be rendered ineffective if unsavory practices that underlie it are exposed.

47 47 APPENDIX A AD TRANSPARENCY- FACEBOOK

48 48 APPENDIX B STUDY 1 INSTRUCTIONS

49 49

50 50 APPENDIX C STUDY 1: LIST OF AD PRACTICES Items 1. A company’s request to target people like me using the information that I stated on my profile 2. A company’s request to target people like me using what Facebook inferred usage based on my Facebook 3. Another company’s website that I’ve logged in without using my Facebook ID 4. Another company’s website that I’ve logged into using my Facebook ID Facebook advertisements that I click on 5. 6. Facebook groups that I am part of 7. belong to (e.g., school, workplace) Facebook networks that I 8. Facebook pages (of companies, celebrities, etc.) that I have liked 9. Facebook pages (of companies, celebrities, etc.) that I have visited 10. Facebook pages (of companies, celebrities, etc.) that my Facebook friends have li ked Facebook pages (of companies, celebrities, etc.) that my Facebook friends have 11. visited 12. How often I log into Facebook My age that Facebook inferred based on my Facebook usage 13. 14. My age that I stated on my profile 15. My current location that Facebook inferred from my computer’s unique IP address My current location that I stated on my profile 16. My Facebook messaging activities 17. My family members that Facebook inferred from my Facebook usage 18. My family members that I listed on my profile 19. 20. Facebook inferred based on my Facebook usage My gender that 21. My gender that I stated on my profile 22. My past browsing history on another company’s website 23. My past purchase history on another company’s website 24. My past search history on a search engine 25. another company’s website My past visits to 26. My relationship status that Facebook inferred based on my Facebook usage 27. My relationship status that I stated on my profile 28. My sexual orientation that Facebook inferred based on my Facebook usage 29. My sexual orientation that I stated on my profile 30. Other people’s Facebook profiles that I visit * Items excluded from the final analysis : 1, 2, 10- 12, 17, & 30

51 51 APPENDIX D STUDY 1: FACTOR ANALYSIS RESULTS Factor Variance Items Dimension Name Loading Explained Within - website 7 . 17 . 53 % Facebook pages (e.g., companies, 8 1 tracking celebrities) that I have liked .77 Facebook networks that I belong to (e.g., school, workplace) .71 Facebook groups that I am part of .58 Facebook pages (of companies, celebrities, etc.) that I have visited .78 Facebook advertisements that I click on .56 Another company’s website that I’ve logged into using my Facebook ID .59 My current location that I stated on my profile Cross - website 19.81 % My past purchase history on another . 6 8 2 tracking company’s website .87 My past browsing history on another company’s website .82 My past search history on a search engine .84 My past visits to another company’s website .78 Another company’s website that I’ve logged in without using my Facebook ID Stated personal 17.8 .8 % My gender that I stated on my profile 3 1 9 on my information .83 My sexual orientation that I stated profile My relationship status that .81 on my I stated profile on my profile .76 My age that I stated My family members that I listed on my .71 profile 4 Inferred 6 .8 Facebook 1 9 .9 4 % My sexual orientation that personal inferred based on my Facebook usage information .73 My relationship status that Facebook inferred based on my Facebook usage .86 My age that Facebook inferred based on my Facebook usage .86 My gender that Facebook inferred based on my Facebook usage .59 My current location that Facebook inferred from my computer’s unique IP address .74 My family members that Facebook inferred from my Facebook usage APPENDIX D

