Likelihood Ratio Test for Logistic Regression

[Sara Moradi]

Hello everybody.

I have a binary outcome variable and several independent variables, and I have to build a logistic regression model. First, I put all independent variables in the model. Then, I want to pull out variables one-by-one to evaluate if I can have a smaller model. For this purpose, I have to run a likelihood ratio test.

Could anyone please help me with interpreting the following LRtest?
A is a full model with all variables. B is my reduced model.
I can't figure out which one of them is better according to this LRtest.

Thank you,

[Carlo Lazzaro]

Sara:
the rule says "the lower AIC (BIC), the better".
Therefore, model B seems to be the way to go.

[Felix Bittmann]

Carlo is right. If you wanna focus on the LR test: it is not significant (p = 0.288) hence there is probably no statistically significant difference between the models overall.

[Øyvind Snilsberg]

going from model A to model B does not significantly reduce model fit.

Then, I want to pull out variables one-by-one to evaluate if I can have a smaller model.

you don't need the LR test for that, just see whether the variable is significant in model A.

[Felix Bittmann]

Another idea: since the models are somewhat equal statistically, I would prefer the simpler model (with fewer variables), unless you have strong theoretically reasons not to.