Overview
McNemar's test checks whether a yes-or-no answer really changed between two measurements on the same people, such as before and after a campaign.
Enter how many people changed in each direction. People who gave the same answer both times don't affect the test.
Worked examples
| Inputs | Result |
|---|---|
| Changed from yes to no (b)6Changed from no to yes (c)15Significance level5% | p-value: 0.0809Chi-square: 3.048 with continuity correction, 1 degree of freedomNot significant at the 5% level.With fewer than 25 changes, an exact binomial test is more reliable. |
Show 1 more example
| Changed from yes to no (b)40Changed from no to yes (c)18Significance level5% | p-value: 0.00583Chi-square: 7.603 with continuity correction, 1 degree of freedomSignificant at the 5% level: the change is unlikely to be chance. |
How it works
Only the people who changed their answer carry information about a shift. If nothing really changed, those switches should split about evenly in each direction. The statistic measures how far they are from an even split, and the p-value comes from a chi-square distribution with 1 degree of freedom.
The formula
| b | People who changed from yes to no |
| c | People who changed from no to yes |
Practical tips
- The test is for paired data, where the same people are measured twice.
- With fewer than 25 changes in total, use an exact binomial test instead.
- Report the size of the change as well as the p-value.
Frequently asked questions
Why are the people who didn't change ignored?
They say nothing about whether answers shifted in one direction more than the other.
When should I use McNemar's test instead of chi-square?
When the two measurements come from the same people. The standard chi-square test needs independent groups.
What does the continuity correction do?
It subtracts 1 to adjust for using a smooth distribution with whole-number counts.