Joshua Starmer on StatQuest, Storytelling, Next-Gen Learning & His Iconic BAMs

Chapter 1 — 00:00 | Introduction & The Origin Story of StatQuest

Josh recounts how StatQuest began not as a YouTube ambition but as a practical solution to a lab problem — getting coworkers to understand data analysis. Nine views and two subscribers in year one felt like a roaring success.

Raja Iqbal: Hello everyone, I’m Raja Iqbal, I’m your host. Today my guest is Joshua Starmer from StatQuest. Josh, great to have you.

Joshua Starmer: Hello, it’s great to be here. Thanks very much for having me, Raja.

Raja Iqbal: So Josh, I know you get this question a lot — how did this all get started? StatQuest is one of the most renowned channels out there. Anyone learning AI or stats fundamentals goes to StatQuest to brush up.

Joshua Starmer: Yeah, it started a long time ago. I used to work in a genetics laboratory at the University of North Carolina. I was in charge of a lot of the data analysis, and I wanted my coworkers to understand what I was doing. I didn’t want them to think I was working magic. I wanted them to understand it — partly because when they go to conferences, they’d have to explain it themselves. So I started doing these Friday morning stat chats, as I called them at the time.

They were great — at least I thought they were. But I worked in an academic laboratory, which meant new people were always coming in. And the other problem was the gap between when I had time to teach and when people actually needed to know the information. If I teach it tomorrow morning but they don’t need it for another six or eight months, they’re just going to forget it and have to learn it all over again.

So I thought YouTube would be a useful way to solve both problems — instead of repeating the same lectures every semester, and to allow people to learn when they needed to, rather than when it was convenient for me to teach. I envisioned it like a virtual bookshelf. In the lab there was a bookshelf of technique pamphlets — if you needed to do PCR, you’d pull down the pamphlet and follow it step by step. I imagined StatQuest would be that virtual shelf: someone’s reading a paper, they see a statistical analysis, they go, “what is this?” — they watch my five-minute video, learn it, and move on.

In the first year, I got 9 views and 2 subscribers. And that was actually a huge success for me, because that meant people in my lab were watching, and two of them liked it enough to want more. It was the 9 views that counted — the 9 views I wanted. And then it just so happened that other people started watching too, and after a couple of years I was able to leave my job and do StatQuest full-time. I’ve been doing that for six years now.

Raja Iqbal: That’s amazing. I was doing a fireside chat with Sal Khan from Khan Academy at a conference, and Sal started in a very similar way — tutorial videos for a cousin who was struggling with math. More people wanted to learn, he uploaded the videos, one thing led to another, and here we have Khan Academy. You didn’t start out wanting to be a YouTuber.

Joshua Starmer: Not at all. This is definitely not what I dreamed of doing. I mean, it is a dream job, I love it — but it was nothing I ever aspired to. It just happened because of the way I was using YouTube as a tool. It was free to upload, and it solved some problems I was having at work. It just so happened that it took off. That wasn’t really the goal.

Chapter 2 — 05:40 | From Computer Science to Bioinformatics

Before StatQuest, Josh studied computer science and music composition simultaneously, stumbled into biology through an informal deal with a professor, pursued a PhD in bioinformatics at NC State, and ended up in the exact lab job he’d read about in an alumni magazine years earlier.

Raja Iqbal: You pivoted from computer science to bioinformatics, right?

Joshua Starmer: Yeah. As an undergraduate I studied computer science and music composition — two degrees pursued at the same time. I found that if I only did computer science I got a little blue, and if I only did music I got a little blue. I always needed a bit of both. After I graduated I got a job in a hospital, and some people there tried to encourage me to go to medical school. I said, I don’t know — let me at least take a biology course first. I’d never taken one before.

I contacted a professor at the local university and made a deal: if you let me sit in on your 8 a.m. Biology 101, I’ll come back in the afternoon and do whatever lab work you need done. I just want to learn what goes on in a lab. He agreed. I loved the class, working in the lab was fun — and that led me to a PhD program in bioinformatics at NC State.

Raja Iqbal: So you did become a doctor — just not the kind that helps people feel better.

