I’m sure this is an easy question for econometricians rather than me as a statistician, but when using non-parametric correlation GRETL gives significance values and only one correlation at a time, but when using parametric correlation, we get a matrix but no significance values.

Hi, let's first make sure we're talking about the same things here: When you say "non-parametric correlation", I guess you're using the GUI and went to Tools/Nonparametric Tests/Correlation, where you have the choice between Kendall's tau and Spearman's rho (and can only enter two variables). And I suppose that "parametric correlation" means going to View/Correlation matrix, where you can choose an arbitrary number of variables. Please correct me if that's not what you're doing.

Here are some remarks about that:

- I can confirm the limitation of two variables for the Kendall / Spearman tests. If scripting is an option, then it's fairly easy to write a loop like this:

<hansl>

</hansl>

Admittedly, output in the form of a matrix would be a little more work. Perhaps at some point the 'corr' command could be extended to cover this.

- About the parametric thing, I'm noting that you get the pairwise test for n=2, and a (significance) test for non-correlation also for n>2, but the latter appears to be a joint test with respect to all off-diagonal correlations. If I understand correctly, you want a battery of pairwise tests. Of course, you could simply reuse the loop above without the --kendall option. But given that we even have a heatmap button in the correlation output, I agree that it would make sense to compute the pairwise tests automatically.

I also ask this question as when teaching using GRETL, a matrix for non-parametric correlation would be a preferred output that students can put in a document rather than creating one themselves from individual answers.

 

Thanks for your feedback

cheers

sven