More Money, More Problems? Can High Pay Be Coercive and Repugnant? Plain text version generated from the source manuscript and clean appendix LaTeX source. Main paper source: /Users/sandroambuehl/Dropbox/Repugnance_Sandro_Muriel_Al/submission/Ambuehl, Niederle, Roth MANUSCRIPT.docx Online appendix source: /Users/sandroambuehl/Dropbox/Repugnance_Sandro_Muriel_Al/writeup/AER P&P/Repugnance AER Appendix V6.tex Online appendix PDF matched by source: /Users/sandroambuehl/Dropbox/Repugnance_Sandro_Muriel_Al/submission/Ambuehl, Niederle, Roth; ONLINE APPENDIX.pdf ===== MAIN PAPER ===== More Money, More Problems? Can High Pay be Coercive and Repugnant? By Sandro Ambuehl, Muriel Niederle, And Alvin E. Roth* * All authors are at Stanford University, 597 Serra Mall, Stanford, CA 94305. (e-mails: sambuehl@stanford.edu, niederle@stanford.edu, alroth@stanford.edu.) We thank Debra Satz and participants at the Stanford Behavioral and Experimental Lunch and the ASSA annual meeting 2015 for helpful comments and discussions. Sometimes two people would voluntarily agree to transact a good or service for compensation, but an unaffected third party would prefer to prevent this transaction. Paid kidney donation, prostitution, and paid participation in medical trials are examples. We use a vignette study to explore how respondents' assessments of such repugnant transactions change as we alter the seller's compensation. We then sketch a model of how people judge the ethics of such transactions. We focus on transactions involving two parties, for which there are no material negative externalities. By considering a single transaction we abstract from general equilibrium effects (cf. Kaushik Basu 2007). We focus on the seller of the good and not the buyer. I. Survey Our survey concerns paid participation in medical experiments. The extant regulatory literature cautions against substantial monetary compensation for participation, particularly for the socioeconomically disadvantaged. According to the National Bioethics Advisory Commission (2001), "benefits threaten... the voluntary nature of the choice,...raise...the danger that the potential participant’s distributional disadvantage could be exploited [and]...lead some prospective participants to enroll...when it might be against their better judgment and when otherwise they would not do so." The medical ethics guidelines of jurisdictions as diverse as the European Union, India, and Kenya contain such language. For economists, these arguments are hard to understand. According to revealed preference theory, enlarging an agent's choice set can only make him better off. Hence preventing transactions that have no negative material externalities cannot improve welfare. We wish to understand whether there is a widespread perception in the population that high payments for participation in clinical trials are indeed ethically inappropriate, and if so, why? We presented 1445 subjects on Amazon Mechanical Turk with a fictitious medical trial that compensates participants with $50, $1,000, or $10,000. We described it as a test for side-effects of a vaccine that requires a total of 40 hours of a participant's time, and characterized it as low but non-zero risk. Each respondent was randomly displayed one of the three payment amounts and answered several questions, including how they would decide as a member of the IRB responsible for approving the experiment, before answering the same questions for each of the remaining amounts. The last page of the survey directly juxtaposed the three payment amounts, as well as the policies of paying $0 and covering participants' opportunity costs. Respondents rated the ethical appropriateness of each of these. Since respondents might not independently think of pertinent ethical aspects, we added an experimental condition for 55% of our respondents presenting the following argument, inspired by Satz (2010), and framed as a discussion between the designers of the study: "Does...payment...draw people into the study...who do not entirely understand what they are getting into?...[Is the] decision to participate …truly voluntary when a substantial payment is offered?" A. Results We use respondents' ratings of how ethical is each payment to categorize them as one of three types. "Economists" rate a payment of $10,000 strictly more ethical than a payment of $1,000. "Ethicists" make the opposite assessment. The remaining subjects rate both equally. When (no) arguments were provided, (57% and 14%) 34% and 27% of respondents were "economists" and "ethicists", respectively. We now show that “ethicists” on the IRB would reject the experiment more when the highest payment was offered, and feel that high payments damage decision making. Figure 1 plots respondents' reactions to varying payment amounts separately for "economists" and "ethicists". We use only the first stage of the survey, and hence each respondent appears only for one payment amount. We pool data over the two treatments. Figure 1. Responses of "Economists" and "Ethicists" Both types believe higher payments are more effective incentives (p < 0.1), and agree on the sign of the comparative statics of raising payment from $50 to $1,000 on all questions. They differ in assessing the effects of raising incentives to $10,000. "Ethicists" think very high incentives make participants who enroll more likely to subsequently regret their decision (p < 0.01), consistent with the hypothesis that very high incentives may lead to worse decision making. They think more prospective participants would be better off if they had never seen the offer to participate in this case (p < 0.01). They also consider the decision to accept less voluntary (p = 0.01) when very high incentives are offered. The opposite comparative statics apply to "economists" (p < 0.05 in all cases). Consistent with this, "ethicists" ("economists") state that as a member of the IRB they would approve the study less (more) if the incentive is $10,000 rather than $1,000 (p < 0.01 for both types). Respondents' own characteristics are predictive of their type. In a joint regression, higher income, education, and age all increase the likelihood a respondent is an "ethicist" (p < 0.1, 0.05, and 0.05, respectively). The 56% of respondents