choosing method for CI and p-value in tab_model
p-mq opened this issue · comments
When summarising a negative binomial model (fit with HMisc::fit.mult.impute
/ MASS::glm.nb
, inherits from "glm") with tab_model
I get significant p-values even though the confidence interval for likelihood ratio spans 1 or non-significant ones even though it does not. The same incongruence of CI and p-value seems to happen in plot_model
.
I have identified the problem as follows: When passed this kind of model, tab_model
uses Wald estimation for the p-values but profiled confidence intervals. This likely happens due to unforeseen circumstances in tidymodel
.
Besides fixing the confusion in the default setting, it would be great if the user could manually specify the method used to calculate p-values/confidence intervals.