Browse Subject Areas

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Fortune Favours the Bold: An Agent-Based Model Reveals Adaptive Advantages of Overconfidence in War

  • Dominic D. P. Johnson ,

    Affiliation Politics and International Relations, University of Edinburgh, Edinburgh, United Kingdom

  • Nils B. Weidmann,

    Affiliations Jackson Institute for Global Affairs, Yale University, New Haven, Connecticut, United States of America, Centre for the Study of Civil War, International Peace Research Institute Oslo, Oslo, Norway

  • Lars-Erik Cederman

    Affiliation International Conflict Research, Eidgenössische Technische Hochschule (ETH) Zurich, Zurich, Switzerland

Fortune Favours the Bold: An Agent-Based Model Reveals Adaptive Advantages of Overconfidence in War

  • Dominic D. P. Johnson, 
  • Nils B. Weidmann, 
  • Lars-Erik Cederman


Overconfidence has long been considered a cause of war. Like other decision-making biases, overconfidence seems detrimental because it increases the frequency and costs of fighting. However, evolutionary biologists have proposed that overconfidence may also confer adaptive advantages: increasing ambition, resolve, persistence, bluffing opponents, and winning net payoffs from risky opportunities despite occasional failures. We report the results of an agent-based model of inter-state conflict, which allows us to evaluate the performance of different strategies in competition with each other. Counter-intuitively, we find that overconfident states predominate in the population at the expense of unbiased or underconfident states. Overconfident states win because: (1) they are more likely to accumulate resources from frequent attempts at conquest; (2) they are more likely to gang up on weak states, forcing victims to split their defences; and (3) when the decision threshold for attacking requires an overwhelming asymmetry of power, unbiased and underconfident states shirk many conflicts they are actually likely to win. These “adaptive advantages” of overconfidence may, via selection effects, learning, or evolved psychology, have spread and become entrenched among modern states, organizations and decision-makers. This would help to explain the frequent association of overconfidence and war, even if it no longer brings benefits today.


War is deemed a puzzle because if states were rational, they could avoid the costs of fighting in a pre-war bargain that reflected their relative power [1]. Of course, there are several caveats to such an expectation: strong states may anticipate the spoils of conquest [2], [3], and weak states may anticipate an improvement in their bargaining position, favourable intervention by third parties, or superior strategies and resolve [4], [5], [6], [7]. Nevertheless, war remains puzzling because it occurs even when these caveats are absent or unlikely, and both sides would be better off avoiding violence.

One solution to the war puzzle, long noted by historians and political scientists, is that people and states tend to be overconfident about their chances of success, reducing the perceived costs of war and increasing its perceived benefits [8], [9], [10], [11]. A general bias towards overconfidence has also been noted in economics [12], law [13], management [14], finance [15], and negotiation [16]. Indeed, the phenomenon of overconfidence is a standard result within the psychological literature, which finds that most normal people tend to exhibit cognitive and motivational biases exaggerating their capabilities, their illusion of control over events, and their perceived invulnerability to risk, collectively termed “positive illusions” [17], [18]. If anything, such individual level biases appear to be further exacerbated at the group, organizational and state levels [6], [19], [20], [21], and historical analyses suggest that states and organizations also frequently fail to update their behaviour given past failures [22], [23], [24]. Nobel Laureate Daniel Kahneman recently concluded that, of all the psychological biases he and his colleagues have uncovered over the last 40 years of the “cognitive revolution”, all of them promote hawkish decision-making [25].

There is growing support for a causal link between overconfidence and war: experimental war games found that overconfident individuals were more likely to make unprovoked attacks on their opponents [26]; recent case study analyses found that variation in confidence among state decision-makers during crises correlated with whether or not war broke out [8], [9], [11], [27]; and quantitative analyses of inter-state wars show that initiators have lost one-quarter to one-half of the wars they started since 1500 [28], and this has increased to a majority of wars since 1945 [7], [29].

The larger question that remains unresolved is why people exhibit a bias towards overconfidence in situations of conflict, given that overconfidence appears to invoke significant costs—preventing peaceful outcomes and increasing the frequency, expense, and risks of war (for example, by provoking more powerful opponents) [1], [9]. One possibility is that although overconfidence may sometimes lead to mistakes, on average and over time, overconfidence may in fact promote advantageous decisions, signals or behaviour after all (or may have done so in the past), in which case we would expect to observe it. Although they amount to systematic errors in assessment, positive illusions have been argued to serve the interests of those that hold them because they promote ambition, creativity, persistence, and performance in a variety of tasks and contexts [17]. Recently, it has specifically been suggested that positive illusions were favoured by natural selection because they were adaptive in conflict settings in our evolutionary past, serving to increase resolve and persistence [30], to bluff opponents [30], [31], [32] and, where the stakes are high enough, to exploit risky opportunities that generate a higher net pay off despite the costs of occasional failures [33], [34]. Of course, overconfidence can also be a conscious political tactic to boost morale, rally support, or deter rivals. But, as Robert Trivers has argued, an evolved psychological bias towards genuine, self-deceptive overconfidence would be especially effective because it generates more convincing beliefs and signals, reducing behavioural “leakage” from sham confidence that might give the game away to opponents [31], [32]. Whether overconfidence serves adaptive advantages in the setting of international conflict, however, remains an untested question.


