anestistouloumis / simcormultres Goto Github PK

I've noticed that the correlation estimate from fitting a GEE model (package gee) is consistently smaller than the true correlation parameter I enter in the rbin statement. In the example below, my true rho = 0.5 (exchangeable correlation structure) but the estimated rho in GEE is 0.2999. I've conducted a small simulation (n = 1000) and the estimated rho is consistently smaller. Fixed effects seem asymptotically unbiased. I would like to obtain unbiased correlation estimates as well as unbiased fixed effects. Please advise.

library(SimCorMultRes)
library(gee)

N <- 100
x <- rbinom(N, 1, prob = 0.5)
xvec <- rep(x, each = 3)

intercepts <- -1
betas <- 0.5

clsize <- 3
rho <- 0.5

cor.matrix <- matrix(ncol = 3, nrow = 3,
c(1, rho, rho,
rho, 1, rho,
rho, rho, 1))

CorBinRes <- rbin(clsize = clsize, intercepts = intercepts, betas = betas,
xformula = ~xvec, cor.matrix = cor.matrix, link = "logit")

binGEEmod <- gee(y ~ xvec, family = binomial("logit"), id = id,
corstr = "exchangeable",
data = CorBinRes$simdata)
summary(binGEEmod)

summary(binGEEmod)

GEE: GENERALIZED LINEAR MODELS FOR DEPENDENT DATA
gee S-function, version 4.13 modified 98/01/27 (1998)

Model:
Link: Logit
Variance to Mean Relation: Binomial
Correlation Structure: Exchangeable

Call:
gee(formula = y ~ xvec, id = id, data = CorBinRes$simdata, family = binomial("logit"),
corstr = "exchangeable")

Summary of Residuals:
Min 1Q Median 3Q Max
-0.4184397 -0.4184397 -0.3270440 0.5815603 0.6729560

Coefficients:
Estimate Naive S.E. Naive z Robust S.E. Robust z
(Intercept) -0.7215851 0.2145247 -3.363646 0.2303582 -3.132448
xvec 0.3924033 0.3048868 1.287046 0.3020483 1.299141

Estimated Scale Parameter: 1.006711
Number of Iterations: 1

Working Correlation
[,1] [,2] [,3]
[1,] 1.000000 0.299852 0.299852
[2,] 0.299852 1.000000 0.299852
[3,] 0.299852 0.299852 1.000000

Great package, thank you, may you share links to similar software to compare performance of your package with other packages
Even if another software is written in another computer language
Thanks a lot in advance

Anestis
great code thank
seems only you managed to have such a package for several last years
only
can you share example how to generate for example for 5 variables
when each variable has 3 or 7 values
and some how correlation ( dependency, similarity) between each value pairs are given
for example for 2 values for 3 variables
variable 1 : green 30%, red 70%
variable 2 : big 40%, small 60%
variable 3 : table 10%, car 90%

Corrections are:
green big - 0.3
green small - 0.1
green table - 0.2
green car - 0.8

red big - 0.7
red small - 0.6
red table - 0.5
red car - 0.2

big table - 0.3
big car - 0.2

small table - 0.3
small car - 0.2

it it correctly set problem for your package?

Thank you very much in advance