who have "thought about participating in a medical research study as a means to earn money" are more likely "ethicists" (p < 0.05). II. Sketch of a Model A prospective seller is offered units of money in exchange for an amount of a good such as health, and can freely decide to accept or reject. An observer who is richer than any seller judges the ethicality of the offer by judging the expected welfare of a participant. Agent 's endowment is and his utility from h units of health and units of money is with and . There is heterogeneity in both and . The distribution of sellers who participate is the population distribution conditional on being willing to participate. While the observer is able to correctly predict the seller's behavior, he does not assume that choice is utility maximizing. Instead, he partially takes the seller's perspective, and asks how he would feel if he had to live with the seller's choice. Hence the observer is (partially) paternalistic in his assessment of the welfare of the seller. Formally, he judges welfare from the point of view of someone with monetary endowment , where . Assessed welfare is. The ethicality of the transaction is . The transaction is considered repugnant if . Suppose the seller is offered a transaction he is just willing to accept. As the observer assesses welfare from the point of view of someone with a lower marginal utility of money, he thinks that the seller loses from accepting this transaction. Hence, he judges offering this transaction to the seller as repugnant. This shows that money is considered repugnant to the extent that it incentivizes the transaction, not per se. Indeed, paying the seller more for providing the same amount is judged as more ethical. Both results are in line with the ethics literature cited above. A. Comparative statics We first study the effects of incentivizing a given seller, and then consider selection. Incentivizing the sale of a larger amount is judged as less ethical. The observer thinks that the seller loses from a transaction that makes him indifferent. The larger the payment needed to make him indifferent, the more he loses. This comparative static is stronger for richer observers. Incentivizing poorer sellers is judged as less ethical (if v has non-increasing absolute risk-aversion) since the observer disagrees with poorer sellers more about the marginal value of money. This explains why medical ethics guidelines are particularly concerned about incentivizing socioeconomically disadvantaged populations. And is decreasing in the observer's endowment because richer observers take the seller's perspective to a lesser extent when assessing welfare. Regarding market design, the model implies that in-kind incentives will be judged as most ethical. Once there is no tradeoff between good h and money, there is no scope for disagreement about the appropriate rate of substitution. The model also explains why 73% of our survey respondents consider remunerating subjects with the opportunity costs as worse than (at least one of) paying $1,000 or $10,000. Starting from two prospective sellers with different endowments who are given the respective least amount of money that just makes them participate, can be increased to without affecting any participation decisions. Since high payments are viewed as unethical only to the extent they serve as incentives, this raises . When allowing for selection, increasing incentives for selling a given amount has three effects. (i) Sellers who would have participated anyway now receive a larger consumer surplus. (ii) Richer sellers, who would not have participated before, now do. (iii) Sellers with higher utility weight on good h now participate. The first two effects increase e. The third is akin to incentivizing a larger transaction and decreases e. The total effect depends on the population distribution of and , and can take the hump shape observed for "ethicist" survey respondents. A promising direction for further empirical research may be to study the attitudes of subjects drawn from the same population as potential participants in a study. An IRB might wish to know if people with the same income as the participants of concern feel that high incentives are repugnant. REFEREnces Basu, Kaushik. (2007). Coercion, contract and the limits of the market (Vol. 29). Soc Choice Welfare . National Bioethics Advisory Commission. (2001). Ethical and Policy Issues in Research Involving Human Participants. Bethesda, MD. Roth, Alvin. E. (2007). Repugnance as a Constraint on Markets (Vol. 21). Journal of Economic Perspectives. Sandel, Michael J. (2012). What Money Can't Buy. The Moral Limits of Markets. New York. Satz, Debra. (2010). Why Some Things Should Not Be For Sale. The Moral Limits of Markets. Oxford University Press. ===== ONLINE APPENDIX (LaTeX SOURCE) ===== \documentclass[smallheadings]{article} \usepackage{graphicx} \usepackage{mathrsfs} \usepackage{mdwlist} \usepackage{graphicx} \usepackage{amssymb} \usepackage{amsmath} \usepackage{eucal} \usepackage{amsthm} \usepackage{epstopdf} \usepackage{dsfont} \usepackage{bm} \usepackage{enumerate} \usepackage[usenames]{color} \definecolor{blue}{rgb}{0,0,1} \usepackage{natbib} \usepackage{multirow} \usepackage[margin=1.2in]{geometry} \usepackage{pdfpages} \usepackage{hyperref} \usepackage{pdflscape} \usepackage[nottoc,numbib]{tocbibind} \clubpenalty = 100000 \widowpenalty = 100000 \DeclareMathOperator{\plim}{plim} \renewcommand{\theenumi}{(\roman{enumi})} \renewcommand{\labelenumi}{\theenumi} \newcommand{\bmath}{\begin{eqnarray}} \newcommand{\emath}{\end{eqnarray}} \newcommand{\bmathnn}{\begin{eqnarray*}} \newcommand{\emathnn}{\end{eqnarray*}} \newcommand{\bi}{\begin{itemize}} \newcommand{\ei}{\end{itemize}} \newcommand{\I}{\mathcal{I}} \newcommand{\var}{\mathrm{var}} \newcommand{\logit}{\mathrm{logit}} \newcommand{\abs}{\mathrm{abs}} \newenvironment{lmss}{\fontfamily{lmss}\selectfont}{\par} \setcounter{MaxMatrixCols}{10} \theoremstyle{plain} \newtheorem{proposition}{Proposition} \newtheorem*{example}{Example} \setcounter{section}{0} \newtheorem{theorem}[proposition]{Theorem} \newtheorem{lemma}{Lemma} \newtheorem{corollary}[proposition]{Corollary} \newtheorem{result}{Result} \newtheorem{hypothesis}{Hypothesis} \theoremstyle{definition} \newtheorem*{definition}{Definition} \newtheorem{assumption}{Assumption} \newtheorem*{remark}{Remark} \theoremstyle{plain} \DeclareMathOperator*{\argmax}{argmax} \DeclareMathOperator*{\argmin}{argmin} \DeclareMathOperator{\core}{Core} \DeclareMathOperator{\cov}{cov} \usepackage{setspace} \setstretch{1.4} \begin{document} \begin{center} \huge{ONLINE APPENDIX}\\ \vspace{0.5cm} \large{More Money, More Problems? Can High Pay Be Coercive And Repugnant?