Democracies are aggressive---this is true in all scenarios, including against other democracies.
Chiba ’21 [Daina and Erik Gartzke; February 19; Associate Professor of Political Science in the Department of Government and Public Administration at the University of Macau, Ph.D in Political Science from Rice University, LL.M in Jurisprudence and International Relations from Hitotsubashi University; Professor of Political Science at the University of California, San Diego, PhD in Political Science from the University of Iowa; Office of Naval Research, “Make Two Democracies and Call Me in the Morning: Endogenous Regime Type and the Democratic Peace,” p. 1-44]
We now turn to the results from the outcome stage, where militarized conflict initiation is regressed on democracy measures and other covariates. The univariate clog-log model 32 that ignores the endogeneity, shown in column (1) in Table 1, successfully replicates the standard, dyadic democratic peace finding that democracies are peaceful, though only toward other democracies. Note that, while individual democracy measures have either a positive or insignificant coefficient, joint democracy has a negative coefficient that overwhelms the positive coefficients of individual democracy measures in the univariate model. As a result, the univariate model produces a result that, while democracy may increase conflict against a non-democracy, it decreases conflict against a democracy.
To illustrate this, we calculate the average treatment effect of joint democracy for the challenger and for the target based on the univariate model. These effects are calculated by comparing the predicted probabilities of conflict initiation when changing the regime type of self (challenger or target) from non-democracy to democracy, holding constant the regime type of the other (target or challenger) as democracy. 33 Gray, hollow circles in Figure 4 show the treatment effects of challenger’s and target’s democracy. We can see that both effects are negative and statistically significant at the 95% confidence level.
Once we correct the endogeneity, however, the data no longer support such conclusions. In column (2) in Table 1, the negative coefficient for joint democracy no longer overwhelms the positive coefficient of challenger’s democracy. Challenger’s democracy now appears to increase conflict even against a democratic target. Red, solid circles in Figure 4 show the average treatment effects of challenger’s and target’s democracy, calculated from the trivariate model. The effect is positive and statistically significant for challenger’s democracy, although the effect is indistinguishable from zero for target’s democracy.
Whether we correct for endogeneity thus makes a significant difference in our estimates of the effect of joint democracy on conflict. The key to understanding why these changes occur lies in the estimated correlations between the error terms for different equations. The estimated error correlation between equations for conflict and challenger’s democracy, 12, is negative and statistically significant. This suggests that unobservable or unmeasured determinants of a country’s democracy make it less likely for that country to attack another country. A failure to control for such factors would generate a negative omitted variable bias, making it look as if challenger’s democracy has a pacifying effect on conflict behavior. On the other hand, the estimated error correlation between conflict and target’s democracy equations, 13, is indistinguishable from zero, suggesting that the endogeneity problem does not seem to operate for target’s regime type.