Sometimes pvalues = 0

[jarbet]

I am finding that sometimes logistf returns chisquare test pvalues of 0 in the summary(fit) output, and the chisquare test stat will be Inf. This is a problem since p-values can't equal 0, especially if doing FDR multiple-testing adjustments (e.g. using p.adjust).

My apologies but I can't share the data that produces this problem. A simple solution I found was to calculate Z-test pvalues in this case:

zstat <- fit$coefficients / sqrt(diag(fit$var));
pvalue.zstat <- 2 * pnorm(abs(zstat), lower.tail = FALSE);

Similarly, anova.logistf sometimes returns pvalues of 0. A simple solution was to manually calculate the chisquare pvalue:

pchisq(anova.result$chisq, anova.result$df, lower.tail = FALSE);