All methods have their limitations. Our inferences–whether design—or model-based—rely on a set of assumptions, only some of which are testable. It follows that we can scrutinize any commonly used method in terms of the plausibility of its assumptions; these assumptions tend to be more heroic under some research designs than others. In our view, the debate over dyadic design in International Relations (IR) matters a great deal. Although we agree with some of the criticisms advanced in this debate, we oppose abandoning the use of dyadic design altogether. Work in the field of international trade shows how the integration of theoretic and methodological advances in IR may provide a productive way forward.
The now-sizable literature on the use and limitations of dyadic data in political science has motivated the adoption of better practices in specifying models that treat the dyad as the unit of analysis. Nevertheless, Cranmer and Desmarais (2016) correctly question the plausibility of the assumption of conditional independence even despite recent advancements. We would also hasten to add the important assumption of conditional ignorability of "treatment assignment" (conditional on covariates) to generate unbiased estimates of "treatment effects" in dyadic models. Failure of either assumption to obtain (among others) may bias our estimates.
So where do these issues leave practitioners? Greater awareness of the limitations of those assumptions that undergird dyadic design should drive the adoption of different approaches when these prove feasible. The series of recent developments outlined in the present articles—including ERGM models, k-adic models, and other methods to improve inference with dyadic data—showcase some of the ways forward. Nevertheless, these discussions reveal outstanding issues. For example, the exaggerated Type-I error rate of all of the estimators assessed in the Monte Carlo analyses by Cranmer and Desmarais indicates much room for further development.
However, it should also lead us to better understand the limitations of our inferences and to find ways to research using dyadic designs.The use of (dyadic) gravity models in the trade literature in International Political Economy (IPE) exemplifies this approach. These gravity models have played a central role in IPE for decades. They challenge Cranmer and Desmarais' assertion that the availability of data drives theoretical innovation toward dyadic analysis. Gravity models of trade derive explicitly from longstanding theoretical models of trade. That is, they closely map theoretical models of trade to research design (SeeHead and Mayer  for an overview of this literature). Since existing formal theories of trade tend to feature two players, the possibility of amending models to account for potential "hyperdyadic" influences may present a theoretically fertile approach. These are precisely the influences underlying Cranmer and Desmarais’ concerns about the potential violation of the assumption of conditional independence in empirical models. Therefore, our sense is that theory should guide the choice of empirical strategy in this domain—and, in doing so, facilitate the best use of new methods.
In sum, the contributions by Cranmer and Desmarais (2016), Poast (2016), andDiehl and Wright (2016) further an important debate on the use of dyadic data in IR. The thorough discussion of the issues and limitations inherent to dyadic analysis provides a helpful resource to practitioners and methodologists. But the practicable implications for empirical IR scholarship remain less clear. Certainly, practitioners should be aware of limitations of the empirical strategies and models they utilize. At the same time, it is not evident that we have arrived at—or will soon arrive at—a one-size-fits-all solution to the ills of dyadic data. In this respect, work on international trade may provide a useful model, as it closely matches theoretical propositions to research methodologies.
Head, Keith, and Thierry Mayer. "Gravity equations: Workhorse, toolkit, and cookbook." (2013). CEPR Discussion Paper No. DP9322