1
$\begingroup$

I understand that the difference between a chi-square test of homogeneity and the McNemar test is that the chi square is used for unpaired data and the McNemar is used for paire data.

Here's my situation. I am running some tests on some devices and I am recording pass and fails, so basically 0 and 1 for each device. Week 1: I have let's say 5,000 devices and I have pass/fail results for each of them. Week 2: additionally to those 5,000, I have another 3,000 devices.

My goal is to test whether or not there is difference between those 2 groups. The first group here being the 5,000 devices from Week 1. What should my second group be? 3,000 devices or 8,000 devices (i.e. combination from week 1 and week 2)

Here's my question: How do I set up my groups? 1- Do I have my first group being the 5,000 devices? 2- Can I have my second group being the 8,000 devices, does it have to be the new 3,000 devices? 3- if answer to question 2 is that my second group has to be the new 3,000 devices, then the test to apply is a chi squared, correct? (unpaired data)

$\endgroup$
1
$\begingroup$

Without knowing the difference between the 5000 and the 3000, I will have to be very general. However, I think you already answered this here "My goal is to test whether or not there is difference between those 2 groups."

Group A (n = 5000) and Group B (n = 3000) should be separate. Group C (n = 8000) can also be added, but any differences between B and C (for example) could be the result of the differences between A and B. Keep in Mind you'd be testing if A is different than A+B, which just means 'is A different than B?'. Hopefully that makes sense? Again, I don't know if you are separating based on technology, brand, manufacturer, etc.

SO, to answer your question: the chi-squared for independence is correct. Keep the two groups separate, otherwise you are testing a portion of the same devices against themselves.

$\endgroup$

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.