Abstract
In this article, we describe a hitherto undocumented fallacy—in the sense of a mistake in reasoning—constituted by a negativity bias in the way that people attribute motives to others. We call this the “worst-motive fallacy,” and we conducted two experiments to investigate it. In Experiment 1 (N = 323), participants expected protagonists in a variety of fictional vignettes to pursue courses of action that satisfy the protagonists’ worst motive, and furthermore, participants significantly expected the protagonist to pursue a worse course of action than they would prefer themselves. Experiment 2 (N = 967) was a preregistered attempted replication of Experiment 1, including a bigger range of vignettes; the first effect was not replicated for the new vignettes tested but was for the original set. Also, we once again found that participants expected protagonists to be more likely than they were themselves to pursue courses of action that they considered morally bad. We discuss the worst-motive fallacy’s relation to other well-known biases as well as its possible evolutionary origins and its ethical (and meta-ethical) consequences.
Keywords
When we judge the moral status of an action, we routinely take into consideration the motives of the person who performs it. In moral philosophy, otherwise divergent views nonetheless agree that the motives behind an action are crucially relevant to its moral status: The agent-based virtue ethics of Slote (1995) take motives to be the exclusive determinants of the goodness of an action, whereas according to Kant’s deontological account, an agent cannot perform a moral action without a good “will.” Even consequentialist views (according to which an action’s moral status is to be judged on the basis of its outcomes) leave some room for the agent’s motives; Sidgwick (1884), for example, suggested that “a man who prosecutes from malice a person whom he believes to be guilty, does not really act rightly; for, though it may be his duty to prosecute, he ought not to do it from malice” (p. 200).
To put these philosophical views into practice, however, we would need to be confident that people are actually good at discerning an agent’s motives. If we are not very good at impartially identifying the motives of other people or, worse, if we are systematically biased in our attribution of motives, then we would need to be much more cautious about adopting a moral theory that recommends or requires us to do so.
In this article, we outline the results of two experimental studies suggesting that people are, in fact, subject to such a bias, which we call the worst-motive fallacy. We found evidence that people display a tendency to assume that an agent’s worst motive for an action is their main motive. We argue that this constitutes a hitherto undocumented cognitive bias.
The Worst-Motive Fallacy: Background
The folk aphorism known as Hanlon’s razor states that one should “never attribute to malice that which is adequately explained by stupidity.” The sentiment is often expressed as a necessary warning against concluding without further evidence that agents acted with bad motives when their actions could well be explained simply by incompetence. How likely is it that such a bias actually exists?
One might have predicted that such a bias existed on the basis of several other well-documented psychological effects. First, there is a general bias toward negativity across a wide range of psychological phenomena (e.g., Baumeister, Bratlavsky, Finkenauer, & Vohs, 2001; Rozin & Royzman, 2001). Emotions, attention, motivation, information processing, and memory are all more strongly influenced by negatively valenced stimuli than neutral or positive ones. The impact of this negativity bias has been documented for decision-making (Kanouse & Hanson, 1971) and even differences in political ideology (Hibbing, Smith, & Alford, 2014). It is therefore not surprising that we might also pay more attention to negative motives in our assessment of other people’s actions.
Second, within social psychology, there are known biases when it comes to the attribution of agency simpliciter (e.g., Hewstone, 1983; Shaver, 1985). We are more likely to view events as having been caused by an agent when those events are negative rather than positive (Morewedge, 2009), and even 6-month-old infants are more likely to attribute agency to a mechanical claw when its actions are “bad” rather than “good” (Hamlin & Baron, 2014). Similarly, in the side-effect effect (Knobe, 2003a, 2003b), an action is more likely to be regarded as intentional when it has a harmful side effect than when it has a beneficial side effect. Pettit and Knobe (2009) further demonstrated that this effect generalizes to desire; we are more likely to think that an agent desired a side effect when that outcome is bad.
Third, another closely related bias is found in the so-called actor/observer difference or the fundamental attribution error of motive attribution (Ross, 1977). Actors’ views of their own behavior emphasize the role of external situational and environmental factors (“I failed the test because a barking dog kept me awake last night”), whereas actors’ views of others’ behavior emphasize internal factors such as character and motives (“She failed the test because she’s lazy and stupid”; see Jones & Nisbett, 1971; Monson & Snyder, 1977; Nisbett, Caputo, Legant, & Maracek, 1973). We are more inclined to attribute our own failures to negative aspects of our environment, whereas we attribute other people’s failures to negative aspects of their character.
