# Re: st: negative and positive maximum likelihood ratios

 From [email protected] To [email protected] Subject Re: st: negative and positive maximum likelihood ratios Date Wed, 31 Jan 2007 09:06:51 +0100

```Sorry for the delay in my replay and for potential double posting.
I am not sure if I understand correctly your questions.
-lrtest- provides a value (that reported besides "LR chi2") which is minus the double of the difference of the Log likelihood (that value usually reported before the table of the estimates; if you don't find it, you can have it by typing -estat ic- and looking at the column "ll(model)") between the two models you want to test. Simple example: if L0 is the log likelihood for the constant-only model and L1 the log likelihood for the full model,
LR chi-squared = - 2 x (L0 - L1).
Given that it is a chi-squared, this value should be positive (I don't understand what "1" matters: as I have shown above, despite being called "ratio", it is not). But I have readen something among Stata's help files (sorry, do not remember which one, probably an user-written command) suggesting that a negative chi-squared points to strongly accepting the null hypothesis.
Nicola

At 02.33 29/01/2007 -0500, Don Spady wrote:
>Can someone please tell me what the difference is between a negative
>and a positive maximum likelihood ratio.  My intuition tells me that a
>'negative' MLR is one that is less than 1 and a 'positive' MLR is
>greater than 1 but I cannot find anything that describes these terms.
>   As a second request: how does one calculate these ratios when doing
>logistic regression.  I know that you can test the ratios using lrtest
>but how can one actually calculate the ratio.