ABSTRACT Title of Dissertation: ANTECEDENTS OF DISHONEST CONSUMER BEHAVIOR In Hye Kang, Doctor of Philosophy, 2019 Dissertation directed by: Ralph J. Tyser Professor of Marketing Amna Kirmani, Department of Marketing Consumers engage in a wide range of dishonest behaviors, such as cheating or lying to companies for financial rewards. These dishonest behaviors are costly for companies and consumers. However, relatively little research in marketing has paid attention to consumer dishonesty. In this dissertation, we enhance the understanding of dishonest consumer behavior by examining a few prominent antecedents: a company’s corporate social responsibility (CSR) initiatives and construal level. The first essay examines how a company’s CSR initiatives impacts consumers’ dishonest behavior toward the company. Companies are proclaiming their values by taking stands on controversial issues in their CSR. We examine a novel way with which consumers respond to a company’s polarizing CSR: dishonest behavior toward the company. We demonstrate that when the CSR cause is congruent with the consumer’s self-concept, CSR (vs. no-CSR) decreases dishonest behavior by increasing anticipatory self-threat (i.e., if I cheat the company, I will feel like I am a bad person). In contrast, when the CSR cause is incongruent, CSR (vs. no-CSR) increases dishonest behavior by decreasing anticipatory self-threat. We demonstrate an asymmetric effect such that the effect of incongruent CSR is larger than the effect of congruent CSR. Building on the anticipatory self-threat mechanism, we identify a boundary condition in which the backfiring effect of incongruent CSR is attenuated: situational salience of moral identity. The second essay investigates how construal level—the extent to which people’s thinking about a situation is abstract or concrete—influences dishonest consumer behavior. We show that the effect of construal level on dishonest behavior is moderated by the importance of moral values. We find that compared to concrete construal, abstract construal reduces dishonest behavior when the importance of moral values is high but not when the importance of moral values is low. Importance of moral values is measured as individual differences and situationally primed. These essays provide valuable insights into consumer dishonesty by demonstrating that different types of factors (characteristics of a company such as CSR and contextual factors such as construal level) influence dishonest consumer behavior. Moreover, these essays provide practical implications for companies seeking to reduce dishonest consumer behaviors. ANTECEDENTS OF DISHONEST CONSUMER BEHAVIOR by In Hye Kang Dissertation submitted to the Faculty of the Graduate School of the University of Maryland, College Park, in partial fulfillment of the requirements for the degree of Doctor of Philosophy 2019 Advisory Committee: Ralph J. Tyser Professor of Marketing Amna Kirmani, Chair Dean’s Professor of Marketing David Godes Dean’s Professor of Marketing Rebecca Ratner Associate Professor of Marketing Rosellina Ferraro Associate Professor of Psychology Edward Lemay © Copyright by In Hye Kang 2019 Acknowledgements I would like to express my gratitude to the many people who helped me complete this dissertation. First and foremost, I would like to express my deepest gratitude to my advisor, Dr. Amna Kirmani, for making this dissertation work possible. I have been extremely fortunate to have an advisor who always provided insightful, knowledgeable academic guidance and showed support and faith in me in hard times. It has been a great pleasure to work with such a great scholar and sincere individual. I would also like to thank Dr. Rebecca Ratner for her thoughtful feedback and support. I am also grateful to Dr. David Godes for his valuable feedback and generous guidance as a PhD advisor. I would also like to express my sincere thanks to Dr. Rosellina Ferraro who always provided constructive feedback and generous support while working on several projects together throughout my doctoral program. I also feel grateful to Dr. Edward Lemay for having generously served as a committee member and providing valuable feedback. I would also like to thank the faculty, students, and staff in the Marketing Department at the University of Maryland for all their helpful advice, feedback, and support. Finally, and most importantly, I feel deeply grateful to my husband, Chan Young Park. None of this would have been possible without his constant support, encouragement, and patience. ii Table of Contents Acknowledgements ....................................................................................................... ii Table of Contents ......................................................................................................... iii List of Tables ................................................................................................................ v List of Figures .............................................................................................................. vi Chapter 1: Essay I – When Taking a Stand on Controversial Social Issues Hurts or Helps the Company: Anticipatory Self-Threat and Dishonest Consumer Behavior .... 1 Introduction ............................................................................................................... 1 Theoretical Background ............................................................................................ 6 Anticipatory Self-Threat from Dishonest Behavior .............................................. 7 The Moderating Role of Cause-Related Identity .................................................. 8 Hypotheses .......................................................................................................... 10 Boundary Condition: Salience of Moral Identity................................................ 12 Alternative Explanations ..................................................................................... 13 Study Overview ...................................................................................................... 14 Study 1 .................................................................................................................... 15 Pretest .................................................................................................................. 15 Method ................................................................................................................ 17 Results ................................................................................................................. 19 Discussion ........................................................................................................... 22 Study 2 .................................................................................................................... 23 Method ................................................................................................................ 23 Results ................................................................................................................. 26 Discussion ........................................................................................................... 31 Study 3 .................................................................................................................... 32 Method ................................................................................................................ 33 Results ................................................................................................................. 34 Discussion ........................................................................................................... 39 Study 4 .................................................................................................................... 41 Method ................................................................................................................ 41 Results ................................................................................................................. 43 Discussion ........................................................................................................... 47 General Discussion ................................................................................................. 48 Theoretical Contribution ..................................................................................... 49 Managerial Contribution ..................................................................................... 53 Conclusion .......................................................................................................... 54 Chapter 2: Essay II – Thinking About Why Leads to Less Cheating When the Importance of Moral Values Is High .......................................................................... 55 Introduction ............................................................................................................. 55 Theoretical Background .......................................................................................... 57 Construal Levels in the Moral Domain ............................................................... 57 Moderating Role of Importance of Moral Values............................................... 62 Study Overview ...................................................................................................... 64 Study 1 .................................................................................................................... 65 iii Method ................................................................................................................ 65 Post-Hoc Test of Manipulation Check of Construal Level ................................. 67 Results ................................................................................................................. 68 Discussion ........................................................................................................... 70 Study 2 .................................................................................................................... 71 Method ................................................................................................................ 71 Post-Hoc Test of Manipulation Check of Construal Level ................................. 74 Results ................................................................................................................. 74 Discussion ........................................................................................................... 78 Study 3 .................................................................................................................... 79 Method ................................................................................................................ 79 Results ................................................................................................................. 80 Discussion ........................................................................................................... 82 Study 4 .................................................................................................................... 82 Method ................................................................................................................ 83 Results ................................................................................................................. 84 Post-Hoc Test ...................................................................................................... 85 Discussion ........................................................................................................... 87 General Discussion ................................................................................................. 87 Theoretical and Practical Contributions.............................................................. 88 Limitations and Future Research Agenda ........................................................... 90 Appendices .................................................................................................................. 92 Bibliography ............................................................................................................. 113 iv List of Tables Table 1. Indirect effect of anticipatory self-threat on intention for dishonest behavior at different levels of LGBT support identity ............................................................... 29 v List of Figures Figure 1. Moderating role of military veteran support identity on dishonest behavior (i.e., difference score) (study 1) .................................................................................. 21 Figure 2. Moderating role of LBGT support identity on intention for dishonest behavior (study 2) ....................................................................................................... 27 Figure 3. Moderating role of LGBT support identity on intention for dishonest behavior (study 3) ....................................................................................................... 36 Figure 4. Interactive effect of refugee support identity and salient identities on intention for dishonest behavior (study 4) .................................................................. 44 Figure 5. The effect of construal level and importance of benevolence on dishonest behavior (study 1) ....................................................................................................... 69 Figure 6. The effect of construal level and importance of moral values on dishonest behavior (study 2) ....................................................................................................... 78 Figure 7. The effect of construal level and importance of moral values on dishonest behavior (study 3) ....................................................................................................... 81 Figure 8. The effect of construal level and social consensus about the acceptability of dishonest behavior on dishonest behavior (study 4) ................................................... 85 vi Chapter 1: Essay I – When Taking a Stand on Controversial Social Issues Hurts or Helps the Company: Anticipatory Self- Threat and Dishonest Consumer Behavior Introduction Increasingly, companies are proclaiming their values by taking stands on controversial social issues, such as gender and ethnic diversity, veterans’ programs, and refugee support. Directed at both internal and external stakeholders, these actions often take the form of corporate social responsibility (CSR), defined as “the responsibility of enterprises for their impact on society” (European Commission, 2011, p. 6). While some CSR initiatives may lead to positive outcomes for the company, many others have the potential to be controversial and may hurt the company. For instance, when Target expressed its inclusivity by installing gender- neutral bathrooms in stores, many stakeholders who opposed gender neutrality reacted negatively. Meanwhile, Starbucks tried to mitigate the negative reaction from announcing a plan to hire refugees worldwide by highlighting its much larger initiative to hire military veterans. In general, prior work on companies’ CSR shows that it has positive benefits for a variety of stakeholders. For example, CSR improves employees’ attitudes toward their company and job satisfaction (e.g., De Roeck, Marique, Stinglhamber, & Sawen, 2014), and commitment towards the company (e.g., Brammer, Millington, & Rayton, 2007). In addition, CSR increases consumers’ attitudes towards the company 1 and purchase intentions (e.g., Sen & Bhattacharya, 2001) and enhances behavioral loyalty (e.g., Bolton & Mattila, 2015; Du, Bhattacharya, & Sen, 2007; Habel, Schons, Alavi, & Wieseke, 2016). This research has focused on uncontroversial CSR causes, such as philanthropic giving and community development programs. However, prior research has not examined individuals’ reactions to controversial CSR initiatives, which is the characteristic of many social issues today. We address this gap by examining how a company’s external stakeholders, specifically its consumers, respond to controversial CSR initiatives. Although CSR can influence consumers’ purchase behavior toward the company (e.g., through boycotting and buycotting), we propose a novel way in which consumers may respond to CSR: dishonest behavior, i.e., the extent to which consumers lie and cheat the company for financial rewards. Dishonest behaviors are highly costly for companies. For example, return fraud and abuse, including wardrobing (i.e., returning the used clothing), were expected to cost retailers about $22.8 billion in 2017 (Total Retail, 2018); meanwhile, fraudulent or inflated property/casualty insurance claims cost insurance companies about $34 billion each year from 2011 to 2015 (Insurance Information Institute, 2017). Moreover, honest consumers are penalized by paying higher prices for products and services to cover firms’ costs from dishonest consumer behavior. Thus, investigating factors that can accentuate or mitigate dishonest behavior is important from both theoretical and practical perspectives. We propose that when boycotting is not feasible because of the lack of readily available substitutes, consumers may 2 engage in dishonest behaviors, such as lying about the reason for returns, or lying about meeting qualifying criteria for a promotion. In particular, we propose that the effect of CSR on dishonest behavior depends on the consumer’s cause-related identity, which refers to the degree of fit between the CSR cause and the consumer’s self-concept (e.g., Sen & Bhattacharya, 2001). Although prior work on CSR considers the effects of cause-related identity, it does not examine incongruent or oppositional identity, such as that represented by causes that are contrary. In contrast, we conceptualize incongruent cause-related identity as one that is the opposite of the consumer’s self-concept. We propose that the moderating role of cause-related identity in the effect of CSR on dishonest behavior toward the company is mediated by anticipatory self- threat. Anticipatory self-threat refers to the expectation that the individual’s positive self-view cannot be maintained after he or she acts in a certain way (Sherman, Nelson, & Steele, 2000; White, Argo, & Sengupta, 2012). We propose that when the consumer’s self-concept is congruent with the cause, he or she will experience stronger anticipatory self-threat about acting dishonestly toward a company engaged in CSR than toward a company with no CSR, thereby reducing dishonest behavior. In contrast, when the cause is incongruent with the self-concept, CSR will backfire; incongruent CSR (vs. no-CSR) will decrease anticipatory self-threat and thus increase dishonest behavior. Moreover, drawing on the literature on negativity bias (Baumeister, Bratslavsky, Finkenauer, & Vohs, 2001) and self-verification (Swann, Pelham, & Krull, 1989), we propose an asymmetric effect of cause-related identity. 3 Specifically, the effect of incongruent CSR on increasing dishonest behavior will be bigger than the effect of congruent CSR on decreasing dishonest behavior. Finally, building on the anticipatory self-threat mechanism, we propose that making moral identity situationally salient will reduce dishonest behavior in response to incongruent CSR. When moral identity is salient, dishonest behavior is likely to be incorporated in consumers’ moral self-view (Mazar, Amir, & Ariely, 2008). This suggests that when moral identity is salient, anticipatory self-threat will be heightened from considering acting dishonestly because dishonest behavior itself is contradictory to the moral identity. Thus, we expect that when moral identity is salient, anticipatory self-threat will be elevated regardless of a company’s CSR and the consumer’s cause- related identity, attenuating the moderating role of cause-related identity. Collectively, our research provides both theoretical and substantive contributions. First, there are several major contributions to work on CSR. Whereas prior literature generally finds positive responses of CSR (Du et al., 2007; De Roeck et al., 2014), we demonstrate negative effects when companies take moral stands on polarizing issues. By doing so, this research answers a call for additional research on the dark side of CSR (Rupp & Mallory, 2015), augmenting work that shows that inauthentic CSR (Mallory & Rupp, 2014; McShane & Cunningham, 2012) and CSR that has poor fit with the company (Torelli, Monga, & Kaikati, 2012) have deleterious effects. We further answer a call for research on the dark side of CSR by demonstrating a novel way that individuals react to a company’s controversial CSR initiatives: dishonest behavior toward the company. Specifically, consumers act more dishonestly toward a company engaged in incongruent CSR (vs. not engaged in 4 CSR). In addition, whereas prior literature looks at the effects of congruent or neutral identities on consumers’ responses to CSR (Sen & Bhattacharya, 2001; Winterich & Barone, 2011), we investigate the effect of oppositional cause-related identity. Examining this oppositional identity is theoretically important as it provides a more complex view of identity in the context of CSR. From the substantive and managerial perspectives, this research suggests that since consumers are divided on a variety of social issues, companies must take into account oppositional identities when selecting CSR initiatives. Our research also contributes to the literature on dishonest behavior. Prior literature has examined various antecedents of dishonest behavior, including characteristics of the actors (e.g., gender, creativity) and contexts (e.g., feelings of disgust, salience of moral values or money, self-regulatory resources) (e.g., Gino & Ariely, 2012; Gino, Schweitzer, Mead, & Ariely, 2011; Kouchaki, Smith-Crowe, Brief, & Sousa, 2013; Mazar et al., 2008; Schaumberg & Wiltermuth, 2014; Winterich, Mittal, & Morales, 2014). We add to this list by demonstrating that the fit between the consumer and the cause (more generally, the fit between the characteristics of an actor and the target of dishonest behavior) is an important antecedent of the company-directed dishonest behavior. This is consistent with literature showing that dishonest behavior is a social phenomenon between an actor who does something wrong and the recipient of the wrong action (Brass, Butterfield, & Skaggs, 1998). 