@@NStripleseven I'd be surprised if that wasn't a Baba is You reference. (If it wasn't: Whoa is a level in a world called Meta, in that game XD) I don't have a prior though, (your knowledge about that game) so the odds of it not being a reference, and me being surprised, are unknown.
This is actual gold content being uploaded for free. It's like I'm unlearning what I learnt in all my classes and seeing Maths in a whole new way. I was asked in an interview one concept that is often confused but makes sense in general. I spoke about Bayes' Theorem. And this has given me something more to talk about. Quite possibly the best educational channel on UA-cam.
As a doctor I'm so happy you're using your platform to get this information out. Let me tell you though... it gets way more complicated! Unfortunately prevalence estimates aren't always known and are constantly changing (especially in pandemics). Another thing to consider is the gold standard. If your test looks for breast cancer you can cut out the lump and look at it under a microscope. Some diseases aren't as easily clarified. For instance, since we don't have a highly accurate, easy test for pancreatic cancer we rely on imaging, demographics, blood markers, symptoms (or lack thereof) as multiple things that form a conglomerate test to increase our Bayes factor. Despite all these things we can't always get a great prediction on whether that scar in your bile duct is cancer or just a residual scar from pancreatitis you had 10 years ago. So we offer the patient a huge surgery to remove the head of their pancreas and duodenum only to find that it wasn't cancer. You can imagine the patient is happy it's not cancer but not so happy they don't have half their pancreas and have abdominal pain and maybe diabetes. Medicine is a tricky thing. Another tricky thing is operator error. Some tests depend on the skill of the lab tech, radiologist, or surgeon. The complexity of the human body and the uniqueness of each individual also plays a role. Your test may be false positive in a particular patient 100% of the time because they have some strange protein mutation. It's tough!
"Your test may be false positive in a particular patient 100% of the time because they have some strange protein mutation." This one bugged me - When you take one test then this isn't an issue, but surely when you take two, just multiplying the probabilities together won't be accurate, right? Just multiplying them works for independent probabilities, but there is a factor linking them, the person being examined. I didn't think of the rest, and I don't work in med, but in situations like covid, I can see how prevalence rate is hard to determine - seems like it would be heavily skewed by people just staying home with a cough for one reason or another.
All too true. We deceive ourselves that medicine is as simple as motor mechanics, and even that is not that simple. Diagnosis is fraught with challenges and it is always 'best opinion at the time with the evidence available and that not present' to give the differential result, which could still be wrong, particularly with rare diseases.
Yeah just like we go for triple test for ca breast/ quadruple test for thyroid nodule or as in the case of whipple's described by you.. We still have Diagnostic uncertainties
Regarding having a mutation that rules out any chances of having a disease, that would in itself would be tested for, i.e. essentially applying a Bayes factor based on the genetic test for that mutation. The odds is based on the information we have on hand until new information is known for that patient to update the odds estimate. However, you do hit on one factor not addressed in this video which is some medical assessments are not hindered by the uncertainty inherent in many tests when the assessment directly shows-often by imaging or pathology-that an individual definitively has or does not have a condition. To add to your examples, angiograms can identify the exact location of a clot or genetic sequencing can specify whether an agent will work against a patient's cancer. (Then again, the angiogram may fail to find the clot and the tissue biopsy may miss the neoplastic cells.) I see one promise of medical advancement is to push this boundary between certainty and uncertainty.
I hope Grant read this 😇 I am an MD and Associate Professor of internal medicine. I teach medical students, residents, and fellows. I used to be a program director for a fellowship at a prestigious American university. This is a recurring lesson I teach. The example I usually use is the DNAJB9 kidney biopsy stain sensitivity and specificity for a disease called Fibrillary GN., and I do the exact walkthrough with my students. I never get bored when I see how surprised they are with the final conclusion. Which is, by the way, is: you can't use a test willy-nilly without considering the pre-probability (you are referring to it here as ”prior.” And I also tell my students that you can increase the prevalence of the disease is by applying it to the right population (signs and symptoms). I am thrilled that Grant validated this with this awesome video
I'm going to apply this to the world of dating. Everything I learn about a potential match updates my prior about our compatibility. I call this Bae's rule.
As a medical student who's done this exact thing in a FAR more complicated way, thank you! In medical terms, the post-test odds = pre-test odds (the prior) * the positive likelihood ratio (the Bayes factor) This is an essential video for any medical professional to watch and understand! I'll be sending this to my instructor because you did such a great job at explaining an otherwise very confusing topic.
As a young doctor, thank you so much. I understood the distinction between the different accuracy parameters and PPV beforehand, but this has fundamentally changed how I view testing. This is a very useful thing to understand as a medical professional.
You should definitely checkout precision and recall. They are machine learning terms but they essentially mean the same things as mentioned because the formulae are the same as well. One technique that we ML practitioners use is F1 score. As you would've seen that there an inverse relationship between PPV (what we call precision) and sensitivity (what we call recall). If you plot them for different threshold (some min value for which you classify something), you'll clearly see the inverse relationship. If you use TP, FN, and FP and put them in F1 score (TP/ (TP + 0.5(FP+FN))), you'll be able to find an optimal threshold value. This threshold we refer to as probability of something being detected as positive. And this takes into consideration both FP and FN which are crucial for medical examinations.
This perspective makes the 'update' concept so much cleaner. I've long believed that until something is utterly obvious to you, you still don't truly understand it. I just got much closer to understanding bayesian updating. I could already do it, and even explain it, but it wasn't the same. True understanding is precious. Thank you for what you do, and as always, I look forward to the next video.
As doctors, we use this every day, often without thinking about the mathematical foundations. Unfortunately, very few diagnostic tests ou exams are indeed both sensitive AND specific. So, if we think a diagnostic unlikely (based on prevalence, physical exam, previous tests, ...), we choose first the more sensitive test in order to exclude this diagnostic. On the contrary, if a diagnostic is very probable, we choose first a specific test to confirm. It is not always easy for technical exams, as we can often only choose between them (if there are several !), without changing their sensitivity and specificity. But for biological tests, we can adjust our cutting values to improve either sensitivity or specificity.
@@aitotem the video isn't moot at all. You just didn't understand the video nor the comment above. The video makes a very legit point about how a lot of people and doctors can have an absurd amount of faith in the accuracy of a test which can lead to a lot of problems. And panic. While the comment above is simply describing how doctors tread around this problem.
@@1994mrmysteryman I would ignore that guy, he seems to have an issue with the channel and is commenting lots of unnecessarily hostile and useless replies to comments on 3b1b’s videos. Haven’t a clue why, but some people are just broken wastes of space.
As a scientist who does medical tests, I'm amazed that any of the doctors asked got the right answer. Every doctor I work with assumes tests are 100% accurate.
I work in the medtech industry. In my experience, it depends on who you are working with. The more research-focused physicians (e.g. in university hospitals) tend to be better. But generally, I agree: The vast majority of physicians is shockingly ignorant about the limitations of the technology they use everyday.
as a medical student this is very instructive. if I were in that seminar room I’d probably go with the “more intuitive” answer without thinking first then act surprised when they explain that this is wrong.
Hey Grant, I am a third semester bachelor's student in physics and I find your videos very intuitive and absolutely inspiring. I find it very hard to deal with spherical harmonics this semester and as they are algebraicly complex but easy to visualize I thought maybe you can make a video about it too. It would help me very much at least. I can imagine these videos take a lot of effort and so I appriciate it very much. Thank You.
My favorite UA-cam video of the year. I have already used this like a dozen times. Even used it to explain to someone how my personal beliefs evolved over time as new evidence updated my prior odds. Love it thank you!
@@gabrielbn 3b1b translated those terms into the abbreviations for True Positive Percentage, False Positive Percentage, in most of the animations. I'm not sure on the script anymore though
@@gabrielbn I don't think of it as a flaw in the video. The disconnect in my brain between language processing and math processing is stronger than I would like! And it takes a while to just roll with new vocab, especially when you have a "which s word is which concept" situation!
