My aim in this chapter is to investigate what justice requires of agents who find themselves in situations that have the following general form. There are many agents who by acting together can avert some anticipated harm. Together, then, they share responsibility for avoiding that harm, but furthermore this collective responsibility can be divided fairly between them so that each knows what he or she must do to discharge that fair share. Despite this, however, some comply but others do not; the ones who do not comply could have chosen to do so, but they have shirked their share of the responsibility, and in doing so have acted unfairly. The question then arises: How should we understand the position of those who have already complied or those who have yet to decide what to do? What are their responsibilities now, in the face of partial compliance? Must they take up the slack, by doing more than they were required to do according to the original fair division of responsibility? What does justice demand in this situation?

Situations of this general kind are not merely hypothetical: one can readily think of a number of real‐world circumstances that either already do or easily could take the form just described. For instance, we might think of rescue cases where there are numbers of people awaiting rescue and several potential rescuers and it is reasonably clear how much fairness requires each to contribute to the operation. Or we might think of world poverty and the position of all those who could contribute to poverty relief by (p.231) voluntary contributions to aid agencies.² Or again if we switch focus from individual agents to collective agents such as nation states, we could think of measures to conserve natural resources, such as agreements to reduce the catch of ocean fish, where each country is given a target quota such that if all countries comply, fish stocks will not be further depleted. Finally we could consider the case of global warming, and think of an agreement to cut greenhouse gas emissions with each nation being given a target to aim at, targets being set to reflect some underlying principle of fairness. In all of these cases, it is unfortunately only too easy to anticipate that because of bad faith, selfishness, weakness of will, or whatever, some agents will find reasons not to do their share, so that the goal in question—rescuing everyone at risk, keeping fish stocks at the required level, and so forth—will not be reached unless other agents take up the slack. But are they obliged, as a matter of justice, to do this?

I think that our immediate intuitions are likely to suggest different answers in different cases. If we take rescue cases, for instance, then assuming it was a matter of justice to do one's fair share in the first place—something we will need to investigate as we proceed, but assume for the moment that if the potential victims had a right to be rescued, and the costs involved in the rescue were not excessive or disproportionate to the harm averted, then justice requires that it be undertaken—many people would say that justice also requires taking up the slack if the costs stay within the same limits. If justice requires you to jump into the pond, spoiling your suit but not risking your life to pull out the first child, then if your companion declines to pull out the second child, you must at least be willing to sacrifice another suit of comparable value to you to rescue that one. On the other hand, in the case of conserving fish stocks, many would think that a nation that conscientiously sticks to its quotas has done all that justice requires, and that reducing catches still further because others have failed in their duty would be supererogation. As long as we have done our fair share, they might say, the decline in fish stocks is entirely the responsibility of the countries that haven't. But what can explain these conflicting intuitions, apart from the rather too obvious fact that one case has to do with human beings and the other with fish? Or can we restore consistency by showing that one or other intuition is mistaken?

Before I try to tackle the normative issue directly, let me expand a little on how the situations that interest me are to be characterized. First, although they are multi‐agent cases, they are not standard prisoner's dilemmas, nor are they what we might call altruistic prisoner's dilemmas. By an altruistic prisoner's dilemma I mean a situation in which there is some good to be achieved or harm to be avoided but it is indeterminate which agent or agents should take action to bring that result about. So one might think of a case where there are several people standing near the pond when the child falls in, (p.232) and only one person needs to jump into the pond to save him. Assuming each bystander wants the child to be saved, and would jump in if he were the only rescuer at the scene, but is reluctant to sacrifice his suit, we have a dilemma where each has a reason to hold back in the hope that someone else will carry out the rescue.³ Or in a variant of this case, a car accident may require only a sub‐set of those who witness the accident to care for the injured, but it is indeterminate who should make up that sub‐set. The situations I am investigating are different from this, because by stipulation it is clear what justice requires each person to do by way of contribution. This might be because a formal agreement has been reached, on the basis of criteria that each party is willing to accept, as in the case of the fish quotas. Or it might simply be because it is evident to everyone what fairness demands in the circumstances—for example an equal sacrifice from each agent. This does not of course prevent people from acting strategically—deciding not to contribute in the hope that others will take up the slack. But it will be clear in these situations that such people are behaving unfairly, and that prima facie at least those who decide to take up the slack are doing more than justice requires.

