Transcript: EP 02 — The World Cup Bracket Machine

10,410 words · 63:09 runtime · every timestamp jumps into the video.

0:00Thomas I'm 90% confident my bracket will be your bracket. I've been a World Cup guy since 06.

0:07Eric Oh my God. How sad will it be if you lose to just someone that doesn't know what they're doing? Yeah, I have

0:13Thomas France playing Spain. Yes, and I have Spain going forward. You have France going forward. So that will be...

0:19Eric Yeah, if one of those get knocked out early, that's gonna call it for us.

0:24Thomas So Eric, are you ready to jump into our second episode?

0:29Eric I'm ready. I'm pretty hype too.

0:31Thomas That sounded super genuine.

0:52Thomas This is Episode 2, The World Cup, and the question I put to Eric was, Can you predict World Cup winners better than I, question mark? Because I believe with my general knowledge of soccer but specific World Cup and my reading of the vibes in the room, I can be more successful at predicting more wins accurately than Eric can with some BS modeling. That's what we're going to test and it sounds like Eric, you put together some modeling TBD if it's BS or not, to predict who's going to win and just to set expectations for everybody here. The first round, the group stages were not including so he, Eric, completed a full bracket starting at the round of 32 and I completed a full bracket starting at the round of 32 and what we're going to do is Eric's going to walk through the model he put together that he then applied to pick all of those winners and then I'll walk through my model but we're only going to review the outcomes of our models after that round of games has occurred. So right now all of the games completed for round of 32 so right now we're going to look at those and compare and contrast but then after each of the rounds or in some combination of them we'll come back and update to see until the ultimate full winner of episode two dash the World Cup until that winner is decided. I think that's how we're doing it, right Eric?

2:23Eric Yeah, I think that makes a lot of sense. I'm feeling pretty good after winning episode one, you know, explaining how airports work and that the variability is quite comprehendable. So you know I'm feeling good, I'm feeling math has all the answers really and I don't even need to watch games to figure out who's going to win.

2:44Thomas And that's the joy of sports is not even watching the games and just applying a model to predict. So I'm glad you've reached that point because that's what life's about. Exactly. It's a good job.

2:55Eric Now that I do have a bracket it does make the games a little more interesting to watch and I think that's why you know sports betting is a thing. You know people… Yeah, this is

3:05Thomas just a first step into what will become a crippling addiction to gambling so apologies for that but that's not my problem. Also side note, I said you would have a crippling addiction to gambling. I can say that word because I'm disabled and I have a spinal cord injury, just be clear for the audience. I want to make sure I don't get cancelled.

3:30Thomas Okay, walk me through what you got and I'm going to just punch, I'm going to punch it. I'm going to beat it. I'm going to beat it up.

3:37Eric Okay. That sounds good. Well, I just want to start off. This is a 32 team bracket so there is a lot of possibilities. I went and crunched some numbers and there's over 2 billion possible brackets that we have here. Yeah, so I just wanted to gauge first, you know, like how confident are you in your bracket?

3:59Thomas I'm 90% confident my bracket will be your bracket.

4:05Eric All right,

4:06Thomas let's see how it is. How confident are you in your bracket?

4:10Eric You know, I'm feeling pretty good and why I'm feeling good is that, you know, there's a lot of different games being played and I just can't believe that you know everything about all of these different teams. So and you are sometimes kind of a little too vibey. So you're like, oh, this small change is going to affect the game when in reality these games are, you know, somewhat of a crapshoot sometimes and the statistics does tell part of the story and I'm hoping that it's enough of the story to give me another point on the board.

4:45Thomas But yeah, you know, keep it on 32 teams. Oh, it's so hard to keep 32 track of 32 teams, this is impossible.

4:56Eric So, you know, I started by thinking through the different types of models I could use. You know, there's a whole lot of options these days. Of course,

5:04Thomas we all know that, definitely all the models.

5:06Eric And people are getting really fancy with these like ML models and they're like doing all these large scale data things. And I'm like, you know, after I did so well on episode one, I was like, I don't need some huge model to kind of do well. All I need is a single number for each team. And if, you know, people are familiar with these types of things, they're often called ELO ratings. So I'm just going to try to fine tune a nice ELO rating for all the different teams and then just do straight analytical probabilities for each matchup. No coin flips, no Monte Carlo. I don't need fancy stuff to beat your vibes, to be honest. So it

5:47Thomas sounds like you're vibing on the models already, just to be honest. But what is an ELO model?

5:54Eric Well, from my perspective, I don't know what it is. All I know is that we get a single rating for each team. So I'm sure it's got some,

6:02Thomas you know, what do you fucking mean? You don't know what it is. Did you just plug this in the fucking Claude? Are you just using your

6:08Eric brain? No, I did not just plug this in the Claude. I just, you know, ELO, I'm actually not sure what it stands for, but I do know that we have a single rating for each team. Are you going to look it up for us?

