Appendix C. Summary of model fits of several competing models.
Estimation was done using Markov chain Monte Carlo simulation methods as implemented in the program WinBUGS.
DIC is the deviance information criterion as discussed in Spiegelhalter et al. (2002). It is a measure of error plus a penalty term of 2 *times effective number of parameters in a model. The smallest DIC is considered optimal.
The following is the most parsimonious model that emerged as best model using reversible jump MCMC:
Model 1: DIC: 2989.90 | |
Posterior Means of standard deviation of (NonShoals , Shoals): 0.664 , 0.983 |
The following includes a spatially varying residual field:
Model 2: DIC = 3009.77 | |
spatially correlated with exponential correlation function. | |
Posterior Means of standard deviation of (NonShoals , Shoals): 0.697 , 2.017 | |
Posterior Means of correlation range in km: (NonShoals, Shoals): 5.19, 39.34 * | |
* these posteriors were highly skewed with medians (1.98, 1.71) |
The following has a spatially varying ln.gulls coefficient:
Model 3: DIC = 3008.70 | |
spatially correlated with exponential correlation function. | |
Posterior Means of standard deviation of (NonShoals , Shoals): 0.984 , 91.28 | |
Posterior Mean of standard devation of: 14.58. | |
Posterior Mean of correlation range ofin km: 33.6 |
LITERATURE CITED
Spiegelhalter, D. J., N. G. Best, B. P. Carlin, and A. vanderLinde. 2002. Bayesian measures of model complexity and fit. Journal of the Royal Statistical Society, B 64:583–639.