I am seeking to confirm that staff members are more likely to perform pro-environmental behaviour if they perceive support from their manager. The behaviours are measured by an ordinal scale (very frequently, frequently, occasionally, rarely and never), and the perception is measured by a likert-scale (strongly agree, agree, neutral disagree and strongly disagree).

What is the appropriate analysis for this? My N is 25 but I can collect more if necessary.

  • 1
    $\begingroup$ Do you want a model relating perceived support & environmental behaviors, or do you only need to test if they are related? $\endgroup$ Oct 6, 2015 at 20:59
  • $\begingroup$ I think test if they are related would be fine. Of course it is better with the model relating but I think it 's too complicated for me. $\endgroup$
    – Nhat Kha
    Oct 7, 2015 at 0:24
  • $\begingroup$ what is the scheme for scoring the responses for each of two variables? $\endgroup$
    – user10619
    Oct 7, 2015 at 13:37
  • $\begingroup$ the behaviours scored by very frequently, frequently ... never; and the perceived-support scored base on Strongly agree, Agree ... Strongly disagree $\endgroup$
    – Nhat Kha
    Oct 7, 2015 at 19:07

1 Answer 1


You have two ordinal variables and you want to see if they are associated. You should just compute an ordinal correlation, such as Spearman's, and test it against a null hypothesis of no association.

  • $\begingroup$ Thanks for your reply, so how about the sample size? $\endgroup$
    – Nhat Kha
    Oct 7, 2015 at 19:06
  • $\begingroup$ There is nothing necessarily wrong with N=25. The question is simply about how big an effect you want to be able to detect. $\endgroup$ Oct 7, 2015 at 19:34
  • $\begingroup$ My supervisor did not allow me to use Spearman, she told my sample is too small, if I try to use in my dissertation she will fail me as it against the principle of Spearman test :(. Do you have any reference so I can show her that N=25 is ok/ $\endgroup$
    – Nhat Kha
    Oct 12, 2015 at 22:10
  • $\begingroup$ @NhatKha, I've never heard of such a thing. I'm not aware of it. My copy of Hollander & Wolfe refers to the large sample approximation for the sampling distribution, but gives no guidance on what N is required to be 'large'. The prove that the mean & variance are correct w/ N=4, but that doesn't mean the shape of the sampling distribution is right, & refers to the case w/o ties, which you would have. You could always simulate the null distribution to be sure. $\endgroup$ Oct 12, 2015 at 22:47

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.