1. Background on the Nonidentity Problem

The nonidentity problem arises because there are some acts that strike us as intuitively immoral, and yet the acts effectively change which people come into existence. For example, consider the following case, inspired by Derek Parfit (Reference Parfit1984):

It is intuitively plausible that the US government's choice was immoral. The initially plausible reason is that the future people under Depletion have been wronged by that choice. And the natural reason behind this judgment is that the people within that future population have been harmed by that policy choice, because they were left with a dirty and depleted environment. But, on closer inspection, this is a difficult case to make. The people within the population under Depletion were not made worse off, for if Conservation had been chosen, a completely different set of people would have existed by 2200. So, given the natural thought that you cannot harm someone if you do not make that person worse off, no particular individual within the future population under Depletion was harmed by the US government's choosing that policy. So, it looks like no individuals in that population were wronged, and it seems that the choice was not immoral after all. Yet this clashes with our strong initial intuition about the case.

What, precisely, is the problem here? As we see it, it involves a clash between the strong initial intuition we tend to have about cases like these and a series of other seemingly reasonable judgments or assumptions. First, we have a strong intuition that the act or policy in question is immoral. Second, we tend to assume that if an act or policy is immoral, this must stem from the fact that the act or policy wrongs someone. Third, we also tend to assume that wronging another person requires doing harm to that person. And, finally, we tend to assume that doing harm to another person requires making that person worse off than they would have been otherwise. But in these nonidentity cases, no one is made worse off, and, if our other assumptions are correct, the act could not have been immoral in the first place. So either our initial intuition is wrong or one of our other reasonable seeming assumptions or judgments must be wrong.

Philosophers have attempted different solutions to the problem by tackling each of these four collectively inconsistent pieces of the puzzle. Taking them in reverse order, some philosophers have argued that harming another does not require making that person worse off, (among them, Hanser Reference Hanser1990; Harman Reference Harman2004, Reference Harman, Roberts and Wasserman2009; Rivera-López Reference Rivera-López2009; Shiffrin Reference Shiffrin1999), or that we can otherwise account for how the person/people in the nonidentity cases are indeed harmed, (such as Gardner Reference Gardner2015). Some have attempted to deny that wronging a person requires that you have harmed that person (Kumar Reference Kumar2003, Reference Kumar2015), and others have offered somewhat related contractualist solutions (Finneron-Burns Reference Finneron-Burns2016; Gibb Reference Gibb2016; Hurley and Weinberg Reference Hurley and Weinberg2015). Some have attempted to explain how the act or policy could be immoral without strictly speaking wronging anyone, which was Parfit's own attempted solution (Reference Parfit1984, Reference Parfit2017) (see also, for example, Buchanan et al. Reference Buchanan, Brock, Daniels and Wikler2000; Steinbock Reference Steinbock, Roberts and Wasserman2009). And some have argued that our strong intuition—that the act or policy is immoral in much the same way that it would be if the same person or people were affected—is simply mistaken (among them, Boonin Reference Boonin2014; Weinberg Reference Weinberg2014).

But regardless of which strategy these philosophers prefer, each of them shares the strong intuition that nonidentity cases are immoral in much the same way as their parallel same-person cases are immoral. Even those, like David Boonin (Reference Boonin2014), who end up arguing that this intuition is mistaken, and that the actions in nonidentity cases are not actually immoral, still agree that this intuition is both strong and widespread. So, in a sense, this assumption that the act or policy does not receive moral absolution simply because different people come into existence seems to be a central driving force in the literature on the nonidentity problem. If we take these authors at their word, they all share the strong intuition, their colleagues tend to share the intuition (Parfit Reference Parfit1984: 359, 363), and their students tend to share the intuition (Boonin Reference Boonin2014: 25).

As alluded to earlier, we have had a different experience: it is difficult to motivate the nonidentity problem as a genuine puzzle because a number of students see an intuitive moral difference once they realize different people will come into existence depending on which choice is made. In other words, it seemed to us that the central intuition that drives the whole literature on the nonidentity problem might not be widely shared outside of our (relatively small) community of philosophers. If our hunch is correct, perhaps this could raise some debunking worries for that central intuition, since it could be that philosophers are relying upon intuitions not widely shared in the general public. (For some examples where the intuitions of philosophers seem to differ from the folk, see Tobia, Buckwalter, and Stitch [Reference Tobia, Buckwalter and Stitch2013]; Machery [Reference Machery2017: chapter 2].) Although we think there is probably something to that line of concern, our worry here is of a more practical nature.

