On the NIST page about McNemar's Test, the third listed assumption is that "The difference P(Xi = 0, Yi = 1) - P(Xi = 1, Yi = 0) is negative for all i or zero for all i or positive for all i."

Let's say I have a sample that is half Democrats and half Republicans and I'm going to present them a video discussing both sides of a controversial issue. Have I violated the assumption above because the Democrats are more likely to move a certain direction on the issue whereas the Republicans are more likely to move the opposite direction?

  • $\begingroup$ That page has a peculiar style muttering all the way, "for all i". It is easier to say, the test is testing the symmetry of the count (or fraction) table in the population. The test is repeated-measures, bot between-group test. It assumes that the population is one. So, if you just mix Dem. and Rep. in one sample to test how the sample prefers either "pro" or "con" before the video and after the video, - you are assuming your sample is "homogeneous" for you: it represents one population, "party people". $\endgroup$ – ttnphns Sep 26 '17 at 19:08

Your Answer

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

Browse other questions tagged or ask your own question.