II. How do we test for necessity in war?

This section presents a working analysis of necessity, and a methodology for testing the empirical premise of the necessity-based argument. We must begin by clearing up some terminological ambiguity. In international law and military discourse, military necessity is typically understood as simply meaning military advantage.Footnote ¹⁹ On this view, anything that increases the likelihood and reduces the costs of victory is considered militarily necessary. Obviously it is not this understanding of necessity in war that grounds the necessity-based argument for non-combatant immunity. The argument depends instead on a more general normative principle, which is indeed frequently recognised in international law and military discourse, even if not always under this description. The principle of necessity is defined by its inverse: the infliction of unnecessary suffering is always impermissible. Undeserved, involuntary human suffering is without exception bad, and is prima facie proscribed; if it is unnecessary, then there is nothing to override this prima facie prohibition.Footnote ²⁰

The necessity constraint comprises three individually necessary and jointly sufficient conditions for ensuring that the infliction of some harm H, on a victim V, by an agent A, to achieve a goal G is necessary.

Condition 1: the harm done must advance some goal. This means two things. First, the harm must be aimed at achieving a goal – it must not be wholly wanton, or an end in itself. Second, it must be somewhat effective at achieving that goal. Harm that contributes nothing to achieving its objective is inescapably unnecessary.

Condition 2: there must be no less harmful course of action with equal or better prospects of achieving the goal. If A could have as good a chance of bringing about G by inflicting less than H harm, then the additional harm done by inflicting H serves no purpose, and so is unnecessary.

Condition 3: if there is a less harmful course of action available that is less likely to succeed, then the difference in prospects of success – or effectiveness – must be sufficiently weighty to justify the difference in harm inflicted.Footnote ²¹ Suppose that, if A inflicts H harm, he has a 0.8 chance of achieving G, but if he inflicts H/2 harm, he has a 0.3 chance of success. In this case, both H and H/2 are somewhat effective at achieving G, but the less harmful option involves costs to A. The more harmful option satisfies necessity only if the additional harm inflicted on V is justified by avoiding the cost to A of a 0.5 reduction in his prospects of success.

For the necessity-based argument for non-combatant immunity to be true we must show that harms inflicted on non-combatants in war (almost) never satisfy these three conditions.

With this working analysis of necessity in hand, the next step is to ask how it can be tested. This is no easy task. We must first formulate a clear sense of the type of anti-civilian attacks that are our focus; then consider the specific methodological issues raised by showing that each of the three necessary conditions for necessity is satisfied (or otherwise).

The first task is to find an operationalisable and morally plausible definition of non-combatant status. Revisionists might seek to defend a version of non-combatant immunity by redefining what is meant by a non-combatant, or proposing some alternative distinction, such as that between participants and non-participants, contributors and non-contributors, or indeed just liable and non-liable.Footnote ²² Whatever the merits of those moves, they are irrelevant to this article, since our interest is in whether revisionism is consistent with conventional beliefs about the protection of non-combatants in war, which presupposes the conventional classification of combatants and non-combatants. This is the line drawn by international law, according to which there are two jointly necessary and sufficient conditions for having non-combatant status: you must be neither a member of the belligerent armed forces, nor directly participating in hostilities. Operationalising this definition is not easy, since the notion of ‘direct participation’ admits of degrees – the familiar problem of how to classify an individual who is a farmer by day, and a guerrilla by night.Footnote ²³

Having decided who the non-combatants are, we need next to consider what tactics the necessity-based argument proscribes. Here I think the laws of war are too inclusive for our purposes. The principle of non-combatant immunity in international law enjoins the protection of non-combatants against all indiscriminate attacks, where indiscriminate attacks fall into two main categories: those where the non-combatants are the intended target, and those

Revisionists do not invoke the necessity-based argument to protect non-combatants against these latter unintended, disproportionate harms, focusing instead on harms that are purportedly a means to achieve a military objective. Our focus, then, should be on the intentional attacks on non-combatants proscribed in the Basic Rule, article 48 of the first additional protocol:

It is worth noting, in passing, that revisionists’ views have troubling implications for collateral harms to non-combatants, which cannot be resolved by appealing to the necessity-based argument. If many non-combatants are liable to be harmed intentionally, then they (and perhaps others) will also be liable to suffer equivalent, perhaps even greater harms when they are inflicted collaterally.Footnote ²⁶ This means that a greater magnitude of harm can proportionately be inflicted to achieve the same end, than if we denied non-combatant liability. Those who worry about the implications of revisionism for non-combatant immunity should be equally concerned that it threatens to make the proportionality constraint on collateral harms to non-combatants (at least those who are liable to be harmed collaterally) so much more permissive. However, for the purposes of this article, and assessing the necessity-based argument, I will focus on intentional attacks on non-combatants. Notice that this includes what most people think of as acts of terrorism within its ambit.Footnote ²⁷

To satisfy the necessity-based argument's empirical premise we must consider whether anti-civilian attacks in war ever satisfy the three necessary and sufficient conditions for necessity in war. The first task is to ask whether anti-civilian attacks are ever effective at advancing states’ goals in war. This again raises both descriptive and normative problems. First, how do we know whether a tactic is effective in war? How can we demonstrate the causal connection between the deployment of some tactic and the realisation of a strategic outcome? Obviously experimental verification is out of the question, so all we can do is interpret the historical and contemporary record. But how can we make the relevant causal attributions, given the over-determined nature of success and failure in war? Where anti-civilian attacks are followed by strategic success, how can we be sure of the role they played in bringing about that success? Where defeat follows, how can we know whether the anti-civilian attacks were to blame, or conversely whether they were effective, and defeat resulted from other factors? These are familiar problems with causal attribution, but they are all exacerbated by the confusion and empirical obscurity endemic to armed conflict.Footnote ²⁸

These descriptive problems are compounded by a difficult moral question, which the revisionist advocates of the necessity-based argument will feel particularly keenly. The necessity constraint outlaws unnecessary harms, because all human suffering (except perhaps when deserved or consented to), is a bad thing, which we have prima facie reasons to avoid. If that harm does not advance a goal, then we have no moral reason that can outweigh that prima facie prohibition. Notice, though, that this is also true if the harm will bring about a morally objectionable goal. Suppose A is a thief, V a security guard, and G is the outcome where A steals the contents of a jewellery shop. Even if there is no less harmful way to bring about G than to inflict H on V, since G is a morally objectionable goal A still has no moral reason that can outweigh the prima facie reason not to harm V. In other words, while H was necessary to bring about G in a value-neutral sense, it is in a broader sense still morally unnecessary.

