var.test and bayes factors

robpetrosino · August 2019

I ran an F-test to compare the variances of two samples from normal populations with the base R command var.test. I wanted to extract the Bayes Factor (BF) from it by using the oneWayAOV.Fstat command, but I got weird negative numbers such as:

oneWayAOV.Fstat(2.5764, 72, 2)$bf # 10BF
[1] -0.5450246

1/oneWayAOV.Fstat(1.2587, 72, 2)$bf # 01BF
[1] -0.8737903

As far as I could understand, BFs should never be negative... What is happening here?

Thanks!

EJ · August 2019

I'll pass this one on to Richard...

richarddmorey · August 2019

Those are log Bayes factors (see https://www.rdocumentation.org/packages/BayesFactor/versions/0.9.12-4.2/topics/oneWayAOV.Fstat).

robpetrosino · August 2019

Well, I did not catch that. Sorry, that was pretty easy to debug. Can I ask what the purpose of calculating log BFs?

EJ · August 2019

For raw BFs, the scale is asymmetrical: evidence for H0 (say) goes from 0 to 1, whereas evidence for H1 goes from 1 to infinity. So a BF of 1/200 is of the same strength as a BF of 200. If you take the log, evidence for H0 is -x, and the same strength of evidence for H1 is +x. In some very specific situations, you can also add the log BFs (instead of multiplying the BFs).

Cheers,

E.J.

Howdy, Stranger!

Categories

var.test and bayes factors

Comments

Howdy, Stranger!

Quick Links

Categories

var.test and bayes factors

Comments