WarningAboutNudged

Transcript

1 Finding Warning: You are about to be nudged George Loewenstein, Cindy Bryce, David Hagmann, & Sachin Rajpal Summary. Presenting a default option is known to influence important decisions. That includes decisions regarding advance medical directives, documents people prepare to convey which medical treatments they favor in the event that they are too ill to make their wishes clear. Some observers have argued that defaults are unethical because people are typically unaware that they are being nudged toward a decision. We informed people of the presence of default options before they completed a hypothetical advance directive, or after, then gave them the opportunity to revise their decisions. The effect of the defaults persisted, despite the disclosure, suggesting that their effectiveness may not depend on deceit. These findings may help address concerns that behavioral interventions are necessarily duplicitous or manipulative. udging people toward particular decisions by recent study, defaults also proved robust when seriously 7 N presenting one option as the default can influence ill patients completed real advance directives. 8 important life choices. If a form enrolls employees in nudges The use of such defaults or other behavioral retirement savings plans by default unless they opt out, has raised serious ethical concerns, however. The House 1 people are much more likely to contribute to the plan. report produced in the of Lords Behaviour Change Likewise, making organ donation the default option United Kingdom in 2011 contains one of the most signif- 9 rather than just an opt-in choice dramatically increases It argued that the “extent to which an icant critiques. 2 The same principle holds for other rates of donation. intervention is covert” should be one of the main criteria major decisions, including choices about purchasing for judging if a nudge is defensible. The report consid- 3,4 insurance and taking steps to protect personal data. ered two ways to disclose default interventions: directly Decisions about end-of-life medical care are similarly or by ensuring that a perceptive person could discern a susceptible to the effects of defaults. Two studies found nudge is in play. While acknowledging that the former that default options had powerful effects on the end-of- would be preferable from a purely ethical perspective, life choices of participants preparing hypothetical the report concluded that the latter should be adequate, advance directives. One involved student respondents, “especially as this fuller sort of transparency might limit 5,6 In a more and the other involved elderly outpatients. the effectiveness of the intervention.” Philosopher Luc Bovens in “The Ethics of Nudge” noted that default options “typically work best in the Loewenstein, G., Bryce, C., Hagmann, D., & Rajpal, S. (2015). Warning: You 10 (1), pp. 35–42. are about to be nudged. Behavioral Science & Policy, 1 Bovens observed the lack of disclosure in a study dark.” 35 a publication of the behavioral science & policy association

2 in which healthy foods were introduced at a school by stating their overall goals for end-of-life care and cafeteria with no explanation, prompting students to eat their preferences for specific life-prolonging measures fewer unhealthy foods. The same lack of transparency such as cardiopulmonary resuscitation and feeding existed during the rollout of the Save More Tomorrow tube insertion. Participants were randomly assigned to program, which gave workers the option of precom- receive a version of an advance directive form on which mitting themselves to increase their savings rate as their the default options favored either prolonging life or income rose in the future. Bovens noted, minimizing discomfort. For both defaults, participants were further randomly assigned to be informed about If we tell students that the order of the food the defaults either before or after completing the form. in the Cafeteria is rearranged for dietary Next, they were allowed to change their decisions using purposes, then the intervention may be less forms with no defaults included. The design of the study successful. If we explain the endowment enabled us to assess the effects of participants’ aware- effect [the tendency for people to value ness of defaults on end-of-life decisionmaking. amenities more when giving them up than We recognize that the hypothetical nature of the when acquiring them] to employees, they advance directive in our study may raise questions may be less inclined to Save More Tomorrow. about how a similar process would play out in the real world. However, recent research by two of the current When we embarked on our research into the impact authors and their colleagues examined the impact of 7 and obtained results of disclosing nudges, we understood that alerting defaults on real advance directives similar to prior work on the topic examining hypothetical people about defaults could make them feel that they 5,6 All of these studies found that the defaults choices. were being manipulated. Social psychology research has provided on advance directive forms had a major impact found that people tend to resist threats to their freedom on the final choices reached by respondents. Just as to choose, a phenomenon known as psychological 11 Thus, it is reasonable to think, as both the the question of whether defaults could influence the reactance . House of Lords report and Bovens asserted, that people choices made in advance directives was initially tested in would deliberately resist the influence of defaults (if hypothetical tasks, we test first in a hypothetical setting informed ahead of time, or ) or try to undo preinformed whether alerting participants to the default diminishes their influence (if told after the fact, or ). postinformed its impact. Such a reaction to disclosure might well reduce or even To examine the effects of disclosing the presence of eliminate the influence of nudges. we recruited via e-mail 758 participants (out defaults, But our findings challenge the idea that fuller trans- of 4,872 people contacted) who were either alumni of parency substantially harms the effectiveness of defaults. New York Times Carnegie Mellon University or readers If what we found is confirmed in broader contexts, fuller who had consented to be contacted for research. disclosure of a nudge could potentially be achieved Respondents were not paid for participating. Although with little or no negative impact on the effectiveness of not a representative sample of the general population, the intervention. That could have significant practical the 1,027 people who participated included a large applications for policymakers trying to help people make proportion of older individuals for whom the issues choices that are in their and society’s long-term interests posed by the study are salient. The mean age for both while disclosing the presence of nudges. samples was about 50 years, an age when end-of-life care tends to become more relevant. (Detailed descrip- tions of the methods and analysis used in this research Testing Effects from Disclosing Defaults are published online in the Supplemental Material.) We explored the impact of disclosing nudges in a study Our sample populations are more educated than the - of individual choices on hypothetical advance direc U.S. population as a whole, which reduces the extent to tives, documents that enable people to express their which we can generalize the results to the wider popu- preferences for medical treatment for times when lation. However, the study provides information about they are near death and too ill to express their wishes. whether the decisions of a highly educated and presum- Participants completed hypothetical advance directives ably commensurately deliberative group are changed behavioral science & policy | spring 2015 36

