|
|
Dipak Dey Presentations |
|
| Abstract |
Bayesian Meta-analysis for Consensus Mean Estimation
Problem Under Grouped Random Effects Model
Dipak Dey The problem of determining a consensus mean based on the data for two or more laboratories is a common one in NIST and other measurement laboratories. Frequently, results from dissimilar studies are inappropriately combined, resulting in suspect inferential synthesis. Comparative studies investigating luminousity of photo diodes in different laboratories, including NIST are studied and observed that there are few distinct groups which have more variability. A hierarchical Bayesian model is developed with grouped random effects to incorporate the interlaboratory studies. Bayesian approaches for such meta-analysis are preferable because of the small number of studies are prevalent in the data. Diffuse proper prior and hyperprior distributions are specified to assure posterior propriety. Sampling based methods using BUGS are used to generate samples from the relevant posteriors. Posterior summaries and plots of model parameters including consensus mean are analyzed to suggest solutions to questions of interest. |
|
Date created: 8/28/2001 |
|