Compare influence of covariates in model

[torkar]

@tester

Okay, so the direction of the effects for a model based on standardized data would be the same as when fitting with non-standardized data, just the magnitude changes? I might have misunderstood this before.

Yes, it’s a standardized measure, but you can always get it back to its unstandardized format.

@tester

Yes, using conditional_effects gives a nice understanding of the effect of a predictor on the outcome. But my understanding is that this is under the condition that the other predictors are kept at their mean values.

Indeed, their mean values or the first level for factors.

@tester

Are you aware of some kind of measure, like a single value, that can be calculated that expresses what one can “see” (visually) in the plots from conditional_effects?

Well, you can ask any questions you want and the posterior will provide a reply (sometimes the reply is gibberish due to asking the wrong question ;) If you do posterior_predict(), setting the covariates to their means (and factors to their first level) you will basically get what conditional_effects() will give you.

I think this post by @jonah summarize things in a very nice way:

Reproduce posterior_predict from posterior_linpred for a neg binom spline model

Hi Angelo, here’s a brief explanation of the difference. Suppose you have a negative binomial regression with a log link: \eta_i = a + bx_i \\ \lambda_i = \exp(\eta_i) \\ y_i \sim {\rm NegBinom}(\lambda_i, \phi) posterior_linpred() gives you posterior draws of \eta. posterior_linpred(transform=TRUE), or equivalently posterior_epred(), gives you posterior draws of \lambda, that is, it does the inverse link transformation for you. posterior_predict() gives you draws from {\rm NegBinom}(\l…