Individual contributions to collective harm: how important is causation?

ABSTRACT In the last chapter of Responding to Global Poverty, Barry and Øverland argue that there are moral reasons against overdetermining harm. They define overdetermining harm as conduct that makes no apparent difference to the occurrence of harm but is of the type that brings about harm when many people engage in it. An individual’s greenhouse gas emissions are a prime example. Barry and Øverland’s proposal is that reasons to refrain from overdetermining conduct exist because of the probability that the agent will become an element of the set of actual conditions that in fact brings about the outcome. This paper aims to show that by focusing on causation-related considerations, Barry and Øverland base their account on the wrong reasons. More specifically, I argue that this focus leads to three difficulties. First, the account is not able to justify overdetermination-based constraints in all overdetermined harm cases. Second, the probability of being in the set may be too easily outweighed by the costs of refraining and the benefits the conduct would bring. Third, the probability of being in the actual set may not be highly morally relevant, given that an individual’s contribution to the harm caused by the set can only be very limited in large-scaled overdetermined harm cases. Barry and Øverland are right in arguing that it is important to undermine scepticism about the wrongness of overdetermining harm. However, a non-causation-based account might be more successful in doing so.


Introduction 1
Let's start with Christian Barry and Gerhard Øverland's story about Robinson and Tom. Robinson is living relatively well on a small island in a lake, until Tom and fifty others shovel excess waste from their gardens into the lake, making water levels rise and eventually causing Robinson to drown. One version of this story, in which the situation is transparent to Tom and others, goes as follows: Robinson 2: Tom's disposing of his waste makes no apparent difference to the drowning of Robinson. Fifty others shovel waste into the lake and that number, or less than that number, is enough to drown Robinson. If Tom were to have abstained, the island would still have been flooded, and Robinson would still have drowned. (Barry and Øverland 2016, 226, italics in original) CONTACT Anne G. Polkamp anne.polkamp@kuleuven.be 1 An earlier version of this paper was presented at the workshop Responding to Global Poverty: Harm, Responsibility and Agency at KU Leuven, 7-8 June 2016. I thank all participants and especially Christian Barry for their helpful comments. I am also grateful to Holly Lawford-Smith, RJ Leland and two anonymous referees for their valuable feedback on earlier versions of this paper.
In Barry and Øverland's words, Tom is an overdeterminer of the harm. This means he is an agent whose conduct 'makes no apparent difference to the occurrence of harm, but is of a type that brings about harm because many people engage in it' (221). 2 They use Tom's example because his overdetermining conduct mirrors the way in which affluent individuals might be claimed to contribute to poverty. Individual contributions to collectively caused harms raise the question whether there are moral reasons against overdetermining harm (which Barry and Øverland dub 'overdetermination-based constraints') and, if so, what their basis is. The sceptic claims that overdetermination-based constraints do not exist, because if one's action makes no difference at all, there is nothing wrong with performing it (Barry and Øverland 2016, 227). Barry and Øverland argue that this is a powerful position (229), but that it is counterintuitive in various ways. For example, it conflicts with the intuitions that, all else equal, Tom did something wrong when he shovelled his waste into the lake and that force may be used to stop Tom and the fifty others in order to save Robinson (228). Barry and Øverland's aim is therefore to blunt the appeal of scepticism by providing a positive account of overdetermination-based constraints (229). In short, their proposal is that these constraints 'are based on the possibility that some agent will become an element of the set of actual conditions that in fact brings about the overdetermined outcome' (237, emphasis in original).
For reasons I will turn to later, this proposal is more successful in justifying overdetermination-based constraints than some of its alternatives. Moreover, I agree that it's important to undermine the appeal of scepticism about the wrongness of overdetermination. After all, there are many situations in which people together cause harm and yet no one appears to make a difference. It seems problematic to say that no one does anything wrong in those cases. However, my worry is that Barry and Øverland's focus on causation creates several difficulties. These problems lead me to believe that an account of overdetermination-based constraints that is not based on causation might be more successful in undermining scepticism.
Below, I will first elaborate on the details of Barry and Øverland's account. I will then compare it to Shelly Kagan's and explain why Barry and Øverland's version has some important advantages. Despite these merits, I worry that both accounts turn on causation and therefore face similar difficulties, which I discuss in the third section. The last section focuses on the problem that it is unclear why exactly being in the actual set is wrong in large-scale overdetermination cases.

