Is Gibbs Sampling More Efficient Than Metropolis-Hastings?

[natski]

Hi all,

I have used Metropolis-Hastings rules in a Markov Chain Monte Carlo (MCMC) algorithm for a few months now. I feel quite confident with this method and I use it frequently for fitting various model parameters to observational data. Already I feel that computational time is really hindering the MCMC method and I require something more efficient.

I have heard that Gibbs sampling is a more efficient method than Metropolis-Hastings, can anyone confirm this first of all? Second, most of the literature on Gibbs sampling I have Googled is quite confusing to me and I would really appreciate it if anyone knows of a very good and simple guide (i.e. for "dummies") on how to make the upgrade from Metropolis-Hastings to the more advanced Gibbs sampling.

Thanks in advance,

Natski

 

[Evalesco]

Hi Natski,

Gibbs sampling is indeed more efficient than the Metropolis-Hastings algorithm. It is a Markov Chain Monte Carlo method of sampling from a multivariate probability distribution. You can find a good and simple guide to Gibbs sampling on the website of John D. Cook, where he explains the basics of Gibbs sampling for beginners. He also provides some examples which can help you understand how to use Gibbs sampling in practice. I hope this helps. Best of luck!