18
$\begingroup$

I'd like to estimate the correlation between:

An ordinal variable: subjects are asked to rate their preference for 6 types of fruit on a 1-5 scale (ranging from very disgusting to very tasty) On average subjects use only 3 points of the scale.

A continuous variable: the same subjects are asked to quickly identify these fruits, which results in an mean accuracy for the 6 fruits.

Is Spearman rho the best method to analyze these data and/or are there other good methods I could consider?

Improve this question
$\endgroup$
2
  • 1
    $\begingroup$ Welcome to the list. The ordinal variable looks like it is actually 6 variables (one for each fruit). It's also not clear to me how the identification variable is created, nor that it is continuous. $\endgroup$
    – Peter Flom
    Dec 4, 2013 at 12:34
  • $\begingroup$ Thanks thats quick! So for each subject I indeed have 6 preference ratings, and 6 accuracy ratings. Accuracy is the mean hitrate over 16 identification trials (16 for each type of fruit). A hit is when they select the right fruit, miss is when they select the wrong type of fruit. Hope that this made it more clear. $\endgroup$
    – San
    Dec 4, 2013 at 13:04

2 Answers 2

Reset to default
7
$\begingroup$

You could use Spearman's, which is based on ranks and therefore OK for ordinal data. You would then have six results.

If you want to take a different approach, you could get complex and look at a multilevel model, with subject being repeated. It sounds like "accuracy" would depend on "preference". So, a mixed model could look at that and account for the non-independence of the data. But, as noted, that's a much more complex model to implement.

Improve this answer
$\endgroup$
1
$\begingroup$

According to this paper* "Measures of Association: How to Choose?" (doi:10.1177/8756479308317006), you should consider kendall's tau-b if the number of items in your ordinal variable is low (<5 or <6 ... this is a bit arbitrary).

If you have a large number of items in your ordinal variable, Spearman correlation would work well.

*the paper may be behind a paywall. But I tried to summarize the essence in my post.

Improve this answer
$\endgroup$

Your Answer

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

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