Joshua Starmer: Yeah, a different kind of doctor. So I got the PhD, then got a job at UNC. There’s a funny story about that. Before I even went to grad school, my parents had both gotten graduate degrees from UNC and still got the alumni magazine. I was flipping through it one day and came across an article about a guy they’d just hired to start the genetics department — Terry Magnuson. On the side of the article was a little blurb about the person who did the computer work in that lab. And I thought, that’s the job I want. I want to be the computer person in a lab.

So I emailed that guy. Found his address, sent him a message: “I want a job just like yours. What do I need to do?” He told me to get a doctorate in bioinformatics. So I did. Funny thing is — he’d actually left UNC before I got there, but I ended up with his old job, in the exact same lab he’d worked in. Very pleasing.

Raja Iqbal: And StatQuest happened after your PhD?

Joshua Starmer: After. I worked at UNC for probably 10 years before it happened.

Chapter 3 — 13:30 | Knowing Your Audience & Making Videos That Actually Teach

Josh reflects on being told he was a “terrible storyteller” in middle school — and why that blunt feedback became one of the most useful things that ever happened to him. He breaks down his video-making process: testing rough drafts on real people, the challenge of updating YouTube videos, and the central question he asks himself for every single concept.

Raja Iqbal: Did you always know you had it in you as a communicator? Not everyone who is capable can actually communicate ideas.

Joshua Starmer: Communicating is actually a big struggle for me. But I’ve been very fortunate that throughout my life, people have been blunt and told me I wasn’t good at it — and that’s helped me try to be better. A long time ago in middle school I was telling a story to my friend Jake Roberts, and Jake just looked at me and said: “You are a terrible storyteller. You start in the middle, you jump around, it’s impossible to follow.” He was very blunt, but he was my friend and he was actually a good writer, so he could tell me exactly what I was doing wrong. That ended up being very influential. I started thinking — maybe I can fix that. Maybe I can communicate better. It’s something I’ve worked on my whole life and continue to work on.

Raja Iqbal: Does Jake know how wrong he was now?

Joshua Starmer: He was right at the time. I actually haven’t seen him since — he moved to California the very next year.

Raja Iqbal: So, as an educator myself, I was watching your bootstrap sampling video while getting ready for this podcast. I found the way you approached it incredibly clear and methodical — I actually learned from it. What goes into making your videos?

Joshua Starmer: There’s a lot to it. One thing I stumbled into early on — which is very critical to communicating well — is knowing exactly who your audience is. When I’m making a video, I think about very specific people I used to work with. I actually still go back to the lab I used to work in and show them rough drafts. If I look out at them and they’ve fallen asleep, or I see that look of confusion, I know I need to fix things. Even on my own I just imagine how they would react — I’ll think, oh, this is going to confuse them, I need to be very careful here.

The other thing: YouTube makes it very difficult to update videos. If there’s a confusing part and I make a new version, it starts with zero views, zero likes, zero comments. It has to crawl back to where the original was. It’s sort of a nightmare. For minor issues — a typo, a small error — I’ll leave the video up and note the correction in the comments and description. For major problems, I bite the bullet and take down the original even though it has all the momentum. It’s only happened two or three times. The goal isn’t just to get views. The goal is to educate people, and I need to be a reliable source.

But the thing I always ask myself when making a video is: can I make it any simpler without dumbing it down? Do I really need this fancy terminology? Do we need linear algebra for this, or is there another way? But without losing the truth of the original algorithm — I want it to be true to the original intent of the method. When I find a simpler way to do that, it feels like I’ve solved a puzzle. It’s a big rush.

Raja Iqbal: Explaining things without losing the spirit of the concept — that’s always a hard balance. But if you think hard enough, you can usually get there.

Joshua Starmer: Yeah, exactly. That’s the ideal goal — simplify it without dumbing it down. That’s what we’re shooting for.

Chapter 4 — 20:00 | The BAMs, Sound Effects & Why 3 Ideas Per Video Is the Magic Number

What started as a way to keep audiences awake became a full teaching system. Josh explains how BAM, double BAM, and triple BAM evolved into a structural tool — and how discovering there’s one secret quadruple BAM in existence led to breaking logistic regression into 8 separate videos.

