florianhartig
commented
2 years ago Hello sssiv93,
to me, this doesn't look concerning. These deviations are probably barely significant, could also be by chance.
If there is something, it's probably in the variable Impressions, so you could play around with adding a nonlinear (e.g. sort, quadratic) term for this variable, but this will make the interpretation of this variable harder, so if your main goal is to make inference about the effect direction, I would probably leave the model as it is.
Best, F
sssiv93
Hi Florian,
Massive thank you for your work in putting together this package! It is hugely useful.
I was wondering if you could help me interpret the simulated residual plots for my Gamma GLM model with a log link. As you can see, the QQ plot looks great but I am seeing quantile deviations in the quantile plot (in the 1st attachment), as well as in the quantile plots of the residuals against the predictors (in the 2nd and 3rd attachments). Would you consider these deviations large enough to discard this model, considering that this is based on a training dataset of 144 observations? Do you have any insights into the implications of such deviations for this model?