Loyalty for sale? Military spending and coups d’etat

For example, see the seminal papers by Acemoglu et al. (2001) and Jones and Olken (2005).

We define coups as attempts to overthrow the government by a small military coalition. As it will become clear later, coups are quite distinct from civil wars.

As part of their seminal research on leadership, Jones and Olken (2009) use the outcome of assassination attempts to estimate the impact of changes in leadership on institutions and war. They restrict their sample to cases in which a weapon was used (e.g., the gun was fired, the bomb exploded), so that the outcome of the assassination attempt can be taken as random.

We look at a number of other variables one and three years after a coup, and check whether the means differ significantly depending on the coup’s outcome. We find no significant differences across success and failure except in institutions. This suggests that other policies and outcomes do not change differentially depending on a coup’s outcome.

We have tried to follow the growth regression literature as closely as possible, and hence our use of five-year periods. By looking at a five-year interval we allow for a window during which military spending may have an effect on coups.

We also find that income measures have a significant negative relationship with coups when no fixed effects are entered, which is consistent with the literature (e.g., Londregan and Poole 1990). However, this result vanishes when we enter fixed effects.

There are a number of alternative mechanisms that could connect military spending and coups, and our findings allow us to refute a number of them (as we discuss in Sect. 6).

However, he finds that changes in military spending do not affect the probability of a coup attempt.

Tables A1–A4 in the online Appendix (http://​www.​caths.​cam.​ac.​uk/​personal/​gleon/​loyalty_​appendix) summarize the data in more detail.

Table A5 in the online Appendix (http://​www.​caths.​cam.​ac.​uk/​personal/​gleon/​loyalty_​appendix) shows that most coups are staged against non-democracies. However, whether a coup succeeds or fails appears to be unrelated to whether the target regime is democratic. Perhaps not surprisingly, the vast majority of successful coups lead to non-democratic regimes, while most failed coups result in the regime type remaining unchanged.

Table A6 in the online Appendix (http://​www.​caths.​cam.​ac.​uk/​personal/​gleon/​loyalty_​appendix) shows average military spending as a percentage of GDP by country-year, separated by region and decade.

That is, it allows us to say that if one of the successful coups had instead failed, military spending would have changed as in the control group of failed coups.

We use the binary variable for democracy created by Cheibub et al. (2010). As they have argued in this journal, changes within the middle range of the Polity scale are difficult to interpret, a problem they avoid by defining a binary variable. Not all non-democracies are military regimes; for example, there are a large number of civilian autocracies.

The online Appendix can be found at http://​www.​caths.​cam.​ac.​uk/​personal/​gleon/​loyalty_​appendix.

In Table A7 in the online Appendix (http://​www.​caths.​cam.​ac.​uk/​personal/​gleon/​loyalty_​appendix) we show the results when we include coups in consecutive years. We count these coups as one event, so that the pre-coup year is the one before the first coup, and the post-coup year is the first after the last coup. We treat the event as successful if any of these coups succeeded. We also include a control variable equal to the number of years included in the event. Again we find a significant difference in the change in military spending depending on a coup’s outcome. In Table A8 in the online Appendix (http://​www.​caths.​cam.​ac.​uk/​personal/​gleon/​loyalty_​appendix), columns 1–4 repeat the main estimation but include data from three years before and three years after a coup. This enables us to better estimate the trends in the data, but the number of coups goes down to 80. (We cluster the standard errors to correct for serial autocorrelation, as suggested in Bertrand et al. 2004.) Once again, we find that successful coups lead to changes in military spending that are larger than the changes following failed coups. These results are significant in columns 1, 2 and 4 (and the p-value for the interaction term in column 3 is 0.11). The estimated magnitudes are similar: roughly 1 percentage point in columns 1 and 2 and over 30 % in columns 3 and 4. In columns 5–8 we allow for regional time trends, and we find that our estimates remain largely unchanged once we include these trends.

Tables A9 and A10 in the online Appendix (http://​www.​caths.​cam.​ac.​uk/​personal/​gleon/​loyalty_​appendix) show that the effect is entirely due to changes following successful coups against non-democracies; changes in military spending following successful coups against democracies are indistinguishable from those following failed coups.

Table A11 in the online Appendix (http://​www.​caths.​cam.​ac.​uk/​personal/​gleon/​loyalty_​appendix) replicates this table but looking at the means 3 years after a coup.

The results are similar if we shift the start of the five-year periods one year backwards or forward. The year dummies are for these seven years.

We use the binary variable for democracy created by Cheibub et al. (2010). In the online Appendix (http://​www.​caths.​cam.​ac.​uk/​personal/​gleon/​loyalty_​appendix) we show that our results are robust to using the polity2 measure of democracy instead of the binary variable.

This eliminates the bias that would arise from the omission of time-invariant country-specific characteristics, which include institutions, whether the country is a primary commodity exporter (O’Kane 1987), and ethnic fractionalization (Jackman 1978). In recent years new evidence has come to light suggesting that existing cross-country correlations disappear once fixed effects are entered. For example, Acemoglu et al. (2008) show that fixed effects eliminate the observed cross-country correlation between income and democracy.

See Table A12 in the online Appendix (http://​www.​caths.​cam.​ac.​uk/​personal/​gleon/​loyalty_​appendix) for more details.

In both cases an F-test rejects the null hypothesis that all fixed effects are equal, suggesting that they need to be included.

A surprising result is that income has no impact on coups once fixed effects are entered. It is also interesting that the size of the military and the type of regime (whether democratic or whether a military dictatorship) lose their predictive power in the presence of country fixed effects. Only instability remains significant, with its coefficient largely unchanged. There is the concern, however, that this variable is mechanically related to coups. The results are largely unaffected when this variable is removed.

The coefficients are insignificant and have a negligible impact on our estimates, while the inclusion of a lagged dependent variable introduces a number of econometric complications. For example, Nickell (1981) showed that a lagged dependent variable causes the parameter estimates to be inconsistent when fixed effects are entered.

In Table A13 in the online Appendix (http://​www.​caths.​cam.​ac.​uk/​personal/​gleon/​loyalty_​appendix) we show that the results are robust to using the polity2 variable as a measure of democracy (instead of a binary definition). In Table A14 in the online Appendix (http://​www.​caths.​cam.​ac.​uk/​personal/​gleon/​loyalty_​appendix) we repeat the estimation but separate the impact of military spending depending on whether the country is democratic or not the year before the five-year period begins.

Although it could be possible that successful coups lead to changes in policy that completely prevent these shocks, this seems unlikely.