The McNemar Test is mentioned in the eighth edition, volume 2. And in most
standard statistics texts.
-----Original Message-----
From: Sten Öhman [mailto:[log in to unmask]]
Sent: Monday, November 26, 2001 1:39 PM
To: [log in to unmask]
Subject: Re: Populating contingency tables
At 2001-11-26 14:49 +0000, Trevor Tickner wrote:
>The McNemar test has the advantage of requiring only the data from the two
>analyses in a single population (i.e. those with the target disorder). It
>compares the 'failure' rates of the two analyses when the other analysis is
>positive. Its disadvantage is that one has to select cut-off points where
>the analysis result is a quantitative but that is what we often do anyway!
This is essentially the same as the "nameless" method I mentioned. Instead
of "failure rate" I used "discordant results". Both methods requires a
cut-off value (e.g. a reference limit) to be determined before the test,
and this can only be done using the reference group.
>I don't know if Documenta Geigy is still produced but it used to have the
>McNemar test in the statistical section.
It is not mentioned in the sixth edition, maybe in later editions.
>http://ourworld.compuserve.com/homepages/jsuebersax/mcnemar.htm
As an example of a McNemar test is this 2x2 table:
- +
- 40 10
+ 20 50
As can be seen there are 90 concordant results (40+50) and 30 discordant
(10+20). A rapid look in the sign test table on N=30 indicates that 0-8
observations in the "minor" group is significant on the P<0.05 level and
0-6 observations is significant on the P<0.01 level.
Using the McNemar test John Uebersax found the exact P-value in this case
of 0.099 which is in accordance with the sign test. Yes, if you need an
exact value of P the MacNemar method yields such a value, but isn't this a
bit of overkill?
The sign test can simply be performed without computer only with the aid of
a sign test table.
Mr Sten Öhman PhD
|