Bearing these biases in mind, it might not be so surprising were we to systematically commit what we have called the worst-motive fallacy. Sometimes an agent can have several different reasons for the very same action. In such cases, the reasons can often be ranked, morally speaking: Some may be more praiseworthy or noble than others, and the agent may perform the same action for better or worse reasons. Furthermore, the agent—or an observer—may regard one of the reasons for action as the main reason for action (where the main motive is not just the strongest desire but rather the motive that would be pursued above all others if the available courses of action could satisfy only one). But if an observer were to systematically suppose, all else being equal, that the worst of these motives was the main motive or that other people are primarily motivated by worse motives than they have themselves, this would be a mistake. The two experiments we present here explored whether people have a cognitive bias toward committing the worst-motive fallacy.
The Worst-Motive Fallacy: Two Experiments
Experiment 1
In our first experiment, we required participants to read a short story about a protagonist who has two motives for carrying out some action. In each case, one of the motives is good, and one is bad. The protagonist then discovers that they cannot, after all, pursue the action they had planned: The protagonist must choose between two alternatives, one of which satisfies the original “good” motive, whereas the other satisfies the original “bad” motive. Participants were then asked which option they thought the character in the story would pick and asked to rate the goodness or badness of each motive. Finally, participants were asked which action they would prefer to do themselves. We expected that if the worst-motive fallacy is a real bias, then the worse participants rated a motive, the more likely they would be to expect the agent to choose an option that satisfied that motive rather than a competing one. We also predicted that participants would expect the characters in the story to choose a worse course of action than they would prefer themselves.
Method
Procedure
Each participant was given one of four different vignettes. Here is an example
1
: A politician has some funding left over from her campaign, and she decides to use it to hire a computer engineer that she knows. She does this for two reasons. First, the engineer has recently lost his job and is in need of new work, and the politician wants to help him out. Second, the politician wants the engineer to send misleading messages to her opponent’s supporters to send them to vote on the wrong day. When she describes the work to the engineer, however, the engineer says he will not do it. The politician has two further options. She could hire the unemployed engineer anyhow, to do ordinary computer maintenance work. This will help the engineer who needs income, but won’t help the politician to mislead voters. Or, she could hire a computer hacker who has no problem sending misleading messages. This will help the politician to mislead voters, but will not help out the unemployed engineer.
After they read this vignette, participants were asked, “Which option do you think the politician will pick?” They could choose from two options: “Hire the engineer, to give him work” or “Hire the hacker, to mislead the voters.” After deciding which option the protagonist will pick, participants were then asked to rate the goodness or badness of the motives that were described in the story, using a scale from −10 (very bad) to 10 (very good). Our hypothesis was that participants would expect the protagonist to pick the option that satisfied the motive they rated as worse.
We interpreted participants’ answers to the question about which option the agent would pick as indicating their assessment of the protagonist’s main motive in the original action: Because the protagonist can satisfy only one of the stated original reasons, whatever action they choose would presumably be the one that satisfies the motive that is most important to them. Of course, one might worry that merely mentioning the bad motive might prejudice participants against the protagonist; the mere fact that the protagonist could even entertain such bad motives might be taken as evidence that he or she is a bad person generally. But this is precisely the point; participants know that the protagonist has both good and bad motives because they are both stated explicitly. If our hypothesis about the worst-motive fallacy is correct, participants will indeed be biased toward thinking that the bad motives are the main ones.
