Explaining Confidence Intervals and The Critical Region

[@very-normal]

ERRATA: - 7:51: this translation of the confidence interval is a common misconception. It's NOT a probability. It's better interpreted as the proportion of intervals that will actually contain the true population difference. This comes up later in the video, but I didn't correct this during editing. Proportions and probabilities are not quite the same thing in this situation. Sorry about that everyone!

[@MKhan-zo8xo]

going off of this, is the left side of the equation supposed to be t with a subscript of 97.5% or 2.5%? asking because I was curious that the notation changed

[@very-normal]

Since the t-distribution is symmetric, the percentiles have the same value, but have different signs. I usually think about the calculation in terms of the positive one, so that’s why I put the 97.5th percentile on the left too. Hope this helps clarify!

[@sjuns5159]

At 1:16 you label both the null and alternative hypothesis as H_0, then keep the alternative on the screen, still as H_0 :)

[@aangelo]

If you change "values" to "random bounds", the translation will be correct.

[@MKhan-zo8xo]

yes! thanks for the answer :D @@very-normal

[@nintendofa]

wake up babe, very normal just posted

[@very-normal]

we’re so back

[@erenjaegersrightbicep63]

Man inferential stat don't get no more intuitive than this :)) I used many of your explanations to draft the methodology section of my Bachelors thesis, and I'm reallyy confident that I'll ace the defence having watched your content. Thank you so much for your hard work!! Would really love to see more Bayesian and Causal Inference content in the future.

[@very-normal]

Good luck on your defense!

[@consumeentertainment9310]

It's weird that your videos are free. Appreciate your content and jokes!! 🤣🤣

[@somteezle1348]

Keep them coming man!

[@MKhan-zo8xo]

this is mana from heaven, great work!

[@LintangNur-x8r]

Thanks hero!!! Had been having troubles understanding the intuitive concept and this cleared it up!!!

[@OmnivorousPancake]

Love your channel dude, watched nearly all the videos in a couple of days and finally many things started to make sense! Can you make a video about Bayesian and Frequentist paradigms, their differences and commons?

[@very-normal]

Thanks! It’s a little surreal to see people want to binge my videos, but I’m glad you’re getting value from them! And yea! A frequentist-Bayesian video is in the works! It might be a while before it’s out, so keep your eyes peeled. Thanks again for your viewership!

[@santiagodm3483]

The fact that we don't know that our hypothesis are actually true is one of the reasons we should i encourage replication in the journals in order to be more sure about what we know.

[@Featherlicht]

That’s why meta analysis is a thing. But the way studies are conducted or reported can be very heterogeneous which makes meta analyzing more difficult.

[@erikross-rnnow5517]

Cool video, I think CI's are super hard to get intuitively as a beginner, especially the differences between two sided and one sided. I would love a deeper dive into onesided CI and when those come up IRL

[@very-normal]

Thanks! I’ll try to figure out a way to fit a one-sided CI section into a future video. For now, one place they come up often is when a pharma company is trying to demonstrate that some new drug has a high enough response rate in phase 2 trials. For example, 30% might be the minimum response rate that might be acceptable. The company only really cares if the response rate is greater than 30%, so they’ll probably opt for a one-sided CI here. It wont matter if there’s evidence that the response rate is lower than 30%, since it would be axed anyways.

[@erikross-rnnow5517]

@@very-normal Makes sense, cool example :)

[@aangelo]

Great video! I noticed one typo: you write H_0 when you speak about the alternative, approximately between the 88s and 102s.

[@very-normal]

Oops, sorry about that! Thanks for letting me know!

[@nickleland2131]

Is that... animal crossing music in the background?? Phenomenal video as always!

[@very-normal]

You got some good ears

[@vrl9037]

Where were you during my degree bro 😂 These videoes are so good

[@bcs1793]

Could you clarify what is the difference between level for Pearson-Neyman and significance level for Fisher?

[@very-normal]

hello, sorry for my late reply. I see you've been going through this series, so thank you! FIsher's significance is kind of like a line in the sand. It marks a point that the p-value would be too low to ignore how improbable the data/statistic would be under the null. Neyman-Pearson's level can be thought of as an "acceptance level"; that is, if we were to repeat this experiment many times, at what rate would we accept (or tolerate) a type-I error among these experiments. It's more focused on keeping bad decisions to a minimum, rather than deciding that a probability conditioned on the null is getting too low. That being said, today statistical referees will rarely distinguish between the two, so I've found that "statistical significance" is just the catch-all

[@RagaarAshnod]

Didn't receive a notification from KZbin :c

[@very-normal]

KZbin hates confidence intervals :(

[@santiagodm3483]

We went bayesian in 7:51🙈🙈.

[@very-normal]

Broke Bayesian for a bit

[@psl_schaefer]

Maybe this is being too particular, but at around kzbin.info/www/bejne/m5KlZ4ajgK95qM0 you say that the left point is for very "low differences", whereas I would rather say for "very negative differences". I.e. intuitively low difference implies for me low absolute difference. Anyway, thanks for making those videos :)

[@very-normal]

Ah yeah, that makes sense lol, I’ll try to remember this for future videos. I think your phrasing is clearer, thank you!

[@xenonmob]

The "translation" at 7:51 is absolutely the incorrect interpretation of a confidence interval. I'm surprised you subscribe to this very common misconception.

[@very-normal]

You’re right, this one fell through the cracks, that translation shouldn’t be phrased that way. I’ll add an addenda to clear that one up

[@santiagodm3483]

Everyone makes mistakes, thank goodness he realized it.

[@demiangrams]

Yeas but it is not really "absolutely incorrect", 95% of the confidence procedures will contain the true value, but that's a pre-data statement, and it is about the procedure, not the parameters. Interpreting any particular observed confidence interval is very hard, for those interested, read "The fallacy of placing confidence in confidence intervals" for some considerations.

[@LordDockerton]

How do I give you money?

[@very-normal]

your viewership is all that’s needed 🫡