We can call necessity ‘morally neutral’, when our focus is on bringing about G, without prejudice as to G's moral value or disvalue, and ‘morally rich’, when it is a necessary condition of some harm satisfying necessity that the goal aimed at be sufficiently morally valuable.Footnote ²⁹ This contrast between morally neutral and morally rich necessity could have profound implications both for the necessity-based argument and for our attempt to evaluate it. If any attacks that serve an unjustified goal inescapably fail the relevant necessity constraint, then the necessity-based argument for non-combatant immunity must again be restricted in scope, and the task of assessing its empirical premise would become considerably more challenging.

If the necessity-based argument deploys the morally rich necessity concept, then it cannot support a neutral principle of non-combatant immunity, which applies to all sides in a conflict, regardless of whether their aims are justified. Since any force used to achieve an unjustified goal is a priori unnecessary, belligerents whose objectives are unjustified have no reason to refrain from attacking non-combatants which does not equally forcefully proscribe attacking combatants. This is a significant shortcoming – not only because it gives inadequate guidance to otherwise unjustified belligerents, but because it cannot support the very intuitive view that even if all the killing you do in war is wrongful, killing non-combatants is other things equal an especially egregious, aggravated form of wrongdoing.

This moralised understanding of necessity also raises problems for the argument's empirical claims. It implies that we should be interested only in cases of anti-civilian attacks that aim at justified goals. Should our dataset therefore include only cases where anti-civilian attacks are aimed at a good sufficient to justify that harm? This would pose serious problems for the argument's falsifiability. For each historical example, we would need to determine not only whether the war fought by a belligerent was justified as a whole, but also whether the specific aim of this attack was justified. Resolving either of these matters involves considerable controversy: indeed, to do it properly each example would require a book-length discussion of its own. We would need first to settle on the correct principles for assessing the attack, and then authoritatively determine whether the broader objectives of the belligerent were justified, and similarly for the specific objective attained by these tactics. Not only would this make our evaluation of the necessity-based argument contingent on first resolving moral questions that are each as complex and controversial as those around non-combatant immunity, it would also potentially reduce the dataset from which inferences could be drawn to an infinitesimally narrow field. This is especially likely to be true for revisionists if, as I have argued elsewhere, their view would render almost no actually existing wars justified.Footnote ³⁰

However, this worry can be overcome, if the core objectives of justified and unjustified conflicts (and the vast set that fall somewhere in between) are the same. More precisely, provided the goal at which a belligerent's tactics are aimed is the sort of objective that a justified belligerent could be aiming at, then if a given tactic is the necessary means of achieving that goal, then if most other things besides overall justification are equal, we can infer from its being necessary (in the morally neutral sense) in an unjustified or ambiguous conflict that it could also be necessary (in the morally rich sense) in a justified one. In many wars, the methods by which victory is achieved are the same – belligerents seek to capture strategically important assets and locations, and to subdue the adversary's ability to fight back. Moreover the outcomes are the same: one side seeks to establish a sufficient threat advantage over the other to force it to make further territorial and political concessions. Notice that this does not entail taking a position on the ‘independence thesis’, that assessments of in bello conduct can be formulated without regard to whether one's cause is ad bellum justified.Footnote ³¹ It is merely an empirical claim, that the intermediate objectives that must be secured for military victory to be achieved are the same in many wars, regardless of whether they are justified or unjustified.

Imagine (not implausibly) that the history of warfare offered only unjustified wars, but that we were faced, now, with the imperative to fight a war that is justified. Suppose you are the commander-in-chief of the justified belligerent, and you ask your generals to plan a strategy composed of tactics that satisfy military necessity. They could surely draw lessons from the history of warfare, from scenarios that were sufficiently similar to their own except that the relevant belligerent had unjustified goals, about what had proved necessary in the neutral sense and so would for them be necessary in the morally rich sense. The mere fact that the dataset differs in this way from their own experience is not sufficient to block the inference from one set of cases to the other.

Of course, sometimes there are specific reasons to block that inference. If the objective aimed at by the belligerent providing the data was the sort that could not possibly be aimed at by a justified belligerent, then the necessity of victimising civilians to achieve that objective is irrelevant to determining necessity in justified conflicts. This will most obviously be true when the belligerent's objective is itself to cause harm to civilians – if the goal is genocide or ethnic cleansing, then obviously anti-civilian attacks will be necessary in the neutral sense, but since these objectives could not possibly be aimed at by justified belligerents, this is irrelevant to the necessity-based argument.

Provided the dataset only includes cases where the objectives are such as could be aimed at by a justified belligerent, we can infer from it conclusions about necessity in future justified wars. This means that instead of focusing only on cases where we can (per impossibile) show that the overall objectives of the conflict, and the specific near-term objectives of the tactical manoeuvre being assessed, were justified, we can draw on a much wider pool of evidence. Interstate conflicts over disputed territory, insurgencies against incumbent powers, and attempts to suppress violent insurgencies are conflicts where the objectives could be aimed at by a justified belligerent. This does not mean that all the specific historical cases discussed in what follows could have been justified – indeed, we need make no controversial claims along those lines at all. It is merely to say that if a future belligerent (FB) faces a situation like that faced by some historical belligerent (HB), which is sufficiently similar in all morally relevant respects except that FB's goals are justified, while HB's otherwise similar goals were not, then we can reasonably infer from it being necessary in the neutral sense for HB to use some tactic (T), that T can be necessary in the morally rich sense for FB.

