Question about summary stats

NumCog

Hi,

First of all, thanks for a great JASP course earlier this week!

I have two questions about the new summary stats module. I am not 100% sure how JASP calculates the BF for an existing correlation from just the sample size and the Pearson's r-value. Does this mean that for every sample of X and r-value of Y the BF is Z?

Also, could I use the summary stats module to create a qualitative meta-analysis? I was thinking about calculating the BF for correlations in a given domain and then calculating the percentage of them that exceed a BF=10 to assess the strength of evidence across studies. Do you think this would be a legitimate approach to go?

Thanks,
Daniel

EJ

Hi Daniel,

Yes, that is correct: for the BF and the posterior distribution, r and N alone are sufficient. See http://arxiv.org/abs/1510.01188 for details.

As far as the quantitative meta-analysis is concerned, I'd prefer to see all of the BFs; using a cut-off loses information. But otherwise, yes, these kinds of endeavors are what the module was meant for.

Cheers,
E.J.

NumCog

Hi EJ,

Thanks so much for your response and the link to the paper!

If I may ask a quick follow up: for the quantitative meta how would you go about creating some kind of aggregate BF? So if I was comparing correlations in domain X with correlations in domain Y, how would I provide some kind of summary stat of the evidence (an aggregate BF)? Any suggestions for this?

Best,
Daniel

EJ

First, Richard Morey and Jeff Rouder have a paper on BF meta-analyses (2011 in PBR, a response to Bem). Second, are you interested in a BF for the difference in correlations?
E.J.

NumCog

Thanks! I will read that paper.

Yes, I am interested in a BF for the difference in correlations!
Best,
Daniel

EJ

Ah yes. Well that requires a separate development. You could plot the two posteriors and eyeball them, but that is not a real test. We have it on our radar but it will take some time to get done.
E.J.

NumCog

Great. Thanks!