I am not a practicing bayesian when it comes to my research, but I've
always wondered why it is made difficult to input bayesian priors. This
seems like one of those cases where for once the gui interface is actually
useful.
Why not allow a user to use their mouse to graph the prior. The person
would type in upper and lower bounds (which could be negative and/or
positive infinity, or something lesser than those extremes). Then, they
could use the mouse to sketch the prior distribution in a window on the
screen, where the window invokes those upper and lower bounds. Then, some
mathematical engine would take over, and output posterior distributions,
hopefully with some standard errors. One could then easily allow the
researcher to 1)input priors for several coefficients; 2)play around with
changing the prior to see whether it matters--this would require allowing
the researcher to save the actual distribution they used, for possible
later publication and possible later editing; and 3)produce random
perturbations in the researcher's prior, or even produce a randomly
generated prior, again to check robustness. Although this next idea is a
bit dangerous, and may turn the approach into a cookie cutter idea, one
could allow the researcher to set upper and lower bounds, and then have
the machine generate a prior. Or, one could allow the researcher to input
parameter estimates from dozens of other studies, and generate a prior
distribution on that basis.
In sum, it would seem to me that if the idea of bayesian analyses is that
a prior distribution pertains to the parameter and reflects our current
knowledge, why not use the GUI to allow people to directly draw their
prior distribution? I believe once people become exposed to bayesian
thinking, thinking in terms of distributions is easy. The hard part is
getting that distributional information into the analysis. The GUI
interface would seem to greatly simplify that process.
Is this a bonehead idea? Am I missing something crucial? Comments
appreciated.
Take care.
Sam
On Sun, 5 Mar 2006, Maarten buis wrote:
> I agree with Greg here. I have been playing with the thought, but I got stuck while devising a
> syntax that is both flexible enough and easy to use.
> Maarten
>
> --- Gregor Franz <[email protected]> wrote:
> > I use matlab. I think it would be hard to incorporate anything but
> > mostly trivial cases for conjugate priors into a packaged program like
> > stata. That's the problem with subjective beliefs, you can't generalize
> > them.
>
> > > Quoting "Benjamin M. Craig, Ph.D." <[email protected]>:
> > >> As was mentioned in the fall, "Stata is not much of a vehicle for doing Bayesian things."
> > >> Should this change?
>
>
> -----------------------------------------
> between 1/2/2006 and 31/3/2006 I will be
> visiting the UCLA, during this time the
> best way to reach me is by email
>
> Maarten L. Buis
> Department of Social Research Methodology
> Vrije Universiteit Amsterdam
> Boelelaan 1081
> 1081 HV Amsterdam
> The Netherlands
>
> visiting adress:
> Buitenveldertselaan 3 (Metropolitan), room Z214
>
> +31 20 5986715
>
> http://home.fsw.vu.nl/m.buis/
> -----------------------------------------
>
>
>
> ___________________________________________________________
> Yahoo! Photos � NEW, now offering a quality print service from just 8p a photo http://uk.photos.yahoo.com
> *
> * For searches and help try:
> * http://www.stata.com/support/faqs/res/findit.html
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
>
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
