(The title is as clickbait-y and tongue-in-cheek as ever. The cover illustration is by Midjourney, prompted by “I love my baby. My baby loves statistics.” . . which is, of course, a wink to that good old U/F/O track sampling Ken Nordine.)
This post discusses the unfortunate and detrimental division of labor between statisticians and quantitative scientists. You see this explicitly as a reviewer (“Does this paper need an additional review by a statistician?”). But even more often, it is evident from the low level of statistical integrity in science, where people just follow a pipeline defined (to some extent by statistics, but ever so much more) by the norms of the field.
My point is that one cannot separate a part of the methodology so central as statistics from the science itself. We quantitative scientists have to have full command of the methods we use and choose not to use. We have to stand against the groupthink-dumbing-down reliance on predefined methods that brought us publication biases and replication crises instead of knowledge.
Statistics is fantastic. It alone enables us to report scientific findings with the infinite precision of numbers rather than feeble English words. But the catch is that statistics is not the language of science. It’s not a language. We cannot translate words into statistics, no matter how much we want p < 0.05 to mean the English word “significant.”
Thank you, professor, for a wonderful talk. I have three questions . . no comments, really.
First, because science is a team effort, we shouldn’t necessarily try to optimize the quality of every single paper. It is more important that other researchers quickly and unambiguously understand what a paper has achieved. For that, standardized methodologies are good.
Second, a wealth of knowledge is hidden in the brains of people who don’t like stats. That’s an asset we should not lose by raising the bar too high.
Third, it’s getting better.
Well, to your first comment, much of the behavioral and social sciences have become islands of statistical monoculture. There is a tradeoff between optimizing the collective intelligence and the individual articles, yes, but we are currently at the very extreme end of the spectrum, which can’t be the optimum.
The second point assumes that one can be an expert on something without understanding where one’s knowledge comes from. I doubt it. We are talking about science, not AI-esque black-box predictions.
Third, I agree! Not only has the causal-inference boom made scientists more stats savvy, I just saw an X discussion among statisticians where they nodded in agreement to the proposal that they need to be more involved in experiment design. That obviously solves the problem of what to do with all unemployed statisticians when we have taken statistics back. We turn them into scientists! That’s what they want to be anyway. Just let ’em read a couple of Bordieu and they’re good to go. 😉
Petter Holme 