Hello All
I am using R to find the impact of ignoring clustering, when indicated by these
tests, on bias, variance and inference of estimates of interest, including
population totals and regression coefficients. But when U have used the apVar
function using the command
test.lme1$apVar
to find the var-cov matrix I got the follwing error
"Non-positive definite approximate variance-covariance"
I think the problem was with my data which contain the following two columns
m <- 250
g <- as.factor(rep(1:m,each=10))
y <- rnorm(10*m)
test <- data.frame(cbind(g,y))
the lme model used was
test.lme1 <- lme (y~1, data = test, random = ~1| g)
This error affects the t-statistic value, which was NA for more than 240 value
of the 250 values which I expect to have. Some times all t-stat values are NA
and sometimes I don't have the NA result at all.
I'll send you all of the simulation program which I've used.
I hope you can help to solve this problem
Thanks
LOAI
PhD student
University of Wollongong
Australia