This preview has intentionally blurred sections. Sign up to view the full version.View Full Document
Unformatted text preview: Random Effects Model Notes Andrew Womack November 21, 2011 1 Basic Model We have the following hierarchical model y ij | μ j ,μ,σ 2 ,τ 2 ∼ N ( μ j ,σ 2 ) μ j | μ,σ 2 ,τ 2 ∼ N ( μ,τ 2 ) where j = 1 ,...,J is our group index and i = 1 ,...,n j are the individuals in group j . This can also be viewed as putting a positive correlation between indi- viduals in the same group y · j | μ,σ 2 ,τ 2 ∼ N n j ( μ 1 n j , Σ j ) where Σ j = σ 2 I n j + τ 2 B n j where I n j is the identity matrix and B n j is a matrix of ones. This gives a variance which is σ 2 + τ 2 for a unit and correlation ρ = τ 2 τ 2 + σ 2 for units in the same group. Units in different groups have a different zero autocorrelation. Before discussing priors for μ,τ 2 ,σ 2 (or μ,ρ,σ 2 ), we should consider the likelihood and the hierarchical structure for possible conjugate prior struc- tures. First, consider the likelihood itself. It is f ( y | μ,σ 2 ,ρ ) = J Y j =1 " 1- ρ 2 πσ 2 n j 2 (1 + n j ρ )- 1 2 exp- 1- ρ 2 σ 2 ( y · j- μ 1 n j ) I n j- ρ ( n j- 1) ρ- 1 B n j ( y · j- μ 1 n j ) 1 From this likelihood, it is clear that prior for μ and σ 2 are fairly easy to consider, for example it is easy enough to use a normal prior for μ and a gamma prior for σ 2 . What remains to be considered is the prior for....
View Full Document
This note was uploaded on 01/29/2012 for the course STAT 6866 taught by Professor Womack during the Fall '11 term at University of Florida.
- Fall '11