Please stop telling me about risk factors. 🙏😖 (ICYMI) statsepi.substack.com/p/sorry-what... - ThreadSky

statsepi.bsky.social • 37 days ago

Please stop telling me about risk factors. 🙏😖

(ICYMI)

https://statsepi.substack.com/p/sorry-what-was-the-question-again

Comments

seanpmackinnon.bsky.social•37 days ago

Years of journal reviewers have taught me that the only acceptable word to use for every statistical analysis is "association" or sometimes "correlation"

This is true of surveys, longitudinal work, and experiments in my experience.

God help you if you try to use the word "predict" for a regression

karlpettersson.bsky.social•37 days ago

How would an article such as this be titled with a saner terminology? “Etiological factors”? “Causes” also seems awkward in this context.

sebbate.uk•37 days ago

The worst version is you take a multivariable regression, find that 2+ variables have p<0.05 and call them 'independent risk factors'. 🤮

tah-sci.com•37 days ago

"AI-identified clinically relevant subsets of the exposome"

(Apols for meeting later btw I'm on a train.)

brucebradbury.bsky.social•37 days ago

You're not going to get rid of 'risk factors'. Has too much history (eg in actuarial studies). Better to simply insist it means that same thing as prediction, and as you say, is definitely not the same as causation.

brucebradbury.bsky.social•37 days ago

Nick Huntington-Klien in The effect discusses these 'weasel words':
"If we don’t want to say causality but we do want the reader to hear it, we might say: X is linked to Y, X is followed by Y, X has ramifications for Y, X predicts Y, people who X are more likely to Y, Y happens as X happens"

chrisp88.bsky.social•37 days ago

I'd argue a 'risk factor' is inherently causal in language (a factor which leads to/causes an increase in risk of outcome), whereas a 'predictor' should be the terminology for prediction modelling.

phdemetri.bsky.social•37 days ago

Booo, don’t say “predictor” because if you do then I’m going to ask how you know it predicts the outcome.

And if you respond with a P value, I will think you are pulling my leg and be unkind.

tjkelman.bsky.social•35 days ago

Sigh. This is rampant in the veterinary literature. One day I will convince colleagues that fitting a model is not prediction. Or not... *sob*

tjkelman.bsky.social•35 days ago

It occurred to me that you may (also) be addressing the fruitless practice of interpreting individual covariates for prediction within the context of a multivariable model, which if so... agreed. Evaluate overall model performance via internal/external validation and leave it at that.

sim0ngates.bsky.social•37 days ago

And especially don't say "independent risk factor."

jeremylabrecque.bsky.social•37 days ago

I've been reading a bunch of research questions recently. You know, the first sentence of the last paragraph of the intro. And it's incredible how when you really stop and parse them word for word, so many actually don't make sense.

sim0ngates.bsky.social•37 days ago

Currently in an argument about an entire 70 page research protocol that has no research question at all.

bobcalder.bsky.social•37 days ago

Needs one of those web interfaces that won't let you click the submit button without filling in the pertinent blank. ;-)

noamross.net•37 days ago

Ecologists like to say nothing by calling correlates "drivers."

Comments

Posting Rules

Reply