-1
$\begingroup$

I am working on a systematic review and want to compare the correlation of different devices.

Both Pearson as Spearman correlations are reported in papers. Because they both represent a correlation, is it possible to put them all in one database and compare them? Or is this not possible.

Improve this question
$\endgroup$
3
  • 4
    $\begingroup$ What do you mean by "putting them all in one database"? What do you want to compare? Is it theoretical differences or some kind of performance comparison? $\endgroup$
    – Tim
    Commented Jul 8, 2016 at 12:12
  • 1
    $\begingroup$ I'm not sure what you're asking exactly, but if you wanted to do a meta-analysis, eg, you wouldn't use Pearson and Spearman interchangeably. $\endgroup$
    – gung - Reinstate Monica
    Commented Jul 8, 2016 at 12:33
  • 1
    $\begingroup$ It's always useful to point out that the assumptions underlying these metrics differ. Pearson captures linear association while Spearman is a measure of monotonic (ordinal) association. Many other measures of nonlinear, pairwise association are out there including the MIC (mutual information criterion), distance correlations, Brownian correlation, reproducing kernel Hilbert spaces, etc. $\endgroup$
    – user78229
    Commented Jul 8, 2016 at 12:40

1 Answer 1

Reset to default
1
$\begingroup$

I suggest you read this thread, which sheds an interesting light on differences between both coefficients in rather simple datasets. I guess, that afterwards you will probably decide not to mix them:

What is the explanation for having a Pearson's correlation coefficient significantly larger than the Spearman's rank correlation coefficient?

Edit: Greenparker is right in his comment, that I should explain here. In the thread I linked, I give an example of a dataset in which Pearson and Spearman differ a lot, as much as one being positive an the other being negative. A shorter example for the purpose of this thread would be the following in R:

> x=c(1.0, 1.01, 1.02, 1.03, 1.04,  100)
> y=c(0.04, 0.03, 0.02, 0.01, 0, 100)
> cor(x,y, method="spearman")
[1] -0.1428571
> cor(x,y, method="pearson")
[1] 0.9999998

As you can see, for the same six simple data points, Spearman is almost zero and negative whilst Pearson is almost 1. I think this should be illustrative enough, not to mix them as "almost the same".

Improve this answer
$\endgroup$
3
  • 2
    $\begingroup$ If you are only going to point to this thread, this should be a comment, not an answer. $\endgroup$
    – gung - Reinstate Monica
    Commented Jul 8, 2016 at 12:35
  • $\begingroup$ While this link may answer the question, it is better to include the essential parts of the answer here and provide the link for reference. Link-only answers can become invalid if the linked page changes. - From Review $\endgroup$
    – Greenparker
    Commented Jul 8, 2016 at 13:15
  • $\begingroup$ @Greenparker You are right. I edited and added a short example that has the same essence as the much longer answer I gave in the other thread. I hope, that for the purpose of this thread, the shorter example is sufficient. $\endgroup$
    – Bernhard
    Commented Jul 8, 2016 at 13:33

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