Since it is not possible to conduct real-world experiments on whether overconfidence is adaptive or not in international conflict, an alternative analytical tool is offered by agent-based models (ABMs) [35], [36]. ABMs have enjoyed increasing popularity in recent years as a way of exploring problems that remain beyond the reaches of experimental or analytical methods. Many such models have been inspired by evolutionary approaches [37], in which agents with different attributes compete with each other according to predefined behavioural rules. Successful strategies are replicated and spread while unsuccessful strategies die out, leading to evolutionary change in the proportions of each strategy in the population. ABMs have been applied to a wide range of problems, including studies of cooperation as well as conflict, identifying conditions that give rise, for example, to grouping behaviour [38], moralistic behaviour [39], cooperation in noisy conditions [40], and cooperation in spatial public goods games [41].

For our purposes, ABMs are useful because they allow us to compare the performance of overconfident states in competition with unbiased and underconfident states in a simulated spatial environment. Using a custom written ABM and following previously established protocols [35], [42], [43], we examined the relative performance of states in competition with each other on a 30×30 spatial grid (see Supporting Information for results with alternative parameter settings). The cells of the grid constitute “provinces”, and actors are represented by states of ≥1 contiguous provinces (see Figure 1). At each time step, states assess their neighbours and interact according to predefined decision-rules (described in full below). They attack if they identify a weaker opponent, and conflict outcomes are determined by a function of the warring states' relative resources (R). The process is then repeated over many generations.

Figure 1. Initial starting conditions for a random simulation run, showing the simulation boundary (outer edges), state borders (black lines), and state capitals (black dots).

Colours indicate different types of states (red: overconfident states, green: unbiased or underconfident states).

We operationalize overconfidence by assigning each state a “confidence factor” α. A state's own perceived resource level is given by αR, so states with α>1 are overconfident, states with α = 1 are unbiased, and states with 0<α<1 are underconfident. While states with α≠1 distort the perception of their own strength, other states are not gullible and always see rivals' true strength. This is important for two reasons. First, the psychological literature on positive illusions shows that people tend to overestimate their own capabilities and prospects, but people are not biased when evaluating the attributes of third parties [17], [18], [44]. Second, if other states were gullible and believed the overconfident claims of aggressors, they would simply back off in the face of a bluff and concede, making overconfidence automatically advantageous; obviously we did not want to prime the model towards this trivial outcome.

αR is only used in deciding whether or not to fight. Actual R is used in determining war outcomes. Initial α values are randomly drawn from a log-normal distribution, which bounds values at zero but allows some states to have very high levels of overconfidence (this mimics reality: values less than zero are meaningless, while the long positive tail allows for a few very overconfident states). With an underlying μ = 0 and σ = 1, this distribution means that simulations begin with a population that is unbiased as a whole, with median α = 1.

For each of the initial N states (default N = 50), one province is chosen to host the capital. In each time step, all states synchronously execute 5 sub-procedures [35], [42]: (1) resource extraction phase—the state extracts one unit of resources from each of its provinces and adds it to its current resource level (all states begin life with 10 units per province); (2) decision phase—states assess the probability p of defeating each neighbour, and attack the state conferring the highest p (as long as p exceeds a given “attack threshold” w, default w = 0.5); (3) resource-allocation phase—states divide resources among all “active fronts” (wars with neighbouring states, whoever initiated them) in proportion to the size of each of those states; (4) interaction phase—war outcomes are determined by a logistic conflict success function (CSF) [45]:in which the likelihood p of winning for an attacker depends on the share r of its own resources RA allocated to the fight, compared with the total resources r0 allocated to the fight by both the attacking and target state (RA+RT). r0 controls the proportion of resources at which the odds of winning are equal, and is kept constant at 0.5 throughout all simulations. A slope parameter k determines the decisiveness of conflict (i.e., how much resource differentials influence outcomes; default k = 5). The CSF is public information, so when making decisions about whether or not to fight, states can compute their true probability of defeating an opponent with a given resource level (although a states' perceived resources may differ from their actual resources, if α≠1). The only thing that states cannot anticipate is whether some of their own or their opponent's resources will have to be diverted elsewhere (if they, and/or their opponent, are attacked by a third party and thereby forced to open up a second front in that same time step). This uncertainty is, however, quite realistic (consider the uncertainty about many states' intentions prior to 1914, the US in 1917, Britain in 1938, or Russia, Japan, and the US in 1941). Finally, (5) structural change phase—the winning state gains a randomly selected adjacent province from the loser's territory. Provinces become independent states if their capital is: (1) captured; or (2) geographically cut off [35]. Such “newborn” states inherit the strategy (α) of their former state. No other processes of geopolitical change, such as secession of a subset of a state, can occur in the current model.

Overconfident states overestimate their relative resources (and hence their chances of winning wars against other states, see Figure 2), and will therefore attack more frequently since they perceive more viable targets. This means that overconfident actors should do worse than unbiased or underconfident states because they fight extra wars, and these extra wars will always be against stronger states, which they will tend to lose. The null hypothesis, then, is that overconfident states should be wiped out of the population. However, does this prediction hold in an n-player, spatial setting?

Figure 2. Estimated success probability for various confidence factors (α) in a hypothetical conflict between an attacker and target of equal strength (RA = RT).

