Dear All,
There is evidence in the literature to the effect that if you fit multiple
regression models with heteroscedastic errors (assumed to be Normally
distributed) AND some of the observations are right-censored, you get bias
in the regression-coefficient estimates. Some simulations I have done
indicate that if you attempt to remove the bias by modelling the variance as
a function of (some of) the covariates, you still appear to get bias in the
coefficients of both the mean and variance parts of the model.
I was wondering if anyone can point me to a literature, or personal
knowledge or work in progress, that indicates how such bias may be overcome.
Thank you
Patrick Royston.
____________________________________
Professor Patrick Royston
Dept of Medical Statistics and Evaluation
Imperial College School of Medicine
Hammersmith Hospital, Ducane Road,
London W12 0NN, UK.
email: [log in to unmask]
Fax: +44 208 383 8573
____________________________________
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|