Hi All
I can specify whatever inits, it has no effect on the estimation. I am
replicating a textbook example. The result is completely trash, having
estimates
of -58.7 (sd=59.3), where it should be closer to an ml estimate of 0.585
(SE=0.063).
The two chains within one run are different, but with different inits for
different runs, I get exactly the same chains, and I mean exactly.
If I set strong (high-precision) priors using the ML estimates, I get the
result
I want. I dont want to set strong priors, but I want to set reasonable inits;
ones that are not taken by sampling from the prior, but ones that I! specify.
Thanks Toby
data <- list("n", "n3", "y", "person", "occ")
model <- function() {
...
}
write.model(model, "example.txt")
inits <- list(list(m1=0.585, m2=0.718, m3=0.672, m4=0.639),list(m1=0.585,
m2=0.718, m3=0.672, m4=0.639))
parameters <- c("m1", "m2", "m3", "m4", "sig21", "sig22")
bugs1 <- bugs(data, inits, parameters, "example.txt", n.chains=2, n.iter=15000,
clearWD=TRUE, debug=1, DIC=0)
display(log)
check(J:/project/ps/code/example.txt)
model is syntactically correct
data(J:/project/ps/code/data.txt)
data loaded
compile(2)
model compiled
inits(1,J:/project/ps/code/inits1.txt)
this chain contains uninitialized variables
inits(2,J:/project/ps/code/inits2.txt)
this chain contains uninitialized variables
gen.inits()
initial values generated, model initialized
thin.updater(15)
update(500)
set(m1)
set(m2)
set(m3)
set(m4)
set(sig21)
set(sig22)
update(500)
coda(*,J:/project/ps/code/coda)
stats(*)
Node statistics
node mean sd MC error 2.5% median 97.5% start
sample
m1 -58.66 59.27 9.001 -120.8 0.4013 0.7102 501 1000
m2 -58.53 59.27 9.001 -120.7 0.5417 0.8314 501 1000
m3 -58.58 59.27 9.0 -120.7 0.4783 0.7871 501 1000
m4 -58.6 59.27 9.0 -120.7 0.3555 0.7695 501 1000
sig21 0.07409 0.01034 3.039E-4 0.05636 0.07319 0.09558 501
1000
sig22 7206.0 7500.0 1097.0 0.07909 8436.0 19810.0 501 1000
history(*,J:/project/ps/code/history.odc)
History
save(J:/project/ps/code/log.odc)
save(J:/project/ps/code/log.txt)
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
