How to HUMILIATE a Fake Expert

So did you do that first step in the — that is the sound of a man's own debate partner desperately whispering the answer he should have known all along — but it's too late. — So are you controlling for age, socioeconomic status, smoking status? pre-existing conditions? Are you controlling for health? — All right, guys. I think you guys uh did you say you don't need to did you just to? Antivax campaigner Steve Kersh has just been asked whether he checked if the people in his study were actually comparable. That is whether the vaccinated group and the unvaccinated group were similar enough that any difference in outcome could be actually attributed to the vaccine and not the fact that one group was older or sicker or wealthier. It's the most basic requirement of the method and he's just said that he doesn't need to do it. — Did you just say you don't need to? Why hello my fellow apes. I hope you are well. If you've ever been in an argument where someone drowns you with data in technical language in jargon you don't recognize and you don't know how to respond, this one is for you. — DMR, I looked at GLM Kaplan Meyer. — You didn't make sure your groups were comparable — because what you're about to watch is a lesson in how to hold someone to the burden of proof and what happens when they can't meet it. We are back with part two of the Pangburn philosophy debate on vaccines. In one corner, antivax tech entrepreneur Steve Kersh, a man who believes his millions and his ability to download data sets makes him an expert in epidemiology. And Pierre Corey, a disgraced former physician. In the other, Dr. Dan Wilson of Debunk the Funk, a molecular biologist who's made a career out of methodically dismantling antivaccine claims. And Dave Fina of Professor Dave explains, who confronts pseudocience with the energy of someone who took it personally. Why don't you get in a car and drive for 24 hours and over 24 hours veer by one degree? Are you going to feel that — that would be equivalent to this? People say, "Oh, I'm just trying to think of the easiest thing. " — something to say. — In this episode, we're focusing on a particular exchange between Steve and Dan. Steve is used to overwhelming his opponents with rapidfire data points and obscure references. And when challenged, he doesn't tend to defend his own work. He demands that you disprove it. That is, he tries to shift the burden of proof onto his opposition. — Where's your other check? — But Dan isn't here to play that game. He's here to hold the line. You made the claim, Steve, and so you must prove it stands up. And he can't. It starts with Dan citing a study of 46 million adults that found no increased risk of cardiovascular events after vaccination. Steve's response, demand the raw data. But Dan wants to know if Steve would even know what to do with it. Yet, I would Yeah, I'd protect it with private internet access. Recently, several major AI companies have been scraping pretty much everything that you've done online to train their models. Browsing history, search queries, the lot. Which means that somewhere out there, a future super intelligence is being shaped in part by my 2-hour investigation into whether a medieval peasant could survive being launched from a trebuche. Spoiler, they could not. You're welcome, Skynet. The unsettling part isn't that it's happening. It's that nobody's particularly surprised. Your internet service provider logs your traffic. Advertisers build profiles on your habits. And apparently your 1:00 a. m. doom scrollings are being compiled for our robot overlords. That is why I use Private Internet Access, a VPN that roots your internet connection through an encrypted tunnel, masking your IP address from hackers, your internet service provider, and anyone else with an unhealthy interest in your browsing history. Here's the thing about VPN companies. Anyone can claim that they don't log your data. Private Internet Access, however, has had to prove it in actual court cases. And in an independent audit by Deote, they have proven it. They cannot hand anything over because there's nothing to hand over. That is a structural guarantee. With servers across 91 countries and 50 US states, peer-to-peer support, and unlimited device protection, everything is covered under just one account. Right now, my viewers get 83% off, just $2. 3 a month with four extra months free and a 30-day money back guarantee, risk-free. With over 30 million downloads and 24/7 customer support, Private Internet Access has earned its reputation. Go to piaavvpn. com/rationality rules to get 83% off Private Internet Access with 4 months free. Thanks. — Do you know how to identify problems with it? — Yes. — And what is the problem with this data