6:19Thomas Yeah, we educate on this pod as well.

6:23Eric Yeah, thank you. Thank you. You got me. You got me trying to fly off the cuff of my

6:28Thomas seat. Yeah, I'm ready. I'm coming fucking ready to strike.

6:31Eric You did say you were going to come in hot. All right. All right.

6:35Thomas ELO's rating system is a method for calculating relative skill levels of players, originally designed for rating chess players. I mean, so I get what you're saying. I just want to know what it stands for. It's

6:47Eric commonly used in a lot of different systems now. So just to give a little general high level overview of what's going to happen here. And why it's so beautiful is it's a single number per team. So people might start off at 1,500 points. You can see there, the bar is calibrated here. So I went and fetched a bunch of data from historical FIFA. That's all I need is, you know, I didn't watch any games. I just went and downloaded some data.

7:15Thomas And

7:15Eric everyone started at 1,500. So these are like kind of points in your bag and stuff. It's kind of like poker. You know, everyone's got their 1,500 tokens to start

7:24Thomas the game. Yeah, I'm with you. By the way, it came after our R-pad ELO, a Hungarian-American chess master and physics professor.

7:32Eric I figured it was

7:33Thomas just a name. Light up on your speed, yeah, okay.

7:35Eric Actually, I thought it might be an abbreviation. That's what I was like, but it's just named after a son. Makes sense. So we got 1,500 points, and it's kind of nice in that, you know, it's a zero-sum game. So say United States goes and plays Canada. Whoever wins that gets points from the loser, and they just kind of trade points. And we can go do that at scale, and I'm just going to start this replay. So you can see I grabbed data from 1870 and on. Wow. I didn't even know they were playing football for that long.

8:08Thomas One, the thing is there are at least eight teams that probably don't have much data in here at all because they've only recently started participating in FIFA, and definitely it's the first time they've participated in the World Cup. Because as you may or may not know, this is the first World Cup where there's been 48 teams total. Previously it was always 32, so there's like 40 more games being played and new people who are not in this historical data set or not represented completely. So this is going to be absolutely biased towards the big dogs, which normally would be fine, but we already saw some upsets in the group stages where the big dogs, the classic European teams, South American teams got down. So your data's already fucked. It's already dirty. Dirty dated.

8:51Eric To put it lightly. To

8:52Thomas put it lightly. Application of ELO on historical data going back to the 1880s is just, it's honestly, it's disgusting is what it is. But

9:05Eric I mean, I'm glad you brought some of this stuff up. So I did learn early on that the format changed, and I went and did some field research. I was talking to the few soccer fans that I know and was like, oh, what do you think about this new system? And I was a little, no one seemed to care much. But because there used to be a group stage, the only thing that happened is they have a round of 32 instead of just round of 16,

9:32Thomas right? And more groups because there's just more teams. So then you had the funnel, they had to add another knockout round.

9:39Eric Yeah. So some of the groups, some of the people, so yeah, 16 of them don't actually make it to the round of 32 then. It used to happen where some didn't make it to the round of 16. It's just like the numbers have changed or no?

9:52Thomas Yeah. Yeah. There were just fewer teams. So now they had to add another layer to help funnel people down, and therefore they added an additional game that teams need to play to move through, which also has an extreme impact because teams with older players now might not be able to perform as well. But yeah, that's a big- I did

10:14Eric see a game where Ronaldo was checking out of the game, and the substitutes are serious business. Can you only do a- you never can sub back in, and it seems like you can only do a certain amount of subs. Yeah,

10:25Thomas yeah. There's a max. I think it's- Chat will fill us in here, but I think it's four per half or maybe it's four total. It looks like

10:33Eric four or five. I wasn't sure if it was per half or just the whole game, but-

10:37Thomas And I think maybe he gets more in extra time, but I'm not sure. But Ronaldo's an old man. He's 41. That's why he's out. But anyway, keep going. Yeah, because he

10:44Eric just doesn't have the endurance, and soccer is such a game of endurance.

10:48Thomas Except for Messi, who walks 65% of the time, he already moves less than most players and walks 65% intentionally, so he's still the top goalscorer in the tournament so

10:58Eric far. He still scores goals. He doesn't need to. It's the strategy.

11:03Thomas Exactly.

11:04Eric He knows where he needs to be, and most of the time he doesn't need to run to get to

11:08Thomas - Operation of energy. Physics 101, baby.

11:10Eric Smart guy. Smart guy. Okay. You brought up a couple of interesting things. So one, are you thinking the data from 1890 is not relevant? Did I hear that?

