[ADMB Users] problems running mcmc simulations

Felice Griffiths fmg1 at sfu.ca
Mon Jul 19 16:53:44 PDT 2010


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



More information about the Users mailing list