I fitted a mixture denstiy of two gaussians two my data. I now want to
calculated the standard errors of the estimates via the boot.se command of
the mixtools package. My question is now, if the output is correct? It
seems a bit odd to me, so is this correct what I am doing and can I rely on
the values?
My data: http://s000.tinyupload.com/?file_id=09285782882980618119
My code:
normalmix<-normalmixEM(dat,k=2,lambda=c(0.99024,(1-0.99024)),fast=FALSE,maxit=10000,epsilon
= 1e-16,maxrestarts=1000)
normalmix$loglik
normalmix$lambda
se<-boot.se(normalmix,B=1000)
se$lambda.se
se$mu.se
se$sigma.se
final results:
lambdahat = 0.990238663
mu1hat= -0.00115
mu2hat= 0.040176949
sigma1hat= 0.012220305
sigma2hat= 0.003247102
My problem now is - and thats why I feel uncomfortable about relying on the
values - that the output of boot.se(normalmix) varies quite strong. So
without changing the code and rerun it (with the same normalmix, so
normalmix is not rerun again) I get different estimates of the standard
error. I increased the default value for B from 100 to 1000. In the manual
there is nothing said about any other randomness. So where does it come
from? What should I do now?
[[alternative HTML version deleted]]