5 Theoretical Background Because CSR entails activities that promote social good, companies engaged in CSR, especially in the domain of uncontroversial causes such as philanthropic activities, are likely to be perceived as prosocial and thus moral. Consistent with this, Hoeffler and Keller (2002) suggested that CSR should lead consumers to view the company as more sincere and caring. Moreover, Chernev and Blair (2015) found that CSR enhanced the perceived performance of the company’s products via a halo effect, where consumers’ perceptions of the company’s morality extended to other aspects of the company. Although they did not measure the moral perception of the company, they showed that when consumers perceived CSR to be motivated by self- interest rather than altruism, consumers had less favorable perceptions of the company’s product performance. Because of this moral nature of CSR and dishonest behavior, one may think that CSR may influence dishonest behavior toward the company because consumers perceive the company engaged in CSR (vs. not engaged in CSR) as more or less moral. More generally, one may also think that CSR can influence dishonest behavior by affecting consumers’ attitudes (i.e., liking) toward the company engaged in CSR. We show that these accounts based on the consumer’s perception of the company do not drive the effect of CSR on dishonest behavior. Rather, we propose an identity- related account of CSR, such that CSR can affect the degree to which consumers expect that their positive self-view will be threatened after acting dishonestly, thereby reducing dishonest behavior toward the company. In other words, consumers act more or less dishonestly toward the company engaged in CSR due to egoistic motives 6 (e.g., to avoid feeling bad about themselves) rather than altruistic motives (e.g., liking or concern for the company). Anticipatory Self-Threat from Dishonest Behavior An opportunity to act dishonestly for financial rewards presents a conflict between the possibility of attaining financial rewards and a threat to one’s positive self-view (Mazar et al., 2008). Prior work shows that one’s positive self-view is threatened when negative information about a relevant identity is presented, because this information makes it difficult to maintain the positive self-view (Steele, 1988; White et al., 2012). Acting dishonestly creates a self-threat by making it difficult to maintain the positive self-view that one is moral (Sachdeva, Iliev & Medin, 2009; Zhong & Liljenquist, 2006). To overcome this self-threat in the moral domain, consumers engage in subsequent compensatory behavior, such as acting prosocially (Sachdeva et al., 2009) and physical cleansing (Zhong & Liljenquist, 2006). Whereas self-threat may occur after acting dishonestly, consumers also anticipate self-threat when thinking about acting dishonestly (i.e., “I would feel bad about myself if I acted dishonestly”) (Gino, Ayal, & Ariely, 2013; Mazar et al., 2008). This anticipatory self-threat serves as a force in deterring dishonest behavior. For example, since people anticipate more self-threat from cheating a lot, they cheat only a little (Mazar et al., 2008). When consumers can attribute their dishonest behavior to external factors rather than to self-interest, they expect less self-threat and act more dishonestly (Gino et al., 2013). Self-threat is often associated with negative emotions, such as guilt (Cheryan & Monin, 2005; Peloza, White, & Shang, 2013; Trudel, Argo, & Meng, 2016). When 7 consumers recognize that their ethical standards diverge from their actual behavior, they feel anticipatory guilt from not purchasing ethical products (Peloza et al., 2013). In fact, researchers sometimes measure negative emotion to assess self-threat (Cheryan & Monin, 2005; Trudel et al., 2016). For example, Trudel et al. theorized that throwing away (vs. recycling) identity-related products poses greater identity threat. They measured identity threat with negative emotions such as guilt, shame, and regret along with the more direct measure of one’s sense of self-view. Thus, anticipatory self-threat can be manifested either cognitively (e.g., cheating will make me a bad person) or emotionally (e.g., I will feel guilty if I cheat). We expect that this baseline level of anticipatory self-threat evoked from contemplating acting dishonestly will be magnified or attenuated depending on the consumers’ cause- related identity. We develop this idea next. The Moderating Role of Cause-Related Identity Consumers vary in terms of their cause-related identities, that is, the degree to which a particular cause is perceived as congruent or incongruent with the self- concept (Sen & Bhattacharya, 2001; Winterich & Barone, 2011). Although the literature uses different terms, such as personal support for the cause (Sen & Bhattacharya, 2001) and cause-involvement (Koschate-Fischer, Stefan, & Hoyer, 2012), cause-related identity captures the notion of closeness to the self-concept. Prior work shows that the positive effects of CSR are attenuated (but not reversed) as the consumer perceives less congruency between the consumer’s self-concept and the cause. For example, Sen and Bhattacharya (2001) showed that when consumers’ self- concept was congruent with the CSR cause, CSR increased company evaluation by 8 enhancing the perception of the consumer-company congruence. In contrast, when consumers’ self-concept was less congruent with the CSR cause, CSR did not influence company evaluation. Similar attenuation of positive effects has been found in the cause-related marketing literature (Koschate-Fischer et al., 2012; Winterich & Barone, 2011). We hypothesize both positive and negative effects of cause-related identity on dishonest behavior. A key difference from prior literature is our definition of incongruent causes as contrary to the consumer’s self-concept. Prior work defines incongruent cause-related identity as a low degree of overlap, not as an oppositional identity (e.g., Sen & Bhattacharya, 2001). However, we view cause-related identity as a continuum from congruent to neutral to incongruent. Congruent cause-related identity involves a high degree of overlap between the self-concept and the cause; neutral cause-related identity involves a low degree of overlap (i.e., what prior research calls incongruent); and incongruent cause-related identity involves a high degree of separation between the self-concept and the cause, one that is the opposite of the consumer’s self-concept. This conceptualization is based on the notion that consumers are polarized about many social causes (Cone Communications, 2017) and companies take stands on these polarized issues. For example, Starbucks supports controversial CSR causes such as refugee hiring and climate change (Starbucks, 2017); and AT&T publicly states their commitment to LGBT groups (AT&T, 2017). In our conceptualization, an incongruent cause-related identity reflects oppositional or undesired identities (Reed, Forehand, Puntoni, & Warlop, 2012). Consumers actively dissociate themselves from having certain characteristics or 9 reference groups (White et al., 2012). In contrast, consumers have a neutral cause- related identity when the cause is unrelated to their self-concept. Next, we develop hypotheses about the relationship between CSR and dishonest behavior, which is moderated by cause-related identity and mediated by anticipatory self-threat. Hypotheses As mentioned earlier, an opportunity to act dishonestly presents a conflict between attaining financial rewards and maintaining one’s positive self-view (Mazar et al., 2008). In the absence of CSR, consumers’ anticipatory self-threat and decision to behave dishonestly may depend on contextual factors because the context can make it easier or more difficult for consumers to justify that their dishonest behavior does not reflect who they are. For example, consumers are more likely to behave dishonestly when the beneficiary of the dishonest behavior is someone else rather than themselves (Gino et al., 2013). These situations reduce anticipatory self-threat by enabling consumers to view their behavior as not reflecting their identity as a moral person. We consider this the baseline level of anticipatory self-threat. When consumers see that a company supports a certain cause in its CSR activity, their cause-related identity becomes salient (Winterich & Barone, 2011), affecting anticipatory self-threat. When a CSR cause is congruent with the consumer’s self-concept, the consumer will view acting dishonestly toward the company engaged in CSR as contradicting their cause-related identity as well as their self-view as a moral person. Dishonest behavior would hurt their self-view because their dishonest behavior does not reflect who they are, both by the fact of acting unethical (I’m not moral) and because it means they do not support the cause (I’m not 10 a person who supports diversity). To reduce this anticipatory self-threat, they would be less likely to act dishonestly toward a company engaged in CSR than one not engaged in CSR. In contrast, when the cause-related identity is incongruent, consumers may feel less anticipatory self-threat. We expect that when the cause-related identity is incongruent, consumers will view acting dishonestly toward the company and causing harm to the company as consistent with their cause-related identity. Although dishonest behavior would still hurt their moral self-view (I am not moral) as in the case of our baseline, i.e., no-CSR, it would also boost their overall self-view by affirming their cause-related identity (I am a person who opposes diversity). Thus, CSR (vs. no-CSR) would reduce the anticipatory self-threat elicited by acting dishonestly toward the company engaged in CSR more than one not engaged in CSR, increasing dishonest behavior. Thus, we hypothesize the following: H1: The effect of CSR (vs. no-CSR) on dishonest consumer behavior toward the company will be moderated by the consumer’s cause-related identity. Specifically, when the cause-related identity is congruent, CSR (vs. no-CSR) will reduce dishonest behavior. When the cause-related identity is incongruent, CSR will increase dishonest behavior. H2: Anticipatory self-threat will mediate the effect of CSR (vs. no-CSR) on dishonest consumer behavior toward the company. Moreover, we expect an asymmetric effect of CSR under incongruent and congruent cause-related identity. Consumers have strong motivation for self- verification, which posits that consumers are motivated to receive information that is 11 consistent with their self-concept (even when the information is about a negative aspect of the self, such as “I am a lazy person”) and motivated to avoid information that is inconsistent with their self-concept (Swann et al., 1989). Thus, consumers will consider the information that a company supports the identity-congruent [identity- incongruent] CSR cause as positive [negative]. Prior work in psychology has consistently demonstrated a negativity bias such that consumers react more strongly to negative than to positive information (Baumeister et al., 2001; Rozin & Royzman, 2001). This negativity bias has also been observed in the moral domain (Janoff- Bulman, Sheikh, & Hepp, 2009). Consumers react more negatively to the violation of moral values (e.g., harming others) than they react positively to the compliance with moral values (e.g., helping others). Thus, we expect that the effect of identity- incongruent CSR (vs. no-CSR) on reducing anticipatory self-threat and thus increasing dishonest behavior will be bigger than the effect of identity-congruent CSR on increasing anticipatory self-threat and decreasing dishonest behavior. Specifically, H3: CSR (vs. no-CSR) will have a greater impact on dishonest behavior when the cause-related identity is incongruent rather than congruent. Boundary Condition: Salience of Moral Identity Given that incongruent CSR can lead to dishonest behavior, how can we attenuate this effect? The proposed anticipatory self-threat mechanism suggests that if anticipatory self-threat is made salient by other situational factors, the effect of CSR on dishonest behavior should be attenuated. We propose that one such situational factor is the situational salience of moral identity. The literature shows that making 12 moral identity salient by asking consumers to recall the Ten Commandments or highlighting the ethical norms of the company reduces dishonest behavior (Aquino, 1998; Mazar et al., 2008). When moral identity is salient, any dishonest behavior will be more likely to be incorporated in consumers’ self-view, reducing dishonest behavior. This suggests that salient moral identity will heighten anticipatory self- threat elicited from considering acting dishonestly itself, regardless of the company’s CSR activity or the consumer’s cause-related identity, because dishonest behavior itself is contradictory to the salient moral identity. Thus, we expect that H4: When moral identity is salient (vs. not salient), the moderating role of cause-related identity in the effect of CSR on dishonest behavior will be attenuated. Alternative Explanations Besides anticipatory self-threat, we consider four potential alternative processes: liking for the company, the perceived morality of the company, the expected harm to the company resulting from people’s dishonest behavior, and anticipatory satisfaction. It is possible that CSR affects dishonest behavior toward the company by influencing how consumers view the company. For example, when the cause-related identity is congruent [incongruent], consumers may see a company engaged (vs. not engaged) in CSR as more [less] moral or likeable. As a result, they may view acting dishonestly toward the company as more [less] morally wrong, which will reduce [increase] dishonest behavior. Relatedly, it is possible that CSR leads consumers to anticipate that their dishonest behavior would cause more or less harm to the company, which in turn influences dishonest behavior. These alternative 13 accounts suggest that the effect of CSR on company-directed dishonest behavior is driven by more altruistic motives such as genuine liking or concern for the company rather than by more egoistic motives, such as avoiding anticipatory self-threat. By ruling out these alternative explanations, we further show that our identity-based account explains the process better than a liking-based account. Moreover, anticipatory satisfaction may also serve as an underlying psychological process. Consumers may expect to be satisfied with dishonest behavior since it provides them with financial gains, which can trigger positive emotions such as satisfaction (Meloy, Russo, & Miller, 2006). It is possible that when the cause- related identity is congruent [incongruent], consumers may anticipate less [more] satisfaction from acquiring financial rewards from the company, and accordingly be less [more] likely to act dishonestly. We test these four alternative explanations in our studies. Study Overview We tested the hypotheses in four studies; study 1 uses actual dishonest behavior, while the other three studies measure intentions. In study 1, we provide support for the moderating role of cause-related identity on the effect of polarizing CSR on dishonest behavior toward the company (H1) and the proposed asymmetric effect (H3). In studies 2 and 3, we replicate the findings of study 1 and demonstrate the mediating role of anticipatory self-threat (H2). In study 4, we show the effect of incongruent CSR on increasing dishonest behavior is attenuated when moral identity was made situationally salient (H4). In studies 2-4, we rule out the above mentioned 14 four alternative explanations. Across the studies, we use different polarizing CSR causes (military veteran support, LGBT support, and refugee support) and different dishonest behaviors (lying about performance, wardrobing, and lying about one’s child age to get a discount) allowing us to generalize the findings. Study 1 The primary purpose of study 1 was to test H1, which posits that a consumer’s cause-related identity moderates the effect of a polarizing CSR on dishonest behavior toward the company, and H3, that CSR will have a greater impact on dishonest behavior when the cause-related identity is incongruent rather than congruent. Moreover, to test H4, we measured the importance of moral identity, defined as the extent to which morality is an important part of one’s self-concept (Aquino & Reed, 2002). If moral identity becomes more accessible for those with stronger moral identity when consumers encounter an opportunity to act dishonestly, H1 should be attenuated for individuals with a highly important moral identity. To examine company-directed actual dishonest behavior in the lab, we adapted an experimental paradigm from Mazar et al. (2008), in which participants could over-report their performance for monetary gains. To assess CSR causes that may lead to polarization, we first ran a pretest. Pretest CSR activities fall into several broad categories. For example, the MSCI ESG Stats Database (2010), which yearly rates CSR performances of about 3,000 largest 15 U.S. companies, classifies CSR domains in the following seven major categories: community, diversity, environment, employee relations, human rights, product, and corporate governance. We focused on three categories: community, environment, and diversity. We selected nine causes that fell into these categories: community (providing support for hungry people and providing education for children in need); environment (ensuring environmental sustainability, actions that tackle climate change, and animal testing); and diversity (promoting racial and ethnic diversity, promoting rights of sexual minorities (e.g., gay and lesbian, transgender, and bi- sexual), giving preferential treatment to refugees, and giving preferential treatment to military veterans). One hundred participants recruited from Amazon MTurk were asked to categorize each cause into one of the boxes: a) I describe myself as a person who supports this issue, b) I describe myself as a person who opposes this issue, and c) I describe myself as a person who neither supports nor opposes this issue. The three categories indicate congruent, incongruent, and neutral cause-related identity, respectively. We also measured the importance of moral identity (internalization dimension; α = .76; Aquino & Reed, 2002) and political ideology, in order to assess whether they are correlated with cause-related identities. Political ideology was measured by a single item, “Do you consider yourself as politically liberal or politically conservative (1 = very liberal, 7 = very conservative) (Graham, Haidt, & Nosek, 2009). The causes with relatively equal sizes of congruent and incongruent participants were promoting rights of sexual minorities (LGBT groups) (58%, 31%), 16 giving preferential treatment to military veterans (58%, 29%), and giving preferential treatment to refugees (11%, 36%). Therefore, we use support for military veterans in study 1, promoting rights of sexual minorities in studies 2 and 3, and support for refugees in study 4. The importance of moral identity was not significantly correlated with the cause-related identities for these three causes (0 = incongruent, 1 = neutral, and 2 = congruent cause-related identity) (ps > .29). Political ideology was not significantly correlated with the cause-related identity for support for military veterans (r = .09, p = .41). Political ideology was significantly negatively correlated with the cause-related identity for promoting rights of sexual minorities (r = -.35, p <.001) and positively for supporting refugee (r =.22, p = .031). Appendix A shows the detailed results. Method Participants and design. The study used a 2 (CSR activity: present vs. absent) by (cause-related identity—military veteran support identity; measured continuously) between-subjects design. Participants were 182 undergraduates (44.5% male, Mage = 20.19, SD = 2.05) at a large American university who were randomly assigned to the CSR activity condition. Procedure. Participants sat at individual computer stations. The experimenter announced that the study was conducted by an American educational program development company, EduValley, which was trying to test the difficulty of one of its educational programs. The experimenter then distributed a company introduction sheet. All participants read a general description of EduValley (see appendix B for 17 description). Participants in the CSR present condition read an additional paragraph about EduValley’s CSR activity. In 2015, EduValley was rated as among the best in the industry on support for military veterans. This rating was given by an independent company that evaluates companies every year. Since the beginning of the company, EduValley has annually donated a portion of their profits to veteran-benefiting charities to provide veterans with necessary medical cares and job training. Moreover, EduValley has provided its educational training programs for veteran-benefiting charities so that they can be used for veterans’ job training. At the computer station was a brown envelope with an instruction sheet, a matrix sheet, an answer sheet, and an envelope containing $5 (see appendix C for stimuli). Participants read from the instruction sheet that all of the materials were funded by EduValley and that they would earn $0.25 for each correctly solved matrix. To make participants aware of the opportunity to cheat without getting caught, they also read that they would throw away the matrix sheet in the recycling bin before submitting the answer sheet. Unbeknownst to them, the matrix and answer sheets were marked with invisible IDs viewable under an ultraviolet light so that their actual performance could be linked to their reported performance. Then, participants took out the matrix sheet, which contained 20 matrices with 12 three-digit numbers (e.g., 3.91). Participants had 4 minutes to find two numbers per matrix that added up to 10. After 4 minutes, they counted the number of correctly solved matrices, threw away the matrix sheet in a recycling bin, and recorded the number of correctly solved matrices on the answer sheet. Participants then paid themselves using the provided money. Participants completed other unrelated filler studies for 30 minutes. Then, they responded to various identity measures, including the target military veteran support 18 identity, the university student identity (α = .82), moral identity (α for internalization dimension = .74, α for symbolization dimension = .78; Aquino & Reed, 2002), and political ideology. The target military veteran support identity was measured using three items (1= strongly disagree, 7 = strongly agree): (1) I identify strongly as a person who cares about and supports military veterans; (2) Being a person who cares about and supports military veterans is an important part of who I am; and (3) Being a person who cares about and supports military veterans is opposite to who I am (reverse-coded) (adapted from Bolton & Reed, 2004) [M = 4.82, SD = 1.28, α = .75]. Note that the third item represents oppositional identity. After participants left, the matrix sheets were matched with the answer sheets. The number of matrices correctly solved was compared to the number of matrices reported. The difference between the two represented the extent to which participants had overstated their performance, serving as the main dependent variable. Results To ensure that the CSR manipulation did not influence identity measures, we ran one-way ANOVAs with each of the identity measures as the dependent variables and CSR as the independent variable. No one-way ANOVA was statistically significant (ps > .19), allowing us to use these identity measures as a moderator. Dishonest behavior. On average, participants correctly solved 6.41 matrices and reported that they solved 7.58 matrices. The comparison between the numbers of matrices actually solved and reported showed that 37.9% (n = 69) of participants over-reported their performance by one or more matrices, that 61.0% (n = 111) showed no difference between the actual and reported matrices, and that 1.1% (n = 2) 19 under-reported their performance (one participant under-reported the performance by one matrix, and the other under-reported the performance by two matrices). In the case of under-reporting, we coded the difference score as 0 since under-reporting was not considered a cheating behavior. To test H1, regression analysis was performed using CSR (0 = absent, 1 = present), mean-centered military veteran support identity, and the two-way interaction term as the independent variables and dishonest behavior (i.e., the difference score) as the dependent variable. There was no significant main effect of CSR on dishonest behavior (b = -.13, t(178) = -.35, p = .73). The main effect of military veteran support identity was not significant (b = .12, t(178) = .62, p = .54). Importantly, the two-way interaction was significant (b = -.72, t(178) = -2.52, p = .013) (see Figure 1). The Johnson-Neyman analysis (Spiller, Fitzsimons, Lynch, & McClelland, 2013) revealed that CSR (vs. no-CSR) significantly decreased dishonest behavior toward the company for participants with military veteran identity greater than 6.02 (MCSRpresent = 0.38 vs. MCSRabsent = 1.36; bJN = -.98, SE = .50, p = .050). In contrast, CSR (vs. no- CSR) significantly increased dishonest behavior for participants with military veteran identity less than 2.73 (MCSRpresent = 2.33 vs. MCSRabsent = 0.96; bJN = 1.37, SE = .69, p = .050). Stated differently, when support for the military veterans was congruent [incongruent] with the individual’s self-concept, CSR (vs. no-CSR) significantly decreased [increased] individual’s cheating behavior toward the company. Thus, H1 is supported. To test the proposed asymmetric effect (H3), we compared the sizes of the coefficients for CSR at different levels of military veteran support identity. When 20 military veteran support identity was 1 versus 7, the coefficient for CSR was larger in the absolute size (b = 1.06 [-.68] when military veteran support identity was 1 [7]). Also, when military veteran support identity was 2 versus 6, the coefficient for CSR was larger in the absolute size (b = .77 [-.39] when military veteran support identity was 2 [6]). This suggests that the effect of incongruent (vs. congruent) CSR on increasing (vs. decreasing) dishonest behavior is bigger, supporting H3. Figure 1. Moderating role of military veteran support identity on dishonest behavior (i.e., difference score) (study 1) Note. Johnson-Neyman region of significance shows that when the consumer’s military veteran support identity was greater than 6.02, dishonest behavior was significantly lower in the CSR present condition than in the CSR absent condition. However, when the consumer’s military veteran support identity was smaller than 2.73, dishonest behavior was significantly higher in the CSR present condition than in the CSR absent condition. 