set? — I haven't seen it. I can't — I sent you the paper. — You sent You didn't send me the data. — So the analysis found among 46 million adults in England and Wales that there is no increased risk for cardiovascular events including heart attacks. — Then what's then why don't they disclose the data? — Steve, you can't take public health care records and disclose it to everybody. — Why can't youidentify the Czech Republic? Did anyone die in the Czech Republic when they disclose the data to everybody? — Steve, you you've tried this with publicly with data sets that you've gotten your hands on and your analyses are terrible. You don't understand what you're doing. — Really? Really, Dan? — The audience claps and by this point of the debate, they know the pattern. Dan isn't attacking Steve's conclusions. He's attacking his method. There's a difference. You can be wrong about the answer, but right about the process. Dan is saying that Steve doesn't even understand the process. — You don't understand what you're doing. — Really? Really, Dan? — Yeah. — What? They're terrible. What What's the correct analysis for the Czech Republic data, Dan? Where where's your Where is your publish where your study of the Czech data? — This is a sneaky yet strategic move from Steve here. He's attacked on method and he fires back. What's your correct analysis then? Where's your study? He wants Dan to prove that he can do better. — Where's your study of the Czech data? — And notice where he runs. The Czech Republic data is Steve's crown jewel, a publicly available government data set of vaccinated records and mortality outcomes. Antivax advocates treat it like a smoking gun because it's publicly available. Anyone can download it and run their own analysis, regardless of whether they have the training to do so. So Steve, unsurprisingly, thinks this data proves that vaccines are killing people. So, so what would a measure of the Czech data? — So, what's the first step in doing a good analysis of say the Czech Republic data? What's the first step? — Can you see the two conversations happening? Dan is sticking to methodology. He's saying, Steve, if I gave you the data, what would you actually do with it? — What's the first step in doing a good analysis of say the Czech Republic data? — But Steve is now demanding something completely different. That Dan produce his own analysis of the Czech data. — Where's your check data? He's no longer defending his analytical skills. He's trying to shift the burden of proof, making this about Dan's analysis of a completely different data set. But Dan doesn't take the bait. Instead, he applies more pressure. What is the first step? If Steve knows what he's doing, this should be easy. If he doesn't, he's about to expose himself. And just to be clear on what that first step is, when you're comparing two groups, say vaccinated versus unvaccinated, you need to make sure that any difference in outcome are actually caused by the thing you're studying, not by other factors. If the vaccinated group is older, sicker, or wealthier than the unvaccinated group, those differences will skew your data. I mean, you can't put all the babies in one group and all the seniors in another and then act shocked when the older group dies at higher rates. Controlling for confounders means accounting for these variables. So you're measuring the vaccine effect, not just noise. It's not a fancy optional extra. It is the foundation. — What's the first step? — Yeah, there's there — you can put it into all sorts of different models. — Yeah, sorry. Methods of analysis. — Be specific. You did this. What did you do? — A simple, sharp, and devastating question. Dan is saying, "Come on, Steve. Tell us specifically what you did. — You can put it into all sorts of different models. " — And Steve gives an incredibly vague answer, implying that he skipped the basics, perhaps even that he doesn't know them. — There Dan's force specificity is relentless. He's pinning Steve down, leaving no room to escape into abstraction. — Be specific. You did this. — Um, I analyzed it all sorts of different ways. So, so at first I looked at time series data. So did what was the first — time series that was — So you just looked at a time series. — No, that was the first thing, right? There are lot all sorts of different ways to analyze the data. It's all and all the methods that I used are published on my GitHub. It's all in public view. Everything I did is in — And when pressed again, Steve retreats even further. This time to it's on GitHub. — All the methods that I used are published on my GitHub. — First it was all sorts of models. — All sorts of different models. then all sorts of different ways. — I analyzed it Now it's everything I did is in public — It's all in public view. — He's gone from not being able to describe his method to simply pointing to a URL and hoping that counts. To those not paying attention, it's a clever dodge. He gets to sound transparent and open while avoiding the question entirely. But transparently bad maths is still bad maths. — It's all in public view. Yeah. — Everything I did is in public view. — Yeah. And you know, there should be