Think back to the choice between the two policies, Conservation and Depletion. Regardless of what analysis any particular philosopher settles upon concerning that case, we assume that this philosopher's intuition tells her that a nation that chooses Depletion over Conservation has done something seriously immoral, all else being equal. Now, many philosophers working on the nonidentity problem see it as having genuine practical relevance. The idea is that if we cannot solve the nonidentity problem, then we will have a hard time justifying why people ought to support Conservation over Depletion. But, from a motivational perspective, the solution they seek might be somewhat irrelevant. If the general public shares the intuitions and judgments of the philosophers working on the nonidentity problem, then they should also think that the fact that different populations come into existence under each policy does not make a moral difference. So if the public's moral motivation in policy choices is predominantly driven by how wrong a policy intuitively seems to them, then we should assume that the public would be just as motivated to support Conservation over Depletion after they grasp that the policies change who exists as they were before they had that realization. So, if philosophers’ intuitions are fairly typical among the population at large, then the nonidentity problem may prove to be a mere theoretical puzzle of little practical relevance. (However, this is not to say that there will not be other issues regarding climate change distinct from the nonidentity problem that are practically relevant; see Gardiner [Reference Gardiner2011: chapter 5] for various others.)

If, on the other hand, the intuitions of philosophers in nonidentity cases are not representative of the population, we are in a very different situation. If the public's moral intuitions and judgments quickly shift once they grasp that they are dealing with a nonidentity case, then we should expect that they will be less likely to support policies that involve some sacrifice on their part. This, in turn, would have some implications for how we, philosophers, ought to talk about these kinds of problems with the public. For example, it might be a morally bad idea to broadcast this particular philosophical problem to the public, for example, in public talks, podcasts, or popular periodicals. (Indeed, it might be a morally bad idea for us to have written this article.) As will become clearer, it also might speak against the seemingly promising strategy of focusing on considerations of harm when attempting to motivate the public to act more conscientiously toward the environment, as suggested by Joshua Rottman, Deborah Keleman, and Liane Young (Reference Rottman, Keleman and Young2015), among others. If the public's intuitions on this matter are very different from those of philosophers, we might be in real trouble if we focus our moral motivational efforts on harm considerations and then inform the public about the nonidentity problem.

So, do the intuitions and judgments of philosophers mirror those of the general public?

2. Two Experiments

Our experiments were designed with two main goals in mind: to determine whether individuals are less likely to make altruistic sacrifices in identity-affecting choice problems and to examine the role that considerations of harm might play in any changes of behavior when making identity-affecting choices. To satisfy these dual goals, we relied on a mix of traditional survey methods and experimental methods from economics. While philosophers are largely familiar with survey methods, experimental methods from economics have received significantly less attention. (We note some exceptions: Bicchieri and Xiao Reference Bicchieri and Xiao2009; Bicchieri and Chavez Reference Bicchieri and Chavez2010; Bruner et al. Reference Bruner, O'Connor, Rubin and Huttegger2018.) And yet, the experimental methods of economics, since they explicitly were devised to explore behavior, are particularly promising given our goal of identifying the behavioral consequences of identity-affecting considerations. Such experiments proceed by observing how subjects behave in a particular scenario, as opposed to merely reporting how subjects believe they would behave. To do this, the subjects’ decisions must have real consequences. In typical economics experiments, subjects make decisions with full knowledge that those decisions have financial consequences for themselves and other participants.

To understand more fully whether identity-affecting choices will encourage more self-interested behavior, we ideally would place subjects in conditions that approximate the salient features of the actual scenario of interest. This is why methods from experimental economics are so helpful for this kind of examination. Theories with genuine practical implications only rarely involve how subjects say they would behave in some environment (that is, only in those cases where the phenomenon of practical relevance is itself what people tend to utter in such and such circumstances). Instead, they make predictions about how subjects will actually behave. And there is substantial evidence that, in many cases, what agents say they will do and what they will actually do come apart (FeldmanHall et al. Reference FeldmanHall, Mobbs, Evans, Hiscox, Navrady and Dalgleish2012; Vlaev Reference Vlaev2012). Second, and more importantly, methods from experimental economics have an established record for effectively probing the other-regarding preferences of individuals. For obvious reasons, we could not conduct an experiment that forces individuals to make choices that in turn directly cause different individuals to come into existence. Instead, we recreated strategic scenarios that approximate the relevant and salient features of the nonidentity cases of interest. Behaviors in this proxy condition provide useful insights.

