Date: Wed, 15 Oct 2008 15:42:53 -0400
Reply-To: Kevin Viel <citam.sasl@GMAIL.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Kevin Viel <citam.sasl@GMAIL.COM>
Subject: Re: Proper use of proc logistic
On Wed, 15 Oct 2008 10:55:54 -0700, cgoldhaw <cgoldhaw@HOTMAIL.COM> wrote:
>Once again, huge thanks to everyone for replying. I'm new to SAS and
>statistics does not come easy to me.
>The variables that I am dealing with would logically be highly
>correlated. They are feed intake, average time spent feeding and
>average number of visits to a feeder (all of which are related, so now
>that I have read your helpful comments, I see that it makes little
>sense to have them all in the same model).
>To make practical recommendations about which variable is "the best"
>at predicting disease/has the most impact on disease outcome, i
>thought that it made the most sense to run each DV independently. But
>as Sigurd Hermansen mentioned, the significance of a model with one DV
>is suspect. When I plot the residuals for each DV there are no
>patterns and each DV shows a normal distribution. There are ten
>animals in the healthy group and ten in the diseased group, balanced
>for age and other factors known to influence disease and/or feed
>Would it be correct to run each DV independently and based on the
>significance level or r^2 make recommendations as to which DV is best
>suited for identifying animals at increased risk of disease??
Peter pointed out a likely typo by Sig. You mentioned matching, so one IV
might not be unreasonable. Some authors suggest 10 observation per
estimated beta, but you might be able to go lower. That is really a
judgement call and subject to (intense) criticisim by your co-authors and
reviewers. However, I think the sample size is very small for this
training investigation. You might want to adjust your criterion based on
the effects of a false positive or false negative....