AIC vs. BIC revisited and other updates

April 22, 2009

The winner: BIC. See Raftery (1995;1996) and Kadane and Lazar (2004). There are two reasons that have convinced me that the BIC is the better choice. First, As N increases to infinity the probability of choosing the true model is 1 with the BIC. Second, BIC has a Bayesian justification whereas the AIC does not. Finally, BIC favors parsimony. Clearly I am moving away from my ecological roots of AIC and the influence of Burnham and Anderson.

This summer I intend to update this blog more frequently. These updates will include Bayesian analyses for typical analyses performed in the social sciences (I plan to spend a great deal of time on Bayesian Hierarchical Modeling); Win/OpenBUGS code (i.e. MCMC simulations); and R code. My intention is to post weekly with an example of a Bayesian analysis of a typical problem, compare it with a frequentist analysis, and provide the code. I will try to keep it objective and allow the procedures/results to speak for themselves. Apparently there are at least a few folks that read this, so if you’re interested in seeing a Bayesian approach to a problem common in the social sciences please comment to let me know and I will try my best to address it.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: