Date: Mon, 22 Jun 2009 10:19:29 -0700
Reply-To: Bminer <b_miner@LIVE.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Bminer <b_miner@LIVE.COM>
Subject: Re: Deviance versus log likelihood
Content-Type: text/plain; charset=ISO-8859-1
On Jun 21, 5:29 pm, art...@NETSCAPE.NET (Arthur Tabachneck) wrote:
> The variable names are different in the two assignments and I'm not sure
> what the -2(etc, etc.) mean.
> If you're just multiplying the same number by two, I would think that both
> the means and variance would simply be doubled as compared to the first
> On Sun, 21 Jun 2009 12:17:54 -0700, Bminer <b_mi...@LIVE.COM> wrote:
> >I am used to testing constraints on parameters (two nested models) in
> >logistic regression as either:
> >Deviance_reduced - Deviance_full
> >-2(LL_reduced - LL_full). Both give the same value to compare to
> >critical values of chi-square.
> >Using genmod and a normal error, identity link model the two values
> >above are not the same. Why?
> >Which should be used in a likelihood ratio test?
Let me re-express my question. I was taught that for any model
estimated with ML the difference in deviance was the same as -2* the
difference in log likihoods. This was the manner to test the inclusion
of a predictor (compare two nested models) as a chi-square with df
equal to the difference in df between the larger and smaller model.
I notice that when using proc genmod, this is not the case. It appears
to be with link=logit and dist=bin but not others.
I have since read a description that one should use (Deviance_smaller
- Deviance larger) / scale from larger. This does not equal -2*
the difference in log likihoods either.
Can anyone tell me how to take the data from genmod to compute the LR