REPLICATING AND RECONSIDERING THE PBS FINDINGS

PBS hypothesize that because democracies have superior war-fighting capabilities, conflicts between personalist dictatorships and democracies ought to be especially frequent. They argue that democracies, confident in their abilities to win, challenge personalist dictatorships, the least politically constrained and militarily weakest of the three types of authoritarian regimes (the other two being single-party regimes and military regimes). PBS (19) argue that “(i)f war-fighting capabilities are key, personalist regimes may be precisely the kids of easy targets that democracies are most likely to attack.”

To test this and other hypotheses, PBS examine annual data from 1946 to 1992 for all pairs of states in the international system.¹

We replicate their results in Model 1 in Table 1 (Model 4 in Table 3 in PBS 24). PBS found that the dyads with a democracy and a personalist dictatorship are significantly more likely to have a dispute than other dyads. PBS (25) interpret this as evidence that “strong and capable democracies pick weak personalist regimes as targets.” There is a gap, however, between this conclusion and the nature of their dependent variable. Their dependent variable is the occurrence, not the deliberate initiation, of a dispute between two states. They do not identify which side initiated the dispute. In the PBS data set, an observed dispute between a personalist dictatorship and a democracy may be an instance of the dictatorship challenging the democracy or the democracy challenging the dictatorship. While PBS contend that their results indicate the latter rather than the former, their research design does not allow them to test this claim.

Fortunately, we can test PBS' claim, as the MID data set codes which states are initiators and which are targets (Bennett and Stam 2000; Jones, Bremer, and Singer 1996). We use these initiator data to test the PBS argument. We create a new “directed dyad year” data set, in which a dyad-year between state A and state B appears twice, first as dyad A–B, in which state A is the potential dispute initiator and country B is the potential dispute target, and then as dyad B–A, in which state B is the potential initiator and state A is the potential target. We use a dichotomous dependent variable coded 1 if the potential initiator (the first listed state) in the directed dyad initiates a dispute and 0 otherwise. We used the EUGene program to build the data set (Bennett and Stam 2000).³

Table 1 has the new results. In Model 2, we use our data set of directed dyad years with the PBS nondirected dependent variable. The results are similar to those in Model 1 (the PBS results). Next, we use our new dependent variable, coded 1 if state A in the dyad initiated a dispute and 0 otherwise. We also replace the Personalist Dictatorship–Democracy variable with two dummy variables. The first (Democ/Pers) is coded 1 if state A is a democracy and state B is a personalist dictatorship and 0 otherwise. The second (Pers/Democ) is coded 1 if state A is a personalist dictatorship and state B is a democracy and 0 otherwise. If democracies target personalist dictatorships with a greater likelihood than the baseline probability, then the first new variable should be positive and significant; if dictatorships are more likely to target democracies, then the second new variable should be positive and significant. Note that with the inclusion of the Personalist, Military, Single Party, and Democracy dummy variables, we exclude the category of mixed state pairs other than Personalist Dictatorship–Democracy pairs.

The results in Model 3 overturn the PBS claim about the bellicose nature of democratic states toward personalist dictatorships. Pairs of states in which state A (the potential initiator) is democratic and state B (the potential target) is a personalist dictatorship are not more likely to experience conflict than the reference group, other mixed pairs of states. Conversely, pairs of states in which state A is a personalist dictatorship and state B is a democracy are significantly more likely to experience conflict than the baseline. In short, personalist dictatorships are significantly more likely to initiate disputes against democracies than vice versa. Democracies are not especially likely to initiate disputes against personalist dictatorships. These results do not change if we instead use rare events logit (King and Zeng 2001; Tomz et al. 1999) to account for the dependent variable's highly skewed distribution of being 99.79% zeros and 0.21% ones (results not shown) or if we exclude all dyads that include Iraq.

There is, however, some evidence for the PBS argument that personalist dictatorships fight wars poorly. From 1946 to 1992, 19 personalist dictatorships fought wars.⁵

Are the relationships within other types of democracy–autocracy dyads similar? That is, are military and single-party regimes also more likely to challenge democracies? We look at this possibility in Table 1. We include four dummy variables. The first (Democ/Mil) is coded 1 if state A, the potential initiator, is a democracy and state B, the potential target, is a military regime; the second (Mil/Democ) is coded 1 if state A is a military regime and state B is a democracy; the third (Single/Democ) is coded 1 if state A is a single-party regime and state B is a democracy; and the fourth (Democ/Single) is coded 1 if state A is a democracy and state B is a single-party regime.

