This logit transformation (from a probability scale to unbounded) does some trimming around extreme values.
logit_ghitza(x, digits = 5)
input
The number of digits to round by. Larger values indicate more closer to a theoretical inverse logit.
posthoc_error calib_oneway
logit <- function(x) log(x/(1 - x))
logit(0.000001)
#> [1] -13.81551
logit_ghitza(0.000001, digits = 5) # the default
#> [1] -11.51292
logit_ghitza(0.000001, digits = 1)
#> [1] -2.197225
logit_ghitza(0.000001, digits = 10)
#> [1] -13.81551