Rational ignorance: the strategic economics of military censorship.

I. Introduction

News coverage of the Persian Gulf War was the subject of controversial restrictions by the military [1]. Opposition to these restrictions from the journalistic community springs in the first instance from the perception that these restrictions make it more difficult for reports to do their jobs. Part of that opposition may spring from a sense of professional duty; much of it may be part of the character of the journalist. Correspondents, particularly war correspondents, probably would not have chosen their careers unless they received personal as well as professional satisfaction from bearing witness to the action.

Nevertheless, opposition to these restrictions does not refer only to personal and professional satisfaction. Moral force is invoked by appealing to the rights of citizens to know what their government and military are doing. In the language of ethics, citizens cannot fulfill their duties as moral actors without being able to control how others act on their behalf--especially in wartime, when those actions themselves raise profound moral concerns. In the more mundane terms of economic efficiency, an argument would be that a principal (the citizen) is better off if to the degree it can effectively monitor the action of its agents (the government and military).

The leading and obvious argument presented against wartime disclosure of military activities is that such disclosure may inform the enemy as well as the citizens [4]. An informed antagonist may be better able to place its offensive and defensive capabilities to increase the likelihood that the protagonist will fail. A less appealing argument is that publicizing military activities may weaken political support in the protagonist's home country. A democracy's need for disclosure of government activity itself may make it more difficult for democracies to prosecute wars. This difficulty is amplified when altruistic or pacifist citizens have the political wherewithal to limit or halt the action or to punish the military for excessive effort. The Viet Nam experience suggests that social ostracizing of returning soldiers was potential punishment for military effort at least some of the citizens regarded as inordinate.

In principle, the military should not escalate a conflict or employ tactics that a democracy's citizens find reprehensible, even if that means losing the war. An obvious example would be refraining from using nuclear weapons in Viet Nam; perhaps Agent Orange should have been on the forbidden list as well. But public awareness of a military's tactics, when many may be averse to their use, introduces a potential strategic cost. If citizens forbid the military from using a particularly destructive tactic, then the military cannot credibly threaten to use that tactic against an enemy. Knowing this, an enemy that might have been deterred from using extreme measures (e.g., chemical weapons) by appropriate retaliation now may credibly threaten to use them. In some cases an enemy might have been deterred from a conflict altogether, but if the deterrent lacks credibility, it may take its chances and fight.

This raises the possibility that the citizens might choose not to know, to be "rationally ignorant" by means of censoring news of military actions. If the citizens could commit not to care about their military's tactics, the problem would disappear. There is extensive work on the benefits of seemingly irrational behavior that runs counter to one's objectives [3; 9]. It is difficult, however, to imagine how citizens could consciously commit not to have the preferences they have, as they might consciously commit simply to avoid information. The interesting issue is to investigate narrower commitments to resist information consistent with the original preference against extreme measures.

Some differences in the prosecution and the outcome in the Iraqi war as compared with the Viet Nam war may be germane. CNN and the major broadcast networks brought much of the Iraqi conflict into U.S. homes in real time. However, one fact not well publicized was the extent of Iraqi casualties. Only after the completion of Desert Storm did estimates surface of over one hundred thousand Iraqi casualties--on the order of one thousand U.S. conflict-related casualties. by contrast, during the Viet Nam conflict the media supplied "body counts" with the regularity of baseball scores. There are many differences, to be sure, but perhaps the military outcome of the Viet Nam war would have been more in the U.S.'s favor had Americans been willing to inflict casualties in a thousand-to-one ratio. Perhaps Saddam Hussein underestimated American willingness to inflict that order of casualties on Iraqis--or perhaps he overestimated the ability of broadcast news organizations to let Americans know how much human damage its army was inflicting.

The purpose of this paper is to explore the feasibility of rational ignorance as a military strategy. If it is not feasible, an important argument against disclosure grounded on the virtue of keeping the citizens ignorant (as opposed to keeping the enemy ignorant) disappears. First Amendment-based criticism of military censorship, particularly when the enemy already has the information, would be justified. On the other hand, a successful model may identify conditions necessary to justify restricting media access to military information.

We begin with a simple model that illustrates some of the basic criteria necessary for a military censorship policy to be rational. In this model, military censorship restricts access to information about either side's efforts. We will then examine more complex scenarios in which the enemy can supply information regarding its own efforts while the citizen's military controls information regarding war-related damages or its own military effort. To prevent the citizens from inferring what the military's effort must have been as a response to the enemy's expected effort, we must introduce explicit uncertainty, leading to more complex models. We end with some observations regarding further complications and the special conditions necessary to justify censorship.

