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.
When summarising a negative binomial model (fit with
HMisc::fit.mult.impute
/MASS::glm.nb
, inherits from "glm") withtab_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 inplot_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 intidymodel
.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.