Raja Iqbal: You use a lot of sound effects and musical instruments. Is that intentional because you think it makes tutorials more engaging — or is it just who you are?

Joshua Starmer: It’s a little bit of both. Saying “bam” started as me trying to make sure the audience was still awake — just throwing something out there: hey, wake up, BAM! The sound effects followed along those lines, but that’s also just very me. The songs, interestingly, didn’t start in the early videos. That evolved over time. And I’ve actually heard that it’s very useful — a lot of people watching my videos are intimidated by the subject. They’re stressed, they’re anxious. And the silly songs at the start help a lot of people calm down and realize: okay, this guy with this ridiculous song can do this — maybe I can too. It helps people relax and go, I can do this.

Raja Iqbal: Your “bam” system also turned into a teaching device.

Joshua Starmer: It did. Whenever I have a major point, it gets a big BAM. The second major concept, a double BAM. The third, a triple BAM. And over the years I’ve discovered that having 3 main ideas in a video is the right number. If I need to get to a quadruple BAM, that probably means I’ve got too much information and I need to break the video up.

Raja Iqbal: Has there ever been a quadruple BAM?

Joshua Starmer: It’s very, very top secret. There is one — in my most recent video, actually. But generally, if I have to do a quadruple BAM, that means I’m trying to teach too much. I discovered this when I was making my first video on logistic regression. I kept cramming more in: bam, bam, bam, bam, bam. I ultimately broke it into 8 different videos, each with its own 3 BAMs worth of information. So I use it as a teaching aid for myself — stay focused, don’t try to teach too much. And it’s a way of emphasizing the main ideas. Here’s a BAM, here’s a double BAM, here’s a triple BAM. Those are the three things we’re talking about today.

Raja Iqbal: Do you plan the BAMs ahead of time, or do they happen as you go?

Joshua Starmer: Both. Sometimes I’ll plan it — what are the three main ideas I want to convey? Other times I discover them while creating the slides: oh yeah, that’s a BAM right there. But I know one when I’ve got one, that’s for sure.

Chapter 5 — 26:00 | Storytelling Is the Real Teaching Tool

Josh makes the case that every effective explanation — from a sporting event to a statistics lecture — lives or dies on dramatic arc. State the problem clearly, establish the stakes, then introduce the tool. People may forget the eigenvalues, but they’ll remember what PCA is for.

Raja Iqbal: How important is storytelling to teaching STEM?

Joshua Starmer: Personally, I think it’s very important. Regardless of whether you’re teaching on YouTube or in person, it’s good to have some dramatic arc to what you’re talking about. There has to be some story to it, because that’s human nature — we like stories. For thousands of years we probably sat around a fire and told each other stories, and it helps keep our attention focused. It also improves general communication, because if you can tell something as a story, it becomes easier for people to understand what’s going on.

A standard storytelling technique in teaching is: clearly state the problem, and why we need to solve it. State the stakes. Then: this is the tool — we’re going to use principal component analysis to solve this problem. But you have to really clearly state the problem and why it matters. People may not remember what an eigenvalue is or an eigenvector, but they’ll come away knowing: PCA solves this type of problem, and it’s important that we solve it. Storytelling gives people a framework to organize all the new information in their brain, because a dramatic arc is a format we’re already very familiar with.

Raja Iqbal: Even beyond teaching — Steve Jobs, Elon Musk — storytelling is almost a prerequisite for anyone trying to communicate at scale.

Joshua Starmer: Completely agree. Anything that isn’t boring probably has a dramatic arc to it. A close sporting match has a dramatic arc. Pretty much every movie that didn’t make you want to go to bed has one. You know what they’re trying to accomplish — you’re just dying to find out how they’re going to do it. Clear communication means using a story that actually is a story, that actually has drama. Even though we’re talking about math and machine learning, there’s drama — and we shouldn’t forget that.

Chapter 6 — 30:00 | The Future of Education & The Risks of AI Tutors

Josh is genuinely excited about AI-powered education — virtual philosophers, on-demand visual explainers, learning exactly when you need it. But Raja pushes back on a harder question: what happens when the person teaching you isn’t neutral?

