Deviance for Binary Data Notes

# An approximate 95 condence interval for i is i 196se

This preview shows page 1. Sign up to view the full content.

This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: are obtained by software package. An approximate 95% conﬁdence interval for βi is ˆ ˆ βi ± (1.96)SE (βi ). To test H0 : βi = 0 vs Ha : βi = 0, Z= ˆ βi − 0 ˆ SE (βi ) which is approximately a N (0, 1). If zobs is the observed value of Z , p-value=2P (Z > zobs ). SAS reports Z 2 ≈ χ21) ( UNM Saturated model. Model with the the maximum number of parameters that can be estimated. Y1 , Y2 , . . . , YN are independent and Yi ∼ Binomial (ni , πi ), log-likelihood is (except for constant), N [yi log (πi ) − yi log (1 − πi ) + ni log (1 − πi )] l (β ; Y ) = i =1 Saturated model: All πi s are different and β = (π1 , π2 , . . . , πN )T The maximum likelihood estimates are πi = yi /ni (bmax ). ˆ The max. log-likelihood is N [yi log (yi /ni )−yi log (1−(yi /ni ))+ni log (1−(yi /ni ))] l (bmax ; Y ) = i =1 UNM For model with p < N parameters, estimates πi . ˆ ˆ Fitted values, yi = ni πi ˆ The log-likelihood, N ˆ ˆ ˆ [yi log (yi /ni ) − yi log (1 − (yi /ni )) + ni log (1 − (yi /ni ))] l (b ; Y ) = i =1 The deviance (textbook) D = 2[l (bmax...
View Full Document

## This note was uploaded on 01/27/2014 for the course STAT 574 taught by Professor Gabrielhuerta during the Fall '13 term at New Mexico.

Ask a homework question - tutors are online