2.2 Experiment 2: The Treatment Study

Our second experiment investigated a variant of the standard dictator game designed to add identity-affecting considerations into the choice problem. As in experiment 1, a dictator (Player A) was allocated $1.00 to divide. However, unlike the previous experiment, there was a total of eleven Player B participants paired with each Player A participant. We referred to these eleven experimental subjects as the Player B Group, and assigned each a label (B0, B1, …, B10). Each member of the Player B Group was matched to a particular outcome of the dictator game: Player B0 was matched to the outcome where Player A transfers no funds, Player B1 was matched to the outcome where Player A transfers $0.10 to their counterpart, and so on. As in experiment 1, Player A then selected an allocation, which was later revealed to the member of the Player B Group predetermined to receive that amount. Importantly, this revelation was the only form of contact between the Player As and any members of their respective Player B Groups.

Those Player B Group participants that did not match the selected allocation were simply paid the $0.50 participation fee and asked to take an unrelated survey. These Player B participants were not provided with any information about the game they had just participated in or the behavior of the Player A they were paired with. In other words, all but one of Player B Group participants were kept completely ignorant of the strategic scenario. Importantly, each Player A was told that those members of the Player B Group not matched to the selected allocation would never find out about Player A's behavior. As a result, this variation of the game mimics some of the salient identity-affecting considerations that generate the nonidentity problem. Player A's decision not only determines both how much money is transferred but also which individual will be drawn into the strategic scenario. And, importantly, any member of a Player B Group chosen to receive a bonus through this process had no chance of doing any better than she did. If Player A had chosen a more generous allocation, a completely different member of the Player B Group would have reaped that reward. Lastly, those not chosen to receive a bonus in the process never find out about the decision, just as those who do not come into existence because of some choice will never have the opportunity to find out about that fact.

To make these details a bit more concrete, figure 1 illustrates the difference between the choice problem faced by Player A in experiment 1 and experiment 2. Hereafter, we refer to experiment 1 as the standard dictator game and experiment 2 as the nonidentity dictator game.

Figure 1

Representation of the choice problem faced by Player A in experiment 1 (standard dictator game, left) and experiment 2 (nonidentity dictator game, right)

As noted above, we found in experiment 1 that subjects appeared to be driven at least partially by other-regarding behavior. Our question was whether the identity-affecting nature of our nonidentity variant of the dictator game would have a noticeable effect on the behavior of our subjects.

2.2.1 Experiment 2: Experimental Set-Up

We again used subjects from MTurk, recruiting a total of 176 Player A participants (male = 100, female = 72, other/decline = 4) and 1936 Player B participants. Once again, our interactions with subjects took place over the online interface, and subject anonymity was guaranteed. All subjects received a $0.50 participation fee upon successful completion of the experiment and were told that they might receive up to an additional $1.00. On average, the experiment took the active subjects just over four minutes to complete.

Player A participants were presented with a detailed description of the nonidentity dictator game. They were then administered a three-question quiz to ensure that they not only understood how their choice affected their own compensation but also determined which of the eleven Player B participants would receive the relevant allocation. Participants had to answer all of these comprehension questions correctly before they were allowed to proceed and choose their preferred allocation. Finally, subjects answered a slightly revised set of debriefing questions, and a demographic survey.

As noted above, each of the eleven Player B Group participants was randomly assigned to one of the eleven possible allocations. The Player B Group participant who corresponded to the allocation chosen by Player A (whom we call here ‘active Player B’) was told about the experimental set-up and was also required to complete the set of comprehension questions. The active Player B was then informed of their predetermined position in the Player B Group and that their position matched Player A's allocation. All of the remaining participants of the Player B Group were simply given an unrelated survey to complete and paid the $0.50 participation fee.

The final task for Player As and active Player Bs was an exit survey, which was once again required for payment. The survey included slightly revised versions of the specific harm, generic harm, and fairness questions from experiment 1, and the same demographic questions.

