Jorbs, as you asked for feedback. When you discuss your uncles fish story, split the slide after you describe the problem, then ask the class you are talking to for some ideas for what could be going wrong, before revealing the answer. The students will read to the end of rhe slide before you get through explaining what was happening. Great stuff as always
I was going to comment this same thing, yeah. Even if you don't ask the class and just keep the presentation the same other than hiding parts of the slide, I feel that would make the presentation feel better. Now time was spent on talking about the mystery while the answer was visible to everyone.
Agree with OP, and I’ll add that the slides in general are far too wordy. A true presentation to an audience should be limited to speaking prompts, images, and/or core concept emphasis. Jorb’s standard presentation style (few slides, many words) is good/fine for engaging an already engaged audience. We all choose to be here. High school students are not that audience, haha. I like the TED talk “death by PowerPoint” that goes through some good examples of how to make a good PRESENTATION.
Yes 100% this. Showing information only when the students are prepared for it by the context of the rest of the lecture will also make it easier to follow the flow of logic, because it will be clearer which point you're discussing right now. As a student, it's a very different effect to be hit with a wall of text vs being drip fed the text one point at a time. Even though it's the same course content, I always find it easier to absorb the information in a math-focused lecture when the professor writes the notes on the board as we go rather than presenting slides that already have the information written out beforehand, even if it has all the requisite steps. Anyway, keeping what's on the slide to just a short prompt or the general idea of what you're saying will help because the student can read the bottom most point and then get back on track with the context more quickly than with full sentences. If you still want the full slides available to the students, you can provide these slides, or have the points in the presenter notes, for the teacher to provide to the students if they want to look back on them with more context.
jorbs, I just want to let you know that I am thoroughly enjoying this statistics mini series despite the situation it arose out of. As someone majoring in data science, one of the things that I get nerdy about is what exactly statistics means, as it is often misunderstood. Thank you for these great lessons, hope you’re doing well.
As someone who is pretty shit at stats (I failed out of AP stats in highschool despite my best efforts lol) and admittedly doesn't really give much of a shit about statistics in general, these videos have been really fun and are great content to put on while I'm at work. It's something good to pay attention to that puts me in a productive mindset honestly Seconding the last part as well, hope all is well and thanks for the neat content despite the original intent
your slides are very static: all of the bullet points are already on the slide as soon as it comes up. for example the the fish story (Slide 3) the "answer" is already on the slide, meaning that the students could just read ahead of your story, and not directly pay attention to the lecture itself. I know statistics can be dry, and I also know that you can have a fair amount of energy.... "sometimes" maybe set some of the bullet points to be on cue, and then for that slide ask the class if anyone has ideas for "why was this students data 'wrong'?" and then when you reveal the bullet point "The data wasn't wrong, everyone else was"
You can relate the "update our understanding slide" to the coin landing on its side, i.e. lets say you thought it was 100% false that a coin can land on its edge, but seeing it happen one time, flips that to 100% true that a coin can land on its edge. Furthermore, if you see the coin *almost* land its edge but tip over, that might change your understanding from 100% false to maybe 90% false. Everyone loves callbacks
22:34 So the thing with standard deviation is, while it is possible to calculate the average absolute deviation (that is, the expected value of |x-u| where x is the actual value and u is the mean), it's often unfeasible to do so for more cumbersome distributions such as the normal distribution. It's actually a better measure than standard deviation due to being more robust (less susceptible to infrequent but extreme outliers), just algebraically harder to calculate.
It really depends on what you want to quantify, the standard deviation is especially important to how it interacts with the central limit theorem. It's not just easier to calculate. It is indeed often the most informative measure of how much uncertainty a random variable has. You just have to interpret is correctly. E[|X-EX|] is by the way often feasible, in particular for the normal distribution. See en.wikipedia.org/wiki/Half-normal_distribution in this case the difference this just a constant factor between both measures. What I want to say, it's not arbitrary. Unfortunately it needs a lot of math to see why it is. Maybe you can get 3b1b to make us an more intuitive explanation ;)
I remember looking the average absolute deviation up in uni, because I was really unsure about standard deviation being the measure we use There were a few reasons, and another one of them was that if you have to mess around with differentiating and whatnot for some reason, the absolute value term can get really inconvenient (as a graph of absolute value will have a kink in it, making the function non-differentiable at that point)
Chiming in to reinforce the point about splitting the conclusion to the fish story. I'm sure I've heard you talk about this on stream/in another video but forgotten the answer, and being able to read ahead to find out what it is deflated a lot of my engagement with the potential causes. I think I would have been absolutely rapt if i'd just read the 'N=1 was correct' part but not been told why until you got there. love the stats content btw.
Hey Jorbs, since you explicitly said you're looking for feedback: I thought this was a really interesting presentation when it stayed more on the topic of sample sizes. The insights of how we can't just ignore anecdote when shaping our understanding, and how big data's vague promise of "more data" can be used to paper over shoddy collection methodology, were extremely valuable. In the 30 to 3000 range, I think the presentation wandered a bit, and you spent most of the time talking about what you can *do* with statistics (normal and binomial distributions, etc) rather than *why* 30-3000 is the normal statistical range. I think it would be pretty powerful to touch on the mathematics and show exactly how scaling up the sample size affects the certainty of statistical models and why you don't even need to approach a census in order to be quite confident in your results. This would then segue into your point on big data and how a census can actually be *worse* than a small SRS because any holes in the data become very systemic. Semi-unrelatedly, the discussion of how a normal distribution is basically just secretly the noise of a bunch of binomial distributions was interesting, but a bit of an aside and didn't really have the time to be fleshed out into its own argument. I'm not sure I quite follow it either even after thinking pretty hard about it. For example, one of the biggest factors for adult male height is "How tall were your parents?" and I don't see a way to map that information to a yes/no question that isn't extremely tortured. It's an interesting thought but left me more confused than enlightened--if it's meant to be an oversimplifcation then I think it could benefit from shouting that out explicitly because it wasn't clear to me, and if it isn't an oversimplification then maybe leave it out until you can dive deeper into it some other time. Overall this was an interesting talk and I enjoyed listening. I'm sure it'll go great. Good luck cramming if you do try to incorporate feedback tomorrow morning :)
A tip: Keep the amount of text on each slide to a minimum. The more text, the more people's attention will split between reading and listening. It can also work even if you have a fair amount of text to not show all of it at once, but rather reveal bullet points through several steps.
