Abstract:
|
In broad strokes, Bayesian adjustment for measurement error in epidemiological contexts can be straightforward. With exposure measurement error, for instance, one can hierarchically bring together a model for the outcome given the error-free (but latent) exposure, a model for this exposure itself, and a model for the measurement error. Turning the Bayesian crank then does the rest, yielding inferences about parameters in the model for the outcome given the error-free exposure. Often, however, some subtlety lies in the question of how much must be known or assumed for this adjusted inference to be effective. One issue surrounds assumptions about structure, such as whether the measurement error can be assumed to be nondifferential. Another issue surrounds the strength of a priori knowledge about the magnitude of measurement error, or alternately the availability of validation data with which to infer this magnitude. In this talk then, we take a detailed look at the circumstances under which adjustment for measurement error is worthwhile.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.