2.2.2 Experiment 2: Results

On average, Player As transferred $0.155. A total of eighty-one Player As (45 percent) chose to keep the entire endowment for themselves, while only twenty-one (12 percent) opted for an equal division. Regarding the responses to the harm/fairness questions, 85 percent of Player As and 77 percent of active Player Bs responded negatively to the generic harm question. This indicates that most subjects do not think zero transfers in the nonidentity dictator game result in any harm. On the specific harm question, we found that 92 percent of Player As thought they did not do harm, while 84 percent of active Player Bs thought the transfer did not harm them. The latter is rather surprising, given that offers tended to be very low. On the fairness question, 68 percent of player As and 57 percent of active Player Bs took the transfer to be fair.

2.3 Comparison of the Experiments

We compare the behavior and attitudes of participants in the standard dictator game and those in the nonidentity dictator game. First, Player As were significantly more generous in the standard dictator game than Player As in the nonidentity version (see figure 2). In particular, many more Player As were willing to take an even split in the standard dictator game than they were in the nonidentity version, and many more Player As made exceptionally low transfers in the nonidentity dictator game. This clearly suggests that, at least in the aggregate, identity-affecting choice problems tend to generate more self-interested behavior on the part of the dictator. The differences between amounts given in the two studies scored as highly significant (with p < 0.001) on every statistical test we ran on the data. (For example, on the full dataset, for the Student's t-test p = 0.00043, for the Komogorov-Smirnov test p = 0.00027, and for the Wilcoxon rank sum test, p = 0.00027.) See figure 2 for the full histogram.

Figure 2

Histogram for giving: proportion of player A transfers in standard and nonidentity dictator games

Attitudes about harm were also substantially different between the two experiments. Both Player As and Player Bs were much more likely to register a harm in the standard dictator game than were their counterparts in the nonidentity dictator game despite the fact that offers in the nonidentity game were much lower than those made in the standard dictator game. As noted, 31 percent of Player Bs in the standard dictator game felt they were harmed by the specific transfer, but only 16 percent of active Player Bs in the nonidentity version felt harmed by their corresponding Player As’ choice. Likewise, substantially more Player As in the standard dictator game felt their behavior resulted in harm than did Player As in the nonidentity dictator game (19 percent and 8 percent, respectively).

Figures 3 and 4 provide a more fine-grained look at the harm attitudes of both Player A and Player B participants. In particular, for both experiments we list the proportion of Player A and Player B participants who believed that low transfers ($0.30 or less) resulted in harm. We find that both players in the nonidentity dictator game were much less likely to think that harm was done than did their counterparts in the standard version. This difference can be quite substantial.

Figure 3

Proportion of Player Bs receiving a low transfer who believed the transfer did them harm

Figure 4

Proportion of Player As giving a low transfer who believed the transfer did the recipient harm

All subjects in the standard dictator game were more likely than were subjects in the nonidentity dictator game to agree with the statement that a Player A participant who transfers $0 to her counterpart harms her counterpart (45 percent vs. 19 percent; Student's t-test: 7.72, p < 0.00001).

3. Practical Implications for the Nonidentity Problem

These studies were intended to examine the practical importance of the nonidentity problem. Recall that while various authors claim that the nonidentity problem is of great practical importance, this claim is slightly puzzling, because all parties in the debate seem to agree that the paradigm identity-affecting actions are intuitively just as immoral as their analogous same-person actions. Even authors who ultimately argue that such intuitions are mistaken also admit to sharing the intuition. If the public generally shares these intuitions of the philosophers, we should expect that the realization that they are in an identity-affecting choice problem would make little difference to their respective moral evaluations and actions. In other words, if they saw a certain choice as being immoral in the first place, that choice should still seem immoral once they realize it is an identity-affecting choice. There admittedly may still be grave practical roadblocks to generating the motivation to help future generations on this version of the story, but the nonidentity problem wouldn't raise any special concerns on that front. And even if the public's intuitions are slightly different from those of philosophers, the public might not generally be able to grasp that a choice is identity-affecting in the first place. Grasping the nonidentity problem, after all, requires some rather sophisticated reasoning. And if the public generally cannot grasp that they are in an identity-affecting choice problem, it is unlikely that the identity-affecting nature of the problem will cause any changes in actual behavior.

