# McNemar - is it the right test for this example? Effect size?

I have two dependent groups that could have a disease before and after a treatment. My sample size is 12214 subjects. Before the treatment, 7 of them had the disease and after the treatment 14 of them had the disease.

Percentages of total sample are very low although the number of patients with disease doubled. Does it make sense that McNemar gives me a significant p-value here? Why?

Here is my R code:

mat=matrix(c(12207, 7, 12200, 14),2)
mcnemar.test(mat)


Is there any size test I can run after McNemar? Have I chosen the right test?

• McNemar would test that the 2x2 frequency table is symmetric, that is, that the number of people having the desease before but not after equals the number of people having the desease after but not before. The test is blind to people having the desease both before and after. So what do you want to test? – ttnphns Jun 2 '14 at 10:05
• I want to answer the question: is there a significant difference before the before and the after? – DroppingOff Jun 2 '14 at 13:30
• Yes, you may use the test. – ttnphns Jun 2 '14 at 17:57
• Thanks for the confirmation. Does it make sense a significant pvalue with such a low positive results (7 and 14)? Is there a size effect test to run after McNemar? – DroppingOff Jun 2 '14 at 18:18
• As far as I know McNemar totally ignores diagonal frequencies. P-value must be the same whether you have diagonal 12207 12200 or 122070 122000 or even 12207 120. Check this. – ttnphns Jun 2 '14 at 18:49