Probably too late to include in the presentation by now, but one thing to consider on the inclusion of women in medical trials is that historically periods have been seen as a complicating, difficult to control factor, and so that had also played some role in their being left out of some trials. This means that around half the population has been under represented in studies due to a regular, recurring fact of life that made the math more difficult. And that matters because due to differences in part related to how sexed bodies absorb and process chemicals that impact the endocrin system means that women are regularly given higher than necessary dosages of medications, the classic example of which is Ambien, the sleeping drug that was linked to an increased risk of car accidents following ambien usage, because it would linger longer in women than in the men that made up the test population prior to FDA approval. Since 2013 the dosage approved for women has been halved, but ambien had been on the market since 1992
Hi Jorbs. I teach college math, so I have advice I can give. Please read this when you are in a good headspace for feedback since I always tend to come across as highly critical when I really mean the best. Also as an educator, my apologies for “grading” with this advice way later that you probably wanted it; hopefully it can be taken into account for a “next time.” Overall your presentation is very good. You have many ideas that you flesh out very well. You are passionate, and it’s clear that you know what you are talking about. Mathematically, it all checks out. For a *UA-cam* audience, your presentation is perfect. For an *academic* audience, there’s a handful of things you should change. First, your slides need some help. For a UA-cam audience, you have nothing to change. I can pause/rewind you at will. I can read your slides at will. For UA-cam, *change nothing about what you are doing unless you want to.* For an academic talk, I cannot pause/rewind you. If you write paragraphs on slides (with the exception of quotes that you should read aloud), you will split your audience’s attention. This diminishes the value of both your spoken words and your slides. You also should making “building” slides for a talk like this; i.e., you can keep the same “big topic” slide, but bring up the bullet points one-by-one. Bringing up all points at once can be confusing because your slide isn’t “tracking with” your current spoken topic. What I would suggest to you is to imagine what you would write on a chalkboard as you were giving this lecture without a PowerPoint. What types of main points would you want to emphasize by writing them? What points would you omit or shorten? Which points are worth your time to write? Your slides should only be an *enhancement* of the things you would otherwise be writing. Also remember in this setting you have note-takers, so you need to make points on your slides note-taker-friendly. Here’s an example of what I would do to change one of your slides: Slide title: Some Observations With N=1 Are Extremely Powerful •Example: Uncle’s Fish Observations •N students, N-1 had matching data, 1 had very different data •What explanations are there for mismatching data? •Made up data? NO •Recording the data wrong? NO •Different time of day? NO •Actual answer: Her height affected fish behavior! •Takeaway: The N=1 data was right, so all the other data was wrong. When presenting this slide, you should only bring up one bullet point at a time, maybe bringing up some of the sub-points together. The rest of the time, you are talking about each point. Describe (with speech) the fish experiment and the problem. Describe (with speech) why the common explanations don’t cover this situation. Describe (with speech) why the actual answer is surprising and how it leads to your main takeaway. Notice I also omitted the coin part of the slide; you made a brilliant point with the fish story, and the coin was taking away from that. If you want to do the coin landing on its edge example, work it in somewhere else. The second major thing I want to bring up is knowing your audience. For *UA-cam*, you nailed it. You are speaking to an audience where you are trying to get to a “common denominator,” so you have to explain things like binomial distribution, mean, standard deviation, etc. For an *AP Stats class*, you went over a bit too much. They likely are already familiar with binomial distribution, so a quick review is good enough. For mean and standard deviation, you don’t need much of a review for an AP Stats audience. Taking time to explain basic concepts to an audience that doesn’t need it will take away valuable time and attention from the points you want to make. On the note of audiences, you have the unfortunate conundrum of a high school audience. While I like your examples about Viagra and believe that they should be appropriate for your audience, you have to be careful. Whether you or I or anyone else likes it or not, it is best to not step on the toes of any parents/guardians or school admin with examples that get into topics of sexual health if you are not teaching a health class. I would also hesitate to bring up any examples that start to “get political.” If this were for a college audience, then I would not be bringing this up at all. For future reference, just be *very* careful with the real-world examples that you choose to talk about if your audience is K-12. I hope this is helpful and not overly critical; I want to help you give the best in-person lectures that you can. You are very good at communicating your unique and insightful ideas. I really appreciate all of your UA-cam lectures and want to see them continue!
Good job on describing stats. The first example you raised is called bayesian logic where you update your priors based on new information. if your current knowledge is that green doesn't exist and then you experience green then its a strong indicator that green exists. on the other hand if you experience something yellow then you might start to think green exists. Your prior still has a large sway over your view but it has been updated with the possibility that green does exist.
I was also given that argument for why standard deviation is a squared term. we don't want the negatives. Also if i am remembering correctly, on the normal model, one standard deviation is at the two points of inflection of the curve.
You think you're just enjoying your Spire content but then a couple of hours pass and you look to your left, take a peak at your right, everyone is taking notes. You look at the board as the tutor explains the differences between the distributions displayed on the board. You begin to question why you took statistics in the first place.
As an undergrad, I spent a lot of time learning the mechanics of stats: how to propagate errors and do the different tests and so on. And because I did all that math, I got tricked into thinking that doing stats was like casting a magic spell. If it had error bars or a standard deviation, it was to some extent Truth. So when I got deeper into academia and I had to apply this stuff in a situation where the answers were unknown, I learned... well, that stuff isn't so important. Like, you publish your standard deviations and draw error bars on graphs, but in most situations, you can never be certain that you've sampled things correctly. You can't control all your variables. You can't even know if you got the important ones that would change your distribution. Did you even ask the right question/a sensible question to start with? In fact, you probably didn't. I think that's a good lesson, and you had some nice discussion in here. It's worth emphasizing how devilishly hard it really is. One thing I don't think you mentioned, but which might warrant a few words, is the interpretation of a statistic. You've differentiated between descriptive and inferential stats in other videos, and the assorted fallacies that come from not understanding the difference. Since I observe (and therefore I conclude confidently that we all observe) those fallacies on a daily basis, I suggest that it might be nice to tell those students, hey, as you're going through this material, keep an eye out for this. Don't trick yourself into thinking these numbers mean more than they do, and look for situations where others are doing that. Thank you for another excellent video. This and your book are getting me through night shift tonight. ❤
Hi Jorbs! Thank you for working so hard to make statistics accessible. It has genuinely had a profound impact on my life! Starting when I first watched your video "How I think about strategy games" a couple years ago, I really developed a passion and fascination with statistics. That video in particular marked the first time I really came to understand how little I understand, and that made it possible for me to truly appreciate the way you think about and talk about statistics and probability, and to challenge and outgrow my own limited view of it. It flipped a switch of sorts, and opened some paths for me to expand my boundaries! So it might seem dramatic to say, but the results of this paradigm shift and your content have had a profound effect on me! At first it was just how I played StS, then it was other games, then it was anything involving probability, and now I can confidently say that it has changed how I see the world. Its not an exaggeration to say that learning to understand what stats mean in the real world has broadened my perspective in nearly all topics. It has even led to some really awesome creative breakthroughs with challenges at my work as a software developer! Anyways, all that to say, I am extremely grateful, and wanted to share this little way that you've made a big impact. You'll always have a fan in me! :)
I think there is an error in the slide at 22:35. Standard deviation is the square root of the average variance, and variance is the distance from the mean SQUARED. So if you don't want to use the word "variance" in the slide, then it should read: Standard deviation is the square root of the average squared distance between the data points sampled from the distribution and the distribution's mean.
Along with the tips that everyone else had about managing the flow of information and avoiding spoiling the next points early, I would reccommend thinking about how you might use the whiteboard/chalkboard/smartboard in the classroom to supplement visuals where possible. When pointing to the graphs in the mean/varience sections, it would help me personally as a student if you marked off visually the information such as where the mean is and roughly how far the deviation goes out from the mean. I know that these are AP stats students, so they're pretty familiar with variance and mean, but I think it would help.
In your section on mean and varience, you had to refer a lot to a previous slide, and did some pointing at the graph. I think it might be helpful to have a visual representation of mean and standard deviation on the mean and variance slide
It's unsportsmanlike to let Twitch chat help you with game strategy but it is completely fair game to let UA-cam comments help you with your homework. This sounds sarcastic but it is simply a true statement that I agree with which is funny.
my biggest note for this presentation is that once you start talking about the power of statistics to do good and evil you can relate that back to the metaphor of coin flipping, society making people "flip the coin wrong" the n=1 case was fascinating, I've never thought of it that way
Hey jorbs! Not sure if you want to include this tie-in, but many "machine learning" models are effectively sampling human behavior and then trying to replicate that outcome, with similar caveats. So if you want a machine model to rate a writing sample, you have the machine sample a lot of human ratings of writing samples - and that also has the "bias in, bias out" effect.
Minor nitpick. It might be more clear to use "dispersion" when referring to the general concept of distributional width than variance which might imply a specific meaning (i.e. the square of the standard deviation).
Nice Spireside Jorbs Chat on my PC with Mega Crit's new Spire Yule Log on my tv EDIT: Dreams shattered. I can't believe Jorbs said I'm too tall to be a marine biologist.
I really enjoy that you're humanizing statistics and explaining some of the reasoning behind it (especially explaining that standard deviation is arbitrary)! One thought on the "Sampling is like this, too" slide: don't reveal that the country is 55/45 at the start. If you're at a polling company, your job is to figure out whether the population is likely to be 60/40 or 50/50 based on a sample. So, saying "165 people polled said they would vote for candidate A and 135 said candidate B, which might mean that the population is 55/45... or it might not" might make your explanation match the actual statistical work being done, and it might also be a more natural lead-in to your comments about bias (165 vs 135 in the sample doesn't imply 55/45 in the population).