Our measure of goodness or badness of a motive also allowed us to control for what we call the extremeness of a motive. Suppose you were told that someone had two motives in mind when setting out on some course of action: wanting to end world hunger and also wanting to mildly annoy a coworker. If you were told that this person found that they could do only one of these things, it is obvious that you would expect the person to pursue the action that would satisfy the motive of ending world hunger. But this is because the motive is so much more extreme than the alternative. Mildly annoying a coworker is a bit bad, but ending world hunger is extremely good. All else being equal, we should expect that if someone has such an extreme motive in mind, they will likely prefer an action that would satisfy it over an action that would satisfy a competing but not very extreme motive. If the good motives in our vignettes were, without our realizing it, all considered more extreme than the bad ones, then this would create a reason for participants to expect the agents in the story to follow the good motives. And if the bad motives were all rated as more extreme, that would create a reason to expect the agent to follow the bad motives. Any bias to expect the agents in the story to follow the bad actions over the good could therefore be revealed only if the extremeness of the motives were controlled. To do this, we treated the distance from zero in either direction as a measure of the extremeness of the motive. If the motive were considered extremely bad, we expected that it would be rated as −10, but if it were just mildly good, it might be rated as +2. The extremeness of the first motive would therefore get a score of +10 (10 points from zero) and the extremeness of the second as +2. Overall, we wanted to make sure that any difference in extremeness was not explaining any bias we found to favor the bad over good motives, so we added this measure as a control.
Notice that participants were not simply asked, “Which of the motives described do you think is the protagonist’s main motive?” This was to avoid alerting the participants to the purpose of the study. Once participants know what a study is exploring, there is a risk that they will adjust their answers to give what they think is a favorable representation of themselves, thereby disguising their real attitudes (Krumpal, 2013). Asking participants what the protagonist in the story will likely do next allows us to explore participants’ assumptions about the protagonist’s motives without asking them about this directly.
Finally, we asked participants which of the options they would choose themselves if they were in the position of the protagonist. We predicted that participants’ responses to this question either would fail to match the course of action that they expected of the agent in the story or would negatively correspond to that action. This allowed us to rule out the possibility that participants were simply ascribing to the protagonists the course of action that the participants would prefer themselves.
In each vignette, the motives ascribed to the character were counterbalanced so that in half of the vignettes, the bad motive was described first, and in the other half, the good motive was described first. The number of words used to describe the good and bad motives and actions was the same, so that, overall, we gave participants no reason to suppose that one of the motives described was the primary motive. Similarly, the contexts described were very different—we used four different vignettes: a politician (described above); a man who must decide whether to take the bus to town to buy his friend a present or take the train to rob a pensioner; a child going to a party who must decide whether to wear a dress that will embarrass the host or a pair of jeans that will make her mother happy; and a college student who has to decide whether to go to France for the summer, where he expects that he can cheat on his girlfriend, or go to stay with cousins in Argentina, where he will learn Spanish to improve his studies (vignettes can be read in Section S1 in the Supplemental Material available online). 2 In spite of the diversity of the contexts described and the counterbalancing of the presentation of the motives, we predicted that participants would be more likely to expect the character to pursue the action that satisfies the worst of the two motives.
Participants
The study was run using the Qualtrics online survey platform (https://www.qualtrics.com/), and participants were recruited using Amazon’s Mechanical Turk. Participants were paid 25¢ for their participation, which took about 1 to 2 min. In a pilot study with 52 participants, we found a significant difference between the rating of motives behind actions that participants expected to be chosen and not chosen, t(51) = −2.132, p = .037. A power analysis indicated that 75 participants would produce a significant difference with an alpha of .01. Because we designed four vignettes and counterbalanced the order in which the motives and options were mentioned in the vignettes in eight ways, we aimed to collect data from 80 participants for each vignette, which would allow the eight counterbalanced versions to be distributed evenly across the group. As a result, we aimed to collect data from 320 participants. We put three attention-check questions at the beginning of the experiment and excluded participants who did not correctly answer all of those questions. We also excluded participants who did not complete the whole test. Finally, if the same participant took the survey more than once, we excluded the second response. In total, we collected 408 responses and excluded 85, resulting in 323 responses.
Analysis
We analyzed the results with a generalized linear model run in the R programming environment (Version 3.4.3; R Core Team, 2017) and using the lme4 package (Bates, Mächler, Bolker, & Walker, 2015). This allowed us to model the main predictor along with control predictors all at once and avoid the problems associated with running multiple analyses. In our full model, we included choice of option (good or bad) as the dependent variable. The main predictor was the difference between the participant’s ratings of the motives behind the two options. We expected that the worse participants rated the bad motive relative to the good one, the more likely they would be to expect the protagonist to pursue the action that satisfied that motive. To control for extremeness, we included the difference between the absolute ratings of the motives (the distances from zero in either direction). We included this further predictor in the model as a random effect, along with the interaction between extremeness and rating.
