This excellent paper led by @connorjmccabe shows that *almost all* interaction tests for non-linear outcomes (binary, counts, etc.) in psychology are wrong.

It shows why, and develops R functions to help you correctly estimate interactions for these outcomes.

https://twitter.com/connorjmccabe/status/1247994922038218753

Why is this the case?

Recall that one assumption of linear regression is that predictor effects are independent.

IOW, the slope of the regression for X on Y does not depend on the value of any other covariate.

Interactions (tested with a product term) test that assumption

OK, you just realized you panic bought a 25 lb bag of dried beans a couple weeks ago, and now you have no idea how to cook them.

You might have just looked at a recipe for beans and seen that it's a multi-day process of soaking, rinsing, and cooking.

WRONG. First principle.

Wash your beans. Yes, they might be "triple washed", but they're still an agricultural commodity.

Check for stones. I've never found one, but still, check. Your dentist will thank you.

ams.usda.gov/mnreports/lswb…

I'm giving a lecture on reproducibility issues in the field to a group of very savvy clinicians at a clinic that exclusively provides EBPs (and does research too).

They've asked me some great questions before the lecture, and I'd like your input on any of them:
1/ Is there any standard by which we can say “yes, the data clearly show X” in our field?

2/