Ive heard that theres some denomination of coin somewhere that actually can land on its side, but I dont remember which one it is. It makes sense that it would depend on the coin.
Jorbs inspiring the next generation of statisticians one UA-cam lecture at a time! I echo most people’s comments about not showing the answer to the uncle anecdote right away. Great stuff!
Thanks Jorbs! I love these videos, I dont watch your slay the spire content as much anymore but I love this style. You have a great way of communicating math and how to incorporate into decisions. I've genuinely used some of your advice in my job and I would like you to know that it led to better maritime safety.
At first it wasn't clear to me who was observing which fish where. The immediate and obvious false explanation was that the outlier data were from a totally different species or lake full of fish. The fact that they were all in a fish tank in a university was either unclear, or I snoozed through that somehow and inferred it later as the story went on.
Gosh. Just this morning I was googling something about sample size evaluation (I'm not great with statistics - in case this sounds funny). And just a few hours later UA-cam gives me this. I thought that it's my search history affecting my feed, but nope - it's Jorbs posting a video. Don't tell me it's some sot of confirmation bias again.
I might be a bit late to the party here, but one thing I would add is that you have to make sure what the data you actually want is (a.k.a. your population). So before you even think about how to draw from your population you have to define it well. Easy example: all American residents eligible to vote. Asking a bunch of kindergarteners about how they would vote is not skewing the coin toss. It's not tossing at all. Not doing this is very problematic and leads to shit results (and it's way too prevalent in my field).
It'd be great to ask the class if they have any questions by the end of the presentation! Also, I saw another comment say this and I thought it was a great idea, so I'll reinforce it here: it'd be really cool when you're talking about your uncle's story to ask the class what they think the answer to the outlier is! It's an effective and simple way to make the class more involved on the lecture
Just want to say thanks for these informational videos on statistics. I took some (fairly basic) stats classes many years ago and your videos been a great refresher.
Great video! I often think in ~all of the work in statistics is getting the right sample, the p-values and inferences are often just formulas you copy once you feel your sample is good. Since you asked for feedback, the use of a 100% or 0% probability is colloquially okay, but I try to avoid it since mathematically it makes little sense. One can’t do a Bayesian update from 100% to 95%, and someone who believes a 100% probability would be willing to take a 1 trillion-to-1 bet on the event, which is almost never the case
I'll write this as I watch along, hope it's helpful feedback. 1. Cut the tangents, teenagers have short attention spans, this is a chance to lose it. Parachutes, coins dropped into mud, wondering about the accuracy of the average man's height, I don't feel these details are relevant to statistics nor add good flavor to the presentation. 2. A good number of students will read your slides instead of listening to you. I would suggest not showing the entire slide at once, but introducing each bullet as you start talking about it instead. 3. 'We Can Update Our Understanding Of the World Constantly!' is a 10/10 slide, especially for a highschooler to hear. 4. Have a quick look at the wiki page for distributions so you can speak about it with more certainty. The audience will be more engaged if they feel the speaker really knows their stuff, and this is an easy way to score points, or lose them. 5. I would not spend as much time discussing the intuition of how many heads we expect in 10, 100, 1000 coin tosses, unless you explicitly want to talk about the reduced variance with bigger n. I actually would strongly suggest taking this chance to do so, if you can prepare this in time. Upon hearing the title, I would have expected some discussion of how the size of a sample changes it's behavior. 6. Limit the number of times you repeat "lots", especially after the first time I would only say it once. The audience gets the point. 7. On standard deviation, I would suggest giving the intuitive explanation of the graph before bringing in any rigorous details or math, and definitely before reading the bullet. 8. The air quotes "artificial intelligence" went over my head, I imagine the same will be true of much of the audience. I think this bit should be better explained, or else not done. Overall it's a great presentation, best of luck.
I agree with some of this and disagree with other points. 6. I think this is perfectly fine and realistic to expect when speaking. "Lots" is even a word which is very relevant to statistics. While it could on one hand lead to students checking out, it could also on the other hand reinforce what they are hearing via that repetition, and help to actually clue them in. 8. I think air-quoting AI is 1000% a thing that should be done. It is just literally true that what people call AI is pretty much nothing more than machine learning, it is not even close to intelligent. This also took all of 1 second to do, and has the upside of possibly sparking curiosity in a different subject, I.E. "Why did he put air-quotes around that?" Just my thoughts, without too much thought put into them (ha), take 'em or leave 'em.
20:30 im sure lots of people have said this by now but the standard deviation js the square root of the average *squared* distance from the mean. I.e. the variance is the average squared distance, and the sd is the (positive) square root of variance. When you talk about it in the examples you say squared distance, but the slide text is incorrect. I only say this because you seemed to indicate you noticed something was a little bit wrong with the slide. I think it doesnt really matter in the larger context of the video
Hi Jorbs, thanks for the video. The slide on standard deviation was a bit confusing. My recollection is that SD is the sqrt of the square of the distance between data and the mean.
In the section on elections you imply that the majority of countries (you speak of "a country" without qualifiers) have two-candidate first-past-the-post presidential elections. I don't know if that's true - I'm not an expert on international political systems - but I think it's useful to add qualifiers here; see your other statistic lecture where you show the inherent weaknesses of this system using the Nash equilibrium.
I always felt that people cannot comprehend how much sample size matters. We should be very careful when we explain matters like this because people sometimes do not pay proper attention in order to actually take the information and then apply it to their everyday life correctly(correctly is the key word here) That is why Stephen is an incredible human, he is dissecting concepts that can be overwhelming and explains them in a way that makes them understandable even to the people that are complete foreigners to e.g. statistics Keep doing what you are doing, Jorbs, love you.
on "sampling like this too" slide, you mention bias, but there is a high chance that the students are unfamiliar with the term. Its an important term and should probably be its own point on the slide. I.e. "bias is a term to describe *how* unfairly statistics were done, either in sampling or in analysis". (Also minor pet peeve but ultimately irrelevant given the audience, it should be "garbage in -> garbage out" and "bias in -> bias out", since its not equal, but it infers.)
I'm someone who always felt that statistics and probability where magic. I could intuitively understand algebra and even calculus but probability never made sense to me. It seems like trying to predict the future which is magic. 😅
on the section about "representation" where you bring up the % of Americans (T=339,996,563 Woldmeter), it might be meaningful to include just how much 1% of American population is (3.4 million), and further put into context of the a local place (for example giving this talk in Seattle "excluding 1% of US population is like ignoring the entire population of Seattle 4.6 times over, or excluding the population of Washington state nearly twice" (World population review)
- At ~ 8:30 ish the bulletpoint "eventually the class worked out...." should be on another slide so they don't get spoilers like you did/so they have more time to think of possibilities while you speak. - At 19:35 ish on that slide, instead of showing average height of males and females, have you considered showing the distribution of heights including both sexes, and showing it being less normal? At 21:00 ish when talking about standard deviation, it seems you're struggling to find a way to explain what it is easily because of your point around it being an arbitrarily made up measurement. Consider "Standard deviation is a number that tries to describe the distance that data points are (likely to be ? ) away from the mean, or middle of this curve. There are some different ways you could describe this distance, but this way (show formula) is the useful tool that people have invented to describe this concept and use with one another" Overall your presentation is quite clear and well-delivered. It really stands out that this is Jorbs' YT channel so I and everyone else who watched is someone who already likes watching you do these types of presentations. You know the audience of who's in the class better than I do but I worry that if this was a presentation to HS students generally they would probably get disinterested quick, have not much dragging their attention back, and then have no idea why you're talking about viagra and menstruation. If the class is is nerdy AP stat students who are actively engaged no matter what then they might not be the kind of students who need to be taught what standard deviation is. But you're obviously in a better situation to judge the audience. Overall I was surprised that your stats presentation to HS students involves no jokes/fun references. Consider instead of "real life" examples like "honey last night helped my throat", use examples from gaming or culture or whatever. Anyway I'm super nit-picky but hey you asked for feedback. :)
Fish anecdote is lovely, but I'd advise that you chop it up into two slides, so as not to have students read the slide and mentally check out because they now know the point you're trying to get across.
