Meta-studies prove interdependency doesn’t create peace

Christopher Schwartz and Nicholas R. Davis. Davis holds a PhD in Political Science and is a visiting instructor at Marquette University. Schwartz is a Scientist at the University of Wisconsin, Madison. “Decisions, Decisions: Re-examining the Trade-Conflict,” Prepared for the Annual Meeting of the International Studies Association in Atlanta, March 16th– 19th 2016. Panel TB20: The Initiation, Dynamics, and Termination of International Conflict. March 17, 2016. http://democracyobserver.org/papers/schwarz_davis_ISA2016.pdf

Results

The results of our 48 logistic regression iterations appear in the appendix. Due to space concerns and readability, we have outlined our findings in Table 2 below. We enter the logistic regression model results in the table as their temporal domain restriction and trade measure, located under the column that corresponds to spatial dimension choice and row that corresponds to the outcome of interest. Overall, the posterior distribution of results is rather shocking given that our prior expectations almost uniformly supported the OR (liberal) position. In twelve of our fourteen hypotheses we expected a confirmation of the liberal position that trade reduces conflict. All but one (H12) of our fourteen hypotheses fail to find supporting evidence. In all models, we resisted the practice in the literature to use robust standard errors, which should bias our results towards the liberals,9 yet still found no support for the pacifying effect of trade on conflict. We have demarcated the results which had the hypothesized sign but failed to gain statistical significance at p < 0.05.

Our findings can be discussed in terms of our three types of hypotheses. First, the general hypothesis H1 and the the unidimensional design choice hypotheses (H2, H3, and H4) were the a priori most likely to find support. Since we have no findings that support the liberal position in any results, H1 is not supported. Neither is H2, which examines the pre-World War II temporal constraint, nor H3 that represents the broad spatial domain. H4 examines the role of the liberal trade measure, dependence, also with a null finding.

The second group of hypotheses involve bi-dimensional choices. These introduce spatial differences while holding temporal domain constant (H5 and H6) and then explore spatial differences and trade measure differences without specifying temporal domain. As stated above, none are supported. Perhaps the most interesting finding here is that for H7 there was a prior expectation of an equal probability of pacifying and null outcomes. While this is not born out by the results (again, no liberal outcomes supported) the posterior probability of the classical realist position is quite a bit higher than the prior. H9 and H10 explore choices in temporal constraint and measure choice while allowing the spatial dimension to vary. The results for the test of H10, located in Table 6 (model 8) come the closest to a finding supporting the liberal position, but at a lower level of statistical significance.

The third and final group of hypotheses explores how tri-dimensional choices affect outcomes. Again, the prior expectation of support for the liberal position is not realized. These hypotheses (H11, H12, H13, and H14) all select on the spatial, temporal, and measurement dimensions. H12 is the only hypothesis that is supported, as our prior expectation of null findings across each of the trade measures in the post-World War II politically relevant dyad sample is born out in the model results, located in Table 6 models 5 through 8. Notably, the replication of the OR design (H14) has no support. Interestingly, the neoMarxist position finds limited support (not hypothesized) in samples without temporal or spatial constraints for two of the non-OR trade measures, share and flow.

Discussion

Despite trade being recognized as an important factor in the contemporary study of interstate conflict, there is still a debate over whether economic interdependence pacifies, aggravates, or has no effect on interstate relations. Contemporary scholarship has suffered from a persistent lack of consensus in the literature, despite the debate evolving to revolve around how the decisions researchers make affect their results. Even as the literature acknowledges that design choices matter, researchers continue to make heterogeneous choices without clearly stated expectations as to how these choices have constrained their findings. We sought to empirically asses the way authors analyze different groups of states, different time periods, and different measures as they seek to contribute to the overall dialogue on trade and conflict. Rather than approach this literature with a normative bias, we reviewed the major contributions to the literature over the past two decades and began our analysis with a prior expectation that the liberal position would be the most likely outcome supported despite manipulation of design choices. The results presented above reveal that the posterior distribution of results almost entirely lies within the classical realist position (null effect).

As noted by Reuveny, Pollins, and Keshk (2010), the weight of evidence in favor of the liberal commercial peace uses similar statistical models, indicators, and data. As stated above, a number of findings in the literature which are frequently cited as support for the liberal perspective (primarily early OR studies) make research design decisions that that appear not to be advisable. Our prior beliefs have now been updated regarding the validity of their research designs, and their findings surrounding the liberal capitalist peace are overstated. The role of spatial and temporal choices particularly condition results. We hope to eventually identify the temporal or spatial conditions under which the liberal position is supported, but in light of our systematic analysis, including good-faith efforts to replicate typical OR design choices, we can say with a degree of confidence those conditions have not been obtained.