1

I computed Bayes Factors for a repeated measures ANOVA in JASP and the R package BayesFactor. There were two between-subjects groups (factor "group") with multiple measurements. However, the results were different:

JASP output:

Model comparison - dependent
Models      P(M)        P(M|data)       BF M            BF 10       % error
group       0.200       5.618e -56      2.247e -55      0.174       0.944

Analysis of Effects - dependent
Effects     P(incl)     P(incl|data)    BF Inclusion 
group       0.600       0.169           0.135

vs.

BayesFactor model:

bf = anovaBF(val ~ group + subj, data = dat, + whichRandom="subj")

BayesFactor output:

Bayes factor analysis
--------------
[1] group + subj : 0.1693252 ±0.69%

This is my first Bayesian analysis, so I may be missing some obvious points. The analyses were performed according to the manuals though.

Update: I now noticed that the BayesFactor output is identical to the P(incl|data) value provided by JASP. I am trying to find evidence for the null hypothesis. I want to report the BF 10, but I do not know how it relates to the BayesFactor output. I would be very thankful for a clarification which portions of the output are relevant.

Stephan Kolassa
  • 95,027
  • 13
  • 197
  • 357

1 Answers1

2

I think what we have here is just the effect of Monte Carlo sampling noise. If you look at the % error, you can see that each Bayes factor is estimated with about 1% error. These Bayes factors are not exact; they will change due to the sampling from computation to computation. Try re-running the anovaBF function; you'll see the value change.

The BF10 column in the JASP output gives you the same Bayes factor as the anovaBF output, just with different sampling noise. I suspect that the 1% sampling error accounts for this difference.

richarddmorey
  • 556
  • 2
  • 4