II. Case I: Ignorance of Enemy Resistance

There are two sides to a potential conflict, the "citizens" and the "enemy." The "citizens" hire an agent to fight this conflict for them; we refer to this agent as the "military." The military and the enemy can choose either high effort or low effort. A scenario in which a commitment for military secrecy is both sensible and increases the likelihood for a favorable outcome from the citizens' perspective may be developed as follows.

Stage 1. The enemy decides whether or not to fight. If the enemy chooses not to fight, the citizens win with no effort on either part. The value of a fight to the enemy depends upon the expected value of the outcome and the costs resulting from the effort expended by both sides.

Stage 2. The citizens decide whether or not to fight back. If the citizens do not fight back, the citizens lose with no effort on either part. The citizens place a value on victory and a cost on the damages suffered on both sides.

Stage 3. The military decides whether or not to make its services available. (We assume a volunteer military.) Its expected compensation must exceed its opportunity cost.

Stage 4. The enemy and the military simultaneously choose their respective military effort levels, which can be high or low. The outcomes of the conflict are ordered pairs ([O.sub.m], [O.sub.e]), where [O.sub.m] and [O.sub.e] respectively are the outcomes for the military and the enemy. The possible outcomes are a win (W) or loss (L) depending upon efforts expended. [O.sub.m] can equal W or L; [O.sub.m] equals W if and only if [O.sub.e] equals L. Payoffs to the military and the enemy depend upon the outcome of the conflict (W or L) and, possibly, the effort each expends.

Stage 5. The citizens compensate or penalize the military on the basis of the outcome and anything else the citizens observe about the conflict.

Following Kreps [5], we can describe the strategic options in extensive form with Figure 1. At decision nodes, E refers to the enemy, C to the citizens, and M to their military. The top dotted line within Stage 4 indicates the enemy's effort level is unknown to the military at the time it elects its effort. This, with placing the military's effort choice after the enemy's, ensures that each chooses its effort without the other's knowledge. The bottom dotted lines at Stage 5 indicate our "rational ignorance" assumption that the citizens do not know after the conflict how much effort the military or enemy expended.

We made the following assumptions:

ASSUMPTION 1. Losing is worse for both citizens and the enemy than not initiating the conflict at all.

ASSUMPTION 2. The military would rather win with a high level of effort than to lose, once conflict starts.

ASSUMPTION 3. To the citizens, the value of victory won with greater damage caused by high military effort is less than the value of a loss under any other condition. To establish the potential for rational ignorance, we make:

ASSUMPTION 4. The citizens cannot commit not to base their compensation on what they learn about the conflict.

ASSUMPTION 5. The citizens observe efforts expended by either side or the damages incurred in the war (as diagrammed in Figure 1).

If Assumptions 3 or 4 do not hold, a threat by the citizens to back high military effort knowingly could be credible. A "rational ignorance" strategy would not be necessary. Assumption 5 specifies the ignorance the citizens could achieve through censorship.

Suppose the military outcomes in Stage 4 are as shown in Figure 2. As this conflict stands, the military can guarantee victory, no matter what the enemy does. From Assumption 2, the military's threat to exert high effort is credible. Assumption 1 implies that the enemy will desist, pre-empting conflict and guaranteeing victory for the citizens (Stage 1).

Deterrence depends upon the credibility of the military's threat to use a high level of effort. If the military were in charge, credibility is ensured under our assumptions. Suppose, however, that the citizens could eliminate the "high effort" option, e.g., through ex post punishment for violating orders. By Assumption 3, citizens would want to penalize a military that exerted great effort. Assumption 4 implies that if the citizens could observe high military effort and punish accordingly, they would do so. This would make the military's threat not credible. As a result, the enemy could guarantee victory by employing a high level of effort. The enemy can achieve its goals without resistance, leading the citizens to refrain from the conflict and accept defeat (Stage 2).

In contrast, suppose the citizens can commit to being censored with regard to their own military's activity. This establishes Assumption 5. As a consequence, the citizens could not reward or punish the military on the basis of the effort they expend. If this were sufficient to make a payment based only on winning or losing credible, the enemy be deterred.

However, other available information can make the threat incredible. In particular, suppose that the citizens could still observe the enemy's effort. If the enemy's were high and the military still won, the citizens could infer (assuming they know that outcome matrix in Figure 1) from a victory that the military exercised a high level of effort. If ignorance is necessary to commit not to punish high effort, this inference eliminates that commitment. Knowing the citizens would punish it for a strong response, the enemy would dismiss the military's threat to respond to high enemy effort in kind. The enemy can thus guarantee victory. Hence, a news blackout covering both the military's effort and the enemy's would be necessary.