I think phrasing it the way 3b1b did in this video helps a lot in making it more intuitive. When you think of concrete sample populations instead of probabilities, it makes a lot more sense.
I remember having been presented a similar problem with concrete examples in a admission test and without knowing anything I just figured out a formula that would work for the example and gave the correct answer, however when I came to study this in statistics it became so confusing and failed so many exercises by applying a formula I didn't understood
@@asdfghyter one major problem the students have is that they are careless with the meanings of the terms and mix them. You can be somewhat sloppy in algebra and analysis, but if you do this in probability, you're setting yourself up for a lot of pain.
If 1% of the population knows how to correctly use Bayes Theorem, and 80% of your students get the correct answer on a Bayes Theorem problem on the final exam, what percentage of your students will be able to recall Bayes Theorem, three semesters after having taken your class?
Thank you so much for this video. This concept of updating odds is what finally got Bayes' Rule to finally click in my head. I've used it a million times at work and it always bothered me how I couldn't do these simple calculations without pen and paper.
This is so much better than the Veritasium video on this topic. You went into so much more detail and presented it in a very clear, easy to understand way. Nice job!
Unfortunately all of them get scared after listening one word of math and never give the opportunity to think it through and see its actually cool, understable and so useful.
I'm a physician and I'll admit that I always knew about these facts (i.e. highly sensitive and specific test does not necessarily mean a high predictive value, the prevalence of the disease needs to be taken into account) and yet I always ignore what I (vaguely) know to be true and just assume that high sensitivity/specificity means that test has a high positive predictive value. I can tell you that a ton of physicians don't even bother to use these concepts at all (obviously that highly depends upon the institution and many other factors) Thanks for explaining it well!! It was nice refresher to what I learned in med school....
It seems that these tests should be explicitly described in terms of 1) the likelihood of a positive diagnosis given a positive result (1 in 11) and 2) the likelihood of a negative diagnosis given a negative result (901in 902), or some similar combination of easily-understood probabilities. If people were to walk out of a positive screening knowing that statistically they're >90% likely to be fine, they'd be able to face the further testing with an informed and calm mindset. If talk of specificity and sensitivity confuses even doctors, why hasn't someone just dumbed it down for everyone with a table that shows "for test x given prevalence y, a positive result means z". Surely that's not too much to ask.
You won me with the short of this video! I'm a psychology students whose knowledge in statistics is minimal but i needed someone that make me understand these topics. Amazing
I just finished learning about biostatisics last month as part of first year medecine in france , and it was truly wonderful to see all the concepts I’ve learned being used in such clear and concise way Great job 3b1b !
This is such an important video, especially during these times. I am a Med-Student and have never had this so clearly explained to me. This definitely confused me during lectures and I look forward to using this new perspective in my practice. Thank you!
Thank You. I'm in my late 50s and I've seen ALOT in life and I know people like you and your efforts are one of the things make life worth living. You have an incredible intellect and work ethic that has helped millions of people.
My Goodness !! Grant, there are few people who has a sharp mind to understand complex stuff, but there are even fewer people who has a sensitive mind to redesign stuff to educate people. You are spot on ! Bayesian rule is such a mind boggling thing that it always remains a source of confusion. But you showed that, instead of beating our mind around the equation, maybe we can beat the equation to wrap around our mind. Take a bow !
Wonderful exposition! I have always felt that the 'paradox' was manufactured as a consequence of highlighting the wrong metrics, but never thought about it deep enough to suggest an alternative. The Bayes factor is the right way to go. Great job!
I really like how you provide more intuitive perspectives to well known math stuff this video reminded me somewhat of your video about the different way to write down exponentials/roots
As a young teacher/educator with a strong passion for math and critical thinking : thank you. It was a legit "It all makes sense now" moment as far as I'm concerned. You gave me a whole new intuition behind this apparent paradox, one that I will gladly share around me. Cheers from France
Hands down the best maths video I've seen until now especially at 12:23 While I understood why it's true, what every part meant and can easily prove it, it still felt wrong, and I still couldn't imagine the whole process that the equation does on the prior at once. Thank you Grant and those who came up with the idea of the Bayes Factor and using odds.
This is great video, very impressive! At first I was really confused that accurate tests don't give good predictions. Then I realised that what it actually says is that the rarer the thing you're trying to detect is, the more accurate your test has to be in order to detect it. If you look at the first example, you'll see that the sensitivity and specificity (both around 90%, or 9 in 10) is an order of magnitude worse than the prevalence (1%, or 1 / 100). Trying to use that test to predict the illness is like bringing a knife to a gunfight. But if you raise sensitivity and specificity to 99%, you'll find that you have 50% chance of detecting. In general, if x% percent of the population has the illness, then we can prove that if the specificity and sensitivity is 1-x%, we'll have a 50% chance of detecting the illness. The reason is that if you have a prevalence of x%, then your odds are x:(1-x). So if your sensitivity and specificity is 1-x, then your Bayes' factor is x/(1-x), and multiplying the odds by the Bayes' factor gives (1-x):(1-x) = 1:1, i.e 50%.
Great video! This reminded me of the book The Drunkard's Walk: How Randomness Rules Our Lives, where the author was tested positive for a disease, and doctor told him that he would not survive. But on calculating the probability of the he actually having the disease, he finds out that it was low and in fact he survived. The book is an interesting discussions on use of probability in different scenarios.
4 роки тому+8
For a quick reference: the equations for calculating between probabilities and odds are: O = P / (1 - P) = 1 / (1 - P^-1) and P = O / (1 + O) = 1 / (1 + O^-1). BTW, thank you so much for making this video, Grant!
Grant, I am rewatching your videos about Bayes’s a theorem and I want to underline the quality and details (and the amount of work!) you put in your videos. We are blessed to have people like you to do such exceptional work
Using odds is quite an elegant analogy to the Bayes Rule traditionally taught. Just as you said the traditional Bayes Rule has its merits in a wide array of applications because it basically is the definition of conditional probability. However, the Bayes formula becomes confusing when we delve into medical testing where we are trying to get probabilities from the prior and test accuracy. It would be interesting to see how the odds analogy could improve our intuition of other Bayes Rule application. Great video as always Grant! I've always struggled with thinking about the Bayes Rule in medical testing.
Often when I see a video from you pop up in my abo box, I hesitate to click on it, because almost every time I get so involved that a say 20 minute video consumes, like, 1 hour of my time to really get through all of it and to understand it and to really get it into my mind to apply it in an every day fashion. But in hinsight, you NEVER fail to make the invested time worth it. Thanks for that.
As someone who's only been introduced to the terms "specificity" and "sensitivity" in March by the pandemic, IMO I really prefer just calling it True Negative Rate and True Positive Rate.
Gotta disagree. “True positive rate” seems to suggest the rate that positive results are true - but this is the positive predictive value, not the TPR :/
@@yinge101 op wasn't saying that PPV = TPR. They were saying TPR is a better phrase for sensitivity, which I have to agree with, as it implies the other three possible outcomes right there in the name.
@@SimonBuchanNz I well aware of that. My point, and a reason medicine and biostatistics prefers sensitivity/specificity, is that the term “true positive rate” is likely to misleadingly suggest to a layperson that it is the rate that positive results are true.
I started watch your channel a month ago, and now you're my favourite math channel! You talk about math in a not too formal and technical way, but at the same time, speaking about technical math things which encourage and help me to go on study math at the university. Thanks for each your videos!
Two ear ago i was start study English. One year ago i was know about that channel, but i cannt understand anything. And now I happy, because I can understand your. Thank you for your visualization of math. It really helps me for understanding math. Lineral geometry, differential equations and mathematics analysis all of that part of math seems for me enough. You are great!
