# When to use F-Test versus Chi-Square Test in Analysis of Deviance

I'm choosing a GAM, and when I try to do

anova(model.1, model.2, test="F")


It gives me a warning message saying that the F test is inappropriate for binary outcomes. Why is this?

I get virtually the same p-value either way.

• The discussion in the help ?anova.glm looks like it is relevant to your question (though not completely covering the 'why'). – Glen_b May 5 '14 at 23:42