Say I did a survey of 100 people asking if they eat vegetables daily. Out of my sample population, 60 say they do, 40 say they don't. Now, I also have the breakdown of males and females. Say in the 60 who eat veggies daily, 45 are females and 15 are males. In the 40 who don't eat veggies daily, 15 are females and 25 are males.

Now, my initial study didn't control for gender but I now want to see if gender has any effect on whether they eat veggies daily or not.

Can I use Chi-Squared/Fischer's exact test with [45,15;15,25] as my contigency table? Or would I need to use a McNemar's Test and do paired matching?

Also if i did chi-squared would this be considered a "Chi-Square Goodness of Fit Test" or a "Chi-Square Test of Independence"?


It's a chi-square test of independence. You only have one test per male subject and one test per female subject. The test will tell you if the probability of eating vegetables is dependent on sex.

| cite | improve this answer | |
  • $\begingroup$ Thanks! But could I use McNemar's test if I manually paired up each male to 1 female? So one member of each pair is the "male" member and the other is the "female" member - making 40 pairs. Then I could just find the number of discordant pairs and compute a p-value. Would doing that give me better results or using a chi-square/Fisher exact test? $\endgroup$ – BYS2 Jul 18 '13 at 14:45
  • $\begingroup$ You need a pre-existing reason to pair. So, if they were married couples that might be OK. You need a reason like that. But it modifies what your test means. It's not that vegetable consumption depends on sex, it's that sex in a couple influences vegetable consumption. $\endgroup$ – John Jul 18 '13 at 16:33

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.