3 by their awareness of being defaulted, that is, having the • mechanical ventilator use, described as “machines default options selected for them should they not take that assist spontaneous breathing, often using action to change them. Prior research has documented either a mask or a breathing tube.” larger default effects for individuals of lower socioeco- 1,12 which suggests that the default effects nomic status, The advance directive forms that participants we observe would likely be larger in a less educated completed randomly defaulted them into either population. accepting or rejecting each of the life-prolonging treat- ments. Those preinformed about the use of defaults were told before filling out the form; those postinformed Obtaining End-of-Life Preferences learned after completing the form. Participants completed an online hypothetical advance One reason that defaults can have an effect is that directive form. First, they were asked to indicate their they are sometimes interpreted as implicit recommen- 2,13–15 This is unlikely in our study, because both broad goals for end-of-life care by selecting one of the dations. groups were informed that other study participants had following options: been provided with forms populated with an alterna- • I want my health care providers and agent to tive default. This disclosure also rules out the possibility pursue treatments that help me to live as long as that respondents attached different meanings to opting possible, even if that means I might have more pain into or out of the life-extending measures (for example, or suffering. donating organs is seen as more altruistic in countries in I want my health care providers and agent to pursue • which citizens must opt in to donate than in countries in treatments that help relieve my pain and suffering, 16 or the possi- which citizens must opt out of donation) even if that means I might not live as long. bility that the default would be perceived as a social norm I do not want to specify one of the above goals. • (that is, a standard of desirable or common behavior). My health care providers and agent may direct the After completing the advance directive a first time overall goals of my care. (either with or without being informed about the default at the outset), both groups were then asked to complete Next, participants expressed their preferences the advance directive again, this time with no defaults. regarding five specific medical life-prolonging interven- - Responses to this second elicitation provide a conser tions. For each question, participants expressed a pref- vative test of the impact of defaults. Defaults can influ- erence for pursuing the treatment (the prolong option), ence choices if people do not wish to exert effort or declining it (the comfort option), or leaving the decision are otherwise unmotivated to change their responses. to a family member or other designated person (the - Requiring people to complete a second advance direc no-choice option). The specific interventions included tive substantially reduces marginal switching costs the following: (that is, the additional effort required to switch) when compared with a traditional default structure in which • cardiopulmonary resuscitation, described as people only have to respond if they want to reject the “manual chest compressions performed to restore default. In our two-stage setup, participants have already blood circulation and breathing”; engaged in the fixed cost (that is, expended the initial • dialysis (kidney filtration by machine); effort) of entering a new response, so the marginal cost feeding tube insertion, described as “devices • of changing their response should be lower. The fact used to provide nutrition to patients who cannot that the second advance directive did not include any swallow, inserted either through the nose and defaults means that the only effect we captured is a esophagus into the stomach or directly into the carryover from the defaults participants were given in stomach through the belly”; the first version they completed. intensive care unit admission, described as a • In sum, the experiment required participants to “hospital unit that provides specialized equipment, make a first set of advance directive decisions in which services, and monitoring for critically ill patients, a default had been indicated and then a second set such as higher staffing-to-patient ratios and venti- of decisions in which no default had been indicated. lator support”; and 37 a publication of the behavioral science & policy association