We identified significance levels using likelihood-ratio tests to compare the fit of different models. The full model included terms for rating, extremeness (the absolute value of rating), and their interaction, along with a measure of the participant’s own preference (which choice the participant said they would prefer to take), the vignette read by the participant, and duration (how long the participant took to complete the test). The null model included only duration and vignette.
Results
The comparison of the full and null models was significant, χ2(4, N = 323) = −19.16, p < .001. To isolate the source of the effect, we compared a model including the interaction between the absolute rating and the basic ratings but found no effect of the interaction, χ2(1, N = 323) = −0.541, p = .4616. We therefore removed the interaction from the model. We then compared models that differed only by the inclusion of the basic rating but which both featured the absolute rating and the participant’s own preference. We found that including the basic rating in the model significantly improved the fit, χ2(1, N = 323) = −5.8005, p = .016. This outcome confirmed our hypothesis: Participants expected the agents in the story to follow the course of action that satisfied the motive that they considered relatively worse, in spite of any effect of the extremeness of the action. 3 There was indeed such an effect of extremeness: The more extreme the motive, the more likely the participants were to expect the agent to follow it, χ2(1, N = 323) = 8.7769, p = .003. But because this was included in the same model as the basic rating, it cannot account for the bias toward the bad motives by itself. 4 These results are plotted in the left side of Figure 1.
Results from Experiment 1. The x-axis shows the options that participants expected would be chosen or not chosen by the protagonist (left) or by themselves (right). The y-axis shows how participants rated the motives behind those actions. Disks represent individual participants’ ratings of a given motive; in each box, the horizontal line represents the mean, and the top and bottom edges of the box indicate the 95% confidence interval.
Finally, we found a significant effect of the participant’s own preference: Participants predicted that they themselves would have chosen the opposite of what they expected the protagonist to choose, χ2(1, N = 323) = −6.607, p = .01, as can be seen in the right side of Figure 1. This is consistent with findings that we generally expect other people to act in worse ways than we would prefer to act ourselves.
There was no effect of vignette and no effect of duration of the experiment. All data, scripts, and modeling results can be viewed at https://osf.io/mjrpy/.
Experiment 2
Method
The Action Editor for this article suggested that we conduct a preregistered replication of our first experiment to test its robustness. We therefore did so using the same setup as in Experiment 1 (see https://osf.io/9djav), in which we ran a total of 12 vignettes: the original four as well as eight further vignettes (reproduced in Section S1; one of these was based on a recommendation by a reviewer, and the other seven were written by the authors). To analyze the new data, we used a generalized linear mixed model in R (Version 3.4.3) with the lme4 package and the logit link function. This allowed us to include a random intercept of vignette in our analysis, thereby ensuring that any effects are generalizable to the full set of vignettes. In total, we collected 1,657 responses and excluded 690 (41%), resulting in 967 responses. Participants were excluded for not passing attention-check questions (25.7%), taking the test more than once (14.6%), or not finishing the test (1.7%). The high exclusion rate raises some suspicions about the data quality, and the results should be considered in that light.
Results
The comparison between the full and null models was highly significant, χ2(4, N = 967) = 141.98, p < .0001. We found no significant effect of the rating: As can be seen on the left side of Figure 2, the 95% confidence intervals are overlapping. Therefore, the first main effect of Experiment 1 was not replicated, as we thought it might be. We did find an effect of extremeness: The more extreme an option, the more likely participants were to follow it, χ2(1, N = 967) = 39.712, p < .0001. However, this time, the good motives were treated as more extreme than the bad motives. As we discuss below, this may explain why our prediction that participants would pick bad over good motives was not replicated. Again, we found a significant overall effect of the participant’s own choice across all vignettes. Participants expected the protagonist in the story to pursue actions that they rated as worse than the actions they would have preferred to pursue themselves, χ2(1, N = 967) = 97.141, p < .0001. This is depicted in Figure 2.
Results from Experiment 2. The x-axis shows the options that participants expected would be chosen or not chosen by the protagonist (left) or by themselves (right). The y-axis shows how participants rated the motives behind those actions. Disks represent individual participants’ ratings of a given motive; in each box, the horizontal line represents the mean, and the top and bottom edges of the box indicate the 95% confidence interval.