Raja Iqbal: AI assistants are becoming more pervasive. We have live interactive avatars. How do you see the future of education in STEM and liberal arts?

Joshua Starmer: Yeah, I definitely think it’s going to be different. A friend of mine built a program that uses AI to teach philosophy — a virtual environment where all these philosophers are hanging out, and you can walk up to them and talk to them. Each philosopher is basically a language model trained to describe things the way that person’s philosophy would. And I thought: what a cool way to learn. I can’t actually talk to Socrates in person, but if I could talk to him in a game, I’d probably learn a lot more.

I think over time, instead of just getting text back from AI, you’ll be able to say “explain this” and it’ll give you an illustrated concept, walk you through it step by step. And that’s very exciting, because it means more people will be able to learn more stuff — especially when they need it, just like I was saying about YouTube. They’ll be able to use these AI tools to learn on the fly, in data science or AI or whatever field they’re in. That’s absolutely fantastic to me.

Raja Iqbal: My concern is less the technology side and more intentional manipulation. If I use DeepSeek, certain things get filtered out. If I use Grok, certain things get filtered out. ChatGPT too. These companies are big, and they control what gets taught. Does that worry you?

Joshua Starmer: In the field I work in — machine learning, AI, statistics — there’s not much motivation to censor that content. It’s not contentious. It doesn’t make a country look bad. But I can see what you’re saying: if you were a historian, relying on models to give you an unbiased perspective on history that wasn’t particularly flattering to someone — that could absolutely be a problem.

I think over time, models will get better at signaling confidence. Something like: this first part is high confidence, but flag this other part in red — use that with caution. If AI companies took that seriously, the tool would be much more useful. We wouldn’t have to take everything at face value. Nothing’s perfect — even original sources have errors. The XGBoost paper, for example — great model, geniuses behind it, but not the best proofreaders. Errors all over it. The goal, as an optimist, is to minimize those effects and maximize the learning.

Chapter 7 — 40:00 | Inspiring Young Learners & Teaching the Math of the Real World

Josh just wrote a children’s book for 7- and 8-year-olds, starring a monster named Squatch and a french fry eating contest — designed to introduce variance and uncertainty before fear has a chance to set in. He makes a case that statistics isn’t just math: it’s the only tool we have for the world as it actually is.

Raja Iqbal: Where does the problem start? Why do we fail to inspire kids? So many of them decide math is not for them.

Joshua Starmer: You asked two questions there — how do we inspire confidence in mathematical skills, and how do we help people generalize beyond a single use case. On inspiring young people: I just wrote a children’s book targeted at 7- and 8-year-olds that I’m hoping will help with that. It introduces statistical concepts in a completely low-stakes way.

The main character is Squatch — a big fluffy monster — who wants to enter a french fry eating contest. Squatch has to eat 10 fries to win, but they only have 7 fries to practice with. So from those 7 fries they have to guess how long it’ll take to eat 10. That’s an experiment. Timing yourself eating fries is data collection. Predicting from that data is making an educated guess. It’s easy, it’s fun, and it introduces the concept that not every french fry takes the same amount of time — some are bigger, some are smaller. That variation matters. That’s the whole concept of variance, introduced before the word “variance” even has a chance to be scary.

I’m targeting that age group for another reason too. When I started taking statistics for the first time, every math problem I’d ever done up to that point ended in a single number. A hard, clean number. Statistics gave me fuzzy outputs — a number plus or minus something — and I was resistant to it. I thought: I’ll just report the mean, ignore the rest. It took me a while before I understood that standard deviation was actually the useful part.

Joshua Starmer: The math we learned in elementary school and high school — that’s the mathematics of heaven, where everything’s perfect. Statistics is the math of the world. The reality we live in is: nothing is perfect, and we’re always dealing with incomplete information, measurement devices that aren’t precise, estimates and guesses. Variation is the world we live in. Statistics is the only tool we have to quantify it and make better decisions.

Raja Iqbal: I use Amazon reviews to explain this. When you buy something, how do you make a decision? You look at the average — but is that all? No. You look at how many reviews there are. You look at the extremes. How much variability is there?

