I am familiar with the calculation of likelihood/evidence ratios (ER) using Akaike's Information Criterion (AIC) and its corrected version (AICc):

ER = $\frac{1}{e^{-0.5*ΔAIC}}$

Is it possible to use a relative likelihood/evidence ratio for information criteria other than AIC?

I would like to compare the relative likelihood of one model against another, but with an information criterion derived from leave-one-out cross validation (LOOIC), rather than AIC. Can I just exchange ΔLOOIC for ΔAIC in the above equation?

If so, is there a citation to which I could reference this manipulation?

  • $\begingroup$ There's an interpretation of something like this for BIC. $\endgroup$
    – Glen_b
    Commented Oct 28, 2016 at 2:15


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.