American Economic Review: Papers and Proceedings 2011, 101:3, 423–428 http://www.aeaweb.org/articles.php?doi 10.1257/aer.101.3.423 =

The Limits of Transparency: Pitfalls and Potential of Disclosing Conflicts of Interest

By George Loewenstein, Daylian M. Cain, and Sunita Sah*

A conflict of interest of the type discussed whether in business, government, media, or aca- here occurs when one’s professional responsibil- demia—include, or are limited to, disclosure. ities conflict with one’s personal interests, which In a world of perfectly rational and informed are often material but could also involve other advisors and advisees, disclosure would proba- goals, such as power or status. Conflicts of inter- bly be an effective remedy for conflicts of inter- est have been at the heart of many recent eco- est see, e.g., Vincent P. Crawford and Sobel nomic crises. For example, the sudden demise 1982( . In principle, disclosure should allow of high-flying corporations such as Enron and the recipient) of advice to discount that advice. WorldCom has been attributed, at least in part, Anticipating such a reaction, advisors who want to the fact that accounting firms auditing those their advice to be valued should be motivated corporations were also providing lucrative con- to steer clear of conflicts of interest or provide sulting services that could have been jeopardized credible assurances that conflicts have been pro- by an unfavorable audit. In the recent mortgage fessionally managed. In fact, prior research on crisis, credit-rating agencies that evaluated the public information disclosure in contexts such mortgage-backed securities were hired and fired as health and safety warnings has generally by firms whose bonds they were rating—a situ- found that beneficial effects when they occur ation that still exists in auditing. In the dot-com are more likely to result from( the behavior of) bubble, firms that were underwriting IPOs were the targets rather than of the recipients of disclo- also giving investment advice to their retail cli- sures Archon Fung, Mary Graham, and David ents. Further, the dramatic increases in health- Weil 2007( . care costs in the United States over past decades However,) there are also many reasons that have been fueled in part by conflicts of interest, disclosure can have unintended consequences, including payments from medical device and some of which involve errors people make in pharmaceutical companies to physicians and responding to the disclosed information, and fee-for-service remuneration schemes Jerome others that involve reactions of advisors upon P. Kassirer 2005 . ( disclosure of their extraprofessional interests. There have been) many policy efforts to deal These psychological factors are described in the with conflicts of interest—including regula- first section of this paper, and have been the focus tory interventions that led to the separation of of our research on the impact of disclosure. We auditing and consulting functions at accounting outline our experimental studies in Section II, firms—as well as legislative initiatives, includ- and suggest ways to enhance the effectiveness ing large parts of the Sarbanes–Oxley Act and of disclosure policies in Section III. the Financial Reform Bill of 2010. Across the diversity of policy interventions, however, there I. How Disclosure Can Backfire is one striking constant: the ubiquity of disclo- sure. Virtually all policies intended to mitigate A. Impact of Disclosure on Advisors the negative effects of conflicts of interest— There are two major psychological mecha- * Loewenstein: Department of Social and Decision nisms through which disclosure can influence Sciences, Carnegie Mellon University, Pittsburgh, PA conflicted advisors. Strategic exaggeration is the 15213 e-mail: [email protected] ; Cain: Yale School ( ) tendency of advisors to inflate the bias in their of Management, Yale University, New Haven, CT 06511 advice to counteract any discounting that might e-mail: [email protected] ; Sah: Fuqua School of Business,( Duke University, The )Fuqua School of Business, occur because of disclosure. Perhaps strategic 100 Fuqua Drive, Box 90120, Durham, NC 27708 e-mail: exaggeration does not make much logical sense, [email protected] . ( especially if we posit no self-restraint in the ) 423 424 AEA PAPERS AND PROCEEDINGS MAY 2011 advice-giver: if the advice-giver could increase model of advisor behavior to predict the impact his or her payoff by giving more biased advice, of the conflict—let alone the disclosure of that why not do so maximally, even in the absence conflict—on the advice. Lacking such a model, of disclosure? Perfect rationality aside, however, advice recipients will not know what to do with one can easily imagine a doctor compensating the