This preview shows pages 1–5. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: STAT5044: Regression and Anova Inyoung Kim 1 / 13 Outline 1 Inference for GLMs 2 / 13 Deviance and Goodness of Fit The saturated GLM has a separate parameter for each observation. It gives a perfect fit. This sound good, but it is not helpful model It does not smooth the data or have the advantage that a simpler model has. Nonetheless, it serves as a baseline for other models, such as for checking model fit. 3 / 13 Deviance and Goodness of Fit A saturated model explains all variation by the systematic component of the model Let denote the estimate of for the saturated model, corresponding to estimated means i = y i for all i . For a particular unsaturated model, denote the corresponding ML estimates by and i For maximized log likelihood L ( ; y ) for that model and maximized log likelihood L ( y ; y ) in the saturated case, 2 [ L ( ; y ) L ( y ; y )] describes lack of fit. It is the likelihoodratio statistic for testing the null hypothesis that the model holds against the alternative that a more general model holds 2 [ L ( ) L ( y ; y )] = 2 i [ y i i b ( i )] / a ( ) 2 i [ y i i b ( i )] / a ( ) Usually, a ( ) has the form a ( ) = / i , and this statistic equals 2 i i [ y i ( i i ) b ( i )+ b ( i )] / = D ( y ; ) / This is called the scaled deviance and...
View
Full
Document
 Fall '11
 Staff

Click to edit the document details