I need to make two further clarificatory comments here. I assume that for each agent, discharging his or her fair share of responsibility is significantly costly, but not so costly that it becomes questionable whether it could be a matter of justice to contribute. Furthermore this also applies to the additional costs involved in taking up the slack—they are significant, but not so significant as to push the agent across the threshold such that it would be beyond the call of duty to contribute more. In other words, I am not going to be addressing the problem of demandingness that has occupied the attention of many philosophers—the problem of when carrying out a duty becomes so onerous that the agent in question can justifiably refuse to discharge it.⁴ My interest is in questions of fairness—in how one should think about cases in which one is asked to bear additional costs only because others have failed to carry out their share of a collective responsibility. If there were no significant costs involved, this would not be an interesting problem—it would merely be pique to refuse to flick an extra switch to avoid some great harm merely because it was somebody else's job to flick that particular switch. On the other hand, when the cost becomes very high, fairness may be obliterated by the demandingness issue—that's to say, the question for each agent may simply be ‘how much must I contribute to this collective task before it becomes so onerous that my personal prerogative cuts in?’ I think that the examples (p.233) I used to illustrate the fairness problem—standard rescue cases, world poverty, conserving natural resources, and combating global warming—all fall into the middle ground where the costs that agents are being asked to bear, both initially (given a fair distribution of responsibility) and as a result of non‐compliance, are significant but not excessive.

One last introductory remark: someone might respond to the situations I have described by saying that what is needed is some mechanism that can oblige people to contribute their fair share, so that the question of taking up the slack need never arise. For example, we should create institutions that can supervise the behaviour of individual people, or of countries in policy areas such as resource conservation and climate change. In some cases this will be the correct response. However, I am confident that there will be other cases—many other cases, unfortunately—in which even if setting up such a mechanism were justifiable in principle, it would not be feasible. In rescue cases, for instance, you cannot in the end force human beings to perform actions they are unwilling to perform—they may simply sit on their hands and refuse to help. In the country‐level cases, no international institution is likely in the foreseeable future to have either the authority or the coercive power to oblige unwilling countries to cut their greenhouse gas emissions or to conserve fish stocks or other resources (what such an institution can do, importantly, is to define fair shares of responsibility by setting targets for each country, but this is very different from enforcing the targets that have been set). So the problem of partial compliance as I have described it is a real one—it cannot be circumvented by ingenuity in designing institutions or other mechanisms that could guarantee full compliance.

In the situations I am considering, agents face a choice between three broad possibilities: they can discharge their original fair share of responsibility despite others' non‐compliance; they can do more than they were originally required to do in order to compensate for the non‐compliance; or they can do less than they were originally required to do, so that they fall more closely into line with the defaulters. Each of these options has something to be said for it, so let me elaborate a little further. The argument for the first option is that one's responsibility is only to do one's fair share regardless of what others decide. Liam Murphy, who on the whole supports this position in his book Moral Demands in Nonideal Theory, notes that a stricter version of the first option would require people in conditions of partial compliance to perform the best available action subject to the condition that the overall costs of doing so would not be greater than the costs they would have borne under full compliance.⁵ This accommodates the fact that the world will be different in certain ways when there is only partial compliance, so what one actually has to do to discharge one's fair share of responsibility may be different than under full compliance, but the key point is that the net level of sacrifice should be the same. And, it might be said, whatever the arguments for doing more, this is all that justice specifically requires of the agent.