To show that the first condition of necessity is satisfied, we need show only that anti-civilian attacks are effective at achieving the sort of goal that a justified belligerent could aim at. This is challenging enough; evaluating whether the second and third conditions for necessity are satisfied adds further empirical difficulties. An effective attack is unnecessary if there is either some other equally or more effective option available that would be less harmful, or if there is a less harmful option such the difference in effectiveness between them does not justify the additional harm inflicted. This means that to know whether a tactic satisfies necessity we should compare it with options that were not taken: counterfactual scenarios about which we can only speculate. Even in favourable conditions, counterfactuals are (to put it charitably) indeterminate. But in the chaotic context of war saying that some other strategy not taken would have been just as or more effective, and equally or less harmful, is extremely difficult, if not impossible.

Rather than engage in this sort of complex speculation, I think we should keep a tighter handle on our counterfactuals, and so I propose a correspondingly more modest research question. Does the historical record suggest that there can be situations where belligerents, aiming at the sort of objectives that justified belligerents might aim at, find anti-civilian attacks to be effective when no other effective option is available to them? Showing that such cases will not arise is a necessary condition for justifying the necessity-based argument; conversely a strong argument that they will would help refute it.Footnote ³²

The remainder of this article seeks to answer this question. I begin by asking whether anti-civilian attacks can be effective, considering in turn the case against and the case for this proposition. I then ask whether, in cases where anti-civilian attacks do look effective, there might be no other effective alternatives. I conclude by considering how the necessity-based argument stands, in light of this enquiry.Footnote ³³

III. The case against anti-civilian attacks

It is harder to show that anti-civilian attacks are never effective than to show they sometimes work. And yet if they are ineffective in a range of conflicts, then absent countervailing evidence, that would support the necessity-based argument. In this section I discuss research that supports this response, looking at aerial bombing in interstate conflict, then at counterinsurgency and terrorism. In section IV, I respond to these arguments, and present the opposite case.

In Bombing to Win (1996), Robert Pape famously argues that aerial bombing of civilians is strategically redundant. Distinguishing between punishment and denial tactics – the former inflict civilian suffering to coerce concessions (either directly or through popular uprising); the latter neutralise the adversary's military capacities, from production, through supply line to battlefield – he shows that in 40 campaigns between 1917 and 1991, punishment did not work.Footnote ³⁴

Pape's principal target, besides mid-1990s’ US strategic doctrine, is Giulio Douhet, whose 1921 book Command of the Air is the locus classicus of punishment-based aerial bombing in strategic theory.Footnote ³⁵ In the Second World War, British Bomber Command explicitly endorsed this approach: the bombing of Germany left seven and a half million people homeless; 305,000 civilians were killed, and 780,000 wounded. Cologne, Hamburg, and Dresden were flattened, in the hopes of prompting the people to rise up against their leaders to stop the war.Footnote ³⁶

And yet, Pape argues that the anti-civilian bombing of Germany played little role in bringing about Allied victory. It neither compelled the Nazi regime to make concessions, nor prompted an uprising against them. Similarly, Pape's reading of internal records suggests that anti-civilian attacks in Japan (including the atomic bombs) were largely irrelevant to their surrender, which was determined more by the advancing Soviet army, rather than worries about continued civilian suffering. He offers detailed analysis of three other cases that might be thought counterexamples (Korea, Vietnam, Iraq (1991), all bombed by the US), and argues that in none did punishment bombing contribute to victory.Footnote ³⁷ Bombing civilians, he concludes, has never led to uprisings against the adversary regime.Footnote ³⁸

Why does Douhet bombing fail? Regarding Germany, Pape notes that nationalist sentiment, government propaganda, and state repression stopped anti-civilian attacks leading to anti-war protests.Footnote ³⁹ Additionally, terrible attacks on civilians leave them primarily concerned with survival, rather than bigger political issues; indeed, it may make them more loyal, since only their government can protect them from these attacks and their effects.Footnote ⁴⁰ More generally, Pape argues that nation-states can tolerate high costs for their vital interests and national pride. They can also mitigate the bad effects of punishment bombing – perhaps not to the civilians’ benefit, but enough at least to minimise their impact on fighting capacity.Footnote ⁴¹

Some researchers of counterinsurgency have reached similar conclusions about the effectiveness of anti-civilian attacks, identifying three plausible objections.

First, counterinsurgents need the support of those whom the insurgents claim to represent. The US Counterinsurgency Field Manual notes that ‘counterinsurgents that use excessive force to limit short-term risk alienate the local populace. They deprive themselves of the support or tolerance of the people. This situation is what insurgents want. It increases the threat they pose.’Footnote ⁴² Insurgents intend their adversaries to overreact, because that strengthens their position. Excessive force generates anger, which triggers high-risk behaviour.Footnote ⁴³

The second argument extends the first: the domestic population is not the only relevant audience; often, the international community is watching too. If the counterinsurgent is internationally regarded as illegitimate, that raises new obstacles to success. When counterinsurgents engage in anti-civilian attacks, international disapprobation, economic sanctions, even foreign intervention might ensue – Libya is the obvious recent example; Syria may be the next.

