« Multivariate multilevel analysis | Main | The great chain of being »
March 21, 2008
Multilevel priors with no hierarchical data?
Yan Jiao writes,
I have a question about multilevel models and expect you to offer me some help.
Have you ever seen examples with multilevel priors but no hierarchical data? For example in the book of Gelman et al 2004, chapter 5, the rat tumor example. If I lost the data or the previous 70 experiments are not designed in the same way, so the original data from the 70 experiments are not available or comparable with the 71th experiment, can I still use multilevel priors from the 70 experiment as the multilevel prior for the 71th experiment. I think we can, but I can't find an example to support. Any help from you will be greatly appreciated.
First, I gotta say, I'm getting a reputation as a soft touch if people expect free help. Anyway . . . yes, you can do the above; it just puts more burden on your prior distribution, which has to represent this multilevel uncertainty. We did this in our toxicology example; see my 1996 paper with Bois and Jiang, and Sections 9.1 and 20.3 of Bayesian Data Analysis.
Posted by Andrew at March 21, 2008 8:58 PM
Trackback Pings
TrackBack URL for this entry:
http://www.stat.columbia.edu/~cook/movabletype/mt-tb.cgi/1502
Comments
Given the name of the writer, I'm guessing English might not be their first language. In which case you could probably translate the first sentence as "I'm sure your opinion would be helpful" or something like that.
Posted by: Daniel Lakeland at March 21, 2008 10:26 PM.
Probably, it was a mistake by a non fluent english writer. Long time ago, I wrote "I expect you ..."; but I wanted to say "I hope you ...".
Posted by: Martin at March 22, 2008 1:57 PM.
I read it first as "I expect you could offer some help" which is country for a polite request which also compliments the expertise of the other person.
There may be a similar colloquialism in his or her language.
Posted by: ZBicyclist at March 22, 2008 6:35 PM.