Have an idea?

Visit Sawtooth Software Feedback to share your ideas on how we can improve our products.

How many chains does the MCMC algorithm use in HB estimation?

asked Jul 26, 2019 by anonymous

1 Answer

0 votes
I'm not sure I understand the question.

If you are asking which MCMC algorithms we support, we currently are using Metropolis Hastings but are developing support for other methods.

There isn't a 'chain count' per se.  It's an algorithm and not an object of the estimation.  In a sense each respondent has their own 'chain' of betas throughout the estimation.

Can you clarify?
answered Jul 30, 2019 by Walter Williams Gold Sawtooth Software, Inc. (18,880 points)
If you run multiple chains, after a while they will mix in high probability regions of the distribution. If you are just running one chain, and the chain starts in a region of low density, how can you be sure convergence is achieved? Since the 1990s much has been done to advance monitoring parallel simulations of multiple chains to assess convergence (see Gelman and Rubin's convergence diagnostic for example). See Fig 1 here http://www.stat.columbia.edu/~gelman/research/published/med4.pdf