I am interested in quantifying measurement error jointly for a set of exposures and outcomes. The exposure is binary and the outcome can be binary or estimated as a rate. I can derive a set of expected associations (taken from well conducted cohort studies) and a set of observed associations (taken from routinely collected data where some mismeasurement is likely). So, I end up with two matrices.
For a single association, it is not possible to tell whether attenuation in the effect estimate (observed/expected) is a result of measurement error in the exposure, in the outcome, or a bit of both. However, I wonder whether approaches exist that can leverage all of the information in the matrices to flag which of the set of exposures and outcomes are not well measured?
This preprint describes something close to what I envisage but with continuous variables and implemented in R. Any pointers to relevant papers or statistical packages would be appreciated. I can work in R, but would rather work in STATA.
Thanks,
Tom
For a single association, it is not possible to tell whether attenuation in the effect estimate (observed/expected) is a result of measurement error in the exposure, in the outcome, or a bit of both. However, I wonder whether approaches exist that can leverage all of the information in the matrices to flag which of the set of exposures and outcomes are not well measured?
This preprint describes something close to what I envisage but with continuous variables and implemented in R. Any pointers to relevant papers or statistical packages would be appreciated. I can work in R, but would rather work in STATA.
Thanks,
Tom