*I will advise you invest with Manager Thompson and earn huge like I do and he's currently handling my Portfolio earning 200% profit with his strategies and guideance you will achieve alot*
There's nothing I can do to make my sister (to be doctor) watch anything, but I play this in front of her, she forgets everything else😁 3b1b is a magician, through and through!!
Impressive video. As assistant professor in digital processing, I’m going to recommend this video to my students. The misinterpretation of frequentist probability is very extended (sometimes is the only way among policy makers). Thank you a lot!
One small improvement - putting the Bayes factor to the left of the prior odds gives better intuition about how to multiply it out. 10 * 1:99 = 10:99 vs 1:99 * 10 = 10:99
"I got the results of the test back, I definitely have breast cancer." "No mom, you've actually just updated the probability of you having breast cancer."
As a fourth year medical student, this is THE BEST explanation I've seen for sensitivity/specificity, and actually applying those terms usefully. I'm sharing it with my class. Also, it's always useful to be reminded that doctors can't math. Just remember that, people, in all your future conversations with your physicians.
Some summary notes: Sensitivity: how often the test is correct for those WITH the disease. So, Sensitivity + FNR = 100%. The false NEGATIVE rate applies here because you're evaluating the samples that were ACTUALLY positive (i.e. WITH the disease), but TESTED negative. Specificity: how often the test is correct for those WITHOUT the disease. So, Specificity + FPR = 100%. The false POSITIVE rate applies here because you're evaluating the samples that were ACTUALLY negative (i.e. WITHOUT the disease), but TESTED positive. This video illustrates why you have to keep in mind which evaluation metric you're using to evaluate your test. Accuracy is not a great metric for the reasons described at 3:48. For something like breast cancer or Covid-19 virus detection, we actually don't care about the test's accuracy as much as we care about its Sensitivity: we want the test to have a high probability of detecting the ACTUAL positive cases for cancer/virus. We don't necessarily care (as much) about the Specificity of the test. If we have a low Specificity, it just means the test will give more false alarms. Having a false alarm is (I think we'd all agree) much better than missing an Actual Positive. This is also similar to why Accuracy is a misleading metric for situations when your data is heavily imbalanced. For example, say the airline industry wants a test to classify if a person is Terrorist vs Non-Terrorist. Well we know that the vast, vast, VAST majority of passengers are Non-Terrorist. Like 99.99% of passengers are not a terrorist. So if you had a simple "test" or "model" that simply classified each passenger as Non-Terrorist, that test would technically be very accurate: an accuracy of 99.99%. Sounds great right? Everyone agrees it's a great model with such high accuracy. WRONG! That test would naively miss Every. Single. Actual. Terrorist. Evaluation metrics matter. 12:18: Algorithm for doing Bayes Rule: Step 1: Convert your Prior Probability to Odds Step 2: Calculate your Likelihood ("Bayes Factor") := Sensitivity / FPR Step 3: Multiply
Specificity can be important - for example if someone has an elevated PSA score, the next step is a fairly unpleasant and harmful prostate biopsy. If it turns out you don't have cancer you've had an unnecessary medical procedure that will injure you and cause pain.
@@derekdreery That's a great point. It's about a trade off. Ideally we want high sensitivity and high specificity, but when dealing with a deadly cancer diagnosis, given the choice, it's better to not miss a true positive. We obviously don't want to subject patients to unnecessary biopsies. But most patients would rather undergo an annoying biopsy that ends up with a negative result rather than not get that biopsy and wind up developing prostate cancer. Neither is ideal but in the first scenario you lose a half day of time, whereas in the 2md scenario, you potentially get cancer and die.
This is just straight amazing. I love Bayes rule and this 'paradox' generally but thinking about things with odds instead of probability is a super useful insight!
Amazing video! Honestly, as someone who recently graduated from a top engineering program, took 30+ applied math courses, and saw Bayes Theorem explained in multiple ways... THIS IS INCREDIBLE. I absolutely believe it should be taught this way, and I had no idea it was possible. Three terms instead of four, that actually reflect what we care about: Prior, Update, Posterior. Finally, it all makes so much more sense! I will say, without good teaching, this is tough stuff to reason about. And it's a bad sign when even for me, having taken all those courses just 3 years ago, can feel out of my comfort zone describing Bayes Theorem to a random person. I never will again! I also think if we just taught this younger - which is probably only possible through a GREAT video like this - that it would stick around in the brain of children. And something to be revisited year after year. Learning it later, even from ages 18-21 as I did, perhaps is just too late to have this kind of thinking become really fluid and natural. So final question - do you guys at 3Blue1Brown have age recommendations for content like this? I'm more curious like... if I have kids one day, when should I show them this video? I'm guessing you guys agree, it would be great to introduce these things pre-college, but reduced as intuitively as you've done here, could this go all the way down to middle school? I swear, my parents exposed me to multiplication flashcards when I was 6, and that propelled my entire math journey. Your content, I hope, is really making a difference in how young people really get into math!
But really, I didn't even get to the end, but now seeing your comparison of probability vs odds formulas, it's so clear. The factoring out of the prior is SO CONCEPTUALLY IMPORTANT. But it's absolutely destroyed in the other one, which, as I can attest, makes for a great computation engine when solving a problem during a test, but doesn't actually promote understanding!
@@speedstone4 the actual prevalence would be unknown if you are only given the positive/negative results, not knowing what an acturate rate for false positives is. Or false negatives for that matter.
Thanks buddy, I'm a science teacher and my entire life I been using the Bayes theorem as a formula instead of intuitively. You are tha man! My students will get this topic quite more easily thanks to you.
Man, this video is pure gold! Right now, EVERYONE should know this. And this is how tests should advertise their value: The updating value of this test is 30. After taking the test your knowledge of having a desease will be 30 times better than before.
Maybe less a "paradox" and more a "glaring gap in how we teach statistics to medical students (and to everyone else)." Maybe instead of trying to bury people in organic chemistry homework to "weed out the weak ones," we could teach them simpler but far more important concepts like this one, as well as see how well they are able to care for patients, instead of being gross and Darwinian about it?
I think that people do okay with actual statistics if they actually calculate the numbers, but then they tend to find all kinds of reasons for why their numbers don't apply to the case at hand.
I WISH I had seen this in high school! I wondered why anyone cares about odds at all! How I could have gotten this far and not seen how Bayes' theorem works on odds is an embarrassment on my part, on the part of my teachers/colleagues, or both! Geez! It makes so much more sense now why bookies deal with odds. They have to constantly update their payouts!
I enrolled in medical school nearly 30 years ago. I've been presented with-and lectured on-the Bayesian concepts of probability innumerable times. I have never been exposed to the "Bayes' Factor" in this way before. Only at the very end of the video, did my brain start to realize that the "Bayes' Factor" was the same as the Likelihood Ratio, then you just casually stated that they were, indeed, the same thing. I absolutely understand "Bayes' Factor", but have only commited to memory the definition of 'positive likelihood ratio". Like so many of 3B1B videos, this was a concept that was clear and intuitive to see, but had been elusive for decades. probability". I have been unconsciously treating those as synonyms. Thank you for yet another bit of clarity in thinking.
This really helps in terms of understanding the world by thinking probabilistically. Ever since I watched Vertasium's video on Bayes Theorem, I have been searching for ways to build on my intuition in Bayesian thinking. I have been visualising on top of my head a box being divided according to the probability, and each portion being subdivided again based on how likely a piece of evidence being true or false. However this method is still not as simple as the one shown in the video. Thank you Grant!
Grant has made the greatest service ever to the so-poorly-understood Bayes rule, making it accessible to practically anyone. This explanation is absolutely beautiful.
I predict this is going to be one of your most important video to date. This clears up a lot of confusion around Bayesian thinking. Using odds neatly side step the need for normalizing the posterior probability, which to me makes Bayes rule overly complicated as an everyday tool.
Do you know that investing in forex can earn you upto 300% profit. For example investing $5,000 will earn you upto $30,000 and above depending on the success of trade
OH my god, I've never really thought much about odds terminology before. I always assumed it was the same as probability, ie 2:1 is the same as saying one in two, meaning 50% rather than 1/3. Good thing I don't gamble, holy crap!