21 Role of other identities. We tested the moderating role of the other identities (university student identity, internalization, symbolization dimension of moral identity, and political ideology) using separate regression analyses. No regression revealed a significant two-way interaction (ps > .19), suggesting that only the cause- related identity affected dishonest behavior. Moderating role of importance of moral identity. To examine the possibility that the moderating role of cause-related identity (H1) may be attenuated for those with a highly important moral identity, we ran a regression analysis using CSR (0 = absent, 1 = present), mean-centered military veteran support identity, mean-centered importance of moral identity (internalization dimension), the three two-way interaction terms, and the three-way interaction term as the independent variables and the difference score as the dependent variable. The three-way interaction of CSR, military veteran support identity, and importance of moral identity was not significant (b = -.01, t(174) = -.03, p = .98). When running the same regression after replacing the internalization dimension of moral identity with the symbolization dimension, the three-way interaction was not significant (b= .32, t(174) = 1.08, p = .28). This suggests that the moderating role of cause-related identity holds regardless of different levels of moral identity. We discuss these results in Discussion. Discussion Using over-reporting of performance to acquire more money from a company, study 1 provides support for H1 and H3. Specifically, we demonstrated that the CSR cause congruent [incongruent] with the consumer’ self-concept significantly reduced [increased] dishonest behavior toward the company. Moreover, we found that CSR 22 had a greater impact on dishonest behavior when the cause-related identity was incongruent rather than congruent. In contrast to H4, we found that these results held across consumers with different levels of importance of moral identity, suggesting that even participants who perceived themselves as highly moral engaged in dishonest behavior when the CSR cause was incongruent with their self-concept. We revisit this issue in studies 3 and 4. One limitation of study 1 may be that we measured cause-related identity after manipulating CSR. Although the CSR manipulation did not affect cause-related identity, in the next study we measured this identity before the CSR manipulation. We also examined cause-related identity in a different cause domain (LGBT support) and examined a different dishonest behavior (wardrobing). We tested the mediating role of anticipatory self-threat (H2) while ruling out alternative explanations: liking for the company, the moral perceptions of the company, and anticipatory satisfaction. Study 2 Method Participants and design. Two hundred forty participants (45.8% male, Mage = 34.13, SD = 11.70) from Amazon MTurk completed a 2 (CSR activity: present vs. absent) by (cause-related identity—LGBT support identity; measured continuously) between-subjects design study in exchange for a small payment. After participants indicated their LGBT support identity, they were randomly assigned to the CSR activity condition. 23 Procedure. Participants responded to three items measuring their LGBT support identity (1= strongly disagree, 7 = strongly agree): (1) I identify strongly as a person who cares about rights of gender minorities (e.g., gay and lesbian, transgender), (2) Being a person who cares about rights of gender minorities is an important part of who I am, and (3) I would describe myself as a person who cares about rights of gender minorities [M = 4.84, SD = 1.81, α = .95]. To mitigate demand effects, we embedded these items with five other identity measures (businessperson identity, athletic person identity, trendy person identity, and identification with local community). After completing unrelated filler items, including a game, participants were asked to complete another survey. All participants read a short description of a clothing retailer, Lawson (see appendix B for description). Participants in the CSR present condition also read a description of Lawson’s CSR activities in the domain of support for LGBT groups. Lawson was rated as among the best in the industry on diversity and inclusion in 2015. Lawson was among the first to adopt a written nondiscrimination policy and to provide benefits for transgender employees and same-sex marriage spouses. Moreover, Lawson has annually donated a portion of its profits to charities representing LGBT communities (lesbian, gay, bisexual, and transgender). Participants indicated their liking for the company with three items: negative/positive, unfavorable/favorable, and dislike/like (7-point scales) (α = .97). Then, all participants read a scenario where they could engage in wardrobing. The scenario stated that: Imagine that you bought a shirt at a local Lawson store. After wearing the shirt once, which looks clean and still has a tag on it, you see another shirt of a similar design and quality for a substantially lower price at another clothing 24 store. The return policy of Lawson is that “Customers can return an unworn merchandise for a full refund. After reading the scenario, participants indicated their intention to act dishonestly, anticipatory self-threat, perceived morality of the company, manipulation check for CSR, and demographics measures (gender and age). The intention to act dishonestly was measured by one item (“How likely would you be to return the shirt to Lawson for a full refund? 1 = very unlikely, 7 = very likely). Two items assessed anticipatory self-threat (1 = not at all, 7 = very much): (1) If I return the shirt to Lawson for a full refund, I would feel bad about myself, and (2) if I return the shirt to Lawson for a full refund, I would feel that I am not a good person (adapted from Gino et al. (2013)) (r = .85). We measured anticipatory guilt and anticipatory satisfaction by asking how they would feel if they returned the shirt for a full refund. Three items (guilty, ashamed, and remorseful) measured anticipatory guilt (α = .96) and another three items (satisfied, excited, and happy) measured anticipatory satisfaction (α = .92). Consistent with our expectation, an exploratory factor analysis on anticipatory self-threat and anticipatory guilt measures showed that they loaded on one factor, suggesting they reflect the same construct of anticipatory self-threat. Due to space constraints, we report only the anticipatory self-threat results. The anticipatory guilt results were similar and are reported in appendix D. Then, participants rated the morality of Lawson with three items: immoral/moral, not trustworthy/trustworthy, and not caring/caring (α = .92). As a manipulation check for CSR activity, participants were asked to indicate how much they agreed with the following two statements: (1) Lawson has a positive record of promoting rights of gender minorities, 25 and (2) Lawson cares for promoting diversity in the domain of gender. (1 = strongly disagree, 7 = strongly agree). These were averaged (r = .94). Results Manipulation check for CSR. We performed a regression analysis using CSR activity (0 = absent, 1 = present), mean-centered LGBT support identity, and the two- way interaction term as the independent variables and the manipulation check as the dependent variable. The main effect of CSR activity was significant (b = 2.24, t(236) = 14.90, p < .001), with participants in the CSR present condition (M = 6.33) perceiving that Lawson supported LGBT groups more than those in the CSR absent condition (M = 4.09). The main effect of LGBT support identity was also significant (b = .18, t(236) = 2.99, p = .003). This suggests that participants who felt that supporting LGBT groups was more congruent with their self-concept perceived Lawson as supporting LGBT groups more than did participants who felt that supporting LGBT groups was incongruent with their self-concept. The two-way interaction was not significant (b = -.09, t(236) = -1.05, p = .29), suggesting that the CSR manipulation worked as intended. Intention for dishonest behavior. Regression analysis revealed no significant main effects of CSR activity (b = -.45, t(236) = -.1.59, p = .11) or LGBT support identity (b = .11, t(236) = 1.00, p = .32). However, as predicted by H1, the two-way interaction was significant (b = -.47, t(236) = -3.04, p = .003) (see Figure 2). The Johnson-Neyman analysis revealed that CSR (vs. no-CSR) significantly decreased intention to act dishonestly for participants with LGBT support identity greater than 5.08 (MCSRpresent = 3.44 vs. MCSRabsent = 4.00; bJN = -.56, SE = .28, p = .050). In 26 contrast, CSR (vs. no-CSR) significantly increased dishonest behavior for participants with LGBT support identity smaller than 1.35 (MCSRpresent = 4.79 vs. MCSRabsent = 3.59; bJN = 1.20, SE = .61, p = .050). Figure 2. Moderating role of LBGT support identity on intention for dishonest behavior (study 2) Note. Johnson-Neyman region of significance shows that when the consumer’s LGBT support identity was greater than 5.08, the intention to act dishonestly was significantly lower in the CSR present condition than in the CSR absent condition. However, when the consumer’s LGBT support identity was smaller than 1.35, the intention to act dishonestly was significantly higher in the CSR present condition than in the CSR absent condition. To test the proposed asymmetric effect (H3), we compared the sizes of the coefficients for CSR at different levels of LGBT support identity. When LGBT support identity was 1 versus 7, the coefficients for CSR were not different in the absolute size (b = 1.37 [-1.46] when LGBT support identity was 1 [7]). When LGBT 27 support identity was 2 versus 6, the coefficients for CSR were not different in the absolute size (b = 0.90 [-0.99] when LGBT support identity was 2 [6]). We think that this finding may have been caused by the continuous measure of cause-related identity, which did not clear distinguish between neutral and incongruent cause- related identity when participants indicated “1 = strongly disagree” to the items such as “I identify strongly as a person who cares about rights of gender minorities (e.g., gay and lesbian, transgender).” This could reduce the effect size of CSR on dishonest behavior at the left-hand incongruent cause-related identity side. We discuss this issue more in the Discussion and address it in study 3. Anticipatory self-threat. Regression analysis revealed a marginally significant main effect of CSR activity (b =.49, t(236) = 1.83, p = .068; MCSRpresent = 3.85 vs. MCSRabsent = 3.36) and no significant effect of LGBT support identity (b = -.11, t(236) = -1.07, p = .29) on anticipatory self-threat. Importantly, the two-way interaction was significant (b = .47, t(236) = 3.17, p = .002). The Johnson-Neyman analysis revealed that CSR (vs. no-CSR) significantly increased anticipatory self-threat for participants with LGBT support identity greater than 4.93 (MCSRpresent = 3.88 vs. MCSRabsent = 3.35; bJN = .52, SE = .27, p = .050). In contrast, CSR (vs. no-CSR) significantly decreased anticipatory self-threat for participants with LGBT support identity less than 1.34 (MCSRpresent = 2.61 vs. MCSRabsent = 3.75; bJN = -1.34, SE = .58, p = .050). Mediating role of anticipatory self-threat. We tested mediation via anticipatory self-threat (H2) using the PROCESS macro for SPSS (Model 8; Hayes, 2018). We included CSR (0 = absent, 1 = present) as the independent variable, LGBT support identity as the moderator, anticipatory self-threat as the mediator, and 28 intention for dishonest behavior as the dependent variable. To generate 95% confidence intervals (CIs), we used 5,000 bootstrap samples. Since we expected different indirect effects of anticipatory self-threat depending on the consumer’s cause-related identity (i.e., LGBT support identity), we tested indirect effects at every interval of LGBT support identity (see Table 1). The results show that when LGBT support identity was greater than or equal to 5, the indirect effect of anticipatory self- threat was significantly negative (b = -.37, SE = .18, 95% CI: -.722 to -.019). When LGBT support identity was 1, the indirect effect of anticipatory self-threat was significantly positive (b = .86, SE = .43, 95% CI: .012 to 1.699). That is, when CSR was congruent (incongruent) with the consumers’ self-concept, CSR reduced (increased) dishonest consumer behavior by increasing (reducing) anticipatory self- threat. This provides support for H2. Table 1. Indirect effect of anticipatory self-threat on intention for dishonest behavior at different levels of LGBT support identity LGBT Support 95% CI 95% CI Identity Effect Boot SE Lower Upper Limit Limit 1 0.861 0.434 0.012 1.699 2 0.553 0.340 -0.112 1.212 3 0.245 0.256 -0.257 0.742 4 -0.063 0.194 -0.443 0.323 5 -0.371 0.178 -0.722 -0.019 6 -0.679 0.218 -1.124 -0.263 7 -0.987 0.293 -1.578 -0.428 Note: Significant indirect effects are in bold. 29 Alternative explanations: liking for the company, perceived morality of the company, and anticipatory satisfaction. We conducted the same regression analysis with liking for the company as the dependent variable. The two-way interaction of CSR and LGBT support identity was significant (b = .32, t(236) = 3.66, p < .001). The mean pattern of liking for the company was similar to that of intention for dishonest behavior. However, the mediation analysis indicated that the indirect effect of liking for the company was not significant across any level of LGBT support identity. Thus, liking for the company did not serve as a mediator. For the perceived morality of the company, the same regression analysis revealed a significant two-way interaction (b = .24, t(236) = 2.84, p = .005). However, the mean pattern was different from that of intention for dishonest behavior and anticipatory self-threat, suggesting that the perceived morality of the company is not likely to serve as a mediator. Specifically, when the participant’s LGBT support identity was low, there was no significant difference in the perceived morality of the company between CSR present and CSR absent conditions (e.g., when LGBT support identity was 1, MCSRpresent = 4.73 vs. MCSRabsent = 4.80, p = .84). Moreover, the mediation analysis indicated that the indirect effect of the perceived morality of the company was not significant across any level of LGBT support identity. Thus, the perceived morality of the company did not serve as a mediator. For anticipatory satisfaction, regression analysis revealed that none of the treatments was significant (ps > .23), indicating that they did not affect anticipatory satisfaction. Moreover, the mediation analysis was insignificant across any level of 30 LGBT support identity. Thus, anticipatory satisfaction did not mediate these effects. Moreover, when we tested all four mediating variables simultaneously (anticipatory self-threat, liking for the company, perceived morality of the company, and anticipatory satisfaction), the results remained the same. In other words, only anticipatory self-threat served as a significant mediator. Discussion Using support for LGBT groups as the cause-related identity, the results of this study provide support for H1 and H2. Anticipatory self-threat was a significant mediator when the CSR was congruent or incongruent with consumers’ self-concept. This provides support for our identity-based account of CSR. Study 2 also ruled out alternative explanations based on liking for the company, perceived morality of the company, and anticipatory satisfaction. This shows that moral perception of the company and liking for the company does not explain the relationship between CSR and dishonest behavior. One limitation of the first two studies may be that the interpretation of the continuous measure of cause-related identity might be ambiguous. Participants may have checked “strongly disagree” to statements such as “I would describe myself as a person who cares about and supports military veterans” to indicate that the cause was either neutral to or incongruent with their self-concept. This may explain why we did not find support for the proposed asymmetric effect in study 2. To separate neutral and incongruent identity more clearly, we use a categorical measure in the next study. Another limitation is that the anticipatory self-threat measures (e.g., If I return the shirt to Lawson for a full refund, I would feel bad about myself) did not allow us 31 to capture the possibility that consumers may anticipate feeling good about themselves when considering acting dishonestly to the company engaged in the identity-incongruent CSR activity. We address this in the next study. Study 3 The primary objective of study 3 was to replicate the findings of studies 1 and 2 by using a different measure of cause-related identity that allows us to better distinguish between neutral and incongruent identity. Also, we used a different measure of anticipatory self-threat to capture the possibility that consumers may anticipate feeling good about themselves after acting dishonestly towards the company. We also ruled out alternative explanations based on liking for the company and perceived morality of the company. We further examined whether moral identity moderated the effect of CSR on dishonest consumer behavior. In study 1, we found that stronger moral identity did not attenuate the effect of CSR on dishonest consumer behavior. To further examine this, we measured both moral identity and the benevolence subscale of Schwartz’s value scale (Schwartz & Bardi, 2001). The benevolence subscale is intended to assess the importance of values related to the preservation and enhancement of the welfare of other people, such as honesty, helpfulness, and loyalty (Schwartz & Bardi, 2001). We wanted to assess whether it was our specific measures of the importance of moral identity which may have led to the null effects in study 1. 32 Method Participants and design. A total of 453 participants (39.3% male, Mage = 36.12, SD = 11.65) from Amazon MTurk completed a 2 (CSR activity: present vs. absent) by 3 (cause-related identity: LGBT support congruent, neutral, incongruent - measured) between-subjects design study in exchange for a small payment. After participants indicated their LGBT support identity, they were randomly assigned to the CSR activity condition. Procedure. Participants first completed a “survey on various social issues.” They saw a list of three social causes: promoting rights of gender minorities such as LGBT rights (focal identity), ensuring environmental sustainability, and helping others in need. In order to measure cause-related identity, they were asked to drag each cause into one of three boxes, as in the pretest. The three categories indicate congruent, incongruent, and neutral cause-related identity, respectively. Consistent with the pretest, 56.3% of participants (n = 255) indicated that LGBT support was congruent with their self-concept, 17.9% (n = 81) indicated that LGBT support was incongruent, and 25.8% (n = 117) indicated that LGBT support was neutral to their self-concept. After they completed the survey and other filler surveys, all participants read the description about the clothing retailer, Lawson. Participants in the CSR present condition also read the description of Lawson’s CSR activities in the domain of support for LGBT groups. Then, all participants read the scenario where they could engage in wardrobing. The descriptions and scenarios used in study 2 were used. 33 They then indicated their intention to act dishonestly, anticipatory self-threat, anticipatory guilt (α = .93), liking for the company (α = .98), perceived morality of the company (α = .92), manipulation check for CSR (r = .88), individual difference measures (internalization dimension of moral identity and benevolence subscale of Schwartz’s value scale), gender, and age. We measured anticipatory self-threat by asking how participants would feel if they returned the shirt to Lawson using two bipolar scales: “bad about myself – good about myself” and “I am a bad person – I am a good person” (reverse-coded, r = .86). For intention to act dishonestly, anticipatory guilt, liking for the company, perceived morality of the company, and manipulation check for CSR, the same measured used in study 2 was used. The benevolence subscale was measured by asking participants to rate the importance of five values (honest, helpful, forgiving, loyal, and responsible) from -1 (opposed to my principles) to 7 (of supreme importance) (α = .84; Schwartz & Bardi, 2001). Results Manipulation check for CSR. We conducted a 2 (CSR activity: present vs. absent) by 3 (cause-related identity: LGBT support congruent, neutral, incongruent - measured) ANOVA analysis on the CSR manipulation check measure. The main effect of CSR activity was significant (F(1, 447) = 265.85, p < .001, η 2p = .37), showing that participants in the CSR present condition (M = 6.19) perceived that Lawson supported LGBT groups more than those in the CSR absent condition (M = 4.14). The main effect of LGBT support identity was significant (F(2, 447) = 4.63, p = .010, η 2p = .02; Mcongruent = 5.29 vs. Mneutral = 5.05 vs. Mincongruent = 4.90). Unexpectedly, the two-way interaction between CSR and LGBT support identity was 34 also significant (F(2, 447) = 7.48, p = .001, η 2p = .03). An examination of the means revealed that in all three identity conditions, the simple main effect of CSR was significant (ps < .001; Congruent: MCSRpresent = 6.47 vs. MCSRabsent = 4.12; Neutral: MCSRpresent = 6.03 vs. MCSRabsent = 4.09; Incongruent: MCSRpresent = 5.55 vs. MCSRabsent = 4.26). Thus, the CSR manipulation worked in each condition. Just to make sure that there were no confounds, we assessed whether the manipulation check measures mediated the effect of CSR on intention for dishonest behavior in all three cause- related identity conditions. It did not, suggesting that the CSR manipulation worked as intended. Intention for dishonest behavior. A two-way ANOVA revealed an insignificant main effect of CSR activity (F(1, 447) = .97, p = .33) and a significant main effect of LGBT support identity (Mcongruent = 3.00 vs. Mneutral = 3.36 vs. Mincongruent = 4.07; F(2, 447) = 7.45, p = .001, η 2p = .03). However, these effects were qualified by the significant two-way interaction (F(2, 447) = 6.25, p = .002, η 2p = .03; see Figure 3). Follow-up analyses showed that in the congruent identity condition, CSR (vs. no-CSR) significantly decreased intention to act dishonestly (MCSRpresent = 2.70, SD = 1.97 vs. MCSRabsent = 3.30, SD = 2.23; F(1, 447) = 4.78, p = .029, d = .29). In contrast, in the incongruent identity condition, CSR (vs. no-CSR) significantly increased intention to act dishonestly (MCSRpresent = 4.77, SD = 2.26 vs. MCSRabsent = 3.39, SD = 2.50; F(1, 447) = 7.99, p = .005, d = .58). In the neutral identity condition, CSR (vs. no-CSR) did not affect intention to act dishonestly (MCSRpresent = 3.31, SD = 2.33 vs. MCSRabsent = 3.41, SD = 2.27; F(1, 447) = .06, p = .81). This provides further support for H1. Moreover, the effect size of CSR was bigger in the incongruent (d = 35 .58) than in the congruent identity condition (d = .29), providing further support for the proposed asymmetric effect (H3). Figure 3. Moderating role of LGBT support identity on intention for dishonest behavior (study 3) 7 6 5 4 3 2 1 Incongruent cause- Neutral cause-related Congruent cause- related identity identity related identity CSR absent CSR present Anticipatory self-threat. A two-way ANOVA revealed an insignificant main effect of CSR activity (F(1, 447) = 1.05, p = .31) and a significant main effect of LGBT support identity (Mcongruent = 4.83 vs. Mneutral = 4.62 vs. Mincongruent = 4.14; F(2, 447) = 4.85, p = .008, η 2p = 0.02). More importantly, there was a significant two-way interaction (F(2, 447) = 6.60, p = .002, η 2p = 0.03). CSR (vs. no-CSR) significantly increased anticipatory self-threat in the congruent condition (MCSRpresent = 5.09, SD = 1.55 vs. MCSRabsent = 4.58, SD = 1.75; F(1, 447) = 5.32, p = .022, d = .30), but significantly decreased anticipatory self-threat in the incongruent condition (MCSRpresent = 3.58, SD = 1.90 vs. MCSRabsent = 4.70, SD = 1.87; F(1, 447) = 8.20, p = 36 Intention for Dishonest Behavior .004, d = .59). Although not statistically significant (t(39) = 1.42, p = .17), anticipatory self-threat (M = 3.58) was directionally lower than the midpoint of the scale, four. This suggests that when the CSR cause is incongruent, individuals anticipate they would feel good from acting dishonestly. In the neutral condition, anticipatory self-threat was unaffected (MCSRpresent = 4.65, SD = 1.78 vs. MCSRabsent = 4.60, SD = 2.00; F(1, 447) = .02, p = .89). Mediating role of anticipatory self-threat. We tested mediation via anticipatory self-threat (H2) (Model 8). We included CSR (0 = absent, 1 = present) as the independent variable, LGBT support identity as the moderator, anticipatory self- threat as the mediator, and intention for dishonest behavior as the dependent variable. The results show that the indirect effect of anticipatory self-threat was significantly negative in the congruent condition (b = -.44, SE = .18, 95% CI: -.801 to -.092), and was significantly positive in the incongruent condition (b = .97, SE = .37, 95% CI: .256 to 1.711). In the neutral condition, the indirect effect of anticipatory self-threat was not significant (b = -.04, SE = .31, 95% CI: -.646 to .564). Thus, H2 is supported. Alternative explanations: liking for and perceived morality of the company. For liking for the company, a two-way ANOVA revealed a significant main effect of LGBT support identity (F(2, 447) = 35.69, p < .001, η 2p = 0.14) and a significant two- way interaction (F(2, 447) = 28.96, p < .001, η 2p = 0.11). CSR (vs. no-CSR) significantly increased liking for the company in the congruent condition (MCSRpresent = 6.21, SD = 1.09 vs. MCSRabsent = 5.08, SD = 1.47; F(1, 447) = 39.20, p < .001, d = .87) and decreased liking in the incongruent condition (MCSRpresent = 3.38, SD = 1.84 vs. MCSRabsent = 4.93, SD = 1.80; F(1, 447) = 23.18, p < .001, d = .85). In the neutral 37 condition, liking for the company was unaffected (MCSRpresent = 4.77, SD = 1.37 vs. MCSRabsent = 4.98, SD = 1.59; F(1, 447) = .63, p = .43). The mediation analysis