(p.234) In defence of doing more, it can be argued that because of the non‐compliance of the other agents, we are now in a position where the harm we were seeking to avert will not be averted. However much we might regret the non‐compliance and condemn the non‐compliers, this is simply a fact of life, so what we must now do is to recalculate contributions, distributing responsibilities fairly among the coalition of the willing, so to speak. What justice requires in the new circumstances is that each agent should discharge his fair share of the responsibility as determined by these calculations. Unless he is the only willing complier, nobody is required to take up all of the slack by himself, but must contribute fairly to filling the gap left by the non‐compliers. Since presumably there may be some agents who are willing to comply with the original distribution of responsibilities, but not to contribute to taking up the slack, there might have to be further iterations in which additional assignments of responsibilities are made, but to keep things simple let's assume that everyone who was originally compliant is also willing to help take up the slack. Our question will be whether this might be what justice requires of such agents.

The third possibility is that one might do less, in the face of others' non‐compliance (I shall call this ‘grouching’). What would doing less mean? One possibility is that one should try to place oneself at the average level of compliance revealed by the actions of others. This is unlikely to lead to a determinate result in the case of rescues—one can't pull half or two‐thirds of a child out of the water—but in the other examples I gave it would certainly be feasible to adjust one's level of contribution to the going rate, so that if others are on average contributing one‐third of what they would need to give to abolish global poverty, one would also give one‐third of whatever one's full contribution was calculated to be. In defence of this policy, a person could say that he was behaving fairly given other people's actual behaviour—if they were prepared to do more he would do so as well, but justice should not require us to contribute more to a collective task than others are on average contributing. After all, to contribute the full amount in these circumstances would put the person in question at an unfair disadvantage relative to these others.⁶ Of course, somebody might take this line of thinking further and refuse to contribute more than the least compliant of the other parties, but again to keep things simple let us have compliance at the average level on the table as our third alternative.

Having sketched these three possible responses to situations of partial compliance at a general level, let me now consider some of the factors that might affect our judgement about which response is required by justice. These in other words are factors that potentially might separate cases in the real world such as those I described earlier.⁷ (p.235)

1. 1. The first factor that might be relevant is the causal relationship between individual contributions and the result that is being sought. In other words, how far do successive contributions by different agents affect the overall outcome? One possibility is that each act of compliance has the same net effect, so that the relationship between level of compliance and outcome is linear. For example, if we take the case of conserving fish stocks, each reduction of the catch, measured in weight, say, might have more or less the same effect on the size of the long‐term stock of fish in the sea. Another possibility, at the other end of the spectrum, is that each contribution taken by itself has zero effect unless all the contributions are made—one might think of a rescue that involves a human chain of rescuers such that if any one of them declines to participate the chain will not reach as far as the victims. In between there will be cases where early contributions count for more than late ones, and others where later ones count for more than early ones (early and late referring here to the position of a contribution in the sequence of contributions, not to the passage of time as such). In cases such as global warming, for example, there may be thresholds which it is important not to cross in terms of atmospheric concentrations of greenhouse gases, in which case emission cuts that would keep the world just below such a threshold would be more significant than others further below or beyond the threshold that might be desirable but not so critical.

   The relevance of this factor to our problem is relatively straightforward. Before we can decide whether justice requires us to take up the slack, or conversely permits us to reduce our contribution to the average, we should ask what the actual effect of doing so would be, in either case. In particular, would making the agreed contribution, or taking up the slack, mean crossing some significant threshold in terms of its effect? Or at the other extreme, would the contribution in either case simply be useless, given what others can be expected to do?

2. 2. The second factor that might be relevant is whether the collective responsibility that our group of agents bear is itself a responsibility imposed by justice. Do they, in other words, owe a collective duty of justice to those who will suffer if the prospective harm is not averted? This raises the question of whether there can be such duties in situations that have the shape I am considering. I shall address that in a moment. But assuming a positive answer, it seems clearly relevant to ask whether the harm that would be averted by successful collective action is a harm that involves injustice, or merely some other kind of loss, as arguably at least (p.236) would be the loss of fish in the sea. Where the collective responsibility we are examining is a responsibility imposed by justice, two issues of justice will arise with respect to individual compliance or non‐compliance: first, justice within the collective (is each agent contributing their fair share or not?); second, justice between the collective and those to whom it is owed. Someone who contributes but refuses to take up the slack might defend herself by pointing out that she is doing her fair share and that to do more would put her at an unfair disadvantage relative to others (indeed at a double disadvantage relative to the non‐compliers). But it might be said in reply that by refusing to take up the slack she is contributing to injustice against the victims, and this is a more serious form of injustice. Whatever we conclude about this, no such issue appears to arise in cases where the harm to be averted is simply a loss that does not involve injustice.⁸