disclosed information, and, as a result, may for anticipated discounting caused by disclosure simply ignore it. by “playing up” the benefits of a new drug. Finally, as we describe below, disclosure can Moral licensing refers to the undermining create a “burden of disclosure” effect whereby of professionalism that can occur as a result of advice recipients who learn of an advisor’s con- disclosure. Although prior research has shown flict do become less trustful of advice, yet feel that the introduction of a subsidy or fine can more pressured to follow that advice. In the undermine nonmaterial e.g., altruistic or ego- absence of disclosure, for example, a patient’s istic motives Roland Bénabou( and Jean Tirole rejection of participation in a drug trial would 2006) , disclosing( a conflict of interest can like- likely be attributed to risk aversion or satisfac- wise )undermine the advisor’s motivation to tion with a currently used drug. The same rejec- adhere to professional standards. Experimental tion, following disclosure of a conflict, might be research suggests that after engaging in moral attributed to the patient’s distrust of the doctor— behavior people feel “licensed” to act immorally an implicit insinuation of corruption—which a in subsequent interactions Benoit Monin and patient is likely to be reluctant to communicate. Dale T. Miller 2001 . Disclosure( also introduces a possible rationalization) for unethical behavior: II. Experimental Studies of Disclosure a person who has received disclosure should, perhaps, “expect” bias—caveat emptor. A. Increased Bias with Disclosure

B. Impact of Disclosure on Advice Recipients In two papers, Cain, Loewenstein, and Don A. Moore 2005, 2011 examined the impact of Psychological research offers similar grounds disclosing (an experimental) facsimile of a con- for pessimism when it comes to advisees’ dis- flict of interest. The first paper 2005 examined counting of advice following disclosure of a disclosure in an experimental ( setup) in which conflict of interest. First, research on judgment “estimators” guessed the value of a jar of coins suggests that advisees are likely to “anchor” and were paid according to the accuracy of their on the advice they receive and then adjust i.e., estimates. “Advisors” were given better infor- discount insufficiently, even though they (know mation for evaluating coin-jar values and then the advice) may be biased Amos Tversky and conveyed suggested valuations to the estimators. Daniel Kahneman 1974 . ( In the no-conflict condition, advisors’ incentives Second, most research) on disclosure of con- were aligned with those of the estimators; they flicts of interest suggests that advice recipients were paid whatever their estimator was paid. are not very concerned about the information In the two conflict conditions, however, advi- they receive e.g., Lisa Bero, Stanton Glantz, and sors were paid only to the extent that estima- Mi-Kyung Hong( 2005; Christine Grady et al. tors overestimated the value of the coins. In the 2006 . Thus, for example, one study Lindsay A. disclosed-conflict condition, the misalignment of Hampson) et al., 2006 found that “more( than 90 incentives was conveyed to estimators via hand- percent of patients expressed) little or no worry written disclosures that came with the advice. In about financial ties that researchers or institu- the undisclosed-conflict condition the misalign- tions might have with drug companies.” Indeed, ment of incentives was not communicated. disclosure can lead to an increase rather than a The central results of the study, consistent with decrease in trust if the disclosure is interpreted psychological mechanisms discussed above, as a sign of honesty or if the fact that the advisor were that advisors exaggerated more when the is receiving payments is interpreted as an indica- misalignment of incentives was disclosed, and tion of professional standing Steven D. Pearson, estimators did not discount sufficiently to com- Ken Kleinman, and Donna Rusinak( 2006 . pensate for this greater exaggeration. Hence, Third, discounting advice appropriately) for a advisors ended up with higher payoffs with dis- disclosed conflict of interest requires a mental closure than without, and estimators ended up VOL. 101 NO. 3 The Limits of Transparency 425 with lower payoffs—exactly the opposite of the down advisors’ recommendations when a con- intended effect of disclosure. flict has been disclosed because they fear the The second paper 2011 presented an eco- rejection will signal the belief that the conflict nomic model of the dynamics( ) involved and, in of interest has corrupted the advisor. We call the main study, used a more naturalistic situa- the second mechanism the panhandler effect; tion somewhat akin to the interactions