0
$\begingroup$

I have some normally distributed data x of length N. I want to test 2 hypotheses: that $\mu_1$=0 or that $\mu_2>0$. I can calculate the likelihood for all N data points for each hypothesis (for model 2 I use numerical integration to get it). Now I have N likelihoods under each model, how do I condense this to one Bayes Factor? I could avg it or count the number of times each model has a higher factor?

$\endgroup$
1
$\begingroup$

If you look at the definition of a Bayes factor (as, e.g., in my book), $\mathfrak{B}_{01}$ it is expressed as the ratio of two marginal likelihoods $$ m_0(x_1,\ldots,x_N)=\int_{\Theta_0} \ell_0(\theta_0|x_1,\ldots,x_N)\pi_0(\theta_0)\text{d}\theta_0 $$ and $$ m_1(x_1,\ldots,x_N)=\int_{\Theta_1} \ell_1(\theta_1|x_1,\ldots,x_N)\pi_1(\theta_1)\text{d}\theta_1 $$ [with generic notations]. Hence you have to integrate one likelihood per model for the entire dataset or sample. In each model you thus end up with a single numerical value.

$\endgroup$

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.