Stata Tutorial: Basics of Instrumental Variables and Two-Stage Least Squares

[@DevikaSharma-bv9ro]

Absolutely invaluable explanation, thank you! Also loved the Venn diagram, it's locked in my brain now :)

[@sayedalimsamim551]

Dear professor, thank you very much for your scientific explanation.

[@matthewdouglas6988]

Your videos have been a great place for me to come after I've read the textbook and attended lecture. You really help clarify things - much appreciated!

[@mikejonaseconometrics1886]

Thanks! Glad they are helpful.

[@emmanuelakuoko-konadu6469]

keep up the good work Mike. we're so so much enjoying the teaching.

[@mikejonaseconometrics1886]

Thank you! What other topics would be helpful?

[@elodiegradlife6904]

1:47 have a correlation between y2 regressor and error term, then any OLS coefficient estimator will be biased 2:21 assume x1 is exogenous uncorrelated with the error term, and y2 is endogenous 2:50 three properties for instrmental variable Z 8:36 2SLS and IV estimator 9:38 save the fitted (predicted) value of Y2 in stage one 10:53 do in STATA 13:10 error term might inlcude talent or hard-working that can also influence wage in addition to education 14:18 look for factors that related to education (Y2) but not affect wage (Y1) 15:37 eststo ols 16:05 ivreg wage exper (educ=sibs) 16:50 eststo iv; reg educ sibs exper; predict educhat (first stage) 17:29 reg wage exper educhat; eststo twostage; estout (second stage)

[@Cookiethemongrel]

Really useful and simple-to-understand explanation, especially the venn diagram! The use of STATA at the end was really helpful as well, so practical! Really appreciate the effort :)

[@AI_Masterpiece_]

Mike! You're saving me man! Such a clear explanation. Thank you!

[@AI_Masterpiece_]

Have my subscribe as a token of my appreciation :)

[@Laura-vn1qo]

Hi Mike, your videos have been superhelpful! Thank you so much

[@AI_Masterpiece_]

Have a cup of coffee on me! You really saved me here!

[@belenguerrero2087]

thank you! it was very helpful

[@samsayn5170]

you are the man, a helpful man thx

[@mikejonaseconometrics1886]

Happy to help

[@rizkyyusviento4021]

Thanks a lot, it helps me to conduct 2SLS. God Bless

[@scotthu9833]

nice job Mike! I hope you were my Econometrics teacher !

[@mikejonaseconometrics1886]

Thank you!

[@danielkrupah]

Please after performing the two-step system GMM method, is it necessary to use the IV method to solve the endogeneity problem again?

[@iskandaramanov189]

Thank you Sir!!!!

[@bafanashabangu9725]

Thanks man👍🏿

[@mikejonaseconometrics1886]

No problem 👍

[@popi20101]

Thank you, it's quite clear. But I have question, should we use ivreg2 or ivreg or ivregress?

[@mikejonaseconometrics1886]

Good question. ivreg2 and ivregress give more options, mostly useful for over-identified 2SLS models. For "normal" IV models, ivreg will work fine.

[@popi20101]

@@mikejonaseconometrics1886 ok i'll try to simulate again

[@sirawinpobsook3669]

Thank you very much Mike. I 've a question about the second stage (10:44). What is B2(IV)?

[@sirawinpobsook3669]

Sorry I have another question, Is standard error of IV always less than 2SLS? Thank you in advance

[@EvilAbrasax]

Dear Mike, thank you very much for this great video. I wish I found that earlier. If I may, I have a question. Maybe you could help me with that: As you explained, in the first step of the 2SLS I separate Y2 in two parts - a "good" part (Area D / predicted values) and a "bad" part (everything else / residuals). In the second step, I use the predicted values instead of Y2 in the orginal regression. But, what if the coefficient of the predicted values of Y2 is now insignificant? Especially if the augmented regression (Davidson and MacKinnon, 1993) shows me that the residuals are insignificant as well, and the Hausman test allows me to not reject the null hypothesis of exogeneity. Is this a major endogeneity problem? Shitty instrumental variable? Or can I assume that I have no endogeneity and, thus, the "correction" of the 2SLS is incorrect in this case? Thank you very much.

[@jenai20]

so the point is to have the beta of interest (gunna refer to it as beta1) change due to the IV which indicates an unbiased beta1. and you want the ivreg code's adjusted beta1 to be the same as the two stage code adjusted beta1. Feel free to verify this for me even if you aren't mike. ALSO, would there be a time when the ivreg beta1 is NOT equal to the twostage beta1? I've been running some code and it seems to be a trend that the adjusted betas are equal and other than the rsquared i'm wondering if i should be looking for other indicators of a good IV. i know some of my IV should not be valid and i should check them by some other means i just don't know which means exactly

[@popi20101]

What about to get estout with star (level of significancy)?

[@mikejonaseconometrics1886]

Try this: estout, cells(b(star) t) to get the coefficients with stars, and t-stats. Type "help estout" in Stata and scroll to the bottom to see other examples of modifications.

[@scotthu9833]

Hi Mike...I just encountered a question...I find that that my OLS estimator and IV estimator have opposite sign (The First stage F value is 12)..May I ask what are the possible reasons?

[@mikejonaseconometrics1886]

In theory, that means that the endogeneity bias in OLS is so large that it pushes the estimate to the opposite sign of the “true” value. This won’t mean so much if neither OLS or IV estimates are significantly different from zero. I suggest trying IV with different instruments if possible, and if the results are robust, trust the IV coefficient.