revealed that in all identity conditions, the indirect effects of liking for the company were not significant (congruent: b = -.05, SE = .09, 95% CI: -.224 to.115; neutral: b = .01, SE = .03, 95% CI: -.040 to .086; incongruent: b = .07, SE = .12, 95% CI: -.151 to .339). When we tested the mediating role of anticipatory self-threat and liking for the company simultaneously, only the anticipatory self-threat was significant in the congruent and incongruent identity conditions. For perceived morality of the company, we found the similar pattern of results to that of liking for the company. The mediating role of perceived morality of the company was not significant when tested as a sole mediator or when tested with anticipatory self-threat as parallel mediators. Detailed results are reported in appendix E. Moderating role of importance of moral identity and benevolence. We first confirmed that the manipulation of CSR did not influence importance of moral identity and benevolence (ps > .17). To examine the possibility that the moderating role of cause-related identity (H1) may be attenuated for those with a highly important moral identity, we ran a regression analysis using CSR (0 = absent, 1 = present), two dummies for LGBT support identity (dummy1: 1 = congruent, 0 = neutral, 0 = incongruent; dummy2: 0 = congruent, 0 = neutral, 1 = incongruent), mean-centered importance of moral identity, the five two-way interaction terms, and the two three-way interaction terms as the independent variables and intention to act dishonestly as the dependent variable. Replicating the findings of study 1, the three- 38 way interactions of CSR, LGBT support identity, and importance of moral identity were not significant (ps > .32). When we ran the same regression after replacing importance of moral identity with a composite value of five benevolence measures, the three-way interactions were not significant (ps > .28). These results further suggest that the moderating role of cause-related identity (H1) does not change depending on different levels of importance of moral identity and values. Discussion Using LGBT support as the cause-related identity, this study provides support for the first three hypotheses. Anticipatory self-threat was a significant mediator of the effects of CSR and cause-related identity on dishonest behavior. We also ruled out alternative explanations based on liking for the company and perceived morality of the company. In fact, our effects remained significant even when controlling for the influence of liking or perceived morality (ps for the interaction of CSR and cause- related identity < .009). Thus, we find support for our identity-based account of CSR rather than a liking or morality-based account. Interestingly, in the study 1 and 3 where we measured the importance of moral identity, moral identity did not attenuate the effects of incongruent cause-related identity on dishonest behavior, as predicted by H4. This suggests that moral identity was not made salient even for participants with a highly important moral identity. Although these findings seem inconsistent with prior work showing that moral identity increases moral behavior (Aquino & Reed, 2002), they reflect an important nuance in the literature. Aquino, Freeman, Reed, Lim, and Felps (2009) found that when a situation activated motives conflicting with morality, such as when monetary 39 rewards were salient, higher moral identity did not reduce dishonest behavior because moral identity was not accessible even to those with a highly important moral identity. This suggests that individual differences in moral identity are not appropriate to capture salience of moral identity in our context. Thus, in study 3 we manipulate salience of moral identity as a state rather than a trait. Moreover, it is possible that the measure of importance of moral identity might have not been sensitive enough to capture the importance of self-concepts relevant to dishonest behavior. Moral identity incorporates a variety of moral values including care, fairness, friendliness, and honesty. It is possible that people could differ in terms of which specific value they regard as an important part of their self- concept. For example, work by Haidt and his colleagues (e.g., Graham et al., 2009) identified several foundational moral values, arguing that it is important to distinguish different moral values. To address this possibility, we considered the following two moral identities: honesty and care. If honesty, not care, is indeed relevant to one’s decision to act dishonestly, we should find that only the salience of honesty identity, not of care identity, attenuates H1. However, if both honesty and care are relevant because they are schematically closely associated with a higher-level construct of morality (Aquino & Reed, 2002), the salience of both honesty and care identities should attenuate H1. 40 Study 4 Method Participants and design. A total of 491 participants (44.2% male, Mage = 38.82, SD = 13.62) from Amazon MTurk completed a 2 (cause-related identity: refugee support congruent, incongruent – measured) x 3 (salient identity: competence, honesty, care) between-subjects design study in exchange for a small payment. Support for refugees was the CSR cause, while lying to get a discount at a theme park was the context. Since our focus was on testing whether salient moral identities attenuated the effect of incongruent cause-related identity on increasing dishonest behavior, we included only the CSR present condition. If H4 is supported, we should find a significant interaction between cause-related identity and salience of moral identities. Specifically, the effect of incongruent (vs. congruent) cause-related identity on increasing dishonest behavior should be attenuated in the honesty and care (vs. competence) identity salience conditions. Procedure. Participants read that they would complete unrelated several short surveys. As in study 3, participants first completed a “social issues survey.” They saw a list of three social causes: providing support for refugees (focal identity), ensuring environmental sustainability, and gun control. Then, they were asked to drag each cause into one of the two boxes that represent either congruent or incongruent cause- related identity. 69.7% of participants (n = 342) indicated that refugee support was congruent with their self-concept, and 30.3% (n = 149) indicated that refugee support was incongruent. Note that they did not have the option of picking neutral identity. 41 Since neutral identities are weak, forcing participants to identify as congruent or incongruent makes this a conservative test of the hypothesis. Then, participants completed a “writing task,” which manipulated salient identities (procedure adapted from Reed, Aquino, & Levy, 2007). Participants in each salient identity condition were given a word (honesty, care, or competence) as a characteristic of a person, respectively. They were told to take a few moments to think about this characteristic and write a brief story about themselves using this word at least once. Competence is one of the self-enhancement values, which are diametrically opposed to self-transcendent moral values such as honesty and care (Schwartz & Bardi, 2001). Thus, we expected that salience of competence identity, unlike of honesty and care identities, would not heighten anticipatory self-threat elicited from considering acting dishonestly and thus not attenuate the moderating role of cause-related identity. Next, participants completed a “consumer behavior survey.” They were asked to read a news article on a local theme park, Great Wonderland. The article described Great Wonderland’s CSR activities in the domain of refugee support (see appendix B for the news article). Then, all participants read a scenario where they could lie about their child’s age to get a discount on the ticket. The scenario stated: Imagine that you are on the way to Great Wonderland, with your child. You and your child arrive at a ticket counter. You read the posted information about ticket prices. It says that children of ages 3-6 are eligible to receive a 20% discount on the adult ticket fair. Your child is 7 years old, but looks younger than his age. They indicated their intention to act dishonestly, anticipatory self-threat (r = .87), expected harm of individual’s action to the company, liking for the company (α = .97), gender, and age. The intention to act dishonestly was measured by one item 42 (“How likely would you be to say that your child is 6 years old at Great Wonderland? 1 = very unlikely, 7 = very likely). We measured anticipatory self-threat and liking for the company as in study 3. We measured the expected harm to the company using one item (“If you say that your child is 6 years old at Great Wonderland, how much would Great Wonderland be harmed by your actions? 1 = not at all, 7 = very much). We included this measure to rule out an alternative account that the effect of CSR on dishonest behavior is driven by a concern for causing harm to the company. Results Intention for dishonest behavior. A two-way ANOVA revealed a significant main effect of refugee support identity (Mcongruent = 3.18 vs. Mincongruent = 3.87; F(1, 485) = 11.30, p = .001, η 2p = .02) and a significant main effect of salient identities (Mcompetence = 3.59 vs. Mhonesty = 3.32 vs. Mcare = 3.25; F(2, 485) = 3.23, p = .040, η 2p = 0.01). However, these effects were qualified by the significant two-way interaction (F(2, 485) = 3.15, p = .044, η 2p = .01; see Figure 4). Follow-up analyses showed that in the competence identity salience condition, replicating the findings from studies 1- 3, participants with incongruent (vs. congruent) refugee support identity showed significantly higher intention for acting dishonestly toward the company (Mcongruent = 3.18, SD = 2.25 vs. Mincongruent = 4.72, SD = 2.34; F(1, 485) = 15.11, p < .001, d = .67). In contrast, in the honesty and care salience conditions, intention for dishonest behavior was not significantly different between incongruent and congruent conditions (honesty condition: Mcongruent = 3.19, SD = 2.25 vs. Mincongruent = 3.53, SD = 2.28; F(1, 485) = .88, p = .35; care condition: Mcongruent = 3.15, SD = 2.18 vs. Mincongruent = 3.50, SD = 2.10; F(1, 485) = .80, p = .37). This result shows that when 43 moral identities (both honesty and care) were externally made salient, the effect of cause-related identity on dishonest behavior was attenuated (H4), providing further support for the anticipatory self-threat mechanism. Moreover, the fact that both honesty and care identity salience attenuated the effect suggests that both honesty and caring are associated with a higher level construct of morality (Aquino & Reed, 2002). Figure 4. Interactive effect of refugee support identity and salient identities on intention for dishonest behavior (study 4) 7 6 5 4 3 2 1 Competence identity Honesty identity Care identity salience salience salience Congruent cause-related identity Incongruent cause-related identity We further compared intention for dishonest behavior between different salient identity conditions within each level of refugee support identity condition. In the congruent condition, the simple main effect of salient identities on dishonest behavior was not significant (F(2, 485) = .01, p = .99). However, in the incongruent condition, the simple main effect was significant (F(2, 485) = 4.45, p = .012, η 2p = 44 Intention for Dishonest Behavior .02). Specifically, intention for dishonest behavior was significantly higher in the competence (M = 4.72) than in the honesty (M = 3.53; F(1, 485) = 7.11, p = .008, d =.52) and care identity salient conditions (M = 3.50; F(1, 485) = 6.66, p = .010, d = .55). This suggests that even when individuals have incongruent cause-related identity, salient moral identities prohibited individuals from acting dishonestly, supporting H4. Moreover, we conducted a series of planned contrasts by collapsing two salient moral identity (honest identity and moral identity) conditions. Supporting our prediction, the two-way interaction was significant (F(1, 485) = 6.30, p = .012, η 2p = .01). Follow-up contrast analyses showed that in the moral identity salience condition, intention for dishonest behavior was not significantly different between incongruent and congruent conditions (F(1, 485) = 1.68, p = .20). Also, in the incongruent condition, intention for dishonest behavior was significantly higher in the competence than in the moral identity salient conditions (F(1, 485) = 8.87, p = .003, η 2p = .02). These results provide convergent evidence that salience of moral identity, measured by combining honesty identity and care identity, attenuated the backfiring effect of incongruent cause-related identity on intention for dishonest behavior. Anticipatory self-threat. A two-way ANOVA revealed a significant main effect of refugee support identity (Mcongruent = 5.18 vs. Mincongruent = 4.72; F(1, 485) = 7.74, p = .006, η 2p = .02) and an insignificant main effect of salient identities (F(2, 485) = 1.40, p = .25). Although the two-way interaction was not significant (F(2, 485) = 1.30, p = .27), the pattern of means is similar to the pattern of means for intention for dishonest behavior. In the competence identity salience condition, participants 45 with incongruent (vs. congruent) identity showed significantly lower level of anticipatory self-threat (Mcongruent = 5.18, SD = 1.59 vs. Mincongruent = 4.28, SD = 1.87; F(1, 485) = 8.08, p = .005, d = .52). In contrast, in the honesty and care identity salience conditions, anticipatory self-threat was not significantly different between incongruent and congruent conditions (honesty condition: Mcongruent = 5.21, SD = 1.84 vs. Mincongruent = 4.93, SD = 1.98; F(1, 485) = 1.00, p = .32; care condition: Mcongruent = 5.15, SD = 1.78 vs. Mincongruent = 4.87, SD = 1.76; F(1, 485) = .84, p = .36). This shows that when moral identities were salient, the effect of cause-related identity on anticipatory self-threat was attenuated. Moreover, we conducted a series of planned contrasts by collapsing two salient moral identity (honest identity and moral identity) conditions. The two-way interaction was not significant (F(1, 485) = 2.59, p = .11). However, consistent with our predictions, follow-up contrast analyses showed that in the moral identity salience condition, anticipatory self-threat was not significantly different between incongruent and congruent conditions (F(1, 485) = 1.82, p = .18). Also, in the incongruent condition, anticipatory self-threat was significantly lower in the competence than in the moral identity salient conditions (F(1, 485) = 3.67, p = .056, η 2p = .01). Mediating role of anticipatory self-threat. We tested mediation via anticipatory self-threat (H2) (Model 8). We included refugee support identity (0 = congruent, 1 = incongruent) as the independent variable, salient identities as the moderator, anticipatory self-threat as the mediator, and intention for dishonest behavior as the dependent variable. Consistent with our predictions, the results show that only in the competence identity salience condition, the indirect effect of 46 anticipatory self-threat was significant (b = .78, SE = .27, 95% CI: .261 to .1.319). In the honesty and care salience identity conditions, the indirect effect was not significant (honesty: b = .25, SE = .27, 95% CI: -.284 to .793; care: b = .25, SE = .27, 95% CI: -.279 to .779). Alternative explanations: liking for the company and expected harm. A two- way ANOVA only revealed the significant main effect of refugee support identity (Mcongruent = 6.07 vs. Mincongruent = 4.45; F(1, 485) = 186.48, p < .001, η 2p = 0.28;. However, the main effect of salient values and the two-way interaction were not significant (ps > .15). This pattern of the means for liking for the company was different from that for intention for dishonest behavior, suggesting that it is unlikely to drive the effect. Moreover, in all three salient identity conditions, liking for the company did not mediate the effect of refugee support identity on dishonest behavior when it was tested as a sole mediator or when it was tested with anticipatory self- threat as parallel mediators. With regard to the expected harm to the company from dishonest behavior, the two-way interaction between refugee support identity and salient values was not significant (F(2, 485) = .30, p = .74). Moreover, in all three salient identities conditions, expected harm did not mediate the effect of refugee support identity on dishonest behavior when it was tested as a sole mediator or along with anticipatory self-threat as parallel mediators. Discussion Using refugee support as the cause-related identity, this study shows that when moral identities (both honesty and care) were situationally made salient, the 47 backfiring effect of incongruent CSR on increasing dishonest behavior was attenuated because salient moral values heightened anticipatory self-threat. This finding provides further support for the anticipatory self-threat mechanism. We further ruled out alternative explanations based on liking for the company and expected harm of individual’s dishonest behavior to the company. General Discussion The objective of the paper was to examine the effects of polarizing CSR on consumers’ decisions to act dishonestly toward the company. Across four studies using different dishonest behaviors and CSR causes, we found that consumers’ cause- related identity moderated the effect of CSR on dishonest behavior toward the company. When the CSR cause was congruent with the consumer’s self-concept, CSR increased anticipatory self-threat from acting dishonestly toward the company; as a result, dishonest behavior decreased. Thus, CSR helped the company. When the CSR cause was incongruent, CSR decreased anticipatory self-threat, thereby hurting the company. Moreover, we found that CSR hurt more than it helped. The effect of incongruent CSR on increasing dishonest behavior was bigger than the effect of congruent CSR on decreasing dishonest behavior. Finally, building on the anticipatory self-threat mechanism, we identified a boundary condition: the situational salience of moral identity. When moral identity, which can be both honesty identity and care identity, was situationally salient, the backfiring effect of incongruent cause-related identity was attenuated. We discuss the theoretical and managerial implications of the findings. 48 Theoretical Contribution Corporate social responsibility. Our findings contribute to the literature on CSR. Prior work on CSR has focused on causes that are largely non-controversial. Most of this work has found positive results of CSR on a variety of stakeholders (e.g., Brammer et al., 2007; De Roeck et al., 2014; Sen & Bhattacharya, 2001). When negative effects have been found, they have been due to the perceived inauthenticity of CSR, such as perceived egoistic firm motives (Mallory & Rupp, 2014; McShane & Cunninghamm, 2012). We augment this literature by demonstrating the effects of polarizing CSR. By doing so, we answer a recent call for more research on the negative effects of CSR (Rupp & Mallory, 2015). We find that CSR can have either positive or negative effects when the company supports a cause that is polarizing. Polarizing issues are theoretically important because polarizing CSR is likely to activate either congruent or incongruent identities, thereby leading to different effects. In this way, we add to prior CSR research that does not consider oppositional identity (e.g., Sen & Bhattacharya, 2001). Those studies show that CSR that is weakly related to the consumer’s identity attenuates positive effects rather than creating negative effects. These results likely stem from the fact that those studies measured neutral rather than incongruent identities. By separating neutral from incongruent identities, we show that CSR can backfire. Moreover, we provide insights into the psychological process through which CSR affects dishonest behavior by identifying a mediator – anticipatory self-threat. Building on this mechanism, we show that moral identity that is situationally salient 49 at the time when consumers make a decision to act dishonestly can attenuate the backfiring effect of CSR. Future research could examine other ways to attenuate the backfiring effect of CSR on increasing dishonest behavior. In this research, we focused on a hypothetical company, and thus the consumer–company identity was weak. We showed that when the consumer–company identity was weak, consumers increased or decreased their dishonest behavior toward the company depending on their cause-related identity. Future research could examine the interaction between the consumer–company identity and cause-related identity when the former is strong. Previous work shows that affective commitment with the company reduces dishonest behavior toward the company (Steenhaut & Van Kenhove, 2003) by increasing consumers’ guilt-related feelings. This suggests that affective commitment with the company could heighten anticipatory self-threat, thus attenuating the backfiring effect of CSR. Further, future research could examine how consumers respond when companies support multiple causes, some of which are congruent while others are incongruent with their identity. For instance, Starbucks increased support for military veterans after some individuals objected to the company’s support for hiring refugees. As shown in our work, both of these issues are controversial. Do these offset each other, or do they make everyone unhappy? Dishonest behavior. The research also contributes to the literature on dishonest behavior by identifying a company’s CSR as an antecedent. Prior literature on dishonest behavior has focused on factors related to individual characteristics (e.g., creativity) and contexts (e.g., salience of moral values) (e.g., Gino & Ariely, 2012; 50 Mazar et al., 2008). Relatively little research has examined how the characteristics of the company affect dishonest behavior (Wirtz & McColl-Kennedy, 2010). We address this gap by showing that the fit between the characteristics of the individual (cause-related identity) and the company (CSR) is an important antecedent of company-directed dishonest behavior. Investigating the characteristics of the company is important since companies have control over what they do. Future research may identify other company characteristics that might affect dishonest behavior. Moreover, we contribute to the literature on dishonest behavior by examining the underlying process by which a company’s CSR influences dishonest behavior. Prior work on prosocial behavior has long debated whether prosocial behavior is driven by egoistic motives (e.g., to feel a warm glow) or altruistic motives (e.g., genuine concern for others) (e.g., Batson et al., 1998). A similar question can be asked about dishonest behavior. Individuals can act more or less dishonestly toward the company engaged in CSR due to egoistic motives (e.g., to avoid feeling bad about themselves) or altruistic motives (e.g., concern for the company). This research provides evidence for the former motive. We found that CSR influenced dishonest behavior by affecting how individuals anticipated that they would view themselves after acting dishonestly. This is consistent with prior work showing that self-concept maintenance plays a key role in influencing dishonest behavior (e.g., Mazar et al., 2008). This research enhances our understanding of the effect by examining the role of importance of moral identity in relation to cause-related identity. Interestingly, 51 individual differences in moral identity did not moderate the effect of cause-related identity on dishonest behavior (study 1 and 3). Moreover, moral identity was not correlated with dishonest behavior (study 1), although moral identity was correlated with intention for dishonest behavior (study 3). The presence of a financial incentive leads individuals to lie regardless of their moral identity, suggesting that more research examining under which conditions the importance of moral identity influences dishonest behavior is needed (Reynolds & Ceranic, 2007). We show that when moral identities are situationally made salient, the moderating role of cause- related identity on dishonest behavior was attenuated (study 4). The paper also makes a contribution to research on how individuals respond to a company’s activity. Our findings suggest that individuals may engage in dishonest behavior against a company when abandonment is difficult. An individual who likes the company’s products (e.g., is addicted to Starbucks coffee) may also find it difficult to abandon the company. It could be hard for employees to leave the company by quitting their current job. For such individuals and employees, acting dishonestly may allow them to maintain their self-concept as someone who stands for/against a particular cause. In the case of Starbucks, for instance, the individual whose self-concept is incongruent with supporting refugees may behave dishonestly by claiming that a drink was not made to their taste (even when it was), thereby getting a free drink, by claiming they are getting a coffee refill when they are actually getting a new cup, or by picking up someone else’s order. 