@@GrizzliusMaximus Calculating payouts in gambling is often more intuitive with odds, which is why they use it there. For example, in craps, if you have a 2:1 payout, you give the gambler twice their stake. This would be for paying the odds bet for, say, a point of 4. This is where you get paid if the shooter throws a 4 (3/36 chance for two dice-- 1-3, 2-2, 3-1 are the ways to get a 4, with 6x6=36 possible throws) before they throw a 7 to "seven out" (6/36). The odds of this happening are 3:6, or 1:2, and since it pays true odds (no house advantage), the payout is 2:1. If you have $20 on the table, they give you $40 (and you keep your original bet, so you walk away with $60). Of course, if the shooter throws a 7 first, they keep your $20. This is way easier to do in your head than if you try to work with probability. The probability of successfully making this bet is 33%: 3/(3+6), but going from "33%" to "give them twice what's on the table" is a bit more convoluted to think through.
This is a great video. People are so unfamiliar with proper use of statistics that it’s trivially easy to mislead. I completely agree that the Bayes factor of a test is what we should discuss in daily use, not the sensitivity or specificity. Sensitivity and specificity are clearly important in creating the Bayes factor, but they can be disturbingly misleading, as you demonstrated quite effectively. It has the advantages of being a single number which encapsulates the false positive and false negative rates into a single useful indicator of it’s diagnostic value.
Thank you for going over the Bayes' Rule. I never completely understood it while learning it in some of my classes, especially under the context of false positives
This was wonderful, thank you very much! Best lecture to understand what is it is meant by prior, and what is meant by updating a belief (a probability), and what Bayesian is all about. It is appreciated. I'll look for more of your posts.
Wonderful! The idea of updating but not obliterating my prior beliefs based on evidence has been percolating in the back of my mind for years (literally decades). Bayes rule seemed proper but I couldn't see how it could be incorporated into daily decision-making. This! This feels life-changing! It is a workable and logical framework to update you beliefs based on evidence.
Good luck with that! Most of them have no scientific or technical background whatsoever, which is one reason they make such lousy decisions. It would help if the media understood these things too, but they don’t. It’s tragic.
This will literally only happen when past generations are replaced with generation more savvy to these facts. The problem is that we can’t push it any faster because people who don’t know any better or people who don’t care will not. Of course, as misinformation thrives in America, teaching the correct facts will also take such a long time. Bad terminology is also a part of this apparently.
@Hand Grabbing Fruits exp(x) is the same as e^x. The former notation is nicer in some ways because it makes the fact that the exponential function is a _function_ explicit. It also goes hand in hand with the notation log or ln.
It's definitely not linear. It is: O=(P)/(1-P) where O are the odds, and P is the probability. You can think about multiplying the top and bottom by the denominator of P to get them in the "colon" format, where both parts are natural numbers.
I greatly appreciate your hard work in demonstrating the beauty of mathematics. The presentation is exquisite and the passion is definitely broadcasted through your videos.
This is awesome. I have now updated my odds of correctly answering a Bayesian probability question.
Meta.
Whoa
@@NStripleseven I'd be surprised if that wasn't a Baba is You reference. (If it wasn't: Whoa is a level in a world called Meta, in that game XD)
I don't have a prior though, (your knowledge about that game) so the odds of it not being a reference, and me being surprised, are unknown.
ML255TV Well now you need to rewatch the video to help update your odds of knowing
That’s the spirit!
This is actual gold content being uploaded for free. It's like I'm unlearning what I learnt in all my classes and seeing Maths in a whole new way. I was asked in an interview one concept that is often confused but makes sense in general. I spoke about Bayes' Theorem. And this has given me something more to talk about. Quite possibly the best educational channel on UA-cam.
I totally agree: Educational content should be free!
@@brightsideofmaths this episode is sponsored by brilliant
@@pewien_internauta Don't get me started...
@@brightsideofmaths something needs to pay for it, somehow. Unless we are going to value skilled educators time at 0
@@12Rman21
Thank you captain obvious
As a doctor I'm so happy you're using your platform to get this information out. Let me tell you though... it gets way more complicated! Unfortunately prevalence estimates aren't always known and are constantly changing (especially in pandemics). Another thing to consider is the gold standard. If your test looks for breast cancer you can cut out the lump and look at it under a microscope. Some diseases aren't as easily clarified. For instance, since we don't have a highly accurate, easy test for pancreatic cancer we rely on imaging, demographics, blood markers, symptoms (or lack thereof) as multiple things that form a conglomerate test to increase our Bayes factor. Despite all these things we can't always get a great prediction on whether that scar in your bile duct is cancer or just a residual scar from pancreatitis you had 10 years ago. So we offer the patient a huge surgery to remove the head of their pancreas and duodenum only to find that it wasn't cancer. You can imagine the patient is happy it's not cancer but not so happy they don't have half their pancreas and have abdominal pain and maybe diabetes. Medicine is a tricky thing. Another tricky thing is operator error. Some tests depend on the skill of the lab tech, radiologist, or surgeon. The complexity of the human body and the uniqueness of each individual also plays a role. Your test may be false positive in a particular patient 100% of the time because they have some strange protein mutation. It's tough!
This needs to be pinned (=
"Your test may be false positive in a particular patient 100% of the time because they have some strange protein mutation."
This one bugged me - When you take one test then this isn't an issue, but surely when you take two, just multiplying the probabilities together won't be accurate, right? Just multiplying them works for independent probabilities, but there is a factor linking them, the person being examined.
I didn't think of the rest, and I don't work in med, but in situations like covid, I can see how prevalence rate is hard to determine - seems like it would be heavily skewed by people just staying home with a cough for one reason or another.
All too true. We deceive ourselves that medicine is as simple as motor mechanics, and even that is not that simple. Diagnosis is fraught with challenges and it is always 'best opinion at the time with the evidence available and that not present' to give the differential result, which could still be wrong, particularly with rare diseases.
Yeah just like we go for triple test for ca breast/ quadruple test for thyroid nodule or as in the case of whipple's described by you..
We still have Diagnostic uncertainties
Regarding having a mutation that rules out any chances of having a disease, that would in itself would be tested for, i.e. essentially applying a Bayes factor based on the genetic test for that mutation. The odds is based on the information we have on hand until new information is known for that patient to update the odds estimate.
However, you do hit on one factor not addressed in this video which is some medical assessments are not hindered by the uncertainty inherent in many tests when the assessment directly shows-often by imaging or pathology-that an individual definitively has or does not have a condition. To add to your examples, angiograms
can identify the exact location of a clot or genetic sequencing can specify whether an agent will work against a patient's cancer. (Then again, the angiogram may fail to find the clot and the tissue biopsy may miss the neoplastic cells.) I see one promise of medical advancement is to push this boundary between certainty and uncertainty.
I hope Grant read this 😇
I am an MD and Associate Professor of internal medicine. I teach medical students, residents, and fellows. I used to be a program director for a fellowship at a prestigious American university. This is a recurring lesson I teach. The example I usually use is the DNAJB9 kidney biopsy stain sensitivity and specificity for a disease called Fibrillary GN., and I do the exact walkthrough with my students. I never get bored when I see how surprised they are with the final conclusion. Which is, by the way, is: you can't use a test willy-nilly without considering the pre-probability (you are referring to it here as ”prior.” And I also tell my students that you can increase the prevalence of the disease is by applying it to the right population (signs and symptoms). I am thrilled that Grant validated this with this awesome video
Medical Student here, THIS IS GOLD. THANK YOU, this is going to help with my boards and future patients
I'm going to apply this to the world of dating. Everything I learn about a potential match updates my prior about our compatibility. I call this Bae's rule.
Severely underrated comment
haha this is so good!
