In the classical errors-in-variables problem, we are looking at the effect of $x^*_i$ on $y$, but $x^*_i$ is misreported. We have observations $x_i = x^*_i +u_i$, where $u_i$ has zero mean mean, variance $\sigma^2_u$, and is independent from $y_i$ and $x^*_i$. In this case, OLS produce biased and inconsistent estimates of the coefficients.

A common solution is to obtain a second measurement $z_i = x^*_i +v_i$, where $v_i$ has zero mean, variance $\sigma^2_v$ and is independent from $y_i, x^*_i, $ and $u_i$ and then use $z_i$ as an instrumental variable for $x_i$ to obtain consistent estimates of the coefficients.

My question is, what if there is a biased measurement error; for example what if $x_i = \gamma x^*_i +u_i$ where $\gamma>0$ and we still have $z_i = x^*_i +v_i$. Is there an estimation procedure that would give a consistent estimator of the coefficients? What about bias?




Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.