}\\ \large{Sandro Ambuehl, Muriel Niederle, Alvin E. Roth} \end{center} \setstretch{1} \tableofcontents \setstretch{1.4} \section{Vignette} \begin{lmss} \textit{Suppose that you are a member of an ethics committee, and you will have to decide whether or not to approve of the following study. Pay close attention. All the following questions will be based on this text.} The E.M.C.A. Medical Research Institute has developed a new vaccine to prevent infection with the Ebola virus. In rats and chimps the vaccine successfully prevents infection with the virus and causes no measurable side effects. The institute now seeks to enlist 100 female participants to investigate whether the vaccine causes side effects in women. This is important to know, as it will determine whether the vaccine can be given to female healthcare workers in regions affected by the disease. Each of the 100 participants will be injected with the vaccine and then monitored in weakly intervals for two months. The total time required to participate if no side effects occur is about 40 hours. Participants will not be exposed to the virus; the study only tests for side effects of the vaccine. Since no side effects occurred in the animal studies, the Institute's experts consider it unlikely that they will occur in humans. However, nobody knows for sure. This is why the experiment needs to be run. In case that unexpected side effects occur, they might range from very mild such as a day of nausea to very severe, such as persistent migraines. Side effects will be treated free of charge, if treating them is medically possible. An affected woman will not, however, receive treatment for any unrelated medical problems, and she will not receive any other compensation for suffering these side effects. The only compensation to any participant is the money paid to her when she agrees to take part in the study, before she is injected the vaccine. Study participation invitations will be posted throughout the mid-western city in which the Institute is located. Invitations will be put up in both rich and poor neighborhoods. The Institute will compensate each woman who participates with [\$50 / \$1,000 / \$10,000] for the risk the participants take, and the total of 40 hours required to participate in the study. [Only in ``arguments'' condition: Before posting the participation invitations, the researchers of the Institute discuss the conditions of the study in a meeting. Some researchers are concerned whether this way of inviting participants to the study is appropriate. While serious side effects seem unlikely, nobody knows this for certain. (Otherwise, there would be no need to run a study). They wonder: Does offering payment for the study inevitably draw people into the study who do not entirely understand what they are getting into? Is it at all possible to make participants adequately aware of the risks they take by participating? Can participants at all imagine what their lives would be like if an unanticipated side effect occurred, and how they would feel about that? Hence, the researchers wonder whether a woman's decision to participate is truly voluntary when a substantial payment is offered for participation; in particular when that person has a very low income. Might the Institute be exploiting those people's economic disadvantages? Do they participate in the study against their own better judgment?]\\ [new page]\\ \textit{Suppose you are a member of the ethics committee that has to approve the Institute's study with [payment1]. How would you decide?} [Answer choices: 7-point Likert scale with extremes labelled ``definitely reject'' and ``definitely approve''] \textit{How much do you personally approve of the Institute's proposal to enlist and compensate study participants from both rich and poor neighborhoods in this way?} [Answer choices: 7-point Likert scale with extremes labelled ``strongly disapprove'' and ``approve without reservation'']\\ [next page]\\ A.S. is a woman who lives in a poor part of the city. For the past 20 years she has worked in various minimum-wage jobs. She currently earns \$1,500 per month, which is barely enough for her to get by. A.S. encounters one of the study participation invitations that the Institute has posted on bulletin boards in her neighborhood. A.S. considers signing up for this study. She is on the fence about whether or not to do so. She is afraid of possible unexpected serious side effects of the vaccine. But then again, she would be paid [as much as she earns in her job in a day / almost as much as she earns in her usual job in an entire month / more than six times as much in her usual job in an entire month]. \\ [new page]\\ \textit{Suppose that 10 women similar to A.S. see the institute's study participation invitation. How many of the 10 would be better off if the institute had never posted the study participation invitation?} [Answer choices: 1 - 10] \textit{How many of the 10, do you think, will eventually participate in the study in exchange for [payment1]?} [Answer choices: 1 - 10] \textit{If A.S. decides to participate in the study for [payment1], how would you describe her decision?} [Answer choices: 7-point Likert scale with extremes labelled ``She was coerced'' and ``Her decision was entirely voluntary''] \textit{If A.S. decides to participate in the study, how likely is it that she will later regret her decision? } [Answer choices: 7-point Likert scale with extremes labelled ``Extremely unlikely'' and ``Extremely likely''] \textit{If A.S. decides NOT to participate in the study, how likely will she later regret her decision? } [Answer choices: 7-point Likert scale with extremes labelled ``Extremely unlikely'' and ``Extremely likely'']\\ [Repetition of the above questions, introduced with ``Researchers at the institute discuss offering \emph{payment} for participation instead.'' Prospective subject described as ``J.K. is a woman similar to A.S. She works in a minimum-wage job and encounters the study participation invitation.'']