   So could there be a collective duty of justice towards those who will suffer the results if collective action does not succeed, or only succeeds partially, in cases of the kind I am considering? To keep things simple, let us focus on cases where the prospective victims have a right to whatever outcome the collective action would produce, and therefore, apparently, can demand it as a matter of justice.⁹ This seems plainly true in the case of rescues—people have a right to life and bodily integrity—and in the case of world poverty—people have a right to a minimally decent standard of living. It may also well be true in the case of global warming, where it has plausibly been argued that among the results of unchecked global warming will be large‐scale violations of the human rights of those who are unable to cope with rising sea levels, droughts, and so forth.¹⁰ So on one side we have demands that are potentially demands of justice. But we have now to identify agents to whom the corresponding obligations can be attached. In standard cases, justice involves a claim made by one person against another, or by a person against an institution such as the state. In the present case, the obligation must attach to the whole group of potential contributors, who do (p.237) not together constitute an agent in the same sense: there is no collective decision to contribute or not to contribute, but rather each individual member decides whether to contribute his or her fair share, or more, or less. What we do have, on the other hand, is an agreed assignment of responsibility within the collective: by stipulation there is no dispute over what the fair share of each member amounts to. So the collective obligation to protect the rights of the potential victims does not simply hang in the air, but is translated into a series of individual obligations, and these it seems clear to me are obligations of justice. That is, if you are part of a collective group which owes something as a matter of justice to another group, and if your share in that obligation is fair and well‐defined, then there is also a more specific obligation of justice that falls on you as an individual. What we should say about individual duties when there is only partial compliance remains of course to be seen, but enough has been said to establish the point that it matters whether the group's collective responsibility is a matter of justice specifically, or something less urgent.

3. 3. The third factor that seems relevant to our question is whether the position of the non‐compliers is reversible or non‐reversible. Have they managed to rule themselves out of the picture altogether, so to speak, or might they be brought back into compliance? In a rescue case, for instance, we might contrast the position of those who quickly leave the scene and are then uncontactable, with those who stand around watching the rescuers but for the moment appear unwilling to contribute. More generally, we might contrast a one‐off decision, where each party has to choose at a certain moment whether to opt in or opt out, with an ongoing practice where agents' behaviour can adapt over time: we might not meet our greenhouse gas emissions targets this year, but next year we can do more by way of compensation. Reversibility makes a difference for the obvious reason that where it obtains, compliers have a choice between taking up the slack themselves, and trying to persuade or cajole or force the non‐compliers to make their fair contribution. But there may also be a more subtle difference, which I shall explore shortly, having to do with the assignment of responsibility. Where the non‐compliers remain on the scene—they have so far chosen not to comply, but they could also choose differently—it might be thought that responsibility for the effects of partial compliance remains firmly with them. They are the ones who by doing nothing are causing some of the victims not to be rescued or some of the starving not to be fed. If they manage to rule themselves out, however, responsibility appears to shift, in some sense, to those who might take up the slack. Whether this is indeed so (and what sense of responsibility is involved here) remains to be seen, but intuitively it seems to make a difference whether the decision the non‐compliers have made is reversible or not.

Having explored what seem to me to be the most salient dimensions of our problem, it's now time to begin working towards a solution. I want to argue broadly in favour of (p.238) option 1—that what justice requires is contributing your fair share, neither more nor less—while conceding that there may be special circumstances in which justice overall permits option 3—grouching by doing less—and other circumstances in which there is an obligation (though, I shall argue, not an obligation of justice) to choose option 2—taking up the slack by doing more. So let me begin by looking critically at the argument in favour of grouching, which I defined as doing only as much as others are doing, on average, given partial compliance.