11:22Thomas I think it's less relevant for this application because we're talking about competing against other teams head to head. And in 1890, there was probably like nine teams that are still involved right now. So I think it's relevant, but I think it's less relevant. And I actually think it could flip negative because it will inform stuff incorrectly. So yeah, I'm telling you. Well, you're going to give me a time to frickin' dial right now, and you're going to be like, oh, well, let me do it since 1980 or some bullshit.

11:52Eric No, I was just going to say, oh, I think you're probably right. But one of the beauties of the rating system is that it ends up just prioritizing recency quite a bit because it's a zero-sum game. They're trading points. So while I went and fetched all that old data, it might not matter at all, but the rating system just handles itself. I think like FIFA, when I was doing some research, FIFA used to have like this fancy system and they like only use certain amount of data and stuff. But I went and tested a bunch of different rating systems and never was like my fancy like DK, like, oh, let's only use like prioritize the four most recent data. Never did that even really do much better than just using all the data, even though that 1890 data, I'll give it to you, it probably doesn't.

12:42Thomas Probably not actually impacted. So just be clear, you ranked data ranking tools.

12:51Eric I mean, that was a lot of my research. I was testing out different ranking systems.

12:55Thomas I mean, then the question obvious is, what tool did you use to rank the ranking tools?

13:01Eric Well, so basically I created a CSV file, you know, with a bunch of this data and would try to use it to predict the outcome of the game. So each row could have one of my spreadsheets had like forty eight columns. So you can imagine like keeping track of for a given match, the preceding matches between these teams, you know, and like whether it's home or away, which region is it in? What's the elevation and stuff? OK, OK,

13:36Thomas I was going to say that is a huge impact to where they're playing. OK,

13:39Eric yeah, yeah. So I went and tried these and I was like, you know what? This is a little too fancy. I don't think I need this. OK, so that's when I went back to a single number.

13:48Thomas He went simple. I still think it's not going to work, but I'm glad he did all that legwork for nothing. So,

13:54Eric yeah, I mean, the improvement was just, you know, not as much as I was hoping. And maybe I was getting pretty fancy at the end. I had some marginal improvement, but, you know, it's just not elegant enough. I do like simplicity and elegance. And the fact that I can just take here, let's see this.

14:16Thomas Almost

14:17Eric 50,000 games, 145,000 goals scored, compress it into 32 numbers that matter to me and then use that to predict the back bracket. You know, it's all good.

14:30Thomas So this is all the history of soccer played by all these teams leading up to this very moment and then applying that to the forthcoming bracket matches they have.

14:39Eric Exactly.

14:40Thomas Yes,

14:41Eric I don't include any club games. I don't do roster changes like track players. Yeah, it's

14:46Thomas purely based on, yeah, I understand what you're doing. I think it's, I think it's a little silly. I think you should have gotten more elegant because I think there's more variables at play here than just simple win loss. But that's what that's what the outcomes will prove. So,

15:02Eric yeah, I'll come back to one of your other points later, but I think that gets us to where we need to go. Here you can see in blue the 32 teams. So I just played that and they're jiggling left and right at the end. These are the 32 teams that are in the round

15:21Thomas of 32. Who's all the way on the left?

15:25Eric Macau, is that how I

15:26Thomas pronounce it? Yeah, a super small country.

15:28Eric Bhutan.

15:29Thomas Bhutan, yeah. Dude, so...

15:32Eric Timor-Leste.

15:33Thomas You know what's insane? Cape Verde is probably a similar population to Bhutan and they went through the group stage. They tied Spain and almost beat Argentina. So that, my point is, lightning strikes, my friend. And you can't predict lightning, even though I don't be sure. I will give

15:54Eric you that. That was honestly a crazy game. And

15:58Thomas one of the most dominant teams ever with the greatest player of all time still playing super well. The country is 600,000 people who's never been in the World Cup, never played these tech teams, pushes it. So that's what I'm saying. That's pure vibes, dude. And you can't predict that. So

16:13Eric you were vibing on them. You were contemplating them upsetting Argentina.

16:21Thomas No, but I at least would have contemplated it more than your model would have.

16:27Eric That's fair. They're literally, they might be the last, no, 1691. I mean, they're the third bottom team. And

16:35Thomas it

16:36Eric wasn't like it was a 0-0 game. There's many goals scored. It's like, this is a

16:41Thomas real,

16:42Eric it's a real result.

16:44Thomas And as you'll hear in my vibes explanation, I have a deep bias vibe against, or for Argentina since I live there and know people there. So like, had I not had that, I probably would have vibed more on Cape Verde. But, all right. So you ran the data, Argentina, Spain, France, shocker, crazy out, crazy in sight, dude, that they're up top. Who would have thought? Can

17:06Eric I step back and maybe explain a little bit on how my elo rating type system goes? You know, it's like most, but I just want to give you a flavor for what's going on behind the scenes. That sound good?