\\ For each of the following ways of compensating study participants, please indicate how ethically appropriate you think it is. Recall that the study test for side effects of a vaccine, and although nobody expects such side effects to occur, if this were known, there would be no need to run a study. Recall that there is no special compensation if side effects occur. [Only in ``arguments'' condition: Please consider professional ethicists' argumentation: 1. By advertising to pay money to participants, the Institute might entice women to participate in the study who don't fully understand what they are getting into. 2. By advertising to pay money to participants, the Institute might lead people to participate in the study who would not otherwise choose to do so] \bi \item Do not pay money for participation \item Pay \$50 for participation \item Pay \$1,000 for participation \item Pay \$10,000 for participation \item Pay everyone the amount of money for participation that he would earn at his job in 40 hours. [Only in ``arguments'' condition: This means that an employee at McDonalds will be paid about \$300 for participation whereas an attorney will be paid about \$3,000.] \ei [Answer choices for each of the above compensation schemes: 7-point Likert scale with extremes labelled ``completely unethical'' and ``completely ethical'']\\ [new page]\\ \textit{For your preferred way of compensating participants, please briefly explain why you think it is the most ethical way to do it.}\\ [new page] \\ \textit{Have you ever participated in a medical research study?} \textit{Have you every thought about participating in a medical research study as a means to earn money?} \textit{Would you participate in the experiment about the Ebola vaccine described in this study for a payment of \$50?} \textit{Would you participate in the experiment about the Ebola vaccine described in this study for a payment of \$1,000?} \textit{Would you participate in the experiment about the Ebola vaccine described in this study for a payment of \$10,000?} [Answer choices: ``Yes'', ``No'', ``I do not know''. In the first two of the above five question, the choice ``prefer not to answer'' was also available.] \end{lmss} \section{Data Analysis} \paragraph{Survey} We fielded the survey in the first two weeks of December 2014 on weekdays on Amazon Mechanical Turk with a total of 1570 respondents. Each respondent was paid \$1.50 lump sum, and additionally received an incentive payment of \$0.25 for answering at least 3 out of 5 attention check questions correctly. We retain the 1445 respondents who answered all five questions correctly for analysis (92\%). The average time to completion for these subjects is slightly over 12 minutes (s.e. 5 minutes and 25 seconds). \paragraph{Stable types} From each respondent we have three measures of approval for each payment amount: (i) As a member of the IRB, to what extent would they approve of the study with the given payment amount, (ii) How much do they personally approve of the study with the given payment amount, and (iii) How ethical do they consider compensation with a given payment amount. While the first two of those questions were asked in immediate succession (but separated temporally for different payment amounts), the last was asked in the direct juxtaposition of the compensation schemes at the end of the survey. We define $d_{i}^k$ as the difference in respondent $i$'s approval of the payment of \$10,000 vs. \$1,000 according to question $k$. The Cronbach alpha coefficient of $(d^1, d^2, d^3)$ is 0.81 when including all respondents, and 0.77 and 0.84 on the subsamples of those who were not, and were given the arguments, respectively. These compare favorably to the standard benchmark of 0.8 (\citet{Kline}) and thus indicate high interitem correlation. As an alternative measure, we regress $d^3$ on the first principal component of $d^1$ and $d^2$. Slope coefficients are 0.83 (0.03), 0.72 (0.04), and 0.86 (0.04) on the subsamples of all respondents, those who have not, and those who have been presented with the arguments, respectively (s.e. in parentheses). The associated $R^2$ coefficients are 0.37, 0.32, and 0.39, respectively.\footnote{Factor analysis reveals a single factor with an eigenvalue that exceeds one. This suggests that the three variables measure the same underlying concept.} Both of these analyses suggest that respondents who consider an incentive of \$10,000 less ethical than one of \$1,000 do so consistently. \paragraph{Classification} As explained in the main text, we use $d^3$ to define a respondent's type. This is the most appropriate single variable for this purpose, since only for this variable were respondents asked to \emph{directly} compare the ethical appropriateness of a payment of \$10,000 compared to a payment of \$1,000. In the full sample, there are 21\% ``ethicist'' types, 34\% neutral types and 44\% ``economist'' types. The corresponding numbers in the subsamples of those who were not, and were provided with the arguments, are 14\%, 32\%, 57\%, and 27\%, 39\%, 34\%, respectively. \paragraph{Determinants of types} Respondents' assessment of the ethical appropriateness of payment is determined by their own characteristics. We first consider the hypothesis suggested by our model, and include (logarithmic) income, and the one obvious confound, education. We consider two specifications: (i) an OLS regression with $d^3$ as the dependent variable including all subjects, and (ii) a Probit regression with $sgn(d^3)$ as the dependent variable, including only ``economist'' and ``ethicist'' subjects. Table \ref{determinants} displays the results. Columns 1 and 4 of show that both income and education significantly increase the likelihood of being an ``ethicist'' type. We then test for the robustness of these factors by successively adding demographic characteristics (columns 2 and 5), as well as measures of political orientation and religion (columns 3 and 6). Age is significantly predictive of type, with older respondents being