Barry and Øverland's argument for overdetermination-based constraints
While there are various forms of overdetermining conduct, including cases in which the type of conduct is beneficent, Barry and Øverland limit their discussion to cases in which the type of conduct does harm, 3 using climate change as a running example. After evaluating various 2 Note that Barry and Øverland's use of the term 'overdeterminer' should not be confused with the interpretation of this term in the literature on causation. In that literature, an event is overdetermined when there is more than one antecedent cause, any of which would be a sufficient condition for the event to occur. In the cases Barry and Øverland discuss, the antecedent causes are not by themselves sufficient for the event to occur. 3 They note that straightforward cases of doing harm typically have two characteristics. The first is what they call relevant action: in clear-cut cases of doing harm, there is an answer to the question how an agent's action was relevant to the harm. The second is a complete causal process linking the relevant action to the harm (Barry and Øverland 2016, 224).
existing proposals for reasons not to overdetermine harm, Barry and Øverland go on to propose an alternative approach. They start by questioning whether overdetermination even exists. The authors observe that it sure seems like there are many situations in which an individual's conduct overdetermines harm, but argue that perhaps this is only because we do not have a 'detailed microanalysis of the situation and how it has come about' (Barry and Øverland 2016, 235). After all, harms can be described in various ways. They can be described generally -Robinson drownsor more specifically -Robinson drowns at a particular time and in a particular way (235). According to Barry and Øverland, 'it is very often the case that a person is not an overdeterminer with respect to an outcome described in a more finegrained fashion' (235). 4 They illustrate this claim by showing how Tom is not an overdeterminer with respect to Robinson's death when it is described in a fine-grained fashion. Suppose that forty piles of waste suffice to flood the island. There are two possible scenarios: in the first, fifty-one people throw in their waste at different times; in the second, they do so simultaneously. In either case, there are no overdeterminers. In the first scenario, the first forty together kill Robinson by flooding the island, whereas the other eleven don't contribute.
In the second scenario, in which the fifty-one dispose of their waste simultaneously, each of them likely contributes to the drowning. There would be intact causal sequences between each person's conduct and the harm: a part of each load of waste would be involved in raising the water level. In this case there are no overdeterminers either, because if any one person would have refrained from waste-throwing, Robinson would have died in a slightly different way (236-237). Barry and Øverland conclude: 'A fine-grained analysis seems to us to provide the correct diagnosis of who contributed to Robinson's death' (237). However, they claim that '[l]earning whether or not Tom actually did or did not contribute to Robinson's death as described in this fine-grained way doesn't really explain why it is wrong for him to dispose of his waste in the first place' (Barry and Øverland 2016, 237). This is where their account of overdetermination-based constraints comes in. On Barry and Øverland's view, these constraints 'are based on the possibility that some agent will become an element of the set of actual conditions that in fact brings about the overdetermined outcome' (237, emphasis in original). If an agent is an element of the set of actual conditions causing harm, she is a contributor rather than an overdeterminer. Therefore, the argument seems to be that agents have overdetermination-based constraints because of the possibility that they will turn out not to be overdeterminers (of the harm described in a fine-grained way). The authors add that the significance of the overdetermination-based constraint depends on both the probability of being in the actual set and the badness of the outcome. An overall appraisal of the conduct also depends on the costs of refraining from performing the act, on the agent's knowledge of the situation and on the benefits the conduct may bring to others (237).

Kagan versus Barry and Øverland
Although there are some important differences, Barry and Øverland's account of overdetermination-based constraints bears similarity with Kagan's (2011), whose account is one of the alternatives they evaluate. It is worth discussing these resemblances, because I believe they lead to similar difficulties. Kagan develops a consequentialist argument for refraining from overdetermining harm. On his view, cases comparable to Robinson's drowning are in fact triggering cases in which most individual acts do not make a difference, until one act triggers more harm, like the straw that breaks the camel's back (Kagan 2011, 118). To illustrate his argument, he gives the following example. Suppose that most individual acts of chicken purchasing do not make a difference to the number of chickens being killed, but that it is likely that for every n number of chickens bought at a store, the store orders n more chickens from the factory. For example, suppose that for every 25 th purchase per day, 25 more chickens are ordered and killed. In that case, none of the first 24 purchases make a difference, but every 25 th purchase triggers more harm (121-124). Moreover, if exactly 25 (or 50, or 75…) chickens are bought on a certain day, it is not only the 25 th purchase that makes a difference but everyone's purchase. While the 25 th , 50 th and 75 th purchases are the ones that trigger another order, any one of the chicken buyers could have prevented the last trigger from taking place. For if any one of the buyers would have refrained, there would have been only 74 purchases and 25 fewer chickens would have been ordered and killed. In the end, all of the buyers thus make a difference. The question is therefore whether an agent's purchase is part of a cohort of an exact multiple of 25 purchases (126). If it is, then it makes a difference to the outcome. When the chance of making a difference outweighs the benefits of the act, i.e., when the expected utility of the act is negative, Kagan condemns the act (127).
