Uncertainty of the Liberal Peace

Abstract

Variable selection is a crucial aspect of formulating a model to empirically examine data, as omitted variables can create spurious association, while inclusion of irrelevant variables can bias the results of one’s estimates. To mitigate such problems, researchers rely on theory to guide their selection of variables to include in their models. Unfortunately, in social science, there often exist several plausible theories to explain actions, and hence several models that researchers can use in their empirical work. This lack of unique theory is evident in examining trade’s effect on conflict, as there are three main theories on this and each suggests a different effect for trade interdependence on conflict. Empirically, the effect of trade on conflict remains uncertain, as researchers Barbieri and Oneal & Russett have come to disparate conclusions using different measures of trade interdependence (models). Each of their inferences is based on the belief that the variables they select form the ‘true’ model that generates the data. The problem is that theory is unable to indicate whether one model is more appropriate than another, which creates uncertainty over the empirical effects of trade on conflict. To account for uncertainty in model selection, the author allows for several models by applying Bayesian model averaging (BMA) to the study of conflict. Accounting for this uncertainty, he finds that trade interdependence does not have a significant effect on the prediction of militarized conflict, whereas joint democracy continues to reduce conflict.