Novelty is Overrated

By Connie Helfat, with comments from Jerry Davis and Myles Shaver

(Note: this article was first published on the Collective Impact Virtual Salon website in December 2022. Reposting here to facilitate commenting and subscriptions.)

Murray Davis’ classic 1971 article “That’s Interesting!” asserts that a theory must be interesting to be considered great. He goes on to say that all interesting theories challenge routinely held assumptions. By implication, counterintuitive theories, which by their definition deviate from common assumptions, are far more likely to be great than theories that accord with our intuition. This makes little sense: counterintuitive theories may be counterintuitive because they don’t accord with the world around us. In other words, they may be wrong.

Nevertheless, management scholars learn early that they must develop novel, preferably counterintuitive, theoretical ideas, and present counterintuitive results. Researchers face pressure to overstate the novelty of their ideas to get published. Even worse, as Rich Bettis observed in 2012, researchers may search datasets for asterisks, i.e., variables with significant p-values, so they can report novel findings—which may well fail to generalize to other settings, in addition to their statistical invalidity.

Many times, I’ve said to graduate students and junior faculty in public forums that “novelty is overrated.” After they get over the initial shock of hearing me say this, they look dubious because it goes against the prevailing norm. What can we do as a field to change this? For starters, we can encourage quasi-replications, stop requiring novel hypotheses, identify stylized facts (empirical regularities) as a basis for theory building, and reward empirical research that poses questions rather than taking the usual approach of testing theory.

Research that poses questions by its nature will deemphasize the importance of novel, often counterintuitive, theory. Qualitative researchers pose questions in their work; there is no reason that quantitative researchers cannot also pose questions. Further, the answers to those questions do not need to challenge routinely held assumptions to be worthwhile. Do we really have much of value to tell the world if we only look for counterintuitive findings—which may be counterintuitive because they are idiosyncratic?

Richard A. Bettis (2012). The Search for Asterisks: Compromised Statistical Tests and Flawed Theories. Strategic Management Journal 33(1): 108-113.

Murray S. Davis (1971). That’s Interesting! Towards a Phenomenology of Sociology and a Sociology of Phenomenology. Phil. Soc. Sci. 1: 309-344.

*****

Comment from Jerry Davis (Dec 2022):

As I say elsewhere on this site, management is the only field I know of where Murray Davis’s article is taught as a how-to guide rather than a cautionary tale about what can go wrong when we write our articles to amuse reviewers and readers rather than to convey findings and insights. If the point of publication is to spark surprise in readers (like a detective novel)…that seems like a sign we have lost our way. Recall that Watson and Crick’s announcement of one of the most fundamental discoveries in the history of science is less than 2 pages long (https://www.nature.com/articles/171737a0). If they had tried to publish it in a management journal, imagine their conversation about how they might “frame” that paper, walk readers down the garden path, and then spring the surprise.

I agree with Prof. Helfat: if we want our science to cumulate, it’s hard to imagine a more nihilistic dictum than to be novel and to demonstrate how things we thought were true are actually false. (On the other hand, if our goal is to amuse each other — keep teaching this article as a how-to guide.

*****

Comment from Myles Shaver (Dec 2022):

I agree with you Connie – to the point that a subtitle for your post could be “That’s Interesting – but not very helpful.”

Your post also highlights and important insight from Arturs Kalnins’ 2018 SMJ paper (Multicollinearity: How common factors cause Type 1 errors in multivariate regression). There is reason to believe that some (maybe many) of the counterintutive findings in our field are statistical artifacts due to multicolinearity (which VIF tests won’t identify). I’d recommend that everyone acquaint themselves with Arturs’ article.
But that makes me think that maybe a better subtitle would be, “That’s Interesting – but not very helpful – and probably wrong”