For military censorship to matter, the military's preferences for measures must differ, at least in some cases, qualitatively from the preference of the citizens. Obviously, if citizens and the military shared the same aversion to high effort, even if that were necessary for victory, citizen ignorance could not make a strong military response credible. However, suppose that the citizens and military would rank tactics in the same order. Then, the citizens could limit the ex post payment to a victorious military so that the military would not find it worth while to employ measures undesired by the citizens. This payment scheme would render incredible military threats to use these measures, even if the citizens were ignorant about the specific tactics the military uses. Since our model begins with the assumption that citizens care about tactics, we cannot assume they would avoid such payment schemes.

We therefore need to justify the possibility that the military and citizens have different preference orderings. The most direct way to do this would be through the following:

ASSUMPTION 6. The number of enemy casualties is an increasing function of both military and enemy effort.

ASSUMPTION 7. Citizen utility is decreasing as a function of enemy casualties; military utility is non-decreasing, and possibly increasing, as a function of enemy casualties.

Assumptions 6 and 7 together qualitatively specify how military effort affects citizen utility. Assumption 7 implies that a punishment mechanism in which the citizens penalize the military on the basis of enemy casualties could render a high effort threat not credible. It also implies that if citizens are ignorant of casualties or the military effort responsible for them, the citizens cannot design a compensation scheme to deter the military from exerting a high level of effort and inflicting heavy enemy losses once the conflict begins.

III. Case II: Ignorance of Circumstance

In the above model, commitment to remain ignorant of battlefield operations is the difference between certain victory and certain defeat. Credibility of that commitment requires not only direct censorship regarding the citizens' own military activity, but censorship of media coverage of enemy activity as well. But the enemy may be able to thwart such censorship by supplying information to the citizens. (This, perhaps, was the basis for criticism of CNN reporter Peter Arnett's wartime reports from Iraq.)

We can construct models in which the citizens learn the activity of the enemy. In the model above, if the citizens high enemy effort they can infer how much effort their military exerted from the outcome. For censorship to be strategically effective, there must be some other source of uncertainty. To introduce such a source, we posit that after Stage 3 and before Stage 4, "Nature" chooses between alternative states of the world. We introduce two potential sources of uncertainty--that states of the world affect the level of damages, and that they affect the level military effort necessary to achieve a win for the citizens.

Level of Damages

We can indicate the introduction of uncertainty between Stages 3 and 4 as Stage 3a. To model uncertainty regarding level of damage or casualties, we can add to our model:

Stage 3a. "Nature" chooses one of two states of the world, [S.sup.l] or [S.sup.h]. For any given levels of effort by the military and civilians, damage is lower in [S.sup.l] than in [S.sup.h]. The ex ante probability that Nature chooses [S.sup.l.1] is p; the probability that Nature chooses [S.sup.h] is 1 - p. The enemy, military, and citizens know p before the conflict takes place.

We retain the earlier assumptions except for the following modifications of Assumptions 3 and 5:

ASSUMPTION 3'. The citizens value victory won with high effort exerted on both sides less than a loss only if the high damage state [S.sup.h] occurs.

ASSUMPTION 5'. The citizens may observe the effort exerted by the military and the enemy, but they can observe neither damages directly nor whether the state of the world is [S.sup.l] or [S.sup.h.].

By Assumption 3', the citizens would penalize their military sufficiently to lose the conflict if [S.sup.h] occurred. But Assumption 5' guarantees that the citizens cannot base post-conflict rewards or penalties to the military on actual damages, only on the expected damage given the effort exerted. Suppose that:

ASSUMPTION 8. The probability of the low damage state [S.sup.l], p, is sufficiently large that the citizens value a victory with expected damages, even with high effort levels on both sides, greater than a loss.

If Assumption 8 holds, the threat of citizen punishment will not deter the military from exerting great effort. If the outcomes are as Figures 2 describes, a high military effort would remain credible, deterring enemy aggression. The strategic gains from censorship, however, follow only if the military would know prior to choosing its level of effort how severely the citizens would penalize it for high damages if the citizens were to learn either the level of casualties or the state of the world during the conflict. Otherwise, the military has no knowledge to censor. To establish the case for rational ignorance, therefore, we need:

ASSUMPTION 9. The military and enemy choose their efforts in Stage 4 knowing whether the state of the world is [S.sup.l] or [S.sup.h].