Dear Grant, Plz pin this comment. Please.... please..... please.
lmfaooo
you’re the real MVP
As a medical student who's done this exact thing in a FAR more complicated way, thank you! In medical terms, the post-test odds = pre-test odds (the prior) * the positive likelihood ratio (the Bayes factor)
This is an essential video for any medical professional to watch and understand! I'll be sending this to my instructor because you did such a great job at explaining an otherwise very confusing topic.
As a young doctor, thank you so much. I understood the distinction between the different accuracy parameters and PPV beforehand, but this has fundamentally changed how I view testing. This is a very useful thing to understand as a medical professional.
And the ones who dont get it helped create the pandemic ...
You should definitely checkout precision and recall. They are machine learning terms but they essentially mean the same things as mentioned because the formulae are the same as well. One technique that we ML practitioners use is F1 score. As you would've seen that there an inverse relationship between PPV (what we call precision) and sensitivity (what we call recall). If you plot them for different threshold (some min value for which you classify something), you'll clearly see the inverse relationship. If you use TP, FN, and FP and put them in F1 score (TP/ (TP + 0.5(FP+FN))), you'll be able to find an optimal threshold value. This threshold we refer to as probability of something being detected as positive. And this takes into consideration both FP and FN which are crucial for medical examinations.
This perspective makes the 'update' concept so much cleaner. I've long believed that until something is utterly obvious to you, you still don't truly understand it. I just got much closer to understanding bayesian updating. I could already do it, and even explain it, but it wasn't the same. True understanding is precious. Thank you for what you do, and as always, I look forward to the next video.
This is the first presentation of Bayes' theorem that didn't leave me feeling both like it was trivial and like it was inscrutable magic.
As doctors, we use this every day, often without thinking about the mathematical foundations.
Unfortunately, very few diagnostic tests ou exams are indeed both sensitive AND specific.
So, if we think a diagnostic unlikely (based on prevalence, physical exam, previous tests, ...), we choose first the more sensitive test in order to exclude this diagnostic.
On the contrary, if a diagnostic is very probable, we choose first a specific test to confirm.
It is not always easy for technical exams, as we can often only choose between them (if there are several !), without changing their sensitivity and specificity. But for biological tests, we can adjust our cutting values to improve either sensitivity or specificity.
So basically the video is moot and based off a trick question given to some tired doctors. So glad I wasted my time.
Thanks for this -- adds some very interesting context!
@@aitotem the video isn't moot at all. You just didn't understand the video nor the comment above.
The video makes a very legit point about how a lot of people and doctors can have an absurd amount of faith in the accuracy of a test which can lead to a lot of problems. And panic.
While the comment above is simply describing how doctors tread around this problem.
@@1994mrmysteryman I would ignore that guy, he seems to have an issue with the channel and is commenting lots of unnecessarily hostile and useless replies to comments on 3b1b’s videos. Haven’t a clue why, but some people are just broken wastes of space.
The mnemonic I learned is SNOUT (SeNsitive tests rule OUT) and SPIN (SPecific tests rule IN)
As a scientist who does medical tests, I'm amazed that any of the doctors asked got the right answer. Every doctor I work with assumes tests are 100% accurate.
I work in the medtech industry. In my experience, it depends on who you are working with. The more research-focused physicians (e.g. in university hospitals) tend to be better. But generally, I agree: The vast majority of physicians is shockingly ignorant about the limitations of the technology they use everyday.
as a medical student this is very instructive. if I were in that seminar room I’d probably go with the “more intuitive” answer without thinking first then act surprised when they explain that this is wrong.
Thanks for this thread @KX36
@@Roman_4x5 isn't it H1N1?
@@Roman_4x5 hey!😂 i haven't seen it!
Grant never misses. He's always brilliant
You forgot a g
@@absence9443 what does this mean?
He don't miss
Hey Grant, I am a third semester bachelor's student in physics and I find your videos very intuitive and absolutely inspiring. I find it very hard to deal with spherical harmonics this semester and as they are algebraicly complex but easy to visualize I thought maybe you can make a video about it too. It would help me very much at least. I can imagine these videos take a lot of effort and so I appriciate it very much. Thank You.
My favorite UA-cam video of the year. I have already used this like a dozen times. Even used it to explain to someone how my personal beliefs evolved over time as new evidence updated my prior odds. Love it thank you!
One of the greatest educators on UA-cam!
For sure! He has a great way of conveying complicated ideas in a simple manner
Top priority among my 200-odd subscriptions!
@ali PMPAINT in the Universe*
@ali PMPAINT of Madness?
what really helped me was not using words like "sensitivity" but instead "true positive"
I had to go back to the definitions a few times to follow. I get lost by the words more than the math. :/
Strongly agreed. For a channel so focused on simplicity and clarity, this is a flaw in an otherwise excellent video.
Be aware that “sensitivity” is not the same as “true positive”
@@gabrielbn 3b1b translated those terms into the abbreviations for True Positive Percentage, False Positive Percentage, in most of the animations. I'm not sure on the script anymore though
@@gabrielbn I don't think of it as a flaw in the video. The disconnect in my brain between language processing and math processing is stronger than I would like! And it takes a while to just roll with new vocab, especially when you have a "which s word is which concept" situation!
This topic always breaks the brains of my students. Every year. Every time. This IS pretty hard stuff to wrap your mind around.
I think phrasing it the way 3b1b did in this video helps a lot in making it more intuitive. When you think of concrete sample populations instead of probabilities, it makes a lot more sense.
I remember having been presented a similar problem with concrete examples in a admission test and without knowing anything I just figured out a formula that would work for the example and gave the correct answer, however when I came to study this in statistics it became so confusing and failed so many exercises by applying a formula I didn't understood
@@asdfghyter one major problem the students have is that they are careless with the meanings of the terms and mix them. You can be somewhat sloppy in algebra and analysis, but if you do this in probability, you're setting yourself up for a lot of pain.
@@MrAntifascista23 "i didnt understood"
?
If 1% of the population knows how to correctly use Bayes Theorem, and 80% of your students get the correct answer on a Bayes Theorem problem on the final exam, what percentage of your students will be able to recall Bayes Theorem, three semesters after having taken your class?
Thank you so much for this video. This concept of updating odds is what finally got Bayes'
Rule to finally click in my head. I've used it a million times at work and it always bothered me how I couldn't do these simple calculations without pen and paper.
This is so much better than the Veritasium video on this topic. You went into so much more detail and presented it in a very clear, easy to understand way. Nice job!
I was just feeling bad seeing your videos when I`ve should be studying medicine. Now you`ve done the best of the both worlds.
Certainly drives home the point of why running a test twice after getting a positive result is so important when possible.
This is such an important takeaway from this lesson!
Even then, be careful. The results of multiple tests on the same person are likely correlated.
@@michaelguenot6177 And that really drives home why you want multiple distinct methods to verify a claim
It also depends on the priors and accuracy of test...it can be an overkill to repeat the test
@@paolobassi544 how much is a overkill?
this is exactly the kind of content that every non-mathematician working with statistics needs, thank you!
Unfortunately all of them get scared after listening one word of math and never give the opportunity to think it through and see its actually cool, understable and so useful.
Unfortunately, that still includes a lot of medical researchers (not to mention the editors of their peer-reviewed publications).
I'm a physician and I'll admit that I always knew about these facts (i.e. highly sensitive and specific test does not necessarily mean a high predictive value, the prevalence of the disease needs to be taken into account) and yet I always ignore what I (vaguely) know to be true and just assume that high sensitivity/specificity means that test has a high positive predictive value.
I can tell you that a ton of physicians don't even bother to use these concepts at all (obviously that highly depends upon the institution and many other factors)
Thanks for explaining it well!! It was nice refresher to what I learned in med school....
It seems that these tests should be explicitly described in terms of 1) the likelihood of a positive diagnosis given a positive result (1 in 11) and 2) the likelihood of a negative diagnosis given a negative result (901in 902), or some similar combination of easily-understood probabilities.
If people were to walk out of a positive screening knowing that statistically they're >90% likely to be fine, they'd be able to face the further testing with an informed and calm mindset.
If talk of specificity and sensitivity confuses even doctors, why hasn't someone just dumbed it down for everyone with a table that shows "for test x given prevalence y, a positive result means z". Surely that's not too much to ask.
You won me with the short of this video! I'm a psychology students whose knowledge in statistics is minimal but i needed someone that make me understand these topics. Amazing
As a current medical student, this is absolute gold
As a retired doctor, agreed :)
No its BITCOIN
Spread that shit my future nigga medic
@@harshvardhanwagare5663 you mean highly volatile, purely speculative and with no intrinsic value?
@@666Tomato666 I meant new
I just finished learning about biostatisics last month as part of first year medecine in france , and it was truly wonderful to see all the concepts I’ve learned being used in such clear and concise way
Great job 3b1b !
This is such an important video, especially during these times. I am a Med-Student and have never had this so clearly explained to me. This definitely confused me during lectures and I look forward to using this new perspective in my practice. Thank you!
Thank You. I'm in my late 50s and I've seen ALOT in life and I know people like you and your efforts are one of the things make life worth living. You have an incredible intellect and work ethic that has helped millions of people.
My Goodness !! Grant, there are few people who has a sharp mind to understand complex stuff, but there are even fewer people who has a sensitive mind to redesign stuff to educate people.
You are spot on ! Bayesian rule is such a mind boggling thing that it always remains a source of confusion. But you showed that, instead of beating our mind around the equation, maybe we can beat the equation to wrap around our mind.
Take a bow !
This actually really educated me on how tests work. This is great stuff! Literally everyone today should watch this video
Wonderful exposition! I have always felt that the 'paradox' was manufactured as a consequence of highlighting the wrong metrics, but never thought about it deep enough to suggest an alternative. The Bayes factor is the right way to go. Great job!
I really like how you provide more intuitive perspectives to well known math stuff
this video reminded me somewhat of your video about the different way to write down exponentials/roots
As a young teacher/educator with a strong passion for math and critical thinking : thank you.
It was a legit "It all makes sense now" moment as far as I'm concerned. You gave me a whole new intuition behind this apparent paradox, one that I will gladly share around me.
Cheers from France
Hands down the best maths video I've seen until now especially at 12:23
While I understood why it's true, what every part meant and can easily prove it, it still felt wrong, and I still couldn't imagine the whole process that the equation does on the prior at once.
Thank you Grant and those who came up with the idea of the Bayes Factor and using odds.
The soothing music and Grant's charming voice has more than made up for my bad day today.
Hope your evening improves!
The music is annoying. The voice should be deeper.
This is great video, very impressive!
At first I was really confused that accurate tests don't give good predictions. Then I realised that what it actually says is that the rarer the thing you're trying to detect is, the more accurate your test has to be in order to detect it.
If you look at the first example, you'll see that the sensitivity and specificity (both around 90%, or 9 in 10) is an order of magnitude worse than the prevalence (1%, or 1 / 100). Trying to use that test to predict the illness is like bringing a knife to a gunfight. But if you raise sensitivity and specificity to 99%, you'll find that you have 50% chance of detecting.
In general, if x% percent of the population has the illness, then we can prove that if the specificity and sensitivity is 1-x%, we'll have a 50% chance of detecting the illness. The reason is that if you have a prevalence of x%, then your odds are x:(1-x). So if your sensitivity and specificity is 1-x, then your Bayes' factor is x/(1-x), and multiplying the odds by the Bayes' factor gives (1-x):(1-x) = 1:1, i.e 50%.
Great video!
This reminded me of the book The Drunkard's Walk: How Randomness Rules Our Lives, where the author was tested positive for a disease, and doctor told him that he would not survive. But on calculating the probability of the he actually having the disease, he finds out that it was low and in fact he survived. The book is an interesting discussions on use of probability in different scenarios.
For a quick reference: the equations for calculating between probabilities and odds are: O = P / (1 - P) = 1 / (1 - P^-1) and P = O / (1 + O) = 1 / (1 + O^-1). BTW, thank you so much for making this video, Grant!
Grant, I am rewatching your videos about Bayes’s a theorem and I want to underline the quality and details (and the amount of work!) you put in your videos. We are blessed to have people like you to do such exceptional work
Using odds is quite an elegant analogy to the Bayes Rule traditionally taught. Just as you said the traditional Bayes Rule has its merits in a wide array of applications because it basically is the definition of conditional probability. However, the Bayes formula becomes confusing when we delve into medical testing where we are trying to get probabilities from the prior and test accuracy. It would be interesting to see how the odds analogy could improve our intuition of other Bayes Rule application. Great video as always Grant! I've always struggled with thinking about the Bayes Rule in medical testing.
Not an analogy.
Often when I see a video from you pop up in my abo box, I hesitate to click on it, because almost every time I get so involved that a say 20 minute video consumes, like, 1 hour of my time to really get through all of it and to understand it and to really get it into my mind to apply it in an every day fashion. But in hinsight, you NEVER fail to make the invested time worth it. Thanks for that.
As someone who's only been introduced to the terms "specificity" and "sensitivity" in March by the pandemic, IMO I really prefer just calling it True Negative Rate and True Positive Rate.
They are completely meaningless unless you know the prevalence of the disease in the group selected to be tested.
@@AlexB-dg9vv so are all the terms, but you can estimate them given historical data on presumably similarly affected populations.
Gotta disagree. “True positive rate” seems to suggest the rate that positive results are true - but this is the positive predictive value, not the TPR :/
@@yinge101 op wasn't saying that PPV = TPR. They were saying TPR is a better phrase for sensitivity, which I have to agree with, as it implies the other three possible outcomes right there in the name.
@@SimonBuchanNz I well aware of that. My point, and a reason medicine and biostatistics prefers sensitivity/specificity, is that the term “true positive rate” is likely to misleadingly suggest to a layperson that it is the rate that positive results are true.
I started watch your channel a month ago, and now you're my favourite math channel!
You talk about math in a not too formal and technical way, but at the same time, speaking about technical math things which encourage and help me to go on study math at the university.
Thanks for each your videos!
Two ear ago i was start study English. One year ago i was know about that channel, but i cannt understand anything. And now I happy, because I can understand your. Thank you for your visualization of math. It really helps me for understanding math. Lineral geometry, differential equations and mathematics analysis all of that part of math seems for me enough.
You are great!
*I will advise you invest with Manager Thompson and earn huge like I do and he's currently handling my Portfolio earning 200% profit with his strategies and guideance you will achieve alot*
+-- 1-- 5 --1 --8 --6 --5-- 0-- 0 --1 --3 --9
Was wondering when he would cover this, awesome!
If I could recommend any one video to watch in 2020, cat or otherwise, this would be it. Thank you!
Sh*t, this hit me hella funny!😹
There's nothing I can do to make my sister (to be doctor) watch anything, but I play this in front of her, she forgets everything else😁 3b1b is a magician, through and through!!
Impressive video. As assistant professor in digital processing, I’m going to recommend this video to my students. The misinterpretation of frequentist probability is very extended (sometimes is the only way among policy makers). Thank you a lot!
¡Gracias!
Yes! I love this alternate form of Bayes' Rule. It's so much easier to do the math with.
Focusing on odds reminds me of my Dad telling me on winning the lottery:
"It's 50%. Either you win or you don´t."
ah yes, half of the population wins everytime ofc
Seems like you guys are super good with statistics or math, but desperately fail at understanding sarcasm.
Makes me want to smash my head in
That's why you should detect sarcasm with a function of odds, instead of probability.
@2C (02) Chan Kwan Yu Well, it's either sarcasm or not sarcasm. 1:1 😎
One small improvement - putting the Bayes factor to the left of the prior odds gives better intuition about how to multiply it out.
