|
Abstract:
|
Measurement error models assume a response Y, latent predictor X, and measured predictor W = X + U with U the measurement error. The problem is to infer E(Y|X) from data (Wi, Yi) and an estimate of the measurement error variance, possible calculated from replicate measurements Wij. Implicit is the assumption that Y depends on only the subject-specific mean X=E(W). Replicate measurements arise in longitudinal studies, where variation around the mean is intrinsic, not instrumental, and it is reasonable to question whether Y might also depend on intrinsic variation. Such models have been studied since 1995, with only Lyles et al. (1997) addressing the measurement error aspect of the problem. This talk will highlight some problems unique to modeling with subject-specific variances and summarize approaches using standard measurement error modeling approaches.
|