To explore more clearly why our main prediction was not replicated, we organized the data in the replication into subsets to distinguish the original vignettes (that were also used in Experiment 1) from the new vignettes (that were not used in Experiment 1). We found that in the original vignettes, the main effect of rating was again significant, χ2(1, N = 332) = 5.9072, p = .015; therefore, in this sense, the precise outcome of Experiment 1 was indeed replicated. Participants reading the vignettes from the first study again expected the characters in the stories to be more likely to act on the bad rather than the good motive. The effect was stronger, in fact, than the effect in the first study under the same analysis, χ2(1, N = 323) = 4.434, p = .035, in spite of the sample size being almost identical (323 participants in Experiment 1 and 332 participants in Experiment 2 reading the same vignettes; see Note 4). There was also an effect of extremeness: As before, participants expected the agent to follow the action that they rated as more extreme, χ2(1, N = 332) = 18.589, p < .0001. But again, because this is included in the model that displays the effect of basic rating, it cannot explain away that effect.
In the new vignettes considered on their own, on the other hand, there was no effect of rating. There was, however, again a significant effect of extremeness, χ2(1, N = 635) = 20.034, p < .0001. This could mean that the participants may have treated the extremeness of the good motives in the new vignettes as stronger than the bad ones, and this may explain why the tendency to rate the bad motives as more likely to be acted on was not found in the new vignettes. To check this, we looked at the mean values of the good and bad motives in the original and the new vignettes from the replication. Sure enough, for the original vignettes featured in the replication, the bad motives were considered more extreme than the good motives (absolute bad: M = 8.2; absolute good: M = 7.45), but for the new vignettes, the good motives were considered more extreme than the bad motives (absolute bad: M = 5.63; absolute good: M = 7.38).
We think this explains why the effect was not found for the new vignettes; any bias that participants may have had to consider the bad motives as more likely to be acted on was counteracted by their consideration of the good motives as being more extreme than their alternatives. Recall that absolute rating was included in the model that revealed the effect of rating in Experiment 1 and for the original vignettes in Experiment 2, which means that absolute rating or extremeness could not explain the effect of rating that was found in those analyses. The effect of badness, in which participants expect the agents in the story to act on the worst motive, goes beyond the effect of extremeness, even though they both go in the same direction. But where there is an effect of extremeness in the opposite direction, this could explain why we found no effect of rating in the new vignettes, because considering the good motives as more extreme than the bad motives will have pushed participants’ intuitions about which motive was more likely to be acted on in the direction of the good motives, thus canceling out any effect of badness. These exploratory analyses were not included in the preregistration. Further details on methods in Experiment 2 can be found in Section S2 in the Supplemental Material.
Overall, then, we found clear evidence of a tendency for participants to expect other people to be more likely to act on the worst motive attributed to them, all else being equal. We were not able to replicate the effect in our new vignettes, but because the good motives were rated as far more extreme than the bad motives in those cases, this failure does not cast doubt on the finding. However, future studies, involving new vignettes with motives that are more evenly balanced, will be needed to make the case that the effect will generalize indefinitely to new cases.
General Discussion
The results of our experiments suggest that the tendency against which Hanlon’s razor warns is, in fact, a real tendency in our judgments of other people’s motives. Across a range of contexts, we found evidence that people were inclined to expect that agents are motivated primarily by the worst of the reasons that they have for a given action and that people expect others to be motivated by worse reasons than they are motivated by themselves. Although we were unable to replicate this effect for a broader range of vignettes than we considered in Experiment 1 (the first four from Experiment 1 plus an additional eight), it was replicated for the original vignettes when these were analyzed separately. Additionally, it is clear that there was a difference in the new vignettes, which could explain this failure: Participants rated the good motives as more extreme than the bad motives in the new vignettes. This difference may have counteracted any tendency to be biased to expect the agent to act primarily on the worse motive. We also found that in all cases, participants expected the agents in the story to be more likely to act on the motives that they found more extreme, as we suspected we might when we set up our study. And finally, the second effect we reported, in which participants expected other people to be more likely to pursue actions that they consider bad than they would prefer to do themselves, was replicated for all of the vignettes in Experiment 2.
