To run a certain activity, a population of companies use a finite set of inputs. There are 8 different inputs, but each company may not use all of them. For instance, company 1 uses inputs A, B and D, whereas company 2 uses inputs B, D, E and F

I have conducted a survey to ask 30 different companies about which inputs they use, and I would like to statistically evaluate the most used input. What statistical test do you recommend for this?

  • 1
    $\begingroup$ It is customary to 'accept' an answer by clicking on the checkbox next to the most useful answer (in this case, there is only one). This helps other find answers to their questions more quickly, as well as letting people know that your question has been resolved. Note that you can go back and do this for your other questions as well, by clicking on your name at the top of the screen. $\endgroup$
    – Jeff
    Aug 28, 2011 at 2:53

1 Answer 1


I don't think the chi-square test is appropriate because your counts / proportions are not independent. They are generated by the same sample of companies.

  • You might consider the McNemar test (see wikipedia). It is a statistical test specifically designed to compare two proportions that are based on the same sample of individuals. This could be used to see whether the input with the largest sample proportion of use is significantly greater than the proportion of use of any one other input.
  • The Cochran test (see wikipedia) extends the above idea to the case of three or more proportions. This could test the overall null hypothesis of whether the proportions for all inputs are equal.

Here's a tutorial example in R of using Cochran Q first followed by McNemar as a post-hoc test.


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.