The third objection is more abstract. Counterinsurgents use force against civilians to compel compliance. But indiscriminate force is a blunt tool: if those who comply and those who refuse are equally likely to be killed, then why comply?Footnote ⁴⁴ To work, coercive sanctions must be prompt, focused, proportionate, and consistent. Indiscriminate attacks on civilians meet none of these criteria.Footnote ⁴⁵ Far from compelling compliance, they are likely to force civilians into the insurgents’ arms, as the only people who can offer protection.Footnote ⁴⁶

In counterinsurgency, scepticism about the efficacy of anti-civilian attacks is common. In research on terrorism, the dominant view is probably the reverse.Footnote ⁴⁷ Those arguments are discussed below; here I focus on a prominent critique of this consensus.Footnote ⁴⁸ It begins with a methodological objection: much of the research on terrorism's effectiveness ‘rests on game-theoretic models, single case studies, or a handful of well-known terrorist victories’, rather than broader empirical foundations.Footnote ⁴⁹ In response, Max Abrahms examines 28 organisations classed by the US State Department as Foreign Terrorist Organisations before 2001, and assesses their success rate, as well as their methods. He identifies 42 policy goals for the 28 organisations, and concludes that only three were either partially or fully achieved.Footnote ⁵⁰ He notes, further, that organisations that primarily target civilians achieve even limited successes only very rarely (again, he finds three examples).Footnote ⁵¹ He reasons that terrorists are more likely to succeed when they have limited territorial goals, but targeting civilians miscommunicates your policy objectives, suggesting that your aims are maximalist and ideological.Footnote ⁵² By killing civilians, terrorists engage the adversary's vital interests, and so invoke the same resilience and resistance that makes Douhet-style bombing so ineffective.Footnote ⁵³

IV. Anti-civilian attacks can be effective …

Notwithstanding the arguments of the foregoing section, other evidence suggests that anti-civilian attacks can be effective in terrorism, counterinsurgency, and interstate conflict. The contrary conclusion is based in part on too restrictive conceptualisations of the relevant attacks, and of what it is for a tactic to be effective. In this section I expand on these observations, before noting some cases where even by the criteria set out above, anti-civilian attacks appear effective.Footnote ⁵⁴

The work by Kalyvas, Abrahms, and Pape discussed above operates with too narrow an understanding of anti-civilian attacks to afford compelling support to the necessity-based argument (though that understanding was undoubtedly appropriate for their own research objectives). Each focuses on the efficacy of indiscriminate civilian attacks, but this is not the only relevant form: selective, or discriminate anti-civilian attacks are also possible, and are no less morally troubling. They can certainly be massive in scale. The Vietcong, for example, selectively assassinated as many as 50,000 people in 15 years.Footnote ⁵⁵ In the Ukrainian resistance to Nazi Occupation selective killings by the Germans cumulatively rivalled the death toll in massacres.Footnote ⁵⁶ Additionally, detailed research shows that what history classes as massacres are often in fact quite selective – as an observer of the 1997 massacres in Algeria noted: ‘Massacres are not blind. They are planned and target specific families. They bypass other families.’Footnote ⁵⁷ Terrorists and insurgents also often use selective anti-civilian attacks – witness the rise in assassinations of individual non-combatants by the Afghan Taliban in recent years.Footnote ⁵⁸ Unfortunately there is not enough research to draw firm conclusions about the effectiveness of selective anti-civilian attacks, but a priori it has one obvious advantage over indiscriminate attacks: if your objective is to coerce a target audience into some set of actions, targeting specific civilians who help the adversary gives others clear incentives to defect to your side.Footnote ⁵⁹

Still more important than this narrow conceptualisation of anti-civilian attacks is a too-restrictive understanding of what makes a tactic effective. Kalyvas, Abrahms, and Pape all define effectiveness by contribution to overall strategic success. A tactic is not effective, on their view, unless it leads to victory. But this is obviously too restrictive – not only because of the difficulties identifying the causal contribution of one tactic to overall victory in war, the paucity of clear data in war, and the potential over-determination of every outcome, but also because tactics can be effective insofar as they realise some intermediate objective, regardless of whether they ultimately yield strategic success.Footnote ⁶⁰

This is especially clear when intermediate goals are achieved, but strategic success is not – for example, capturing town X would help us win, and using tactic T_X we do capture it, but because we lose town Y, we lose the war regardless. It does not follow that T_X was ineffective.Footnote ⁶¹ It is also clear when victory is ultimately achieved through some other avenue. Suppose victory requires capturing the adversary leadership, who are hidden either in town X, Y, or Z. Tactic T_X helps us capture town X, T_Y helps capture town Y, and T_Z helps us capture Z. Our targets turn out to be in Z. Nonetheless clearly T_X and T_Y should count as effective.

In less general terms, there are clear intermediate goals that anti-civilian attacks can yield, which suffice to qualify them as effective. Consider asymmetric conflicts in which insurgents or other non-state groups confront a powerful incumbent. Anti-civilian attacks by insurgents can mobilise and galvanise their own community, empowering them despite the crushing imbalance of power with the adversary.Footnote ⁶² They attract financial backers, as well as possible fighters.Footnote ⁶³ They can also use attacks on their own community to enforce compliance with their agenda – the Algerian civil war is an example.Footnote ⁶⁴

Insurgents can also use anti-civilian attacks as a means to win international support – this might seem counterintuitive, but when small-scale attacks on civilians provoke disproportionate responses from strong governments, insurgents can portray themselves as underdogs, victims of an unfair imbalance of power.Footnote ⁶⁵ Regardless of whether they achieve overall strategic success – and their battles might be ongoing, or might be decided by other factors – this tactic of provoking the adversary can definitely be effective. The Kosovo Liberation Army, for example, increased solidarity among Kosovar Albanians by attacking Serbian police, leading to vicious crackdowns, which ultimately led to international intervention.Footnote ⁶⁶ Kurdish attacks on Turkish targets, by prompting Turkish government repression, have slowed ‘the assimilation of the Kurdish population into Turkish culture’, which is a valuable intermediate goal of Kurdish nationalism.Footnote ⁶⁷ Hindu nationalists have used terrorism to divide Hindus from Muslims in cycles of retaliation.Footnote ⁶⁸ Hamas increased violence before Israel's 1996 and 2001 elections, encouraging support for Likud, who predictably responded with crackdowns that further galvanised Palestinian resistance.Footnote ⁶⁹

