Omitted variable bias - example 1

[@marston2020]

To clarify the confusion - this is tricky because B1 is negative. Thus, in this case, the estimate of B1 has a negative bias (since x1 & x2 are negatively correlated, and y & x2 are positively correlated -- + * - = - bias in estimate of B1), it is downward biased and not biased toward zero. E(B1) < B1.

[@vireshramcharanvr]

thank you very much because of you i passed my test im so happy!

[@SpartacanUsuals]

Hi, thanks for your comment. No the estimator in this example should definitely be downward-biased. The intuition in the video example is that 'class size' is taking more credit for better test scores than it should, due to its negative correlation with school funding which positively affects test scores. If you still need convincing I can send you a Matlab simulation which demonstrates this. Hope that helps! Ben

[@BerlinDnB]

Many thanks for your quick answer and overall you great videos. I was just confused about the sense of what "downward bias" or "negative bias" meant (though it was about "absolute" value). Btw a video explaining the general principle (4 cases, and also impact in conclusions i.e. no impact if the bias is in the direction contrary to the hypothesis) may be useful ;D (basically sumarizing what's in here: people.ucsc.edu/~aspearot/Econ113F12/Lecture%204.pdf)

[@pigeonwing9172]

You're completely incorrect. If the omitted variable is negatively correlated with the included but has a positive correlation with Y then class size is taking less credit than it should, it is indeed as you say downward biased, however you have mixed up the meaning of downward bias with up. Downward is underestimating. You are completely fucked up as you say in the video its overstimating and in your comment you say downward biased, you're all over the place.

[@SpartacanUsuals]

Hi, I think perhaps the confusion here stems from the use of the words 'downwardly-biased'. I mean it to signify that the class room size estimate will be more negative (a larger effect size in magnitude) than it would otherwise be. Forgive any confusion caused here. Best, Ben

[@katarzyna2254]

Ben Lambert Wow wow: Ben, this is my first comment on youtube ever. Not only you are great at explaining things, but also extremely patient and kind (even if some youtubers are being quite rude towards you). All the best to you and thank you for your amazing classes! 😊

[@anahitahosseini2403]

So if the class size is taking more credit for better test scores, that means because of funding being omitted, beta is becoming less negative (cuz it’s taking credit for a positive impact on test scores, even though class size itself has negative impact.) But if funding was also taken into account and was brought in to the regression, the beta for class size would be more negative. Meaning it would be something like -15. So it’s actually upwardly biased. Right?

[@jakescott8768]

Thank you Thank you Thank you- you're absolutely carrying my personal statement

[@sondao8495]

Thank you so much for the video! This helped me so much more!

[@gracetchabi]

You are so awesome. The explanations are so clear and organized. Thanks very much

[@1982sadaf]

But the class size & funding are negatively correlated! So the estimated beta may not be "over"-estimated.

[@CasperThalen]

Very helpful, thank you Ben!

[@doanhaibui]

Why my University don't have the funding for teacher like you Ben :(

[@DAS_92]

Wow - just THANK YOU

[@mansikumari4954]

Thanks It really really helped

[@skg2109]

When you specify the Beta at -10 or -5, how do you make sure that the model does not predict negative scores? How do you constraint it, to give you logical results, ie scores = [0 - 100] ? Thanks!

[@debit342]

Cov ( u, class size ) =/= 0

[@sonalnayak7836]

Thanks Ben

[@polychenko8717]

Also known as confounding

[@muhammadharoon5460]

Hi Ben. Is there any textbook required for this course?

[@BerlinDnB]

HI, I don't understand the bias direction. As the Omitted Variable (OV) is positively correlated with the Dependent Variable, and the OV is negatively correlated with the other IV, should the bias for Beta not be, in absolute value, an underestimation rather than an overstimate (I'm not sure for the rule, I read it here : www.albany.edu/faculty/kretheme/PAD705/SupportMat/OVB.pdf) Many thanks for your help.

[@andrewsmith3613]

No, increasing class seize is associated with a (1) fall in attainment and (2) a fall in funding. If we omit funding as a variable, then class size is doing the work of funding, meaning that the true effect of increasing the class size is overestimated.

[@imranhakimi1202]

thank you

[@zoozolplexOne]

Cool !!