Introduction to treatment effects: Part 1

[@juliogaonasalas4151]

Bro I’m glad I found this channel !

[@leonardom6840]

Great! Thanks so much! This really helped me in a moment of lost hope while I was writing my dissertation.

[@tomnewman9306]

One more little error: @ 4:38 Chuck says, "...had she chosen not to smoke," but he means "had she chosen TO smoke."

[@tomnewman9306]

This is starts out as an excellent overview, but then gets confusing. @ 6:12 The part on inverse probability weighting is confusing. Chuck says, “The first step is to estimate the probability that each subject had belonged to the other treatment group, given the covariates.” He then goes on to say that the inverse of these probabilities would be used as the weights. But this is not correct. We estimate the probability that each subject had belonged to their own treatment group by estimating the probability of treatment (the propensity score, P), and then weighting by 1/P if the person was treated and 1/(1-P) if the person was not treated. @ 6:46. The picture seems to be the same as Figure 6 on page 42 of the teffects intro of the Stata 13 Treatment Effects manual, which explains why it does not match his explanation. If age were the only covariate, we would expect the size of the circles to depend only on age. In the pdf manual, you can see that the other covariates were included in the model as well. @7:13 “The reason why this works is that we don’t observe any smokers in the younger age range, so we give the younger smokers extra weight to compensate.” Actually, this is a nice example of when you shouldn’t use these methods, because there is no overlap between smokers and nonsmokers in the youngest and oldest age groups. You can’t weight the observations in young smokers more heavily and fix the problem because there aren’t any! In this dataset the probability of smoking is 0 at young ages and 1 at old ages, thus violating the positivity assumption.

[@nip5554]

Hi Tom, I would like to get those graphs up with my data. Do you happen to know the commands by any chance?

[@zhehabeshascience3066]

best teacher

[@MegaBexus]

Could you make your code of how you created those graphics public?

[@elham4223]

if you find the code, would be great if you share it..

[@palsshin]

lovely explanation, many thanks

[@kalebnbrown]

Thanks, this is extremely helpful!

[@RojoPirata]

Thanks, your videos are very helpful!

[@moonflower_1408]

Thank you so much!!!!

[@NASIRABBASKHAN]

wonderful video! thanks for uploading, can you please make a video on calculating marginal effects and partial elasticity in a binary logistics reg. model

[@tantruongkhanh1312]

❤

[@lokeshposti2242]

how to average treatment effect on the untreated in Stata

[@statacorp]

You can obtain the average treatment effects among treated subjects (ATETs) by specifying the *atet* option with *teffects* . We don't have an option for obtaining the average treatment effect on the untreated, but you can obtain it by reverse coding your treatment indicator and using the *atet* option. Note that you will also have to reverse the sign of the estimated effect. Please type *help teffects* to learn more about this command. Contact [e-mail] if you have further questions.

[@SunnyOlga]

it would be more useful if Chuck did not talk that fast. When you provide educational video you should not speak like you are a Formula 1 participant.