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 DocumentThis 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 DocumentThis 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: PUBH 7430 Lecture 14 J. Wolfson Division of Biostatistics University of Minnesota School of Public Health October 20, 2011 Fitting GEE models: Examples FEV data Mean model: E ( FEV ij  initheight i , age ij ) = β + β 1 initheight i + β 2 age ij • Focus on β 1 (influence of initial height, adjusted for age) • Fit Gamma family with identity link using: 1 “Naive” GLM 2 “Naive” GLM w/ sandwich variance estimates 3 GEE w/ independence working correlation matrix 4 GEE w/ other working correlations Fitting GEE models in R/SAS • Assume that data are in long format • Variable subjID identifies observations taken from the same cluster (girl) subjID init.height age fev 1 1.20 9.34 1.24 1 1.20 10.39 1.45 1 1.20 11.45 1.63 1 1.20 12.46 2.12 1 1.20 13.42 2.30 1 1.20 15.47 2.44 1 1.20 16.37 2.39 2 1.13 6.59 1.36 2 1.13 7.65 1.42 2 1.13 12.74 2.13 Fitting GEE models in R Use geeglm() function from geepack library: library(geepack) geeglm(fev~initheight+age, family=Gamma(link=’’identity’)’, id=subjID, corstr=’’) where can be independence , exchangeable , ar1 , unstructured , etc. Fitting GEE models in SAS Add REPEATED statement to PROC GENMOD : PROC GENMOD DATA=FEVDAT; MODEL FEV = INITHEIGHT AGE / DIST=GAMMA LINK=IDENTITY; REPEATED SUBJECT=SUBJID / TYPE=  where can be IND , EXCH , AR(1) , UN , etc. FEV example: “Naive” GLM In R glm(fev~initheight+age, family=Gamma(link=’’identity’’)) In SAS PROC GENMOD DATA=FEVDAT; MODEL FEV = INITHEIGHT AGE / DIST=GAMMA LINK=IDENTITY; FEV example: “Naive” GLM Variable Coef 95% Conf Int Intercept1.9 (2.11, 1.69) Initial Height (dm) 0.148 (0.13, 0.165) Age (years) 0.191 (0.186, 0.195) “Adjusting for age, two girls differing by 1 dm (10 cm) in initial height are expected to differ by 0.148 L (95% CI: 0.13 to 0.165 L) in FEV.” FEV example: “Naive” GLM For this analysis... • Estimate of initial height effect (0.148 L): Likely close to correct ( coefficient estimates are consistent ) • Confidence interval (width = 0.035 L): Likely incorrect ( sensitive to variance function and independence assumptions ) FEV: Naive GLM w/ sandwich variances • Coding “trick”: Use GEE code, but tell software that each observation is its own cluster (and use independence working correlation). • Define variable obsID : obsID subjID init.height age fev 1 1 1.20 9.34 1.24 2 1 1.20 10.39 1.45 3 1 1.20 11.45 1.63 4 1 1.20 12.46 2.12 5 1 1.20 13.42 2.30 6 1 1.20 15.47 2.44 7 1 1.20 16.37 2.39 8 2 1.13 6.59 1.36 9 2 1.13 7.65 1.42 10 2 1.13 12.74 2.13 FEV: Naive GLM w/ sandwich variances In R library(geepack) geeglm(fev~initheight+age,...
View
Full Document
 Fall '04
 Prof.Eberly
 Variance, Orders of magnitude, Correlation and dependence

Click to edit the document details