The main argument in its favour, as I suggested, is that it preserves horizontal equity within the group: I am not assuming more of the burden than others are. Another way of dressing the argument up is to say that significant non‐compliance changes the meaning of a fair share of responsibility. One may be tempted into this way of thinking by considering cases such as collecting money for a colleague's leaving present where somebody might suggest an appropriate figure per head but looking into the collection box as it comes round you notice that there seem to be rather few notes and rather a lot of coins and so you swiftly revise your previous good intentions downwards. But the cases we are considering are not like this: they are cases in which the group has a collective responsibility to avoid some harm, and so if you do less than your original fair share you remain responsible for some portion of the harm that remains. Where that harm amounts to injustice towards the victims, as I have suggested it does in several of our real‐world cases, this seems decisively to outweigh the relative disadvantage that a complying agent suffers with respect to the non‐compliers in the group. Recall here that we are considering situations in which the cost of contributing to the collective effort is not only smaller than the cost that the victims will bear if the effort fails, but is moderate in absolute terms. So option 3 only looks plausible at all in two circumstances: one is where very little will be achieved by one agent's compliance given existing levels of non‐compliance; the other is where the harm that non‐compliance will produce does not involve injustice towards the victims, and the would‐be groucher can give good reasons why doing more than others have done is unduly burdensome to her (for example, that it leaves her at a significant competitive disadvantage in some other sphere of life). Otherwise it is dominated by option 1, doing your fair share.

You might, however, think that the argument I have just given for choosing option 1 over option 3 also entails choosing option 2 over option 1: if horizontal equity among members of the group isn't such a weighty consideration, why aren't we required to take up the slack in the face of partial compliance, at least where this will have the effect of averting injustice towards the potential victims? So now I need to show that option 1, doing your fair share, is all that justice requires.

The key argument here is that because the collective responsibility to avert injustice has been fairly distributed, ex hypothesi, by doing my fair share I have discharged my obligation, and the injustice that remains, because of partial compliance, is the (p.239) responsibility of the non‐compliers, and only theirs.¹¹ As a general matter, we are not required as a matter of justice to correct the injustice that others perpetrate, although we may have reason to do so. If Bert steals Anne's money, justice does not require Charles to right this wrong, although if Charles happens to be so placed that he can direct the money back to Anne, this would very likely be the right thing for him to do.¹² Charles commits no injustice if he fails to secure the return of Anne's money, because responsibility for Anne's loss rests entirely with Bert.¹³ How are things different in the partial compliance cases we are considering? Suppose that I am a member of a group with collective responsibility for averting some harm. Initially, then, I share in that collective responsibility. But now it is divided up so that I know what my fair share is, and I discharge it conscientiously. Others, however, do not. Does some share of the collective responsibility now revert to me, or does it remain entirely with them? One must avoid being misled here by cases in which some of the group are simply unable to discharge their share of responsibility. For example, in rescue situations, some potential rescuers might turn out to be terrified by what they were being asked to do, and become paralysed by fear. Under these circumstances, clearly, collective responsibility does return to the whole group and must be divided up afresh with each of the remaining rescuers having to carry a larger share. Similar considerations apply if external circumstances change, for example if the rescue group is alerted to the existence of a second group of potential victims, so that some of the rescuers need to hive off and form a separate party. By contrast, the cases we are considering are ones in which the non‐compliers choose not to comply, even though they are able to do so and have no competing obligations that would prevent them from contributing. How can such a choice shift responsibility wholly or partly off of their shoulders and on to those of the willing compliers?

As I indicated earlier, it might seem to make a difference here whether it still remains within the power of the non‐compliers to comply if they choose. If it does remain within their power, then it would be natural to say that what is causing the injustice that remains after partial compliance is the continuing unwillingness of the non‐compliers to do their bit; the responsibility therefore remains wholly with them. Suppose on the other hand that they have left the scene or somehow disabled (p.240) themselves irreversibly from making a contribution. Clearly they are morally at fault for doing this, and if we were in the business of ascribing responsibility in order to attribute blame for the bad outcome, these are the ones we would hold responsible. Nevertheless, given that they have in fact acted in this way, the question still arises of responsibility for the ongoing injustice. Those who have already complied could if they chose take up the slack and do more; the non‐compliers are out of the picture for practical purposes. Are the compliers not then responsible for the injustice if they decide not to do this?