Would it perhaps be beneficial to split up the anecdotal slide about the fish? I read ahead and did not feel as much of an impact from the reveal of the story. I don't think it changes the information but it might change the impact and attention-grabbing ability of the information, if that seems like it might be important or meaningful at all. P.S. Loved the presentation, this era of content is great for however long you keep tackling it!
Hey Jorbs, hope you're well. I'm going to give some feedback and thoughts - I hope that they're useful to you one way or another (I also hope that I'm coherent enough, because I am certainly not awake yet). As a primer, my background is Physics and Biology, with specialisation for Bioinformatics (using Big Data for biological problems) Above all else!! I think it's a really good presentation and I'm only nit-picking because you asked. Will the average listener get confused by the unbiased nature of the coin, and wonder how you can get results that are skewed one way or the other. i.e An accurate representation of voting preference. - I feel you could tighten up the slide on mean + standard deviation. Flipping the slides back and forth made it more difficult to follow, and following your mouse to understand what you were saying wasn't easy. If you're doing this presentation in person against a big screen your gestures would likely suffice, but perhaps adding one more, annotated graph would help with it. - You touched on this at the end, but you specifically showed one side -> ADHD studied in boys rather than girls. I'll bring up ASD, which is a related topic (but one I am more knowledgeable of) to make my point. We had a good number of seminars, debates, and lectures regarding ASD. With all the statistics we have, we still don't have clear evidence or a good consensus of how to grapple ASD in girls. The raw data shows fewer diagnoses in girls than boys, but we still have no clear indication of what the ratio of male/female ASD is, and whether/ how we're underserving girls with ASD (as compared to boys). -> I suppose I am saying "Women's health can be systematically under-prioritized..." *is* leading. Whether it is correct to lead, I'm ultimately unknowable with our current datasets. I suspect you're cool with this, because I know you'd rather err on the side of more mental health study than not, but I figured I'd flag it regardless. _____________________ Line break because now I'm going to talk a bit more about stuff that might be out of the scope of this. - As a trade for looking up statistical distributions on Wikipedia, I would highly recommend looking through the Wikipedia rabbit-hole of Quantative Genetics, Sir Ronald Fischer (Biologist & Statistician), Genetic Variance, Heritability. -> I think it holds the chance of greatly enriching your knowledge pool (although I know I'm ridiculously biased) - Your last statement "The power of statistics..." does hold true in many respects and should be how most people conceive of the field. *However*, this completely discounts the field of Bioinformatics; a growing field of study that is already providing tangible benefits to people globally. - The type of data you are talking about - voting intentions - has a remarkably even chance of happening, either way. *Even if* a country got so skewed as to be a 95/5 split, similar methodologies would work. - The issue arises when you look at one of two types of events: a) Small effect size -> i.e It's *definitely* doing something, but only very slightly b) Rare events -> i.e. This thing causes x10 risk of disease but is a mutation in only 1/10000 people In this field, we have the ability to estimate with a reasonable level of accuracy (twin studies and the like), how much additive genetics contributes to height. "Scientists estimate that about 80 percent of an individual’s height is determined by the DNA sequence variations they have inherited", this suggests that we could, theoretically, add up the effect sizes of all the genetic variants associated with height that someone has and determine their height within 80%. From PMC2955183, they found 180 loci, "Our data explain ∼10% of the phenotypic variation in height, and we estimate that unidentified common variants of similar effect sizes would increase this figure to ∼16% of phenotypic variation (∼20% of heritable variation)." 20%!!!! Out of 80% total!!! And to find even this amount they had almost 200,000 participants. You might shrug this off, 'who cares about understanding where height comes from', but this is only an easy example. -> Here's the main takeaway. Over the next century and certainly within our lifetimes we will work towards a more personalised medicine. There are hundreds of tiny changes in our genetic code, which might each act as drivers for cancers, dementia, Parkinson's. We will eventually reach a point (bear in mind for me, NHS) that we will gain an unprecedented level of predictive accuracy, aimed at ACTIONABLY helping best treat and even prevent some of the worst diseases known to humanity. Think about how impactful the identification of the BRCA gene has been, and how many lives have been saved. -> To get that kind of predictive power we need LARGE datasets. We need well annotated, well collected datasets. We can use the power of statistics to be able to analyse huge amounts of data; to accurately and actionably analyse huge populations to find, a) small but additive, as well as b) rare occurrences that when collated can have massive effects.
Hey jorbs, As a HS math teacher myself, it's generally a bad idea to have so many words on a page that all pop up at once. Consider, if you do this again, making the information pop up point by point. This is especially important when you ask a question and answer it on the same slide, like the fish example--if they know the answer already, they will zone out, as they believe math is about The Answer.
the 70.9 inches being the mean and whether it is too high is amusing to me, as if we talk the frequency, it is a high frequency of instances of people who are that tall, and this is kind of the point of the mean, but if we are taking what is being measured, the literal height, then it becomes purely subjective as to whether 180cm is too high
Hey, you should talk about Krippendorf's Alpha and why it sucks balls because it tries to measure four different things in a single number making it practically meaningless for any real work.
@5:00 I think it would be nice to not show the 3rd point at the beginning of the slide, rather, only show it when you reveal what the actual problem was. That makes the students wonder and be more invested in knowing what the deal was with the anecdote. Posting this since you specifically asked for feedback. Hope you it goes well!
The women stuff is inherently more complicated tbh. CT scans are a big one If hospitals gave them to women at the same rate as men some of those women will sue hospitals because they had a fetus that was killed during the process
On the topic of feedback: I noticed that your slides tend to have a lot of words, and they're presented in full, which can lead to people reading ahead and that can cause unintended consequences like getting bored or tuning out, and I tend to prefer one revealing the summary on the slide after I've said the thing I want to say.
It just occured to me that i never saw anyone elses reaction to your other stats video or the lifecoach video. Lol. They could be seething mad and i have mo idea. I dont watch other streamers
So you're saying it's not about the size but how you use it? Funny how this kind of deal keeps showing up, it's like we're obsessed with "how much" rather than "how good" :D
I think it's a good idea to avoid involvement with drama.. to approach all people (whether they are friendly or antagonistic) with openness and kindness and humility. Lead by example, and don't lose sleep over vague squabbles. Ultimately, the trials and obstacles that we face today will rapidly fade to insignificance. But if we can be kind and thoughtful to each other, we can plant those seeds to form a better world.
If you read this comment, you will become a weeb. If you take a look at the past ten videos, you will see, that you have become a weeb (idk exactly) about five times. This would indicate a rate of about fifty percent. However this observation would change, if you look at the past one hundred videos, where you would gain a different dataset. However the question is, whether or not the sample size is even relevant for this statistic. There is also the question of whether or not this is even relevant to turn into a statistic as i know, that one hundred percent of people reading this are weebs.
Jorbs, as you asked for feedback.
When you discuss your uncles fish story, split the slide after you describe the problem, then ask the class you are talking to for some ideas for what could be going wrong, before revealing the answer. The students will read to the end of rhe slide before you get through explaining what was happening.
Great stuff as always
I was going to comment this same thing, yeah. Even if you don't ask the class and just keep the presentation the same other than hiding parts of the slide, I feel that would make the presentation feel better. Now time was spent on talking about the mystery while the answer was visible to everyone.
Agree with OP, and I’ll add that the slides in general are far too wordy. A true presentation to an audience should be limited to speaking prompts, images, and/or core concept emphasis.
Jorb’s standard presentation style (few slides, many words) is good/fine for engaging an already engaged audience. We all choose to be here. High school students are not that audience, haha.
I like the TED talk “death by PowerPoint” that goes through some good examples of how to make a good PRESENTATION.
