3
$\begingroup$

My field regularly demonstrates a certain type of result with pairwise Pearson's correlation matrices between predicted and measured data. As soon as such correlations become high, Fisher-transforming the correlations will (visually) pronounce differences between the pairs better than leaving the correlations uncorrected. The magnitude of these differences between pairwise correlations is relevant when interpreting the matrices as a visual result.

I can understand the motivation behind including the matrices in this way in a paper, but is it really correct to report Fisher-transformed correlation matrices in this case?

Improve this question
$\endgroup$
1
  • 1
    $\begingroup$ You might find some of the answers and comments in this stats.stackexchange.com/questions/62824/… helpful although they do not directly answer your very specific question. $\endgroup$
    – mdewey
    Commented Feb 22, 2017 at 13:35

1 Answer 1

Reset to default
3
$\begingroup$

The reasonableness of using the transformation this way depends more on the theoretical context it is used in than on purely statistical considerations. For example, if the differnce between a correlation of .99 one of .995 is not theoretically interesting, then you probably would not want to do the transformation. On the other hand, if this difference is meaningful and theoretically more important than a difference between .79 and .795 then the transformation makes sense. In short, this is not about whether using the transformation is "correct" statistically but rather whether it gives you values that are more meaningful in the context of your theoretical framework.

Improve this answer
$\endgroup$

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.