Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: McNemer's Test with OR = 0

From   Steve Samuels <>
Subject   Re: st: McNemer's Test with OR = 0
Date   Wed, 15 Aug 2012 10:42:14 -0400

McNemar's test can certainly be used if there is a zero cell.

Assume the 2x2 table has cells a,b,c,d, as usual, with baseline attitude (+/-)
for rows and follow-up attitude (+/-) for columns. Positive attitude before is
p_1 = p_a +p_b  Positive attitude after is p_2 = p_a + p_c.  The difference
(after minus before) is p_c - p_b, estimated by (c - b)/n.

The estimated OR (after/before) is b/c. This is the stratified Mantel-Haenszel
estimate of the OR when each stratum is a matched pair.   OR = 0 means
that b = 0. Notice that the upper endpoint of the CI is non-zero.   A small
p-value would be expected if OR = 0,  provided also that "c" is large.

However the odds ratio is _not_ an appropriate measure for a change
study, although it is for a case-control study. Report the risk difference (or
ratio) and its CI. All this will be covered in any good epidemiology text. 

See also Fleiss et al. (2003) and the Formulas and Methods section of the Manual
entry for -epitab-

Ref:  Fleiss, Joseph L., Bruce A. Levin, and Myunghee Cho. Paik. 2003.
Statistical methods for rates and proportions. Hoboken, N.J.: J. Wiley.


> On Aug 14, 2012, at 10:18 AM, Jeanette Somlak Pedersen wrote:
> Hello,
> I have two questions about the McNemar's test. I ran the test to determine change in attitude.
> 1) Can a McNemar's test be used if one of the four cells = 0?
> 2) The Stata output shows that the p value is significant (0.000) and the OR = 0. I have never seen an OR of 0 before, is this an issue?
> Thank you,

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index