Johnson-Neyman plots for glmer models (logistic, mixed effects) in R I have an R glmer (logisitc mixed-effects) model that looks something like:
mod <- glmer(choice ~ (x1+ x2)*log(x3) +x4 +(x1 + x2 | id), data = dat, family = binomial)

I can get the slopes of the interaction between x1 and x3 by doing:
slopes <- modelbased::estimate_slopes(mod, trend = "x1", at = "x3", length=50)

And then plot the significance of x1 across values of x3 (i.e. look at the Johnson-Neyman plot):
plot(slopes)

This works fine, but is this kind of plot valid for logistic mixed-effects models? From my understanding, it is based on assumptions of normality, which do not hold true for mixed-effects models.
 A: 
From my understanding, it is based on assumptions of normality, which do not hold true for mixed-effects models.

The normality assumptions for test of significance in generalized linear mixed models (GLMMs) have to do with the assumed distributions of the modeled coefficient estimates, not the distributions of residuals. Models fit by maximum likelihood have an asymptotic multivariate normal distribution of the fixed-effect coefficients. Agresti says in Section 12.6.6 of Categorical Data Analysis, 2nd edition:

After fitting the model, inference about fixed effects proceeds in the usual way. For instance, likelihood-ratio tests can compare nested models. Asymptotics for GLMMs apply as the number of clusters increases, rather than as the numbers of observations within the clusters increase.

In principle there is no theoretical problem with what you're doing, provided that your data set is large enough.
Note, however, that the ranges of "significance" in the Johnson-Neyman plot will mostly be a function of the size of your data set. Think carefully about how much displaying that plot actually adds to your description of the model. Your audience might better appreciate a display of results for informative and realistic combinations of predictor values, rather than a display of the range of values over which you reach some sample-size-dependent measure of "significance."
