Computing likelihood ratio pseudo-R2

Question

1 个投票

I wish to compute the likelihood ratio pseudo-R2 from a logistic regression using glmfit.

According to the Wikipedia definition it is defined as

R2 = (DevianceNull-DevianceModel)/(DevianceNull).

Does DevianceModel correspond to dev returned from glmfit?

[~, dev] = glmfit(X, Y, 'binomial');

In that case, is the pseudo-R2 computed as?

[~, devModel] = glmfit(X, Y, 'binomial'); 
[~, devNull]  = glmfit(zeros(length(Y), 1), Y, 'binomial');
R2 = (DevNull-DevModel)/(DevNull);

Many thanks

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Andreas 2014-1-20

Nobody?

Answer 1

Arseny 2017-9-4

编辑：Arseny 2017-9-4

0 个投票

It would be your code, but with

[~, devNull] = glmfit(X, Y(randperm(length(Y))), 'binomial');

Or better yet, this row repeated several times, and devNull estimation averaged over these several attempts.