one thing that i've learned is that you have to have a separate validation
data set to estimate the variance of measurement error.
SIMEX -- simulation extrapolation is very cool for measurement error models.
it's sort of like the bootstrap method for this kind of models.