Yes 100% this. Showing information only when the students are prepared for it by the context of the rest of the lecture will also make it easier to follow the flow of logic, because it will be clearer which point you're discussing right now. As a student, it's a very different effect to be hit with a wall of text vs being drip fed the text one point at a time. Even though it's the same course content, I always find it easier to absorb the information in a math-focused lecture when the professor writes the notes on the board as we go rather than presenting slides that already have the information written out beforehand, even if it has all the requisite steps. Anyway, keeping what's on the slide to just a short prompt or the general idea of what you're saying will help because the student can read the bottom most point and then get back on track with the context more quickly than with full sentences. If you still want the full slides available to the students, you can provide these slides, or have the points in the presenter notes, for the teacher to provide to the students if they want to look back on them with more context.
totally agree
jorbs, I just want to let you know that I am thoroughly enjoying this statistics mini series despite the situation it arose out of. As someone majoring in data science, one of the things that I get nerdy about is what exactly statistics means, as it is often misunderstood. Thank you for these great lessons, hope you’re doing well.
I am also enjoying this dive into statistics. It is helpful especially with the way statistics are misused and poorly acquired
As someone who is pretty shit at stats (I failed out of AP stats in highschool despite my best efforts lol) and admittedly doesn't really give much of a shit about statistics in general, these videos have been really fun and are great content to put on while I'm at work. It's something good to pay attention to that puts me in a productive mindset honestly
Seconding the last part as well, hope all is well and thanks for the neat content despite the original intent
I stand atop the summit of the bell curve, hear my statistically average roar and tremble!
your slides are very static: all of the bullet points are already on the slide as soon as it comes up. for example the the fish story (Slide 3) the "answer" is already on the slide, meaning that the students could just read ahead of your story, and not directly pay attention to the lecture itself.
I know statistics can be dry, and I also know that you can have a fair amount of energy.... "sometimes"
maybe set some of the bullet points to be on cue, and then for that slide ask the class if anyone has ideas for "why was this students data 'wrong'?" and then when you reveal the bullet point "The data wasn't wrong, everyone else was"
Yeah this is the thing I would suggest if possible, I ususlaly read the entire slide before he gets past the first bullet point.
You can relate the "update our understanding slide" to the coin landing on its side, i.e. lets say you thought it was 100% false that a coin can land on its edge, but seeing it happen one time, flips that to 100% true that a coin can land on its edge.
Furthermore, if you see the coin *almost* land its edge but tip over, that might change your understanding from 100% false to maybe 90% false.
Everyone loves callbacks
yes! callbacks, repitions, and throughlines are very helpful
22:34 So the thing with standard deviation is, while it is possible to calculate the average absolute deviation (that is, the expected value of |x-u| where x is the actual value and u is the mean), it's often unfeasible to do so for more cumbersome distributions such as the normal distribution. It's actually a better measure than standard deviation due to being more robust (less susceptible to infrequent but extreme outliers), just algebraically harder to calculate.
It really depends on what you want to quantify, the standard deviation is especially important to how it interacts with the central limit theorem. It's not just easier to calculate.
It is indeed often the most informative measure of how much uncertainty a random variable has. You just have to interpret is correctly.
E[|X-EX|] is by the way often feasible, in particular for the normal distribution. See en.wikipedia.org/wiki/Half-normal_distribution in this case the difference this just a constant factor between both measures.
What I want to say, it's not arbitrary. Unfortunately it needs a lot of math to see why it is. Maybe you can get 3b1b to make us an more intuitive explanation ;)
I remember looking the average absolute deviation up in uni, because I was really unsure about standard deviation being the measure we use
There were a few reasons, and another one of them was that if you have to mess around with differentiating and whatnot for some reason, the absolute value term can get really inconvenient (as a graph of absolute value will have a kink in it, making the function non-differentiable at that point)
Chiming in to reinforce the point about splitting the conclusion to the fish story. I'm sure I've heard you talk about this on stream/in another video but forgotten the answer, and being able to read ahead to find out what it is deflated a lot of my engagement with the potential causes. I think I would have been absolutely rapt if i'd just read the 'N=1 was correct' part but not been told why until you got there.
love the stats content btw.
Hey Jorbs, since you explicitly said you're looking for feedback:
I thought this was a really interesting presentation when it stayed more on the topic of sample sizes. The insights of how we can't just ignore anecdote when shaping our understanding, and how big data's vague promise of "more data" can be used to paper over shoddy collection methodology, were extremely valuable.
In the 30 to 3000 range, I think the presentation wandered a bit, and you spent most of the time talking about what you can *do* with statistics (normal and binomial distributions, etc) rather than *why* 30-3000 is the normal statistical range. I think it would be pretty powerful to touch on the mathematics and show exactly how scaling up the sample size affects the certainty of statistical models and why you don't even need to approach a census in order to be quite confident in your results. This would then segue into your point on big data and how a census can actually be *worse* than a small SRS because any holes in the data become very systemic.
Semi-unrelatedly, the discussion of how a normal distribution is basically just secretly the noise of a bunch of binomial distributions was interesting, but a bit of an aside and didn't really have the time to be fleshed out into its own argument. I'm not sure I quite follow it either even after thinking pretty hard about it. For example, one of the biggest factors for adult male height is "How tall were your parents?" and I don't see a way to map that information to a yes/no question that isn't extremely tortured. It's an interesting thought but left me more confused than enlightened--if it's meant to be an oversimplifcation then I think it could benefit from shouting that out explicitly because it wasn't clear to me, and if it isn't an oversimplification then maybe leave it out until you can dive deeper into it some other time.
Overall this was an interesting talk and I enjoyed listening. I'm sure it'll go great. Good luck cramming if you do try to incorporate feedback tomorrow morning :)
id add more catgirls to the slides they help me pay attention
A tip: Keep the amount of text on each slide to a minimum. The more text, the more people's attention will split between reading and listening. It can also work even if you have a fair amount of text to not show all of it at once, but rather reveal bullet points through several steps.
Probably too late to include in the presentation by now, but one thing to consider on the inclusion of women in medical trials is that historically periods have been seen as a complicating, difficult to control factor, and so that had also played some role in their being left out of some trials. This means that around half the population has been under represented in studies due to a regular, recurring fact of life that made the math more difficult. And that matters because due to differences in part related to how sexed bodies absorb and process chemicals that impact the endocrin system means that women are regularly given higher than necessary dosages of medications, the classic example of which is Ambien, the sleeping drug that was linked to an increased risk of car accidents following ambien usage, because it would linger longer in women than in the men that made up the test population prior to FDA approval. Since 2013 the dosage approved for women has been halved, but ambien had been on the market since 1992
Hi Jorbs. I teach college math, so I have advice I can give. Please read this when you are in a good headspace for feedback since I always tend to come across as highly critical when I really mean the best. Also as an educator, my apologies for “grading” with this advice way later that you probably wanted it; hopefully it can be taken into account for a “next time.”
Overall your presentation is very good. You have many ideas that you flesh out very well. You are passionate, and it’s clear that you know what you are talking about. Mathematically, it all checks out.
For a *UA-cam* audience, your presentation is perfect. For an *academic* audience, there’s a handful of things you should change.
First, your slides need some help. For a UA-cam audience, you have nothing to change. I can pause/rewind you at will. I can read your slides at will. For UA-cam, *change nothing about what you are doing unless you want to.* For an academic talk, I cannot pause/rewind you. If you write paragraphs on slides (with the exception of quotes that you should read aloud), you will split your audience’s attention. This diminishes the value of both your spoken words and your slides. You also should making “building” slides for a talk like this; i.e., you can keep the same “big topic” slide, but bring up the bullet points one-by-one. Bringing up all points at once can be confusing because your slide isn’t “tracking with” your current spoken topic.
What I would suggest to you is to imagine what you would write on a chalkboard as you were giving this lecture without a PowerPoint. What types of main points would you want to emphasize by writing them? What points would you omit or shorten? Which points are worth your time to write? Your slides should only be an *enhancement* of the things you would otherwise be writing. Also remember in this setting you have note-takers, so you need to make points on your slides note-taker-friendly.
Here’s an example of what I would do to change one of your slides:
Slide title: Some Observations With N=1 Are Extremely Powerful
•Example: Uncle’s Fish Observations
•N students, N-1 had matching data, 1 had very different data
•What explanations are there for mismatching data?