The curve shows how the estimated probability of success for actor A changes with its confidence factor α. When α = 1 (signifying an unbiased actor), state A assesses its chances of winning correctly as 0.5. Overconfident states (α>1) overestimate their true probability of winning, and underconfident states (α<1) underestimate their true probability of winning, with asymptotes of winning probability at 0 and 1.


In stark contrast to the prediction, overconfidence consistently emerges as the predominant strategy in the model (see Figure 3; Movie S1 in the Supporting Information shows an example of a single simulation run). This result is robust to large changes in model parameters (e.g. size of the grid, whether it has boundaries or is a continuous surface, initial polarity, k, and distribution of α; see Supporting Information Table S1). However, the best performing states have a middling level of overconfidence: states with extremely high or low α do not perform well, suggesting that there is an “optimal margin of illusion”, as has been suggested in the psychology literature [46]. The superior performance of overconfident states can be attributed to three different phenomena, discussed in turn below: the “lottery effect”, offensive alliances, and attack thresholds.

Figure 3. Results of simulations showing how the confidence of surviving states changes over time.

Panel A shows a single example simulation run, with a typical pattern that the median confidence factor undergoes significant variation over time but stabilizes at a level well above 1. In this case, the strategy that comes to predominate is one that overestimates its resources by a factor of around 4. Panel B shows median confidence factors aggregated over 200 runs of the simulation. Simulations were terminated if there were 50 time steps with no fighting, or if only one state was left. In the majority of cases, only one state remained.

The “lottery effect”

We call the first phenomenon the lottery effect. Even though overconfident states are expected to lose more wars, they also enter more wars than unbiased or underconfident states—effectively “buying more lottery tickets” in the competition for survival. Many of these overconfident states will overreach themselves and be destroyed, but others will, by chance, enjoy consecutive victories and expand quickly. Overconfident states—and their associated high levels of α—are therefore more likely to be represented among the states that survive than are unbiased or underconfident states. If this seems counter-intuitive, note that all states—even overconfident ones—maximize p (the probability of defeating an opponent) in deciding which neighbour to attack (see Methods), so aggressive states tend to choose weaker targets rather than stronger targets just like any other state. However, because they pick more fights overall than anyone else, overconfident states have the best chance of gaining new territory and expanding. By contrast, underconfident or unbiased states are more likely to fight only when they are victims of attack by stronger neighbours and therefore tend to be destroyed by gradual attrition. Overconfident states also benefit from positive feedback: states that gain an early size advantage enjoy increased resources (R), which compounds their advantage as they are increasingly likely to win subsequent wars as a result, as well as accruing relative gains by denying rivals finite resources with which to compete in the future (akin to the advantages of “spiteful” territorial expansion among animals proposed by Verner [47]).

Offensive alliances

The second reason for the success of overconfident states is that their targets will often be the simultaneous targets of other neighbouring states in the same time step (being the weakest in the vicinity). Because such a victim has to split its defensive resources to fight both (or more) attackers, the probability of defeating the weak state is increased even further. Since overconfident states are more likely to attack other states in the first place, they are also more likely to benefit from this effect. Thus, there can be offensive “alliances” that emerge automatically in the model (without any cooperation or planning). Obviously, if the model were contrived to allow defensive alliances, these could help to protect weak states and decrease the advantage of overconfident states. However, what is interesting is that the model shows that offensive alliances can emerge spontaneously on their own, whereas defensive alliances cannot (they would need higher level mechanisms of coordination and commitment to solve the problem of collective action and preventing free-riders).

Attack Thresholds

Another important influence on the relative success of alternative strategies is the attack threshold (w)—the power asymmetry required for states to attack another. All our simulations presented above set the attack threshold at 0.5 (meaning that unbiased states only attack if the odds are in their favour). However, altering w has an important effect on optimum levels of confidence. If w<0.5, then underconfident states are favoured (notwithstanding some persistence of the other phenomena outlined above), because both unbiased and overconfident states would increase their frequency of attacks against stronger states that they are likely to lose against. This would give rise to an especially peaceful world dominated by cautious states. By contrast, if w>0.5, then overconfident states are favoured—even more than they already were in the simulations presented above—because both underconfident and unbiased states would decrease their frequency of attacks against states that they are actually likely to defeat. This would give rise to an especially war prone world dominated by belligerent states. Although overconfidence may be a dangerous strategy for any one state (many such states die), in a world in which wars tend not to happen unless one state has an overwhelming power advantage (w>0.5), it is an overconfident state that is likely to become king. Because of this logic, increasing w makes overconfidence a more successful strategy than it already is. This effect is ironic, because it means that in a world in which states are reluctant to attack without a significant power advantage (w>0.5; which is arguably closer to the real world than the other way around [48], [49]), the states that come to dominate are, paradoxically, more likely to be overconfident (akin to the United States being certain it could defeat North Vietnam or Iraq, but underestimating the costs nevertheless).

Overconfidence versus aggression and risk-taking

Note that the key behavioural difference between overconfident actors and other actors was that they were more likely to attack a given opponent. This suggests that anything that increases aggressiveness or risk-taking—rather than overconfidence—could be advantageous. However, the decision-mechanism leading to overconfidence (and as specified in our model) differs in important ways from these alternatives.

