{[ promptMessage ]}

Bookmark it

{[ promptMessage ]}

# 26REML - REML Estimation of Variance Components Copyright c...

This preview shows pages 1–9. Sign up to view the full content.

REML Estimation of Variance Components Copyright c 2012 (Iowa State University) Statistics 511 1 / 31

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Consider the General Linear Model y = X β + , where N ( 0 , Σ ) and Σ is an n × n positive definite variance matrix that depends on unknown parameters that are organized in a vector γ . Copyright c 2012 (Iowa State University) Statistics 511 2 / 31
In the previous set of slides, we considered maximum likelihood (ML) estimation of the parameter vectors β and γ . We saw by example that the MLE of the variance component vector γ can be biased. Copyright c 2012 (Iowa State University) Statistics 511 3 / 31

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Example of MLE Bias For the case of = σ 2 I , where γ = σ 2 , the MLE of σ 2 is ( y - X ˆ β ) 0 ( y - X ˆ β ) n with expectation n - r n σ 2 . Copyright c 2012 (Iowa State University) Statistics 511 4 / 31
This is MLE for σ 2 is often criticized for “failing to account for the loss of degrees of freedom needed to estimate β .” E " ( y - X ˆ β ) 0 ( y - X ˆ β ) n # = n - r n σ 2 < σ 2 = E ( y - X β ) 0 ( y - X β ) n . Copyright c 2012 (Iowa State University) Statistics 511 5 / 31

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
A Familiar Special Case y 1 , . . . , y n i . i . d . N ( μ, σ 2 ) E n i = 1 ( y i - μ ) 2 n = σ 2 but E n i = 1 ( y i - ¯ y ) 2 n = n - 1 n σ 2 < σ 2 . Copyright c 2012 (Iowa State University) Statistics 511 6 / 31
REML is an approach that produces unbiased estimators for these special cases and produces less biased estimates than ML in general. Depending on whom you ask, REML stands for REsidual Maximun Likelihood or REstricted Maximum Likelihood. Copyright c 2012 (Iowa State University) Statistics 511 7 / 31

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
The REML Method 1 Find n - rank ( X ) = n - r linearly independent vectors a 1 , . . . , a n - r such that a 0 i X = 0 0 for all i = 1 , . . . , n - r . 2 Find the maximum likelihood estimate of γ using w 1 a 0 1 y , . . . , w n - r a 0 n - r y as data.
This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}