 |
 |
Best et al. (1999) indicated a model with a Deviance Information Criterion (DIC) minus the number of estimable parameters (pD) should roughly equal the sample size (n). DIC - pD = n. This supposition is based on Spiegelhalter et al. (1998), which my library has not yet attained. My question for this forum is whether values of DIC - pD considerably less than n and considerably more than n indicate over- and under-fitting, respectively? Best et al. (1999) indicate at least the former is true. If this read of Best et al. is correct, how much of a difference between DIC - pD and n is necessary to suggest under- and over-fit? Anybody use any rules-of-thumb? I ask this question because I have one model that has a Deviance - pD of 3296.8 and an n of 1840, which suggests dramatic underfit based on Best et al. For another model, the problem is worse, with a Deviance - pD of 7001.6 compared to an n of 1840. My concern is that these models are severely underfited. I calculated goodness-of-fit based on Gelman et al. (comparing parameter sets from the original data with those derived for a replicate set) and found GOF > 0.7 for each model. Has anyone considered the discrepancy between these assessments of model fit before (or am I looking at this completely wrong)? Best et al. 1999. Bayesian models for spatially correlated disease and exposure data. Bayesian Statistics 6:131-156. Spiegelhalter et al. 1998. Bayesian deviance, the effective number of parameters, and the complexity of arbitrarily complex models. Tech. Rep., Medical Research Council Cambridge. (this report is not available at http://www.mrc-bsu.cam.ac.uk/cgi-bin/pub-2002.cgi) Gelman et al. 1997. Bayesian data analysis. Chapman and Hall, NY, NY, USA. Best regards, Wayne E. Thogmartin, PhD Statistician (Biology) - Avian/Habitat Modeling USGS Upper Midwest Environmental Sciences Center 575 Lester Avenue Onalaska, WI 54650 608.783.7550 Ext 42 [log in to unmask] www.umesc.usgs.gov/staff/bios/wet0.html/ ------------------------------------------------------------------- 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