•Made up data? NO
•Recording the data wrong? NO
•Different time of day? NO
•Actual answer: Her height affected fish behavior!
•Takeaway: The N=1 data was right, so all the other data was wrong.
When presenting this slide, you should only bring up one bullet point at a time, maybe bringing up some of the sub-points together. The rest of the time, you are talking about each point. Describe (with speech) the fish experiment and the problem. Describe (with speech) why the common explanations don’t cover this situation. Describe (with speech) why the actual answer is surprising and how it leads to your main takeaway. Notice I also omitted the coin part of the slide; you made a brilliant point with the fish story, and the coin was taking away from that. If you want to do the coin landing on its edge example, work it in somewhere else.
The second major thing I want to bring up is knowing your audience. For *UA-cam*, you nailed it. You are speaking to an audience where you are trying to get to a “common denominator,” so you have to explain things like binomial distribution, mean, standard deviation, etc. For an *AP Stats class*, you went over a bit too much. They likely are already familiar with binomial distribution, so a quick review is good enough. For mean and standard deviation, you don’t need much of a review for an AP Stats audience. Taking time to explain basic concepts to an audience that doesn’t need it will take away valuable time and attention from the points you want to make.
On the note of audiences, you have the unfortunate conundrum of a high school audience. While I like your examples about Viagra and believe that they should be appropriate for your audience, you have to be careful. Whether you or I or anyone else likes it or not, it is best to not step on the toes of any parents/guardians or school admin with examples that get into topics of sexual health if you are not teaching a health class. I would also hesitate to bring up any examples that start to “get political.” If this were for a college audience, then I would not be bringing this up at all. For future reference, just be *very* careful with the real-world examples that you choose to talk about if your audience is K-12.
I hope this is helpful and not overly critical; I want to help you give the best in-person lectures that you can. You are very good at communicating your unique and insightful ideas. I really appreciate all of your UA-cam lectures and want to see them continue!
Good job on describing stats. The first example you raised is called bayesian logic where you update your priors based on new information. if your current knowledge is that green doesn't exist and then you experience green then its a strong indicator that green exists. on the other hand if you experience something yellow then you might start to think green exists. Your prior still has a large sway over your view but it has been updated with the possibility that green does exist.
Don't put the answer to the fish story on the slide! Just tell the story. The kids will be much more engaged if they can't read ahead.
I was also given that argument for why standard deviation is a squared term. we don't want the negatives. Also if i am remembering correctly, on the normal model, one standard deviation is at the two points of inflection of the curve.
You think you're just enjoying your Spire content but then a couple of hours pass and you look to your left, take a peak at your right, everyone is taking notes.
You look at the board as the tutor explains the differences between the distributions displayed on the board.
You begin to question why you took statistics in the first place.
As an undergrad, I spent a lot of time learning the mechanics of stats: how to propagate errors and do the different tests and so on. And because I did all that math, I got tricked into thinking that doing stats was like casting a magic spell. If it had error bars or a standard deviation, it was to some extent Truth.
So when I got deeper into academia and I had to apply this stuff in a situation where the answers were unknown, I learned... well, that stuff isn't so important. Like, you publish your standard deviations and draw error bars on graphs, but in most situations, you can never be certain that you've sampled things correctly. You can't control all your variables. You can't even know if you got the important ones that would change your distribution. Did you even ask the right question/a sensible question to start with? In fact, you probably didn't. I think that's a good lesson, and you had some nice discussion in here. It's worth emphasizing how devilishly hard it really is.
One thing I don't think you mentioned, but which might warrant a few words, is the interpretation of a statistic. You've differentiated between descriptive and inferential stats in other videos, and the assorted fallacies that come from not understanding the difference. Since I observe (and therefore I conclude confidently that we all observe) those fallacies on a daily basis, I suggest that it might be nice to tell those students, hey, as you're going through this material, keep an eye out for this. Don't trick yourself into thinking these numbers mean more than they do, and look for situations where others are doing that.
Thank you for another excellent video. This and your book are getting me through night shift tonight. ❤
Hi Jorbs! Thank you for working so hard to make statistics accessible. It has genuinely had a profound impact on my life!
Starting when I first watched your video "How I think about strategy games" a couple years ago, I really developed a passion and fascination with statistics. That video in particular marked the first time I really came to understand how little I understand, and that made it possible for me to truly appreciate the way you think about and talk about statistics and probability, and to challenge and outgrow my own limited view of it. It flipped a switch of sorts, and opened some paths for me to expand my boundaries!
So it might seem dramatic to say, but the results of this paradigm shift and your content have had a profound effect on me! At first it was just how I played StS, then it was other games, then it was anything involving probability, and now I can confidently say that it has changed how I see the world. Its not an exaggeration to say that learning to understand what stats mean in the real world has broadened my perspective in nearly all topics. It has even led to some really awesome creative breakthroughs with challenges at my work as a software developer!
Anyways, all that to say, I am extremely grateful, and wanted to share this little way that you've made a big impact. You'll always have a fan in me! :)
Couldn't have said it better, I've had the exact same experience.
I know I'm in for fun relaxing ride when Slideshow Jorbs uploads.
new jorbs variant identified?
I think there is an error in the slide at 22:35. Standard deviation is the square root of the average variance, and variance is the distance from the mean SQUARED. So if you don't want to use the word "variance" in the slide, then it should read:
Standard deviation is the square root of the average squared distance between the data points sampled from the distribution and the distribution's mean.
Along with the tips that everyone else had about managing the flow of information and avoiding spoiling the next points early, I would reccommend thinking about how you might use the whiteboard/chalkboard/smartboard in the classroom to supplement visuals where possible. When pointing to the graphs in the mean/varience sections, it would help me personally as a student if you marked off visually the information such as where the mean is and roughly how far the deviation goes out from the mean. I know that these are AP stats students, so they're pretty familiar with variance and mean, but I think it would help.
In your section on mean and varience, you had to refer a lot to a previous slide, and did some pointing at the graph. I think it might be helpful to have a visual representation of mean and standard deviation on the mean and variance slide
It's unsportsmanlike to let Twitch chat help you with game strategy but it is completely fair game to let UA-cam comments help you with your homework.
This sounds sarcastic but it is simply a true statement that I agree with which is funny.
my biggest note for this presentation is that once you start talking about the power of statistics to do good and evil you can relate that back to the metaphor of coin flipping, society making people "flip the coin wrong" the n=1 case was fascinating, I've never thought of it that way
Hey jorbs! Not sure if you want to include this tie-in, but many "machine learning" models are effectively sampling human behavior and then trying to replicate that outcome, with similar caveats. So if you want a machine model to rate a writing sample, you have the machine sample a lot of human ratings of writing samples - and that also has the "bias in, bias out" effect.
Minor nitpick. It might be more clear to use "dispersion" when referring to the general concept of distributional width than variance which might imply a specific meaning (i.e. the square of the standard deviation).
Nice Spireside Jorbs Chat on my PC with Mega Crit's new Spire Yule Log on my tv
EDIT: Dreams shattered. I can't believe Jorbs said I'm too tall to be a marine biologist.
I really enjoy that you're humanizing statistics and explaining some of the reasoning behind it (especially explaining that standard deviation is arbitrary)! One thought on the "Sampling is like this, too" slide: don't reveal that the country is 55/45 at the start. If you're at a polling company, your job is to figure out whether the population is likely to be 60/40 or 50/50 based on a sample. So, saying "165 people polled said they would vote for candidate A and 135 said candidate B, which might mean that the population is 55/45... or it might not" might make your explanation match the actual statistical work being done, and it might also be a more natural lead-in to your comments about bias (165 vs 135 in the sample doesn't imply 55/45 in the population).
Ive heard that theres some denomination of coin somewhere that actually can land on its side, but I dont remember which one it is. It makes sense that it would depend on the coin.
hi jorbs! you have done enough of these mini-lectures that you should make a playlist with them
Jorbs inspiring the next generation of statisticians one UA-cam lecture at a time!
I echo most people’s comments about not showing the answer to the uncle anecdote right away. Great stuff!