Aggression implies a willingness to attack whether you believe you will win or not. Risk-taking implies a willingness to attack despite a (recognized) low probability of winning. Overconfident states do not fit either description—they only attack when they believe they will win. Overconfidence therefore represents a very different assessment and decision-making mechanism from aggression and risk-taking. Our model focuses specifically on overconfidence, operationalized as states believing they are stronger than they are in reality, and this bias directly influences their decision to fight a given opponent or not.

This leads to a related but separate question. If a propensity to attack other states is advantageous, as it was in our simulations, then is aggression, risk-taking, or overconfidence the best means to achieve this behaviour? We suggest that overconfidence may offer the best proximate mechanism than either aggression or risk-taking, for two reasons. First, overconfidence does take probabilities of winning into account, even if they are somewhat distorted, and this will lead to fewer defeats than a pure aggression strategy which ignores them. Second, overconfidence relies on a very simple rule: overestimating one's strength by a fixed amount. This avoids the need for extensive and accurate information about true capabilities and probabilities of all actors and outcomes typically required of a risk preference approach. Overconfidence is bounded, efficient, and fast—considerations that may have been particularly important if it emerged though an evolutionary process.


Contrary to intuition, a bias towards overconfidence can be an advantageous strategy. Despite wide variations in the basic parameters of our model, overconfident states consistently came to predominate over time at the expense of unbiased and underconfident competitors. The extent to which overconfidence pays off may actually be rather conservative in our model because, in the real world, believing or signalling that one has exaggerated strength through overconfidence can also serve to: (1) deter rivals [30], [31]; (2) attract allies; (3) extract greater concessions in bargaining; (4) increase resolve [17], ; (5) hedge against worse errors [33], [34]; (6) garner public support for war; and (7) win elections. None of these effects are included in our model, but all suggest additional mechanisms by which overconfidence can lead to adaptive advantages.

Overconfidence may seem an implausible strategy because it violates conventional formulations of rationality [1]. However, the appropriate metric of success in competitive situations is “ecological rationality”—the strategy that best exploits the prevailing environment, whatever that strategy may be [37], [50]. It is also important to recognize that overconfidence can spread via more than one mechanism. For example, if the strategy of overconfidence represents an ideology (akin to a gene), and the states represent the entities that carry these ideologies (akin to an individual organism) then, as in conventional natural selection, overconfidence as a strategy can spread even if it causes many of its bearers to die [51]. This is supported by recent modelling which suggests that a trait for risking death in war could arise through cultural group selection where there is strong inter-group competition [52], [53].

Overconfidence is not the best strategy under all conditions. We have already noted that unbiased or underconfident strategies would do better if w<0.5. To further examine the constraints of overconfidence, we added war costs c, making violent interaction between states increasingly expensive. As well as the gains and losses from the outcomes of war (the win/loss of a province, the latter of which already represented a cost of war within our model, but was limited to the loser only), the act of fighting now inflicts a given damage to each opponent, which is deducted from its resource level. War costs for a given state A are determined as a share q of the resources invested in the conflict by its opponent (the target state, T): cA = q * RT, 0<q<1. As before, αRA determines decisions for war, while RA determines war outcomes. Clearly, and unsurprisingly, there are limits on the advantages of overconfidence as the costs of fighting increase (see Figure 4). While this may appear to undermine the adaptive advantages of overconfidence in war, note that: (1) overconfidence can remain the predominant strategy even when war is costly, up to a point; (2) overconfidence would remain the predominant strategy over a larger range of war costs if the attack threshold is increased (w>0.5); and (3) importantly, conflict in our model was always assumed to be zero-sum (one state wins 1 unit of territory at the expense of the opponent losing that 1 unit). In the real world, conflicts are often fought over non-zero-sum stakes, such as resources or land that neither actor owns in the first place. As the ratio of the value of the prize increases relative to the costs incurred in trying to obtain it, again overconfidence would become the predominant strategy over a larger range of war costs. Finally, note that political leaders—those making the decisions for war—do not always expect or personally experience any costs of war, even though they tend to reap its spoils. It is therefore unclear whether the costs of war would necessarily impact on the selection of overconfident traits at all, let alone succeed in counteracting evolved psychological propensities towards overconfidence in the modern world.

Figure 4. Median confidence factor at the end of the simulation decreases as war costs increase (line represents the median of 200 simulations for each level of war costs).

Overconfident states prevail in the population as long as war costs are relatively low. As war costs increase they counteract the favourable effect of overconfidence. Note, however, that this decrease would be offset by altering other settings of the model, such as increasing the attack threshold, w, above 0.5, or allowing the spoils of victory to exceed the losses of defeat, instead of the current setup in which there is a zero-sum gain or loss of a province.

Do the advantages of overconfidence identified in our model have real-world empirical validity? The lottery effect certainly has some real-world analogues. Historically, successful conquerors are typically those that made aggressive moves to gain early footholds, which solidified their own position and disadvantaged rivals, as exemplified in the so-called “scramble for Africa” and the conquest of the Indian subcontinent [54], [55]. Moreover, in international relations theory, Stephen Van Evera cites as one of the four main causes of war the situation in which states see the chance to gain some resource that will facilitate future cumulative gains later on [9].