In short, Kagan's overdetermination-based constraint is based on the probability that the act is part of a set of exactly the number of acts necessary to trigger more harm to occur. There are some similarities between this account and that of Barry and Øverland. Both accounts base the constraint on the probability that the act will be involved in causing harm: Kagan focuses on the probability that the agent is part of an exact set of conditions causing harm; Barry and Øverland focus on the probability that the agent is part of the actual set of conditions causing harm. Moreover, on both views the moral appraisal of the act depends not only on this probability but also on considerations like the costs of refraining and the benefits produced by the conduct. 5 A crucial difference is the fact that Kagan's proposal turns on the probability of the agent making a difference to the harm described in a general way, whereas Barry and Øverland focus on the probability of the agent being part of the actual set that brings about the outcome, described in a fine-grained way. In other words, the question whether agents (might) make a difference to the harm suffered by anyone, is irrelevant for Barry and Øverland. The advantage is that their account is applicable to a wider range of cases than Kagan's. As Barry and Øverland (2016, 234) point out: while Kagan's proposal may justify overdetermination-based constraints in cases where agents believe they may make a difference, it cannot explain why agents should refrain from overdetermining conduct when they know they won't make a difference to the outcome. Kagan concedes this point. He illustrates it with a hypothetical example where the chicken-buyers are omniscient with respect to the chicken business and they are certain that more than an exact multiple of 25 chickens are bought on a given day, say 67. In that case, no individual makes a difference because if anyone would have refrained, 50 chickens would still have been ordered and killed that day. (Note that the 25 th and 50 th buyer do not make a difference either, even though it is their purchase that triggers another order. If they would have refrained, someone else's purchase would have been the trigger and 50 chickens would still have died.) Given that the agents know this, there are no reasons to refrain (Kagan 2011, 128). Barry and Øverland's proposal, on the contrary, can still justify overdetermination-based constraints in this case, for there is still a chance that an agent's purchase will be part of the actual set causing harm, even if being part of that set does not make a difference to the number of chickens being killed.

Limitations in applicability and stringency
While Barry and Øverland's account is applicable to a larger range of cases than Kagan's, it nevertheless seems to be limited in a similar way. That is because both accounts focus on the causal role of the relevant conduct: the probability of being part of a set that causes harm. Because Barry and Øverland's constraint turns on such a probability, they note that '[i]f there is no risk that by φ-ing the agent will be among the actual set of conditions that in fact does the harm, then there are no overdetermination-based constraints against his φ-ing', although they add that the behaviour may be wrong for some other reason (2016, 238). Like Kagan, then, Barry and Øverland succeed in justifying overdetermination-based constraints in cases where agents believe they may be part of the set, but not in cases where they know they won't be. For example, in Kagan's case of the 67 omniscient chicken-buyers, Barry and Øverland could only explain why the first 50 have reason not to buy a chicken. The next 17 buyers have no chance of being among the actual set causing harmand know thisso they do not have an overdetermination-based constraint. Alternatively, if only 24 chickens were sold, none of the chicken-buyers would have an overdeterminationbased constraint. Barry and Øverland do not seem to consider these limitations in applicability problematic, which appears to conflict with their criticism of Kagan. Their critique is that it is hard to see how consequentialists can respond to cases where the number of chicken-buyers is not a multiple of 25. In those cases, consequentialists can only base their reason to refrain from chicken-buying on the fact that we often don't know whether we will be part of a cohort of 25 buyers (234). But similarly, it is hard to see how Barry and Øverland can respond to certain cases, like the one in which only 24 chickens are bought. They, too, can only base their reason to refrain on the fact that we often don't know whether we will be part of the actual set causing the harm.