Thanks Jorbs! I love these videos, I dont watch your slay the spire content as much anymore but I love this style. You have a great way of communicating math and how to incorporate into decisions. I've genuinely used some of your advice in my job and I would like you to know that it led to better maritime safety.
At first it wasn't clear to me who was observing which fish where. The immediate and obvious false explanation was that the outlier data were from a totally different species or lake full of fish. The fact that they were all in a fish tank in a university was either unclear, or I snoozed through that somehow and inferred it later as the story went on.
Gosh. Just this morning I was googling something about sample size evaluation (I'm not great with statistics - in case this sounds funny). And just a few hours later UA-cam gives me this. I thought that it's my search history affecting my feed, but nope - it's Jorbs posting a video.
Don't tell me it's some sot of confirmation bias again.
I might be a bit late to the party here, but one thing I would add is that you have to make sure what the data you actually want is (a.k.a. your population). So before you even think about how to draw from your population you have to define it well. Easy example: all American residents eligible to vote. Asking a bunch of kindergarteners about how they would vote is not skewing the coin toss. It's not tossing at all. Not doing this is very problematic and leads to shit results (and it's way too prevalent in my field).
Good guy Jorbs giving free statistics course for us youtube people
It'd be great to ask the class if they have any questions by the end of the presentation!
Also, I saw another comment say this and I thought it was a great idea, so I'll reinforce it here: it'd be really cool when you're talking about your uncle's story to ask the class what they think the answer to the outlier is! It's an effective and simple way to make the class more involved on the lecture
Just want to say thanks for these informational videos on statistics. I took some (fairly basic) stats classes many years ago and your videos been a great refresher.
Great video! I often think in ~all of the work in statistics is getting the right sample, the p-values and inferences are often just formulas you copy once you feel your sample is good.
Since you asked for feedback, the use of a 100% or 0% probability is colloquially okay, but I try to avoid it since mathematically it makes little sense. One can’t do a Bayesian update from 100% to 95%, and someone who believes a 100% probability would be willing to take a 1 trillion-to-1 bet on the event, which is almost never the case
I'll write this as I watch along, hope it's helpful feedback.
1. Cut the tangents, teenagers have short attention spans, this is a chance to lose it. Parachutes, coins dropped into mud, wondering about the accuracy of the average man's height, I don't feel these details are relevant to statistics nor add good flavor to the presentation.
2. A good number of students will read your slides instead of listening to you. I would suggest not showing the entire slide at once, but introducing each bullet as you start talking about it instead.
3. 'We Can Update Our Understanding Of the World Constantly!' is a 10/10 slide, especially for a highschooler to hear.
4. Have a quick look at the wiki page for distributions so you can speak about it with more certainty. The audience will be more engaged if they feel the speaker really knows their stuff, and this is an easy way to score points, or lose them.
5. I would not spend as much time discussing the intuition of how many heads we expect in 10, 100, 1000 coin tosses, unless you explicitly want to talk about the reduced variance with bigger n. I actually would strongly suggest taking this chance to do so, if you can prepare this in time. Upon hearing the title, I would have expected some discussion of how the size of a sample changes it's behavior.
6. Limit the number of times you repeat "lots", especially after the first time I would only say it once. The audience gets the point.
7. On standard deviation, I would suggest giving the intuitive explanation of the graph before bringing in any rigorous details or math, and definitely before reading the bullet.
8. The air quotes "artificial intelligence" went over my head, I imagine the same will be true of much of the audience. I think this bit should be better explained, or else not done.
Overall it's a great presentation, best of luck.
I agree with some of this and disagree with other points.
6. I think this is perfectly fine and realistic to expect when speaking. "Lots" is even a word which is very relevant to statistics. While it could on one hand lead to students checking out, it could also on the other hand reinforce what they are hearing via that repetition, and help to actually clue them in.
8. I think air-quoting AI is 1000% a thing that should be done. It is just literally true that what people call AI is pretty much nothing more than machine learning, it is not even close to intelligent. This also took all of 1 second to do, and has the upside of possibly sparking curiosity in a different subject, I.E. "Why did he put air-quotes around that?"
Just my thoughts, without too much thought put into them (ha), take 'em or leave 'em.
waking up and becoming enthusiastic about watching jorbs videos about statistics... is there something wrong with my life?
20:30 im sure lots of people have said this by now but the standard deviation js the square root of the average *squared* distance from the mean. I.e. the variance is the average squared distance, and the sd is the (positive) square root of variance. When you talk about it in the examples you say squared distance, but the slide text is incorrect.
I only say this because you seemed to indicate you noticed something was a little bit wrong with the slide. I think it doesnt really matter in the larger context of the video
i am kind of loving this powerpoint presentation arc!
Hi Jorbs, thanks for the video. The slide on standard deviation was a bit confusing. My recollection is that SD is the sqrt of the square of the distance between data and the mean.
In the section on elections you imply that the majority of countries (you speak of "a country" without qualifiers) have two-candidate first-past-the-post presidential elections. I don't know if that's true - I'm not an expert on international political systems - but I think it's useful to add qualifiers here; see your other statistic lecture where you show the inherent weaknesses of this system using the Nash equilibrium.
I always felt that people cannot comprehend how much sample size matters.
We should be very careful when we explain matters like this because people sometimes do not pay proper attention in order to actually take the information and then apply it to their everyday life correctly(correctly is the key word here)
That is why Stephen is an incredible human, he is dissecting concepts that can be overwhelming and explains them in a way that makes them understandable even to the people that are complete foreigners to e.g. statistics
Keep doing what you are doing, Jorbs, love you.
on "sampling like this too" slide, you mention bias, but there is a high chance that the students are unfamiliar with the term. Its an important term and should probably be its own point on the slide. I.e. "bias is a term to describe *how* unfairly statistics were done, either in sampling or in analysis".
(Also minor pet peeve but ultimately irrelevant given the audience, it should be "garbage in -> garbage out" and "bias in -> bias out", since its not equal, but it infers.)
MORE STATS TEACHER JORBS!!! we eatin good tonight :)
Good night jorbs, I watch this in the morning on the way to work, so thank you :)
I'm someone who always felt that statistics and probability where magic. I could intuitively understand algebra and even calculus but probability never made sense to me. It seems like trying to predict the future which is magic. 😅
on the section about "representation" where you bring up the % of Americans (T=339,996,563 Woldmeter), it might be meaningful to include just how much 1% of American population is (3.4 million), and further put into context of the a local place
(for example giving this talk in Seattle "excluding 1% of US population is like ignoring the entire population of Seattle 4.6 times over, or excluding the population of Washington state nearly twice" (World population review)
Oh snap! New Jorbs talking to chat video just dropped! 😮
- At ~ 8:30 ish the bulletpoint "eventually the class worked out...." should be on another slide so they don't get spoilers like you did/so they have more time to think of possibilities while you speak.
- At 19:35 ish on that slide, instead of showing average height of males and females, have you considered showing the distribution of heights including both sexes, and showing it being less normal?
At 21:00 ish when talking about standard deviation, it seems you're struggling to find a way to explain what it is easily because of your point around it being an arbitrarily made up measurement. Consider "Standard deviation is a number that tries to describe the distance that data points are (likely to be ? ) away from the mean, or middle of this curve. There are some different ways you could describe this distance, but this way (show formula) is the useful tool that people have invented to describe this concept and use with one another"
Overall your presentation is quite clear and well-delivered. It really stands out that this is Jorbs' YT channel so I and everyone else who watched is someone who already likes watching you do these types of presentations. You know the audience of who's in the class better than I do but I worry that if this was a presentation to HS students generally they would probably get disinterested quick, have not much dragging their attention back, and then have no idea why you're talking about viagra and menstruation.
If the class is is nerdy AP stat students who are actively engaged no matter what then they might not be the kind of students who need to be taught what standard deviation is. But you're obviously in a better situation to judge the audience. Overall I was surprised that your stats presentation to HS students involves no jokes/fun references. Consider instead of "real life" examples like "honey last night helped my throat", use examples from gaming or culture or whatever. Anyway I'm super nit-picky but hey you asked for feedback. :)
People need more information about sample sizes so badly.
