Lou:
There is a danger of mixing the two. Alternatively, one could:
(1) take the subjects who were the SAME and do McNemar;
(2) take the subjects who were in ONLY one of the samples, and do the
difference between two independent proportions, z (note: z = chi, or
z-squared = chi-squared, but is sometimes used when one wishes to impose
a 1-tailed test on the 2 X 2);
(3) just leave everyone and do the independence test.
If you get the same answers, moot; if (1) and (2) differ (and [3] is,
therefore, somewhere in the middle), then you can decide to what degree
the difference makes, or, maybe make a more interesting finding.
If you have low sample size to begin with, then (2) is underpowered
relative to (3), but if you have at least n = 40 for (1) and the
marginals are not too discrepant, you should have enough power for that.
Joe Burleson
-----Original Message-----
From: SPSSX(r) Discussion [mailto:
[hidden email]] On Behalf Of
Lou
Sent: Tuesday, April 03, 2007 7:14 AM
To:
[hidden email]
Subject: Comparing proportions with 'some' dependencies
Dear all,
Just something I've been pondering and would like your thoughts on. If
I
need to compare two proportions and the associated groups contain 'some'
of the same people but also some different people, can I simply assume
independence and use a Chi-squared test, or do I need to somehow take
the
dependencies into account with, say, McNemar's test?
For instance, let's say we have two distinct rounds of a medical exam
and
32.5% complete the exam in the first round whereas 28.6% complete the
exam
in the second round. Some people will have undertaken the exam in both
rounds whereas some people will be new to each round. I want to state
whether there has been a significant decrease in uptake.
Thanks for your help,
Lou
Locked