In a purely causal sense of responsibility, this seems to be true. In other words, if we are asking what will make the difference between the remaining injustice's being remedied and its not being remedied, the answer, given the actual state of the world, is the decision by the compliers whether or not to take up the slack. But this causal sense of responsibility may not be relevant if what we are asking is whether justice requires taking up the slack. As I noted above, the fact that Charles may be able to correct the injustice that Bert has inflicted on Anne does not entail that Charles is himself required by justice to do this. After all, for each of us there are at any moment a large number of injustices that we might be able to correct if we set our minds to it. So the question that arises is what special circumstances, if any, have to obtain in order for the correction of injustice perpetrated by some other agent to be itself required by justice?

In the cases we are considering, there does seem to be one special circumstance, namely that both the compliers and the non‐compliers belong to the same group of agents who were originally held collectively responsible for averting the injustice. In other words, the injustice that the compliers are being asked to redress by taking up the slack is not merely some injustice that might occur at random anywhere, but the injustice that is caused by the failure of those with whom they originally shared collective responsibility to do their bit. However, it is important here to think clearly about the nature of this collective. I have often for convenience described them as a group, but this could be misleading if it suggests a set of agents who are already bound together in some way, by identity or organization for example. But think instead of the collection of people who happen to be on the scene when a traffic accident occurs, or the much bigger collection of people who are so placed that they have a duty to contribute to the relief of global poverty. Neither of these forms a group in the stronger sense. What connects them to each other is simply the fact of being so placed that they are able by sharing responsibility to avert some harm.

This is relevant, because in groups proper we may think that responsibility passes between the members when some of them default. If a team of people has undertaken some task, for example, and some members of the team back out or become unable to continue with the work, we will often think that the rest of the team has an obligation to complete the task on their behalf. But this doesn't apply in the cases we are considering. There may or may not be some kind of signalling or communication between members of the collective—for instance a group of rescuers may need to talk to each other to divide up the necessary labour—but this will not amount to a (p.241) precommitment to cover for defaulters. Indeed if anything the reverse will be the case: everyone has an incentive to send out a signal that he or she is prepared to do a fair share of the work, but nothing beyond that.

Responsibility, then, remains with the non‐complying members of the collective even in cases where they are no longer in a position to contribute by virtue of earlier decisions they have made. Reversibility matters only for the practical reason that where it obtains, the compliers can choose between bending their efforts to get the non‐compliers to play their part and taking up the slack themselves, whereas of course in the opposite case, only taking up the slack is an option. But in either case, compliers are not required to do more as a matter of justice.

This conclusion might, however, be challenged on the grounds that I have misrepresented the nature of collective responsibility in the situations being considered. I have relied on the idea that where collective responsibility can be fairly distributed among members of the relevant group, each has an obligation to do his fair share, but not more than that. If there are five people needing rescue, and five equally able rescuers, each rescuer's responsibility is limited to carrying out a single rescue. But there is another way of describing the situation. Suppose that the cost to each person of rescuing all five would still remain moderate, so that demandingness considerations do not enter the picture. Then it might be said that each rescuer has a responsibility to rescue all of the victims, even though, if fairness prevails, he or she will only need to discharge one‐fifth of that total. In other words, the situation is more like one in which there is just a single victim in need of help and many possible rescuers—in that situation each of the rescuers has a responsibility to help, but in the event may well not have to discharge the corresponding obligation if somebody else steps in first.