more likely an ``ethicist'' type. Surprisingly, conservatives are significantly more likely ``ethicist'' types than centrists or apolitical subjects (but not significantly more likely than liberals). Finally, the 56\% of respondents who claim to have previously thought about participating in a medical study as a means to earn money are significantly more likely ``ethicist'' types. (These respondents' mean annual household income is also smaller by a highly significant \$6089.) The probit estimates exclude respondents that are classified as neutral types. \paragraph{Beliefs} For each type we estimate how responses to various survey questions change when the payment offered is increased from \$1,000 to \$10,000. We use only the first stage of the survey for these regressions. Hence, in each regression, each respondent appears as at most one observation. Table \ref{compStaticsAcross} displays the results. They correspond to the graph and discussion in the main text. Neutral types are intermediate on each question. The sole question that is not discussed in the main text is the assessed likelihood that a prospective participant who rejects to participate will regret this decision. All types believe that this likelihood is higher for the \$10,000 payment than for the \$1,000 payment. These results are not qualitatively different when we perform the regressions separately for those respondents who were and were not provided with the ethical arguments (see below). When we perform the estimations using the within-subjects data our results change slightly. First, ``ethicists'' responses to how likely a prospective participant would be better off if he had never encountered the opportunity to participate in the trial is not significantly different across payment amounts. Second, even ``ethicists'' are less likely to think that a participant who enrolls in the trial will regret her decision when payment is \$10,000 rather than \$1,000, although to a much lesser extent than ``economists''. Third, even ``economists'' consider the decision to accept less voluntary when payment is \$10,000. Note that for each of these questions, the differences in the comparative statics between ``economists'' and ``ethicists'' remain unchanged relative to the across subjects data. \paragraph{Effects of providing the arguments} Columns 1 - 3 (4 - 6) of table \ref{determinantsArguments} replicate the respective columns of table \ref{determinants} for the subsample of respondents who were (were not) provided with the ethical arguments. For the former subsample, income and education remain significant predictors of type, but the coefficients on age and conservativeness drop by about half and lose significance. Additionally, the coefficient on having thought about participating in paid medical trials remains similar in magnitude, but loses significance. Finally, gender emerges as a weakly significant predictor of type, with males more likely to be ``ethicist'' types. In contrast, for the latter subsample, the effect of education intensifies, and the coefficients on (log) income drop to just over half their previous value, and lose statistical significance. The remaining predictors retain comparable magnitudes and significance levels as in the full sample, with the exception that the race dummies now have a significant effect. Finally, while providing the arguments changes the incidence of the three types, it does not substantially affect the beliefs of the types. Table \ref{compStaticsAcrossArguments} shows that all the signs in table \ref{compStaticsAcross} remain unchanged when they are estimated on the subsamples of respondents who have and have not been provided with the arguments, respectively, and the magnitudes remain comparable. \section{Model and Proofs} \paragraph{Setup} The model consists of a prospective seller $s$ and an observer $o$ who judges the ethicality of the transaction. Agent $i$'s utility is defined over bundles $(h,m)$ where $h$ is a good such as health, and $m$ is money. The utility function of agent $i\in\{s,o\}$ is \bmathnn U_i(h,m) = \alpha_i u(h) + v(m) \emathnn with $u \geq 0, u' > 0, v' > 0$, and $v'' < 0$. Agent $i$ has endowment $(h,m_i)$. The seller's preference parameter $\alpha_s$ and monetary endowment $m_s$ are jointly distributed according to probability measure $P$. The observer is richer than the seller, $m_o > m_s$. The seller is offered to exchange amount $\bar{h}$ of good $h$ in exchange for $\bar{m}$ units of money. He accepts the transaction iff $U_s(h - \bar{h}, m_s+\bar{m}) \geq U_s(h,m_s)$. We assume that the distribution of sellers that select into the transaction equals the population distribution conditional on being willing to accept. The observer judges the ethicality of offering the transaction to prospective sellers. While he is perfectly able to predict sellers' decisions to accept or reject the transaction, he judges seller $s$'s welfare from accepting the transaction by partially taking the seller's perspective. Specifically, for parameters $\rho\in [0,1]$ and $\gamma\in[0,1]$, we define $m^s_{\rho} = \rho m_s + (1 - \rho) m_o$, and $\alpha^{s}_{\gamma} = \gamma \alpha_s + (1 - \gamma) \alpha_o$. The observer judges seller $s$'s welfare from accepting the transaction as \bmathnn w(\bar{h},\bar{m}) = \alpha^{s}_{\gamma} u(h - \bar{h}) + v(m^s_{\rho} + \bar{m}) \emathnn The observer's partial perspective taking is motivated by paternalistic concerns. For instance, he might believe that the seller suffers from projection bias (\citet{Loewensteinetal.ProjectionBias}), and account for that presumed bias when judging welfare. The observer judges the ethicality of offering the transaction to the seller by \bmath\label{ethicality} {e}(\bar{h},\bar{m}) = E\big[\Delta w(\bar{h},\bar{m})\big| U_s(h - \bar{h}, m_s+\bar{m}) \geq U_s(h,m_s) \big] \emath where $\Delta w(\bar{h}, \bar{m}) = w(\bar{h}, \bar{m}) - w(0,0)$ is the assessed change in welfare from accepting the transaction. The transaction is judged as unethical or repugnant if ${e}(\bar{h},\bar{m}) < 0$. \paragraph{Implications} We first study the comparative statics of incentivizing a given seller. \begin{proposition}\label{Transactions}(Transactions)\ \begin{enumerate} \item Transactions that require a larger amount of money to incentivize a given seller to participate are judged as less ethical if $\alpha_o \geq \alpha_s$ or if $\alpha_o < \alpha_s$ and $\gamma$ is sufficiently close to 1: $\left. \frac{d{e}}{d\bar{m}} \right|_{U_s(h_s-\bar{h}, m_s+\bar{m}) = U_s(h_s,m_s) } <0$ \item The richer the observer, the stronger this comparative static. \item Paying more money for provision of the same amount of $\bar{h}$ is judged as more ethical:$\left.