between consistent with the literature on “reluctant altru- a prospective home buyer and a conflicted real ism” e.g., Jason D. Dana, Cain, and Robyn estate agent who would benefit from a home- M. Dawes( 2006 , advice recipients may feel buyer’s high (valuation . Estimators in the study increased pressure) to help an advisor satisfy attempted to guess the) actual selling prices of his or her personal interests once these interests houses in Pittsburgh, and were given advice become common knowledge. by advisors who had much better informa- In experiments with real payoffs, advisors tion, including prices of comparable houses in advised “choosers” whether to choose “die-roll the neighborhood and the houses’ tax-assessed A” or “die-roll B,” each of which would result values. Four experimental conditions varied in a different, specified set of prizes e.g., a $5 whether or not there was a conflict i.e., whether Starbucks gift card if a “5” was rolled( on die the advisor gained by the estimator( giving a B . Die-roll A was superior to die-roll B, as it high or accurate estimate and whether or not had) more than twice the expected value, and the advisor’s incentives were) disclosed. Results pilot tests revealed that approximately 92 per- of this study were strikingly parallel to those of cent of participants preferred it over die-roll B. the coin-jar study; again, estimators earned less Advisors gave choosers a communication form when the conflict was disclosed than when it that contained information about the prizes asso- was not. ciated with both die-rolls and also gave a recom- Two other studies in the second paper 2011 mendations about which die-roll the choosers tested the specific psychological mechanisms( ) should pick. Some advisors were subject to a for how disclosure might bias advice. The first conflict of interest; they were informed that they, study examined strategic exaggeration and too, would get a die-roll of their choice if the found that while many advisors who changed choosers picked die-roll B the inferior choice . their advice following disclosure gave more If the choosers picked die-roll( A, the advisors) biased advice consistent with strategic exag- would receive nothing. Of these conflicted advi- geration , another,( albeit smaller, group exhib- sors, half were required to disclose their conflict ited what) could be called strategic restraint; of interest to the choosers by writing out a word- they reined in the bias of their advice, anticipat- for-word disclosure statement on the communi- ing that advisees informed of the conflict would cation form. Other conflicted advisors were told discount advice if it was too obviously exagger- not to disclose their self-interest to the choosers. ated. The second follow-up study, in contrast, In addition, there was a no-conflict condition in produced strong evidence in support of moral which advisors were not subject to a conflict of licensing; participants judged it as less unethi- interest and were rewarded regardless of the die- cal to offer biased advice with the intention of roll the choosers picked. misleading advisees if the advice was preceded Not surprisingly, nearly all unconflicted advi- by disclosure. sors recommended die-roll A 93 percent , and all choosers who received this (recommendation) B. The Burden of Disclosure complied without feeling increased pressure. Of those choosers who received bad advice from In a series of studies Sah, Loewenstein, their conflicted advisors i.e., a recommendation and Cain 2010 , again involving( interactions to pick die-roll B , 52 (percent complied with between advisors) and advice recipients, we this recommendation) in the no-disclosure con- explored the burden of disclosure—the distrust dition, but this number increased to 81 percent of advice coupled with increased pressure to with disclosure. After making their choice of comply with it—as a result of disclosure. The die-roll, but before rolling for their prize, choos- first mechanism behind such increased compli- ers answered a series of questions regarding ance is what we call insinuation anxiety; advice how they felt about their choice and the advice recipients experience greater ­discomfort ­turning they received. Choosers who received disclosure 426 AEA PAPERS AND PROCEEDINGS MAY 2011 statements reported that they trusted the advice ­significantly more uncomfortable turning down less and were also significantly less pleased the doctor’s recommendation for fear of insinu- about their choice with disclosure; but they also ating that the doctor was corrupt. Disclosure cre- indicated that they felt significantly more pres- ated a significant burden on the patient through sure to help their advisors and also felt that it these conflicting forces. would be considerably more