I'm wondering if anyone has any experience using ggeffect() with lots of fixed effects. Although my independent variable is strongly related to my dependent variable, I get these massive confidence intervals when I add in state fixed effects.
A toy example:
set.seed(200)
indvar <- runif(500, min = 0, max = .5)
state <- as.factor(rep(c(1:50), 10))
statev <- as.integer(state) * runif(500, 0, 0.02)
depvar <- round(indvar + statev)
data <- data.frame(indvar, state, depvar)
m1 <- glm(depvar ~ indvar, data = data, family = "binomial")
margin <- ggeffect(m1, "indvar")
plot(margin)
Plot with expected confidence intervals
This gives me nice clean confidence intervals around the independent variable. However, as soon as I add in the state fixed effects, the confidence interval stretches essentially from 0 to 1, even though there remains a super strong relationship.
m2 <- glm(depvar ~ indvar + state, data = data, family = "binomial")
margin <- ggeffect(m2, "indvar")
plot(margin)
Plot with super wide confidence intervals
Thoughts much appreciated!