The power of offensive alliances also has real-world analogues. States and political elites have historically found it easy to combine forces where there is an opportunity to exploit weak rivals, while the converse of forming defensive alliances to help others in danger is extremely hard and suffers from an intense free-rider problem [56], [57]. Although defensive alliances have commonly occurred in the face of great mutual threats (such as against Germany in WW II), they depend on complex agreements or treaties that require considerable coordination and credible signals of commitment. States—especially weak states—often have a greater incentive to “bandwagon” with a powerful aggressor rather than taking the risk of “balancing” against them, since others may defect on such collective action leaving a balancer exposed and vulnerable. International relations theorists have also noted an interaction with the relative advantages of offense and defense. Perceived or actual defensive advantages (where w>0.5) leads to more free-riding and the avoidance of defensive alliances [58]—precisely the kind of world in which we expect overconfidence to spread.

Our model obviously lacks many aspects of realism—for example, states may not always be searching for opportunities to attack each other at every turn, and we do not allow states to learn from their own experience, or from observing others. Allowing periods of peace or learning in the model, however, would not change our results. Reducing the frequency of war would simply slow down the model, but does not alter the differential advantages of alternative strategies when wars occur. Conversely, allowing states to copy the strategies of successful competitors would simply speed up the model, since states would learn that overconfidence is the winning strategy and this would be rapidly copied from state to state rather than having to wait for the selection effects of pre-existing overconfident states to spread, one province at a time. Finally, as discussed above, in the real world expansionist states tend to be balanced by alliances of other powers [57]. However, the interesting thing about this is that offensive alliances can emerge automatically in our model whereas defensive alliances cannot. Defensive alliances require special conditions to be added ex ante. International security regimes such as NATO arise precisely because of the danger of expansionist states and the inability of weak targets to protect themselves without binding prior commitments to solve the collective action problem.

Our model suggests that the broad macro-historical process of inter-state competition may have selected for overconfidence, simply because it was—at some point—a successful strategy in competition with alternative strategies. However, while overconfidence may have been adaptive state behaviour in historical contexts, we do not claim that it remains adaptive behaviour today. While states and their ruling elites may have reaped great personal or national benefits from war until recent times [2], [3], and/or learned that bellicosity is an effective strategy (from their own or other states' histories), wars in the 21st century have increasing domestic, international, and economic costs—low casualty tolerance, norms against conquest, legal responsibilities, intervention by collectives such as the UN or NATO, and the great resolve of nationalist insurgencies [7], [20], [59], [60], [61]. It is therefore likely that, today, overconfidence does nothing more than hinder political adjustment to the increasing costs of modern war. Unfortunately, given a deeply rooted human psychology that tends to bias decision-making towards hawkish behaviour [25], [30], overconfidence is likely to remain a prevalent political phenomenon, even if it causes considerable death and destruction for little gain.

Supporting Information

Movie S1.

Quicktime movie of a single simulation run as displayed in Figure 3a , showing the actual step-by-step interactions of states on the grid.


Table S1.

Simulation results using alternative parameter settings. As reported in the main paper, overconfidence consistently emerged as the dominant strategy in our agent-based model, as long as war costs remained relatively low. Here we show that the predominance of overconfidence is robust to large changes in model parameters. The table reports the results of simulations with all combinations of the following alternative parameter values: (1) the size of the grid (20×20, 30×30, or 40×40); (2) whether the grid was a finite square with borders, or a continuous wrap-around Torus with no borders (yes/no); (3) the initial polarity (number of states) on the grid (10, 50, or 100); (4) the decisiveness of conflict, k (3, 5, or 7); and (5) the standard deviation of the initial distribution of confidence factors, α (0.5, 1.0, or 1.5). The final column in each row displays the median confidence after 50 runs with different random seeds. Each individual run continued until one of two termination criteria occurred: (1) 50 time steps with no fighting; or (2) only one state was left. In the majority of cases, only one state remained. In all cases, the median confidence factor at the end of the simulation was greater than 1.0, corresponding to the predominance of the overconfident strategy (for all the simulation results in the table, summary statistics for median confidence factors are: mean = 1.611, standard deviation = 0.467, range 1.047–3.091). In all cases reported here, war costs were zero (see main text for the effects of war costs), and the initial confidence parameter distribution was set to a mean of zero (which corresponds to an unbiased population on average).



We thank Holly Arrow, Robert Axelrod, Dan Blumstein, James Fowler, Luc Girardin, John Londregan, Daniel Nettle, John Orbell, Robert Trivers, Richard Wrangham and anonymous reviewers for comments, help and criticism.

Author Contributions

Conceived and designed the experiments: DDPJ NBW L-EC. Performed the experiments: NBW. Analyzed the data: DDPJ NBW L-EC. Contributed reagents/materials/analysis tools: NBW L-EC. Wrote the paper: DDPJ NBW L-EC. Designed the agent-based modelling software: NW L-EC.


