7 and Y=X1+X2 Problems with generalizability of the models,because they are in general too large. Methods,including the AIC,ignore the sample size,again tending towards models that are too large. Comments on the Bayesian approaches Conclusions independent of the order in which you try out models. Models need not be nested to be compared. Prior knowledge,when available,can be incorporated in the inferential process. Very simple to use,as we will see,there are easy to use R programs that provide extremely useful information in addition to implementing the criteria automatically. Bayes Factors and the BIC lead easily to "model averaging",shown to provide "optimal predictions"if two or more models are plausible. Because models tend not to be too large,these methods tend to lead to models that better generalize to other populations (i.e.,they avoid "over-fitting"the data). Sample size is considered in the calculations (compare AIC to BIC,where it is like AIC has a sample size of 2,regardless of real sample size.).7 and Y = X1 + X2 • Problems with generalizability of the models, because they are in general too large. • Methods, including the AIC, ignore the sample size, again tending towards models that are too large. Comments on the Bayesian approaches • Conclusions independent of the order in which you try out models. • Models need not be nested to be compared. • Prior knowledge, when available, can be incorporated in the inferential process. • Very simple to use, as we will see, there are easy to use R programs that provide extremely useful information in addition to implementing the criteria automatically. • Bayes Factors and the BIC lead easily to “model averaging”, shown to provide “optimal predictions” if two or more models are plausible. • Because models tend not to be too large, these methods tend to lead to models that better generalize to other populations (i.e., they avoid “over-fitting” the data). • Sample size is considered in the calculations (compare AIC to BIC, where it is like AIC has a sample size of 2, regardless of real sample size.)