Bayesian Vs Frequentist: Which should you be

An opinionated take from someone that's had to work with both in practice
Spoilers
They're both useful. Learn both! Be both
Here's the full notebook
github.com/canyon289/causal_i...

[@Mutual_Information]

I like this perspective. I've find myself using both frequently, but only b/c frequentist methods are just simpler and easier to communicate/justify. Unless I'm actually doing hypothesis testing, I always prefer Bayesian methods.. but I can't always afford them due to constraints on compute, time or effort. Practically, I agree - we have to know both well. Also, nice to see you on YT! - just discovered this channel

[@TangerineTux]

4:05 I believe that this conclusion is invalid. It corresponds to misinterpretation #22 from the article “Statistical tests, P values, confidence intervals, and power: a guide to misinterpretations” (DOI: 10.1007/s10654-016-0149-3). The valid conclusion is: if, in many random alternate universes in which we have 8 part deliveries, we were to compute 95% confidence intervals, then 95% of _those intervals_ would contain the mean of the “true distribution” from which those 8 part arrival times are drawn. That they are 95% confidence intervals, in itself, says nothing about what we can conclude from the one confidence interval that we constructed from our sample. An interesting read on the topic is “The fallacy of placing confidence in confidence intervals” (DOI: 10.3758/s13423-015-0947-8) as well as “Frequentism and Bayesianism: A Python-driven Primer” (DOI: 10.48550/arXiv.1411.5018). In my view, this is what is wrong with frequentist statistics: it answers questions so irrelevant that we are not always sure what they are, so we tend to extend those answers to other questions for which it is not justified.

[@ravink]

You're right! I should have said the mean is contained in that interval, not is greater than 2. Thank you for posting this detailed correction

[@sarasantos2811]

Thank you for these articles you suggest, they are very informative!!!

[@snowmonster42]

I'm just starting to try to understand Bayesian perspectives (and had no idea until 15 minutes ago that I'm a Frequentist!), but I'm just stunned by the example you've used in this tutorial. I thought that the right thing to do in this scenario is to call the company to explain that you really REALLY need the part within 48 hours and ask them how confident they are that the part will get to me on time . . . Oh. Well, I can see that I've absolutely framed the question like the frequentist that I am. But I would expect them to provide data to support their assertion and also, doesn't this leave out the possibility of human intervention that might affect the outcome? I thought this was going to turn into a problem of how to choose a shipper when all of them had pretty similar outcomes. I was thinking that you choose the shipper that has the fewest steps in the process because each step is an opportunity to accumulate additional error. But I can see that this is probably the way to choose how you ship all of your stuff. I might still get rid of this comment because I know that I have missed the whole point of the video - I couldn't even focus on your analysis because I was so distracted by my big question, which was "Are there really people who answer questions this way???" I'm going to have to watch it at least one more time. I might actually post the comment because I guess I'm an illustration of a goldfish who knows that it gets oxygen from water, but has no idea that water also contains hydrogen. I really appreciate the way that you have framed your discussion, though. I thought that Bayesian statistics were just a set of methods, but your video suggests that it's bigger than that -- that these approaches might be different world views. I'm reminded of when my kids were in high school. They would come home and tell me that they got an 85% on their test, which was "good!" because the class average was 75%. The first time even I was surprised when the first thing out of my mouth was "Okay . . . but what's the standard deviation?" Our world views are definitely baked in and they definitely drive the kinds of questions that we ask.

[@mattiaarsendi5421]

Thank you for sharing, Ravin! This video made me think about different points of view! Thanks

[@entropygun3661]

Very informative and insightful 💯

[@ravink]

Glad you think so!

[@user-km9vh3wb5p]

Good tutor instructor

[@JuanMrDude]

Awesome!

[@ravink]

Thank you!

[@d_b_]

Having trouble understanding the fully worded frequentist conclusion. Could you restate the slide on 4:24 and relate it to the interval calculated on the prior slide [2.12,3.64]? How did you arrive at 5 of the hypothetical means being over 2 days?

[@alanmainwaring1830]

I have been trying to sort out what all the fuss is about the two interpretations of Frequentist and Bayesian. I am supposed to know this stuff as I have been teaching mainly the frequentist approach. The trouble I have seen over and over again is that Bayes theorem is given as the be all and end all of understanding of what Bayesian statistics is all about . This cannot be correct since it can be derived from the concept of axiomatic probability theory using the concept of the sample space and random variables. R A Fischer called Bayesian methods the error of inverse probability . I agree about using both but philosophically I am still confused, is Bayesian fundamentally about belief? Over the practical approach of repeated experiments?

[@ravink]

There are situations where you cannot do repeated trials, like at SpaceX when we only had so many launches. In situations like that Bayes Theorem was the only approach that could yield useful results!

[@ravink]

Decisions need to be made even when there's only 3 data points available. you can either use no data, or those 3 data points

[@andrewnguyen3312]

I use R