sdmTMB and ggeffects

[seananderson]

Question by email:

I usually derive confidence intervals for predictors from lmer or glmer models using the ggeffects package with something like this:

CI <- data.frame(ggeffect(model, terms = c("sCov1 [pred]", "sCov2 [all]")))

I also tried this with a model fitted in sdmTMB, as ggeffects also supports glmmTMB models. However, it seems not to work. Would you know how I can use ggeffects on a model fitted in sdmTMB? Or how I can derive confidence intervals for plotting covariates?

[seananderson]

ggeffects::ggeffect() works with sdmTMB but currently only for non-smoother terms. There's a vignette here. I have ggeffects::ggpredict() working locally, but it will require me submitting a pull request to the ggeffects package first before it will work for others.

In the meantime, you can:

• Use ggeffects::ggeffect() for non-smoother terms. See the vignette.
• Use the visreg package for smoother or non-smoother terms. See the vignette.
• Predict across a sequence of your predictor variable while holding the other variables at some constant value (which is what visreg or ggpredict are doing for you). As an example, look for "Visualizing a marginal effect" in the examples within predict.sdmTMB().