Dear Wendy
Like you, I am also interested to do Bayesian analysis of multi stage survey
data without using model to incorporate effect of different stages. I was
expecting good suggestion in this regard. Since I could not get any response
from experts, I am trying to extend discussion on this interesting area,
although I have little knowledge in this area.
You can proceed in following way
(1) Draw a random sample by using weight given to you.
(2) Use any of supper population model
a. the normal-normal with known variances, with an inverse gamma for
unknown variances,
b. (ii) the Dirichlet (prior)-multinomial (likelihood) to represent extreme
vagueness about the super-population.
(3) You can use bootstrap method for analyzing survey data through
Bayesian approach (see http://3w.eco.uniroma1.it/OB07/papers/aitkin.pdf )
You may use my e-mail id ([log in to unmask] or [log in to unmask])
for future discussion.
With regards
Nand Kishore
|