52 52 STUDY 1: CROSS -FACTOR LOADINGS 1 ITEMS 4 2 3 0.18 0.86 0.20 0.16 My age that Facebook inferred based on my Facebook usage. 0.86 0.20 0.29 0.10 My sexual orientation that Facebook inferred based on my Facebook usage. 0.18 0.26 0.12 My gender that Facebook inferred based on my Facebook usage. 0.86 0.74 0.36 0.27 0.07 My family members that Facebook inferred from my Facebook usage. 0.73 0.35 0.25 0.15 My relationship status that Facebook inferred based on my Facebook usage. y current M location that Facebook inferred from my computer’s unique IP 0.59 0.38 0.13 0.25 address. 0.87 0.03 0.16 My past browsing history on another company’s website. 0.25 0.22 0.86 0.05 0.17 My past purchase history on another company’s website. My past visits to another company’s website. 0.22 0.84 0.07 0.20 My past search history on a search engine. 0.82 0.06 0.10 0.37 A nother company ’ s website that I ’ ve logged in without using my Facebook 0.15 0.02 0.18 0.78 ID. 0.29 0.01 0.83 0.30 My sexual orientation that I stated on my profile. 0.21 0.20 0.81 0.29 My relationship status that I stated on my profile. 0.03 0.81 0.37 My gender that I stated on my profile. 0.26 My age that I stated on my profile. 0.23 0.05 0.76 0.40 My family members that I listed on my profile. 0.35 0.25 0.71 0.15 0.10 0.04 0.10 0.78 Facebook advertisements that I click on. Facebook pages (of companies, celebrities, etc. ) that I have ‘liked .’ -0.04 0.22 0.25 0.78 0.31 Facebook networks that I belong to (e.g., school, 0.15 0.23 0.77 workplace). 0.27 0.08 0.31 0.71 Facebook groups that I am part of. My current location that I stated on my profile. 0.28 0.04 0.47 0.59 0.39 Facebook pages (of companies, celebrities, etc. 0.15 ) that I have visited. 0.34 0.58 Another company ’s website that I ’ve logged into using my Facebook ID. 0.01 0.48 0.25 0.56

53 53 APPENDIX E STIMULI FROM STUDY 2 For those tasked with browsing movies within- platform: a) interactive image b) pop- up im age -platform , the same for “Kong: Skull Island.” For those tasked with browsing movies cross images served as : a) the blog’s home page b) landing page for “Kong: Skull Island.” a) b)

54 54 APPENDIX F STIMULI FROM STUDY 2 Images used in study 2 for the (a) baseline (b) within -website and (c) cross -website conditions. a) b) c)

55 55 APPENDIX G INFERRED INFORMATION BY GOOGLE

56 56 APPENDIX H STIMULI FROM STUDY 3 Images used in study 3 for the (a) baseline (b) stated and (c) inferred conditions. a) b) c)

57 57 APPENDIX I STUDY 4 DESIGN Validation question: Log into your account. Then, answer the following question. On the top right corner of your Facebook account, there is a pull -down arrow. Which of the following does not appear on this (Correct answer: ‘Help’) pull down list? • Activity log • News Feed Preferences • Settings • Log Out • Help

58 58 APPENDIX J STUDY 4 DESCRIPTIVE STATISTICS Percent of ad transparency messages (N = 414) coded as reveal ing the given information flow. Rows sum to 100% . Yes No I’m not sure Within website - 15.2% 37.7% 47.1% tracking Cross - website 18.9% 34.9% 46.2% tracking 62.8% 25.6% 11.6% Stated attribute 25.1% 51.5% 23.4% Inferred attribute

59 59 APPENDIX K SCREENSHOTS FROM STUDIES 5A- B Study 5a Study 5b

60 60 REFERENCES Dimensions of B rand Personalitym,” Journal of Marketing Research, (1997), “ Aaker, Jennifer L. 356. 34 (3), 347- Acquisti, Alessandro, Leslie K. John, and George Loewenstein (2012), “The I mpact of R elative ropensity to Disclose,” Journal of Marketing Research , 49 ( 2), Standards on the P 160- 174. Aguirre, Elizabeth, Dominik Mahr, Dhruv Grewal, Ko de Ruyter, and Martin Wetzels (2015), “Unraveling the Personalization Paradox: The Effect of Information Collection and -Building Strategies on Online Advertisement Effectiveness,” Journal of Retailing , Trust 91 (1), 34- 49. Altman, Irwin and Dalmas A Social Penetration: The Development of . Taylor (1973), Interpersonal Relationships : Holt, Rinehart & Winston. Archer, Richard L . and John H . Berg (1978), “Disclosure Reciprocity and Its Limits: A Reactance Analysis,” Journal of Experimental Social Psychology , 14 (6), 527- 40. Ashley, Christy, Stephanie M . Noble, Naveen Donthu, and Kathe . Lemon (2011), “ Why rine N t Relate: Obstacles to Relationship Marketing Engagement, Journal of Customers Won’ ” iness Research , 64 (7), 749- Bus 56. Bart, Yakov, Venkatesh Shankar, Fareena Sultan, and Glen L . Urban (2005), “ Are the Drivers and Role of Online Trust the Same for All Web Sites and Consumers? A Large -Scale Exploratory Empirical Study, Journal of M arketing , 69 (4), 133- 52. ” Baumeister, Roy F., Ellen Bratslavsky, Catrin Finkenauer, and Kathleen D. Vohs (2001), “ Bad is Stronger than Good,” Review of General P sychology , 5 ( 4) : 323- 370.

61 61 (2014), “When Identity M arketing Bhattacharjee, Amit, Jonah Berger, and Geeta Menon Identity E Journal of Consumer Research, 41 gency in Backfires: Consumer A xpression,” (2) : 294- 309. Bleier, Alexander and Maik Eisenbeiss (2015), “ Personalized Online Advertising Effectiveness: The Interplay of What, When, and Where,” Marketing Science , 34 (5), 669- 88. yla, M ichael Schmitt, and Richard Harvey (1999) Branscombe, N ervasive , “ Perceiving P Discrimination A rou p Identification and mong African Americans: Implications for G -being,” Journal of Personality and Social Psychology, 77 (1), 135- 149. Well ., Tami Kim, and Chia -Jung Tsay (2016), “ Creating Reciprocal Value through Buell, Ryan W Operational Transparency, ” Management Science . Buell, Ryan W . and Michael I . Norton (2011), “The Labor Illusion: How Operational Transparency Increases Perceived Value, ” Management Science , 57 (9), 1564- 79. Caltabiano, Marie Louise and Michael Smithson (1983), “ Variables Affecting the Perception of -Disclosure Appropriateness,” The Journal of Social Psychology , 120 (1), 119- 28. Self Common F ate, S imilarity, and O Campbell, Donald T. (1958), “ Indices of the S tatus of ther Aggregates of P ersons as S ocial Entities,” Systems R esearch and B ehavioral S cience, 3 (1) : 14- 25. Cazier, Joseph A . St Louis (2003), “ Addressing E -Business ., Benjamin Shao, and Robert D Privacy Concerns: The Roles of Trust and Value Compatibility,” in Proceedings of the 2003 ACM Symposium on Applied C omputing : ACM, 617- 22. Chaudhuri, Arjun and Morris B “The Chain of Effects from Brand Trust and . Holbrook (2001), Brand Affect to Brand Performance: The Role of Brand Loyalty, ” Journal of Marketing , 65 (2), 81- 93.

62 62 . and Raymond G . Sin (2005), “ Chellappa, Ramnath K Personalization Versus Privacy: An Information T ” echnology Empirical Examination of the Online Consumer’s Dilemma, , 6 (2), 181- and M anagement 202. Attention Shoppers: Store Is Tracking Your Clifford, Stephanie and Quentin Hardy (2013), “ ” New York Times , 14. Cell, Communications Consumer Panel (2011), “Online Personal Data: The Consumer Perspective,” http://www.communicationsconsumerpanel.org.uk/Online personal data final 240511.pdf. Costa, Paul T. and Robert R. McCrae (1980), “Influence of Extraversion and Neuroticism on Subjective Well -being: Happy and Unhappy P eople,” Journal of Personality and Social Psychology, 678. 38 (4): 668- Culnan, Mary J . (2000), “ Protecting Privacy Online: Is Self -Regulation Working?,” Journal of Public Policy & Marketing , 19 (1), 20- 26. Dana, Jason, Rober to A. Weber, and Jason Xi Kuang (2007), “ oral W iggle R oom: Exploiting M 33 emonstrating an Experiments D Fairness,” Economic Theory Illusory P reference for (1), 67- 80. Derlega, Valerian J . and Alan L . Chaikin (1977), “Privacy and Self ‐ Disclosure in Social Relationships,” Journal of Social Issues , 33 (3), 102- 15. Desai, Preyas S ., Woochoel Shin, and Richard Staelin (2014), “ The Company That You Keep: When to Buy a Competitor ’s Keyword, ” Marketing Science , 33 (4), 485- 508. Dienel, Eva (2015), “ Transparency, Purpose, and the Empowered Consumer: A New Paradigm