Barry and Øverland might reply that this limitation in applicability is not problematic for their account in the way that it is for Kagan's. They could emphasize that their account works in a larger range of cases, as mentioned above. Moreover, they could argue that a plausible account of overdetermination-based constraints does not need to apply to agents who know they won't be in the set, since they know that their actions will not contribute to any further harm. This reply would mean that Barry and Øverland can only partially respond to cases where agents know they will not be in the set. Moreover, the reply conflicts with the intuition that there is something intrinsically wrong with certain overdetermining conduct, such as buying factory-farmed meat or voting for evil politicians. It seems that there is always a pro tanto reason not to perform these actions, even for people who know they will not be in the actual set causing harm.
A second similarity between Kagan's account and that of Barry and Øverland limits the stringency of their overdetermination-based constraint. Recall that Kagan weighs the probability that an agent's conduct will make a difference against the benefits the conduct will produce. The overall appraisal of the conduct therefore depends on the magnitude of the harms and benefits involved and on the probabilities that those harms and benefits will materialize. Barry and Øverland (2016, 234) criticise Kagan because 'there may be many cases where the chance of making a morally relevant difference by φ-ing is extremely small, whereas the expected value of φ-ing is quite significant. So it is not true that appeals to expected utility will "guarantee" that it is wrong to perform the overdetermining act'. 6 Instead, the likeliness of having reason to refrain from overdetermining conduct 'seems to depend largely on the case that is being considered' (234). However, the same seems to be true of Barry and Øverland's proposal, for they take into account similar outcome-related factors. As mentioned before, they hold that the overall moral appraisal of an agent's conduct does not depend solely on the probability of being part of the actual set. It also hinges on an agent's knowledge of the situation, the costs of refraining from the conduct and the benefits the conduct may produce for others (237). The lower the probability of being part of the set, the weaker the overdetermination-based constraint and the more easily it will be outweighed by the costs of refraining and the benefits of performing the act. If the constraint is outweighed, the overall appraisal of the conduct could be positive. How often that would happen depends on the harms, benefits and probabilities involved. So on Barry and Øverland's account, too, the moral appraisal of overdetermining conduct depends on the case being considered.
In response, Barry and Øverland might argue that it is plausible that overdetermination-based constraints can be outweighed by other factors and that a view guaranteeing that overdetermining harm is wrong, would be too strong. This is shown by one of their own examples, where Tom needs to shovel waste into the lake in order to save his own life. Surely, in this case, it would be permissible for him to do so (Barry and Øverland 2016, 242). While this is correct, the reply misinterprets my objection. I do not expect Barry and Øverland to guarantee that overdetermining harm is wrong, for it is plausible indeed that the constraint can be outweighed. My worry is rather that the overdetermination-based constraint is outweighed too easily. After all, many overdetermining actions may produce relatively large benefits, whereas the probability of being in the actual set may carry relatively little weight. This concern brings me to the following, more fundamental worry.
The moral relevance of being part of the set I wonder whether Barry and Øverland's causation-based approach pinpoints what's most problematic about overdetermining harm. In fact, the question arises why being part of the actual set that causes harm is wrong at all. If the actions of a set of people together cause harm, it seems plausible to say that the set has acted wrongly. It does not follow, however, that it is wrong for an individual to act in a way that makes her part of the set. As Walter Sinnott-Armstrong argues: 'It begs the question here merely to assume that, if it is bad for everyone in a group to perform acts of a kind, then it is morally wrong for an individual to perform an act of that kind '. 7 In what follows, I will discuss two possible responses Barry and Øverland could give and argue that both fail to explain why exactly it is wrong to be in the actual set.
The first response is that the wrongness of being part of the actual set simply follows from the harm principle. Yet an appeal to the harm principle does not suffice, because being part of a group that causes harm does not equal causing harm. To return to the example of Robinson: even if Tom is in the actual set, he does not really harm Robinsonnot if we use a counterfactual account of what it means to cause harm, at least. After all, the harm would still have occurred had Tom refrained from participating (recall that Robinson dies no matter what Tom does). It is therefore not accurate to say that Tom caused Robinson's death; he merely overdetermined it. An appeal to the harm principle therefore fails to explain what is wrong with being part of the actual set that causes harm.
Barry and Øverland might object that, although being in the set does not equal causing harm, it does constitute making a difference to the fine-grained description of the harm. Tom's conduct, for example, changes the circumstances in which Robinson dies. In other words: while Tom is not in a position to prevent Robinson from drowning, he is in a position to avoid Robinson drowning in a certain way and at a certain time. It might be that Robinson will die minutes earlier if Tom throws his waste into the lakea morally relevant difference and a good reason for Tom to refrain. This leads to the second response Barry and Øverland could give: being part of the actual set is wrong because it constitutes making differences to the fine-grained description of the harm.