Besides provoking a vicious response, anti-civilian attacks by the militant wing of insurgent movements can enhance their negotiating factions’ bargaining power. Irish and Basque nationalists, for example, have forced concessions from their adversaries, including regional autonomy and, for the IRA, political recognition for their leaders.Footnote ⁷⁰ In Palestine, Hamas activists clearly regard Israel's withdrawal from the Gaza Strip as a victory, caused by their anti-civilian attacks.Footnote ⁷¹

Finally, anti-civilian attacks can contribute to the intermediate goals of insurgency by rendering a territory ungovernable. As recent years in Iraq and Afghanistan have shown, this is particularly effective against an occupying military force whose objective is nation-building. Every attack demonstrates the occupier's inability to guarantee the most fundamental prerequisite of political life, that is, security from lethal attack, and so delegitimises them.Footnote ⁷² The incumbent's inability to regain control discredits it; it also uses resources on remedial measures, diverting them from prevention.Footnote ⁷³

States too can achieve important intermediate goals through anti-civilian attacks – contribution to overall strategic success is not the whole story. In counterinsurgency, anti-civilian attacks can help reduce casualties among the insurgents’ forces. Jason Lyall has shown that indiscriminate artillery shelling (at random times, from random locations) by Russian counterinsurgents on Chechen villages was effective at reducing rebel attacks between 2000 and 2005.Footnote ⁷⁴ Using the matching technique, he paired shelled villages with otherwise similar, but unshelled villages, to see the difference in insurgent response. One might expect them to be counterproductive, leading to more attacks from shelled than from control villages, but Lyall finds the opposite to be true. Shelled villages drop from a mean of 2.11 attacks to 1.5, in the 90 days before and after a Russian strike, while control villages drop from 2.15 to 2.05 during the same period.Footnote ⁷⁵ This means shelling yielded a 24.2 per cent reduction in insurgent attacks.Footnote ⁷⁶ He shows that this translates into 81 fewer attacks in the 90 days after shelling than would otherwise have taken place. Since the average insurgent attack killed 0.88 Russian soldiers and wounded another 1.21, the reduction in insurgent attacks attributable to shelling saved 71 soldiers’ lives, and a further 107 escaped wounding.Footnote ⁷⁷

Even in interstate conflicts anti-civilian attacks can yield intermediate dividends. Although non-combatants do not pose threats, they do contribute to the war effort, as noted in the introduction. Killing civilians can diminish the adversary's productive capacity, as well as divert resources away from offence towards defence.Footnote ⁷⁸ Indeed, this was the intended goal of US firebombing of Japan in 1943 and 1944.Footnote ⁷⁹ Of course, in short wars the fighting will be over before damage to production capacity can start to tell. But in drawn-out wars of attrition, belligerents are more likely to use up their stockpiles of materiel, so civilian productive capacity ‘becomes directly relevant to success or failure on the battlefield’.Footnote ⁸⁰

Finally, there is some evidence that anti-civilian attacks can in fact contribute to overall strategic success, not only to intermediate goals. Insurgent groups that deploy terrorist methods in asymmetric conflicts can achieve strategic objectives by coercing their more powerful adversaries. Asymmetric conflicts are often asymmetric in two respects: the belligerents’ relative military capacities, and their relative commitment to the issue at stake. The weak actor often seeks concessions from the strong that matter profoundly to the weak, but are peripheral to the strong.Footnote ⁸¹ Although nation-states are resilient when their vital interests are threatened, a disparity of interest makes coercive strategies more effective.Footnote ⁸² Some also argue that democratic states are particularly vulnerable to coercion, since they are more sensitive to casualties.Footnote ⁸³

Pape's research into the effectiveness of suicide bombing between 1980 and 2001 confirms this analysis. In the 1980s, there were 31 recorded suicide attacks worldwide; 104 in the 1990s, and 53 in 2000–1.Footnote ⁸⁴ Since then this number has rocketed – Afghanistan alone saw 280 suicide attacks between 2009 and 2010.Footnote ⁸⁵ Pape asks what explains this increase, and concludes that suicide terrorism is growing because terrorists – justifiably – believe that it works.Footnote ⁸⁶ Suicide terrorists fighting the French and US in Lebanon in 1983, Israel in Lebanon in 1985, Israel in the occupied territories in 1994 and 1995, and Sri Lanka since 1990 made more gains after resorting to suicide operations than they did before.Footnote ⁸⁷ Out of 11 campaigns completed in the test period, more than half ‘closely correlate with significant policy changes by the target state’.Footnote ⁸⁸ Leaders of terrorist groups, other observers in their community, neutral analysts and adversary government leaders ‘often agreed that suicide operations accelerated or caused the concession[s]’.Footnote ⁸⁹ This is an impressive success rate, considering both the general ineffectiveness of military coercion (effective only 30 per cent of the time, according to Pape)Footnote ⁹⁰ and the paucity of alternative options available to terrorists in these cases, faced with such powerful adversaries. Of course, many of the cases discussed by Pape were not anti-civilian – the attacks on the US barracks in Lebanon, for example. But some were, and those that were not still reveal the possibility of achieving success through spectacular coercion, which is the basic model for anti-civilian attacks.

