why do the "flag supported correlations" start with BF above 10?

Simon1

Hi

I want to use Bayes factor analysis of a Pearson's correlation. When using the "flag supported correlations" it uses stars for BF above 10. I would like to know and cite that this is the praxis but cannot find where this is discussed anywhere. From my understanding for other test such as independent sample t-test the praxis has been to use BF above 3 to claim support for either of the H1 or H0. Is there any reason for this? I can not find any source that discusses it.

EJ

Hi Simon1,

These thresholds are somewhat arbitrary, and the evidential value is in the continuous measure. See for instance https://www.bayesianspectacles.org/lets-poke-a-pizza-a-new-cartoon-to-explain-the-strength-of-evidence-in-a-bayes-factor/

So the stars should not be taken too seriously. I am a little hesitant to add a star for the category bigger than 3 and lower than 10, especially because the matrix plot usually has many correlations and a correction for multiplicity may be in order anyway.

Cheers,

E.J.

GeorgesLemaitre

HI E.J.

I would argue against the "stars", since they will only provide a cue that will trigger old, learned behaviors, like hunting for them.

2c,

Georges

EJ

Hi Georges,

It actually was more or less a joke, to tease the Bayesians who (correctly) felt that evidence is continuous, not discrete. However, maybe some people find it useful -- you could argue that the non-diagnostic BFs deserve less attention than the diagnostic ones, and the stars are a way to highlight the diagnostic ones. At least the option is there, for those who want this. If multiple people complain (you are the first) we will take it out.

E.J.

GeorgesLemaitre

well, what a n00b i am. I have no excuses.

```{r, }

holidays %>%

filter (Christmas == 1)%>%

lm(happy ~ holiday!!)%>%

summary()

```