Search code examples
rglmlsmeansemmeans

Extract probabilities from pairwise contrast in emmeans

I'm trying to perform a pairwise comparison after a glm with a binomial dv, and emmeans reports odds ratios, whereas I require a difference in probabilities.

library(magrittr)
library(emmeans)

glm(
  am ~  wt * factor(vs), 
  family = binomial(),
  data = mtcars
) %>% 
  emmeans(
~ vs | wt, at = list(wt = seq(2.6, 3.6, length.out = 10))
  ) %>% 
  pairs(type = "response")

returns

wt = 2.6:
 contrast odds.ratio         SE  df z.ratio p.value
 0 / 1     668.84887 3388.29452 Inf   1.284  0.1991

wt = 2.71111111111111:
 contrast odds.ratio         SE  df z.ratio p.value
 0 / 1     453.70452 2028.70137 Inf   1.368  0.1713
Solution

  • Add %>% regrid() before pairs(). The regrid() call back-transforms all the results to a new set of EMMs as if there had never been a transformation.

    See the vignette on transformations.