I would like your opinion about which analyses to use for the research question.

I've got a repeated measures with binary data. N = 20

H0 = there is no difference between pre and post measurement

HA = there is a difference between pre and post measurement

So, I first wanted to go for the McNemar-test or maybe a Fisher's test because of the small sample size. But now I looked into it, I'm not sure if any of these is the right test. Because within the experiment it's really unlikely that someone scores 1 at T1.

These are my predictions (by rule of thumb)

          |0   |1  |
     T1|0 |15  |5  |
       |1 |0   |0  |
prediction if H0 is correct

          |0   |1  |
     T1|0 |2   |18 |
       |1 |0   |0  |
prediction if HA is correct

1 Answer 1


This is an interesting question. My intuitive thinking is goodness of fit $\chi^2$ test:

  1. Merge these two contingency table
  2. Calculate $$\sum_{i=1}^4\frac{(O_i - E_i)^2}{E_i}$$ Here $O_i$ is the observed counts and $E_i$ is the expected counts.

if there is no difference between pre/post measurement, the above statistic should follow a $\chi^2_3$ distribution, where the degree of freedom is calculated as 4 - 1.

  • $\begingroup$ This Pearson test doesn't take into account information about paired measurements. $\endgroup$
    – chl
    Oct 24, 2020 at 18:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.