Compare the estimated model with the model with intercept-only

[valentina89]

The question is:
How do the test to compare the estimated model with the model with intercept-only ? How can I comment on the result ?

I think we need to calculate the test F...
I read several notes, but I can not understand how I do . The idea is to take Null deviance and Residual Deviance and split in some way ...


glm output:

 

[mmercker]

Hi, there are different possible ways. You can compare the AIC-values of the two models and you can favour the model with the lowest AIC. Or you can perform a likelihood-ratio test, which is approximately Chi^2-distributed. However, there are several other possible techniques to compare / select models, e.g. cross-validation. Which software do you use?

 

[valentina89]

Hello , I use R but exercise is written , without the support of software.

 

[hlsmith]

Mmereck is right, there is a test, I can't remember the exact formula but it is something like the differences in RSS between the model with the difference in DF.

 

[rogojel]

hi,
isn't this what the p-value calculates anyway?At least if we have a type I sum of squares (if memory serves).
regards

 

[Dason]

You only have a single predictor. So the test of the significance of the predictor (AGE) is the same as the test of the model compared against the intercept-only model. Alternatively you could use the likelihood ratio test by taking the difference of the deviances for the two models and comparing against a Chi-square distribution (which is what mmercker was suggesting with the likelihood ratio test).