uncomfortable to turn down their advisors’ recommendation. III. Enhancing the Beneficial Disclosure, therefore, increased the pressure to Impact of Disclosure comply with bad advice. In a follow-up study, we found that this pres- Research has identified a number of factors sure on choosers to help advisors was motivated that determine whether product safety disclo- not by pure concern for the advisors’ welfare but sures have intended or unintended consequences rather by reluctance to appear unwilling to help see Fung, Graham, and Weil 2007 . Our own the advisors once the advisors’ interests were research( has likewise identified variables) that publicly disclosed. We found that the burden of are key in determining whether disclosing a disclosure was significantly reduced i.e., that conflict of interest has beneficial or detrimental choosers were far less likely to follow (advisors’ effects. Cain, Loewenstein, and Moore 2011, bad recommendations when the disclosure was study 4 identified one such factor, finding( that provided secretly by )an external source rather people )took a single piece of conflicted advice than directly from the advisor, suggesting that it pretty much at face value, disclosure or no dis- is the common knowledge of the disclosed inter- closure, discounting only minimally. However, ests—not merely the advice-recipient’s knowl- when given advice from two advisors, one con- edge of those interests—that creates pressure to flicted and the other not, they put less weight satisfy them. on the conflicted advice. This suggests that dis- In a final role-playing scenario rather than closure may be more effective when conflicted incentive-compatible lab study , we( had par- advice is contrasted with unconflicted advice ticipants play the role of hypothetical) patients see, also, Christopher Robertson 2010 ; how- and first read about their medical history and ever,( an in-depth analysis of the results) indi- current symptoms. Each participant then lis- cated that conditions needed to be near perfect tened to a voice recording of their “doctor,” for advice recipients to properly weigh a second who identified two treatment options to the opinion. Disclosure seemed to work only when patient and recommended one of these options four ingredients came together: i the unbi- the one most benefiting the doctor . The ased advice was disclosed as unbiased,( ) ii the manipulation( between the two main conditions) biased advice was disclosed as coming (from) a was that, in the disclosure condition, the doc- conflicted source, iii the conflicted advice was tor disclosed that he would gain financially if clearly biased, and( )iv both pieces of advice the patient chose the recommended treatment were offered simultaneously.( ) option, while in the no-disclosure condition, In a series of follow-ups to the die-roll stud- no extra information was given to the patient. ies Sah, Loewenstein, and Cain 2010 , we also Despite the advice being exactly the same in examined( different remedies for the burden-of-) both conditions, the patients reported trusting disclosure effect. We found that choosers were the doctor significantly less with disclosure, less likely to follow the advisors’ bad advice were less likely to believe that their doctor when there was a “cooling-off” period between had their best interests at heart, and were less getting the advice and choosing, or when choos- likely to indicate that they would consult with ers made their decisions in private rather than that particular doctor again in the future—even in front of their advisors. Applied to medicine, though it was unclear whether the advice was these results suggest that patients should not be best for them as patients. Disclosure had dam- asked to make some decisions e.g., whether to aged the doctor-patient relationship. It could enroll in a clinical trial until they( have had time be argued that decreased trust is the intended to think about it at home,) away from the pressure purpose of disclosure. However, the patients of the doctor’s presence. who heard the disclosure statement from Other research has examined different inter- their doctor also reported that they would feel ventions that could make disclosure more VOL. 101 NO. 3 The Limits of Transparency 427

