Estimating the reliability of an exposure variable in the presence of confounders.