It is just this knowledge the citizens would want to commit not to know, so, that the military could establish a credible threat.

Level of Necessary Effort

A second possibility is that citizens would know damages given the level of combat, but that they would not know the level of military effort necessary to secure victory. In one state of the world, labeled [S.sup.m] to indicate that it favors the military, the military would win the conflict with a low level of effort regardless of enemy effort. In the other state of the world favoring the enemy, [S.sup.e], a high level of military effort is necessary to ensure victory. Figure 3 illustrates the possibilities.

We describe the citizens' uncertainty by replacing Stage 3a described above by:

Stage 3b. "Nature" chooses [S.sup.m] or [S.sup.e]. The ex ante probability that Nature chooses [S.sup.m] is p; the probability that Nature chooses [S.sup.e] is 1 - p. The enemy, military, and civilians know p before the conflict takes place.

To ensure the possibility that the military might use a low level of conflict, we add the following assumption.

ASSUMPTION 10. In state of the world [S.sup.m], the military would rather win with a low effort level than a high effort level.

This assumption may be justified by the military's interest in minimizing its own casualties.

To make the information structure appropriate to this case, we retain our original Assumptions 3, that the citizens would rather lose than have their military exert high effort. However, we modify Assumption 5 again to say:

ASSUMPTION 5". The citizens may observe the enemy's effort, but they can observe neither their own military's effort nor whether the state of the world is [S.sup.m] or [S.sup.e].

T enemy may be able to advertise its efforts to the citizens--the energy's enemy--but the military may be able to control access of information about its own efforts. Citizen ignorance of the state of the world prevents them from inferring the level of their military's effort or from designing state-contingent payments. Suppose, however, the p, the ex ante conditional probability that the military would use low effort in a war, is sufficiently large. it would then be reasonable to modify Assumption 8 from the previous scenario to read:

Assumption 8'. The expected value of a victory to the citizens is greater than the value of a loss, even though there is a positive chance (1 - p) that the military will win the conflict only with high effort.

With this assumption, the citizens would feel that a war is worth prosecuting even if there is some chance that the effort used would be excessive. If the enemy knows that military censorship will be sufficiently effective to keep the citizens from punishing the military after the conflict for excessive effort, it will regard the military's threat of high levels of effort as credible and, presumably, desist from aggression.

IV. Concluding Observations

We can identify some necessary components of a "rational ignorance" scenario. Since the citizens are what the military does, ignorance ex ante is not good enough. They have to be unable to infer ex post what kinds of tactics the military used. Consequently, censorship must limit the ability of the citizens to discriminate between different victorious outcomes, or to learn what the actual combat circumstances were like. For example, the citizens may only learn who won or lost, but not by how much. They may also not be able to know how much effort the enemy put into the battle. The more precise is the information they have following the conflict regarding the battlefield situation, the more likely will they be able to determine whether the military exerted an undesirable amount of effort. To take a recent potentially relevant example, suppose the enemy knows that the military will use chemical weapons if the enemy uses them. If the enemy knows that the citizens know this, then the citizens will now ex post that its soldiers used chemical weapons if they learn that the enemy did so. Hence, if the citizens were to learn that the enemy used chemical weapons, and if the military expects punishment for using chemical weapons, then it could not credibly threaten to respond in kind.

The "rational ignorance" scenario also requires that the citizens cannot commit ex ante not to penalize the military for intolerable tactics discovered ex post. If the citizens can make such a commitment, their situation would be identical to one in which they would not learn the information at all. Commitment to ignorance may be necessary just because they can not commit not to punish the military if they learn of extreme measures or damage. One can view "rational ignorance" as the means to commit to a payment dependent upon outcome but independent of military measure used.

Even if these requirements are met, real-world conflicts and media coverage may not fit the structural requirements necessary to make censorship commitments effective. Most potential armed engagements are not the simple games modelled here, with stages and well-determined outcomes. Military conflicts are not one-shot games, but take place over time. Reactions on each side depend upon the prior history of the conflict as well as expectations regarding present and future conduct. The chance that a military censorship policy will get an enemy to desist from aggression seems smaller than our simple models suggest. It follows that censorship may not bring significant strategic gains.

