dear BUGSers,
I am interested in estimating the expected values for the
variances of a given population. I would use the
results of some experiments as observed values for the variances. I tried the
following model:
S[i] ~ Gamma(alpha[i], beta[i])
alpha[i] ~ Exp(1)
beta[i] ~ Gamma(0.1, 1)
mu[i] <- alpha[i] / beta[i]
where the prior distributions for alpha and beta are as in
Congdon (2001).
The values for mu[i] are not consistent and both the scale and
the variability appear to be very high.
A possible different model I tried was by modelling S[i] as a
lognormal(mu[i], tau), but I can't figure out which prior distribution I can use
for mu[i] (NB mu[i] is my primary interest, as I would like to use these values
as prior for the variances in an other model, where I am trying to estimate
means).
Does anybody have a hint about that?
Thanks in advance,
Marta Blangiardo
PhD student in Applied Statistics
University of Florence (Italy)