Effects of additive covariate error on parameters and covariates of a linear regression model

photophotophoto
Day & Time 1st July 2016, 15:00 ~ 16:00
Venue Room C816, Building of the Faculty of Science Graduate school of Science
Lecturer Prof. Eiji NAKASHIMA (Radiation Effects Research Foundation, Hiroshima University)
Presentation Title Effects of additive covariate error on parameters and covariates
of a linear regression model
Abstract We show that, under independent additive covariate error, the condition that
the log of the true covariate density is up to quadratic polynomial of the
dose on an interval such as standardized normal, exponential, and uniform
distributions on a dose interval, is the distributional condition for the
regression calibration to be linear. Violation of this condition indicates
that the linear dose response in terms of true dose becomes non-linear in
terms of observed dose due to additive dose error. A practical and possibly
important application of this result in radiation epidemiology is shown. A
generalized additive covariate error model using the transform-both-sides
that includes additive and multiplicative covariate error models as special
cases is proposed and the distributional condition for the linear regression
calibaration is considered.