\frac{d{e}}{d\bar{m}} \right|_{d\bar{h} = 0} >0 $ \end{enumerate} \end{proposition} \begin{proof} \begin{enumerate}[(i)] \item We consider a given seller, so that ${e} = \Delta w(\bar{h},\bar{m})$. By the chain rule and implicit differentiation, we derive \bmathnn \frac{d{e}}{d\bar{m}} &=& - \alpha_{\gamma}^s u'(h_{\rho} - \bar{h}) \frac{d\bar{h}}{d\bar{m}}+ v'(m_{\rho}^s + \bar{m}) \\ \left. \frac{d\bar{h}}{d\bar{m}} \right|_{U_s(h_s-\bar{h}, m_s+\bar{m}) = U_s(h_s,m_s) } &=& \frac{v'(m_s + \bar{m})}{\alpha_s u'(h_s - \bar{h})} \emathnn Hence, by $h_s = h_o$, we obtain \bmathnn \left. \frac{d{e}}{d\bar{m}} \right|_{U_s(h_s-\bar{h}, m_s+\bar{m}) = U_s(h_s,m_s) } &=&v'(m_{\rho}^s+\bar{m}) - \frac{\alpha_{\gamma}^s}{\alpha_s} v'(m_s + \bar{m}) \\ \emathnn Because $v'' < 0$, and $m_o > m_s$ (and hence $m_{\rho}^s > m_s$), the above expression is negative if $\frac{\alpha_{\gamma}^s}{\alpha_s}$ is not much smaller than 1. This is the case if either $\alpha_o \geq \alpha_s$ or if $\alpha_o < \alpha_s$ and $\gamma$ is sufficiently close to 1. \item From (i) we derive \bmathnn \left. \frac{d}{dm_o} \frac{d{e}}{d\bar{m}} \right|_{U_s(h_s-\bar{h}, m_s+\bar{m}) = U_s(h_s,m_s) } &=& \frac{d}{dm_o} v'(m_{\rho}^s+\bar{m}) - v'(m_s + \bar{m}) \\ &=& v''(m_{\rho}^s+\bar{m}) (1 - \rho) \emathnn which is negative due to $v'' < 0$. \item This follows trivially since ${e}$ is monotonic in the seller's ex post endowment $m_s + \bar{m}$. \end{enumerate} \end{proof} \begin{proposition}\label{endowments} (Endowments and Preferences)\ \begin{enumerate} \item Richer observers are more likely to judge a transaction as unethical: $\frac{d{e}}{dm_o}<0$. \item If $v''/v'$ is non-increasing, then incentivizing a richer (poorer) seller to sell $\bar{h}$ is judged as more (less) ethical: $\left. \frac{d{e}}{dm_s} \right|_{U_s(h_s-\bar{h}, m_s+\bar{m}) = U_s(h_s,m_s) } \geq 0$. \item If $\gamma$ is sufficiently close to 1, then incentivizing a seller with higher $\alpha_s$ to sell $\bar{h}$ is judged as less ethical: $\left. \frac{d{e}}{d\alpha_s} \right|_{U_s(h_s-\bar{h}, m_s+\bar{m}) = U_s(h_s,m_s) } > 0$. \end{enumerate} \end{proposition} Intuitively, the comparative statics in part (iii) of the above proposition depends on $\gamma$ for the following reason. Regardless of $\gamma$, the observer recognizes that a seller with higher $\alpha_s$ requires a higher monetary compensation to make him indifferent between accepting and rejecting the transaction. If $\gamma$ is sufficiently close to 1, he acknowledges this as welfare relevant. Thus, incentivizing a seller with a higher parameter $\alpha_s$ to give up $\bar{h}$ is judged akin to incentivizing a given seller to exchange a larger amount of good $h$, and hence is judged as less ethical. By contrast, if $\gamma$ is close to 0, the observer mainly considers the fact that a seller with higher $\alpha_s$ is given more money in exchange for $\bar{h}$, but hardly acknowledges the fact that this is to compensate for a larger loss in utility, and hence considers the transaction more ethical. \begin{proof} \begin{enumerate}[(i)] \item \bmathnn \frac{d{e}}{dm_o} = E\Big[ (1 - \rho) \left[ v'(m_{\rho}^s + \bar{m}) - v'(m_{\rho}^s) \right] \Big| U_s(h - \bar{h}, m+\bar{m}) \geq U_s(h,m) \Big] \emathnn By $v''<0$, this is negative. \item We show that $\left. \frac{d{e}}{dm_s} \right|_{U_s(h_s-\bar{h}, m_s+\bar{m}) = U_s(h_s,m_s) } > 0$. \bmathnn \frac{d{e}}{d\bar{m}} &=& \rho \big(v'(m_{\rho}^s + \bar{m}) - v'(m_{\rho}^s)\big) \frac{dm_s}{d\bar{m}} + v'(m_{\rho}^s + \bar{m}) \\ \left. \frac{dm_s}{d\bar{m}} \right|_{U_s(h_s-\bar{h}, m_s+\bar{m}) = U_s(h_s,m_s) } &=& - \frac{v'(m_s + \bar{m})}{v'(m_s + \bar{m}) - v'(m_s)} \emathnn Consequently, \bmathnn \left. \frac{d{e}}{d\bar{m}} \right|_{U_s(h_s-\bar{h}, m_s+\bar{m}) = U_s(h_s,m_s) } &=& - \rho \frac{v'(m_{\rho}^s + \bar{m}) - v'(m_{\rho}^s)}{v'(m_s + \bar{m}) - v'(m_s)} v'(m_s + \bar{m}) + v'(m_{\rho}^s + \bar{m}) \emathnn This is weakly negative if and only if \bmathnn \rho \leq \frac{1 - \frac{ v'(m_{\rho}^s)}{v'(m_{\rho}^s + \bar{m})}}{1 - \frac{v'(m_s)}{v'(m_s+ \bar{m})}} \emathnn The RHS of the above expression weakly exceeds 1 iff \bmathnn \frac{ v'(m_{\rho}^s)}{v'(m_{\rho}^s + \bar{m})} \leq \frac{v'(m_s)}{v'(m_s+ \bar{m})} \emathnn Thus, because $m_{\rho}^s> m_s$, it is sufficient to show that $\frac{v'(h)}{v'(h+\bar{m})}$ is weakly decreasing in $h$. We have \bmathnn \frac{d}{dh}\left[ \frac{v'(h)}{v'(h+\bar{m})} \right] = \frac{v''(h)v'(h+\bar{m}) - v''(h+\bar{m})v'(h)}{v'(h+\bar{m})^2} \emathnn The RHS of this expression is weakly smaller than 0 iff $\frac{v''(h)}{v'(h)} \geq \frac{v''(h+\bar{m})}{v'(h+\bar{m})}$, i.e. iff the coefficient of absolute risk aversion $-\frac{v''}{v'}$ is non-increasing. \item By differentiating ${e}$, we get \bmathnn \frac{d{e}}{d\alpha_s} = \gamma \big( u(h - \bar{h}) - u(h) \big) + v'(m^s_{\rho} + \bar{m}) \frac{d\bar{m}}{d\alpha_s} \emathnn By implicit differentiation of the indifference condition, we obtain \bmathnn \frac{d\bar{m}}{d\alpha_s} = - \frac{u(h - \bar{h}) - u(h)}{v'(m_s + m)} \emathnn Inserting this into the first expression, we get \bmathnn \frac{d{e}}{d\alpha_s} = \big( u(h - \bar{h}) - u(h) \big) \cdot \left[ \gamma - \frac{v'(m^s_{\rho} + \bar{m})}{v'(m_s + \bar{m})} \right] \emathnn By $u'>0$, $u(h - \bar{h}) - u(h) < 0$. By $m_o > m_s$ and $v''<0$, $\frac{v'(m^s_{\rho} + \bar{m})}{v'(m_s + \bar{m})} < 1$. Consequently, if $\gamma$ is sufficiently close to 1, $\frac{d{e}}{d\alpha_s} < 0$, as was to be shown. \end{enumerate} \end{proof} So far we studied the comparative statics of ${e}$ regarding incentivizing a given seller. In the next proposition we explicitly consider selection of heterogenous sellers into the transaction. \begin{proposition} (Selection)\ If sellers are heterogenous regarding both $\alpha_s$ and $m_s$, and $\gamma$ is sufficiently close to 1, then ${e}$ may be non-monotonic in payment. \end{proposition} \begin{proof} The proof is by example. Fix a loss of health from accepting the transaction of $\bar{h} = 1$, and suppose that $u(h) = h$ and $v(m) = \sqrt{m}$. A seller's monetary endowment is either 0 or 1, all agents' endowment with good $h$ is normalized to 0, and a seller's preference parameter $\alpha_s$ is either 1 or 2. Payment amounts under consideration are 1, 3, and 4. The utilities from accepting the transaction with given payment amounts are listed in the following table.