— And you know what you didn't do? You — You didn't do the first step, which is to make sure that your two groups that you're comparing are comparable. — Dan's got him. He spent the last minute getting Steve to admit he just ran models without mentioning controls. Now he drops the hammer and the audience knows it. Steve skipped the first step. A step that isn't just crucial to epidemiology. It's foundational across statistics, experimental science. social sciences, clinical research, basically any field that compares groups to draw conclusions. And Steve is seemingly unaware of this. For context, analyses of the check data have noted that Madna recipients tended to be older than FISA recipients. Exactly the kind of systematic difference that would inflate death rates in one group regardless of vaccine safety. — You didn't do the first step, which is to make sure that your two groups that you're comparing are comparable. Oh, really? What? How? How are they not comparable? — So, in the Sheekch data, I you argue that the vaccines were distributed randomly, right? — I There is no evidence of a systemic — That's what you did, right? — There's no evidence. — Notice the pause. Steve sees the problem. If he says, "Yes, it was random," he's actually wrong. Elderly people got vaccinated first. Sicker people had priority. Availability varied by region. If he says no, he admits he needed to adjust for confounders. Both roads lead off a cliff. So he tries a third path, weasel words. — He's not saying it was random. He's carefully saying there's no evidence it wasn't. It sounds like a defense, but it commits to nothing. — So people can choose their vaccine. There's a preferential recommendation against Merna because we're not all in the pockets of big pharma and we go with the data. — Yeah. So, and there's availability differences across the country, right? — So, it's not random. — No, That admission is fatal. Steve's data is not random. In statistics, if data isn't random, you must control for confounders. You have to ask, are the people who got vaccinated systematically different from those who didn't? If they are, then any difference in outcomes might be due to those differences, not the vaccine. And so just like that, Steve has destroyed his own defense, though he may not realize it yet or ever. — So it's not random. — No, it's not random. — But you argued it was random. At one point you had to be told it wasn't. — No, I said there was no systemic or systematic bias in the distribution of the vaccines. People didn't know which vaccine was going to be safer. Like if you — they were advised there were there was a preferential recommendation against Madna. — No, there wasn't. — Yes, there was. — No, that's news to me. Where is it, — Steve? Really? It's 2025. It's 2025 is an insult. If you're still arguing basic premises that were settled years ago, you haven't been paying attention. It's that simple. — That's news to me. Really? It's 2025. Where where is that recommendation from that shows that there's a bias in the distribution but it doesn't matter because you can look at — it does matter it does — oh okay why — I think — because you the first step of looking at this data set is making you making sure your first two groups are comparable you want to be able to compare them accurately so that it's not just statistical biases you're analyzing — the audience gets it Dan's strategy is working he's anchored the debate to comparability Steve wants to talk about his results and look at the data, but Dan won't let him move past step one. — The first step of looking at this data set is making sure you making sure your first two groups are comparable. — Until the groups are comparable, the results are noise. It's like debating how fast a car is before establishing that it has an engine. — So that it's not just statistical biases. You're analyzing the data. — So do you understand what the how the Kore method works? — Kore method. — Yes. — Is this what you did? Notice how Steve tries to intimidate Dan with jargon. Kc core method. It's a term that sounds technical and sophisticated to his audience as if he knows what he's talking about. It's the same pattern. Shift the burden of proof by making the conversation about whether Dan can keep up rather than whether Steve did his work. — Do you understand what the how the Kore method works? — But Dan just asks, is this what you did? It doesn't matter what the K method is if Steve didn't use it. A core method. — Yes. — This is one of No, I looked at at DSCMR. I looked at uh uh GLM. I looked at — But you didn't make sure your two the groups were comparable. — You didn't make sure your This is a clean gish. Gallup. DCMRR GLM Kaplan Mayor rattled off in quick

succession. Steve isn't explaining any of them. He's listing them, hoping the sheer volume of technical terms will overwhelm the question. — At DSCMRR, I looked at uh uh GLM. I looked at — But you didn't make sure your two the groups were comparable, — but Dan doesn't engage with a single one of them. He just repeats the same line. You didn't make sure the groups were comparable. — You didn't make sure your Oh, the randomiz that I use is better. randomized madna and — it's not random. Yes, it is. It's not random. Pierre jumps in to help claiming that the groups were randomized. — It's not random. Yes, it is. — But this contradicts what Steve said seconds ago. Steve just said, "No, it's not random. " — So, it's not random. No, it's not — On form, Dan swats this away instantly. Ironically, Pierre is actually hurting Steve here by highlighting the central floor. If your ally has to contradict your own words to defend you, you're losing. It's not random. Yes, it is. It's not random. — No. But it doesn't matter, Dan, because — can people choose which to get, and is there a preferential recommendation? — Break it down. It's not random. — I didn't break it down by the vaccine