I want to compare the number of lesions worsening and non worsening on superior to inferior side of the eye of a group of patients. I want to compare if the proportional results (the total sum of worsening and total sum of non worsening in the superior vs inferior side ) have a significant difference. In my data I have:

  • worsening 2618 (66.8%) INFERIOR

  • non worsening 1298 (33.1) INFERIOR

  • worsening 2981 71.4% SUPERIOR

  • non worsening 1190 28.5% SUPERIOR

Which method should I use since these proportions are dependent (it comes from the same patient), but not paired(there is no before after rather different locations)?


1 Answer 1


Those are definitely paired. Pairing is not restricted to 'before-after' comparisons.

Anything where within-pair values (like values from two locations on a single patient) would tend to be more alike than out-pair values (values from the corresponding two locations on two different patients) are paired.

(This is a disturbingly common error -- presumably there's some widely-used elementary books promulgating this mistaken notion.)

  • $\begingroup$ Thanks for clarifying, however when I look up in the literature which method use to compare proportions I found chi square for independent variables, which method should i used for proportions if these are dependent variables? $\endgroup$
    – D.Vega
    Nov 12, 2018 at 22:50
  • $\begingroup$ It's not clear what your null and alternative hypothesis is when comparing four proportions. Can you clarify? (It will likely be some form of chi-squared test, but more information is needed). Is the fact that the first pair of percentages and the second pair don't quite add to 100% just rounding error or is there a (tiny) third category? $\endgroup$
    – Glen_b
    Nov 12, 2018 at 23:22
  • $\begingroup$ You are right, there was a rounding error, well I like to compare first the proportions of worsening category (INF 66.9 vs SUP 71.5) and compare the proportions of non worsening as well (INF 33.1 vs SUP 28.5) and I would have two result for statistical significance $\endgroup$
    – D.Vega
    Nov 12, 2018 at 23:53
  • $\begingroup$ Thanks for the clarification The two comparisons are dependent, since 66.9 = 100-33.1 and 28.5 = 100-71.5. If one is different it would imply that the other is. $\endgroup$
    – Glen_b
    Nov 12, 2018 at 23:55
  • $\begingroup$ I wasn't clear sorry about that, My null hypothesis: no difference in proportions between sup e inf, I want to compare just two proportions but the chi square didn't seem appropriate because it assumes these 2 proportions are independent when in fact are dependent hence my confussion. I really appreciate your help Glen_b!! $\endgroup$
    – D.Vega
    Nov 12, 2018 at 23:57

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.