4 Experimental design Table 1. Group 2: Group 3: Group 4: Group 1: Prolong preinformed Comfort postinformed Prolong postinfomed Comfort preinformed Disclosure Disclosure Choice 1 Choice 1 Choice 1 Choice 1 Comfort default Prolong default Comfort default Prolong default Disclosure Disclosure Choice 2 Choice 2 Choice 2 Choice 2 No default No default No default No default Participants were randomly assigned into one of four Supplemental Material. Here we summarize our most groups in which they were either preinformed or post- pertinent findings, which are presented numerically in informed that they had been assigned either a prolong Table 2 and depicted visually in Figures 1 and 2. default or a comfort default for their first choice, as Participants showed an overwhelming preference depicted in Table 1. for minimizing discomfort at the end of life rather The disclosure on defaults for the preinformed group - than prolonging life, especially for the general direc read as follows: tives (see Figure 1). When the question was posed in general terms, more than 75% of responses reflected The specific focus of this research is on this general goal in all experimental conditions and “defaults”—decisions that go into effect if people both choice stages. By comparison, less than 15% of don’t take actions to do something different. responses selected the goal of prolonging life, with Participants in this research project have been the remaining participants leaving that decision to divided into two experimental groups. someone else. If you have been assigned to one group, the Advance Directive you complete will have The impact of defaults on overall Figure 1. answers to questions checked that will direct e goal for car health care providers to help relieve pain and suffering even it means not living as long. If Percent choosing each option you want to choose different options, you will 100 be asked to check off a different option and place your initials beside the different option you select. 75 If you have been assigned to the other group, the Advance Directive you complete 50 will have answers to questions checked that will direct health care providers to prolong your life as much as possible, even if it means 25 you may experience greater pain and suffering. The disclosure for the postinformed group was the same, 0 except that participants in this group were told that that Comfort preinformed Comfort postinformed they had been defaulted rather than would be defaulted. Prolong No choice Comfort 95% confidence intervals. are to Error included indicate The bars bars Capturing Effects from Disclosing Nudges display exists among data from each group. If two how much variation error of quarter not do (or length total their bars a than less by overlap A detailed description of the results and our anal- overlap), the probability that the dierences were observed by chance is yses of those data are available online in this article’s less than 5% (i.e., statistical significance at p 05). <. behavioral science & policy | spring 2015 38

5 Table 2. Percentage choosing goal and treatment options by stage, default, and condition Choice 1 Choice 2 Comfort default Prolong default Comfort default Prolong default Post- Post- Post- Pre- Pre- Pre- Pre- Post- informed informed informed informed informed informed informed informed Choice Question 76.9% 80.5% 78.2% 76.0% 79.8% 79.7% Overall goal Choose comfort 81.6% 81.7% 15.4% 16.1% 12.5% 12.8% Do not choose 12.8% 14.5% 7.5% 7.5% 5.6% Choose prolong 5.6% 5.8% 12.0% 5.6% 11.2% 7.7% 12.8% 53.8% 30.2% 41.2% 46.9% 50.7% Choose comfort Average of 47.3% 36.3% 45.4% 5 specific 26.6% 30.4% 24.6% 28.2% 22.1% 22.4% 20.9% 28.8% Do not choose treatments 32.5% 22.3% 21.6% 41.6% 37.9% 24.2% 26.9% Choose prolong 37.1% Preferences for comfort in the general directive Unlike the results for general directives, defaults were so fixed that they were not affected by defaults for specific treatments, when the participant is only or disclosure of defaults (that is, choices did not differ informed after the fact, are effective (see Figure 2A in by condition in Figure 1). We note that these results Figure 2). We could observe this after averaging across 7 differ from recent work using real advance directives the five specific interventions that participants consid- in which defaults had a large impact on participants’ ered: On this combined measure, 46.9% of participants general goals. One possible explanation is that the highly who were given the comfort default (but not informed educated respondents in our study had more definitive about it in advance) expressed a preference for comfort. preferences about end-of-life care than did the less By comparison, only 30.2% of those given the prolong educated population from the earlier article. default (again with no warning about defaults) expressed The impact of default on responses to specific treatments Figure 2. Percent choosing each option C. Second choice after B. When aware of default A. When unaware of default being made aware of default 100 75 50 25 0 Prolong Prolong Comfort Comfort Prolong Comfort postinformed preinformed postinformed postinformed preinformed postinformed Prolong No choice Comfort data from If two error bars bars are included to indicate 95% confidence intervals. The bars display how each group. among exists variation much Error overlap 5% than is less chance (i.e., a quarter than less by of their total length (or do not overlap), the probability that the dierences were observed by statistical significance at <. 05). p 39 a publication of the behavioral science & policy association

