DIC and mixture model references

They proposed 8 DIC alternatives in missing data analysis (Celeux, et al, 2003) http://www.ceremade.dauphine.fr/~xian/cfrt05.pdf "DeIorio and Robert (2002) described some possible inconsistencies in the definition of a DIC for mixture models, while Richardson (2002) presented an alternative notion of DIC, again in the context of mixture models." DIC in Mixture likelihood has been discussed in Spiegelhalter (2002), but mixture likelihood can be treated as missing data problem, so the 8 DIC alternatives can be used ( Celeux, et al, 2003). a nice review is here, titled Bayesian Model Comparison: Review and Discussion http://www.stat.auckland.ac.nz/~iase/publications/13/Alston-Kuhnert-Low_Choy-McVinish-Mengersen.pdf

reading some winbugs, and found a new function , logfact(), it was released in May of 1999. I have missed it fro ten years, really?! logfact(e) is ln( e !). Verson .603 has bugs found on loggam and logfact functions, oh, really, why? Here, it has all the functions that bugs can specify. http://mathstat.helsinki.fi/openbugs/Manuals/ModelSpecification.html#ContentsAII

response surface issued by Cox and Wilson in 1950, this design is to display the relationship between multiple factors and the outcomes. It try to understand what situations (combination of the factors and their conditions ) can generate the optimal outcome. The first degree polynomial was suggested to construct the response to approximate the surface. The easiest way of accomplishing the first degree polynomial is to use the factorial experimental design, which can determine which factor has the what effect on the outcome. After model selection, the final model including only the only significant factors can be achieved. A further quadratic or cubic model can be fit to those factors to approximate the outcomes. In Rubin's work. the nonlinear response surface was due to the nonlinear relationship of the X, the matching variables, and Y.

Structural equation models with dichotomous variables

Structural equation models with dichotomous variables The dichotomous items are assumed to associated a or a group of latent factors. The estmiation of this kind of model can be find in the line of studies using EM algorithm. Full infromation maximum likelihood estimation plays a important role in the factor analysis when the items are dichotomous (Bock and Aitkin, 1981).