0
$\begingroup$
  • Can Metropolis sampling be used in conjunction with Gibbs sampling? So for example, if I have three parameters of interest, but only two of them have full conditionals that are known distributions, can I sample using Gibbs for those two parameters and Metropolis using the other parameter for each iteration?
  • If the parameter of interest only takes positive values, would it be wrong (in terms of producing an answer, not efficiency) to use a normal proposal distribution centered on the previous iteration's value of the parameter?
$\endgroup$
1
$\begingroup$
  • Yes. In fact, you can consider a Gibbs sampler to be a special case Metropolis Hasting sampler.

  • If the parameter of interest is strictly positive, using a normal proposal will still work (any negative proposals will be automatically rejected). However, there is a very good chance that using the normal distribution for the proposed log of the parameter will actually be more efficient, as the posterior of the log-parameter is often approximately normal.

$\endgroup$

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.