This function provides an efficient algorithm to calculate the tuning parameters involved in ELCIC under GEE.
Arguments
- x
A matrix containing covariates. The first column should be all ones corresponding to the intercept.
- y
A vector containing outcomes.
- id
A vector indicating subject id.
- beta
A plug-in estimator solved by an external estimation procedure, such as GEE.
- r
A vector indicating the observation of outcomes: 1 for observed records, and 0 for unobserved records. The default setup is that all data are observed. See more in details section.
- dist
A specified distribution. It can be "gaussian", "poisson",and "binomial".
- rho
A correlation coefficients obtained from an external estimation procedure, such as GEE.
- phi
An over-dispersion parameter obtained from an external estimation procedure, such as GEE.
- corstr
A candidate correlation structure. It can be "independence","exchangeable", and "ar1".
Details
If the element in argument "r" equals zero, the corresponding rows of "x" and "y" should be all zeros.
Examples
## tests
# load data
data(geesimdata)
x<-geesimdata$x
y<-geesimdata$y
id<-geesimdata$id
corstr<-"exchangeable"
dist<-"poisson"
# obtain the estimates
library(geepack)
fit<-geeglm(y~x-1,data=geesimdata,family =dist,id=id,corstr = corstr)
beta<-fit$coefficients
rho<-unlist(summary(fit)$corr[1])
phi<-unlist(summary(fit)$dispersion[1])
r=rep(1,nrow(x))
lambda<-lambda.find.gee(x,y,id,beta,r,dist,rho,phi,corstr)
lambda
#> [,1]
#> [1,] -2.602981e-03
#> [2,] 1.321320e-03
#> [3,] -1.115227e-05
#> [4,] -1.213901e-03
#> [5,] 4.672891e-03
#> [6,] -4.575420e-03