52 Managerial Contribution More and more companies are taking stands on controversial social issues in their CSR initiatives. Our findings suggest that since individuals are divided on a variety of social issues, companies must take into account oppositional identities when selecting CSR initiatives. We provide empirical evidence about how individuals respond to companies’ polarizing CSR. Specifically, our finding that CSR can increase dishonest behavior means that companies must be careful about how they deal with target individuals who oppose a particular CSR cause. The obvious strategy would be for the company to support only those causes favored by the target market. Although this may be possible for niche companies, such as Patagonia, it is more difficult for companies with a broader audience, such as Target or Starbucks. These companies clearly have to weigh the benefits and costs of supporting polarizing issues. Our finding that the magnitude of the effect of CSR on increasing and decreasing dishonest behavior is asymmetric provides companies with useful input for calculating the benefits and costs. Moreover, building on the anticipatory self-threat mechanism, we offer insights regarding how to reduce the backfiring effect of CSR. Companies can utilize contextual factors that can heighten anticipatory self-threat, such as temporarily making moral identity salient, in places where dishonest behaviors commonly occur (e.g., customer service areas, ticket booths). Although we focus on individuals’ dishonest behavior, we expect that a similar effect may occur for employees. Companies can highlight ethical company culture and norms to reduce the backfiring effect of CSR on increasing employees’ dishonest behavior toward the company. 53 Conclusion More and more companies are supporting controversial issues in their CSR initiatives. This creates a previously unexamined situation where the individual’s self- concept is incongruent with a CSR cause. We show that when the cause is incongruent with the self-concept, CSR backfires; incongruent CSR (vs. no-CSR) decreases anticipatory self-threat and thus increases dishonest behavior toward the company. Congruent CSR increases anticipatory self-threat and decreases dishonest behavior. Moreover, the backfiring effect of incongruent CSR on increasing dishonest behavior is bigger than the effect of congruent CSR on decreasing dishonest behavior. Given the prevalence of companies taking stands on controversial causes, we hope this research inspires other scholars to examine other benefits and costs of taking stands on controversial CSR for the company. 54 Chapter 2: Essay II – Thinking About Why Leads to Less Cheating When the Importance of Moral Values Is High Introduction Sally and Jane (separately) face the same dilemma at the ticket counter of an amusement park. They can each can get a 20% discount by lying about their respective children’s age. On the way to the park, Sally read an article about why people should eat better, while Jane read an article about how people should improve their diet. Which of the two is more likely to lie about her child’s age? Does the answer depend on how much each of them values morality? Prior research on the effects of construal level in the moral domain provides conflicting results. Although some studies show that an abstract (vs. concrete) construal leads to judgments that other’s unethical behavior is more wrong (Agerström & Björklund, 2009; Eyal, Liberman, & Trope, 2008), others show the opposite (Gong & Medin, 2012). Virtually all these studies examined moral judgements of hypothetical scenarios of moral dilemmas (for an exception, see Rixom & Mishra, 2014). However, moral judgments do not always reflect real behavior (Bostyn, Sevenhant, & Roets, 2018; FeldmanHall et al., 2012). Bostyn et al. (2018) found that participants’ judgments of hypothetical moral dilemmas did not predict their behavior in real-life dilemmas, while FeldmanHall et al. (2012) demonstrated that participants were more willing to hurt others for gains when faced with a real decision than when responding to hypothetical scenarios. Similarly, other research has found that the same contextual factor, such as incidental feelings of disgust, can differentially influence moral judgements and actual behavior (Schnall, Haidt, Clore, 55 & Jordan, 2008; Winterich, Mittal, & Morales, 2014). Moreover, most of the prior studies on the effects of construal level in the moral domain focused on extreme unethical behaviors (e.g., incest, eating one’s dog) rather than more minor unethical behaviors (e.g., lying or cheating to get a small monetary gain) (Eyal et al., 2008; Gong & Medin, 2012). Thus, in the opening example, it is unclear whether or under what conditions the type of articles people read would affect their subsequent actual lying behavior. Therefore, we examine the effect of construal level on actual dishonest behavior conducted for self-interested gains and identify the importance of moral values as a conditional factor that may strengthen and attenuate this effect. Prior literature suggests that abstract rather than concrete construal makes moral values more salient in the context of moral scenarios (which we call the moral value account), as moral values are general and decontextualized in nature and thus they align more with abstract rather than concrete construal (Agerström & Björklund, 2009; Eyal et al., 2008). We propose that this effect is more likely to occur when the importance of moral values is high. When individuals place a high importance on moral values and thus have a complex knowledge structure surrounding the moral values (Aquino & Reed, 2002), abstract construal, which activates cognitive thinking on relevant high-level meanings of a situation, will be likely to make moral values salient in the context where lying and cheating are possible. In contrast, when individuals place a low importance on moral values and thus have limited knowledge structure surrounding the moral values, abstract construal will be unlikely to make moral values salient. Thus, we 56 propose that when the importance of moral values is high, abstract construal, compared to concrete construal, reduces dishonest behavior. We also expect that when the importance of moral values is low, the effect of construal level on dishonest behavior will be attenuated. Collectively, our research provides both theoretical and practical contributions. First, a majority of prior works on construal level in the moral domain examine moral judgments of hypothetical, extreme moral scenarios (e.g., Eyal et al., 2008; Gong & Medin, 2012). In this research, we address this gap by examining actual dishonest behavior such as lying and cheating that individuals encounter in their daily lives. Moreover, we enhance the understanding of the relationship between construal and dishonest behavior by demonstrating the importance of moral values as a moderator. We refine the moral value account proposed by Eyal et al. (2008)— abstract (vs. concrete) construal makes moral values salient—by showing that the moral value account is more likely to occur under conditions in which the importance of moral values is high. Practically, our findings suggest that practitioners can effectively reduce dishonest behavior by priming abstract construal and the importance of moral values together. Theoretical Background Construal Levels in the Moral Domain Construal level theory argues that any action or situation can be construed at different levels of abstraction, from low-level, concrete construal to high-level, 57 abstract construal (Trope & Liberman, 2010; Vallacher & Wegner, 1989). Concrete construal focuses on how an action is performed using incidental, peripheral, subordinate details that are immediately available in the situation. Abstract construal focuses on why an action is performed using general, essential, superordinate information that is not immediately available, and abstracting these general, high- level features conveys the general meaning of the action or situation. For example, traveling to a new country can be represented in terms of preparing logistic procedures such as purchasing flight tickets when it is construed at a concrete level. The same action can be represented as an opportunity to broaden one’s horizons at an abstract construal. Prior work on construal level in the moral domain has primarily focused on how construal level influences moral judgements of others’ hypothetical unethical behavior (e.g., Agerström & Björklund, 2009; Eyal et al., 2008; Gong & Medin, 2012; Lammers, 2012). The findings on the effect of construal level on moral judgments are mixed. One stream of research shows that abstract (vs. concrete) construal leads to more severe judgments of others’ unethical behavior (i.e., judgments that other’s unethical behavior is more wrong), such as sexual intercourse between siblings (Agerström & Björklund, 2009; Eyal et al., 2008). Eyal et al. argued that this effect emerges because abstract rather than concrete construal makes moral values more salient and leads individuals to construe a situation in terms of moral values in a context where they make a judgement relevant to morality (Agerström & Björklund, 2009; Eyal et al., 2008; Rixom & Mishra, 2014), which we call the moral value account. Because moral values are general and decontextualized in nature, they 58 align more with abstract rather than concrete construal. Consistent with this, Agerström and Björklund find that abstract (vs. concrete) construal leads people to attribute others’ unethical behavior to abstract dispositional causes (e.g., being selfish) rather than concrete situational causes, resulting in more severe moral judgments. In contrast, Gong and Medin (2012) demonstrate that abstract (vs. concrete) construal leads to less severe judgments of others’ unethical behavior. They speculate that abstract construal may lead people to perceive the description of others’ unethical behavior less vividly and visually, thereby reducing the severity of moral judgments. Relatedly, Lammers (2012) provides mixed findings on the effect of construal level on moral judgements. Although he focused on comparing moral judgments of one’s own and others’ unethical behavior, the experimental design allows us to assess whether construal level influenced moral judgements. Specifically, when individuals judged others’ unethical behavior, in one out of four studies, abstract (vs. concrete) construal significantly increased the severity of moral judgments of others’ behavior; in the other three studies, however, construal level did not have a significant impact. When individuals judged one’s own unethical behavior, in two out of four studies, abstract (vs. concrete) construal significantly decreased the severity of moral judgments of their own behavior; in the other two studies, however, construal level did not influence the severity of moral judgements. In summary, prior work is inconclusive about the effects of construal level on moral judgments. Moreover, as this work examines moral judgements of hypothetical dilemmas, this provides limited insights on the effects of construal level on actual 59 unethical behavior. Individuals’ moral judgments of hypothetical scenarios often do not predict their actual behavior in a similar situation (Bostyn et al., 2018; FeldmanHall et al., 2012). For example, Bostyn et al. (2012) demonstrated that one’s moral judgments of hypothetical trolley dilemmas—i.e., participants imagined a trolley train on a deadly collision course towards five tied-up people. They were asked which one was more morally wrong: to pull a lever to divert the trolley to another track to save five people, which would kill single innocent bystander, or not to pull a lever—did not have any predictive value for their actual decision to administer an electroshock to a single mouse to avoid five other mice from receiving the shock. This finding is in line with the research showing that the same antecedent can have a different effect on moral judgments of unethical behavior and actual behavior (Cohen, Wolf, Panter, & Insko, 2011; Schnall et al., 2008; Winterich et al., 2014). Although incidental feelings of disgust led to more severe moral judgments of unethical behavior (Schnall et al., 2008), they increased actual lying and cheating behavior (Winterich et al., 2014). Thus, it becomes important to directly examine how construal level influences actual dishonest behavior. One paper by Rixom and Mishra (2014) has examined actual dishonest behavior. They also obtained mixed findings with regards to the effect of construal level on actual dishonest behavior. The focus of their paper was to show that when an abstract construal is activated, people act more dishonestly for social good than for self-interest. They explained this finding using the moral value account proposed by Eyal et al. (2008), that abstract rather than concrete construal makes relevant moral values, such as caring, more salient. However, a close examination of the results 60 shows that the effect of construal level on actual dishonest behavior for self-interested gains was not entirely consistent. Specifically, Rixom and Mishra (2014) found in study 1 that construal level did not significantly influence cheating for self-interested monetary benefits. In study 2, abstract (vs. concrete) construal significantly reduced cheating behavior. Therefore, it is unclear whether construal level affects actual dishonest behavior. This examination of prior literature reveals that the moral value account, one of the most prevalent explanations for the effect of construal level on moral judgments and behavior so far, is not always supported. However, outside the moral context, the literature on construal level provides empirical evidence consistent with the moral value account. Prior work finds that activation of abstract (vs. concrete) construal leads people to act consistently with their long-term goals and values (Fujita, Trope, Liberman, & Levin-Sagi, 2006; Mehta, Zhu, & Meyers-Levy, 2014; Torelli & Kaikati, 2009). For example, Fujita et al. (2006) show that when individuals are confronted with a decision that requires self-control, which involves a conflict between long-term and short-term outcomes, abstract construal increases self-control. Torelli and Kaikati (2009) report that when abstract (not concrete) construal is activated, individuals engage in behaviors (e.g. participating in a volunteer program) that are consistent with their value systems (e.g., benevolence) because abstract construal makes the values salient in a relevant situation. If the findings consistent with the moral value account are consistently reported, a potential reason for the mixed findings about the effect of construal level on dishonest behavior may be undetected moderators that may have attenuated the effect of construal level. 61 Moderating Role of Importance of Moral Values In this research, we propose that the importance of moral values moderates the effect of construal level on dishonest behavior. When a person considers a certain value important, a complex knowledge structure surrounding the value is more likely to be stored in memory (Aquino & Reed, 2002). Thus, if abstract (vs. concrete) construal indeed makes moral values salient and facilitates defining a situation in terms of moral values, this effect will be more likely to occur when individuals place a high importance on moral values and thus have a complex knowledge structure surrounding the moral values. When individuals do not place a high importance on moral values, abstract (vs. concrete) construal will not make moral values salient and thus not impact dishonest behavior. This prediction is consistent with prior work showing that an abstract construal activates the values or long-term goals relevant to a situation only when individuals put a high important on the values or long-term goals (Fujita et al., 2006; Torelli & Kaikati, 2009). For example, Fujita et al. (2006) have demonstrated that abstract construal, compared to concrete construal, leads individuals to choose a healthy option which provides long-term benefits despite a short-term pain only when they value physical health as an important long-term goal. Moral values may be important because of individual differences or contextual factors. Although individuals generally believe strongly in their morality (Greenwald, 1980; Sanitioso, Kunda, & Fong, 1990), they differ in the extent to which they value morality (Aquino & Reed, 2002; Schwartz & Bardi, 2001). For example, moral identity, the degree to which one values morality as a part of one’s self-concept, exhibits stable individual differences (Aquino & Reed, 2002). So does 62 the value of benevolence, which captures preservation and enhancement of the welfare of people with whom one is in frequent personal contact (Schwartz, 1992; Schwartz & Bardi, 2001). Moral values can also be important because situational factors, such as external priming, highlight them. For instance, environmental cues, such as a holy day or reading a book about virtue, may temporarily increase the importance of moral values. So far, we discussed that the effect of construal level on dishonest behavior would vary depending on the importance of moral values. Our theorization can be looked at in a different way. Our theorization also suggests that whether the higher importance of moral values leads to less dishonest behavior depends on the type of construal levels activated. Building on the moral value account, we argued that abstract (vs. concrete) construal makes relevant moral values salient in a situation where dishonest behavior is possible. This suggests that when abstract (vs. concrete) construal is activated, the importance of moral values is more likely to be incorporated into individuals’ decision to act dishonestly. In sum, we hypothesize that H1: Construal level and the importance of moral values will interact to influence dishonest behavior. H1a: When the importance of moral values is high, abstract (vs. concrete) construal will reduce dishonest behavior. When the importance of moral values is low, the effect of construal level on dishonest behavior will be attenuated. H1b: When abstract construal is activated, the higher importance of moral values will lead to less dishonest behavior. When concrete 63 construal is activated, the effect of construal level on dishonest behavior will be attenuated. Study Overview We tested our propositions in four studies using actual dishonest behavior. In study 1, we measured the importance of moral values using the benevolence scale (Schwartz & Bardi, 2001). Consistent with H1a, we found that abstract (vs. concrete) construal reduced dishonest behavior only for those who placed a high importance on benevolence. Also, consistent with H1b, we found that those with the higher importance on benevolence acted less dishonestly only when abstract construal was activated. In study 2, we replicated these findings using a different manipulation of construal level, measure of moral values, and cheating task, providing robust support for H1. Moreover, we included a control condition where we did not manipulate construal level, and showed that the concrete construal was a baseline condition. In studies 3 and 4, we externally primed the importance of moral values to provide a clear evidence for the causal relationship between the importance of moral values and dishonest behavior. In these studies, we found support for H1. Overall, across the four studies, abstract and concrete construals were induced using a well-established why-and-how manipulation (studies 1, 3, and 4) as well as another well-established time horizon manipulation (study 2). The importance of moral values was either measured as individual differences (studies 1 and 2) or experimentally primed (studies 3 and 4). Actual dishonest behavior was measured with two different tasks: a context in which participants could lie to qualify for a paid 64 survey (studies 1, 3, and 4) and a deception game (study 2). These different operationalizations of independent, moderating, and dependent variables enhance the generalizability of our findings. Study 1 The purpose of study 1 was to test the proposition that abstract (vs. concrete) construal would have a stronger impact on dishonest behavior for those who place higher importance on moral values (H1a). We also aimed to test the proposition that the higher importance on moral values leads to less dishonest behavior when abstract construal is activated, but not when concrete construal is activated (H1b). To measure individual differences in the importance of moral values, we used the benevolence subscale of Schwartz’s value scale (Schwartz & Bardi, 2001). The benevolence subscale is intended to assess the importance of values related to the preservation and enhancement of the welfare of other people, such as honesty, helpfulness, and loyalty. Method Participants and design. Four hundred thirteen individuals completed an online study on Amazon MTurk website for a small payment (46.2% male; Mage = 35.87, SD = 12.25). The study was a 2 (construal level: abstract vs. concrete) x (importance of benevolence; measured) between-subjects design. Procedure. Participants read that they would complete several studies. They were first asked to complete eligibility questions to check whether they qualified for one of the studies, a mobile device survey. They read that if they qualified, they 65 would receive a bonus payment of $0.25 to complete the 1-2 minute mobile device survey. After participants completed the qualification questions, they learned that it would take a couple of minutes to calculate whether they qualified for the mobile device survey; meanwhile, they were asked to complete another survey. Participants completed a seemingly unrelated task that manipulated construal level using an established why-how manipulation task developed by Freitas, Gollwitzer, and Trope (2004). As a cover story, participants were informed that they would be completing the survey examining different types of thought processes (called, “survey on thought processes”). In the abstract construal condition, participants were asked to provide three responses about why they would improve their physical health. In the concrete construal condition, participants were asked to do the same about how they would improve their physical health. Participants then read that results for eligibility questions for the mobile device survey were available. In order to create the opportunity to cheat, all participants received feedback that they did not qualify for the mobile device survey due to their response to a specific question. Then, participants read that “We are aware that respondents sometimes check a wrong answer by mistake. To prevent this from happening, we will ask the same eligibility question once again.” Participants had two options to choose: a) changing their response to qualify for the 1-2 minute paid survey and b) not changing their response. This choice served as our binary dependent variable measuring lying. See appendix F for the stimuli. After completing filler tasks for about two minutes, participants provided ratings of benevolence. Specifically, participants rated the importance of each value 66 as a guiding principle in their lives on a nine-point scale (-1 = opposed to my values, 0 = not important, 7 = of supreme importance) (Schwartz & Bardi, 2001). The items to measure benevolence values were presented among other items (power, achievement, and universalism). Benevolence values, measured by five items (helpful, honest, forgiving, loyal, and responsible), were of primary interest. All participants indicated gender, age, political ideology, and religiosity as demographic measures. Post-Hoc Test of Manipulation Check of Construal Level Although the why-how construal level manipulation task has been widely established, we conducted a post-hoc test to make sure this manipulation worked as intended. Two hundred two individuals recruited from Amazon MTurk (39.6% male; Mage = 37.41, SD = 11.62) were randomly assigned to either abstract or concrete construal condition. Participants first completed the why-how construal manipulation task as in study 1. They then completed the Behavioral Identification Form (BIF) as a measure of level of construal (Vallacher &Wegner, 1989; See appendix G for the stimuli). BIF has been widely used to assess the efficacy of the construal level manipulation (Agrawal & Wan, 1989; Fujita et al., 2006; van Kerckhove, Geuens, & Vermeir, 2015; Liberman & Trope, 1998). In the BIF, participants read that any action can be described in many ways: “typing a paper” can be described as “pushing keys” or “expressing thoughts.” Then, they were asked to choose one of two descriptions of the same behavior (e.g., making a list), one that represents abstract construals related to goals, values, or why considerations (e.g., getting things organized) and the other that represents concrete construals related to means, 67 processes, or how considerations (e.g., writing things down). There were a total of 18 items. A construal level score was calculated with the following formula: (the number of abstract options chosen/total items). Participants in the abstract condition (M = 0.69, SD = 0.24), compared to those in the concrete condition (M = 0.62, SD = 0.27), showed a significantly higher score (F(1, 200) = 4.16, p = .043, η2=0.02). This shows that the why-how manipulation systematically altered participants’ construal level. Results To ensure that the manipulation of construal level did not influence the importance of benevolence, we ran a one-way ANOVA, revealing that the main effect of construal level on the importance of benevolence was not significant (F(1, 410) = 1.18, p = .28). Therefore, benevolence could be used as an independent variable. To test H1, we regressed whether participants acted dishonestly (1 = lying, 0 = not lying) on construal level (0 = concrete, 1 = abstract), the mean-centered importance of benevolence (M = 7.62, SD = 1.14), and their interaction using a logistic regression. The analysis revealed nonsignificant main effects of construal level, b = -0.09, Wald χ2(1) = 0.19, p = .67 and of benevolence, b = -0.07, Wald χ2(1)= 0.32, p = .57. As predicted, there was a significant interaction effect, b = -0.42, Exp(b) = 0.66, Wald χ2(1) = 5.05, p = .025 (see Figure 5). Spotlight analysis indicated that when participants rated the importance of benevolence strongly (+1 SD from the mean, corresponding to a value of 8.76 on a scale from 1-9), abstract (vs. concrete) construal marginally significantly reduced lying (b = -0.56, Exp(b) = 0.57, Wald χ2(1) = 3.35, p = .067). Specifically, whereas 23.1% of participants in the abstract condition lied, 34.5% in the concrete condition lied. In contrast, when participants rated the 68 importance of benevolence weakly (-1 SD from the mean, corresponding to a value of 6.48), construal level did not significantly influence lying (b = 0.38, Wald χ2(1) = 1.77, p = .18). 