\\ \begin{center} \begin{tabular}{llcccc} &\multicolumn{5}{c}{$U_s(h - \bar{h}, m_s + \bar{m}) = -\alpha_s + \sqrt{m_s + \bar{m}}$}\\ \hline \multirow{ 2}{*}{Type} &Money $m_s$ & 0 & 1 & 0 & 1 \\ &Preference $\alpha_s$ & 1 & 1 & 2 & 2\\ \hline \multirow{ 3}{*}{Transaction} & $\bar{h} = 1$, $\bar{m} = 1$ & 0 & $-1 + \sqrt{2}$ & -1 &$ -2 + \sqrt{2}$ \\ &$\bar{h} = 1$, $\bar{m} = 3$ & $-1 + \sqrt{3}$ & 1 &$ -2 + \sqrt{3}$ & 0 \\ &$\bar{h} = 1$, $\bar{m} = 4$ & 1 & $-1 + \sqrt{5}$ & 0 & $-2 + \sqrt{5}$ \\ \hline \end{tabular} \end{center}\ A seller with monetary endowment 0 (1) will accept the transaction if the utility from accepting it weakly exceeds 0 (1). Consequently, for payment $\bar{m} = 1$ only types $(m_s, \alpha_s) = (0,1)$ (poor, low marginal utility of health) weakly accept the transaction. For $\bar{m} = 3$, types $(m_s, \alpha_s) \in \{(0,1), (1,1)\}$ (poor and rich, low marginal utility of health) accept the transaction. Finally, for $\bar{m} = 4$, types $(m_s, \alpha_s) \in \{(0,1), (1,1), (0,2)\}$ (poor and rich with low marginal utility of health, and poor with high marginal utility of health) accept the transaction. By proposition \ref{endowments}, part (ii), making a richer seller indifferent between accepting and rejecting is judged as more ethical, all else equal; and by proposition \ref{Transactions}, part (iii), paying more to a type who would have accepted the transaction anyway is judged as more ethical. Consequently, raising payment from $\bar{m} = 1$ to $\bar{m} = 3$ is judged as more ethical, irrespective of the population distribution. By proposition \ref{endowments}, part (iii), if $\gamma$ is sufficiently large, then offering type $(0,2)$ the payment $\bar{m} = 3$ (which makes him just indifferent between accepting and rejecting the transaction) is judged as less ethical than offering type $(0,1)$ the payment $\bar{m} = 1$ (which makes that type indifferent). Because the ethicality of offering the transaction to a seller is the expected welfare (as judged by the observer) of those sellers who are weakly willing to accept, it is apparent that offering the transaction at $\bar{m} = 4$ is judged as less ethical than offering the transaction at $\bar{m} = 3$ if $P\big((m_s, \alpha_s) = (0,2)\big)$ is sufficiently large. \end{proof} \begin{thebibliography}{} \bibitem[\protect\astroncite{Kline}{1999}]{Kline} Kline, P. (1999). \newblock {\em Handbook of Psychological Testing}. \newblock Routledge, London and New York, 2 edition. \bibitem[\protect\astroncite{Loewenstein et~al.}{2003}]{Loewensteinetal.ProjectionBias} Loewenstein, G., O'Donogue, T., and Rabin, M. (2003). \newblock Projection bias in predicting future utility. \newblock {\em Quarterly Journal of Economics}, 118(4):1209--1248. \end{thebibliography} \begin{table} \setstretch{1.3} \begin{center} \begin{tabular}{lcccccc} & (1) & (2) & (3) & (4) & (5) & (6)\\ \hline VARIABLES & $d^3$ & $d^3$ & $d^3$ & sgn($d^3$) & sgn($d^3$) & sgn($d^3$) \\ & & & \\ ln(Income) & -0.143** & -0.128** & -0.124** & -0.146** & -0.138** & -0.127** \\ & (0.0594) & (0.0592) & (0.0606) & (0.0574) & (0.0582) & (0.0600) \\ Education & -0.136*** & -0.133*** & -0.126*** & -0.0875** & -0.0913** & -0.0984** \\ & (0.0406) & (0.0407) & (0.0416) & (0.0390) & (0.0395) & (0.0407) \\ Male & & 0.103 & 0.108 & & 0.121 & 0.136 \\ & & (0.0940) & (0.0960) & & (0.0888) & (0.0920) \\ Age & & -0.0123*** & -0.0139*** & & -0.00948** & -0.0105** \\ & & (0.00414) & (0.00430) & & (0.00390) & (0.00409) \\ African American & & -0.0640 & -0.205 & & 0.175 & 0.0256 \\ & & (0.357) & (0.363) & & (0.348) & (0.368) \\ Asian & & -0.369 & -0.512 & & -0.0483 & -0.219 \\ & & (0.371) & (0.378) & & (0.355) & (0.374) \\ Caucasian & & -0.511 & -0.560* & & -0.327 & -0.406 \\ & & (0.318) & (0.324) & & (0.305) & (0.326) \\ Hispanic & & -0.568 & -0.661* & & -0.262 & -0.435 \\ & & (0.373) & (0.381) & & (0.354) & (0.375) \\ No Religion & & & -0.115 & & & -0.0602 \\ & & & (0.104) & & & (0.0996) \\ Conservative & & & -0.296** & & & -0.295** \\ & & & (0.145) & & & (0.138) \\ Liberal & & & -0.157 & & & -0.149 \\ & & & (0.114) & & & (0.111) \\ Thought about participating & & & -0.217** & & & -0.203** \\ & & & (0.0961) & & & (0.0922) \\ Constant & 2.588*** & 3.049*** & 3.363*** & 2.456*** & 2.767*** & 3.076*** \\ & (0.610) & (0.684) & (0.705) & (0.588) & (0.662) & (0.696) \\ & & & & & & \\ Method & OLS & OLS & OLS & Probit & Probit & Probit \\ & & & \\ Observations & 1,415 & 1,415 & 1,375 & 915 & 915 & 882 \\ $R^2$ & 0.016 & 0.030 & 0.038 &-&-&- \\\hline \end{tabular} \end{center} \caption{\label{determinants} Excluded categories are \emph{female}, \emph{other race}, \emph{has a religion}, \emph{neither conservative nor liberal}. Respondents who did not know or preferred not to answer whether they had previously thought about participating in medical trials are not included in the regressions in columns 3 and 6.