Manna from Heaven or Forbidden Fruit? the (Ab) Use of Control Variables in Research on International Conflict

Abstract

Model specification is a ubiquitous challenge in the social sciences. While this issue of Conflict Management and Peace Science is dedicated to addressing this topic in the context of research on international conflict, the discussion among the contributors also contains general lessons and debates relevant to statistical analyses of political, economic, and social phenomena. The reader might suspect that when Glenn Palmer first approached us with the idea of putting together a roundtable and special issue on using control variables, we yawned. Nothing could be further from the truth. Not only did we think the topic warranted some good debate and dialogue, but we already had in mind some lively, articulate, and bright scholars who would offer clever insight.1 We apparently were not alone in our estimation: a record-breaking, spirited audience attended a roundtable featuring these individuals at the 2004 Peace Science Meeting in Houston, Texas. We are confident that the readers of this volume will similarly be engaged and will find the dialogue on the contributions and weakness of control variables useful for their own research. The origins of this special issue can be traced to James Lee Ray’s 2002 Presidential Address to the Peace Science Society (published in 2003). In that address, Ray argued that international conflict scholars had become obsessed with control variables. He warned us of the perils of interpreting results from models with too many independent variables and set down five guidelines for assessing whether to use controls and how to do so if one must. Returning to that theme, Ray leads off this issue with a paper demonstrating what he would consider the pitfalls of multivariate analysis by examining a particular research program, namely a series of dyadic democratic peace papers. John Oneal and Bruce Russett take Ray to task in the next paper, arguing that their own democratic peace research would not have progressed, especially in terms of being able to evaluate alternative theories, without the use of control variables. In our own contribution to this issue, we make a leap of faith and submit some of our own research on the systemic democratic peace to Ray’s five guidelines,