Carl Schwarz comments

Results 7 comments of


                                            Carl Schwarz

ggpredict for GLM reports the SE from the predicted logit and not from the predicted response.

Thanks.... the SE needs to be back transformed according to the type of link function, i.e. the back transformation for a logit() link is different than the se for a...

ggpredict for GLM reports the SE from the predicted logit and not from the predicted response.

Unfortunately, you have NOT computed the back-transformed SE properly. The anti-logit transformation for the estimate and the confidence limits is computed as p=1/(1+exp(-theta)) where theta is the estimate or upper/lower...

ggpredict for GLM reports the SE from the predicted logit and not from the predicted response.

[ggeffects.bug.pdf](https://github.com/strengejacke/ggeffects/files/5058919/ggeffects.bug.pdf) [ggeffects.bug.txt](https://github.com/strengejacke/ggeffects/files/5058920/ggeffects.bug.txt)

ggpredict for GLM reports the SE from the predicted logit and not from the predicted response.

I think it is dangerous to leave SE for the untransformed data in the table. The note at the bottom of the table will invariably get lost/overlooked. Users will ask...

ggpredict for GLM reports the SE from the predicted logit and not from the predicted response.

The reason for the difference is that emmeans gives MARGINAL means while ggpredict is giving predictions for a SINGLE level of a categorical variable.. What is a marginal mean --...

Difficulties with isfinite() on Windows machines

I'm afraid that you've lost me... I'm a Mac user and have seldom used C++ (except to write TMB functions). I don't have a Makevars file (that I'm aware of),...

Number >1 converted to boolean TRUE without any warning

Yes, I agree. I was recently bitten by this problem - took about a day to figure out! An alternate solution would be return a character vector rather than a...