An additional drawback to the censorship strategy is that it is likely to be ineffective tactically as well as strategically during a protracted struggle. The information set becomes riche as a military conflict becomes more drawn out. The ability of aggressive journalists to ferret out pieces of information, and for media to transmit it to citizens able to infer enemy conduct, military conduct and level of damages, are likely to grow over time. If the enemy determines that the citizens would learn enough about effort or damages to rein in the military, censorship cannot make powerful threats credible.

Information plays a crucial role in supporting the equilibrium. As Kreps [6] has pointed out, however, the reasonableness of an equilibrium depends, somewhat perversely, on what the participants believe about events that should not occur. Our first case clearly illustrates this. We found in that model that if the citizens remain ignorant of effort levels, a military threat to respond with great force becomes credible, thus deterring enemy conduct. However, suppose the enemy were to attack, despite the apparent loss. As characterized by Kreps, an "impossible event" may arise as a simple unintended mistake, through misunderstanding of the game, or perhaps to establish a reputation for future purposes. Suppose that if the enemy made a mistake and began the conflict (Stage I) it would attack only at a high effort level (Stage 4). From Figure 2, the military would match the high effort to prevent defeat. Therefore, the citizens would know that if the enemy made a mistake, that would reveal that the military used high effort. But this implies that a deviation from the equilibrium by the enemy--an attack--would not be repelled, since the citizens would infer merely from the fact of the attack that high effort was used. If the enemy knew, this it would choose to attack, since the citizens would infer that their military used high levels of effort. Thus, "out-of-equilibrium" beliefs that mistaken attacks would involve high levels of effort would fail to support the equilibrium. Censorship would be pointless.

These game theoretic considerations also speak to the ethical issues regarding military censorship. Despite its instrumental virtues, censorship may be unethical per se, keeping citizens from access to information needed to make responsible political decisions and conduct effective political debate [7; 8]. A commitment by citizens to keep themselves in the dark prevents them from exercising their democratic responsibility to control the actions of their government. However, if censorship is an effective policy, it would never have to be undertaken. With military conflict deterred, there would be nothing to censor. One may analogize censorship in this regard to nuclear deterrence. There, a commitment to do something immoral--destroy another country with nuclear weapons--would produce desirable circumstances under which the immoral action never takes place [10]. But even if censorship is not supposed to happen, out-of-equilibrium actions may in fact bring it about.

Therefore, the belief that military censorship can prevent conflict does not mean that censorship will never occur. Despite the potential theoretical benefits of rational ignorance, conflict with political responsibility to maintain a watch on the actions of the government may not be avoidable. Other diplomatic and domestic situations may require a credible ability to keep the military under control This does not mean that the tactic has not been employed to some effect, perhaps even in the recent conflict with Iraq. As General Schwarzkopf said, "I'm anti-body-count. Body count means nothing, absolutely nothing" [2, 92]. The strategic motive behind this expression may not have been that the public would have been upset with a low ratio of enemy losses to ours. Rather, a high ratio might have fueled political opposition to the war, reducing credibility of the coalition's threats against Iraq. While the credibility of that particular threat may well have served morally defensible ends, perhaps the most serious problem with censorship is that, if effective, we may never know whether self-imposed ignorance served good ends or bad ones.

References

[1]--, "Dupont Forum Critiques Gulf Coverage." Broadcasting, February 4, 1991, 39, 46. [2]Drew, Elizabeth, "Letter from Washington." New Yorker, February 25, 1991, 90-95. [3]Frank, Robert H. "If Homo Economicus Could Choose His Own Utility Function, Would He Want One with a Conscience?" American Economic Review, September 1987, 593-604. [4]Haiman, Franklyn S. Speech and Law in a Free Society. Chicago: University of Chicago Press, 1981, 388-409. [5]Kreps, David M. Game Theory and Economic Modeling. Oxford: Oxford University Press, 1990. pp. 13-21. [6]--. "Out-of-equilibrium Beliefs and Out-of-equilibrium Behavior," in The Economics of Missing Markets. Information, and Games, edited by F. Hahn. Oxford: Oxford University Press, 1990. pp. 7-45. [7]Rawls, John. A Theory of Justice. Cambridge, Mass.: Harvard University Press, 1971, pp. 224-25. [8]Scanlon, Thomas. "A Theory of Freedom of Expression," in The Philosophy of Law, edited by R. Dworkin. Oxford: Oxford University Press, 1979, pp. 153-71. [9]Sen, Amartya K, "Rational Fools: A Critique of the Behavioral Foundations of Economic Theory." Philosophy and Public Affairs, Summer 1977, pp. 317-44. [10]Tucker, Robert W, "Morality and Deterrence." Ethics, April 1985, 461-78.