So what do the data from our study suggest about the practical implications of the nonidentity problem? We start with the latter point about the public's ability to track the identity-affecting nature of a choice problem. First, if most people's reasoning skills were inadequate, thus making them incapable of tracking whether they are in an identity-affecting choice problem, then we would expect it to be very difficult for participants to complete the comprehension questions in our nonidentity dictator game. While we did notice that participants in the nonidentity version failed more comprehension questions overall, we required only 32 percent more trials (622 for the standard dictator game versus 824 for nonidentity dictator game) to collect roughly the same number of data points (354 for the standard dictator game versus 352 for the nonidentity dictator game). And some of this increase in comprehension failures are likely due to the simple fact that we asked an additional comprehension question in the nonidentity dictator game. This suggests that while there probably is indeed some portion of the population that finds such details difficult to follow, it is certainly not a large enough proportion to support the claim that the general public is so naive that we should expect identity-affecting choices to make very little impact on their behavior. And the substantial portion who make it through the comprehension questions seem fully capable of seeing the normative difference in an identity-affecting choice problem. For example, the substantial drop in the proportion of those who think that a $0 offer harms the player given $0, which went down to 19 percent from 45 percent in the standard dictator game, suggests that these participants are able to track the moral intricacies of the choice problem. (That said, the experiment was designed only to make some of the nonidentity features salient to the participant, and we admit that those same individuals could fail to register these features when faced with real nonidentity problems.)

Do the data suggest that the identity-affecting nature of a choice problem has an effect on the public's behavior? We believe they clearly do. First, there was a rather pronounced drop in giving behavior, from a mean of $0.238 to $0.155. Although an $0.083 drop in giving might not seem so drastic in absolute terms, it is worth noting that this is a decrease in total giving of roughly one third. This was paired with a 13 percentage point increase in Player As who kept the whole amount and a 7 percentage point increase in those who kept most but not all of it (that is, 80 or 90 percent of the bonus). Even splits decreased by 9 percentage points (from 21 percent in the standard dictator game to 12 percent in the nonidentity dictator game). If we assume that moral pressure is what drives altruistic choices in dictator games—a common assumption in the literature—these results suggest that a substantial portion of the population does feel less moral pressure to be altruistic in identity-affecting choice problems.

We must admit that the kinds of effects we saw in our identity-affecting choice problem are merely suggestive of what we might see when dealing with large-scale identity-affecting problems like when a nation is deciding on climate change policy. Even beyond any worries one might have about how robust and reproducible our results will prove to be, there are a number of important features of climate policy choices that our scenario does not capture. First, there are severity differences: climate policy addresses a very serious problem, and, as such, the relevant outcomes are of a very different sort than small differences in earnings. In particular, the different choices will endow each respective future population with drastically different levels of welfare, and our choice scenario does not capture this kind of serious inequality. Second, there are existence differences: our identity-affecting scenario involves choices that differentially affect existing humans, as opposed to choices that change which individuals exist. Put another way, in the nonidentity dictator game there is a group of living people who are not benefited in the way they would have been if a different choice had been made; in a genuine nonidentity case the possible future people who are not benefited do not even exist. Third there are temporal differences: in our game the outcomes of identity-affecting choices are relatively immediate, whereas the outcomes of climate policy choices take generations to bear their identity-affecting results. And there may be shorter-term reasons we could appeal to in order to generate action on climate change, and this kind of alternative path to generating altruism is lacking in our study.

We believe that each of these differences might ground legitimate concerns, and we feel that all of them call out for further study. For example, regarding severity, one could perform a successor study that either raised the monetary stakes or one that introduced actual harms into the study (perhaps by using a paradigm like the one used by Crockett et al. Reference Crockett, Kurth-Nelson, Siegel, Dayan and Dolan2014), or both. For example, if the choice continuum ranged from a $1,000 gift to ten mild electrical shocks, we might see a much higher level of altruistic behavior in that version of an identity-affecting case. And although we, on balance, prefer our experimental paradigm over the use of surveys that employ vignettes, the latter could yield important insights into whether the severity or inequality of outcomes affects participants’ moral intuitions in identity-affecting cases. (For some insights along these lines in the medical context, see Doolabh et al. Reference Doolabh, Caviola, Savulescu, Selgelid and Wilkinson2019, which we encountered after submission.) Regarding existence concerns, it is much more difficult to imagine how our paradigm could be modified to capture more properly the parallels, so vignettes would be needed to probe those differences further. Our hunch is that the mere fact that some of the affected individuals do not actually exist is not really what is doing the work; the simple fact that those who miss out on the bonus do not ever know they missed out should be enough to get the analogy going. But this is admittedly an open question. And one author has been working on a more recent collaborative project that attempts to tease out whether a phenomenon referred to in the psychological literature as ‘compassion fade’ (see Butts et al. Reference Butts, Lunt, Freling and Gabriel2019) might be partly responsible for increases in selfish behavior in multiplayer dictator games (see Grodeck, Handfield, and Kopec Reference Grodeck, Handfield and Kopec2021). Finally, there are various ways one could adjust our paradigm to capture the temporal aspects better, for example by paying out the second players’ bonuses at later dates and by varying those durations. We predict these temporal changes would further decrease the level of altruism, as would be expected if participants are temporal discounters, but this is again an open question worthy of further study.

