So some interesting off-line responses to my post on the proposed journal LR ≠1J.
Some commentators mentioned pre-study registration of designs. I agree that’s a great practice, and while I mentioned it in my original post I should have linked to the most ambitious program, Open Science Framework, which integrates pre-study design registration into a host of additional repositories aimed at supplementing publication as the focus for exchange of knowledge among researchers.
Others focused on efforts to promote more receptivity to replication studies–another great idea. Indeed, I learned about a really great pre-study design registration program administered by Perspectives on Psychological Science, which commits to publishing results of “approved” replication designs. Social Psychology and Frontiers on Cognition are both dedicating special issues to this approach.
Finally, a number of folks have called my attention to the practice of “adversary collaboration” (AC), which I didn’t discuss at all.
AC consists of a study designed by scholars to test their competing hypotheses relating to some phenomenon. Both Phil Tetlock & Gregory Mitchell (working together, and not as adversaries) and Daniel Kahneman have advocated this idea. Indeed, Kahneman has modeled it by engaging in it himself. Moreover, at least a couple of excellent journals, including Judgement and Decision Making and Perspectives on Psychological Science, have made it clear that they are interested in promoting AC.
AC obviously has the same core objective as LR ≠1J. My sense, though, is that it hasn’t generated much activity, in part because “adversaries” are not inclined to work together. This is what one of my correspondents, who is very involved in overcoming various undesirable consequences associated with the existing review process, reports.
It also seems to be what Tetlock & Mitchell have experienced as they have tried to entice others whose work they disagree with to collaborate with them in what I’d call “likelihood ratio ≠1” studies. See, e.g. Tetlock, P.E. & Mitchell, G. Adversarial collaboration aborted but our offer still stands. Research in Organizational Behavior 29, 77-79 (2009).
LR ≠1J would systematize and magnify the effect of AC and in a way that avoids the predictable reluctance of “adversaries” — those who have a stake in competing hypotheses– from collaborating.
As I indicated LR ≠1J would (1) publish pre-study designs that (2) reviewers with opposing priors agree would generate evidence — regardless of the actual results — that warrant revising assessments of the relative likelihood of competing hypotheses. The journal would then (3) fund the study, and finally, (4) publish the results.
This procedure would generate the same benefits as “adversary collaboration” but without insisting that adversaries collaborate.
It would also create an incentive — study funding — for advance registration of designs.
And finally, by publishing regardless of result, it would avoid even the residual “file drawer” bias that persists under registry programs and “adversary collaborations” that contemplate submission of completed studies only.
Tetlock & Mitchell also discuss the signal that is conveyed when one adversary refuses to collaborate with another. Exposing that sort of defensive response was the idea I had in mind when I proposed that LR ≠1J publish reviews of papers “rejected” because referees with opposing priors disagreed on whether the design would furnish evidence, regardless of outcome, that warrants revising estimates of the likelihood of the competing hypotheses.
As I mentioned, a number of journals are also experimenting with pre-study design registration programs that commit to publication, but only for replication studies (or so I gather–still eager to be advised of additional journals doing things along these lines). Clearly this fills a big hole in existing professional practice.
But the LR ≠1J concept has a somehwat broader ambition. Its motivation is to try to counteract the myriad distortions & biases associated with NHT & p < 0.05 — a “mindless” practice that lies at the root of many of the evils that thoughtful and concerned psychologists are now trying to combat by increasing the outlets for replication studies. Social scientists should be doing studies validly designed to test the relative likelihood of competing hypotheses & then sharing the results whatever they find. We’d learn more that way. Plus there’d be fewer fluke, goofball, “holy shit!” studies that (unsurprisingly) don’t replicate
But I don’t mean to be promoting LR ≠1J over the Tetlock & Mitchell/Kahneman conception of AC, over pre-study design registration, or over greater receptivity to publishing replications/nonreplications.
I would say only that it makes sense to try a variety of things — since obviously it isn’t clear what will work. In the face of multiple plausible conjectures, one experiments rather than than debates!
Now if you point out that LR ≠1J is only a “thought experiment,” I’ll readily concede that, too, and acknowledge the politely muted point that others are actually doing things while I’m just musing & speculating. If there were the kind of interest (including potential funding & commitments on the part of other scholars to contribute labor), I’d certainly feel morally & emotionally impelled to contribute to it. And in any case, I am definitely impelled to express my gratitude toward & admiration for all the thoughtful scholars who are already trying to improve the professional customs and practices that guide the search for knowledge in the social sciences.
Hey, look — I’m not the only person who thinks something like this is a good idea! What’s more, the other guy is pretty darn smart!
Brendan Nyhan, it turns out, proposed the pre-accepted, design-only article at one point as well (along with a host of various other great things, including elimination of the Senate filibuster and construction of thousands of cold fusion nuclear power plants). Over the course of an initial post and a follow up, he makes a good case for the unique benefits that such a reform would have in relation to design registries, “methods only/result blind” review, and promotion of replication studies–all of which are really good ideas that public-spirited groups of scholars are actually undertaking.
The only thing LR ≠1J “adds” to J. Nyhan is the spectacle of fully funded, enlightenment-guaranteed “steel cage” matches: in addition to pre-commiting to publication, LR ≠1J would pay the cost of conducting any study that “adversarial reviewers” agree is designed to produce evidence that warrants adjustment of the probabilities they’d assign the hypothesis–regardless of how the study comes out.
Bringing my own exercise in selfless public service to a close, I hereby propose that someone come up with the money that this journal would require and that Nyhan be appointed the editor.