  1. 1. Fearon JD (1995) Rationalist explanations for war. International Organization 49: 379–414.JD Fearon1995Rationalist explanations for war.International Organization49379414
  2. 2. Mearsheimer JJ (2001) The Tragedy of Great Power Politics. New York: Norton. 576 p.JJ Mearsheimer2001The Tragedy of Great Power PoliticsNew YorkNorton576
  3. 3. Liberman P (1998) Does Conquest Pay? The Exploitation of Occupied Industrial Societies. Princeton: Princeton University Press. 262 p.P. Liberman1998Does Conquest Pay? The Exploitation of Occupied Industrial SocietiesPrincetonPrinceton University Press262
  4. 4. Paul TV (1994) Asymmetric Conflicts: War Initiation by Weaker Powers. Cambridge: Cambridge University Press. 264 p.TV Paul1994Asymmetric Conflicts: War Initiation by Weaker PowersCambridgeCambridge University Press264
  5. 5. Slantchev BL (2003) The principle of convergence in wartime negotiations. American Political Science Review 97: 621–632.BL Slantchev2003The principle of convergence in wartime negotiations.American Political Science Review97621632
  6. 6. Jervis R (1976) Perception and Misperception in International Politics. Princeton: Princeton University Press. 464 p.R. Jervis1976Perception and Misperception in International PoliticsPrincetonPrinceton University Press464
  7. 7. Arreguin-Toft I (2005) How the Weak Win Wars: A Theory of Asymmetric Conflict. Cambridge: Cambridge University Press. 276 p.I. Arreguin-Toft2005How the Weak Win Wars: A Theory of Asymmetric ConflictCambridgeCambridge University Press276
  8. 8. Johnson DDP (2004) Overconfidence and War: The Havoc and Glory of Positive Illusions. Cambridge: Harvard University Press. 288 p.DDP Johnson2004Overconfidence and War: The Havoc and Glory of Positive IllusionsCambridgeHarvard University Press288
  9. 9. Van Evera S (1999) Causes of War: Power and the Roots of Conflict. Ithaca: Cornell University Press. 288 p.S. Van Evera1999Causes of War: Power and the Roots of ConflictIthacaCornell University Press288
  10. 10. Blainey GA (1973) The Causes of War. New York: Free Press. 344 p.GA Blainey1973The Causes of WarNew YorkFree Press344
  11. 11. Ganguly S (2001) Conflict Unending: India-Pakistan Tensions Since 1947. New Delhi: Oxford University Press. 200 p.S. Ganguly2001Conflict Unending: India-Pakistan Tensions Since 1947New DelhiOxford University Press200
  12. 12. Van den Steen E (2004) Rational Overoptimism (and Other Biases). American Economic Review 94: 1141–1151.E. Van den Steen2004Rational Overoptimism (and Other Biases).American Economic Review9411411151
  13. 13. Loewenstein G, Issacharoff S, Camerer C, Babcock L (1993) Self-serving assessments of fairness and pretrial bargaining. Journal of Legal Studies XXII: 135–159.G. LoewensteinS. IssacharoffC. CamererL. Babcock1993Self-serving assessments of fairness and pretrial bargaining.Journal of Legal StudiesXXII135159
  14. 14. Ben-David I, Graham JR, Harvey CR (2006) Managerial overconfidence and corporate policies. I. Ben-DavidJR GrahamCR Harvey2006Managerial overconfidence and corporate policies.Fuqua School of Business, Duke University Working Paper No. 81. Fuqua School of Business, Duke University Working Paper No. 81.
  15. 15. Malmendier U, Tate G (2004) C.E.O. overconfidence and corporate investment. U. MalmendierG. Tate2004C.E.O. overconfidence and corporate investment.National Bureau of Economics Research Working Paper No. 10807. National Bureau of Economics Research Working Paper No. 10807.
  16. 16. Bazerman MH, Curhan JR, Moore DA, Valley KL (2000) Negotiation. Annual Review of Psychology 51: 279–314.MH BazermanJR CurhanDA MooreKL Valley2000Negotiation.Annual Review of Psychology51279314
  17. 17. Taylor SE (1989) Positive Illusions: Creative Self-Deception and the Healthy Mind. New York: Basic Books. 320 p.SE Taylor1989Positive Illusions: Creative Self-Deception and the Healthy MindNew YorkBasic Books320
  18. 18. Fiske ST, Taylor SE (1991) Social Cognition. New York: McGraw-Hill. 672 p.ST FiskeSE Taylor1991Social CognitionNew YorkMcGraw-Hill672
  19. 19. Janis IL (1972) Victims of Groupthink: Psychological Studies of Policy Decisions and Fiascoes. Boston: Houghton Mifflin. 277 p.IL Janis1972Victims of Groupthink: Psychological Studies of Policy Decisions and FiascoesBostonHoughton Mifflin277
  20. 20. Johnson DDP, Tierney DR (2006) Failing to Win: Perceptions of Victory and Defeat in International Politics. Cambridge: Harvard University Press. 360 p.DDP JohnsonDR Tierney2006Failing to Win: Perceptions of Victory and Defeat in International PoliticsCambridgeHarvard University Press360
  21. 21. Van Evera S (1998) Hypotheses on Nationalism and War. International Security 18: 5–39.S. Van Evera1998Hypotheses on Nationalism and War.International Security18539
  22. 22. May ER (1973) Lessons of the Past: The Use and Misuse of History in American Foreign Policy. Oxford: Oxford University Press. 240 p.ER May1973Lessons of the Past: The Use and Misuse of History in American Foreign PolicyOxfordOxford University Press240
