Here are a few quick suggestions regarding Wayne Thogmartin's questions
about fitting, understanding, and checking complex multilevel models.
Andrew
1. Those inverse-gamma prior distributions can create problems; uniform
prior dists on the sd's would be better; see here:
http://www.stat.columbia.edu/~gelman/research/published/taumain.pdf
2. If you have a lot of variance parameters, I suggest modeling them
hierarchically using the half-Cauchy family; see Section 6 of the
above-referenced paper.
3. If convergence is still slow after the above 2 steps, try parameter
expansion (see pp.597-598 in Appendix C of Bayesian Data analysis); the
appendix is also here:
http://www.stat.columbia.edu/~gelman/bugsR/software.pdf
4. To understand the model fit, I recommend plots of data (or estimated
coefficients) and fitted curve, at each level of the model. For
example, see Figures 1 and 2 of this paper:
http://www.stat.columbia.edu/~gelman/research/published/multi2.pdf
or Figures 2 and 3 of this paper:
http://www.stat.columbia.edu/~gelman/research/published/parkgelmanbafumi.pdf
5. You can check model fit by simulating replicated datasets and
comparing to the actual data. Graphical comparisons are usually best,
in my experience. See Chapter 6 of BDA, and p.598 of the aforementioned
appendix for an example in Bugs.
Sorry for referring only to my own stuff. These are topics that I've
thought a lot about, and naturally I find my own writing on the topic
easiest to understand! Good luck.
Wayne E Thogmartin wrote:
>
> I would appreciate if interested folks might comment on the attached
> code which attempts to model factors associated with the
> discrimination between four land cover states. The global model has a
> slew of random effects which makes convergence difficult. The
> attached code is more quickly able to converge, but still has some
> mixing problems. I'm wondering if I might make the code more
> efficient. I would also appreciate thoughts on approaches to
> interpretation; currently, I examine the results to identify patterns
> in the random effects (i.e., where some groups differ in interesting
> ways from the main). I'm also a bit unsure how I might assess model
> fit. Any thoughts would be appreciated.
>
> Regards
> Wayne Thogmartin
>
>
>
>
>
> Wayne E. Thogmartin, PhD
> USGS Upper Midwest Environmental Sciences Center
> 2630 Fanta Reed Road
> La Crosse, WI 54603
> 608-781-6309 (off)
> 608-783-6066 (fax)
> [log in to unmask]
> http://www.umesc.er.usgs.gov/terrestrial/migratory_birds/bird_conservation.html
>
> The contents of this message are mine personally and do not reflect
> any portion of the US government.
>
> "A recurrent theme in the history of science is that novel claims
> conjoined with a paucity of data inevitably attract the attention of
> statisticians, much in the manner that offal attracts flies." W.M.
> Schaffer
--
Andrew Gelman
Professor, Department of Statistics
Professor, Department of Political Science
[log in to unmask]
www.stat.columbia.edu/~gelman
Tues, Wed, Thurs:
Social Work Bldg (Amsterdam Ave at 122 St), Room 1016
212-851-2142
Mon, Fri:
International Affairs Bldg (Amsterdam Ave at 118 St), Room 711
212-854-7075
Mailing address:
1255 Amsterdam Ave, Room 1016
Columbia University
New York, NY 10027-5904
212-851-2142
(fax) 212-851-2164
-------------------------------------------------------------------
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
|