These caveats notwithstanding, we do think what we have uncovered offers some good initial reasons to be careful in how we choose to motivate actions aimed at future generations. The fact that our subjects in the identity-affecting choice problem were substantially more selfish suggests that the public may be less willing to make even small sacrifices, such as modest tax increases or slightly higher prices on certain products, if made aware of the identity-affecting nature of such policy choices. Furthermore, the fact that the increases in selfish behavior appear to correlate with a decreased perception of having harmed anyone gives us a reason to be cautious when we attempt to motivate the population to support conservationist policies by pointing to the harms that inaction could cause over very long timeframes. At least until future studies come along to debunk these possibilities, we feel caution is advised.

4. Implications for Normative Reasoning on Harm

When Parfit (Reference Parfit1984) set the stage, the nonidentity problem was deemed a problem because of the common-sense notion of harm that is often employed in our moral theorizing. According to this common-sense notion of harm, often referred to as the counterfactual comparative account, to harm someone is to make them worse off than they would have been otherwise. And in a nonidentity case, it is simply not true that you have made the relevant person or people worse off than they would have been otherwise. So, if this counterfactual comparative account of harm is correct, then there is no legitimate sense in which you have harmed the person in a nonidentity case. This gets the whole problem rolling. As noted above, one way to solve the nonidentity problem would thus be to devise a generally acceptable account of harm that entails that the affected individuals in nonidentity cases are actually being harmed after all. So various authors have seen the nonidentity problem predominately as a problem for the counterfactual comparative account of harm, and thus as a reason to seek out alternative accounts of harm (see, for example, Harman Reference Harman2004, Reference Harman, Roberts and Wasserman2009; Shiffrin Reference Shiffrin1999; Gardner Reference Gardner2015). Given the parallels, the choice problem we have generated in our game is well situated to offer some insight into which notions of harm members of the public are employing in their normative reasoning, at least at the moment of making their choice. Below we look at some of the competing accounts of harm that have been proposed in the literature and consider what the data we have collected could tell us about which notions of harm might be affecting the public's behavior.

We start with a precise statement of what those like Parfit take to be the common-sense account of harm:

The main competitor for the counterfactual comparative account in the literature is typically referred to as the non-comparative account of harm, which can be stated as follows:

To give an example of how these two accounts of harm can come apart, think back to the environmental policy example from earlier. If the United States adopts the Depletion policy, the US population in 2200 lives in relative squalor. On the counterfactual comparative account, the current US generation has not harmed the citizens of 2200 because it is not the case that these future individuals would have been better off if the Conservation policy had been adopted instead. But the non-comparative account has a very different verdict. By selecting the Depletion policy, the current US generation has caused the citizens of 2200 to be in an intrinsically bad state. So, on the non-comparative account, the 2025 US government did indeed harm the US population in 2200.

Although these are the two main accounts of harm in the literature, there is one other account that is worth mentioning. This account, which is often called the temporal comparative account, can be stated as follows:

Philosophers tend to discredit this account as a legitimate notion of harm because various supposed counterexamples are fairly easy to devise (see, for example, Hanser Reference Hanser2008; Thomson Reference Thomson2011; Shiffrin Reference Shiffrin1999; but see Foddy Reference Foddy2014 for a defense). But because we are interested in which notions of harm play a role in the normative reasoning of the public, philosophers’ counterexamples do not give us a reason to toss out the concept by fiat. It is now well established that the public uses various deeply problematic concepts in their normative thinking. So we treat the temporal comparative account as a live alternative.