  23. 23. Johnson DDP, Madin EMP (2008) Paradigm Shifts in Security Strategy: Why Does it Take Disasters to Trigger Change? In: Sagarin RD, Taylor T, editors. Natural Security: A Darwinian Approach to a Dangerous World. Berkeley and Los Angeles, CA: University of California Press. pp. 209–239.DDP JohnsonEMP Madin2008Paradigm Shifts in Security Strategy: Why Does it Take Disasters to Trigger Change?RD SagarinT. TaylorNatural Security: A Darwinian Approach to a Dangerous WorldBerkeley and Los Angeles, CAUniversity of California Press209239
  24. 24. Van Evera S (2003) Why states believe foolish ideas: nonself-evaluation by states and societies. In: Hanami AK, editor. Perspectives on Structural Realism. New York: Palgrave Macmillan. pp. 163–198.S. Van Evera2003Why states believe foolish ideas: nonself-evaluation by states and societies.AK HanamiPerspectives on Structural RealismNew YorkPalgrave Macmillan163198
  25. 25. Kahneman D, Renshon J (2006) Why Hawks Win. Foreign Policy 34–38.D. KahnemanJ. Renshon2006Why Hawks Win.Foreign Policy3438
  26. 26. Johnson DDP, McDermott R, Barrett E, Cowden J, Wrangham R, et al. (2006) Overconfidence in wargames: experimental evidence on expectations, aggression, gender and testosterone. Proceedings of the Royal Society of London, Series B 273: 2513–2520.DDP JohnsonR. McDermottE. BarrettJ. CowdenR. Wrangham2006Overconfidence in wargames: experimental evidence on expectations, aggression, gender and testosterone.Proceedings of the Royal Society of London, Series B27325132520
  27. 27. Johnson DDP, Tierney DR (In Press) The Rubicon theory of war. International Security. DDP JohnsonDR TierneyIn PressThe Rubicon theory of war.International Security
  28. 28. Wang K, Ray JL (1994) Beginners and winners: The fate of initiators of interstate wars involving great powers since 1495. International Studies Quarterly 38: 139–154.K. WangJL Ray1994Beginners and winners: The fate of initiators of interstate wars involving great powers since 1495.International Studies Quarterly38139154
  29. 29. Lindley D, Schildkraut R (2006) Is War Rational? The Extent of Miscalculation and Misperception as Causes of War. Philadelphia, PA: American Political Science Association Conference. D. LindleyR. Schildkraut2006Is War Rational? The Extent of Miscalculation and Misperception as Causes of WarPhiladelphia, PAAmerican Political Science Association Conference
  30. 30. Wrangham RW (1999) Is military incompetence adaptive? Evolution and Human Behaviour 20: 3–17.RW Wrangham1999Is military incompetence adaptive?Evolution and Human Behaviour20317
  31. 31. Trivers RL (2000) The elements of a scientific theory of self-deception. Annals of the New York Academy of Sciences 907: 114–131.RL Trivers2000The elements of a scientific theory of self-deception.Annals of the New York Academy of Sciences907114131
  32. 32. Trivers R (2011) Deceit and Self-Deception: Fooling Yourself the Better to Fool Others. London: Allen Lane. 320 p.R. Trivers2011Deceit and Self-Deception: Fooling Yourself the Better to Fool OthersLondonAllen Lane320
  33. 33. Nettle D (2004) Adaptive illusions: Optimism, control and human rationality. In: Evans D, Cruse P, editors. Emotion, Evolution and Rationality. Oxford: Oxford University Press. pp. 193–208.D. Nettle2004Adaptive illusions: Optimism, control and human rationality.D. EvansP. CruseEmotion, Evolution and RationalityOxfordOxford University Press193208
  34. 34. Haselton MG, Nettle D (2006) The Paranoid Optimist: An Integrative Evolutionary Model of Cognitive Biases. Personality and Social Psychology Review 10: 47–66.MG HaseltonD. Nettle2006The Paranoid Optimist: An Integrative Evolutionary Model of Cognitive Biases.Personality and Social Psychology Review104766
  35. 35. Cederman L-E (2002) Endogenizing Geopolitical Boundaries with Agent-Based Modeling. Proceedings of the National Academy of Sciences 99: 7296–7303.L-E Cederman2002Endogenizing Geopolitical Boundaries with Agent-Based Modeling.Proceedings of the National Academy of Sciences9972967303
  36. 36. Gilbert N (2007) Agent-based Models. London: Sage. 112 p.N. Gilbert2007Agent-based ModelsLondonSage112
  37. 37. Axelrod R (1997) The Complexity of Cooperation: Agent-based Models of Competition and Collaboration. Princeton: Princeton University Press. 248 p.R. Axelrod1997The Complexity of Cooperation: Agent-based Models of Competition and CollaborationPrincetonPrinceton University Press248
  38. 38. Hammond RA, Axelrod R (2006) The Evolution of Ethnocentrism. Journal of Conflict Resolution 50: 926–936.RA HammondR. Axelrod2006The Evolution of Ethnocentrism.Journal of Conflict Resolution50926936
  39. 39. Helbing D, Szolnoki A, Perc M, Szabó G (2010) Evolutionary establishment of moral and double moral standards through spatial interactions. PLoS Computational Biology 6: e1000758.D. HelbingA. SzolnokiM. PercG. Szabó2010Evolutionary establishment of moral and double moral standards through spatial interactions.PLoS Computational Biology6e1000758
