Why A/B tests and randomized controlled trials work

Рет қаралды 8,374

Күн бұрын

To try Shortform for a free trial, visit shortform.com/..., and you'll receive an additional 20% discounted annual subscription.
How to deal with the unobserved and the unknown: randomness
More context here: • What haunts statistici...
Stay updated with the channel and some stuff I make!
👉 verynormal.sub...
👉 very-normal.se...
FTC Disclaimer: This video was sponsored by Shortform.

Пікірлер: 34

@very-normal 3 ай бұрын

To try Shortform for a free trial, visit shortform.com/verynormal, and you'll receive an additional 20% discounted annual subscription.

@Moe5Tavern 3 ай бұрын

Somehow your videos are so well structured and therefore good to follow, the red line throughout is very clear. Thank you!

@olipito 3 ай бұрын

Your videos are amazing, please keep going!

@MrUCTeja 3 ай бұрын

Amazing explanation..! Thanks a lot..!

@anthonymoore3246 3 ай бұрын

Lovely video as always

@gonzalodiazamor5494 3 ай бұрын

after a clinical trial is great to use propensity score to match the patients. great video! thank you very much! Cheers and thanks for the content! PS. Here I am too with my 'Ventolin

@very-normal 3 ай бұрын

inhaler bros

@iamjinse 3 ай бұрын

Can u make video about difference between confounder mediator and moderator?

@yashagrahari 3 ай бұрын

3:58 Lol! you got me.

@Another_option 3 ай бұрын

Great video!

@theraizor1629 3 ай бұрын

Amazing as always, keep going

@galenseilis5971 3 ай бұрын

I've half of Taleb's Incerto. So far so interesting!

@AustinFragomen-wy8sy 3 ай бұрын

Check out Daniel khaneman he presents taleb’s idea but somehow better

@very-normal 3 ай бұрын

Which ones have you read so far?

@AustinFragomen-wy8sy 3 ай бұрын

@@very-normalBlack swan and fooled by randomness

@galenseilis5971 3 ай бұрын

@@very-normal I've read The Black Swan, Fooled by Randomness, and I am partway through Antifragile. I've also read his "technical incerto": Statistical Consequences of Fat Tails. I actually got interested in the technical analysis of extreme value theory and some of his other technical work before moving onto the pop' books.

@galenseilis5971 3 ай бұрын

@@AustinFragomen-wy8sy I've read Thinking Fast and Slow, and also his book Noise. I really enjoyed the former when I read it back in summer 2018. I could see Noise being a good book for someone that really has never thought about the statistical concepts like the variance-bias tradeoff, but I'll confess I found it repetitive. His books aside, I felt some sadness when I heard he passed away.

@chemistrycapital 3 ай бұрын

Great content!

@youteubakount4449 3 ай бұрын

Yeah chapter 2 deserves more than 1 minute bro.

@lorenzoplaserrano8734 3 ай бұрын

love love your channel ❤

@wanfuse 3 ай бұрын

don't many confounders, confound the relationships? That is if there is enough confounders with overlapping effects, doesn't this break the statistics?

@very-normal 3 ай бұрын

That’s the great thing about randomization! When it works, it doesn’t care about the number of confounders or how these confounders interrelate, the cause-effect relationship gets isolated

@joelbeeby866 3 ай бұрын

Great vid!!

@robertwilsoniii2048 3 ай бұрын

I didn't know randomization breaks confounders. I'm skeptical of that... how do we know the samples themselves are good enough? I guess it's "unlikely" to get a sample that isn't good enough. I'd feel more comfortable first prooortionately stratifying samples, and then randomly sampling them while holding the stratification sizes proportional to get a proportionately allocated stratified simple random sample. I also don't see the difference between statistical controls applied to simple random samples and experimental trials. They seem to be logically equivalent.

@very-normal 3 ай бұрын

Yes! Stratified randomization is definitely another common strategy. I didn’t cover it here, but you’re right

@guitarflori 3 ай бұрын

There is one difference. If I use statistical controls on a simple random sample, I rely on measuring all disjunctive pre-treatment causes+Proxies of Confounders (VanderWeele 2019). Some of those variables might not be measurable, or might be unknown and are therefore not measured. We can only find the exact effect if it is true that all necessary variables (according to above criterion) are measured. Randomized Controlled Trials do not rely on measuring all the necessary variables for random assigment, because the assignment to the Treatment is being done by the researchers directly in the simple random sample. They do(x) on the subjects of the study. Edit note: Of course both have the common problem of selection bias.

@orangedrink-r6f 3 ай бұрын

Great vid🎉🎉🎉❤❤❤😊😊😊

@synchro-dentally1965 3 ай бұрын

Do you have seasonal allergies? Consider consuming more vitamin C if you haven't already. It is a natural anti-histamine.

@very-normal 3 ай бұрын

lol yeah sometimes my allergies get bad and it comes through on voiceovers unfortunately. I’ll try to add some more vitamin C in my routine

@synchro-dentally1965 3 ай бұрын

@@very-normal If they're particularly bad there was also a 2020 study from Thailand on Ginger extract versus Loratadine in the treatment of allergic rhinitis. The only difference found were the side effects from taking the Loratadine.