So what can the data from our study tell us about which notions of harm the public employ? We start with our subjects’ attitudes about harm in the standard dictator game. In the standard version, there is no sense in which Player B is put into an intrinsically bad state when Player A takes all of the bonus money for herself. After all, Player B will still get the show up fee. So if the public predominantly employed a non-comparative notion of harm in their normative thinking, we would expect a very low percentage of our subjects to think that a $0 transfer harms Player B. We found quite the contrary. Because 45 percent of our subjects in the standard dictator game believed that a $0 transfer harmed Player B, this suggests that a rather substantial portion of the population employs something like the counterfactual comparative notion of harm in their normative thinking—at least when faced with certain kinds of choice problems.

Notice further that one way to frame what is going on when Player A takes the whole bonus for herself is that she is failing to benefit the Player B she was paired with. Many philosophers have a strong intuition that simply failing to benefit someone is very different from harming that person, and this intuition has been used as a motivation to reject the counterfactual comparative account, since it tends to conflate the two. For example, Ben Bradley (Reference Bradley2012: 397), who calls this the ‘problem involving omission’, rejects the counterfactual account on just these grounds (but see Feit [Reference Feit2019] for critique of Bradley's argument). It turns out that a substantial portion of the population does not seem to share this intuition, since they are happy to diagnose something that is a clear case of failing to benefit as an instance of harming. This, perhaps, suggests that we should proceed with some caution before rejecting the counterfactual comparative account of harm based solely on the problem of omission, although here we do not intend to take a stand on which notion of harm is the ‘correct’ one.

The substantial drop in harm perceptions when we move to the nonidentity dictator game further supports the claim that a substantial portion of the population employs a counterfactual comparative notion of harm. Recall that a much smaller proportion of Player As believe they have harmed the member of the Player B Group with their choice of transfer, even though the offers were on the whole much lower. And we also saw a sizable drop in the proportion of subjects that thought that a $0 transfer does harm. Both of these facts further support the claim that something like the counterfactual comparative notion of harm is playing a role in these subjects’ normative thinking, at least in their assessments of this particular problem.

But we think the data also suggest that an even larger portion of the population employs some other notion of harm when assessing these kinds of choices. For example, 55 percent of subjects in our standard dictator game did not think that Player A did harm by taking the whole bonus, which you would not expect if these subjects employed a counterfactual comparative notion. So, one thing the data from our study show is that there are substantial limits on what portion of the population could be using the various notions of harm in their normative thinking. For example, the data suggest that it is very unlikely that in future studies researchers will find that the vast majority of the population uses a non-comparative notion of harm in their normative thinking, and similarly for the counterfactual comparative notion. It is much more likely that we will find substantial heterogeneity in the population. It also would be fully consistent with our data if the notion of harm the public employs did not even play the role of a singular criterion concept in the first place. Perhaps harm, in the public's thinking, is a much messier kind of concept, one that gloms onto different features of cases depending on context. (We thank an anonymous referee for this journal for suggesting this possibility.) Either way, it is simply unlikely that we would find one of the singular criterion notions preferred by philosophers dominated the public's thinking.

Much like before, our study came with various limitations for probing the public's ideas of harm, and there are important unanswered questions that call out for further study. For example, even if the public did think of harm as a singular criterion concept, our data would not distinguish which of the alternative accounts various portions of the population employ. While it is true that a Player B who is given no bonus is not caused to be in an intrinsically bad state, it is also true that she is not made worse off than she was before she was given no bonus. After all, she started out with no bonus. Thus, this portion of our subjects could just as well be utilizing a temporal comparative notion of harm as a non-comparative one.Footnote ² In future work, we plan to ask questions to help us to tease apart which of these competitors this substantial portion of the population is using. Additionally, a nonnegligible portion of our subjects in the nonidentity dictator game still thought that a Player A who keeps the whole bonus harms Player B0 by making that choice (19 percent). We would not really expect this on any of the notions of harm covered here (including Boonin's contender mentioned in the previous footnote). It could very well be that there is a notion of harm that is playing a role in the public's normative thinking that philosophers have thus far completely ignored. And, as noted above, intuitive instances of harm might be triggered by a multiplicity of different considerations depending on the specific context the individual encounters. Further study is needed to tease all of this apart.