Is there a favor for reporting BF10 over BF01?
Hi there!
In one of my papers, I calculated Bayes Factors using JASP (great tool!!). Now a reviewer has commented on my results with the following remark:
"I am no expert on Bayesian Statistics, but as far as I’m concerned you usually use the BF01 to report evidence speaking in favour of the null hypothesis, whereas BF10 reports the probability of the data for an alternative hypothesis. In other words, by using BF10 you are trying to find evidence for H1, which you do not find - in parallel to the frequentist analysis. According to me, it would be better to look at BF01 and check whether you can find evidence for your null finding."
I would argue that it doesn't matter to report BF01 over BF10 as ultimately both scores provide the same information as BF01 = 1/BF10. However, perhaps I'm missing some point he/she is trying to make that the direction of the interpretation. Perhaps my phrasing in the original text was a bit unclear.
Here's an excerpt of my own manuscript:
"A JZS Bayes factor ANOVA(ref55) with default prior scales was performed to estimate the likelihood of the interaction effect. This resulted in a BF10 of 0.005, thereby providing very strong evidence for the absence of an effect(ref56). The effect of time and group alone resulted in a BF10 of 0.209 and 0.158 respectively"
Comments
You're 100% right - BF10 = 1/BF01
I always prefer to write the actual values so that they are human-interpolate (which ever is larger than 1).
e.g. it's easier to understand that H1 was 4.784 times less likely than H0 (BF01 = 4.784), than that H1 was 0.209 times more likely than H0 (BF10 = 0.209).
I guess you could write that BF10 = 1/4.784 if they insist...