manufacturer. I'm just looking at people who are vaccinated and unvaccinated. And of course, you know, — did you make sure those two groups were comparable? — Now, that's subtle, but it's a classic case of moving the goalposts. Steve just shifted from comparing different vaccines to comparing vaccinated versus unvaccinated. — Dan ignores it completely. He catches the pivot immediately. Did you make sure those two groups were comparable? Same question, same problem. No escape. — Did you make sure those two groups were comparable? as an expert in this. — Do you know? Do you understand? — Yes or no, Steve? Did you — Dan the — Did you Yes or no? It's the first step. — Yes or no? That is all Dan wants. He's sharply interrupting Steve's attempts to pivot, forcing him to answer a binary question. You can dance around a conceptual debate, but you can't dance around a yes or no. — Did you? Yes or no? Do you understand that the groups are — that's the first step? when you have a vaccination that the two groups are completely different? Do you understand that Dan? So why are you asking me this question as to whether the groups are comparable? — Because you have to make sure they are as comparable as possible. Do you know why? — No. That's the whole point. It's all falling apart as Steve practically screams his denial. — Well, do you know why? — NO. GOD, PLEASE. NO. NO. — He's arguing that making groups comparable, the foundation of the scientific method, is unnecessary. He thinks he's winning by pointing out that the groups are different. He doesn't realize, it seems, that he's confessing to incompetence. you understand that when you have a vaccination that the two groups are completely different. — Yes, the groups are different. That's exactly why you need to control for those differences. That is the entire point of adjusting for confounders. You remove as many differences as possible so that any signal in your data actually points to a cause, not noise. It's not that hard to get. And that is what makes this all so painful to be honest. Steve is treating the problem as the solution. — No. — That's the whole point. — You have to make sure the two groups are as comparable. Steve, why do we do randomized control trials? Why is that the gold standard for clinical trials? — Because that makes it really easy for people to see a signal. — That makes it easy. Because they're randomized. Because — Exactly. Because your two groups are comparable. Because they're random. — Checkmate. Dan has led Steve into agreeing that randomization makes groups comparable and comparability is why randomized control trials work — because that makes it really easy for people to see a signal. — The logic is airtight. Comparisons require comparability. Steve's study lacks comparability. Therefore, Steve's study doesn't work. Steve just endorsed the very principle that destroys his own analysis. because your two groups are comparable because they're random. — So, so Dan, when you don't have that available to you, you have other techniques that normalize for the mortality of the two groups. — Exactly. And so you have to make sure that Exactly. You have to control for confounders. So how did So did you do that? — First step in the first step in

— Did you catch that? Pierce whispers for Steve to say yes. He knows that Steve is on the ropes. The fact that the correct answer is so obvious that his ally has to whisper it proves that Steve is failing. When your corner has to feed you lines, you've lost the room. Everyone in the room. But the bigger problem, he can't say yes as he's already admitted he didn't. It would just be an obvious lie. — Did you step in the first step in the al algorithm is to make them comparable so that you can then compare them. How — you fit the slopes to the death curves. — That's a fixed of fixed cohorts. — That's not controlling for the confounders. That would make the groups comparable. — But it does. — No, it doesn't. — It does, Dan. Because do you understand? — It's you versus them. Steve is totally cornered on confounders, a simple concept. So he pivots to go mortality, a complex one. He's hoping the audience thinks, "Wow, Steve knows big words. He must be right. It's a smoke bomb. And notice the framing. Do you understand Gumpert's mortality? — Do you understand? — He's not defending his method anymore, is he? He's testing Dan, trying to make Dan prove that he's qualified to ask the question. It's yet another attempt to shift the burden of proof. If you can't follow my terminology, you don't get to challenge me. Yeah. When you can't win on the basics, flood the zone with jargon. — It's you versus the epidemiologists of the world on this, Steve. — It's epidemiology 101. — This is the key move. Dan refuses to let this become gonearts versus confounders. He frames it as Steve versus basic science. This is crucial because it gives the audience permission to ignore Steve's jargon. You don't need a PhD to understand that one person is explaining introductory methodology and the other is throwing out terms to evade questions. And the room sees it. It's epidemiology 101. — Do you understand do you understand gumprit mortality? — Sorry. — Do you understand garrance mortality? — Yeah. You understand garrance mortality um with when we have