The results of this analysis are in Model 4 in Table 1. The Model 3 results do not change. Like personalist dictatorships, military regimes are significantly more likely to challenge democracies. Interestingly, democracies are significantly less likely to challenge military regimes, but democracies are not more likely to challenge single-party regimes. Single-party regimes are more likely to challenge democracies, but the result is of only limited statistical significance (p < .10; one-tailed test). If we instead use rare events logit (Model 5), the results are mostly unchanged, with the exception that Democracies are not significantly less likely to challenge military regimes. Notably, if we drop the statistically insignificant variable of Democ/Single (democracy as potential initiator and single party as potential target), as justified by a log-likelihood ratio test, then the Single/Democ variable (single party as potential initiator and democracy as potential target) becomes positive and statistically significant at the .05 level, one-tailed test. These results hold up whether one uses conventional or rare events logit (result not shown). The general picture is remarkably consistent: Democracies are no more likely than other kinds of states to challenge any of the three types of autocracy, and generally, each type of autocracy is more likely to challenge democracies than they are to challenge other types of nondemocratic regimes. This is quite the opposite of what PBS argue.

IMPLICATIONS

PBS (25) argue that “the increased possibilities for conflict found between personalist dictatorships and democracies provides evidence suggesting that it is the war-fighting capabilities of democracies that best explains the distinctive conflict behavior of democracies,” based on the inference that conflict between democracies and personalist dictatorships means democracies are attacking dictatorships (see also Bueno de Mesquita et al. 1999). While we agree with the core assumption that democracies are better war-fighters than personalist dictatorships, we have shown that the inference PBS draw is incorrect. Specifically, with a dependent variable containing more information we have demonstrated the exact opposite of the PBS inference: Democracies are not significantly more likely to target dictatorships than vice versa, but dictatorships are significantly more likely to target democracies. Our finding that dictators target democracies but not vice versa is consistent with several analyses of different data sets (Bennett and Stam 2000; Gelpi and Grieco 2001; Leeds and Davis 1999; Rousseau et al. 1996).

What theoretical inferences should one draw from these results? First, although different kinds of regimes display varying degrees of warfighting effectiveness, this is probably insufficient to explain how regime type affects the likelihood of violent interstate conflict. While domestic political institutions do affect military efficiency (Reiter and Stam 2002), they are not the most important determinants of victory in war (Stam 1996), but they are among the most powerful predictors of conflict initiation (Bennett and Stam n.d.). The unwillingness of democracies to challenge militarily ineffective personal dictatorships demonstrates this point. This is not to say that democracies do not pay attention to the likelihood of victory when picking their fights. On the contrary, democracies pay close attention to the military power of potential opponents. Indeed, democracies become increasingly likely to initiate disputes as their military advantages over potential opponents increases, though this fact does not completely explain the democratic propensity for victory (Reiter and Stam 2002, 48–51). They may pick on very small states or they may be better able to ascertain which large states have inefficient military organizations and target them.

Second, the evidence presented here strongly supports explanations of the democratic peace based on institutional constraints rather than mutual deterrence. Elected leaders know that to remain in power they must avoid costly or unsuccessful policies, which in turn steers them away from launching risky military ventures. Dictators, however, know that they are unlikely to lose power if they launch an unsuccessful diplomatic challenge or even a losing war short of catastrophic defeat (Goemans 1999). This explains why unconstrained dictators challenge democracies more often than vice versa. Dictators more than other state types use military force short of war as a means to try to push democracies into making substantial concessions over some issue. If war does come, dictators hope to win by simply outlasting their cost-averse democratic opponents (Filson and Werner 2002; Reiter and Stam 2002).

More generally, these results are consistent with a wide array of findings stressing the importance of domestic institutional constraints in the causes and outcomes of international conflict. Specifically, democracies start wars only when they are confident of victory (Reiter and Stam 2002); the greater the fraction of society that votes, the less likely are democracies to initiate MIDs (Reiter and Tillman 2002); democracies fight short wars (Bennett and Stam 1996); democracies seek low-casualty wars (Siverson 1995); democracies are likely to initiate only those international challenges that they are likely to win (Gelpi and Griesdorf 2001); and democracies choose military strategies that promise quick, low-cost victory (Reiter and Meek 1999).