AIC, BIC, Mallow's Cp, etc

crossvalidation
modelselection

#1

Hi there!!

I have a set of general questions which have been bugging me for a while.

  1. Amongst the three most commonly used penalties for linear regression(AIC, BIC and Mallow’s cp), which one is relatively more trustworthy? I understand that they are all relative measures and have their own caveats.

  2. What should I do when the bic and cp of my best subsets regression disagree?

  3. How does cross validation tie into this particular process?

Thank you for taking time to answer!!