63 63 for Advertising,” https://www.bsr.org/reports/BSR_Participant_Responsibility_Transparency_in_Advertisi ng_March_2015.pdf . Digital Advertising Alliance (2014), “Your Adchoices,” http://youradchoices.com/ . Trust in B uyer -seller R elationships Doney, Pat Journal ricia M., and Joseph P. Cannon (1997), ” ,” arketing , 61: 35- 51. of M Advertisers Lobby Ftc for Right to Not Tell the Truth, ” 3/9/09. Edwards, Jim (2009), “ “What’s the Latest on the Ico Privacy Seals?, Farmer, Gemma (2015), ” -the -on -the -ico -privacy - https://iconewsblog.wordpress.com/2015/08/28/whats -latest . seals/ (2004), “ Social N orms and H uman Cooperation,” Trends in Fehr, Ernst, and Urs Fischbacher Cognitive Sciences , 8 ( 4), 185- 190. Fehr, Ernst, and Simon Gächter (2002), “ Altruistic P Natur e, 415 ( 6868), unishment in Humans,” 140. 137- “A Theory of Reasoned Action: Some Applications and Implications.” Fishbein, Martin (1979), 116. Nebraska Symposium on Motivation, 27, 65- Foreh , “When is H onesty the B est P olicy? The E ffect of and, Mark R., and Sonya Grier (2003) Stated C ompany Intent on Consumer Skepticism,” Journal of Consumer Ps ychology , 13 (3), 356. 349- García Gómez, Blanca, Ana Gutiérrez Arranz, and Jesús Gutiérrez Cillán (2006), “ The Role of Loyalty Programs in Behavioral and Affective Loyalty, ” Journal of Consumer Marketing , 23 (7), 387 -96. Goldfarb, Avi and Catherine E . Tucker (2011), “Privacy Regulation and Online Advertising, ”

64 64 , 57 (1), 57- 71. Management Science enstein (2017), “Information Avoidance,” agmann, and George Loew Golman, Russell, David H 1), 96- 135. Journal of Economic Literature , 55 ( cy, a Disconnect for the Young,” When It Comes to Online Priva Greenblatt, Alan (2013), “ http://www.npr.org/sections/alltechconsidered/2013/06/10/190433719/when- it- comes -to- -privacy -a-disconnect -for -the -young. online “ Ftc to Ad Tech Industry: More Transparency, User Control over Data Greiff, Felicia (2016), -to-ad-tech Necessary,” http://www.mediapost.com/publications/article/267211/ftc - industry -more -transparency -user -c.html. . Paul, Peter Cole, and Jerry Morgan (1975), “ Logic and Conversation,” in Grice, H Syntax and Semantics 3: Speech Acts, eds . Peter Cole and J. L. Morgan, New York: Academic Press 41- 58. Heider, Fritz (1958), “The Naive Analysis of Action,” in The Psychology of Interpersonal Relations , Hoboken, NJ: John Wiley & Sons, 79– 124. Hast ie, Reid, and Bernadette Park (1986), “ elationship B etween M emory and J udgment The R Psychological hether the ask is M emory Depends on W On -line,” Judgment T -based or Review 93 ( 3), 258 -268. Higgins, E. Tory (1997) , “ Beyond P leasure and Pain,” American P sychologist , 52 ( 12) : 1280- 1300. Hill, Kashmir (2012), “ How Target Figured out a Teen Girl Was Pregnant before Her Father ” http://www.forbes.com/sites/kashmirhill/2012/02/16/how Did, -figured -out -a-teen - -target girl -was -pregnant -before -her -father -did/ - 6e4a8ca334c6. Huang, Karen, Michael Yeomans, Alison Wood Br ooks, Julia Minson, and Francesca Gino