47.5% (38.2%) of participants in the abstract (concrete) condition lied. This provides support for H1a.1 Figure 5. The effect of construal level and importance of benevolence on dishonest behavior (study 1) 50 47.5 45 40 38.2 34.5 35 30 25 23.1 20 15 10 5 0 -1 SD (M = 6.48) +1 SD (M = 8.76) Importance of Benevolence Abstract Concrete Looked at another way, in the abstract construal condition, higher importance of benevolence significantly reduced lying (b = -0.49, Exp(b) = 0.62, Wald χ2(1) = 12.53, p < .001). In the concrete construal condition, the importance of benevolence 1 We also analyzed data using the Johnson-Neyman analysis technique. Consistent with the spotlight analysis, we found that abstract (vs. concrete) construal marginally significantly decreased dishonest behavior for any individual with the importance of benevolence of 9 (bJN = -0.66, SE = 0.34, p = .052). 69 Probability of Lying (%) did not influence lying (b = -0.07, Wald χ2(1) = 0.32, p = .57). This provides support for H1b by showing that only when abstract construal was activated, the importance of benevolence influenced dishonest behavior. To ensure that the moderator was specific to benevolence, we tested the moderating role of the other three values (power, achievement, and universalism), religiosity, and political ideology, using separate regressions. No regression revealed a significant two-way interaction (ps > .21), suggesting that only benevolence affected dishonest behavior. Discussion In study 1, we provided initial support for H1. Consistent with H1a, abstract (vs. concrete) construal reduced dishonest behavior only for those who placed higher importance on moral values. Moreover, consistent with H1b, when abstract construal was activated, not when concrete construal was activated, those with higher importance on moral values acted more honestly. We conducted a direct replication study of study 1 on Amazon MTurk. Although the interaction between construal level and the importance of benevolence was not significant (b = -0.24, Wald χ2(1) = 2.17, p = .14), we found the same pattern of results, enhancing confidence in our finding of study 1. See supplemental study 1 in appendix H for details. 70 Study 2 In study 2, we used a different manipulation of construal level (temporal distance manipulation), dishonest behavior task (deception game), measure of moral values (five moral values based on Graham, Haidt, and Nosek (2009)), and population (undergraduate students) to generalize the findings of study 1. Moreover, as a baseline, we included a control condition that did not manipulate construal level. Building on prior work (Torelli & Kaikati, 2009), we expected that individuals would act similarly in the concrete construal and control conditions, and that both would be significantly different from the abstract construal condition. Method Participants and design. A total of 440 undergraduate students at a large American university participated in a study for course credit (48.9% male; Mage = 20.85, SD = 2.18). We used a 3 (construal level: abstract vs. concrete vs. control) x (importance of moral values; measured) between-subjects design. Procedure. Participants came to the behavioral lab in small groups for a series of studies run on the computer. The first study manipulated construal level, and the second, ostensibly unrelated, study was a deception game adapted from Gneezy (2005). In the deception game, participants decided whether to lie to their anonymous partner to obtain more money. In this study, the manipulation of construal level was based on time horizon. Participants in the abstract (concrete) construal condition were asked to imagine their life a year from now (tomorrow) and write a few sentences about how they expect it to be (Agrawal & Wan, 2009; Cho, Kahn, & Dhar, 2013; Förster, Friedman, & 71 Liberman, 2004). Merely thinking about life in the distant (near) future has been found to activate an abstract (concrete) construal. In the control condition, participants were asked to write any associations with elephants (adapted from Torelli & Kaikati, 2009). In the deception game (called “decision making task”), participants learned that they would be randomly matched with someone at another campus. Both their own and their partner’s identity would remain anonymous. Participants learned that they could win either $1 or $2 based on which option their partner selected. Option A would give $1 to the participant and $2 to the partner; Option B would give $2 to the participant and $1 to the partner. Thus, the participant would win more money if the partner chose Option B. Participants had to choose one of two messages to send to their partner. Message 1 was true, as it stated that “Option A will earn you more than Option B.” Message 2 was false, because it stated that “Option B will earn you more than Option A.” Thus, selecting Message 2 would mean that the participant was lying to the partner. Before making the message choice, they were informed that the partner would not know the amount of the monetary payments associated with each option (Option A or Option B). Thus, the partner would be relying on the participant’s message to make a choice. In addition, to reduce the possibility of strategic honesty (Hershfield, Cohen, & Thompson, 2012), in which a participant sends the true message based on a belief that the partner would not believe the message, participants were told that, according to past research, 80% of partners trusted the message and chose the option 72 in the message. In short, we attempted to ensure that participants believed that they would not be caught, that they would remain anonymous to the partner, and that the partner would believe them. Participants marked their choice of message (Message 1 or 2) on a slip of paper and raised their hand so that experimenters collected the slip to inform their partner of the message. This choice represented the dependent variable of lying. Unbeknownst to participants, there were no partners on the other campus. Participants were then asked what they thought the hypothesis of the study was and whether they had any thoughts they wanted to share with the researchers. These two questions were added to check the possibility of hypothesis guessing and suspicion of the research procedure. Participants then completed other unrelated filler studies for 30 minutes. After that, participants rated the chronic importance of seven values as a guiding principle in their life on a 7-point scale (1 = not at all important, 7 = extremely important). Five moral values are based on Graham, Haidt, and Nosek (2009): caring for someone weak or vulnerable, being an honest person, being loyal to an in-group, respecting the traditions of society, and acting in a virtuous or uplifting way. To assess whether the moderating effect is specific to moral values, we also measured the values of being a smart person and being a successful person. An exploratory factor analysis (eigenvalues >1) identified two factors: one consisting of the five moral values (α = .72) and the other consisting of the two non-moral values (r = .58). Thus, we analyzed data on the two variables: moral values and non-moral values. 73 At the end of research session, participants indicated gender and age. Participants who chose to tell the truth by sending Message 1 were paid $1; those who lied by sending Message 2 were paid $2, consistent with the rules of the game. See appendix I for the stimuli. Post-Hoc Test of Manipulation Check of Construal Level Although the temporal construal level manipulation task has been widely established (Agrawal & Wan, 2009; Cho et al., 2013; Förster et al., 2004), we conducted a post-hoc test to make sure that this manipulation worked as intended. Ninety six undergraduate students (56.3% male; Mage = 20.02, SD = 1.78) were randomly assigned to either abstract or concrete construal condition. Participants first completed the temporal construal manipulation as in study 2 and then completed BIF (see appendix G for the stimuli). Unexpectedly, there was no significant difference between the abstract (M = 0.66, SD = 0.19) and concrete conditions (M = 0.64, SD = 0.19; F(1, 94) = 0.27, p = .60). Given that this temporal construal level manipulation has been widely used, we speculate that this finding may have been driven by the insensitivity of BIF measures. Results Suspicion check. The first author, who was blind to the construal condition, coded the responses to the two items (“what do you think the hypothesis of the experiment is?” and “do you have any thoughts you want to share with the researchers?) for hypothesis guessing and suspicion of the research procedure. Seven 74 participants indicated that they did not believe that there was a partner involved. They were excluded, leaving 433 participants for analysis. Dishonest behavior. First, we ran an ANOVA analysis and confirmed that the manipulation of construal level did not influence the importance of moral values (F(2, 429) = 0.65, p = .52). To test the proposed interaction effect of construal level and the importance of moral values, we conducted a logistic regression, entering two dummy- coded variables for construal level using the concrete condition as a baseline (dummy-abstract: representing abstract construal with a value of 1, otherwise 0; dummy-control: representing control condition with a value of 1, otherwise 0), the mean-centered importance of moral values (M = 5.70, SD = 0.81), and the two two- way interaction terms between each dummy-coded variable and importance of moral value as predictors of the dichotomous variable of dishonest behavior (1 = lying, 0 = not lying). The analysis yielded a significant main effect of dummy-abstract, b = -0.47, Exp(b) = 0.63, Wald χ2(1) = 4.03, p = .045, suggesting that abstract construal, compared to concrete construal, significantly reduced lying. The main effects of dummy-control and the importance of moral values were not significant (ps > .59). Importantly, these effects were qualified by a significant two-way interaction between dummy-abstract and the importance of moral values, b = -0.61, Exp(b) = 0.54, Wald χ2(1) = 4.42, p = .036 (see Figure 6). Spotlight analysis indicated that when participants rated the importance of moral values strongly (+1 SD from the mean, corresponding to a value of 6.51 on a scale from 1-7), abstract construal (34.5%) significantly reduced lying compared to concrete construal (58.0%) (b = -0.96, Exp(b) 75 = 0.38, Wald χ2(1) = 8.55, p = .003). However, when participants rated the importance of moral values less strongly (-1 SD from the mean, corresponding to a value of 4.89), abstract construal (54.7%), compared to concrete construal (54.0%), did not significantly influence lying (b = 0.03, Wald χ2(1) = 0.01, p = .93). This replicates the findings of study 1 and provides further support for H1a. The two-way interaction between the dummy-control and the importance of moral values was not significant, b = -0.05, Wald χ2(1) = 0.02, p = .88, suggesting that the effect of concrete construal (vs. control) on lying did not vary depending on the importance of moral values. This suggests that the concrete construal is similar to the control condition. In order to examine the interaction between the abstract and control conditions and the importance of moral values, we conducted another logistic regression analysis after replacing the two dummy-coded variables for construal level with another set of two dummy-coded variables using control condition as a baseline (dummy-abstract: representing abstract construal with a value of 1, otherwise 0; dummy-concrete: representing concrete construal with a value of 1, otherwise 0). The analysis yielded insignificant main effects of dummy-abstract, dummy- concrete, and the importance of moral values (ps > .11). Importantly, there was a marginally significant two-way interaction between the dummy-abstract and the importance of moral values, b = -0.57, Exp(b) = 0.57, Wald χ2(1) = 3.28, p = .070, suggesting that the effect of abstract construal (vs. control) on lying varies depending on the level of importance of moral values. Spotlight analysis indicated that when participants rated the importance of moral values strongly (+1 SD from the mean), 76 abstract construal (34.5%), compared to control condition (55.1%), significantly reduced lying (b = -0.85, Exp(b) = 0.43, Wald χ2(1) = 5.56, p = .018). In contrast, when participants rated the importance of moral values less strongly (-1 SD from the mean), abstract construal (54.7%), compared to control condition (52.9%), did not significantly influence lying (b = 0.07, Wald χ2(1) = 0.04, p = .83). Overall, the results from the two logistic regression analyses show that the effect of abstract construal (vs. concrete and control) on reducing dishonest behavior was attenuated when consumers placed less importance on moral values, providing support for H1a.2 Moreover, consistent with H1b, higher importance of moral values significantly reduced lying in the abstract (b = -0.52, Exp(b) = 0.59, Wald χ2(1) = 5.73, p = .017), but not concrete (b = 0.10, Wald χ2(1) = 0.28, p = .60) or control conditions (b = 0.06, Wald χ2(1) = 0.06, p = .80). This suggests that it is the abstract construal that makes people behave consistent with the importance of the moral values. To ensure that the moderator was specific to benevolence, we tested the moderating role of the importance of non-moral values (a composite of success and 2 We also analyzed data using the Johnson-Neyman analysis technique, which revealed the results consistent with the spotlight analysis. We found that abstract (vs. concrete) construal significantly decreased dishonest behavior for any individual with the importance of moral values greater than 5.64 (bJN = -0.45, SE = 0.23, p = .050) but not for any individual with the importance of moral values less than 5.64. Abstract (vs. control) construal significantly decreased dishonest behavior for any individual with the importance of moral values greater than 5.80 (b JN = -0.41, SE = 0.24, p = .050) but not for any individual with the importance of moral values less than 5.80. In contrast, when we compared concrete construal with control condition, there was no significantly different region in terms of the probability of lying across any level of importance of moral values. 77 smart) using a separate regression. None of the two-way interaction terms were significant (ps > .17). Figure 6. The effect of construal level and importance of moral values on dishonest behavior (study 2) 70 60 58.054.7 54.0 52.9 55.1 50 40 34.5 30 20 10 0 -1 SD (M = 4.89) +1 SD (M = 6.51) Importance of Moral Value Abstract Concrete Control Discussion In study 2, we replicated the findings of study 1, providing further support for H1. The use of a different construal level manipulation, dishonest behavior task, measure of importance of moral values, and sample enhances the robustness of the findings. Moreover, we found that participants acted similarly in the concrete and control conditions and that abstract construal was different from these two conditions. So far, we operationalized the importance of moral values by measuring the individual differences in the importance of moral values. One limitation of the individual difference measure is that the causal relationship between the importance 78 Probability of Lying (%) of moral values and dishonest behavior is not clear. To address this issue, in the next study, we externally manipulate the importance of moral values using a priming procedure. Study 3 Method Participants and design. Four hundred seventy-six individuals completed an online study on Amazon Mturk for a small payment (47.9% male; Mage = 36.56, SD = 11.49). We used a 2 (construal level: abstract vs. concrete) x 2 (priming of importance of moral values: yes vs. no) between-subjects design. Procedure. The procedure of the study was very similar to that of study 1 where participants could lie to change their response on a survey to qualify for a paid survey except for the following points. First, after participants completed the qualification questions and before manipulating construal level using the why-how manipulation, we manipulated the salience of the importance of moral values via a priming procedure. Specifically, participants in the importance of moral values priming condition read that they would participate in another survey, which was ostensibly called “survey on reading comprehension.” Participants were asked to read the following paragraph, which emphasized the importance of following moral values. “MAKING EVERYDAY DECISIONS” To understand people's decision-making processes, psychologists build models. In such models, psychologists make the following assumptions: (1) that all individuals are moral, which means being truthful and caring about 79 others’ interests; and (2) that all individuals attempt to follow moral principles, such as being good and sincere. Moreover, psychologists have shown that this leads to better decisions. Those in the no priming condition did not participate in this survey and thus did not read the paragraph. Second, we did not measure the individual differences in the importance of moral values. After the moral value priming task and the why-how construal level manipulation task, participants indicated whether they would change their response to qualify for a paid survey, which served as our dependent measure of dishonest behavior. Results To test H1, we ran a logistic regression with lying as the binary dependent variable of dishonest behavior (1 = lying, 0 = not lying) and construal level (0 = concrete, 1 = abstract), the importance of moral values (0 = no priming, 1 = priming), and their interaction as independent variables. The analysis revealed nonsignificant main effects of construal level, b = 0.09, Wald χ2(1) = 0.11, p = .75 and the importance of the moral values, b = 0.24, Wald χ2(1) = 0.84, p = .36. As predicted, there was a significant interaction effect between construal level and the importance of moral values, b = -0.79, Exp(b) = 0.45, Wald χ2(1) = 4.34, p = .037 (see Figure 7). Consistent with H1a, when the importance of moral values was externally primed, participants in the abstract construal condition (29.6%) lied significantly less than those in the concrete construal condition (45.9%) (b = -0.70, Exp(b) = 0.50, Wald χ2(1) = 6.90, p = .009). In contrast, when the importance of moral values was not primed, construal level did not significantly influence lying (b = 0.09, Wald χ2(1) = 80 0.11, p = .75). Specifically, 42.1% of participants in the abstract construal condition and 40.0% of participants in the concrete construal condition lied. Figure 7. The effect of construal level and importance of moral values on dishonest behavior (study 3) 50 45.9 45 42.1 40.0 40 35 29.6 30 25 20 15 10 5 0 No Priming Priming Importance of Moral Values Abstract Concrete Moreover, consistent with H1b, in the abstract construal conditions, moral priming (29.6%), compared to no moral priming (42.1%), significantly reduced lying (b = -0.55, Exp(b) = 0.58, Wald χ2(1) = 4.04, p = .045). In the concrete construal conditions, however, participants in the moral priming (45.9%) and no moral priming (40.0%) conditions did not significantly differ in terms of the probability of lying (b = 0.24, Wald χ2(1) = 0.84, p = .36). This suggests that when abstract but not concrete construal was activated, the importance of moral values that were temporarily made salient was incorporated into participants’ decision to lie, providing further support for our theory. 81 Probability of Lying (%) Discussion In study 3, by externally priming the importance of moral values, we showed that abstract construal, compared to concrete construal, reduced dishonest behavior only when the importance of moral values was externally primed, providing further support for H1a. Moreover, consistent with H1b, only when abstract construal was activated, not when concrete construal was activated, the importance of moral priming reduced dishonest behavior. This suggests that abstract construal led participants to act consistently with the primed importance of moral values. This provides a causal evidence for the interactive effect of construal level and importance of moral values on dishonest behavior. Study 4 In study 3, we manipulated the importance of moral values by the presence or absence of a paragraph that discussed how psychologists view moral values. In the current study, we manipulated the importance of moral values by varying social consensus about the acceptability of dishonest behavior (i.e., whether a majority of other people think small lying and cheating behaviors as morally unacceptable or acceptable). We expected that when participants perceive that a majority of other people think dishonest behavior is morally unacceptable (acceptable), they would temporarily consider moral values as important (unimportant). Consistent with the findings of our prior studies, we initially predicted that abstract construal, compared to concrete construal, would reduce dishonest behavior only in the unacceptable social consensus condition, not in the acceptable social consensus condition. 82 Method Participants and design. A total of 162 undergraduate students at a large American university participated in a study for course credit (31.7% male; Mage = 21.59, SD = 2.14). The design was a 2 (construal level: abstract vs. concrete) x 2 (social consensus: unacceptable vs. acceptable) between-subjects design. Procedure. The procedure of this study was very similar to that of study 3 except for the manipulation of social consensus. After participants completed the qualification questions and before manipulating construal level using the why-how manipulation, we manipulated social consensus of dishonest behavior via a priming procedure. In the social consensus manipulation task, which was ostensibly called “study on short-term memory,” participants were asked to read the following news article on social consensus on various issues (adapted from Yam, Chen, & Reynolds, 2014). Two undergraduate students at the University of Illinois are on a mission to make the opinions of different social issues more public through a new web polling site called “open-vote.com.” Discerning public consensus on social issues can be a challenge, the creators of the site say, and their project attempts to provide that service. Open-Vote co-founder Alex Jacobs said, “We collect data from undergraduate students in the United States. Currently we have surveyed more than 10,000 undergraduates nationwide.” Then, participants were presented with two findings. One finding (“41.3% of the respondents responded that the minimum drinking age should lowered from 21 to 18.”) was the same across two conditions. We varied the other finding to manipulate the social consensus of dishonest behavior. In the unacceptable [acceptable] social consensus condition, participants read that “89.0% of the respondents considered lying or cheating for a small reward (e.g., taking a couple of office supplies from the office, lying to a retailer to get discount) as morally unacceptable [acceptable]). 