10 * 1:99 = 10:99 vs 1:99 * 10 = 10:99
I’m a specialist emergency physician and this is by far the best explanation of a very tough concept I have seen. Great work!
So happy you're back! :) Missed your videos!
"I got the results of the test back, I definitely have breast cancer."
"No mom, you've actually just updated the probability of you having breast cancer."
Haha. What a story Claudette!
As a fourth year medical student, this is THE BEST explanation I've seen for sensitivity/specificity, and actually applying those terms usefully. I'm sharing it with my class.
Also, it's always useful to be reminded that doctors can't math. Just remember that, people, in all your future conversations with your physicians.
Some summary notes:
Sensitivity: how often the test is correct for those WITH the disease.
So, Sensitivity + FNR = 100%. The false NEGATIVE rate applies here because you're evaluating the samples that were ACTUALLY positive (i.e. WITH the disease), but TESTED negative.
Specificity: how often the test is correct for those WITHOUT the disease.
So, Specificity + FPR = 100%. The false POSITIVE rate applies here because you're evaluating the samples that were ACTUALLY negative (i.e. WITHOUT the disease), but TESTED positive.
This video illustrates why you have to keep in mind which evaluation metric you're using to evaluate your test. Accuracy is not a great metric for the reasons described at 3:48.
For something like breast cancer or Covid-19 virus detection, we actually don't care about the test's accuracy as much as we care about its Sensitivity: we want the test to have a high probability of detecting the ACTUAL positive cases for cancer/virus.
We don't necessarily care (as much) about the Specificity of the test. If we have a low Specificity, it just means the test will give more false alarms. Having a false alarm is (I think we'd all agree) much better than missing an Actual Positive.
This is also similar to why Accuracy is a misleading metric for situations when your data is heavily imbalanced. For example, say the airline industry wants a test to classify if a person is Terrorist vs Non-Terrorist. Well we know that the vast, vast, VAST majority of passengers are Non-Terrorist. Like 99.99% of passengers are not a terrorist. So if you had a simple "test" or "model" that simply classified each passenger as Non-Terrorist, that test would technically be very accurate: an accuracy of 99.99%. Sounds great right? Everyone agrees it's a great model with such high accuracy. WRONG! That test would naively miss Every. Single. Actual. Terrorist.
Evaluation metrics matter.
12:18: Algorithm for doing Bayes Rule:
Step 1: Convert your Prior Probability to Odds
Step 2: Calculate your Likelihood ("Bayes Factor") := Sensitivity / FPR
Step 3: Multiply
Outstanding
Specificity can be important - for example if someone has an elevated PSA score, the next step is a fairly unpleasant and harmful prostate biopsy. If it turns out you don't have cancer you've had an unnecessary medical procedure that will injure you and cause pain.
@@derekdreery That's a great point. It's about a trade off. Ideally we want high sensitivity and high specificity, but when dealing with a deadly cancer diagnosis, given the choice, it's better to not miss a true positive.
We obviously don't want to subject patients to unnecessary biopsies. But most patients would rather undergo an annoying biopsy that ends up with a negative result rather than not get that biopsy and wind up developing prostate cancer.
Neither is ideal but in the first scenario you lose a half day of time, whereas in the 2md scenario, you potentially get cancer and die.
This is just straight amazing. I love Bayes rule and this 'paradox' generally but thinking about things with odds instead of probability is a super useful insight!
Amazing video!
Honestly, as someone who recently graduated from a top engineering program, took 30+ applied math courses, and saw Bayes Theorem explained in multiple ways... THIS IS INCREDIBLE. I absolutely believe it should be taught this way, and I had no idea it was possible.
Three terms instead of four, that actually reflect what we care about: Prior, Update, Posterior. Finally, it all makes so much more sense!
I will say, without good teaching, this is tough stuff to reason about. And it's a bad sign when even for me, having taken all those courses just 3 years ago, can feel out of my comfort zone describing Bayes Theorem to a random person. I never will again!
I also think if we just taught this younger - which is probably only possible through a GREAT video like this - that it would stick around in the brain of children. And something to be revisited year after year. Learning it later, even from ages 18-21 as I did, perhaps is just too late to have this kind of thinking become really fluid and natural.
So final question - do you guys at 3Blue1Brown have age recommendations for content like this? I'm more curious like... if I have kids one day, when should I show them this video? I'm guessing you guys agree, it would be great to introduce these things pre-college, but reduced as intuitively as you've done here, could this go all the way down to middle school?
I swear, my parents exposed me to multiplication flashcards when I was 6, and that propelled my entire math journey. Your content, I hope, is really making a difference in how young people really get into math!
But really, I didn't even get to the end, but now seeing your comparison of probability vs odds formulas, it's so clear. The factoring out of the prior is SO CONCEPTUALLY IMPORTANT. But it's absolutely destroyed in the other one, which, as I can attest, makes for a great computation engine when solving a problem during a test, but doesn't actually promote understanding!
Ingenious! You've done a great service to the world, thank you.
Probability of liking this video before watching it is high !!
So how did you update that prior? :P
@@Ricocossa1 based on known quality of this channel
@@thinboxdictator6720 I mean, after seeing the video
@@Ricocossa1 I wish I could like the video again, after watching it. I guess I’ll have to share it with all my friends so they can like it for me. :)
_3B1B quickly doing ctrl+f to replace all the "covid"s in the script to "cancer"_
With covid it's much more complicated, because the prevalence is not well known and constantly changing.
ctrl+h
@@speedstone4 the actual prevalence would be unknown if you are only given the positive/negative results, not knowing what an acturate rate for false positives is. Or false negatives for that matter.
@@speedstone4 Testing without symptoms leads to a casedemic
im a bit surprised this video hasnt been taken down yet given how covi testing works, its not exactly politically kosher
Thanks buddy, I'm a science teacher and my entire life I been using the Bayes theorem as a formula instead of intuitively. You are tha man! My students will get this topic quite more easily thanks to you.
B..T..C a.n.d c.r.y.p.t.o.c.u.r.r.e.n.c.y i.n.v.e.s.t.m.e.n.t 💯.......
Dm
W.... H....A.....T....S.....A.....P.....P + 1.... 6.... 0.... 6.... 2.... 0.... 3.... 1.... 1.... 5.... 4.........
Do well to let him know I reffered you to him his strategies are top Notch.✅..........
Man, this video is pure gold!
Right now, EVERYONE should know this. And this is how tests should advertise their value: The updating value of this test is 30. After taking the test your knowledge of having a desease will be 30 times better than before.
Was waiting long for a video from 3Blue1Brown
Maybe less a "paradox" and more a "glaring gap in how we teach statistics to medical students (and to everyone else)." Maybe instead of trying to bury people in organic chemistry homework to "weed out the weak ones," we could teach them simpler but far more important concepts like this one, as well as see how well they are able to care for patients, instead of being gross and Darwinian about it?
I think it's also just one of those examples of humans being really bad at statistics.
@@hedgehog3180 no, humans are great at statistics when writing it down, they just can't read it back correctly
I think that people do okay with actual statistics if they actually calculate the numbers, but then they tend to find all kinds of reasons for why their numbers don't apply to the case at hand.
There are different types of paradoxes. One of the types is a fact that is true but counterintuitive
I WISH I had seen this in high school! I wondered why anyone cares about odds at all! How I could have gotten this far and not seen how Bayes' theorem works on odds is an embarrassment on my part, on the part of my teachers/colleagues, or both! Geez! It makes so much more sense now why bookies deal with odds. They have to constantly update their payouts!
it doesn't seem to make multiple iterations of updates any easier to use the Odds form... because the Bayes Factor doesn't remain constant
This is probably the most practical piece of math I have ever seen. Brilliant.
