Instrumental Variables - an introduction

[@cristianoruttico7673]

Hi ben, you literally saved my life, the world needs professor like you that are passionate about the subject, and have understand so deeply every detail so they can teach properly and in the easiest way also the most complicated things.

[@flavianepomuceno7465]

Ben, as usual, you are saving me. Thank you for explaining so didactically the paper by Angrist (1990). That was extremely helpful.

[@SpartacanUsuals]

Hi, many thanks for your comment - much appreciated! If you have any ideas or suggestions of material you would like covered in videos then please let me know. Best, Ben

[@GavConnn]

great video, very well explained without going into unnecessary detail.

[@naznazz8979]

thank you so much I am phd student in Iran I find your videos so helpful

[@SomethingSoOriginal]

Thanks man, really helpful for my econometric class

[@SpartacanUsuals]

Hi, glad to hear it was helpful! Best, Ben

[@MartinLichtblau]

So many others factors could be responsible for that phenomena, in so many different ways, with different intensities. E.g. Educational level, pressure when winning the lottery, gambling mentality, believing in destiny, fixed mindset, ... So those 16% young people joining military, after receiving the draft, are for sure people with certain specific characteristics, that's the reason why the followed the call. And those underyling characteristics could explain some share of the equation, not only the trauma caused by the military service.

[@sublime9525]

But you're going to control for all those variables?

[@Kavafy]

Why are all these things any more likely to be found in the group that were eligible, than in the group that were not eligible? They are not. That is why the IV approach works.

[@ericbecker7624]

Thank you so much for this video. It is especially helpful how you name the Angrist paper that was instrumental in this model's development, very helpful as a learner to have literature to refer to.

[@TheGPFilmMaker]

Why does this only have 60k views! I've done advanced economic research but not for a while - needed a quick refresher for a new IV project I'm working on. This is perfect!

[@salomekuchukhidze5421]

Thank you so much! All your videos are incredibly clear and helpful!

[@bryvebasketball6580]

Literally the one thing my professor didn't explain well and it was half of my last midterm. Watching this going into the final

[@sharonmodiba8533]

Hi Mr Lambert, may i please request that you cover the lewbel IV method. thanks

[@georgegao5748]

Hey Ben, have you ran an analysis for the effect of your videos on Econometrics scores? What instrument would you use for the dummy variable of whether the assessment was sat before or after the publishing date of your videos? Thanks heaps.

[@SpartacanUsuals]

Glad to hear it helped! Thanks, Ben

[@centreback13]

Could you please explain again the relationship between the 16% and -$436 value. I didn't quite get it from the video. Thanks Ben :)

[@SpartacanUsuals]

Hi, no problem. The idea is that if Zi = 1 there is a 16% increase in the probability that the individual served. For this group (where Zi = 1) there was an average income which was $436 lower than those for which Zi = 0. Hence, if we want to estimate the effect of participating, then we need to work out what effect a 100% increase in probability of serving has on the income. To do this we divide 436 by 0.16. Hope that helps! Ben

[@centreback13]

Ben Lambert Perfect! Thank you so much. That has made things so much simpler for me. And a massive thanks for all these videos! They've been extremely helpful for my course. You're doing an amazing thing here Ben :)

[@hamaybe]

IDK if this helps anyone (or is even algebraically correct) but wrote this out to help me understand: From regression we found: E(LI|MP=1) - E(LI|MP=0) = -436 BUT we know that above is biased by omitted Z: P(MP|Z=1) = P(MP|Z=0) * 16% Rearrange for bias from Z: 16% (bias from omitted Z) = P(MP|Z=1)/P(MP|Z=0) Want to know E(LI|MP|Z) correcting for this bias: E (LI|MP|Z) = E(LI|MP) / bias from omitted Z = [E(LI|MP=1)-E(LI|MP=0)] / [P(MP|Z=1)/P(MP|Z=0)] = -436/0.16 Can think of LHS above as being E (LI|MP|Z) /100% (correct to find 100% effect)

[@Kavafy]

@@hamaybe It's not biased by omitted Z. Z is the instrument.

[@Digbirt]

Great video. Very helpful for those of us teaching ourselves new concepts (with your help!)

[@faro99ru]

How do we know that the the first individual who was not eligible but participated in the war was going to earn less as an average of their lifetime earnings? How do we know this individual was always going to learn less?

[@thesenate8268]

10 years later you were helpful to me internet is amazing but how do we know if draft eligibility has nothing to do with other factors..

[@MrKohlenstoff]

Because it is randomized. It is not causally downstream of the person's character or socio-economic background or anything like that, but an impersonal randomization, hence there's no way for this variable to be affected by any of the other factors.

[@sumitkumarmaji9158]

Thank you for the lucid explanation. I have one doubt. After the 1st stage when the endogenous variable y2 hat is determined, in the second stage, can we use some more exogenous (Xi) variables? Why I'm asking, that some of the exogenous variables may not be related to Y2 but important for Y1. So can we included them later in the second stage???

[@theskyking619]

That was fantastically explained mate.