83 Then, to check whether participants read and understood the article, we asked to what extent undergraduate students thought that the three behaviors (taking a couple of office supplies from the office, over-reporting the number of hours worked by an hour or two, and lying for a small rewards) were unacceptable or acceptable (1 = Most undergraduates think this is unacceptable, 7 = Most undergraduates think this is acceptable). Participants’ response to the three items were averaged (α = .93). After the social consensus manipulation task, participants completed the why-how construal level manipulation task and indicated whether they would change their response to qualify for a paid survey, which was our dependent measure of dishonest behavior. Results Comprehension check for social consensus manipulation. The degree to which participants perceive that most undergraduates think small dishonest behaviors as unacceptable versus acceptable was submitted to a one-way ANOVA with social consensus as a factor. We did not include construal level as a factor because construal level was manipulated after the comprehension check measures for the social consensus manipulation. Results revealed that participants in the acceptable condition (M = 5.73, SD = 1.24) perceived that most undergraduates thought small dishonest behaviors as more acceptable than those in the unacceptable condition (M = 2.36, SD = 1.86; F(1, 160) = 184.78, p < .001, η2=0.54). This indicates that participants understood the article correctly. Dishonest behavior. To examine the prediction, we ran a logistic regression with dishonest behavior as the binary dependent variable (1 = lying, 0 = not lying) and construal level (-1 = concrete, 1 = abstract), social consensus (-1 = unacceptable, 84 1 = acceptable), and their interaction as independent variables (see Figure 8). Contrary to expectation, only the main effect of construal level was significant (b = - 0.43, Exp(b) = 0.65, Wald χ2(1) = 6.19, p = .013), while the interaction effect between construal level and social consensus was not significant (b = -0.13, Wald χ2(1) = 0.56, p = .45). Specifically, participants in the abstract construal condition lied significantly less than those in the concrete construal condition (24.4% vs. 43.0%). Figure 8. The effect of construal level and social consensus about the acceptability of dishonest behavior on dishonest behavior (study 4) 50 46.2 40.0 40 30 26.8 22.0 20 10 0 Unacceptable Acceptable Social Consensus of Dishonest Behavior Abstract Concrete Post-Hoc Test We speculated post-hoc that we had made the importance of moral values salient under both unacceptable and acceptable social consensus conditions, since both conditions discussed moral values. To test this, we ran a post-hoc study with 152 undergraduate students at the same large American university (38.4% male; Mage = 20.66, SD = 0.84). The design was a 3-cell (social consensus: unacceptable vs. 85 Probability of Lying (%) acceptable vs. control) between-subjects design, with participants randomly assigned to conditions. Participants in the unacceptable and acceptable social consensus conditions read the news articles used in study 4, while those in the control condition did not read any news article. After that, we measured the importance of moral values by asking how much they agreed with the following statement: “It is important for me to behave consistent with my moral values” (1 = strongly disagree, 7 = strongly agree). A one-way ANOVA analysis with the importance of moral values as the dependent variable and social consensus as the independent variable revealed a significant main effect (F(2, 149) = 3.24, p = .042, η 2 p = 0.04). Specifically, participants in the unacceptable (M = 5.88, SD = 0.91) and acceptable (M = 5.82, SD = 1.17) social consensus conditions perceived moral values as more important than those in the control condition (M = 5.41, SD = 0.94; F(1, 149) = 5.51, p = .020, d = 0.51, and F(1, 149) = 4.09, p = .045, d = 0.39, respectively). There was no significant difference in the importance of moral values between the unacceptable and unacceptable social consensus conditions (F(1, 149) = 0.10, p = .76). This shows that both unacceptable and acceptable social consensus manipulations actually increased the importance of moral values. This explains why we found that abstract (vs. concrete) construal reduced dishonest behavior in both social consensus conditions. We inadvertently made the importance of moral values high in both conditions in study 4, which makes the study consistent with our theory and findings. 86 Discussion In study 4, unlike our initial prediction, we found that abstract (vs. concrete) construal reduced dishonest behavior in both high and low social consensus of dishonest behavior conditions. However, a post-hoc test showed that this may be because our manipulation of both high and low social consensus, compared to control condition, increased the temporary importance of moral values. One’s value system is constructed over a life time. Thus, a short article describing that a majority of other people consider dishonest behavior as morally acceptable and unacceptable may have made their innate importance of moral values salient rather than changing their perception of the importance of moral values. If our manipulation of social consensus indeed made the importance of moral values salient, our finding that abstract (vs. concrete) construal reduced dishonest behavior in both social consensus conditions is consistent with our proposition, providing further evidence for H1. General Discussion Under what conditions would people’s decision to act dishonestly be influenced by construal level activated in a prior, unrelated task? We demonstrate that the importance of moral values moderates this effect. Across four reported and one supplemental study that examine actual dishonest behavior, we find that abstract construal, compared to concrete construal and control conditions, reduces dishonest behavior only when the importance of moral values is high. We also find that the higher importance of moral values leads to less dishonest behavior only in the abstract, but not in concrete and control conditions, showing that abstract construal 87 makes people behave consistent with the importance of the moral values. Across these studies, the importance of moral values was either measured as individual differences or experimentally primed. Abstract and concrete construals were induced using a why-and-how manipulation as well as a time horizon manipulation. Dishonest behavior was measured with two different tasks. These different operationalizations enhance the generalizability of our findings. Theoretical and Practical Contributions The present research makes several theoretical contributions. First, our findings contribute to the literature on construal level and morality. A vast majority of prior work on construal level in the moral domain examines moral judgments of hypothetical moral scenarios (Agerström & Björklund, 2009; Eyal et al., 2008; Gong & Medin, 2012; Lammers, 2012). Moreover, prior work focuses on judgments of extreme, uncommon unethical behaviors (e.g., incest, eating one’s dog), which are not directly relevant to a consumer context such as decisions to cheat or lie for self- interested gains (e.g., lying to get a discount, cheating on an exam, and falsely inflating tax deductions). Limited research examines how construal level influences actual dishonest behavior such as lying and cheating (see Rixom and Mishra (2014) for an exception). In this research, we address this gap by examining when and how construal level influences actual dishonest behavior. We also extend the understanding of the relationship between construal and dishonest behavior by demonstrating the importance of moral values as a moderator. By doing so, the current research refines and develops the moral value account (e.g., Eyal et al., 2008), that abstract rather than concrete construal makes relevant moral 88 values more salient in the situation where dishonest behavior is possible. We show that this effect is more likely to occur under conditions in which the importance of moral values is high. Beyond the literature on morality, our findings are consistent with the prior work demonstrating that the effect of construal level on self-control decisions are strengthened when individuals value their long-term goals (Fujita et al., 2006). We also note that our finding on the main effect of construal level on dishonest behavior is not robust. For example, we found an insignificant main effect of construal level in study 1 and a significant effect in study 2. This is actually consistent with Rixom and Mishra’s (2014) findings that documented inconsistent effects of construal level on dishonest behavior. Our findings suggest that this inconsistencies may have been driven by undetected moderators, such as contextual salience of the importance of moral values. This shows the importance of considering the importance of moral values when examining the effect of construal level in the moral domain. Furthermore, the current research provides insights on the relationship between individual differences in the importance of moral values and dishonest behavior. Rather surprisingly, we found that the higher importance of moral values did not lead to less dishonest behavior when concrete construal was activated or none of the construals was activated. However, the close examination of the literature suggests that this is consistent with the prior work. For example, Aquino, Freeman, Reed, Lim, and Felps (2009) found that when a situation activated motives conflicting with morality, such as when monetary rewards were salient, higher moral identity did 89 not reduce dishonest behavior because moral identity was not accessible even to those with a highly important moral identity. Moreover, in our essay 1, we also observed in study 1 that the individual difference in moral identity was not correlated with dishonest behavior. Our findings suggest that the activation of abstract construal is one context under which one’s importance of moral values translates into their actual decision to act dishonestly by making their moral values salient. Practically, the current research has implications for companies and organizations who are interested in reducing dishonest behavior. Our findings suggest that practitioners can effectively reduce dishonest behavior by activating abstract construal and the importance of moral values together. For example, public service campaigns that try to reduce dishonest consumer behavior such as fraudulent insurance claims and tax returns can use copies that encourage consumers to adopt an abstract construal and that stress the importance of moral values (e.g., copy that focuses on why they need to act honestly and mentions about social consensus of acceptability of dishonest behavior). Limitations and Future Research Agenda Our research has several limitations, which provide fruitful avenues for the future research. First, although we provided evidence for the underlying psychological process of salience of moral values by examining the theoretical driven moderator, the importance of moral values, we did not provide evidence for the process by measuring it. Future research can provide more direct evidence for the process by using an implicit measure of salience of moral values, such as reaction time measures. For example, one can measure whether abstract (vs. concrete) 90 construal makes individuals to recognize honesty-related words (e.g., honest, true) more quickly using a response time measure (Fitzsimons et al. 2007). Second, in this research, we did not aim to resolve the mixed findings on the effect of construal level on moral judgments of hypothetical moral scenarios. Our research, however, suggests that our findings are not explained by other accounts proposed in the literature, such as vivid processing of the information (Gong & Medin, 2012) and self-serving justification account (i.e., abstract (vs. concrete) construal allows people to justify their unethical behavior; Lammers, 2012). These accounts were examined in the context of moral judgements of moral dilemmas. It is possible that conditions under which these alternative accounts are likely were not present in our studies that examined actual behavior. In particular, our results demonstrate the importance of examining actual unethical behavior in the context of moral decision-making. 91 Appendices Appendix A Study 1: Pretest Results Table A in this appendix shows the percentage of participants who have congruent, neutral, or incongruent identities for each cause. Relatively more polarizing causes are promoting rights of sexual minorities, giving preferential treatment to refugees, giving preferential treatment to military veterans, and animal testing. For providing support for hungry people, providing education for children in need, and ensuring environmental sustainability, which are the causes that prior work on CSR largely focused on, more than 85% of participants indicated having a congruent identity. In terms of correlations, we found no significant correlations between moral identity [M = 6.08, SD = 1.00] and cause-related identity for any of the causes (0 = incongruent, 1 = neutral, and 2 = congruent cause-related identity) (ps > .12), except that moral identity and cause-related identity for environmental sustainability were positively correlated (r = .20, p = .041). Political ideology [M = 3.57, SD = 1.70] was significantly negatively correlated with cause-related identities for the following causes: ensuring environmental sustainability (r = -.23), actions that tackle climate change (r = -.44), promoting racial and ethnic diversity (r = -.25), and promoting rights of sexual minorities (r = -.35), indicating that participants with more liberal political orientations are more likely to have congruent identities for these causes (ps < .05). Political ideology was significantly positively correlated with the cause-related identity for giving preferential treatment to refugees (r = .22, p = .031). However, the sizes of the correlations were small to moderate, suggesting that they are not 92 interchangeable. Moral identity and political ideology were not correlated (r = .02, p = .82). Table A. Percentage of participants with congruent, neutral, or incongruent cause- related identity for each cause (pretest for study 1) Congruent Neutral Incongruent Causes cause- cause- cause-related related related Identity Identity Identity Providing support for hungry people 94 0 6 Providing education for children in need 97 0 3 Ensuring environmental sustainability 87 3 10 Actions that tackle climate change 72 9 19 Animal testing 14 54 32 Promoting racial and ethnic diversity 70 17 13 Promoting rights of sexual minorities 58 11 31 Giving preferential treatment to refugees 11 53 36 Giving preferential treatment to military veterans 58 13 29 93 Appendix B Study Stimuli Study 1 General description of an educational program development company, EduValley EduValley is an American educational program development company founded in 1990. EduValley has developed various training programs and games that allow people to train their cognitive abilities. Customers of EduValley’s training programs range from kindergarteners to the elderly. Many of EduValley’s training programs and games were rated highly on educational program customer review sites. Study 2 and 3 General description of a clothing brand and retailer, Lawson Lawson is an American clothing brand and retailer. Lawson offers an assortment of women’s, men’s and children’s apparel and accessories, including shirts, sweaters, outwear, pants, suiting, and bags. Lawson operates more than 200 retail stores throughout the United States. Lawson has a reputation for making high-quality apparel and accessories. Study 4 News article on Great Wonderland’ CSR activity in the refugee support domain 94 Appendix C Stimuli for Study 1: Instruction Sheet, Matrix Sheet, and Answer Sheet Instruction Sheet: Incentives and Problem-Solving Task Thanks very much for participating in this study. EduValley is currently developing a problem-solving task. Today, we would like to test the difficulty of the task among a sample of college students. We would also like to test whether monetary incentives influence problem-solving skills. All of the task materials and monetary rewards are provided by EduValley. Below are the instructions for the problem-solving task. Please read the instructions carefully. Instructions: In the large yellow envelope on your desk, there is a matrix sheet (Do not take out the matrix sheet now). In the matrix sheet, you will see 20 matrices like the one below: In each matrix you should look for a unique set of two numbers that sum up exactly to 10. When you find a set, circle the numbers, and mark the corresponding ‘Got It’ Box below, as in the example below: For each matrix you solve, you will receive $.25. You have 4 minutes for this task. Experimenters will notify you when 4 minutes are done. Once 4 minutes are up you have to do the following: 1. Count the number of correctly solved matrices. 2. We do not need to collect the matrix sheet and this instruction sheet. Place the matrix sheet and the instruction sheet in a recycling box in a corner of the room. Then, return to your station. 95 3. Take out the yellow answer sheet from the packet, and write down the number of correctly solved matrices. 4. In the packet, you can find a small white envelope containing 4 $1 bills and 4 $0.25 coins. Pay yourself with the money provided in the white envelope on your desk. The amount of payment you receive will depend on the number of correctly solved matrices. Do not start the task now. Wait for instructions from experimenters. 96 Matrix Sheet: 97 Answer Sheet: I got _______ matrices, which translates to $ _______ (=$0.25 per matrix). * Please see the table below to determine the amount of payment you receive. # of # of # of correctly Payment correctly Payment correctly Payment solved ($) solved ($) solved ($) matrices matrices matrices 1 .25 9 2.25 17 4.25 2 .50 10 2.50 18 4.50 3 .75 11 2.75 19 4.75 4 1.00 12 3.00 20 5.00 5 1.25 13 3.25 6 1.50 14 3.50 7 1.75 15 3.75 8 2.00 16 4.00 Q. How difficult was the task? Very Easy 1 2 3 4 5 6 7 Very Difficult 98 Appendix D Study 2: Results for Anticipatory Guilt Anticipatory guilt. We performed a regression analysis using CSR activity (0 = absent, 1 = present), mean-centered LGBT support identity, and the two-way interaction term as the independent variables and anticipatory guilt as the dependent variable. Regression analysis revealed a marginally significant main effect of CSR activity (b =.49, t(236) = 1.77, p = .078; MCSRpresent = 3.86 vs. MCSRabsent = 3.38) and insignificant effect of LGBT support identity (b = -.09, t(236) = -.80, p = .43) on anticipatory guilt. The two-way interaction was significant (β = .44, t(236) = 2.91, p = .004). The Johnson-Neyman analysis revealed that CSR (vs. no-CSR) significantly increased anticipatory guilt for participants with LGBT support identity greater than 4.97 (MCSRpresent = 3.91 vs. MCSRabsent = 3.36; bJN = .54, SE = .28, p = .050). In contrast, CSR (vs. no-CSR) marginally significantly decreased anticipatory guilt for participants with identity at 1 (MCSRpresent = 2.50 vs. MCSRabsent = 3.71; bJN = -1.21, SE = .64, p = .061). Mediating role of anticipatory guilt. We tested mediation via anticipatory guilt using the PROCESS macro for SPSS (Model 8). We tested indirect effects of anticipatory guilt at every interval of LGBT support identity measures (see Table B). The results show that when LGBT support identity was greater than or equal to 5, the indirect effect of anticipatory guilt was significantly negative (b = -.33, SE = .17, 95% CI: -.673 to -.010). When LGBT support identity was 1, the indirect effect of anticipatory guilt was positive, but not significant (b = .73, SE = .41, 95% CI: -.074 to 1.519). 99 Table B. Indirect effect of anticipatory guilt on intention for dishonest behavior at different levels of LGBT support identity (study 2) LGBT Support 95% CI 95% CI Identity Effect Boot SE Lower Upper Limit Limit 1 0.727 0.406 -0.074 1.519 2 0.462 0.318 -0.163 1.080 3 0.197 0.238 -0.271 0.652 4 -0.068 0.180 -0.422 0.281 5 -0.333 0.168 -0.673 -0.010 6 -0.598 0.209 -1.015 -0.197 7 -0.863 0.281 -1.438 -0.329 Note: Significant indirect effects are in bold. 100 Appendix E Study 3: Results for Anticipatory Guilt and Perceived Morality of Organization Anticipatory guilt. We performed a 2 (CSR activity: present vs. absent) by 3 (CSR-related identity: LGBT support congruent, neutral, incongruent) ANOVA on anticipatory guilt. There were insignificant main effects of CSR activity (F(1, 447) = 1.65, p = .20) and of LGBT support identity (F(2, 447) = 2.07, p = .13). There was a significant two-way interaction (F(2, 447) = 11.93, p < .001, η 2p =.05). Follow-up analyses showed that in the congruent identity condition, CSR (vs. no-CSR) significantly increased anticipatory guilt (MCSRpresent = 4.48, SD = 1.93 vs. MCSRabsent = 3.71, SD = 2.08; F(1, 447) = 8.80, p = .003, d = .38). In contrast, in the incongruent identity condition, CSR significantly decreased anticipatory self-threat (MCSRpresent = 2.72, SD = 2.04 vs. MCSRabsent = 4.52, SD = 2.22; F(1, 447) = 15.47, p < .001, d = .84). In the neutral identity condition, anticipatory self-threat was unaffected (MCSRpresent = 3.88, SD = 2.11 vs. MCSRabsent = 3.68, SD = 2.15; F(1, 447) = .28, p = .60). We tested mediation via anticipatory guilt (H2) using the PROCESS macro (Model 8). We included CSR (0 = absent, 1 = present) as the independent variable, LGBT support identity as the moderator, anticipatory guilt as the mediator, and intention for dishonest behavior as the dependent variable. The results show that in the congruent identity condition, the indirect effect of anticipatory guilt was significantly negative (b = -.53, SE = .18, 95% CI: -.876 to -.183). In the incongruent identity condition, the indirect effect of anticipatory guilt was significantly positive (b = 1.25, SE = .34, 95% CI: .595 to 1.905). In the neutral identity condition, the indirect effect was not significant (b = -.14, SE = .27, 95% CI: -.681 to .395). 101 Perceived morality of the organization. For perceived morality of the organization, a two-way ANOVA revealed an insignificant main effect of CSR activity (F(1, 447) = .07, p = .80), a significant main effect of LGBT support identity (F(2, 447) = 15.79, p < .001, η 2p = 0.06; McongruentCSR = 5.43 vs. MneutralCSR = 4.83 vs. MincongruentCSR = 4.59), and a significant two-way interaction (F(2, 447) = 34.67, p < .001, η 2p = 0.13). In the congruent identity condition, CSR (vs. no-CSR) significantly increased perceived morality of the organization (F(1, 447) = 55.58, p < .001, d = .98; MCSRpresent = 6.07, SD = 1.12 vs. MCSRabsent = 4.80, SD = 1.45). In contrast, in the incongruent identity condition, CSR (vs. no-CSR) significantly decreased perceived morality of the organization (F(1, 447) = 27.46, p < .001, d = 1.06; MCSRpresent = 3.79, SD = 1.57 vs. MCSRabsent = 5.37, SD = 1.42). In the neutral identity condition, perceived morality of the organization was unaffected (F(1, 447) = .65, p = .42; MCSRpresent = 4.93, SD = 1.20 vs. MCSRabsent = 4.73, SD = 1.56). The mediation analysis revealed that in all identity conditions, the indirect effects of the perceived morality of the organization were not significant (congruent: b = .04, SE = .11, 95% CI: -.160 to .265; neutral: b = .01, SE = .03, 95% CI: -.047 to .071; incongruent: b = -.05, SE = .13, 95% CI: -.313 to .224). Moreover, when we tested the mediating role of anticipatory self-threat and perceived morality of the organization simultaneously, only the anticipatory self-threat was as a significant mediator in the congruent and incongruent LGBT support identity conditions, and perceived morality of the organization did not significantly mediate in any LGBT support identity conditions. 102 Appendix F Study 1: Stimuli of Dishonest Behavior Task We are currently conducting a Mobile Device Survey to collect data on mobile device usage behavior. It is very important for us to collect data from participants with different lifestyles and opinions. To do that, we created quotas for different groups of users. Some of the quotas have already been filled. To see whether you are eligible for the survey, we will ask a few eligibility questions. If you are eligible, you will be redirected to an additional 1-2 minutes Mobile Device Survey after completing other studies. If you complete this 1-2 minutes Mobile Device Survey, you will receive an additional $0.25 as a bonus payment, in addition to your baseline payment of $0.85. If you are not eligible, you will not complete Mobile Device Survey, but will complete other studies and receive your baseline payment of $0.85. Please click on the arrow button below to complete eligibility questions. ---page_break----- Eligibility Questions Which of the following devices do you use most frequently? Laptop Smartphone Tablet (e.g., iPad, Surface, Galaxy Tab) Smartwatch None of these Which best describes you about electronic devices and technology? Early adopter - I am among the first people to check out a new electronic device or gadget. Follower - I tend to wait a long time to try a new electronic device or gadget. Which of the following best describes how you would like to use mobile devices? I would like to use mobile devices more often. I would like to use mobile devices at just right amount. I would like to use mobile devices less often. ---page_break----- 103 [Construal level manipulation was inserted here.] ---page_break----- [We provide the stimuli for those who indicated themselves as an “early adopter” in the previous eligibility question.] Qualification Results You indicated that you described yourself as an early adopter of electronic devices and technology. Unfortunately, the quota for early adopter has been met and thus you are not eligible to participate in a subsequent paid survey. We are aware that respondents sometimes check a wrong answer by mistake. To prevent this from happening, we will ask the same eligibility question once again. Please check one of the options below. - I describe myself as a follower of electronic devices and technology. Thus, I am eligible for the Mobile Device survey. (If you select this option, you will be redirected to the 1-2 minute Mobile Device survey after completing other surveys. You will receive $0.25 as a bonus payment.) - I describe myself as an early adopter of electronic devices and technology. Thus, I am not eligible for the Mobile Device survey. (If you select this option, you will not complete Mobile Device survey and not receive $0.25 of a bonus payment.) 