6 a preference for comfort (a difference of 17 percentage directives did not include defaults, so any effect of points, or 36% [17/46.9]). defaults reflects a carryover effect from the first-stage The main purpose of the study was to examine the choice. (More detailed analysis of our results and more impact on nudge effectiveness of informing people information listed by specific treatments are available in that they were being nudged, a question that is best the online Supplemental Material.) addressed by analyzing the effects of preinforming people about directive choices. Figure 2B presents the Defaults Survive Transparency impact of the default when people were preinformed. Despite extensive research questioning whether advance As can be seen in the figure, preinforming people about directives have the intended effect of improving quality defaults weakened but did not wipe out their effective- 17,18 they continue to be one of the of end-of-life care, ness (see Figure 2B). When participants completed the few and major tools that exist to promote this goal. advance directive after being informed about the impact Combining advance directives with default options of the defaults, 50.7% of participants given the comfort could steer people toward the types of comfort options default expressed a preference for comfort, compared for end-of-life care that many experts recommend with only 41.2% of those given the prolong life default (a and that many people desire for themselves. This study difference of 10 percentage points, or 19%). Although all suggests such defaults can be transparently imple- specific treatment choices were affected by the default mented, addressing the concerns of many ethicists in the predicted direction, the effect is statistically signif- without losing defaults’ effectiveness. icant only for a single item (dialysis) and for the average More broadly, our findings demonstrate that default of all five items (see the Supplemental Material). Prein- options are a category of nudges that can have an effect - forming participants about the default may have weak even when people are aware that they are in play. Our ened its impact, but did not eliminate the default’s effect. results are conservative in two ways. First, not only were Postinforming people that they have been defaulted respondents informed that they were about to be or had and then asking them to choose again in a neutral way, - been defaulted, but they also learned that other partic with no further nudge, produces a substantial default ipants received different defaults, thereby eliminating effect that is not much smaller than the standard any implicit recommendation in the default. Given that default effect, as seen in Figure 2C. When participants the nudge continued to have an impact, we can only completed the advance directive a second time (this conjecture that the default effect would have been even time without a default), having been informed after the more persistent if the warning informed them that they fact that they had been defaulted, 47.3% of participants had been defaulted deliberately to the choice that poli- given the comfort default expressed a preference for cymakers believe is the best option. comfort, compared with only 36.3% of those given Second, our results are conservative in the sense that the prolong life default (a difference of 11 percentage the second advance directive that participants completed points, or 23%). Again, postinforming participants about contained no defaults, so the effect of the initial default the default and allowing them to change their decision had to carry over to the second choice. Our experi- may have weakened its impact, but did not eliminate the mental design minimized the added cost of switching: default’s effect. Regardless of whether they wanted to switch, respon- These results are important because they suggest that dents had to provide a second set of responses. Presum- either a preinforming or a postinforming strategy can ably, the impact of the initial default would have been be effective in both disclosing the presence of a nudge even stronger if switching had required more effort for and preserving its effectiveness. In addition, the results respondents than sticking with their original response. provide a conservative estimate of the power of defaults What exactly produced the carryover effect remains because all respondents who were informed at either - uncertain. It is possible, and perhaps most inter stage had, by the second stage, been informed both that esting, that the prior default led respondents to think they had been randomly selected to be defaulted and about the choice in a different way, specifically in a - that others had been randomly selected to receive alter way that reinforced the rationality of the default they native defaults. In addition, the second-stage advance behavioral science & policy | spring 2015 40

