Correlation vs Regression (JASP 0.8)

MSB

Hi there,
I was wondering why correlations and simple regression give different BFs.
For example (using the new shiny Summery Stats in JASP 0.8):

• A correlation of 0.5 (n=50) yields a BF=134
• A similar R^2 = 0.25 (n=50) yields a BF=112
  (this isn't really a big difference, but this did bother me somewhat..)

also, I couldn't find any documentation explaining how the Prior r scale functions?

Thank,

M

EJ

Hi MSB,

The model formulation is different. The regression model is based on the Jeffreys-Zellner-Siow framework, and the correlation model is based on the initial proposal by Jeffreys. The paper by Ruud Wetzels presents a correlation in line with the JZS approach, but for JASP we chose the Jeffreys framework because we felt it was more intuitive.

More info on JZS correlations: http://www.ejwagenmakers.com/2012/WetzelsWagenmakers2012.pdf
More info on Jeffreys-style correlations: http://www.ejwagenmakers.com/2016/LyEtAl2016JMP.pdf

Cheers,
E.J.

MSB

Hi EJ,

Thanks! Will read!