104 Appendix G Studies 1 and 2 (Post-Hoc Test): Manipulation Check of Construal Level Manipulation Task Behavior Classification Survey Any behavior can be identified in many ways. For example, one person might describe a behavior as "typing a paper," while another might describe the behavior as "pushing keys." Yet another person might describe the behavior as "expressing thoughts." We are interested in your personal preferences for how a number of different behaviors should be described. On the following pages you will find several different behaviors listed. After each behavior will be two choices of different ways in which the behavior might be identified. Here is an example: Attending class ____ sitting in a chair ____ looking at the blackboard Your task is to choose the identification that best describes the behavior for you. Please mark only one alternative. Of course, there are no right or wrong answers. People’s preferences for expressing behaviors in one way or another is situation-specific, and we are interested in your personal, momentary preferences. Please proceed to the next page if you understood the instructions. Be sure to mark your choice for each behavior. Remember, choose the description that you personally believe is more appropriate in each pair. ------ Page Break ------ Choose the description that you personally believe is more appropriate in each pair. Be sure to mark your choice for each behavior. Making a list Getting organized Writing things down Reading Gaining knowledge Following lines of print 105 Washing clothes Removing odors from clothes Putting clothes into the machine Measuring a room for carpeting Getting ready to remodel Using a yardstick Cleaning the house Showing one's cleanliness Vacuuming the floor Painting the room Making the room look fresh Applying brush strokes Paying the rent Maintaining a place to live Writing a check Caring for houseplants Making the room look nice Watering plants Locking a door Securing the house Putting a key in the lock Filling out a personality test Revealing what you're like Answering questions Toothbrushing Preventing tooth decay Moving a brush around one's mouth Taking a test Showing one's knowledge Answering questions Greeting someone Showing friendliness Saying hello Resisting temptation Showing moral courage 106 Saying "no" Eating Getting nutrition Chewing and swallowing Traveling by car Seeing countryside Following a map Having cavity filled Protecting your teeth Going to the dentist Talking to a child Teaching a child something Using simple words Pushing a doorbell Seeing if someone's home Moving a finger 107 Appendix H Supplemental Study 1 The purpose of supplemental study 1 was to show the robustness of our findings by conducting a direct replication experiment of study 1. Method Participants and design. Four hundred five individuals completed an online study on Amazon’s Mechanical Turk website for a small payment (46.7% male; Mage = 37.15, SD = 11.85). The study was a 2 (construal level: abstract vs. concrete) x (importance of benevolence; measured) between-subjects design. Procedure. The experimental procedure was the same as that of Study 1. Results To ensure that the manipulation of construal level did not influence the importance of benevolence, we ran a one-way ANOVA. The main effect of construal level on the importance of benevolence was not significant (F(1, 402) = 0.33, p = .563). Therefore, benevolence could be used as an independent variable. To test H1, we regressed whether participants acted dishonestly (1 = lying, 0 = not lying) on construal level (0 = concrete, 1 = abstract), the mean-centered importance of benevolence (M = 7.57, SD = 1.27), and their interaction using a logistic regression. The analysis revealed nonsignificant main effects of construal level, b = -0.09, Wald χ2(1) = 0.19, p = .67, and of benevolence, b = 0.01, Wald χ2(1)= 0.01, p = .91. Contrary to prediction, the interaction effect between construal level and the importance of benevolence was not significant, b = -0.24, Wald χ2(1) = 2.17, p = .14 (see Figure A). However, the pattern of results was similar to that of 108 study 1. Spotlight analysis indicated that when participants rated the importance of benevolence strongly (+1 SD from the mean, corresponding to a value of 8.84), abstract (vs. concrete) construal directionally reduced lying (b = -0.40, Wald χ2(1) = 1.77, p = .18). Specifically, whereas 29.0% of participants in the abstract construal condition lied, 37.8% in the concrete condition lied. When participants rated the importance of benevolence weakly (-1 SD from the mean, corresponding to a value of 6.30), construal level did not significantly influence lying (b = 0.11, Wald χ2(1) = 0.53, p = .47). Figure A. The effect of construal level and importance of benevolence on dishonest behavior (supplemental study 1) 45 42.2 40 37.0 37.8 35 29.0 30 25 20 15 10 5 0 -1 SD (M = 6.30) +1 SD (M = 8.84) Importance of Benevolence Abstract Concrete Moreover, consistent with study 1, in the abstract construal condition, higher importance of benevolence marginally significantly reduced lying (b = -0.23, Exp(b) = 0.80, Wald χ2(1) = 3.65, p = .056). However, in the concrete construal condition, 109 Probability of Lying (%) the importance of benevolence did not influence lying (b = 0.01, Wald χ2(1) = 0.01, p = .91). Moreover, we found that the moderating role of the other two values (power and achievement), religiosity, and political ideology in the effect of construal level on dishonest behavior was not significant, respectively (ps > .46). We note that the interaction between construal level and universalism (b = -0.22, Wald χ2(1) = 2.08, p = .15) revealed a similar pattern of results to that of the interaction between construal level and benevolence. This may be because both benevolence and universalism are conceptually similar in that they are self-transcendence values capturing serving the interest of others (Schwartz, 1992). Or it could be just a random result. We believe that benevolence (vs. universalism) better captures the value directly relevant to dishonest behavior both theoretically and empirically. Theoretically, benevolence is measured by items such as honest and responsible, whereas universalism is measured by items such as world at peace and unity with nature. Empirically, the predicted pattern of interaction between construal level and benevolence was replicated in two studies (study 1 and supplemental study 1). 110 Appendix I Study 2: Stimuli of Deception Game This is a short experiment in decision making. In this experiment, you will be randomly matched with another participant in the research session currently occurring on another campus. Neither of you will ever know the identity of the other. The money that you earn will be actually paid to you at the end of the session, privately and in cash. Two possible monetary payments are available to you and your partner in the experiment. The two payment options are: Option A: $1 to you and $2 to the other student Option B: $2 to you and $1 to the other student The choice rests with the other student who will have to choose either option A or option B. The only information your partner will have is information sent by you in a message. That is, he or she will not know the monetary payments associated with each choice. On the next page, we will ask you to choose one of the following two possible messages, which you will send to your partner. Message 1: "Option A will earn you more than Option B." Message 2: "Option B will earn you more than Option A." We will show the other student your message and ask him or her to choose either option A or option B. To repeat, your partner's choice will determine the payments in the experiment. However, your partner will never know what sums were actually offered in the option not chosen (that is, he or she will never know whether your message was true or not). Moreover, he or she will never know the sums to be paid to you according to the different options. We will pay the two of you according to the choice made by your partner based on the message you sent to him or her. Below, please indicate if you understand the instructions. Specifically, do you understand that you will be choosing to send one message (Message 1 or Message 2) to your partner and they will never know your identity or the monetary amounts of each option? [Yes] [No] - Please read the instructions again before proceeding to the next page 111 ------ Page Break ------ Now, we will ask you to indicate the message you want to send to your partner. Recall the two payment options are: Option A: $1 to you and $2 to the other student Option B: $2 to you and $1 to the other student From past research, we know that over 80% of the time the partner chooses the option that is stated in the message you send to earn him or her more. Therefore, if you send Message 1, there is over an 80% chance that your partner will select Option A, earning you $1 and them $2. If you send Message 2, there is over an 80% chance that your partner will select Option B, earning you $2 and them $1. Please indicate the message you wish to send to your partner on a white slip of paper at your station. Message 1: "Option A will earn you more than Option B." Message 2: "Option B will earn you more than Option A." 112 Bibliography Agerström, J., & Björklund, F. (2009). Temporal distance and moral concerns: Future morally questionable behavior is perceived as more wrong and evokes stronger prosocial intentions. Basic and Applied Social Psychology, 31, 49–59. Agrawal, N., & Wan, E. W. (2009). Regulating risk or risking regulation? Construal levels and depletion effects in the processing of health messages. Journal of Consumer Research, 36(3), 448–462. Aquino, K. (1998). The effects of ethical climate and the availability of alternatives on the use of deception during negotiation. International Journal of Conflict Management, 9(3), 195–217. Aquino, K., Freeman, D., Reed, A., Lim, V. K. G., & Felps, W. (2009). Testing a social-cognitive model of moral behavior: The interactive influence of situations and moral identity centrality. Journal of Personality and Social Psychology, 97(1), 123–141. Aquino, K., & Reed II, A. (2002). The self-importance of moral identity. Journal of Personality and Social Psychology, 83(6), 1423–1440. AT&T (2017). AT&T diversity and inclusion. Retrieved from: . Batson, C. D., Dyck, J. L., Brandt, J. R., Batson, J. G., Powell, A. L., McMaster, M. R., et al. (1988). Five studies testing two new egoistic alternatives to the empathy-altruism hypothesis. Journal of Personality and Social Psychology, 55(1), 52-77. Baumeister, R. F., Bratslavsky, E., Finkenauer, C., & Vohs, K. D. (2001). Bad is stronger than good. Review of General Psychology, 5(4), 323–370. Bolton, L. E., & Mattila, A. S. (2015). How does corporate social responsibility affect consumer response to service failure in buyer-seller relationships?. Journal of Retailing, 91(1), 140–153. Bolton, L. E., & Reed II, A. (2004). Sticky priors: The perseverance of identity effects on judgment. Journal of Marketing Research, 41(4), 397–410. Bostyn, D. H., Sevenhant, S., & Roets, A. (2018). Of mice, men, and trolleys: Hypothetical judgment versus real-life behavior in trolley-style moral dilemmas. Psychological Science, 29(7), 1084–1093. Brammer, S., Millington, A., & Rayton, B. (2007). The contribution of corporate social responsibility to companyal commitment. The International Journal of Human Resource Management, 18(10), 1701–1719. 113 Brass, D. J., Butterfield, K. D., & Skaggs, B. C. (1998). Relationships and unethical behavior: A social network perspective. The Academy of Management Review, 23(1), 14–31. Chernev, A., & Blair, S. (2015). Doing well by doing good: The benevolent halo of corporate social responsibility. Journal of Consumer Research, 41(6), 1412– 1425. Cheryan, S., & Monin, B. (2005). “Where are you really from?”: Asian Americans and identity denial. Journal of Personality and Social Psychology, 89(5), 717– 730. Cho, E. K., Khan, U., & Dhar, R. (2013). Comparing apples to apples or apples to oranges: The role of mental representation in choice difficulty. Journal of Marketing Research, 50(August), 505–516. Cohen, T. R., Wolf, S. T., Panter, A. T., & Insko, C. A. (2011). Introducing the GASP scale: A new measure of guilt and shame proneness. Journal of Personality and Social Psychology, 100, 947–966. Cone Communications (2017). 2017 Cone Communications CSR study. Retrieved from: . Du, S., Bhattacharya, C. B., & Sen, S. (2007). Reaping relational rewards from corporate social responsibility: The role of competitive positioning. International Journal of Research in Marketing, 24(3), 224–241. European Commission (2011). A renewed EU strategy 2011–14 for corporate social responsibility. Retrieved from: . Eyal, T., Liberman, N., & Trope, Y. (2008). Judging near and distant virtue and vice. Journal of Experimental Social Psychology, 44(4), 1204–1209. FeldmanHall, O., Mobbs, D., Evans, D., Hiscox, L., Navrady, L., & Dalgleish, T. (2012). What we say and what we do: The relationship between real and hypothetical moral choices. Cognition, 123, 434–441. Förster, J., Friedman, R. S., & Liberman, N. (2004). Temporal construal effects on abstract and concrete thinking: Consequences for insight and creative cognition. Journal of Personality and Social Psychology, 87(2), 177–189. Freitas, A. L., Gollwitzer, P., & Trope, Y. (2004). The influence of abstract and concrete mindsets on anticipating and guiding others’ self-regulatory efforts. Journal of Experimental Social Psychology, 40(6), 739–752. 114 Fujita, K., Trope, Y., Liberman, N., & Levin-Sagi, M. (2006). Construal levels and self-control. Journal of Personality and Social Psychology, 90(3), 351–367. Gino, F., & Ariely, D. (2012). The dark side of creativity: Original thinkers can be more dishonest. Journal of Personality and Social Psychology, 102(3), 445– 459. Gino, F., Ayal, S., & Ariely, D. (2013). Self-serving altruism? The lure of unethical actions that benefit others. Journal of Economic Behavior & Company, 93, 285–292. Gino, F., Schweitzer, M. E., Mead, N. L., & Ariely, D. (2011). Unable to resist temptation: How self-control depletion promotes unethical behavior. Organizational Behavior and Human Decision Processes, 115(2), 191–203. Gneezy, U. (2005). Deception: The role of consequences. American Economic Review, 95(1), 384–394. Gong, H., & Medin, D. L. (2012). Construal levels and moral judgment: Some complications. Judgment and Decision Making, 7(5), 628–638. Graham, J., Haidt, J., & Nosek, B. A. (2009). Liberals and conservatives rely on different sets of moral foundations. Journal of Personality and Social Psychology, 96(5), 1029–1046. Greenwald, A. G. (1980). The totalitarian ego: Fabrication and revision of personal history. American Psychologist, 35(7), 603–618. Habel, J., Schons, L. M., Alavi, S., & Wieseke, J. (2016). Warm glow or extra charge? The ambivalent effect of corporate social responsibility activities on customers’ perceived price fairness. Journal of Marketing, 80(1), 84–105. Hayes, Andrew F. (2018), Introduction to mediation, moderation, and conditional process analysis: A regression-based approach, NY: Guilford Press. Hershfield, H. E., Cohen, T. R., & Thompson, L. (2012). Short horizons and tempting situations: Lack of continuity to our future selves leads to unethical decision making and behavior. Organizational Behavior and Human Decision Processes, 117(2), 298–310. Hoeffler, S., & Keller, K. L. (2002). Building brand equity through corporate societal marketing. Journal of Public Policy & Marketing, 21(1), 78–89. Insurance Information Institute (2017). Background on insurance fraud. Retrieved from < https://www.iii.org/article/background-on-insurance-fraud>. 115 Janoff-Bulman, R., Sheikh, S., & Hepp, S. (2009). Proscriptive versus prescriptive morality: two faces of moral regulation. Journal of Personality and Social Psychology, 96(3), 521–537. Van Kerckhove, A., Geuens, M., & Vermeir, I. (2014). The floor is nearer than the sky: How looking up or down affects construal level. Journal of Consumer Research, 41(6), 1358-1371. Koopman, J., Howe, M., Johnson, R. E., Tan, J. A., & Chang, C. H. (2013). A framework for developing word fragment completion tasks. Human Resource Management Review, 23(3), 242-253. Koschate-Fischer, N., Stefan, I. V., & Hoyer, W. D. (2012). Willingness to pay for cause-related marketing: the impact of donation amount and moderating effects. Journal of Marketing Research, 49(December), 910–927. Kouchaki, M., Smith-Crowe, K., Brief, A. P., & Sousa, C. (2013). Seeing green: Mere exposure to money triggers a business decision frame and unethical outcomes. Organizational Behavior and Human Decision Processes, 121(1), 53–61. Lammers, J. (2012). Abstraction increases hypocrisy. Journal of Experimental Social Psychology, 48(2), 475–480. Liberman, N., & Trope, Y. (1998). The role of feasibility and desirability considerations in near and distant future decisions: A test of temporal construal theory. Journal of Personality and Social Psychology, 75(1), 5-18. Mallory, D. B., & Rupp, D. E. (2014). Good leadership: using corporate social responsibility to enhance leader-member exchange. In I.T. Bauer & B. Erdogan (Eds.), The oxford handbook of leader-member exchange, UK: Oxford University. Mazar, N., Amir, O., & Ariely, D. (2008). The dishonesty of honest people: A theory of self-concept maintenance. Journal of Marketing Research, 45(6), 633–644. McShane, L., & Cunningham, P. (2012). To thine own self be true? Employees’ judgments of the authenticity of their company’s corporate social responsibility program. Journal of Business Ethics, 108(1), 81-100. Mehta, R., Zhu R., & Meyers-Levy, J. (2014). When does a higher construal level increase or decrease indulgence? Resolving the myopia versus hyperopia puzzle,” Journal of Consumer Research, 41 (2), 475–488. Meloy, M. G., Russo, J. E., & Miller, E. G. (2006). Monetary incentives and mood. Journal of Marketing Research, 43(May), 267–275. 116 MSCI ESG Stats Database (2010). How to use KLD stats & ESG ratings definitions. Retrieved from . Peloza, J., White, K., & Shang, J. (2013). Good and guilt free: The role of self- accountability in influencing preferences for products with ethical attributes. Journal of Marketing, 77(January), 1–52. Reed, A., Aquino, K., & Levy, E. (2007). Moral identity and judgments of charitable behaviors. Journal of Marketing, 71(1), 178–193. Reed, A., Forehand, M. R., Puntoni, S., & Warlop, L. (2012). Identity-based consumer behavior. International Journal of Research in Marketing, 29(4), 310–321. Reynolds, S. J., & Ceranic, T. L. (2007). The effects of moral judgment and moral identity on moral behavior: an empirical examination of the moral individual. The Journal of Applied Psychology, 92(6), 1610–1624. Rixom, J., & Mishra, H. (2014). Ethical ends: Effect of abstract mindsets in ethical decisions for the greater social good. Organizational Behavior and Human Decision Processes, 124, 110–121. De Roeck, K., Marique, G., Stinglhamber, F., & Swaen, V. (2014). Understanding employees’ responses to corporate social responsibility: Mediating roles of overall justice and organisational identification. The International Journal of Human Resource Management, 25(1), 91–112. Rozin, P., & Royzman, E. B. (2001). Negativity bias, negativity dominance, and contagion. Personality and Social Psychology Review, 5(4), 296–320. Rupp, D. E., & Mallory, D. B. (2015). Corporate social responsibility: Psychological, person-centric, and progressing. Annual Review of Organizational Psychology and Organizational Behavior, 2, 211–236. Sachdeva, S., Iliev, R., & Medin, D. L. (2009). Sinning saints and saintly sinners: The paradox of moral self-regulation. Psychological Science, 20(4), 523–528. Sanitioso, R., Kunda, Z., & Fong, G. T. (1990). Motivated recruitment of autobiographical memories. Journal of Personality & Social Psychology, 59(2), 229–241. Schaumberg, R. L., & Wiltermuth, S. S. (2014). Desire for a positive moral self- regard exacerbates escalation of commitment to initiatives with prosocial aims. Companyal Behavior and Human Decision Processes, 123(2), 110–123. Schnall, S., Haidt, J., Clore, G., & Jordan, A. (2008). Disgust as embodied moral judgment. Personality and Social Psychology Bulletin, 34, 1096–1109 117 Schwartz, S. H. (1992). Universals in the content and structure of values: Theoretical advances and empirical tests in 20 countries. In M. P. Zanna (Ed.), Advances in experimental social psychology (Vol. 25, pp. 1-65). New York: Academic Press. Schwartz, S. H., & Bardi, A. (2001). Value hierarchies across cultures: Taking a similarities perspective. Journal of Cross-Cultural Psychology, 32(3), 268– 290. Sen, S., & Bhattacharya, C. B. (2001). Does doing good always lead to doing better? Consumer reactions to corporate social responsibility. Journal of Marketing Research, 38(2), 225–243. Sherman, D. A. K., Nelson, L. D., & Steele, C. M. (2000). Do messages about health risks threaten the self? Increasing the acceptance of threatening health messages via self-affirmation. Personality & Social Psychology Bulletin, 26(9), 1046–1058. Spiller, S. A., Fitzsimons, G. J., Lynch Jr., J. G., & McClelland, G. H. (2013). Spotlights, floodlights, and the magic number zero: Simple effects tests in moderated regression. Journal of Consumer Research, 50(April), 277–288. Starbucks (2017). Social impact. Retrieved from: < https://www.starbucks.com/responsibility>. Steele, C. M. (1988). The psychology of self-affirmation: Sustaining the integrity of the self. In L. Berkowitz (Ed.), Advances in experimental social psychology, Vol. 21, New York: Academic Press, 261–302. Steenhaut, S., & Van Kenhove, P. (2006). The mediating role of anticipated guilt in consumers’ ethical decision-making. Journal of Business Ethics, 69(3), 269– 288. Swann Jr, W. B., Pelham, B. W., & Krull, D. S. (1989). Agreeable fancy or disagreeable truth? Reconciling self-enhancement and self- verification. Journal of Personality and Social Psychology, 57(5), 782-791. Torelli, C. J., & Kaikati, A. M. (2009). Values as predictors of judgments and behaviors: The role of abstract and concrete mindsets. Journal of Personality and Social Psychology, 96(1), 231–247. Torelli, C. J., Monga, A. B., & Kaikati, A. M. (2012). Doing poorly by doing good: Corporate social responsibility and brand concepts. Journal of Consumer Research, 38(5), 948–963. Total Retail (2018). 2017 consumer returns in the retail industry. Retrieved from: < http://www.mytotalretail.com/resource/2017-consumer-returns-retail- industry/>. 118 Trope, Y., & Liberman, N. (2010). Construal-level theory of psychological distance. Psychological Review, 117(2), 440–63. Trudel, R., Argo, J. J., & Meng, M. D. (2016). The recycled self: Consumers’ disposal decisions of identity-linked products. Journal of Consumer Research, 43, 246–264. Vallacher, R. R., & Wegner, D. M. (1989). Levels of personal agency: Individual variation in action identification. Journal of Personality and Social Psychology, 57(4), 660–671. White, K., Argo, J. J., & Sengupta, J. (2012). Dissociative versus associative responses to social identity threat: The role of consumer self-construal. Journal of Consumer Research, 39(4), 240–255. Winterich, K. P., & Barone, M. J. (2011). Warm glow or cold, hard cash? Social identify effects on consumer choice for donation versus discount promotions. Journal of Marketing Research, 48(5), 855–68. Winterich, K. P., Mittal, V., & Morales, A. C. (2014). Protect thyself: How affective self-protection increases self-interested, unethical behavior. Organizational Behavior and Human Decision Processes, 125(2), 151–161. Wirtz, J., & McColl-Kennedy, J. R. (2010). Opportunistic customer claiming during service recovery. Journal of the Academy of Marketing Science, 38(5), 654– 675. Yam, K. C., Chen, X. P., & Reynolds, S. J. (2014). Ego depletion and its paradoxical effects on ethical decision making. Organizational Behavior and Human Decision Processes, 124(2), 204-214. Zhong, C.-B., & Liljenquist, K. (2006). Washing away your sins: Threatened morality and physical cleansing. Science, 313(5792), 1451–1452. 119