What's wrong with this proposed redescription? I think it misunderstands the force of the idea of a fair division of responsibility. This is not just a convenient way of dividing up a task—each gets a specific job to do (rescue one person, say) and in that way the whole task is discharged quickly and efficiently. Instead it reflects the idea that each person is a responsibility‐bearing moral agent, so that no one person is called upon to shoulder the entire burden of averting harm herself in the cases we are considering. Indeed if she does take that burden upon herself—by acknowledging in advance that she has a duty to take up the slack if others default—we might say that she does not treat the others as responsible moral agents. She would be like a mother with small children who asks for their help in clearing up the dinner table, and is pleasantly surprised when they do, but who knows that because they are only children all the responsibility finally rests with her. In contrast, in the five‐victims, five‐rescuers case, one who is left to do all the work as the others decamp can justifiably feel anger and resentment at the way she has been treated. She is not just somebody who is unlucky because as it turns out she has had to discharge her obligation in full; she has been exploited by agents whose responsibility is identical to hers but have simply chosen not to discharge it.

Where responsibility for averting a collective harm can be fairly divided, I have argued, justice can only require each agent to perform his or her fair share. But this (p.242) conclusion runs up against an intuition that many people have, which is that there are situations in which justice does require more, situations in which questions about fair shares of responsibility should be set aside in favour of simply looking at the consequences of acting or not acting.¹⁴ Rescue cases are the most frequently cited example: many think it obvious that if others are refusing to cooperate, the rescuer who has done his bit is duty‐bound to jump back into the river and pull more people out, subject to the usual riders about excess cost or risk. But we might think the same applies to cases like global warming: if the world is reaching some crucial tipping point, such as the melting of a large portion of one of the polar ice‐sheets for example, then the fact that other countries refuse to cut their carbon dioxide emissions in line with an agreed set of targets does not pre‐empt complying countries being obliged as a matter of justice to cut back still further, given the consequences of not doing so.

What should we say about this? Everyone can agree that in these cases compliers have a strong reason to take up the slack, since doing so will have the effect of safeguarding basic rights otherwise under threat. We can also agree that agents who refuse to take up the slack merely because of their understandable indignation at the non‐compliers are morally at fault and can be criticized and blamed for their refusal. But do they have an obligation to take up the slack, and if so what kind of obligation could this be?

One way to approach this question is by asking about possible enforcement. It is characteristic of obligations of justice that they are enforceable in principle by third parties, the nature of the enforcement that is justifiable depending on the case we have in mind.¹⁵ Roughly speaking, the more urgent the duty of justice, as measured by the effects of not fulfilling it, the heavier the sanctions that may be applied by a third party to an unwilling duty‐bearer. In a simple rescue case, where the only potential rescuer refuses to carry out a low‐cost rescue, a third party arriving on the scene but unable to carry out the rescue herself (she can't swim, for example), can justifiably threaten to impose some fairly substantial loss, like pushing the rescuer's car into the water, for example, in order to save a life. This, it seems to me, applies also in the cases of collective responsibility we have been examining, which is why I said earlier on that if it were possible to create some mechanism that would ensure compliance that would very often be desirable. We are examining cases where no such mechanism can be (p.243) created, for practical reasons, so I am appealing to enforceability only in order to test our intuitions about justice. So, consider the situation in which some have fulfilled their obligation to avert injustice but others have not. An enforcing agent arrives on the scene but is unable to get the non‐compliers to act—threats of various kinds do not work. Is the enforcing agent then justified in turning his attention to the compliers and forcing them, if he can, to take up the slack? Simple consequentialism would say that he is so justified, but my intuition is that he is not. Because the compliers have already discharged their share of the responsibility, it cannot be right to force them to do more.¹⁶ So their obligation to take up the slack is not an obligation of justice.

Let's then find a different label for this obligation—‘a humanitarian obligation’ would be my preferred description. Why call it an obligation at all if it is not enforceable? First of all, to distinguish it from cases of supererogation.¹⁷ There is a moral difference between the person who refuses to go back into shallow water to pull out further victims having already done his share, and the person who declines to leap into a raging torrent for the same purpose. The first person can be blamed for his refusal while the second cannot. Second, in order to underline that the humanitarian duty in question is a weighty moral reason that can override others. Thus if the person who is taking up the slack has to disregard other duties for the time being, she could say, rightly, that in the circumstances she was obliged to do what she did. The language of obligation conveys the motivating force that the reason to take up the slack should have for the agent, while at the same time defeating the less urgent demands of others—people to whom promises have been made, or who have legitimate expectations of other kinds, for instance.¹⁸ Third, to indicate that the victims have a legitimate complaint against the agent if she fails to discharge her obligation. Of course their main complaint will be against the non‐compliers who have failed in their duty of justice. (p.244) If restitution comes into question after the event, it is the original non‐compliers who will be required to pay, not those who have merely failed to take up the slack. Nevertheless because those in the latter group have failed to act to safeguard the rights of the victims in circumstances where they could do so without incurring excessive cost, the victims do have a secondary complaint against them as well. By saying that they had a humanitarian obligation to act, we help to make sense of this complaint.

