Confidence intervals for predictions from logistic regression GLMER

Question

I am trying to make a prediction model and I would like to have the confidence intervals around these predictions

This is a mixed generalized logistic regression model

I tried to take the code here : Confidence intervals for predictions from logistic regression

I liked this format because I need to do a for loop, Basically I'm working on a model of mortality due to fire with several variables (flame height, intensity, tree diameter) And I would like to have the predicted mortality rate for each degree of fire intensity.
I managed to make a loop that makes predictions for each of these intensities, but I would like to have a confidence interval to better interpret my results.

But it doesn't seem to work when there are random effects.

I hope someone can help me

Here is the reproducible error:

 foo <- mtcars[,c("mpg", "vs", "cyl")]; names(foo) <- c("x","y", "z") 
    ## Working example data
mod <- glmer(y ~ x + (1|z), data = foo, family = binomial)
preddata <- with(foo, data.frame(x = seq(min(x), max(x), length = 100)))

fixed_z = 5
preddata2 <- data.frame ( x = preddata$x, 
                          z = 6) 
    ## want to fix a variable to see how the other react

preds <- predict(mod, newdata = preddata2, type = "link", se.fit = TRUE)

Then, there is no preds$fit

I saw that many questions on stackflow had answers but all were about glm. Is it possible to have a confidence interval on a glmer?

I tried this code here : Confidence intervals for predictions from logistic regression

I also tried the first one on this page, but in my datan the random factor is a character so

medEff = REquantile(mod, quantile = 0.5, 
                    groupFctr = "ID", 
                    term = "(Intercept)")

doesn't work

score 0 · Answer 1 · answered May 19 '23 at 19:05

The easiest way to do this might be with emmeans (the ggeffects package also offers several options). There is more low-level/do-it-yourself code here.

library(emmeans)
as.data.frame(emmeans(mod, ~x, at = list(x = preddata$x, z=6)))

An important note is that these confidence intervals only take uncertainty in the fixed-effect parameters into account; they assume that the random effects are known without error.

Confidence intervals for predictions from logistic regression GLMER

1 Answers1