Coup-Proofing and Military Inefficiencies: An Experiment

ABSTRACT

Coup-proofing occurs when a leader arranges his military to prevent military leaders from overthrowing him. However, coup-proofing often has the additional effect of lowering the military’s effectiveness in conflict. This article discusses coup-proofing in the context of the Nouri al-Maliki’s regime in Iraq before presenting two formal models. The first model shows when coups are possible, leaders select military commanders with lower ability but higher loyalty. The second model shows that when coups are possible, leaders rotate their military commanders to prevent any one commander from becoming too powerful. The article then presents experimental tests of the models. The results of these laboratory experiments show that leaders are more likely to select loyal commanders or rotate their commanders under the coup treatment relative to groups with no leadership turnover or with leadership turnover according to elections. Thus, when faced with the possibility of a coup, leaders intentionally lower their military effectiveness. This article captures the dynamics behind a fundamental inefficiency introduced into groups when leadership is valuable, delegation is necessary, and powerful subordinates can remove the leader from office.

KEYWORDS:

• Coup-proofing
• coups
• experiment
• game theory

Acknowledgments

I would like to thank Felix Bethke, Kevin Morrison, Jonathan Powell, Kiron Skinner, and Thomas Zeitzoff for helpful comments. Earlier versions of this article were presented at the 2015 Midwest Political Science Association annual conference and the 2015 American Political Science annual conference.

Notes

¹ Egorov and Sonin (2011) argues that the loyalty-competence trade-off is fundamental to dictatorships in particular and extends beyond the military to all positions of power held by subordinates.

² It is important to note that Chiozza and Goemans (2003) find that an increase in the overall risk of losing office makes leaders less likely to initiate conflict.

³ Brown et al. (2016) also use defense burden to proxy for the degree to which militaries have been bought off by the leaders and the presence of a minority regime to account for the role of ethnic politics.

⁴ Harkness (2016) warns that the transition from a broadly recruited army to one relying on ethnic loyalty can increase the chances of a coup in the short term.

⁵ The limits of a laboratory experiment are discussed below.

⁶ However, exceptions are emerging (Bethke 2011; Koga 2014).

⁷ The use of the term “general” here is a convenience and could mean any high ranking officer. Farcau (1994, 44-45, 103) points out that while field grade officers are most likely to undertake the initial planning of a coup, a senior officer must inevitably be brought in for a variety of reasons.

⁸ Having the group engage in war every round biases against coup-proofing in the experiment. The leader, knowing the outcome of the war affects her payoff, is more likely to delegate efficiently than if there were only a chance that the choice of whom to delegate to would affect her payoff. If there was a lower probability of war, the leader would have less incentive to delegate efficiently. Moreover, this setup allows me to show that, in contrast to McMahon and Slantchev (2015), even if there is no disagreement about the nature of the foreign threat, coup-proofing can occur.

⁹ The war outcome here is thought of as a public good and all group members benefit equally. Losing groups receive a positive payoff on the assumption that the leader has provided other public goods or club goods targeted at the military even if the group loses the war.

¹⁰ Powell and Thyne (2011) find almost exactly 50% of coups were successful from 1950 to 2010. Fixing the probability of a successful coup in this model allows inference on the effect of loyalty without complicating this effect by simultaneously varying the probability of success. I assume here that the general and the subordinates necessary to attempt the coup are a unitary actor, and thus the model cannot speak to coups that fail at the plotting stages or how the relationship between generals may affect coup attempts.

¹¹ Targeting public goods in this way is common in Sub-Saharan Africa (Franck and Rainer 2012).

¹² Leaders removed through a coup lose as a form of punishment, but their earnings from the war are not lost. War here is conceived of as a public good, so the leader still benefits from the security of the state. Furthermore, this biases the model against coup-proofing as the leader is not excessively punished for being removed from office.

¹³ She does not receive for a failed coup. This payoff captures the leader punishing the general for attempting the coup by not awarding . However, if the coup is successful, the general that took over power keeps this bonus.

¹⁴ In the model, the leader never loses office due to war. States are rarely existentially threatened by war (Fazal 2004). Meanwhile, Goemans (2008) shows that autocrats, in particular, lower their chance of irregular removal from office by winning wars while losing wars only moderately affects their chances of survival in office.

¹⁵ Leaders enter office at the end of the round and thus have no time to accrue benefits from holding office until the next round.

¹⁶ In each round, a 2-2 tie will occur in equilibrium and the next leader will be decided randomly.

¹⁷ The comparative statics are straightforward: increases in the cost of coup attempts decrease the likelihood of a coup attempt, while increases in the value of and increases in increase the likelihood of coup attempts.

¹⁸ Again, the comparative statics are straightforward: increases in the cost of coup attempts decrease the likelihood of a coup attempt, while increases in the difference between and and increases in increase the likelihood of coup attempts.

¹⁹ If the general from her subgroup has the highest probability of winning, the leader always delegates to that general.

²⁰ The comparative statics are discussed in the appendix. This result assumes that the leader only values the current period. If the leader values the future more, it makes the leader more likely to make a choice that is not optimal for the group.

²¹ This point is consistent with past empirical and formal studies (Powell 2012a; McMahon and Slantchev 2015).

²² An assumption is that battle experience in and of itself is valuable to a general and even if the general loses the war, the general gains skills for future combat.

²³ Increasing the cost of a coup increases the needed threshold for a general to attempt a coup.

²⁴ Comparative statics are discussed in the appendix. Again, this result assumes that the leader only values the current period. If the leader values the future more, it makes the leader more likely to make a choice that is not optimal for the group.

²⁵ The full experimental instructions and screen shots can be found in the appendix.

²⁶ Although 92.4% is quite high, it appears somewhat puzzling that this is not as close to 100% as in the other two treatments. The appendix suggests that a coup attempt by a general from the leader’s subgroup made that general less likely to be delegated to again. This finding may be because the subgroup member lost the trust of the leader after attempting a coup.

²⁷ The lower than expected percentage of elected leaders choosing the best choice is partially accounted for by one leader making 6 of the 21 suboptimal decisions. When this subject is removed from the sample, the percentage of best choices moves up to 92%.

²⁸ In fact, modeling coup attempts revealed a small, insignificant relationship between the probability of a successful coup and whether or not it was attempted.

²⁹ The payout from the lottery has been dropped here since all group members receive it and it does not factor in the strategic decision making.

³⁰ The difference is significant at the level of p < .06.

³¹ The level of significance on the difference not being equal to zero is p < .13.