Youtube Link: http://www.youtube.com/watch?v=X9ve4EdxFJc
Parts:
See also:
Commentary:
The McNemar test is a relatively uncommonly seen in the literature, given that it is based on a research design that is quite common. In effect, the McNemar test is the repeated measures (or within-subjects) design equivalent of the Pearson chi-square test. I suspect that some people simply apply the 2 by 2 Pearson chi-square test to data that actually conform to the McNemar test.
I know when I first studied the McNemar test formula, I struggled to understand how it was testing what it was supposed to be testing. That is, the McNemar test only uses two pieces of information in the 2 by 2 table: the upper right and lower left cells ("the discordants"). From this perspective, the formula may appear a bit misleading. One would perhaps more intuitively expect the formula to make use of the marginal row and column data (which is where the meaningful percentages can be found). However, upon further reflection, it becomes clearer that the formula is clever. That is, if, from time 1 to time 2, just as many cases switch from category A to B as category B to A, then it necessarily implies that the relevant percentages in the corresponding marginal row and columns will have stayed the same.
In the youtube example above, I apply the McNemar test (via SPSS) to "real" data relevant to study published by ... (2001). I tested the hypothesis that fewer children will require soothing at night at the age of 3, in comparison to when they were 1 years old. At age 1, 51.5% of children required soothing. By comparison, at age 3, 33.3% of children required soothing. Is that 18.2% difference statistically significant? Watch the video to find out : )