Going back and forth between slay the spire content and stats content is lovely
A lesson about statistics, exactly what I want from my favourite STS player.
I failed my statistics course is university, but I'll try my best.
Fish anecdote is lovely, but I'd advise that you chop it up into two slides, so as not to have students read the slide and mentally check out because they now know the point you're trying to get across.
Would it perhaps be beneficial to split up the anecdotal slide about the fish? I read ahead and did not feel as much of an impact from the reveal of the story.
I don't think it changes the information but it might change the impact and attention-grabbing ability of the information, if that seems like it might be important or meaningful at all.
P.S. Loved the presentation, this era of content is great for however long you keep tackling it!
We have a playlist with all the videos in this format?
Statistics era Jorbs is very fun
Hey Jorbs, hope you're well. I'm going to give some feedback and thoughts - I hope that they're useful to you one way or another (I also hope that I'm coherent enough, because I am certainly not awake yet). As a primer, my background is Physics and Biology, with specialisation for Bioinformatics (using Big Data for biological problems)
Above all else!! I think it's a really good presentation and I'm only nit-picking because you asked. Will the average listener get confused by the unbiased nature of the coin, and wonder how you can get results that are skewed one way or the other. i.e An accurate representation of voting preference.
- I feel you could tighten up the slide on mean + standard deviation. Flipping the slides back and forth made it more difficult to follow, and following your mouse to understand what you were saying wasn't easy. If you're doing this presentation in person against a big screen your gestures would likely suffice, but perhaps adding one more, annotated graph would help with it.
- You touched on this at the end, but you specifically showed one side -> ADHD studied in boys rather than girls. I'll bring up ASD, which is a related topic (but one I am more knowledgeable of) to make my point. We had a good number of seminars, debates, and lectures regarding ASD. With all the statistics we have, we still don't have clear evidence or a good consensus of how to grapple ASD in girls. The raw data shows fewer diagnoses in girls than boys, but we still have no clear indication of what the ratio of male/female ASD is, and whether/ how we're underserving girls with ASD (as compared to boys).
-> I suppose I am saying "Women's health can be systematically under-prioritized..." *is* leading. Whether it is correct to lead, I'm ultimately unknowable with our current datasets. I suspect you're cool with this, because I know you'd rather err on the side of more mental health study than not, but I figured I'd flag it regardless.
_____________________
Line break because now I'm going to talk a bit more about stuff that might be out of the scope of this.
- As a trade for looking up statistical distributions on Wikipedia, I would highly recommend looking through the Wikipedia rabbit-hole of Quantative Genetics, Sir Ronald Fischer (Biologist & Statistician), Genetic Variance, Heritability.
-> I think it holds the chance of greatly enriching your knowledge pool (although I know I'm ridiculously biased)
- Your last statement "The power of statistics..." does hold true in many respects and should be how most people conceive of the field. *However*, this completely discounts the field of Bioinformatics; a growing field of study that is already providing tangible benefits to people globally.
- The type of data you are talking about - voting intentions - has a remarkably even chance of happening, either way. *Even if* a country got so skewed as to be a 95/5 split, similar methodologies would work.
- The issue arises when you look at one of two types of events:
a) Small effect size
-> i.e It's *definitely* doing something, but only very slightly
b) Rare events
-> i.e. This thing causes x10 risk of disease but is a mutation in only 1/10000 people
In this field, we have the ability to estimate with a reasonable level of accuracy (twin studies and the like), how much additive genetics contributes to height.
"Scientists estimate that about 80 percent of an individual’s height is determined by the DNA sequence variations they have inherited", this suggests that we could, theoretically, add up the effect sizes of all the genetic variants associated with height that someone has and determine their height within 80%.
From PMC2955183, they found 180 loci,
"Our data explain ∼10% of the phenotypic variation in height, and we estimate that unidentified common variants of similar effect sizes would increase this figure to ∼16% of phenotypic variation (∼20% of heritable variation)."
20%!!!! Out of 80% total!!! And to find even this amount they had almost 200,000 participants. You might shrug this off, 'who cares about understanding where height comes from', but this is only an easy example.
-> Here's the main takeaway. Over the next century and certainly within our lifetimes we will work towards a more personalised medicine. There are hundreds of tiny changes in our genetic code, which might each act as drivers for cancers, dementia, Parkinson's. We will eventually reach a point (bear in mind for me, NHS) that we will gain an unprecedented level of predictive accuracy, aimed at ACTIONABLY helping best treat and even prevent some of the worst diseases known to humanity. Think about how impactful the identification of the BRCA gene has been, and how many lives have been saved.
-> To get that kind of predictive power we need LARGE datasets. We need well annotated, well collected datasets.
We can use the power of statistics to be able to analyse huge amounts of data; to accurately and actionably analyse huge populations to find, a) small but additive, as well as b) rare occurrences that when collated can have massive effects.
babe, wake up! new jorbs presentation is up :')
great as always
Hey jorbs,
As a HS math teacher myself, it's generally a bad idea to have so many words on a page that all pop up at once. Consider, if you do this again, making the information pop up point by point. This is especially important when you ask a question and answer it on the same slide, like the fish example--if they know the answer already, they will zone out, as they believe math is about The Answer.
more anecdotes but i have personally seen coin flips land on their edge a couple times in my life that i can recall
the 70.9 inches being the mean and whether it is too high is amusing to me, as if we talk the frequency, it is a high frequency of instances of people who are that tall, and this is kind of the point of the mean, but if we are taking what is being measured, the literal height, then it becomes purely subjective as to whether 180cm is too high
Hey, you should talk about Krippendorf's Alpha and why it sucks balls because it tries to measure four different things in a single number making it practically meaningless for any real work.
@5:00
I think it would be nice to not show the 3rd point at the beginning of the slide, rather, only show it when you reveal what the actual problem was. That makes the students wonder and be more invested in knowing what the deal was with the anecdote.
Posting this since you specifically asked for feedback. Hope you it goes well!
Did you do the thing where you first person the story to make it easier to tell (the scared fish) or did I originally hear that anecdote from you?
it is from my uncle.
The women stuff is inherently more complicated tbh.
CT scans are a big one
If hospitals gave them to women at the same rate as men some of those women will sue hospitals because they had a fetus that was killed during the process
this is about Kramnik, isn't it?
The smallest sample size is N=1? I think you forgot about N=0: I made it up.
Jorbs and Aella team up when?
honey, wake up, jorbs uploaded another chill video
On the topic of feedback:
I noticed that your slides tend to have a lot of words, and they're presented in full, which can lead to people reading ahead and that can cause unintended consequences like getting bored or tuning out, and I tend to prefer one revealing the summary on the slide after I've said the thing I want to say.
It just occured to me that i never saw anyone elses reaction to your other stats video or the lifecoach video. Lol. They could be seething mad and i have mo idea. I dont watch other streamers
So you're saying it's not about the size but how you use it? Funny how this kind of deal keeps showing up, it's like we're obsessed with "how much" rather than "how good" :D
I think it's a good idea to avoid involvement with drama.. to approach all people (whether they are friendly or antagonistic) with openness and kindness and humility. Lead by example, and don't lose sleep over vague squabbles. Ultimately, the trials and obstacles that we face today will rapidly fade to insignificance. But if we can be kind and thoughtful to each other, we can plant those seeds to form a better world.
I am just here for streamer v streamer content.
If you read this comment, you will become a weeb.
If you take a look at the past ten videos, you will see, that you have become a weeb (idk exactly) about five times. This would indicate a rate of about fifty percent. However this observation would change, if you look at the past one hundred videos, where you would gain a different dataset. However the question is, whether or not the sample size is even relevant for this statistic. There is also the question of whether or not this is even relevant to turn into a statistic as i know, that one hundred percent of people reading this are weebs.
the goat
YEEEEEES STAAAAAAAAAAAAAATS
grah i love it@
My favorite answer to the question “how would you improve this study?” In science class was always: increase the sample size lol
Let's goo
You could be more clear by becoming translucent. Or a weeb.