In order to do a LRT between two mixture models with different numbers of components, I need to know the number of parameters. I would like to know the answer both for: a) Gaussian mixture model b) Betabinomial mixture model

I read that for mixtures of Gaussians the number of parameters is 3k-1 (k being the number of components). Where does this number come from?


  • $\begingroup$ k means, k variances, and k weights that sum up to one, hence of effective dimension k-1 $\endgroup$ – Xi'an Mar 26 '18 at 11:19

If there are no restrictions on the parameters of the components of a finite mixture model, then the only restriction is that the mixture weights should sum to one. Hence the total number of independent parameters in a $k$-mixture is:

$k*$(num of parameters of each component) $+ k-1$.

Since the Gaussian has 2 parameters the answer to a) is $3k-1$.

The BetaBinomial distribution is not a finite mixture model and has 2 parameters (if the number of trials $n$ is known)

| cite | improve this answer | |

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.