Joshua Starmer: Exactly. If there’s only one 5-star review, you’re going to be skeptical. The sample size isn’t big enough to have confidence that it’s actually a 5-star product. These concepts come naturally to us as humans. It’s just when we formalize them that they become scary.

Chapter 8 — 51:00 | Lightning Round

The most misunderstood concept in statistics. The hardest to explain. Josh’s favorite educator. His favorite StatQuest video. What he’d change about how stats is taught. And his favorite distribution — plus the theorem that makes it king.

Raja Iqbal: Lightning round — quick answers. What is the most misunderstood concept in statistics?

Joshua Starmer: Probably the p-value. A lot of people think it’s the probability of whatever we observed. But it’s actually the probability of what we observed plus more extreme — and that “plus more extreme” part is quite important.

Raja Iqbal: At Microsoft we used to run online experiments, and the moment you hit a p-value under 0.05, suddenly all the guards were down. No one could question anything.

Joshua Starmer: There are definitely pitfalls.

Raja Iqbal: What is the hardest concept for you to explain?

Joshua Starmer: They’re all really hard. But the one I’ve spent the longest time working on — and I’m still working on today — is linear regression.

Raja Iqbal: Besides Josh Starmer, who is your favorite educator?

Joshua Starmer: I love Luis Serrano. He’s a good friend of mine, but every time I watch one of his videos I’m just awed by how good he is. I see him as a role model.

Raja Iqbal: What is your favorite StatQuest video of all time?

Joshua Starmer: I love them all — but I distinguish them by their theme songs. My favorite theme song is probably the Linear Discriminant Analysis one.

Raja Iqbal: Do you write those songs?

Joshua Starmer: They’re always improvised. That’s why they’re so silly and ridiculous — I make them up on the spot.

Raja Iqbal: If you could redesign how math and stats is taught at the college level, what would you change?

Joshua Starmer: Step one: throw away whatever they’re using and use my stats book instead. Shameless self-promotion. But seriously — I actually stumbled into an Intro Stats 101 course at UNC once while I was lost in a building, had some time to kill, sat in the back row. It was a complete train wreck. That really did inspire me to write the statistics book and do everything I can to improve how it’s being taught. It’s a disaster.

Raja Iqbal: What is your favorite statistical distribution?

Joshua Starmer: The normal distribution. Because of the central limit theorem. Even if your data isn’t normal, the average is normal — and most of the time, we’re comparing averages. So you can get all bent out of shape over normality assumptions, but if you’re looking at averages, it just doesn’t matter. It’s magical that the normal distribution kind of rules them all.

Chapter 9 — 58:00 | What Josh Is Working On Next

Inspired by his non-technical brother Mike’s conversation with an AI, Josh is building a visual, non-technical explainer of how AI actually works — designed to anticipate the exact confusions a business-minded person falls into. Plus: a statistics book with Jupyter notebooks for both R and Python, due in early 2026.

Raja Iqbal: What is the next video you’re working on?

Joshua Starmer: I’ve just made a long to-do list for 2026. The first one is probably going to be inspired by my brother Mike. He’s a business person — he loves technology, enthusiastic about it, but doesn’t have the most technical brain. He wanted to know how AI works, so he asked an AI: “Hey, how do you work?” And he ended up in this conversation with it. He sent me a copy of the transcript.

Through that conversation, I could see the things he wanted to understand — but I could also see the misconceptions and the pitfalls he was falling into as he tried to make sense of it. So I’m making a video basically geared for people who want a very non-technical overview of how AI works, but it’s visual, easier to follow, and it anticipates the specific confusions my brother ran into and tries to help you avoid them. I’m presenting it in person at Carnegie Mellon in about a week and a half, and it’ll probably be out in early 2026.

I also have a statistics book coming out in early 2026, and I’m working on Jupyter notebooks in both R and Python for every chapter. For statistical work I prefer R, but a lot of people only know Python — and there are some nice Python modules that make Python behave in a very R-like way, so it’s a small move for Python people to start doing statistics that way.

Raja Iqbal: Josh, thank you so much for your time. It has been a pleasure.

Joshua Starmer: It’s a pleasure being here. Thank you very much.