Dyads Are Dead, Long Live Dyads! The Limits of Dyadic Designs in International Relations Research

Dyads are dominant. After Bremer (1992)’s seminal work, state-to-state dyad years (such as USA-UK-1972) became the standard research design in quantitative international relations.[1] But the continued acceptance of dyads was not inevitable. Time and again, methodologists wrote papers with the potential to kill off dyadic designs, only to pull back and, so long as one adopted the appropriate "tweak", grant dyads a reprieve.[2] 

Consider some examples. Temporal dependency between dyad-years produced inferential errors, but cubic splines offered dyads a second life (Beck, Katz, and Tucker 1998).[3] Quantal response equilibrium logit estimators overcame the inability of dyadic designs to capture strategic interdependence (Signorino 1999), while ``neural networks’’ addressed the non-constant effect of covariates across dyads (Beck, King, and Zeng 2000). A 2001 International Organization symposium highlighted the problem of fixed unobserved differences between dyads, but, once again, the contributors spared properly adjusted dyadic designs from execution.[4] Subsequent work introduced a host of methods to account for various problematic features of dyadic designs: Bayesian bilinear mixed effects models (Hoff and Ward 2004; Hoff 2005; Ward, Siverson, and Cao 2007), spatial-lags (Neumayer and Plümper 2010), randomized testing (Erikson, Pinto, and Rader 2014), community detection (Lupu and Traag 2013), and nonparametric variance estimators (Aronow, Samii, and Assenova 2015). None of these papers advocated completely jettisoning dyadic designs. 



[1] Dorff and Ward (2013) credit Rudolph J. Rummel with introducing dyads to political science research during the 1960s.

[2] Many of the studies cited in the next paragraph offer replications of the Oneal and Russett (1997) study of the democratic peace.

[3] An alternative to the cubic splines solution is provided by Carter and Signorino (2010).

[4] The contributors disagree on the proper means of addressing the problem.  Green, Kim, and Yoon (2001) advocate using a fixed-effects estimator. Beck and Katz (2001) argue that this is a bad idea (due to the loss of observations with no variation on the dependent variable). King (2001) advocates obtaining better data that captures the heterogeneity between observations.

Full Text PDF   Find Replication Data
Discuss this Article
There are currently no comments, be the first to post one.
Start the Discussion...
Only registered users may post comments.
ISQ On Twitter