65 65 It Doesn’t Hurt to Ask: Qu estion Journal of (2017), “ -Asking Increases Liking,” . Personality and Social Psychology John, Leslie K Strangers on a Plane: ., Alessandro Acquisti, and George Loewenstein (2011), “ Journal of C onsumer Context -Dependent Willingness to Divulge Sensitive Information,” 73. Research , 37 (5), 858- John, Leslie K ., Kate Barasz, and Michael I . Norton (2016), “ Hiding Personal Information Reveals the Worst, Proceedings of the National Academy of Sciences , 113 (4), 954- 959. ” Kelley, H. (1973), “ Processes of C ausal A ttribution ,” American Psychologist , 28, 107- 128. The “The Data Brokers: Selling Your Personal Information,” CBS News , Kroft, Steve (2013), 3/9/2014. Lambrecht, Anja and Catherine Tucker (2013), “ When Does Retargeting Work? Information Specificity in Online Advertising, ” Journal of Marketing Research , 50 (5), 561- 76. Learmonth, Michael (2009), “ Sensing Coming Regulation, Online Ad Groups Unite,” tisingAge Adver , 1/13/09. -Healy, Fern, and Deborah A. Small (2013), “ Nice G uys Finish Last and G uys in Last are ] Lin Nice: The C lash Between D oing W ell and D oing Good,” Social Psychological and Personality Science, 4( 6), 692- 698. Liu, Chang, Jack T Lu, and Chun- Sheng Yu (2005), “ Beyond Concern— a . Marchewka, June Privacy -Trust -Behavioral Intention Model of Electronic Commerce,” Information & Management , 42 (2), 289 -304. Malhotra, Naresh K . Kim, and James Agarwal (2004), “Internet Users’ Info rmation ., Sung S Privacy Concerns (Iuipc): The Construct, the Scale, and a Causal Model,” Information Systems Research , 15 (4), 336- 55.

66 66 Marketing Charts (2014), “Consumers Say They Find Personalized Ads More Engaging and -say -they -find- Memorable,” http://www.marketingcharts.com /online/consumers -more and -memorable -43005/. -ads -engaging- personalized Direct Mail Privacy -Efficiency Trade -Offs Milne, George R . and Mary Ellen Gordon (1993), “ ” Journal of Public Policy & Marketing , within an Implied Social Contract Framework, 206- 15. Montgomery, Alan L . Smith (2009), “ Prospects for Personalization on the . and Michael D Internet,” Journal of Interactive Marketing , 23 (2), 130- 37. Intimate Exchanges: Using Computers to Elicit Self m Moon, Youngme (2000), “ -Disclosure fro Journal of Consumer Research , 26 (4), 323- 39. Consumers,” Morewedge, Carey K. (2009), “Negativity Bias in Attribution of External Agency,” Journal of Experimental Psychology: General , 138 (4), 535- 545. Morey, Timothy, Theodore Theo Forbath, and Allison Schoop (2015), “ Customer Data: Designing for Transparency and Trust,” , 93 (5), 96- 105. Harvard Business Review eorge E., and Daylian M. Cain (2014), “ ltruism: When D oing S ome G ood Newman, G Tainted A valuated as W orse T is E oing No G ood A t All,” Psychological Science , 25(3), 648- han D 655. Nissenbaum, Helen (2004), “ Privacy as Contextual Integrity, ” Wash ington L aw Review , 119, 101- 139. --- (2011), “ A Contextual Approach to Privacy Online,” Daedalus , 140 (4), 32- 48. Paul, Kari (2017), “ When F acebook and Instagram Think You’re Depressed,” https://www.vice.com/en_us/article/when -facebook -and -instagram- thinks -youre - depressed.

67 67 Liang, and Yajiong Xue (2006), “ Understanding and Mitigating Pavlou, Paul A, Huigang -Agent Perspective,” , 31, A Principal MIS Quarterly Uncertainty in Online Environments: 1, 105- 136. Phelps, Joseph, Glen Nowak, and Elizabeth Ferrell (2000), “ Privacy Concerns and Consumer Journal of Public Policy & Marketing , 19 Willingness to Provide Personal Information,” 41. (1), 27- Public Affairs (2015), “Is Online Privacy Over? Findings from the Usc Annenberg Center for the Digital Future Show Millennials Embrace a New Online Reality,” http://annenberg.usc.edu/news/faculty/online -over -findings -usc -annenberg - -privacy -digital- center -show -millennials. future Rainie , Lee and Maeve Duggan (2016), “ Privacy and Information Sharing,” http://www.pewinternet.org/2016/01/14/privacy -and- information -sharing/. Ramirez, Edith, Julie Brill, Maureen K . Ohlhausen, Joshua D . Wright, and Terrell McSweeny Data Brokers –a Call for Transparency and Accountability, (2014), “ Federal Trade ” Commission, Tech. Rep . Rozin, Paul and Edward B. Royzman (2001), “Negativity Bias, Negativity Dominance, and Contagion,” Per sonality and Social Psychology R eview 5 ( 4), 296 -320. Rubin, Zick (1975), ” Journal of “Disclosing Oneself to a Stranger: Reciprocity and Its Limits, Experimental Social Psychology , 11 (3), 233 -60. Sayedi, Amin, Kinshuk Jerath, and Kannan Srinivasan (2014), “Competitive Poaching in Sponsored Search Advertising and Its Strategic Impact on Traditional Advertising, ” Marketing Science , 33 (4), 586- 608. Schoenbachler, Denise D . and Geoffrey L . Gordon (2002), “ Trust and Customer Willingness to