um the um when people are dying right mortality just says oh it's an exponential great but there's competence mortality with depletion. Do you understand that? And you understand that if you look at those curves, the people who are dying might be different in each group. — Do you know what their difference is? — This is boring. — Pierre realizes the Gart's jargon isn't working and Steve is floundering, so he calls it boring and tries to reset the room. It's a submission signal disguised as superiority. When you're winning, you don't need to call the fight boring. — The people who are dying might be different in each group. — Do you, Steve? because you had to control for them. — I do. — What's this? — Yes, I do. And you didn't control for them. — Believe it or not, they're on different parts of the garence curve. Do you know what happens? — You can practically hear Dan roll his eyes. — Do you know what happens when you have a frailty of three or five as to where you are in the conference curve? What does that do to your death slope when you're on the conference? So, so again, we're just going to say this is all you did and you didn't account for epidemiology 101. — Same move, same result. Dample Steve, back onto the question that matters. Did you make your groups comparable? — And you didn't account for epidemiology 101. — No, this is — this is not all I did. The first thing is to normalize the cohorts so that we take away the differences in the cohorts. And so what — so are you controlling for age, socioeconomic status, smoking status, pre-existing condition? — Notice that Steve doesn't answer the question. Dan asks whether he controlled for age, socioeconomic status, smoking status, and Steve responds with a completely unrelated demand. Tell me the negative control. — Are you controlling for pre-existing? — It's a red herring. And another attempt to shift the burden of proof. Instead of answering, he's challenging Dan to address something else entirely. — Are you controlling for pre-existing conditions? Are you controlling for healthy? And there it is. After 4 minutes of evasion, jargon, and deflection, Steve finally answers the question directly. — Are you controlling for healthy? — He admits that he did not control for age, health, or socioeconomic status. He believes his negative control magic exempts him from the laws of statistics, — controlling for pre-existing. — This is the confession. He has admitted his study is flawed by design. The entire structure of his argument has collapsed because he refused to do what every epidemiologist knows is the first

step. — Are you controlling for health? — All right, guys. I think you guys uh did you say you don't need to? Did you just — Did you say that you don't need to? He asks it twice, even making sure nobody in the room can miss what Steve just admitted. — Did you just say you don't need to control that shows that it takes all of that? Let's hang on for a second. — And there we have it. Steve has told us plainly that he does not need to control for confounders because he has a negative control. — You guys control that shows that it takes all of that. — To scientists, this is disqualifying. And to the broader audience, it's the confession that they've been watching him so desperately resist. Pierre, bless him, tries to intervene, but it is too late. The damage is done. First step in the — All righty. So, what did we just watch? A tech entrepreneur with no formal training in epidemiology tried to convince a room full of people that vaccinations are killing millions. And a molecular biologist dismantled him without ever touching the data. Dan didn't need a counter study. He didn't need to debunk a single chart. He just asked one question over and over. Did you make your groups comparable? Did you make sure those two groups were comparable? — As an expert in this, — did you make sure those two groups were comparable? — Do you know? Do you understand? — Yes or no, Steve? Did you, — Dan? The — Did you Yes or no? It's the first step. And Steve couldn't say yes. That's the lesson. When someone makes a big confident claim, whether it's about vaccines, climate, economics, or anything, the burden of proof is on them, not you. You don't need to match their complexity. You don't need to produce your own research. You just need to ask did you do the basics? — You know what you didn't do? — You didn't do the first step which is to make sure that your two groups that you're comparing are comparable — and if they can't answer that everything else the jargon the GitHub repositories — all the methods that I used are published on my GitHub — the Garts curves. — Do you understand the negative controls — are you controlling for pre-existing it's all irrelevant. Well, the king said I was ded to build a castle on A SWAMP, BUT I BUILT IT ALL THE SAME JUST TO show them it sank into the swamp. — Don't argue about the castle when it's built on a swamp. Ask about the foundation first and don't move on until they answer. — There's there you can put it into all sorts of different models. — Be specific. You did this. What did you do? Anyhow, as always, thank you kindly for the view and an extra special thank you to everyone who supports the channel, including today's sponsor, Private Internet Access.