I enrolled in medical school nearly 30 years ago. I've been presented with-and lectured on-the Bayesian concepts of probability innumerable times. I have never been exposed to the "Bayes' Factor" in this way before. Only at the very end of the video, did my brain start to realize that the "Bayes' Factor" was the same as the Likelihood Ratio, then you just casually stated that they were, indeed, the same thing. I absolutely understand "Bayes' Factor", but have only commited to memory the definition of 'positive likelihood ratio". Like so many of 3B1B videos, this was a concept that was clear and intuitive to see, but had been elusive for decades.
probability". I have been unconsciously treating those as synonyms. Thank you for yet another bit of clarity in thinking.
Looking for a secure investement Expert with Good trading strategies to gain profits
Do well to correspond
+ 1 .6. 1. 4 . 9. 9. 6.4 .7 .3 .8⏏️
Bayesian statistics is amazing. It should be included in all undergraduate schools.
Long time 3B1B was thinking about a new upload on this channel today
3:56 "What does really this mean?" Lol
Do really you see any problem with that question?
@@3blue1brown yes, the grammar is wrong
@@3blue1brown Lol I see what you did here.
@@okanyakin5119 Are you sure really about it?
@@3blue1brown Really don't I see any problem at all.
This should be required viewing, at least once a semester, for all medical students. The reformulation using odds ratios is brilliant.
This really helps in terms of understanding the world by thinking probabilistically. Ever since I watched Vertasium's video on Bayes Theorem, I have been searching for ways to build on my intuition in Bayesian thinking. I have been visualising on top of my head a box being divided according to the probability, and each portion being subdivided again based on how likely a piece of evidence being true or false. However this method is still not as simple as the one shown in the video. Thank you Grant!
Thanks for the comments learn more from my team broker +1 4 0 7 6 3 9 2 0 5 8
Few Hours ago I was watching Dream' s response and Geosquare and now I am here.
I so love good, educational content. Thank you!
Me: suggested by UA-cam recommendations
My brain: just click it
Me: but i dont understand
My brain: just watch it
Grant has made the greatest service ever to the so-poorly-understood Bayes rule, making it accessible to practically anyone. This explanation is absolutely beautiful.
I predict this is going to be one of your most important video to date. This clears up a lot of confusion around Bayesian thinking. Using odds neatly side step the need for normalizing the posterior probability, which to me makes Bayes rule overly complicated as an everyday tool.
You had me at: "Picture a thousand women." 1:24
Do you know that investing in forex can earn you upto 300% profit.
For example investing $5,000 will earn you upto $30,000 and above depending on the success of trade
WRITE MY ACCOUNT MANAGER MR. MARTIN LOUIS ON WHAT SAPP TO MAKE GREAT PROFIT.
+ 1 7 0 6 6 3 7 3 4 4 1
OH my god, I've never really thought much about odds terminology before. I always assumed it was the same as probability, ie 2:1 is the same as saying one in two, meaning 50% rather than 1/3. Good thing I don't gamble, holy crap!
I actually was caught off guard here too, I thought the same thing before this video
I think 2:1 would translate into 2/3 in probability and not 1/3
I have been using probability to calculate stuff but hardly odds. I wonder how many more equations out there make more sense with odds
@@GrizzliusMaximus Calculating payouts in gambling is often more intuitive with odds, which is why they use it there. For example, in craps, if you have a 2:1 payout, you give the gambler twice their stake. This would be for paying the odds bet for, say, a point of 4.
This is where you get paid if the shooter throws a 4 (3/36 chance for two dice-- 1-3, 2-2, 3-1 are the ways to get a 4, with 6x6=36 possible throws) before they throw a 7 to "seven out" (6/36). The odds of this happening are 3:6, or 1:2, and since it pays true odds (no house advantage), the payout is 2:1.
If you have $20 on the table, they give you $40 (and you keep your original bet, so you walk away with $60). Of course, if the shooter throws a 7 first, they keep your $20.
This is way easier to do in your head than if you try to work with probability. The probability of successfully making this bet is 33%: 3/(3+6), but going from "33%" to "give them twice what's on the table" is a bit more convoluted to think through.
I've learned most of this in statistics (~6-7 years ago now I think), and this is giving me a fair bit of test anxiety...
You and the channel, Primer, have reignited my interest and math that burned so bright just a year or two ago and I can’t say how thankful I am
B..T..C a.n.d c.r.y.p.t.o.c.u.r.r.e.n.c.y i.n.v.e.s.t.m.e.n.t 💯
Dm
W.... H....A.....T....S.....A.....P.....P + 1.... 6.... 0.... 6.... 2.... 0.... 3.... 1.... 1.... 5.... 4....
Do well to let him know I reffered you to him his strategies are top Notch.✅
This is a great video. People are so unfamiliar with proper use of statistics that it’s trivially easy to mislead. I completely agree that the Bayes factor of a test is what we should discuss in daily use, not the sensitivity or specificity. Sensitivity and specificity are clearly important in creating the Bayes factor, but they can be disturbingly misleading, as you demonstrated quite effectively. It has the advantages of being a single number which encapsulates the false positive and false negative rates into a single useful indicator of it’s diagnostic value.
This reminds me of Daniel Kahneman's book: Thinking, Fast and Slow, absolute recommendation!
The breast example is also on Steven Strogatz's "The joy of X"
Thank you for going over the Bayes' Rule. I never completely understood it while learning it in some of my classes, especially under the context of false positives
4:00
He looks very young for his age.
Says God
Oh I see 👌
Hes only 73
This was wonderful, thank you very much! Best lecture to understand what is it is meant by prior, and what is meant by updating a belief (a probability), and what Bayesian is all about. It is appreciated. I'll look for more of your posts.
Wonderful!
The idea of updating but not obliterating my prior beliefs based on evidence has been percolating in the back of my mind for years (literally decades). Bayes rule seemed proper but I couldn't see how it could be incorporated into daily decision-making.
This!
This feels life-changing! It is a workable and logical framework to update you beliefs based on evidence.
10:58 the disillusioned pi face makes me so sad, don’t give up little guy there’s something good in 3 seconds I PROMISE
+1 4 0 7 6 3 9 2 0 5 8...
Even sadder is the sick pi creature at 18:48 :(
The sad pi and the sick pi made me so sad I just want to hug them
Now we only need to get this into the heads of politicians.
maybe they arent the dumb ones.
@@TheWormzerjr true
Roberto N. Good luck! What are the odds of that? LOL!
Good luck with that! Most of them have no scientific or technical background whatsoever, which is one reason they make such lousy decisions. It would help if the media understood these things too, but they don’t. It’s tragic.
This will literally only happen when past generations are replaced with generation more savvy to these facts. The problem is that we can’t push it any faster because people who don’t know any better or people who don’t care will not.
Of course, as misinformation thrives in America, teaching the correct facts will also take such a long time. Bad terminology is also a part of this apparently.
The more I learn, the dumber I feel.
It's better to have the illusion that you're smart.
Never truly understood Bayes Rule, just used the formula to solve questions but this video makes it so interesting and amazing to learn..!✌️
Wow. This makes using and understanding the relationships between these test statistics much easier. Thanks so much for sharing this!
Oh well, hello there! :3
Is andrew still in your basement
Wow didn’t expect this lol
are you going to make another math-vengers crossover integrating some fun stuff just like last year ? that would be great :)
0:48 That notation for e to the pi i is actually genius.
@Hand Grabbing Fruits exp(x) is the same as e^x. The former notation is nicer in some ways because it makes the fact that the exponential function is a _function_ explicit. It also goes hand in hand with the notation log or ln.
Now I’m curious if there’s something unique about the transformation, either linear or not, from probability to odds.
Oooh that would be interesting.
I think it should be a kind of conformal mapping (from odds to probability).
It's definitely not linear. It is:
O=(P)/(1-P) where O are the odds, and P is the probability. You can think about multiplying the top and bottom by the denominator of P to get them in the "colon" format, where both parts are natural numbers.
This is among the best educational videos I've ever seen on UA-cam, and I've seen a lot of them
I greatly appreciate your hard work in demonstrating the beauty of mathematics. The presentation is exquisite and the passion is definitely broadcasted through your videos.