I am running a Bayesian regression model by WinGUBS via R2WinBUGS package in R. Everything looks fine except for one parameter:
nlssim$summary[37,][c(1,2,8,9)] mean sd Rhat n.eff 3.054326e-05 9.523965e-06 1.000000e+00 1.000000e+00
The number of effective size is 1! It seems to indicate that the autocorrelation between samples are extremely large, but the trace plot looks all fine. And I then try the effectiveSize function in code package and the result is 4590.
I am curious how WinBUGS calculate the n.eff statistic, and if n.eff=1 is symptom of some of my mistakes?
I run 3 chains in parallel, and each chain has 60000 iterations. n.burnin=200, n.thin=30
Thank you very much in advance!