[@sonalnayak7836]

Thanks Ben, very helpful

[@hellomoto170]

Ignore this: hadn't got to the ratio bit yet! Will leave original question for reference. I have a question - I understand that by ignoring those who participate in the military as volunteers we get rid of the correlation between the explanatory variable and the error term, but surely by assuming this we're omitting a variable, which is counter intuitive to the whole process? I mean, when the aim is trying to determine what impact military participation has on lifetime earnings, I get that OLS is problematic if military participation is correlated with, for example not liking office work, but how can we just ignore this (the first bar of the bar chart; people who don't win the lottery but volunteer any way) when using IV?

[@peytonfitzgerald3955]

When you refer to z=1 as being eligible for the draft do you mean that they were chosen for the draft? I'd imagine there are people who were eligible from a technical/regulatory stand point but z=0 for them as their random number wasn't chosen. If you can clarify here I would really appreciate your help!

[@meganmccormack3928]

This was so helpful in understanding the concept!

[@indiekid97]

This was a really clear explanation, thanks!

[@ninitinka]

What process do you use to select an instrumental variable? I understand the concept that the IV should have a casual relationship with X but not Y, but how do you justify it?

[@CANIbirne]

very helpful thanks a lot for taking the time to make it :)

[@ZRG01]

Thank you...It is very helpful, but what if the volunteered individuals under group 1 (MP=1 and Z=0) were omitted in the evaluation, their expenditures (including them) might affect the salary adjustment on overall military earnings and especially for individuals MP=Z=1...right? Accordingly, this will affect their Life earnings if the military budget has a certain threshold.

[@alqaaa0a]

Thank you so much. This really help

[@superfreakmusic7681]

wow this is explained so much better than in Wooldridge! (sorry Wooldridge!) At 13.34 it suddenly makes sense!!

[@eustacheeustache]

Excellent explanation ! Could you kindly share what kind of tool / software do you use for this video ? Thanks !

[@vimigsocrates3066]

Nice explanation! Had a question though: We have learned that there are three conditions required for variables to exist as instruments. Does this not break the condition that being eligible for the draft and lifetime income can have the same cause, if the reason for non-eligibility is a physical condition or something similar?

[@jonathanbower863]

No, because being eligible was entirely because of random assignment to be eligible by lottery - meaning that a disabled person would still be included on that list. Actual participation was not taken into account.

[@thesenate8268]

thanks @@jonathanbower863

[@1982sadaf]

After watching 100 videos of this course, it's the first time I really didn't get the idea, from min 9:00 onwards. :( Why we don't separate the data? Compare average income from those eligible and served (Z=1, MP=1), and those that didn't serve (MP=0)? Then we wouldn't need to worry about 16% etc. And what happens to small sample bias if the size of these two groups aren't the same?

[@botobotoboto]

Great example it really helped illustrate the dry theory behind it. Please be my teacher ;)

[@Moose75528]

Thanks a lot for this mate, I was hoping you could explain something a little bit more clearly. Say I go forth with using OLS instead of IV and force it, why will my Bols be biased upwards?

[@SpartacanUsuals]

Hi, thanks for your message, and kind words. The direction of bias in your OLS estimators will depend on the situation. If, for example, the omitted factor is positively associated with your dependent variable and an independent variable, then the OLS estimator for the effect of that variable will be biased upwards. If the omitted factor is positively associated with the dependent variable, but negatively so with the independent variable (and the independent variable has a positive effect on the dependent variable), then the OLS estimator will be biased downwards.Hope that helps, Ben

[@Кира-в6й]

oh you are a life saver!

[@ekteshafatdaily]

Thank you :)

[@samuk1000]

Useful links: economics.mit.edu/faculty/angrist/data1/data/angrist90 www.personal.ceu.hu/staff/Gabor_Kezdi/Econometrics-2/Material/Angrist-1990.pdf

[@fabriziomaccallini]

Did Angrists took into account that individuals who served had a gap in terms of work experience equal to the average length of the service?

[@superfreakmusic7681]

Just what I was thinking. It could be that those not in the military were able to get on the first rungs of the career ladder while those in the army were less likely to gain experience that was transferable to the workplace

[@aywanghoajie]

very helpful, thank you

[@supermatti]

great! you save my studies!

[@amineazzouz7762]

Thanks!

[@yavagedex]

Awesome

[@kimberlygonzales5025]

great video!

[@truemanjo9654]

U SAVED MY ASS... Which playlist is this video belongs? searching for following vid

[@iWatchYooToobe]

It's unclear to me why did he not simply run the regression LI=B0+B1*MP*Zi+...+Ei, so that your beta will only account for those individuals that were both eligible for the draft AND went to war. I'd appreciate it if someone cleared that up for me.

[@shyuejerng1465]

+TE SKH What you are describing is the interaction term. This would be acceptable ONLY if the other control variables (perception on money, satisfaction towards office work) are not correlated highly with military participation.

[@Me-ji2pn]

Why 366 when there are 365 days in a year?

[@mattsharp5999]

+Me Because there's 366 days in leap years.

[@MrAppdog]

Thanks for the video, but to clear up something very minor, you made a silly mistake in the calculation of the effect of 100% : -436/0.16 = -2725, not -2741. Thanks for sharing these videos. They really do help.

[@l.r.3225]

This (Dit is ...) is Life (Enyou...)

[@EltonJohn-g4w]

im too confused

[@ahmadfawzilafi]

I O U my degree in Finance

[@brucelee7782]

omg this is so complicated example zzz

[@bradley5210]

Stick to karate, bitch.