@1:58 how the units work out? Beta is cov/sigma, here we divide covariances, what happens to sigmas that should be present in the concept of OLS beta? And why we don't divide beta's? Beta(Independent Variable, IV) / Beta(Dependent Variable, IV) ? This way at least sigmas are counted for, and intuitively makes sense too. ???
I have a question. You test the strength of your IV in the 1st stage using t-test to make sure it is a good IV. But should you also regress wages(Y) on the IV to be sure it does not DIRECTLY predict Y in the second stage? I see papers where they do both tests. What are your thoughts? Many thanks!
I have a question, if MPi is determined by latent variables Xi, can we use IV to estimate beta? Also, as MPi is binary, can we use probit/logit in the first stage?
Hi, thanks for your questions. Yes - we can use IV to estimate beta if Xi is determined by a latent variable. However, we can't use that latent variable as a IV since it will likely suffer from the same endogeneity as the variable MP. If is binary we should still use OLS, this is because using probit/logit is very susceptible to mispecifications. You are better off using LS therefore. Hope that helps, Ben
Very clear, as the other videos you posted. Thank you very much for your help and for sharing your knowledge.
To tell the truth Ben...you were my honors econometrics teacher! :)
Thank you so much for all your videos. Very useful indeed!
Thank you for your video!
I think I'm going to pass my final test of Econometrics thanks to you! Great videos!
Thank you very much, very helpful videos with a clear explanation!
Thanks Ben
@1:58 how the units work out? Beta is cov/sigma, here we divide covariances, what happens to sigmas that should be present in the concept of OLS beta?
And why we don't divide beta's?
Beta(Independent Variable, IV) / Beta(Dependent Variable, IV) ?
This way at least sigmas are counted for, and intuitively makes sense too. ???
I think your question is not very clear. We do not really divide the COVs because they are only present in the population, he says.
I have a question. You test the strength of your IV in the 1st stage using t-test to make sure it is a good IV. But should you also regress wages(Y) on the IV to be sure it does not DIRECTLY predict Y in the second stage? I see papers where they do both tests. What are your thoughts? Many thanks!
Who carries the ball in that last play?
Funny, I have no idea what game is even being played!! :))
I have a question, if MPi is determined by latent variables Xi, can we use IV to estimate beta? Also, as MPi is binary, can we use probit/logit in the first stage?
Hi, thanks for your questions. Yes - we can use IV to estimate beta if Xi is determined by a latent variable. However, we can't use that latent variable as a IV since it will likely suffer from the same endogeneity as the variable MP. If is binary we should still use OLS, this is because using probit/logit is very susceptible to mispecifications. You are better off using LS therefore. Hope that helps, Ben