Date: Fri, 13 Jan 2006 18:25:08 -0500
Reply-To: Kevin Roland Viel <kviel@EMORY.EDU>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Kevin Roland Viel <kviel@EMORY.EDU>
Subject: Re: Epidemiological Question
Content-Type: TEXT/PLAIN; charset=US-ASCII
On Fri, 13 Jan 2006, Joe Whitehurst wrote:
> Your problem likely involves many more variables than my problem. All I
> have is counts of a particular kind of event for each month for each site
> while you likely have many predictors (demographic, geographic, economic ,
> etc.). No doubt such factors may be involved in producing the events
> being counted, but no data are available. I think I can use X11
> methodology to adjust for seasonality and some measure of deviations to
> derive a decision rule, but I am still hopeful that someone will have
> already developed some method of using all the available information (60
> months of counts/month for hundreds of sites) to derive a rule for
> reporting when a new month's counts are in some sense atypical for a
> particular site.
It sounds as if you might be interested in determining if two rates may
be statistically different. If so, you'll need to decide what
distribution might provide a good approximate or satisfactory theoretical
grounds (for instance, the poisson).
If you have access to a library, then this article might be an easy and
"Basic models for disease occurrence in epidemiology." Flanders WD,
Kleinbaum DG. Int J Epidemiol. 1995 Feb;24(1):1-7.
Once you choose a distribution, the it seems to me that you would use
the 60 month average to calculate the probability of observed count and
every count more extreme. Depending on the observed count, you will
probably want to take use the compliment, i.e. 1 - p(0...observed - 1).
Department of Epidemiology
Rollins School of Public Health
Atlanta, GA 30322