I performed multivariate logistic regression with this dataset: https://justpaste.it/61vgo

#build model
dataB <- read.table("clipboard", header = T, dec = ".", 
                sep = "\t") #after copying data
fit1 <- glm(outcome ~ A + B, family=binomial, data=dataB)
#get output
tbl_regression(fit1, exponentiate = T)

As you can see below, 95% CI are not the same between both output (logistic.display from epiDisplay and tbl_regression from gtsummary, see below), and moreover, tbl_regression display 95%CI that does not match with p-value from Wald test (pvalue should be < 0.05 if 95%CI does not contain 1) ...

Do you know why? Which one is the good one?






You are comparing Wald p-values with confidence intervals based on profile likelihood, this is the reason why the results don't match.

If you were to estimate Wald confidence intervals, they would include 0 and match your Wald p-value results.

                   2.5 %      97.5 %
(Intercept) -4.505810889 7.827309627
A           -0.008111892 0.794005695
B           -0.420888942 0.001519585

On the other hand, you already got the LR-test p-values (0.01, 0.034) in your call to logistic.display(), which also match the results of the 95% CIs based on profile likelihood.

  • $\begingroup$ Ok thanks ! I understood that when the sample is big enough, Wald pvalues and Likelihood pvalues merge. Am I right? So which one to choose (pvalues and IC) when sample is sample? Moreover, why tbl_regression select pvalue from Wald and IC from Likelihood? $\endgroup$
    – B_slash_
    Apr 16 at 16:15

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.