­effective. Bryan K. Church and Xi Kuang’s Bero, Lisa, Stanton Glantz, and Mi-Kyung Hong. 2009 research suggests that the ability to sanc- 2005. “The Limits of Competing Interest Dis- tion( biased) advisors may help improve disclo- closures.” Tobacco Control, 14: 118–26. sure’s efficacy. Christopher Koch and Carsten Cain, Daylian M., George Loewenstein, and Don A. Schmidt’s 2009 study suggests that many Moore. 2005. “The Dirt on Coming Clean: Per- rounds of feedback( ) may serve to educate advice verse Effects of Disclosing Conflicts of Inter- recipients on how to properly react to disclosure. est.” Journal of Legal Studies, 34 1 : 1–25. However, neither of these approaches seems Cain, Daylian M., George Loewenstein,( ) and Don practical in most real-world situations in which A. Moore. 2011. “When Sunlight Fails to Dis- bias is difficult to identify and in which opportu- infect: Understanding the Perverse Effects of nities for learning from experience are limited. Disclosing Conflicts of Interest.” Journal of For now, and especially for the inexperienced Consumer Research, 37 5 . and vulnerable, disclosure does not appear to Church, Bryan K., and Xi( Kuang.) 2009. “Con- live up to its protective promises. flicts of Interest, Disclosure, and Costly Sanctions: Experimental Evidence.” Journal( of) IV. Conclusion Legal Studies, 38 2 : 505–32. Crawford, Vincent P.,( and) Joel Sobel. 1982. “Stra- These results contribute to a growing array tegic Information Transmission.” Economet- of research that compares the effect of informa- rica, 50 6 : 1431–51. tion provision to other often more substantive Dana, Jason( )D., Daylian M. Cain, and Robyn M. policy interventions in (situations in which indi)- Dawes. 2006. “What You Don’t Know Won’t viduals are prone to make mistakes by impos- Hurt Me: Costly But Quiet Exit in Dictator ing “internalities” on themselves Richard J. Games.” Organizational( Behavior) and Human Herrnstein et al. 1993 . In the realm( of diet, Decision Processes, 100 2 : 193–201. for example, disclosure) of nutritional informa- Fung, Archon, Mary Graham,( ) and David Weil. tion, taxes and subsidies, and “nudges” that 2007. Full Disclosure: The Perils and Promise make healthful food choices more convenient of Transparency. Cambridge: Cambridge Uni- are all options for policies designed to promote versity Press. consumption of healthful foods. Much of this Grady, Christine, Elizabeth Horstmann, Jeffrey research in other domains reaches similar con- S. Sussman, and Sandra Chandros Hull. 2006. clusions to those discussed here; more informa- “The Limits of Disclosure: What Research tion, in general, is not very effective in improving Subjects Want to Know about Investigator decisions. People deserve accurate informa- Financial Interests.” Journal of Law, Medicine tion with which to make informed decisions, and Ethics, 34 3 : 592–99. so disclosure is inherently desirable. However, Hampson, Lindsay( ) A., Manish Agrawal, Steven whether and to what extent information actu- Joffe, Cary P. Gross, Joel Verter, and Ezekiel J. ally improves( economic outcomes) depends criti- Emanuel. 2006. “Patients’ Views on Financial cally on what information is delivered, how it Conflicts of Interest in Cancer Research Tri- is delivered, and how it is utilized by receivers. als.” New England Journal of Medicine, 355: While disclosure has manifest pitfalls, there are 2330–37. also enormous opportunities for designing poli- Herrnstein, Richard J., George F. Loewenstein, cies that will enhance its benefits. Care must be Drazen Prelec, and William Vaughan Jr. 1993. taken, however, to ensure that disclosure does “Utility Maximization and Melioration: not replace more effective measures, such as Internalities in Individual Choice.” Journal working harder to eliminate conflicts of interest of Behavioral Decision Making, 6 3: 149– in the first place. 85. ( ) Kassirer, Jerome P. 2005. “Physicians’ Finan- cial Ties with the Pharmaceutical Industry: A References Critical Element of a Formidable Marketing ­Network.” In Conflicts of Interest: Challenges Bénabou, Roland, and Jean Tirole. 2006. “Incen- and Solutions in Business, Law, Medicine, tives and Prosocial Behavior.” American Eco- and Public Policy, ed. Don A. Moore, Dayl- nomic Review, 96 5 : 1652–78. ian M. Cain, George Loewenstein,­ and Max ( ) 428 AEA PAPERS AND PROCEEDINGS MAY 2011

H. Bazerman, 133–41. Cambridge: Cambridge ­Physicians’ Financial Incentives to Patients.” University Press. Archives of Internal Medicine, 166: 623–28. Koch, Christopher, and Carsten Schmidt. 2009. Robertson, Christopher. 2010. “Blind Expertise.” “Disclosing Conflict of Interest—Does Expe- New York University Law Review, 85: 174– rience and Reputation Matter?” Accounting, 257. Organizations, and Society, 35 1 : 95–107. Sah, Sunita, George Loewenstein, and Daylian M. Monin, Benoit, and Dale T. Miller.( 2001.) “Moral Cain. 2010. “The Burden of Disclosure.” http:// Credentials and the Expression of Prejudice.” papers.ssrn.com/sol3/papers.cfm?abstract_ Journal of Personality and Social Psychology, id=1615025. 81 1 : 33–43. Tversky, Amos, and Daniel Kahneman. 1974. Pearson,( ) Steven D., Ken Kleinman, and Donna “Judgment under Uncertainty: Heuristics and Rusinak. 2006. “A Trial of Disclosing Biases.” Science, 185 4157 : 1124–31. ( )