Two well-known (and related) measures of model complexity from statistics are the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC).

When might AIC = BIC?

  • 10
    $\begingroup$ You should try writing down the formulas and setting them equal to each other :) You will get the answer immediately. $\endgroup$ – guy Mar 14 at 13:18

As a reminder:

$$AIC = - 2 \log \mathcal{L}(\hat{\theta}|X)+2k $$

$$BIC = - 2 \log \mathcal{L}(\hat{\theta}|X)+k \ln(n)$$

So for what values of $n$ is $2 = \ln(n)$?

  • 8
    $\begingroup$ (+1) I noticed that for $BIC$ you write $\log$ and $\ln$ in the same expression. Is this distinction necessary? $\endgroup$ – Sycorax Mar 14 at 19:23
  • $\begingroup$ Both logarithms have $e$ as their basis. It is just that log-likelihood (instead of ln-likelihood) is the term we use to describe the natural logarithm of the likelihood. $\endgroup$ – Stats Mar 14 at 19:27
  • $\begingroup$ @Sycorax: I guess it's to signify that for the $\log$ it really doesn't matter (as long as you're consistent) whereas for the $\ln$, well it has to be $e$. $\endgroup$ – Mehrdad Mar 15 at 0:37
  • 1
    $\begingroup$ But $n$ should be an integer, right? $\endgroup$ – innisfree Mar 15 at 5:44
  • $\begingroup$ @innisfree Not always. E.g., for generalized additive models the estimated degrees of freedom are usually not integers. $\endgroup$ – Roland Mar 15 at 7:12

Your Answer

By clicking "Post Your Answer", you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.