feature request: reality check of formula

[HedvigS]

I noticed that if you specify a formula in either of these two ways

Response ~ Predictor + Response

Response ~ Predictor + (1|Response + Group)

brms::brm() still runs, though naturally both cases generate various warnings that should be heeded.

I can't think of any circumstances where these formulas would make sense (let me know if there are any!), and it should be relatively easy for brms::brm to do a reality check and not proceed if the formulas look like this (or generate a specific warning about formula specification).

Yes, obviously this is something that people shouldn't do. I happened across it as I was running multiple models and copy-pasting poorly and then trying to understand the weird results.

[paul-buerkner]

I like the idea! I will have to think of specific rules that would make sense here.

[HedvigS]

I like the idea! I will have to think of specific rules that would make sense here.

Thanks! That's very understandable.

If there are any legit scenarios where this is reasonable, it could be a warning (e.g. "we noticed that your response variable is also a predictor, consider carefully if this makes sense in your case.") instead of alright stop.

[paul-buerkner]

Yes, indeed. I would only throw a warning.

[HedvigS]

Yes, indeed. I would only throw a warning.

Understandable.

I know brms strives to be quite unrestrictive. Out of curiosity, are there any instances where formulas like this might make sense? I can't think of any.

[paul-buerkner]

I am not sure either, but I will think about it :-D