Pape notes that suicide bombing works best when the bomber's objectives are limited and territorial. Ejecting an occupier is easier than overthrowing a government. This applies beyond just suicide terrorism. Campaigns of national liberation have often made good use of anti-civilian attacks – witness the anti-colonial struggles of the mid-twentieth century, such as the expulsion of (some) colonisers in Cyprus, Palestine, Aden, and Algeria. More recently, many regard Spain's withdrawal from Iraq after the 2004 Madrid bombings as a direct success for terrorism.Footnote ⁹¹ Others counter that Spaniards did not reject the ruling Popular Party because they were cowed by the terrorists, but because of the government's opportunistic handling of the bombings, and rush to blame them on ETA.Footnote ⁹² But this misses the point: that mishandling would not have occurred without the attack, which aimed not merely to coerce the Spaniards through punishment, but to sow discord and dissent, and invite just such a maladroit response.

States too can achieve strategic success through anti-civilian attacks in asymmetric conflicts. Insurgents depend on the civilian population for food, shelter, money, and recruits.Footnote ⁹³ Sometimes this relationship is even formalised, as insurgents create parallel state structures, including a tax regime.Footnote ⁹⁴ Equally important, non-combatants provide ‘human camouflage’, enabling rebels to evade detection.Footnote ⁹⁵ Civilians hide weapons, and deliver messages for the insurgents. Guerrillas might also rely on civilian institutions, including ‘financial, law enforcement, welfare, political, educational, and media institutions’.Footnote ⁹⁶ Media are especially important, as are the homes and Internet cafes from where insurgents’ grievances and achievements are broadcast to the world. Blocking this support can critically disrupt insurgent groups, denying them materiel, communication, and safe havens.Footnote ⁹⁷ Civilian support for insurgency can be blocked in three ways: elimination, coercion, and division. Elimination is obvious: you cannot contribute to insurgency if you are dead. Coercion likewise: if civilians fear reprisals, they may learn to dissociate from the rebels. Division arises when civilians come to blame the rebels for inviting government attacks.Footnote ⁹⁸ Underlying each approach is this core metaphor: if Mao was right, and the civilian population is the sea in which insurgent fish swim, then ‘the surest way to catch the fish is by draining the sea’.Footnote ⁹⁹

Alexander Downes offers compelling analysis of ruthlessly effective anti-civilian attacks in the Second Anglo-Boer War (1899–1902).Footnote ¹⁰⁰ The Boers were swiftly defeated in conventional combat, and turned to guerrilla warfare. The British responded by burning the Boers’ farms, and imprisoning their families. Epidemics in the concentration camps killed over 45,000.Footnote ¹⁰¹ The guerrillas were weakened; even Boer Leader De Wet admitted that ‘had not the English burnt the corn by the thousand sacks, the war could have been continued’.Footnote ¹⁰² The camps also coerced the Boers. Acting President of Transvaal, Schalk Burger, said that it was ‘not the arms of the enemy which directly compelled us to surrender, but … the sword of hunger and nakedness, and … the awful mortality amongst our women and children in the Concentration Camps’.Footnote ¹⁰³

The Italians used similar tactics when suppressing the 1923–32 Sanusi uprising in Cyrenaica (now Libya).Footnote ¹⁰⁴ After seven years of ‘fruitless pursuit’, they interned the entire population. The rebels had been well-supported, and the civilians’ imprisonment both removed this support, and offered brutal leverage: between 85,000 and 100,000 entered the camps, but only 35,000 survived. These concentration camps were ‘probably the key to [Italian] victory in Cyrenaica’.Footnote ¹⁰⁵ Similar tactics paid equal dividends for the US in the Philippines at the turn of the twentieth century.Footnote ¹⁰⁶ Anti-civilian attacks appear clearly effective in these cases. It is obviously a stretch to suppose that these are the sort of objectives that could plausibly be deployed in a justified war – partly because if it is necessary to lock up so much of the population in order to subdue them, then the whole project of counterinsurgency is unlikely to be legitimate. But there can be justified counterinsurgencies against a robust insurgent movement, and we can reasonably infer from the success of these tactics in the past that they could work again.Footnote ¹⁰⁷

Anti-civilian attacks can also yield strategic results in interstate conflicts. Sieges, for example, have a long history of success despite being avowedly and intentionally anti-civilian. Consider, for example, the sieges of Paris in the Franco-Prussian war, of Plevna in the Russo-Turkish War, Adrianople in the First Balkan War, and Beirut in the Lebanon War.Footnote ¹⁰⁸ The headline success is probably the Allied blockade of Germany and Austria-Hungary in World War I, which may have caused 400,000 civilian deaths, but surely hastened the German surrender.Footnote ¹⁰⁹ Moreover, even Pape concedes that the threat of nuclear bombardment is a decisive coercive tool.Footnote ¹¹⁰

Additionally, where territorial control is at stake, and the civilian population is united in opposition, their forced removal might itself be a military objective.Footnote ¹¹¹ When two states contest a territory, in which part of the population shares characteristics with the dominant groups in each state, eliminating or evacuating the civilian population affiliated with the opponent can contribute directly to securing territorial control.Footnote ¹¹² Indeed, ejecting the adversary population in part constitutes the military objective itself. It also reduces potential threats from those civilians, eliminating a seat of resistance, as well as a potential trigger for a future rescue mission by the adversary.Footnote ¹¹³ And it facilitates forming a cohesive nation-state, with a stronger claim to the territory, based on active possession.Footnote ¹¹⁴ The Arab/Israeli conflict is a particularly salient example. Downes cites a Zionist leader saying in 1938 ‘we cannot start the Jewish state with … half the population being Arab … Such a state cannot survive even half an hour.’Footnote ¹¹⁵ Expelling the Arab population from their villages was almost certainly wrong, but it is not plausibly described as ineffective in establishing a Jewish state.Footnote ¹¹⁶ Again, some might think this sort of objective could never be aimed at by a justified belligerent; I am less certain: I think there can probably be cases of justified colonisation in which such dilemmas would arise, whether or not we think this was one of them.Footnote ¹¹⁷

