Likelihood Ratio Test Using The Codeml

Question

[vicky]

Hi I have just started working with the positive selection. I have a query regarding the likelihood ratio test in Codeml.I have two likelihood ratio generated from two models .But how I implement the Chi-Square test.I have the two likelihood values as -1043.83 and other is -1041.69.From here how i have to proceed , I am not understanding.

Answer 1

[Fabian Bull]

First of all: Likelihood-Ratio-Test can only be used for nested models. So convince yourself your models are indeed nested.

Assume likelihood1 is the likelihood of the nested model1 and likelihood0 is the likelihood of the "supermodel" model0

Your test-statistic is:

T = -2 * log ( likelihood1 / likelihood 0)

This is Chi^2 distributed with dgf (parameter model1 - parameter of model0).

To test if model1 is significant better then model0, cdf(T) should be greater then 0.95 .

EDIT: Suppose your models do not differ hugley in there number of parameters if would stick with the simpler(int you case) because the likelihoods are pretty similar.