My question is motivated by Do, Joshi, and Stopler's paper "Can Environmental Policy Reduce Infant Mortality? Evidence from the Ganga Pollution Cases".

A simplified form of their models that describes the question is:

$$y_i =\beta_0 +\beta_1 X_i +\varepsilon_i $$

Which would be estimated (1) using only $Z_{1i}$ as an instrument and (2) using $Z_{1i}$ and $Z_{2i}$ as instruments. The goal is to determine if $Z_{2i}$ is a valid instrument (uncorrelated with $\varepsilon_i$). They "know" $Z_{1i}$ is valid.

Their method is to use a C-statistic, which as far as I can tell is the difference in Sargan-Hansen test statistics between the two models, and would be distributed as $\chi^2_1$ because the number of instruments differs by 1.

My question is, wouldn't the Sargan-Hansen test from (1) just give 0? The equation is exactly identified. Thus, doesn't the C-statistic just reduce to a Sargan-Hansen test of (2)? Is there something I am missing? Are they, perhaps, just trying to show they are aware of complicated techniques to signal intelligence? Or is it more appropriate to call it a C-statistic due to the comparative nature?



Yes, you are correct about the conclusion that $J_{n1}=0$ and hence $C_{n}=J_{n}-J_{n1}=J_{n}$. However, despite the test statistics are the same, the asymptotical distributions and critical values are the same, they are derived from different procedure under different null hypothesis, I prefer to keep using the name C-test instead of Sargan-Hansen test.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.