Date: Fri, 4 Mar 2005 11:51:15 -0500
Reply-To: Dianne Rhodes <RHODESD1@WESTAT.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Dianne Rhodes <RHODESD1@WESTAT.COM>
Subject: Re: knowledge discovery vs. hypothesis
Content-Type: text/plain; charset="us-ascii"
> Li,Tom wrote:
> The analysis methods should be pre-specified, not after the
> data are collected. Most of the results from afterward
> introduced methods are not worth anything.
Let me say "that depends". In the real world, as opposed to academia,
I've made a career of analyzing retrospective health care data. Data is
collected to satisfy the needs of insurance providers such as Worker's
Comp as an example. Their reason for collecting it is to analyze their
spending and predict costs for next year. But there is a wealth of
information there. This is the data we used to identify an alarming
increase in the rates of carpal tunnel syndrome. We analyzed that data
and found that outcomes were not better if the patient had surgery.
Many important findings come from analysis of retrospective data.