Notes from the Editors: Redefining some methodological criteria for the journal⋆

Abstract

There are many things in which Operations Management (OM) researchers can take pride. Since the inception of empirical OM, we have rigorously incorporated measurement reliability and validity into our analyses. In many respects, the OM literature is a few steps ahead of its sister disciplines — incorporating measurement error into analyses is perhaps the best example. We have also made considerable progress in terms of theory development, whether by way of case research or purely conceptual and theoretical analysis. Finally, recent developments in the area of problem solving and design science demonstrate that OM scholars are genuinely interested in solving actual managerial problems and remaining practically relevant. These are all reasons to celebrate the progress in empirical OM. But there are a number of blind spots, many of which continue to be reasons for rejections in the manuscript review process. The purpose of this editorial is to describe some of these issues. Specifically, there are a number of misunderstandings about some of the key methods used in manuscripts submitted to us. There are also some outdated practices that we want to discourage authors from using in their manuscripts. These issues are discussed in this editorial, in a roughly descending order of importance. We all know correlation does not establish causality. It is high time we do something about this. We constantly get manuscripts — based on cross-sectional surveys in particular — where the authors make causal claims. We no longer send to the review process manuscripts that uncritically interpret a cross-sectional correlation of X and Y as support of a causal claim, or more mildly, that the variance of X is driving the variance of Y. This applies to both econometric and structural equation models. The problem with assuming that the variance of X drives the variance of Y is well documented. Ignoring the problem often results in over-permissive tests of substantive hypotheses: we see evidence for our hypotheses even when there is not any. We now require all authors to take steps — theoretical or empirical, preferably both — to address the problem of endogeneity. This is now a standard practice in most top-tier management journals, and it is time for JOM, as a premier operations management journal, to follow suit. The literature on endogeneity is massive, going back almost a hundred years. Roberts and Whited (2013) offer a comprehensive summary of the key issues in the context of corporate finance research. All the issues discussed are directly applicable to OM research as well. In a nutshell, the problem of endogeneity is this: when a researcher is using non-experimental data to test the hypothesis that X has an effect on Y, it is possible that the variance of X is not exogenous but endogenous to the model. The end result is that the model is misspecified. This in fact applies not just to cross-sectional but even longitudinal research. Even if X is measured at t-1 and Y at t, there could be an unobserved variable Z that affects X and t-1 and Y at t. In a recent manuscript submitted to us, authors hypothesized that organizational integration drives employee commitment. Integration was assumed exogenous to commitment. This is a very problematic assumption, because we have many reasons to believe commitment could easily drive integration, making the variance of organizational integration indeed endogenous to the model. The consequence of endogeneity is asymptotic bias in parameter estimation. We must come to terms with the fact that plausible claims about the direction and magnitude of an effect cannot rest on an analysis that completely ignores endogeneity. If our inferences are to be biased, they need to be biased toward being conservative. The problem of endogeneity often has just the opposite effect, it inflates our results. We are not aware of any scientific principles that warrant the use of over-permissive inference. Examination of endogeneity starts with a simple question: What is the source of the variance in the exogenous variables in my model? So far JOM authors have been allowed simply to declare that these sources are exogenous to the model. Authors must take steps toward either demonstrating exogeneity or correcting for endogeneity. Both approaches have the common denominator: they call for addressing assumptions that have thus far gone untested. Endogeneity can probably never be completely eliminated from empirical analysis, and it is well known that many "solutions" create more problems than they solve (Murray, 2006). But there are no good reasons to avoid tackling the issue, at least theoretically. If the problem of endogeneity cannot be addressed empirically by testing for it or using instrumental variables or an experimental research design to mitigate it (Roberts and Whited, 2013), we expect at least a theoretical treatment of the topic in all JOM submissions where the general claim that one variable induces varia