Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I did correlation tests between a number of variables consisting of both interval and categorical scales ($n = 1274$, one sample). I used $\chi ^2$ test for categorical vs ordinal and Spearman's $\rho$ for interval vs ordinal data.

The correlation between climate zone (zone $1, 2, 3, 4$) and daily exposure to air-conditioning (low, medium, high) is $0.35$ (Cramer's V).

The correlation between exposure to AC and pro-environmental attitude score ($1-5$, interval) is $0.21$ (Spearman's $\rho$).

Can I compare the strength of the relationship if I used different correlation tests, say the relationship between climate and exposure to AC is stronger than that between the pro-environmental attiudes and exposure to AC?

share|improve this question
up vote 2 down vote accepted

Two small points:

1) As mentioned before - the two correlation measures are measuring different things, so a strict comparison of the two is risky.

2) If you also want to make a formal inference of the difference between the two, you will probably need to go with a permutation test, since (to my knowledge) there is no formal test that will make this comparison.

share|improve this answer
Many thanks Tal Galili. Could you please explain more about the permutation test? I really have no idea what it is. – tida Sep 23 '12 at 8:47

Up to a point. The closer you get to conditions of interval- or ratio-level variables with normal distributions, the greater the opportunity for correlations to reach high values if the underlying associations are linear and strong. With ordinal variables and distributions that are far from normal, correlations will be attenuated: underlying relationships that are linear and strong will not show up as quite so strong in the observed coefficient. Point-biserial correlations, which connect interval with binary variables, will also suffer from this sort of attenuation. There just isn't as much opportunity for the two variables to co-vary when, for every small increment of the one measured on an interval scale, the other either remains a 0 or jumps to 1.

share|improve this answer
+1, nice points. It's good to be seeing you around again, @rolando2. – gung Sep 22 '12 at 2:20
@gung - Thank you! – rolando2 Sep 22 '12 at 13:33
Thank you very much rolando2. So, you mean it is not fair to compare these correlation coefficients. – tida Sep 23 '12 at 8:51
"It is wrong to state categorically that something is right or wrong" :-) I tried to shed light on conditions that tend to limit the soundness of making such comparisons. – rolando2 Sep 23 '12 at 22:45
goiod answer. ratio measures do not seem to recognise non-linear relation and hence produce higher estimates of correlation coefficient. May be K.P. correlation coefficient - a popular measure of correlation does so. – subhash c. davar Nov 5 '15 at 14:06

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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