  40. 40. Helbing D, Yu W (2009) The outbreak of cooperation among success-driven individuals under noisy conditions. PNAS 106: 3680–3685.D. HelbingW. Yu2009The outbreak of cooperation among success-driven individuals under noisy conditions.PNAS10636803685
  41. 41. Szolnoki A, Perc M (2010) Reward and cooperation in the spatial public goods game. EPL 92: 38003.A. SzolnokiM. Perc2010Reward and cooperation in the spatial public goods game.EPL9238003
  42. 42. Cederman L-E (1997) Emergent Actors in World Politics: How States and Nations Develop and Dissolve. Princeton: Princeton University Press. 290 p.L-E Cederman1997Emergent Actors in World Politics: How States and Nations Develop and DissolvePrincetonPrinceton University Press290
  43. 43. Weidmann NB, Cederman L-E (2008) GeoContest: Modeling Strategic Competition in Geopolitical Systems. Social Science Computer Review 26: 510–518.NB WeidmannL-E Cederman2008GeoContest: Modeling Strategic Competition in Geopolitical Systems.Social Science Computer Review26510518
  44. 44. Taylor SE, Brown JD (1994) Positive illusions and well-being revisited: separating fact from fiction. Psychological Bulletin 116: 21–27.SE TaylorJD Brown1994Positive illusions and well-being revisited: separating fact from fiction.Psychological Bulletin1162127
  45. 45. Hirshleifer J (2001) The Dark Side of the Force: Economic Foundations of Conflict Theory. Cambridge: Cambridge University Press. 366 p.J. Hirshleifer2001The Dark Side of the Force: Economic Foundations of Conflict TheoryCambridgeCambridge University Press366
  46. 46. Baumeister RF (1989) The Optimal Margin of Illusion. Journal of Social and Clinical Psychology 8: 176–189.RF Baumeister1989The Optimal Margin of Illusion.Journal of Social and Clinical Psychology8176189
  47. 47. Verner J (1977) On the adaptive significance of territoriality. American Naturalist 111: 769–775.J. Verner1977On the adaptive significance of territoriality.American Naturalist111769775
  48. 48. Van Evera S (1998) Offense, Defense, and the Causes of War. International Security 22: 5–43.S. Van Evera1998Offense, Defense, and the Causes of War.International Security22543
  49. 49. Mearsheimer JJ (1989) Assessing the conventional balance: the 3∶1 Rule and its critics. International Security 13: 54–89.JJ Mearsheimer1989Assessing the conventional balance: the 3∶1 Rule and its critics.International Security135489
  50. 50. Axelrod R (1984) The Evolution of Cooperation. London: Penguin. 264 p.R. Axelrod1984The Evolution of CooperationLondonPenguin264
  51. 51. Burt A, Trivers R (2006) Genes in Conflict: The Biology of Selfish Genetic Elements. Cambridge: Belknap Press. 632 p.A. BurtR. Trivers2006Genes in Conflict: The Biology of Selfish Genetic ElementsCambridgeBelknap Press632
  52. 52. Bowles S (2006) Group competition, reproductive leveling, and the evolution of human altruism. Science 314: 1569–1572.S. Bowles2006Group competition, reproductive leveling, and the evolution of human altruism.Science31415691572
  53. 53. Smirnov O, Arrow H, Kennett D, Orbell J (2007) “Heroism” in Warfare as a Functionally Specific Form of Altruism. Journal of Politics 69: 927–940.O. SmirnovH. ArrowD. KennettJ. Orbell2007“Heroism” in Warfare as a Functionally Specific Form of Altruism.Journal of Politics69927940
  54. 54. Pakenham T (1992) The Scramble for Africa. New York: Harper. 738 p.T. Pakenham1992The Scramble for AfricaNew YorkHarper738
  55. 55. James L (2000) Raj: The Making and Unmaking of British India. London: St. Martin's Griffin. 736 p.L. James2000Raj: The Making and Unmaking of British IndiaLondonSt. Martin's Griffin736
  56. 56. Walt SM (1987) The Origins of Alliances. Ithaca: Cornell University Press. 321 p.SM Walt1987The Origins of AlliancesIthacaCornell University Press321
  57. 57. Waltz KN (1979) Theory of International Politics. New York: McGraw-Hill. 256 p.KN Waltz1979Theory of International PoliticsNew YorkMcGraw-Hill256
  58. 58. Christensen TJ, Snyder J (1990) Chain Gangs and Passed Bucks: Predicting Alliance Patterns in Multipolarity. International Organization 44: 137–168.TJ ChristensenJ. Snyder1990Chain Gangs and Passed Bucks: Predicting Alliance Patterns in Multipolarity.International Organization44137168
  59. 59. Mack AJR (1975) Why Big Nations Lose Small Wars: The Politics of Asymmetric Conflict. World Politics 27: 175–200.AJR Mack1975Why Big Nations Lose Small Wars: The Politics of Asymmetric Conflict.World Politics27175200
  60. 60. Beckett IFW (2001) Modern Insurgencies and Counter-Insurgencies; Guerrillas and Their Opponents Since 1750. New York: Routledge. 288 p.IFW Beckett2001Modern Insurgencies and Counter-Insurgencies; Guerrillas and Their Opponents Since 1750New YorkRoutledge288
  61. 61. Mueller JE (2004) The Remnants of War. Ithaca: Cornell University Press. 258 p.JE Mueller2004The Remnants of WarIthacaCornell University Press258