} \end{table} \begin{table} \setstretch{1.3} \begin{center} \begin{tabular}{lccc} VARIABLE & ``Ethicists'' & Others & ``Economists'' \\ \hline Voluntariness & -0.5744889 ** & -0.4865654 *** & 0.3735843 ** \\ & ( 0.2332091 ) & ( 0.1772251 ) & ( 0.1517972 ) \\ Better off without & 1.335146 *** & 0.0536728 & -0.6496786 ** \\ & ( 0.4253126 ) & ( 0.3549511 ) & ( 0.2961822 ) \\ Regret accepting & 0.7368421 *** & 0.010728 & -0.3016748 ** \\ & ( 0.1679424 ) & ( 0.1440166 ) & ( 0.1244856 ) \\ Regret rejecting & 0.500435 ** & 0.6359372 *** & 0.544427 *** \\ & ( 0.2279612 ) & ( 0.163064 ) & ( 0.1550434 ) \\ Likelihood of accepting & 0.7070466 ** & 0.3680325 & 0.4403997 * \\ & ( 0.3330673 ) & ( 0.2512439 ) & ( 0.2306782 ) \\ Personal approval & -0.6907351 *** & -0.298645 * & 0.4647339 *** \\ & ( 0.2178791 ) & ( 0.1706136 ) & ( 0.140393 ) \\ IRB Approval & -1.099174 *** & -0.1593124 & 0.8052604 *** \\ & ( 0.2220483 ) & ( 0.1822648 ) & ( 0.1457809 ) \\ Observations & 197 & 350 & 431 \\ \hline \end{tabular} \end{center} \caption{\label{compStaticsAcross} Effects of increasing payment from \$1,000 to \$10,000. Displayed are the coefficients of regressions of the indicated dependent variable on a dummy that equals 1 if the payment is \$10,000. The variables \emph{Better off without} and \emph{Likelihood of accepting} are measured on a scale of 0 to 10, the remaining variables are measured on a scale of 1 to 7. Data are pooled over respondents who have and have not been presented with the ethical arguments. For each subject, we include only the responses to the first payment amount the subject was presented with (across subjects data).} \end{table} \begin{table} \setstretch{1.3} \begin{center} \begin{tabular}{lccc} VARIABLE & ``Ethicists'' & Others & ``Economists'' \\ \hline Voluntariness & -0.9568107 *** & -0.3027344 *** & -0.0917722 ** \\ & ( 0.0775053 ) & ( 0.0427468 ) & ( 0.0456397 ) \\ Better off without & -0.1096345 & -0.2167969 ** & -1.123418 *** \\ & ( 0.1268662 ) & ( 0.1058058 ) & ( 0.1231329 ) \\ Regret accepting & -0.1960133 *** & -0.3203125 *** & -0.7041139 *** \\ & ( 0.064993 ) & ( 0.0525853 ) & ( 0.0510592 ) \\ Regret rejecting & 0.9169436 *** & 0.8828125 *** & 1.037975 *** \\ & ( 0.0861971 ) & ( 0.0581592 ) & ( 0.0589765 ) \\ Likelihood of accepting & 1.774086 *** & 1.615234 *** & 2.025316 *** \\ & ( 0.1052676 ) & ( 0.0750388 ) & ( 0.0714451 ) \\ Personal approval & -1.113726 *** & 0.0169082 & 0.9370078 *** \\ & ( 0.1083016 ) & ( 0.061888 ) & ( 0.0709356 ) \\ IRB Approval & -1.531561 *** & -0.0644531 & 1.376582 *** \\ & ( 0.1065387 ) & ( 0.070405 ) & ( 0.0666284 ) \\ Would participate & 0.2244898 *** & 0.2407407 *** & 0.4433497 *** \\ & ( 0.0307428 ) & ( 0.0237886 ) & ( 0.0246852 ) \\ Observations & 301 & 512 & 632 \\ \hline \end{tabular} \end{center} \caption{\label{compStaticsWithin} Replication of table \ref{compStaticsAcross} including each subjects' responses to each payment amount (within subjects data).} \end{table} \begin{table} \setstretch{1.3} \begin{center} \begin{tabular}{lcccccc} & (1) & (2) & (3) & (4) & (5) & (6)\\ \hline VARIABLES & $d^3$ & $d^3$ & $d^3$ & $d^3$ & $d^3$ & $d^3$\\ Arguments provided & Yes & Yes & Yes & No & No & No \\ & & & \\ ln(Income) & -0.210** & -0.208** & -0.195** & -0.0827 & -0.0697 & -0.0683 \\ & (0.0825) & (0.0827) & (0.0850) & (0.0821) & (0.0818) & (0.0832) \\ Education & -0.105* & -0.108* & -0.106* & -0.187*** & -0.184*** & -0.183*** \\ & (0.0555) & (0.0560) & (0.0569) & (0.0572) & (0.0571) & (0.0585) \\ Male & & 0.222* & 0.242* & & 0.0129 & -0.0143 \\ & & (0.132) & (0.134) & & (0.129) & (0.133) \\ Age & & -0.00614 & -0.00542 & & -0.0168*** & -0.0204*** \\ & & (0.00580) & (0.00599) & & (0.00568) & (0.00593) \\ African American & & 0.189 & 0.157 & & -0.657 & -1.211** \\ & & (0.471) & (0.474) & & (0.537) & (0.559) \\ Asian & & -0.247 & -0.263 & & -0.662 & -1.208** \\ & & (0.485) & (0.488) & & (0.563) & (0.587) \\ Caucasian & & -0.371 & -0.347 & & -0.948* & -1.368*** \\ & & (0.412) & (0.415) & & (0.490) & (0.511) \\ Hispanic & & -0.841 & -0.844 & & -0.808 & -1.323** \\ & & (0.512) & (0.522) & & (0.545) & (0.567) \\ No Religion & & & 0.0661 & & & -0.302** \\ & & & (0.148) & & & (0.139) \\ Conservative & & & -0.171 & & & -0.435** \\ & & & (0.211) & & & (0.193) \\ Liberal & & & -0.0951 & & & -0.133 \\ & & & (0.163) & & & (0.154) \\ Thought about participating & & & -0.189 & & & -0.243* \\ & & & (0.134) & & & (0.132) \\ Constant & 2.847*** & 3.175*** & 3.145*** & 2.566*** & 3.582*** & 4.484*** \\ & (0.846) & (0.936) & (0.964) & (0.845) & (0.975) & (1.009) \\ & & & & & & \\ Observations & 774 & 774 & 755 & 641 & 641 & 620 \\ $R^2$ & 0.017 & 0.032 & 0.035 & 0.022 & 0.047 & 0.074 \\ \hline \end{tabular} \end{center} \caption{\label{determinantsArguments} Replication of columns 1 - 3 of table \ref{determinants} on the subsample of respondents who were and were not provided with the ethical arguments. } \end{table} \begin{table} \setstretch{1.3} \begin{center} \begin{tabular}{lccc} VARIABLE & ``Ethicists'' & Others & ``Economists'' \\ \hline &\\ &\multicolumn{3}{c}{Arguments provided}\\ \hline Voluntariness & -0.4656982 * & -0.0454182 & 0.1687608 \\ & ( 0.264698 ) & ( 0.2330012 ) & ( 0.239059 ) \\ Better off without & 1.170904 ** & -0.3567427 & -0.8222468 * \\ & ( 0.5270833 ) & ( 0.4848331 ) & ( 0.4183129 ) \\ Regret accepting & 0.6761501 *** & -0.1026411 & -0.2562208 \\ & ( 0.1924534 ) & ( 0.1813971 ) & ( 0.1907579 ) \\ Regret rejecting & 0.2958031 & 0.7472989 *** & 0.5362158 ** \\ & ( 0.2722383 ) & ( 0.2202284 ) & ( 0.2440293 ) \\ Likelihood of accepting & 0.3218321 & 0.5660264 * & -0.0103474 \\ & ( 0.4012775 ) & ( 0.3239899 ) & ( 0.3630007 ) \\ Personal approval & -0.6444713 ** & -0.2360944 & 0.6214585 *** \\ & ( 0.261249 ) & ( 0.2312517 ) & ( 0.2156906 ) \\ IRB Approval & -1.08979 *** & -0.0102041 & 1.02981 *** \\ & ( 0.2528326 ) & ( 0.2447896 ) & ( 0.2106807 ) \\ Observations & 143 & 200 & 181 \\ &\\ &\multicolumn{3}{c}{Arguments \emph{not} provided}\\ \hline Voluntariness & -0.8934817 * & -1.069683 *** & 0.5213385 *** \\ & ( 0.4994146 ) & ( 0.2678897 ) & ( 0.1966008 ) \\ Better off without & 1.497615 ** & 0.5764706 & -0.5381348 \\ & ( 0.6414327 ) & ( 0.5189251 ) & ( 0.4118101 ) \\ Regret accepting & 0.9046105 ** & 0.1701358 & -0.3299131 ** \\ & ( 0.3524177 ) & ( 0.2350692 ) & ( 0.1647037 ) \\ Regret rejecting & 1.019078 ** & 0.4968326 ** & 0.543886 *** \\ & ( 0.4130296 ) & ( 0.2437357 ) & ( 0.2010984 ) \\ Likelihood of accepting & 1.81876 *** & 0.0588235 & 0.7832822 *** \\ & ( 0.5788488 ) & ( 0.3973525 ) & ( 0.2970511 ) \\ Personal approval & -0.6073132 & -0.3013575 & 0.3275603 * \\ & ( 0.365104 ) & ( 0.2420997 ) & ( 0.1802602 ) \\ IRB Approval & -0.9984102 ** & -0.3067873 & 0.6370172 *** \\ & ( 0.4638384 ) & ( 0.2698883 ) & ( 0.1998775 ) \\ Observations & 54 & 150 & 250 \\ \hline \end{tabular} \end{center} \caption{\label{compStaticsAcrossArguments} Replication of table \ref{compStaticsAcross} on the subsamples of respondents who were and were not provided with the ethical arguments.} \end{table} \end{document}