7 were presented with (consistent with reference 16). It author affiliation is, however, also possible that the respondents were mentally lazy and declined to exert effort to reconsider Loewenstein and Hagmann, Department of Social and their previous decisions. Decision Sciences, Carnegie Mellon University; Bryce, Although the switching costs in our study design were Graduate School of Public Health, University of Pitts- burgh; Rajpal, Bethesda, Maryland. Corresponding small, such costs may explain why we observed default author’s e-mail: [email protected] effects for the specific items but not for the overall goal for care. If respondents were sufficiently concerned about representing their preferences accurately for supplemental material their overall goal item, they may have been willing to • http://behavioralpolicy.org/supplemental-material engage in the mental effort to overcome the effect of • Methods & Analysis the default. Finally, it is possible that the carryover from the defaults of stage 1 to the (default-free) responses 19 If so, in stage 2 reflected a desire for consistency. then carryover effects would be weaker in real-world contexts involving important decisions. If the practice References of informing people that they were being defaulted Madrian, B. C., & Shea, D. F. (2001). The power of suggestion: 1. became widespread, moreover, it is unlikely that either Inertia in 401(k) participation and savings behavior. Quarterly of these default-weakening features would be common. 1149–1187. Journal of Economics, 116, Johnson, E. J., & Goldstein, D. G. (2003, November 21). Do 2. That is because defaults would not be chosen at random defaults save lives? Science, 302, 1338–1339. and advance directives would be filled out only once, Johnson, E. J., Hershey, J., Meszaros, J., & Kunreuther, H. 3. (1993). Framing, probability distortions, and insurance decisions. with a disclosed default. 35–53. Journal of Risk & Uncertainty, 7, Despite our results, it would be premature to Acquisti, A., John, L., & Loewenstein, G. (2013). What is privacy 4. conclude that the impact of nudges will always persist 249–274. Journal of Legal Studies, 42, worth? Kressel, L. M., & Chapman, G. B. (2007). The default effect in 5. when people are aware of them. Our findings are based end-of-life medical treatment preferences. Medical Decision on hypothetical advance directives—an appropriate first Making, 27, 299–310. 6. Kressel, L. M., Chapman, G. B., & Leventhal, E. (2007). The step in research given both the ethical issues involved influence of default options on the expression of end-of-life and the potential repercussions for choices made treatment preferences in advance directives. Journal of General regarding preferences for medical care at the end of life. 1007–1010. Internal Medicine, 22, 7. Halpern, S. D., Loewenstein, G., Volpp, K. G., Cooney, E., Vranas, Before embracing the general conclusion that warnings K., Quill, C. M., . . . Bryce, C. (2013). Default options in advance do not eliminate the impact of defaults, further research directives influence how patients set goals for end-of-life care. Health Affairs, 32, 408–417. should examine different types of alerts across different 8. Thaler, R. H., & Sunstein, C. R. (2008). Nudge: Improving settings. Given how weakly defaults affected overall . New Haven, CT: decisions about health, wealth, and happiness goals for care in this study, it would especially be fruitful Yale University Press. 9. House of Lords, Science and Technology Select Committee. - to examine the impact of pre- or postinforming partic London, United Behaviour change (2011). (Second report). ipants in areas in which defaults are observed to have Kingdom: Author. Bovens, L. (2008). The ethics of nudge. In T. Grüne-Yanoff & 10. robust impact in the absence of transparency. Those Preference change: Approaches from S. O. Hansson (Eds.), areas include decisionmaking regarding retirement (pp. 207–220). Berlin, philosophy, economics and psychology savings and organ donation. Germany: Springer. Wortman, C. B., & Brehm, J. W. (1975). Responses to 11. Most generally, our findings suggest that the effec - uncontrollable outcomes: An integration of reactance theory tiveness of nudges may not depend on deceiving those Advances in Experimental and the learned helplessness model. Social Psychology, 8, 277–336. who are being nudged. This is good news, because poli- Haisley, E., Volpp, K., Pellathy, T., & Loewenstein, G. (2012). 12. cymakers can satisfy the call for transparency advocated The impact of alternative incentive schemes on completion of 9 with little diminution in the in the House of Lords report health risk assessments. American Journal of Health Promotion, 26, 184–188. impact of positive interventions. This could help ease Halpern, S. D., Ubel, P. A., & Asch, D. A. (2007). Harnessing the 13. concerns that behavioral interventions are manipulative power of default options to improve health care. New England Journal of Medicine, 357, 1340–1344. or involve trickery. 41 a publication of the behavioral science & policy association

8 Johnson, E. J., & Goldstein, D. (2004). Default donation Writing Group for the SUPPORT Investigators. (1995, 17. 14. November 22). A controlled trial to improve care for seriously ill decisions. Transplantation, 78, 1713–1716. McKenzie, C. R., Liersch, M. J., & Finkelstein, S. K. (2006). 15. hospitalized patients: The Study to Understand Prognoses and Psychological Recommendations implicit in policy defaults. Preferences for Outcomes and Risks of Treatments (SUPPORT). 414–420. 1591–1598. Science, 17, JAMA, 274, Davidai, S., Gilovich, T., & Ross, L. D. (2012). The meaning Fagerlin, A., & Schneider, C. E. (2004). Enough: The failure of the 16. 18. PNAS: of default options for potential organ donors. (2), 30–42. living will. Hastings Center Report, 34 Proceedings of the National Academy of Sciences, USA, 109, Falk, A., & Zimmerman, F. (2013). A taste for consistency and 19. 15201–15205. 59, 181–193. survey response behavior. CESifo Economic Studies, behavioral science & policy | spring 2015 42

Related documents