68 68 -Driven Relationship Marketing, ” Provide Information in Database Journal of Interactive -16. , 16 (3), 2 Marketing Targeted Online Schumann, Jan H ., Florian von Wangenheim, and Nicole Groene (2014), “ Advertising: Using Reciprocity Appeals to Increase Acceptance among Users of Free ” Journal of Marketing , 78 (1), 59- 75. Web Services, -Juergen Hippler, and Norbert Schwarz (1992), “ Confidentiality Singer, Eleanor, Hans Assurances in Surveys: Reassurance or Threat?, International Journal of Public Opinion ” , 4 (3), 256- Research 68. a Mae, and Matthew T. Crawford (1998), Skowronski, John J., Donal E. Carlston, Lynd “Spontaneous T ransference: Communicators T ake on the Q ualities T hey Describe rait T in Others,” Journal of P ersonality and Social P sychology , 74 ( 4) : 837- 848. Smith, H . Jeff, Sandra J . Burke (1996), “Information Privacy: Measuring . Milberg, and Sandra J ’ Concerns About Organizational Practices,” 96. MIS quarterly , 167- Individuals . (2003), “ Stewart, Katherine J Organization Science , Trust Transfer on the World Wide Web,” 14 (1), 5- 17. Summers, Christopher A ., Robert W . Smith, and Rebecca Walker Reczek (2016), “ An Audience of One: Behaviorally Targeted Ads as Implied Social Labels, Journal of Consumer ” Research . Tam, Kar Yan and Shuk Ying Ho (2006), “ Understanding the Impact of Web Personalization on User Information Processing and Decision Outcomes, ” MIS Quarterly , 30 (4), 865- 90. Tsai, Janice Y ., Serge Egelman, Lorrie Cranor, and Alessandro Acquisti (2011), “ The Effect of Online Privacy Information on Purchasing Behavior: An Experimental Study,” Information Systems Research , 22 (2), 254 -68.

69 69 Tucker, Catherine (2014), “Social Networks, Personalized Advertising, and Privacy Controls,” -62. Journal of Marketing Research , 51 (5), 546 Turow, Joseph, Jennifer King, Chris Jay Hoofnagle, Amy Bleakley, and Michael Hennessy (2009), “ ” Americans Reject Tailored Advertising and Three Activities That Enable It. Case Western Waldman, Ari Ezra (2016), “ Privacy, Sharing, and Trust: The Facebook Study,” , 67 (1), 193- 233. Reserve Law Review “Trust in and A doption of O Wang, Weiquan and Izak Benbasat (2005), ecommendation nline R Agents,” Journal of the Association for Information Systems , 6 ( 3), 72 -101 . Facebook Now Has 901 Million Users,” Wasserman, Todd (2012), “ now -has -901- http://mashable.com/2012/04/23/facebook- -users/ - million S0aWCJ77U8qT. White, Tiffany Barnett (2004), “ Consumer Disclosure and Disclosure Avoidance: A Motivational Framework,” Journal of Consumer Psychology -2), 41- 51. , 14 (1 . Zahay, Helge Thorbjørnsen, and Sharon Shavitt (2008), White, Tiffany Barnett, Debra L ” “Getting Too Personal: Reactance to Highly Personalized Email Solicitations, , 19 (1), 39- 50. Marketing Letters Yang, Sha, Shijie Lu, and Xianghua Lu (2013), “ Modeling Competition and Its Impact on Paid- Search Advertising, ” Marketing Science , 33 (1), 134- 53. Yoon, Yeosun, Zeynep Gürhan- Canli, and Norbert Schwa rz (2006), “ The Effect of C orporate Social R ctivities on C ompanies with B ad R eputations ,” Journal of esponsibility (CSR) A Consumer P sychology, 16 ( 4), 377- 390. Zhang, Kaifu and Zsolt Katona (2012), “ Contextual Advertising, ” Marketing Science , 31 (6), 980- 94.

Related documents