The worst-motive fallacy fits naturally within the family of general negativity biases mentioned in the Background section, because it suggests, in effect, that we are also negatively biased in our moral evaluation of other people’s motives. Plausibly, we consider the worst reasons for actions to be the main motives because of our more general tendency to place greater focus on negative stimuli rather than positive, coupled with a more pronounced tendency to evaluate other people’s characters more negatively than we do our own.
We think that the worst-motive fallacy may arise because of the adaptive advantages that are gained from paying more attention to negative rather than positive aspects of other people’s behavior, the wisdom of which is recommended by another common folk aphorism 5 : “Hope for the best but prepare for the worst.” A cognitive bias may be selected for when the errors in which it results are less costly than erring in the opposite direction (Haselton & Nettle, 2006). In general, the evolutionary story goes, it is more advantageous to pay attention to negative aspects of our environment and thereby avoid harm, even if that means failing to notice positive aspects and thereby missing good opportunities. In the context of the worst-motive fallacy, although being overly suspicious of other people’s motives may incur the cost of failing to take up cooperative opportunities, the cost of naively entering cooperative partnerships with malicious actors may be higher. It will therefore be more advantageous to err on the side of falsely believing that other people have bad motives than to risk falsely believing that they have good motives.
Given the evolutionary account proposed here, several follow-up studies using similar methods could be conducted to explore the boundary conditions of the effect we have identified. For example, one might expect to find an in-group/out-group effect that leads to more benign interpretations of other people’s motives when they are relatives or when more is known about the protagonist’s history, prior behavior, or decision-making context. If we are generally predisposed to treat other people with suspicion, then it seems plausible that this would extend to an increased negativity bias when it comes to out-group attribution of motives. Such an effect would likely manifest along depressingly familiar prejudicial lines of gender, race, nationality, class, and so on.
Conclusion
What about those philosophical theories, considered at the outset, that appeal to an agent’s motives in the assessment of the morality of actions? The present study suggests that we should be cautious about appealing to our assessment of other people’s motives to judge the morality of their actions. The negativity bias we have uncovered casts doubt on the practicalities of any meta-ethical theory that recommends that our moral evaluation of other people’s actions should be rooted in our assessment of their motives. Similarly, the robust effect of extremeness (i.e., the fact that people are more likely to attribute a motive when it is further away from morally neutral in either direction) gives us a further reason to be wary. Given two competing motives that we have reason to believe an agent has in mind, it seems we are likely to consider not only the bad motive to be the main motive but also the more morally extreme motive to be the main motive. Although such meta-ethical theories could still be correct that, objectively, actors’ motives play an essential role in the goodness of their actions, they should nonetheless carry a user warning, as it were, that our subjective assessment of those motives may be far less reliable than is generally supposed.
These are matters for future investigation beyond the scope of this article. Our present focus has been to formally identify this unnoticed fallacy and to demonstrate for the first time that there is a tendency for people actually to commit it. Of course, the reader might suspect that our main motive in writing the present article was something else again: to publish in a top-ranking, peer-reviewed journal for the purposes of fame, glory, and career advancement. We suggest, however, that to suppose this would be to commit a fallacy, whose cause is a demonstrably commonplace cognitive bias.
Supplemental Material
Walmsley_SOM_Section_S1 – Supplemental material for The Worst-Motive Fallacy: A Negativity Bias in Motive Attribution
Supplemental material, Walmsley_SOM_Section_S1 for The Worst-Motive Fallacy: A Negativity Bias in Motive Attribution by Joel Walmsley and Cathal O’Madagain in Psychological Science
Supplemental Material
Walmsley_SOM_Section_S2 – Supplemental material for The Worst-Motive Fallacy: A Negativity Bias in Motive Attribution
Supplemental material, Walmsley_SOM_Section_S2 for The Worst-Motive Fallacy: A Negativity Bias in Motive Attribution by Joel Walmsley and Cathal O’Madagain in Psychological Science
Footnotes
Transparency
Action Editor: Michael Inzlicht
Editor: D. Stephen Lindsay
Author Contributions
J. Walmsley and C. O’Madagain share equal coauthorship. J. Walmsley and C. O’Madagain jointly conceptualized the research problem, designed the methodology, developed the experimental resources, and wrote and revised the manuscript. C. O’Madagain collected and analyzed the data. Both authors approved the final manuscript for submission.
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
