0
$\begingroup$

This is a well-written blog on how we can fit a mixture of beta distributions to a dataset:

http://varianceexplained.org/r/mixture-models-baseball/

However, it would have been excellent to identify the optimal number of beta distributions one can fit a beta mixture model. I think one could figure this out using AIC/BIC but I am not sure how to do this or if there is already a package (preferably in R) that could do this. Any insights will be much appreciated. Thanks.

Improve this question
$\endgroup$
3
  • $\begingroup$ Thanks, Xi'an. I would appreciate how AIC/BIC applies to this particular problem. $\endgroup$
    – Kasthuri
    Commented Nov 22, 2021 at 20:42
  • $\begingroup$ There are constraints on AIC/BIC, for example, $N>100$ or $300$, that are severe enough that I wouldn't trust them for something as volatile as mixture models of beta distributions without checking first if all the conditions assumed actually apply to the data available. $\endgroup$
    – Carl
    Commented Nov 23, 2021 at 1:29
  • 1
    $\begingroup$ BIC is a constistent criterion for estimating the number of component, AIC is not. $\endgroup$
    – Xi'an
    Commented Nov 23, 2021 at 8:33

0

Reset to default