An identified systemic model of the democracy-peace nexus

In previous empirical work, the basis for the proposition that democratic countries do not fight each other has been a single equation regression of hostility on democracy and other variables. This approach is misleading for two reasons. First, peace and democracy are part of a simultaneous system of relations in which they foster each other. Before quantitative inferences which affect policy conclusions can be reached, a separate structural equation has to be estimated corresponding to each of these variables. The equations must be distinguishable from one another at the same time as they embody the interdependence between peace and democracy. Second, previous regression studies emphasized the statistical association between democracy and peace rather than focusing on the substantive magnitude of that effect. To demonstrate the sensitivity of the results to these concerns, we accepted the data and indexes of one of the most influential statements of the peace-democracy thesis, and studied the Cold War period where it is deemed strongest. We differed from the prevailing paradigm by estimating an identified, simultaneous two-equation interactive system. In this more properly specified model, the dyadic democracy-peace nexus generally was not statistically significant and, more important, was very small in its impact. The alternative, peace causing democracy, was much stronger. Neither equation in the simultaneous system explained more than six per cent of the variance, so that other factors are likely to be much more important. Deterrence of aggression and patient negotiation of differences may be a more important guide to public policy than attempts to export western, democratic institutions to other nations. The analysis of these data show that it is more likely that the most important differences that arise between nations are specific to historic epochs and their political and socio-economic conjunctures. Various other indices have been suggested for conflict and democracy in the literature, as well as numerous modifications of the additional variables to be included in the regression equations. The results have tended to be variable with respect to these modifications, but the failure of single equation model specifications to deal with the problem of simultaneous causation makes their results subject to our fundamental methodological criticism and unable to support the burden of policy recommendations.