[ADMB Users] problems running mcmc simulations

[Felice Griffiths]

Hello,
I am a relatively new ADMB user and I am having a problem running mcmc simulations. I have a population dynamics model with multiple log likelihood components (all are log normally distributed). I can fit parameter estimates, estimate the hessian and get reasonable standard errors, however when I invoke the -mcmc command it runs with an acceptance rate of 0. The objective function contains likelihoods as well as priors (beta distributions on survival and fecundity rates). I discovered that if I multiply the full objective function (likelihood components and prior fits) by a constant of 15 or more, the acceptance rate appears reasonable (around .3). In addition, the larger this constant is, the tighter the posterior distributions of the estimated parameters become (i.e. if I use a weight of 50 instead of 15). Applying the constant only to the likelihood portion of the objective function gives similar results. Any insight as to why I would get an acceptance rate of 0 and
why multiplying the objective function by a constant would appear to solve the problem?

Thanks!

Felice Griffiths, BSc

Masters Student
Resource and Environmental Management
Simon Fraser University
_______________________________________________
Users mailing list
Us...@admb-project.org
http://lists.admb-project.org/mailman/listinfo/users

[dave fournier]

Are you multiplying the likelihood or the log-likelihood by 15?
Multiplying the log-likelihood by 15 sharpens the modes (figuratively
speaking) and is not equivalent to the original problem.

Otherwise without looking at your code and data it is hard to say
much about what is going on.

[Felice Griffiths]

I was multiplying the log-likelihood by 15, so that explains why I was seeing differences in my results with different weights. Turns out the real problem was that I didn't initialize the penalty vectors I have in the model. Putting the weight on the log-likelihood was seemingly solving the problem by getting the mcmc simulations to run, but was not correcting the real issue. Thanks for the insight!