I am trying to get the predicted probabilities from a rare-events logit.
When I try ggpredict, I get a single data-point, instead of the representative values I would expect (see: https://github.com/strengejacke/ggeffects/issues/220)
I have tried updating all the packages from their github sources.
Dataset can be found here: https://ucdp.uu.se/downloads/nsos/ucdp-onesided-231-csv.zip
rm(list = ls())
library(foreign)
library(ggeffects)
library(logistf)
df = read.csv("~/ucdp-onesided-211.csv")
df$best_fatality_estimate = log(df$best_fatality_estimate + 1)
#Making some of the values 0, to reproduce the error
index = sample(1:nrow(df), 100, replace=FALSE)
df$best_fatality_estimate[index] = 0
#Making up some independent and dependent variables
df$dv = rbinom(nrow(df), 1, 0.1)
df$iv = runif(nrow(df), min=0, max=100)
df$iv2 = rnorm(nrow(df), 100, sd=20)
reg <- logistf(dv~best_fatality_estimate+year + iv + iv2, data = df)
ggeffects::ggpredict(reg, terms="best_fatality_estimate")
ggeffects::ggpredict(reg, terms="best_fatality_estimate [all]")
Interestingly, ggeffects works:
ggeffects::ggeffect(reg, terms="best_fatality_estimate")