Does this also mean that compliers who decline to take up the slack are responsible for the harm that they fail to avert, in more than a merely causal sense? Do they share responsibility with the non‐compliers? We need to take great care in our use of the language of responsibility here. Consider the following case. Daniel is searching for Edward with the intention of killing him. Frances knows of Daniel's intention, but can hide Edward in a place where Daniel will not find him. However, she decides not to do so and Edward is killed. In what sense, if any, is Frances responsible for Edward's death? If we were to say that she shares the responsibility with Daniel, this would be quite misleading, because it would assimilate the case to one in which Daniel and Frances have cooperated to kill Edward.¹⁹ It is clear that the primary responsibility for Edward's death rests entirely with Daniel. Frances bears only a secondary responsibility, namely the remedial responsibility to avert harm that will otherwise be caused by another agent.²⁰ She is blameable for not discharging it in the absence of excusing or justifying factors. So she is responsible for Edward's death in a sense that is more than merely causal. However, her responsibility is of a categorically different kind from Daniel's, which is why any talk of shared responsibility is out of place. My suggestion is that this distinction applies also in the cases of partial compliance we have been considering. If the harm is not averted, primary responsibility for this lies with the non‐compliers, and with them alone. If the compliers have a humanitarian obligation to take up the slack but fail to do so, they become responsible in a secondary sense. But it is incorrect to say that responsibility for the harm is shared by the two groups, for this neglects the categorical distinction between primary and secondary responsibility.

There is one final implication that I wish to draw out of this discussion. For reasons of simplicity I have been using examples in which it is individual people who have to decide whether to comply in situations of collective responsibility, or whether to go further by taking up the slack as well. But I said that the analysis should apply too to (p.245) collective agents such as states, and it seems clear that problems such as resource conservation and climate change are more important than individual rescues if we are thinking about their effects on human rights and human welfare. So what does our discussion tell us about collective agents specifically? Notice here that compliance problems will often emerge at two levels: first the collective has to decide whether to discharge its fair share of responsibility, and then the individual members have to decide whether to do what they are asked to do by the collective. Thus Britain, say, has to decide whether to sign a treaty that sets limits to the overall level of greenhouse gas emissions that its population generates in any year, and then individual citizens have to decide whether to make the behavioural changes that are required of them by cutting down on the use of fossil fuels and so forth. Now I take it that where an obligation of justice is involved, the collective is justified in enforcing its decision if necessary. So if we believe that runaway climate change threatens the basic rights of people worldwide, the UK government will be justified in compelling citizens to comply with measures that are necessary to keep gas emissions below the level agreed in the treaty, such as using only low‐emission vehicles for transport. On the other hand, where the harm that we are trying to avert by collective action is not one that involves injustice, or where what we are doing collectively is actually taking up the slack left by others, then internal compulsion becomes more problematic. And this raises difficult questions of political philosophy. Suppose, for example, a nation sets itself more ambitious targets in relation to climate change because of an awareness that other countries are not going to fulfil their obligations: can it be right to force citizens to change their behaviour so as to meet those targets, or must the state at this point rely upon encouraging voluntary compliance? Or to take a different case, what obligations can be compulsorily imposed on citizens in response to humanitarian disasters such as famines or civil wars occurring abroad where primary responsibility lies with the governments of those states?²¹ Although there is no algorithm for answering these questions, I hope that what I have said here provides at least a partial framework for thinking about them.