When money talks: Judging risk and coercion in high-paying clinical trials

Millions of volunteers take part in clinical trials every year. This is unsurprising, given that clinical trials are often much more lucrative than other types of unskilled work. When clinical trials offer very high pay, however, some people consider them repugnant. To understand why, we asked 1,428 respondents to evaluate a hypothetical medical trial for a new Ebola vaccine offering three different payment amounts. Some respondents (27%) used very high pay (£10,000) as a cue to infer the potential risks the clinical trial posed. These respondents were also concerned that offering £10,000 was coercive— simply too profitable to pass up. Both perceived risk and coercion in high-paying clinical trials shape how people evaluate these trials. This result was robust within and between respondents. The link between risk and repugnance may generalize to other markets in which parties are partially remunerated for the risk they take and contributes to a more complete understanding of why some market transactions appear repugnant.

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 weekly 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. If 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 with the vaccine.
Study participation invitations will be posted throughout the 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 she takes and the time commitment required to participate in the study (∼ 40 hours).
How many participants you would expect to have any side effects from taking part in the study? 0-100 [ 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? [7-point Likert scale with extremes labelled "strongly disapprove" and "approve without reservation"]

NEW 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 as in her usual job in an entire month].

NEW PAGE [Reminder of the text on A.S.]
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? [0-10] How many of the 10, do you think, will eventually participate in the study in exchange for [ Have you ever thought about participating in a medical research study as a means to earn money?
Would you participate in the experiment about the Ebola vaccine described in this study for a payment of £50?
Would you participate in the experiment about the Ebola vaccine described in this study for a payment of £1,000? Would you participate in the experiment about the Ebola vaccine described in this study for a payment of £10,000? ["yes", "no", "I do not know". In the first two of the above five questions, the choice "prefer not to answer" was also available.] To what extent do you agree/disagree with the following statements (about clinical trial markets) in general? A "clinical trial market" refers to research institutions being able to offer monetary incentives to prospective participants to take part in a clinical trial, who can freely agree or disagree to take part. [7-point Likert scale with extreme labelled "completely agree" and "completely disagree"] • Clinical trial markets are deplorable.
• Clinical trial markets should be banned.
• Clinical trial markets should be tightly monitored by the government.
• Clinical trial markets are morally permissible.
Note. These are almost identical materials as in Ambuehl et al. (2015). The differences were as follows. We (1) added the side effects question, (2) removed "midwestern" from the city description and expressed the payment in £ due to our primarily British sample, (3) in wave 2, added questions pertaining to how repugnant respondents considered clinical trial markets (see last question set), (4) asked respondents about their willingness to take risks and numeracy in addition to the other demographic variables (also see demographics table) and (5) appended an exploratory task in which respondents estimated their chances of loosing various hypothetical monetary amounts (see osf.io/5kewt/). We did not have predictions about how responses to this task are linked to our questions of interest here; the data will be analyzed and reported at a later point in time.  Table S1: Replication of Table 1 in the manuscript, but including all three payment amounts per respondent (within-respondent analyses). We accounted for random variation between respondents by including Response ID as a grouping factor. Estimated coefficients in columns are main effects (a-d) and interaction effects between respondent types from multivariate fixed-effects regressions (variable ∼ payoff × respondent type); using £1,000 and "trustful" as a baseline (see the Intercept column, a). One regression per dependent variable. Credible effects in bold.

Gender effects
As the vignette concerned female volunteers recruited for a clinical trial, female respondents may assess the clinical trial more critically than male respondents. We explored this possibility below.

Variable
Gender ( Table S2: Replication of interaction effects in Table 1 in the manuscript, but including gender (betweenrespondent analyses). Women were more critical of the clinical trial regarding voluntariness, and compared to men they thought that a woman volunteering may regret accepting the offer. They also approved less of the clinical trial personally; and were less likely to give IRB approval. The interaction effects between payoff and being doubtful were robust.   Table 2 in the manuscript, but including all three payment amounts per respondent (within-respondent analyses). We accounted for random variation between respondents by including Response ID as a grouping factor. Estimated coefficients in columns are main effects (a-d) and interaction effects between respondent types from multivariate fixed-effects regressions (side effect ∼ payoff × respondent type); using £1,000 and "trustful" as a baseline (see the Intercept column, a). One regression per side effect type. Credible effects in bold.
Note. We also tested for order effects. It is plausible that within-respondents effects would be larger when going from low to high payment amounts. Indeed, the within-respondent effects were mostly driven by respondents who were presented with £50 first, and the larger amounts later (b T rustf ul = 2.63, CI = [1.04, 4.24], b Doubtf ul = 3.74, CI = [1.89, 5.60], estimated number of any side effects when £50 was presented first). Because of this, we relied on between-respondent differences in our primary analyses.

