Date: Mon, 10 Jan 2005 08:36:23 -0500
Reply-To: "Chelminski, Iwona" <IChelminski@lifespan.org>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: "Chelminski, Iwona" <IChelminski@lifespan.org>
Subject: Re: multi-site studies
Content-Type: text/plain
Hi Anthony,
Do you know how to increase the number of layers in the SPSS? I have a
version 11.5 and it gives me only 8 layers. Since I have 10 doctors I do
neen 10 layers, right?
Iwona
-----Original Message-----
From: Anthony Babinec [mailto:tbabinec@ameritech.net]
Sent: Friday, January 07, 2005 9:54 AM
To: 'Chelminski, Iwona'
Subject: RE: multi-site studies
It sounds to me like the CMH framework holds in your case.
Consider a table
dichotomous outcome
by
dichotomous factor
by
stratum variable
In your case, the stratum variable could be Doctor. You
want to conduct the test for a common odds ratio in
the outcome by factor table. That is the cochran-Mantel-
Haenszel situation. This was added to SPSS Crosstabs.
From SPSS Help:
Cochran's and Mantel-Haenszel statistics. Cochran's and Mantel-Haenszel
statistics can be used to test for independence between a dichotomous factor
variable and a dichotomous response variable, conditional upon covariate
patterns defined by one or more layer (control) variables. Note that while
other statistics are computed layer by layer, the Cochran's and
Mantel-Haenszel statistics are computed once for all layers.
With small samples, an exact form of the test is preferable. I don't
know without checking whether the SPSS Exact Tests option
includes an exact form of this test. The Gold Standard software
in this area is Cytel's StatXact software. See Agresti's 2nd edition
"Categorical Data Analysis" for a discussion.
Anthony Babinec
-----Original Message-----
From: Chelminski, Iwona [mailto:IChelminski@Lifespan.org]
Sent: Thursday, January 06, 2005 11:47 AM
To: 'Anthony Babinec'
Subject: RE: multi-site studies
Hi Anthony,
Thanks for responding. The reason I'm asking is because I'm struggling with
a similar issue. I have almost 600 cases that came from 10 doctors who
provided information whether in their decision to chose a medication (one of
the four) they considered a certain factor. For example, "Once a day dosing"
(YES/NO), "Samples available" (YES/NO), "Patient's age" (YES/NO). So, for
all these cases I have a bunch of 1's or 0's. I combined the data (ignored
the fact that it came from 10 different doctors) and looked at frequencies,
chi-squares etc. (For example, I found that "avoidance of specific side
effects" most frequently influenced the prescription of medication 1 and so
on). The reviewers want to know whether there are differences in those
trends (my findings) between those ten doctors. How should I proceed?
| 
