Bayesian Linear Regression - Discrepancies between SPSS and JASP Bayes Factor?
Dear all,
I have a problem with discrepancies in the Bayes Factor computed with SPSS and with JASP when running a Bayesian Linear Regression, and would be very thankful for some help.
So, here is what I did:
First, I have run Bayesian Linear Regressions with SPSS for several dependent variables. For this, I added metric covariates and sticked to the default options. The resulting BF10 range from 0.001 to 0.005.
Second, I have run the Bayesian Linear Regression again with JASP. Here again, I used the same metric covariates and sticked to default options. The resulting BF10 ranges now from 0.028 to 0.092. Therefore, there is a difference of about one decimal place when comparing the outputs of both programs...
I already checked whether there might be a difference in the data I used, but when I run a frequentist linear regression, both programs show exactly the same output - so, this is most likely not the problem. Additionally, I compared the default options of both SPSS and JASP and came to the conclusion that it should be generally the same (at least as far as I understand it with basic statistical knowledge), e.g. SPSS uses JZS as prior just like JASP.
Is anyone familiar with this problem? Is there maybe something I might have missed when either conducting the analysis with SPSS or JASP?
Thanking you very much for your help in advance!
Miriam
Comments
Hi Miriam,
It may well be the prior settings (i.e., the width of the JZS prior). I am not sure what procedure is used in SPSS. JASP uses the BAS package by Merlise Clyde.
Cheers,
E.J.
Hi E.J.,
thank you for the tip - it solved the problem!
I adjusted the r parameter of the JZS method to 1 and now I get the same results as in SPSS.
Thank you!
Miriam
Miriam! Thank you for posting this! I've been googling around for a MONTH trying to find the answer to this question, and finally here it is. Thanks!