Lastly, large-scale quantitative research into the strategic effectiveness of anti-civilian attacks over the last two centuries suggests that it is at least as effective as exclusively counterforce tactics in wars of attrition.Footnote ¹¹⁸ Working from their dataset of around 200 conflicts over 200 years, and focusing on tactics including both intentional attacks on non-combatants and obviously disproportionate collateral harms, Downes and Cochran find that these tactics were ‘probably an important component of victory in 30 per cent of the wars of attrition (29 per cent of states) and at least partially contributed to victory in 45 per cent of such wars (50 per cent of states)’.Footnote ¹¹⁹ They then observe that ‘relative to the rates of victory for states or sides that did not target non-combatants in wars of attrition – 40 per-cent – civilian targeting is not radically less (or more) effective than fighting more conventionally’.Footnote ¹²⁰ As I argue in the next section, states attack civilians when they are already on the ropes, with few other alternatives. We should therefore expect anti-civilian attacks to correlate (much) more often with defeat than with victory.Footnote ¹²¹ Valentino et al. concur on this: belligerents ‘are most likely to resort to killing civilians in the most difficult and desperate conflicts – when conventional military means are ineffective or too costly’ so we can predict that ‘victory in these circumstances will be relatively unlikely no matter what tactics a combatant employs’.Footnote ¹²² If states that victimise civilians in interstate wars achieve the same rate of success as those that do not, that itself strongly implies – given the typically direr straits faced by those who do take these extreme measures – that attacking civilians is militarily effective.

V. … But can it be necessary?

The preceding sections have shown that anti-civilian attacks can be effective in insurgencies and terrorist campaigns, counterinsurgency, and interstate armed conflict. If situations can arise in these conflicts when belligerents have no other effective options besides anti-civilian attacks, then those tactics satisfy necessity, and the necessity-based argument fails.Footnote ¹²³ In this section, I show that in conflicts of these types, belligerents’ option-sets can be sufficiently straitened that anti-civilian attacks become necessary. I discuss terrorist campaigns and insurgencies first, then counterinsurgency, and finally interstate conflict.

For insurgents with territorial objectives – aiming to take control of a territory from an occupying power, or an exogenous national group – the case that there are no other options besides anti-civilian attacks is perhaps weakest. After all, in territorial asymmetric conflicts the stronger actor generally has a significant military presence in the disputed territory, which affords the insurgents many other potential targets besides non-combatants. And one could certainly argue that, in Iraq and Afghanistan for example, counterforce tactics by insurgents have been very effective – partly owing to the occupying powers’ discomfort with military casualties.Footnote ¹²⁴ Moreover, in some asymmetric conflicts non-violent resistance would potentially be still more likely to succeed than either form of military response.Footnote ¹²⁵ The question is not, however, whether anti-civilian attacks are generally the only means to prosecute insurgencies, but whether situations can arise where non-violent and counterforce alternatives are not available or, if available, are not effective. Suppose that we rule out non-violent measures, since if these were effective then there would be no basis at all for using lethal force. Why would we think that anti-civilian attacks might be available when counterforce ones are not? Simply put, because it is much easier to launch a terrorist campaign than a guerrilla war.Footnote ¹²⁶ Confronting an adversary's armed forces requires a level of resources, personnel, and training that an insurgent group may lack. Direct confrontation would ensure a swift end to the insurgency. It is considerably easier to attack targets that cannot fight back.Footnote ¹²⁷ And it is very hard to defend a whole population against a few determined individuals (particularly when they are prepared to lose their lives).Footnote ¹²⁸ It took fewer than twenty primary operatives to bring about the 9/11 attacks; the ten Mumbai attackers held India's economic capital to ransom for a day, killing 164 people; the Oklahoma bomber, Timothy McVeigh, killed 168 and injured 450 others acting with one accomplice. We have already seen that anti-civilian attacks in insurgencies and terrorist campaigns can be effective; now we see that they are easier, less costly than counterforce measures. It should follow that situations arise where the resources, personnel, and training required for an effective guerrilla campaign are lacking, but anti-civilian attacks are still possible. In other words, anti-civilian attacks become necessary.

In general, anti-civilian attacks are clearly not the only way to fight a counterinsurgency. Contemporary counterinsurgency doctrine favours winning hearts and minds, gathering intelligence, and focusing on highly targeted (and high-tech) counterforce measures.Footnote ¹²⁹ The question is not whether anti-civilian attacks are the only way ever to win, but whether situations can arise where these other tactics are unavailable, so the only effective alternative is to turn on civilians.

One reason to think such situations possible is that counterforce and non-violent counterinsurgency are very hard to do. Winning hearts and minds requires enormous financial resources, taking additional risks, and a diplomatic skill-set that we can reasonably excuse soldiers for lacking.Footnote ¹³⁰ Confining attacks to adversary combatants is particularly difficult when they avoid direct confrontations and hide among a civilian population from which they are indistinguishable.Footnote ¹³¹ Sometimes there are no military targets to hit; for example, Michael Gross observes that in the 2006 Hezbollah/Israel war, Hezbollah had only 83 command posts, which were destroyed within five days. Hezbollah continued to strike, but there was no way to strike back, he argues, besides aiming at the civilian institutions on which Hezbollah in part relied.Footnote ¹³² When civilian support for the insurgents is sufficiently high that hearts and minds tactics are doomed to fail, and that the insurgents can benefit from the tactical advantages of hiding among civilians and civilian objects, counterforce tactics can prove extraordinarily costly, in both lives and resources.Footnote ¹³³ If anti-civilian attacks are less costly than counterforce measures, then in some cases anti-civilian attacks will be the only option.

