Someone asked:
Do you know when this term was coined or by whom? Kass and Raftery’s use of the tem as the title of their 1995 paper suggests that it was still novel then, but I have not noticed in the paper any information about where it started.
I replied:
According to Etz and Wagenmakers (2016), “The term ‘Bayes factor’ comes from Good, who attributes the introduction of the term to Turing, who simply called it the ‘factor.'”
They refer to Good (1988) and Fienberg (2006) for historical review.
I generally hate Bayes factors myself, for reasons discussed at a technical level in our Bayesian Data Analysis book (see chapter 7 of the third edition). Or, if you want to see my philosophical reasons, see the discussion around Figure 1 of this paper. Also this paper with Rubin from 1995.
Kass and Raftery (1995) hardly suggest that the term is novel, they discuss the Good/Turing/Jeffreys origins in Section 3.1
Pam:
Perhaps “not in common use” would be more accurate than “novel.”
Harold Jeffreys already used it in his 1939 Theory of Probability book, introducing the (in)famous log scale on the strength of the evidence produced by a Bayes factor. (He did not call it a Bayes factor.) See Ly, Verhagen, and Wagenmakers (2016) for details.
Andrew– okay: You hate Bayes Factor. But what do you *like* as a statistic for conveying the *weight of the evidence* associated with a particular study finding? And in particular, how can you like anything more than Bayes Factor, which is just a device for operationalizing the likelihood ratio term in Bayes’s Theorem? What will be more conducive than BF to thinking about the probative value of a study finding in Bayesian terms?
Dan:
The closest thing I’ve got is this.
Just because you would like “a statistic for conveying the *weight of the evidence* associated with a particular study finding” (or for any other purpose) doesn’t mean there is a good statistic that will do that.
+1
+1
But within the model (taking it as true enough for now) there is posterior/prior in full dimension of the parameters with unavoidably _risky_ marginalizations over the nuisance parameters to posterior.marg/prior.marg .
I was wondering is there a dataset and R code for BDA 3, particularly, Chap 14 (Bayesian regression)? I remember seeing them on the homepage somewhere.
http://www.stat.columbia.edu/~gelman/book/ should have everything you need.
I am aware of that. And Chapter 14 R code (incumbent election) is not in it.
Oops, I should have checked more thoroughly. The dataset is there, but you’re right, I can’t see the R code.