While this is a more promising response, my worry is that it is unclear how morally problematic this kind of difference-making is in real-world overdetermined harm cases. In large-scale cases such as climate change, the difference any individual can make is likely much more minuscule than the difference Tom can make. The difference may even be so insignificant that it fails to explain what is wrong with being part of the actual set.
Take the following example. Suppose in her spare time Mary likes to go for joyrides in the desert next to Robinson's lake, using her gas-guzzling SUV. 8 Her greenhouse gas emissions make her an overdeterminer of the harm caused by climate change. The question is what Mary's contribution to this harm could be, supposing that she ends up in an actual set that causes harm. This is an empirical question we cannot answer with any certainty, but there is reason to believe Mary's contribution will likely be very small.
For Mary to make a difference at all, we must make the controversial assumption that there is an intact causal chain between the joyride and some harm, such that if Mary would refrain, the climate change damage would have been slightly less bad. If there is no intact causal sequence, the joyride cannot cause any harm to begin with. 9 So let me assumefor the sake of argumentthat any increase in greenhouse gases raises global temperatures a tiny bit, thus slightly exacerbating certain meteorological events, which is what John Nolt argues (2013;cf. Lawford-Smith 2016). Perhaps, like in an example of his, the joyride's greenhouse gases will affect a drought and make temperatures an infinitesimal fraction of a degree higher, as a result of which countless people will be a tiny bit thirstier (Nolt 2013, 113).
Under these assumptions, Mary's joyride indeed makes a difference to the finegrained description of the harm. Yet there are many differences between this example and that of Robinson: the joyride does not contribute to the suffering of one person, but to that of many; it does not contribute to a death but to a gradual scale of suffering; and it does not cause a perceptible but an imperceptible difference in suffering. The difference Mary makes by joyriding is arguably much more minuscule than the difference Tom might make by shovelling waste into the lake.
There is another reason why the difference Mary makes is likely much tinier than the difference Tom makes. The story above assumes a simplified causal chain where the emissions from the joyride are active at all levels from raising the temperature to exacerbating one specific drought. Given the complexity of the climate system, it seems more likely that such a straightforward causal chain does not exist. Instead, the greenhouse gas molecules from the joyride might have multiple destinations, in which case the effects from the joyride would be scattered even more. This would make the differences in harm caused by the joyride tinier still, perhaps to the point where they don't even seem to be worth mentioning. 10 The argument that being in the actual set is wrong because it constitutes making a difference to the fine-grained description of the harm, is therefore not fully convincing. This kind of difference-making may be significant in cases like Tom's, but in realworld collective harm cases such as climate change, the difference an individual can make by being part of the set may be too small to be morally relevant. At the very least, the claim that this kind of difference-making is wrong is not as self-evident as the claim that causing perceptible harm is wrong. If they are to fully justify their overdetermination-based constraint, Barry and Øverland need a more elaborate storyone that explains why there is something wrong with causing imperceptible differences in harm, or one that gives us another reason why being in the actual set is wrong. For in largescale overdetermined harm cases it is not clear why the possibility of being in the actual set is in itself a sufficient reason for refraining from overdetermining conduct.

Conclusion
Barry and Øverland rightly emphasize the importance of overcoming scepticism about the wrongness of overdetermining harm. Their proposal goes a long way in showing that overdetermining conduct often does contribute to harm, even when it seems like it doesn't. However, I worry that their focus on causation limits the applicability and stringency of their proposal. Moreover, it seems unclear why exactly being part of the actual set that causes harm is problematic in large-scale overdetermination cases.
10 A critic might argue that individuals can make significant differences. For example, Nolt (2011) calculates that the average American is responsible for making one or two future people suffer or die. His calculation is based on the expected number of climate change victims and the average American's share in global emissions. However, as Ronald Sandler (2011) notes, the calculation shows averages rather than the actual harm caused (i.e., the harm that would have been prevented had the person in question not emitted any greenhouse gases).
A critic might insist on the importance of a causation-based approach because of the primacy and strength of negative obligations and because in certain overdetermined harm cases, like climate change, causal influence on the outcome seems to be all that matters. Yet given the insignificance of individual contributions to large-scale collective harm cases, the causal role of individual actions may carry little weight. If so, attempts to overcome scepticism might be more successful when they are based on factors other than causation.

Disclosure statement
No potential conflict of interest was reported by the author.