GLM, likelihood ratio test

[Ark raw]

In sec 3.4 of chapter 10, it talks about
1.) for normally distributed data, the scaled deviance has a chi square distribution so what is the degrees of freedom here? is it n-p, where n is the no. of observations and p is the no. of parameters?
2.) for data that isn't normal, the deviance has only approx ( or rather asymptotically) a chi square distribution, I wanna know how the deviance has a chi square distribution? and what are its degrees of freedom?
3.) in the same section the last paragraph on pg 36, talks about comparing 2 models, in this paragraph, it states that we can compare S_1 and S_2 with 2(p-q), where as in the summary of this chapter it talks about comparing SD_1 and SD_2 with 2(p-q) S_i and SD_i are deviance and scaled deviance of the ith model, so what are we are supposed to you use for comparing 2 models S or SD?

 

[John Lee]

1) Yes that sounds correct. But we never use it in CT6.
2) We're using the central limit theorem - degrees of freedom should be the same.
3) In my summary it says scaled deviance only.