Hi All!
Andrew Gelman's message gives me a nice excuse to share some concerns
that I have always had with you all. I think that he problem might
not be caused only by the use of the inverse-gamma . It has repeteadly
been reported that using a "vague proper prior" which is "close" to
being "non-informative" in the sense of being "close" to an improper,
non-informative prior is a HORRIBLE idea when this improper prior
would produce an IMPROPER posterior. This is the case with the
variance components problem: the prior 1/sigma^2 for the variance
component (corresponding to an inverse-gamma(0,0) ) yields and
improper posterior, and any attempt to get "vague" with an
inverse-gamma with small values of the hyper-parameters is not a good
idea, and will not work. On the other hand, a constant prior on sigma
yields a proper posterior; hence, approximating this improper constant
prior with a "chopped" proper uniform should work much better.
This, I think, should be a subject of some concern and lots of work,
since the use of the dangerous kind of "vague proper priors" is quite
frequent.
Susie Bayarri
--
********************************
M.J. bayarri
Univ. Valencia, Dept. of Statistics and O.R.
Av. Dr. Moliner 50
46100 Burjassot, Valencia, Spain
Ph: (34) 96 354 4309
(34) 96 354 4709
********************************
---------- Mensaje reenviado ----------
Para: [log in to unmask]
Asunto: problem with inverse-gamma prior distribution
Reply-to: Andrew Gelman <[log in to unmask]>
De: Andrew Gelman <[log in to unmask]>
Fecha: Tue, 29 Jul 2003 18:49:52 -0400
Hi all. I just wanted to let youall know that problems can occur with
seemingly-noninformative inverse-gamma prior distributions for variance
components. In many cases, it's more reliable to simply use uniform
prior distributions on the standard deviation parameters.
We have an example in Appendix C of the second edition of our book,
"Bayesian Data Analysis."
To get this appendix, just go to the bugs.R page
(http://www.stat.columbia.edu/~gelman/bugsR/) and
near the top of the page, there's a place to click to download Appendix C.
For example, in a simple 1-way data structure (the "8 schools" example
from "Bayesian Data Analysis"), we use the following model:
model {
for (j in 1:J){
y[j] ~ dnorm (theta[j], tau.y[j])
theta[j] ~ dnorm (mu.theta, tau.theta)
tau.y[j] <- pow(sigma.y[j], -2)
}
mu.theta ~ dnorm (0, 1.0E-6)
tau.theta <- pow(sigma.theta, -2)
sigma.theta ~ dunif (0, 1000)
}
This is a uniform prior distribution on the the sd parameter
(sigma.theta) and works fine.
An alternative uses the inverse-gamma prior, replacing the last 2 lines
in the above model by,
tau.theta ~ dgamma (1, 1)
sigma.theta <- 1/sqrt(tau.theta)
This does not work well at all (see Figure C.3 on page 597).
- Andrew Gelman
P.S. I'm not trying to knock the inverse-gamma model in general. It
can be very effective, especially when applied hierarchically to several
variance parameters. I'm just making a recommendation for what to do
when a noninformative distribution is required.
-------------------------------------------------------------------
This list is for discussion of modelling issues and the BUGS software.
For help with crashes and error messages, first mail
[log in to unmask]
To mail the BUGS list, mail to [log in to unmask]
Before mailing, please check the archive at
www.jiscmail.ac.uk/lists/bugs.html
Please do not mail attachments to the list.
To leave the BUGS list, send LEAVE BUGS to [log in to unmask]
If this fails, mail [log in to unmask], NOT the whole list
-------------------------------------------------------------------
This list is for discussion of modelling issues and the BUGS software.
For help with crashes and error messages, first mail [log in to unmask]
To mail the BUGS list, mail to [log in to unmask]
Before mailing, please check the archive at www.jiscmail.ac.uk/lists/bugs.html
Please do not mail attachments to the list.
To leave the BUGS list, send LEAVE BUGS to [log in to unmask]
If this fails, mail [log in to unmask], NOT the whole list
|