I have a dataset which includes responses to a survey given to two groups. One of the group has 350 respondents, while the other is greater and contain ~ 4000 respondents. An example to a question is: How much time do you spend on TV, out of your total leisure time? The responses are: (a) A lot (b) Some time (c) Not that much etc.

Note that some of the answers as "ordinal" nature (e.g.: Always, often, sometimes, Never), and some are just multiple choice questions.

I am trying to show that both groups have similar proportions for each question.

I am quite confused with the possibilities. I tried to convert the responses I have to a table comparing each level's number of responses between the two groups. For example:

           group1   group2
    Never   20         200
    Often   10       1500
    Always   80      1300

and use this table as a matrix which is passed to chisq.test, but I'm not sure it tests what I want to test. I also tested the possibility of wilcox test and read about prop.test but still - can't find the suitable method.

Any help would be very much appreciated!

  • 2
    $\begingroup$ "Trying to show that both groups have similar proportions for each question." This is not a chisqu situation, as that test, whether there is any difference, no matter how small the difference is. You will have to find some definition of "similar" to make the question more precise. $\endgroup$
    – Bernhard
    Commented Nov 17, 2017 at 12:35
  • $\begingroup$ The Fisher exact test is another possibility for this type of problem. $\endgroup$
    – Dave2e
    Commented Nov 17, 2017 at 15:02

1 Answer 1


For your ordinal responses, look into ordinal logistic regression. See for instance How to understand output from R's polr function (ordered logistic regression)?. Use group membership as predictor.

For the nominal (not ordered) responses a chisquare test could be fine. But maybe multinomial logistic regression is better, see for instance Logistic regression for multiclass.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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