Gender effects
As the vignette concerned female volunteers recruited for a clinical trial, female respondents may assess the clinical trial more critically than male respondents-and expect side effects to be higher. We explored this possibility below.
Side effect type Gender (female) £10,000×Doubtful    Table 2 in the manuscript, but including gender (betweenrespondent analyses). Women generally estimated side effects to be higher compared to men (chosen as the reference group). Consistent with results reported in the manuscript, doubtful respondents still judged side effects to be higher in the £10,000 condition.  Table S5: Results for models linking clinical trial evaluations to respondents' estimated number of side effects. Columns a-c use the variable "any" side effects as the predictor. Column (a) shows the main effect; columns (b) and (c) show β coefficients from a model that included both side effects and voluntariness as predictors ("S&V", modeled as two main effects). Columns (d) and (e) show main effects using "mild" and "severe" side effects as predictors.
*same models as above, but including an interaction with payoff (using £1, 000 as a baseline) -as the δs also refer to payoff-dependent differences in the evaluations.
Note. We also tested a number of other models, for instance controlling for doubtful/trustful × compensation amount, the interaction between side effects estimates and payment amount, or the three-way interaction doubtful/trustful × compensation amount × side effects estimates. The main effects of side effect estimates on the clinical trial evaluations was present in all models. Generally, and as can be seen from Table S5, a higher number of estimated side effects results in a less positive evaluation of a medical trial, and a higher degree of "voluntariness" for prospective participants results in a more positive evaluation of a medical trial.

Clinical trial evaluations (mediations)
We used a mediation approach suggested by MacKinnon et al. (2000) to show how payment information differentially affected IRB approval of doubtful and trustful respondents, given their differential estimates of side effects and voluntariness.

Individual payoff sensitivity
Our findings suggest that doubtful respondents (1) were more likely to utilize payment amount as a cue when estimating side effects; (2) also considered high payments to increase coercion (decrease voluntariness); and (3) evaluated high payments as less ethical (IRB approval). Is there a link between sensitivity to payoff information for IRB approval and side effects; and/or for IRB approval and voluntariness? In an exploratory analysis, we turned to within-respondent analyses for these dependent variables (i.e., using respondents' answers pertaining to each payment amount). For each respondent, we computed a "payoff sensitivity score" that measured how responses changed as clinical trials offered extremely high compensations. We did this for respondents' clinical trial evaluations (IRB approval: δ IRB = IRB £10,000 − IRB £1,000 ), for their inferred number of any side effects [SE] (δ SE = SE £10,000 − SE £1,000 ), and for voluntariness (δ volun. = volun. £10,000 − volun. £1,000 ). As Figure S3 shows  Figure S3: Relation between payoff sensitivity in the estimated number of side effects (normalized δ SE ) and payoff sensitivity in IRB approval ratings (normalized δ IRB ). A payoff-dependent increase in estimated side effects was linked to lower IRB approval rates for doubtful, but not trustful respondents (within-respondents). Each dot represents one participant. Note: Although the relationship is statistically reliable, as the density of the datapoints suggests, most respondents did not infer a different amount of side-effects-which is sensible given they saw the same vignette (just with a different payoff) repeatedly.  Table S6: Columns a-c used estimates of "any" side effects for £10, 000 as predictors. Column (a) shows the main effect; columns (b) and (c) show β coefficients from a model that included both side effects and voluntariness as predictors ("S&V", modeled as two main effects). Columns (d) and (e) show main effects using side effects estimates for £50 and £1, 000.  Table S7: Side effects. Main effects for each type. Each respondent is only entered in the regression once, with their estimate for "any' side effects given the £10,000 trial.    Table S9: Repugnance predicted from side effects, voluntariness, and demographic variables (also see S5 for main effects). As before, the combined model reveals that a higher number of estimated side effects increases repugnance; lower voluntariness (i.e., higher coercion) has the opposite effect. Beyond these predictors, only whether or not respondents had thought about participating themselves and willingness to take health risks predicted additional, unique variance. Being doubtful was not a reliable predictor (CI includes 0).  Taking risks (general) Figure S5: Characteristics of current sample. The sample consisted of a slightly higher proportion of women than men; and was slightly skewed toward younger age groups. The median income was roughly equivalent to the median income in the UK (dashed line). Our sample was risk averse toward taking health risks (a potential consequence of asking questions about clinical trials prior asking about health risks), and risk neutral toward taking general risks.

Sample characteristics: Excluded respondents
We assessed whether there were any systematicities in demographics for participants who were excluded (due to failed attention checks) vs. included. The analysis below predicts the probability of being excluded based on various demographic characteristics (Gender, Income, Education, Age

Sample characteristics: Between conditions
We assessed whether there were any systematicities in demographics for participants who were assigned to the £10,000 condition (compared to the £1,000 condition). The analysis below predicts the probability of being assigned to the £10,000 (high payoff) condition based on various demographic characteristics (Gender, Income, Education, Age). Baseline for Education: Advanced Graduate Work or PhD. Baseline for Gender: Female. All credible intervals included 0; suggesting that the random assignment worked as intended.   Table S10: Demographic variables and relationship to measures of interest. δs are the differences in respondents' evaluations of the clinical trial offering £10,000 vs. £1,000 (δ = Rating £10,000 − Rating £1,000 ). All coefficients are reported with their 95% credible intervals. *only collected in wave 2 Note. As in the original survey, we find that personal approval and IRB approval are highly correlated (β = .76). Moreover, IRB approval and ethicality are correlated, but to a lesser extent (β = .36). The distinction between doubters and trustful respondents is based on δ ethicality .