The fact is that well-supported insurgencies are difficult to defeat;Footnote ¹³⁴ all options are poor – even for strong states. Anti-civilian attacks can be effective, they risk little incidental harm to the attackers (because the targets cannot fight back), and they are an easy target – ‘essentially immobile, almost impossible to conceal, and difficult for the guerrillas to defend’.Footnote ¹³⁵ If counterforce measures fail, or become too costly, anti-civilian attacks might be the only viable options. In the Boer war and the Sanusi uprisings, for example, the British and Italians made anti-civilian attacks their focus only after their counterforce measures had failed. Moreover, research by Valentino et al. shows that this trend in fact applies generally.

Using a dataset of 147 wars between 1945 and 2000, they ask why some states mass killed civilians (defined as intentionally killing more than 50,000 over five years or less). Mass killing is most likely, they show, when states face grave threats from an adversary that cannot be defeated by conventional means, but depends on a civilian population.Footnote ¹³⁶ Conflicts against guerrilla opponents who posed a major threat saw a sixfold increase in likelihood of mass killing; where guerrillas were strongly supported by civilians; mass killings were eight times more likely.Footnote ¹³⁷ When guerrillas both posed a major threat, and were well supported, mass killings were 18 times more likely.Footnote ¹³⁸ Although they note that mass killings seem rarely to lead to victory,Footnote ¹³⁹ they conclude that, when states attack civilians,

In other words, states engage in mass killing of civilians because their leaders believe it to be necessary. This should be a chilling conclusion for those who believe necessity grounds non combatant immunity. They may respond that, given the historical record, leaders were unreasonable to believe mass killings could work. However, examples such as Chechnya, the Anglo-Boer war, the Sanusi uprising, and the Philippines tell against this argument. Moreover, the key point is not whether anti-civilian attacks had high prospects of success, but whether they were more likely to be successful than the available alternatives. Even if they are unlikely to succeed, if there is no more promising option, they can still satisfy necessity.

The general point that anti-civilian attacks are less costly than counterforce measures, because the target is unable to defend itself, and unable to hide, and therefore that there can be situations where the additional costs of counterforce measures make them unavailable as options, holds for interstate conflict too. As does the observation that sometimes states adopt anti-civilian attacks after first trying and failing to achieve their objectives through counterforce measures. Even in Pape's analysis of British bombing of German cities in World War II, he notes that anti-civilian attacks were adopted only after attempts to bomb more accurately (for example, by flying during the day) had failed or proved too costly.Footnote ¹⁴¹ Even if the new strategy was ultimately unsuccessful, the lack of viable alternatives might have allowed it to meet necessity, ex ante.Footnote ¹⁴²

Valentino, Huth, and Croco support this general observation with specific data, showing that states attack civilians when they face great threats, and think they have no other options. Looking at all interstate wars between 1900 and 2003, they conclude that the decision to target civilians is based on ‘incentives … created by the risks, threats, and opportunities associated with the particular conflict’.Footnote ¹⁴³ Killing non-combatants ‘is often a calculated military strategy’.Footnote ¹⁴⁴ Even in war ‘few states actively desire to kill enemy civilians in large numbers. But in certain circumstances combatants may come to believe that doing so is the best way to achieve victory at an acceptable cost.’Footnote ¹⁴⁵

They defend this theory by showing that states engaged in wars of attrition or counterinsurgency ‘killed nearly six times more civilians than did states utilising other strategies’.Footnote ¹⁴⁶ When attrition and counterinsurgency combine, the predicted number of civilian casualties jumps exponentially; when these combined with a long war and maximalist war aims, the numbers multiply still more.Footnote ¹⁴⁷ The inference is clear: when states fight longer, more arduous wars, where the threats they face are grave, and where counterforce measures prove ineffective, they are more likely to engage in anti-civilian attacks, as a desperate measure that might be intrinsically unattractive, but is preferable to accepting defeat.Footnote ¹⁴⁸

As with the same arguments in counterinsurgency, this should worry advocates of the necessity-based view. Clearly, states frequently target civilians in war because they believe that doing so is effective, and that they have no other option – in other words, because they believe that doing so is necessary. If the principle outlawing anti-civilian attacks are grounded in necessity, then it denies non-combatants protection when they need it most.

Revisionists might respond that the leaders’ beliefs about necessity are irrelevant – what matters is whether anti-civilian tactics were in fact necessary.Footnote ¹⁴⁹ This raises some interesting and complex questions, that I cannot do justice to here.Footnote ¹⁵⁰ However, two brief responses are possible.

First, regardless of how we should best interpret the necessity constraint, the fact that it provides so weak a bulwark against political and military leaders who believe attacking civilians to be necessary is a serious concern. If the necessity-based argument were the best argument for non-combatant immunity, we should conceal that fact, because it is so vulnerable to abuse by political leaders who believe attacking civilians will be effective.Footnote ¹⁵¹

Second, if the necessity constraint is applied ex post, assuming all the facts are known, then it would have radical implications for the permissibility of killing combatants in war.Footnote ¹⁵² If all the facts were known, then many combatants killed in war die pointlessly – killing them neither averts a specific threat, nor helps avert the overall threat posed by their state. Even for those whose deaths have some use, it is highly likely that there was some other available alternative that would have had the same result, but done less harm.

An example I gave above is relevant here: victory requires capturing the adversary leadership, who are hidden either in town X, Y, or Z. Tactic T_X helps us capture town X, T_Y helps capture town Y, and T_Z helps us capture Z. Our targets turn out to be in Z. If the necessity constraint is applied ex post, then any deaths inflicted in T_X and T_Y with the aim of capturing the adversary leadership were impermissible.Footnote ¹⁵³ Such a restrictive constraint on killing in war would render it all but impossible to fight permissibly. We must either conclude in favour of some form of pacifism, or weaken the necessity constraint. The most plausible way of doing so is applying it ex ante, and indexing it to reasonable beliefs.