Jose M. Cortina, a professor in the Department of Management and Entrepreneurship at Virginia Commonwealth University,* discusses the origins of his paper (co-written with Christian Dormann, Hannah M. Markell, and Sheila K. Keener) “Endogenous Moderator Models: What They Are, What They Aren’t, and Why it Matters,” published in Organizational Research Methods.
This paper was something of an offshoot of a different Organizational Research Methods paper on models integrating moderation and mediation (MIMMs: Holland et al., 2017). That paper was born of the observation that investigators weren’t always clear in their own minds regarding which stage of a mediation model was moderated, what the effect of the moderation was on the total effect, etc. In reviewing literature for that paper, I had noticed that a related sort of model was becoming common, viz. models that contain moderator variables for which at least one cause is specified. Hannah and I toyed with calling them ICE models (Indirect Conditional Effect), which sounds cool but sounds too much like a first-stage moderation model, so we went with Endogenous Moderator Model instead. We invite the reader to name their next child Emm.
It isn’t immediately obvious why inclusion of a moderator’s cause should make much difference. Only when one does the path analytic math does one see that the obvious approach to testing these models doesn’t work. That led Hannah and me to start searching the literature in earnest (wherever that is). We found first that there was almost no guidance on the handling of such models. In the absence of guidance, investigators did what seemed, on the surface, to be correct – they treated these models as if they were first-stage moderation models. This is perfectly understandable. I daresay that I would have treated such models in the same fashion if I hadn’t gone through the tracing rule math.
But the math shows that endogenous moderator models are, in fact, second stage moderation models. The precise type of second stage moderation model depends on the specific role played by the moderator. Anyway, Hannah and I got the paper together and submitted it, but reviewers wanted a more comprehensive review of the literature. So, we enlisted the help of Sheila Keener (née List), without whom we couldn’t have managed. Then the reviewers wanted Monte Carlos that showed the consequences of treating EMMs as if they were first-stage moderation models. Enter my partner in crime Christian Dormann, who did hundreds of Monte Carlos that allowed us to show exactly where the biggest misses were when one adopts common misspecifications of EMMs.
In the end, we had a paper that showed the common treatments of these models, derived the math that demonstrated the true nature of these models, showed the consequences of misspecification, and provided examples and code that allow investigators to generate the numbers that they need. If everyone in our field takes the time to digest our paper, the world will become a veritable utopia in which a man can stand tall, unless of course he is a woman. Or not. But maybe we would get just a little bit closer to understanding what makes organizations tick.
*For more on Cortina’s background and why he’s not a professional golfer, Business and Management INK recommends his “Reflections on Academic Career Choices: What Might Have Been, What Is, and What May Yet Be” in Academy of Management Learning & Education.