17:18Thomas It sounds great, Eric.

17:21Eric Cool. So as example, we have Argentina here and Mexico here. So Mexico has 2,000 points. Argentina has 2,200 points. So what does a 200 point gap actually buy?

17:38Eric So this curve, and this is a common curve in probability type systems, it's called the logistics curve. It's got a lot of like great properties. So I

17:50Thomas can already tell, I'm already looking at this curve and I'm thinking, look at these properties. These are great, great properties. Honestly, it's like Miami Beach with great properties.

18:01Eric Yeah. So one of these properties I like a lot is that zero maps to 0.5, so 0.5 probability 50%. So if we now think of what this curve's representing and we'll do that, just the difference between two teams ratings. So Argentina minus Mexico, the difference was like 2,000 or minus 2,000. So that would be kind of this blue line here. And we can kind of go up the curve and see it maps to 0.75, so 75%. So based on the rating difference alone, Argentina has a 75% chance to beat Mexico.

18:44Thomas So on the X is the point difference. Yeah. Okay.

18:48Eric Yeah.

18:48Thomas I got you.

18:49Eric So what are the properties that I like about this curve? So one, if you look at the Y axis, you can see that the bottom is zero and the top is one. So it maps to a zero one. So for probability, that's exactly what you need. It's also symmetric. So you can see that plus 200 is the reverse of minus 200. So if I take, run it through the logistics curve, add them up, I get to one, which is 100%. So all the probabilities add up nicely. So it has like a lot of these, like very nice characteristics.

19:21Thomas Does your wife know that you're into curves, like this model's curve to this level?

19:25Eric I don't know. I mean, it's a pretty curve.

19:29Thomas What are you talking about these curves?

19:31Eric If you go and actually chart the probabilities of the games, I went and made these like little red circles. That's the actual probabilities. So if I go and like, look at the ratings between the two teams and then see who actually wins, you can see my curve's not precise, but you know, it's like, how much work do you actually want to go through when the other guy's just going to vibe? You know, do we need to put forward all this effort?

19:59Thomas A straight line would be almost even hitting it more. So wait, the red is the outcome?

20:07Eric Yeah, the red is actual game. So I went and calculated the actual outcomes versus this logistics curve, just a function. And the reason it's nice is it's a very like well-behaved function. So if you take a linear function as example, and then I look at a minus 1000 ratings, it doesn't work because minus 1000 will end up not being like zero to one probability, right? Because it's linear. So at some point, if I have a line,

20:35Thomas at

20:35Eric some point, it's going to go below zero, which is

20:38Thomas not real.

20:40Eric So it's not well-behaved, which is why this function is actually used, unlike your linear functions that you thought might work.

20:47Thomas All right. No, I'm fine. Okay.

20:50Eric There's some slap in the system, but you know, it's like good enough for good enough for me, I think, to to win a point. So we'll see.

20:57Thomas Is that true? Are you really confident? Like, does that give you after seeing this and the fact your line doesn't hit reality? You think that's stronger than a vibe? A vibe?

21:08Eric Yeah, I think

21:09Thomas some of mine would be more wavy and less close to the reality is what you're probably thinking. See,

21:13Eric I don't think you know how to like tailor or like tune your vibes. Your vibes are just like so strong in your life that they just overwhelm like all sane responsible picks you might make.

21:29Thomas Or my vibes are so strong because they're been reinforced and are accurate and they're just right.

21:37Eric I mean, have you done an honest reflection on your life and thought things

21:41Thomas through? That's a different, it's a different video call I have weekly. All right, all right.

21:47Eric We'll keep going here then. Thanks for the

21:49Thomas equation, Eric. To Eric, I really was curious what E equal, so thanks for that.

21:55Eric Yeah, yeah, yeah.

22:01Eric So I mentioned it was a zero sum game, right? So a lot of the tuning comes in when like a team wins or loses, right? And the way I like to think about it is like the amount of points that should be transferred is based on the surprise. You know, it's like if the result is not a surprise, points should not really change hands, right? Like the system worked.

22:27Thomas Okay.

22:28Eric That makes sense.

22:29Thomas I'm with you, yeah.

22:31Eric And if it's a big upset, you'd expect a lot of points to change hands. And this is why I'm saying it actually self-tunes quite well because when a surprising game happens, it will adjust drastically if you have, you know, these parameters tuned, right? And I did read a little bit on FIFA's and, you know, I was like, even FIFA agrees, you know, that they use a similar rating system. And I went and found some of their parameters. I didn't use theirs because they're, I mean, it's FIFA. It's not a real prediction algorithm. Yeah,

23:05Thomas they're basically whoever goes most money to Infantino just gets the favor. Yeah,

23:09Eric yeah, yeah. We'll save the corruption for the next part in this series.

23:15Thomas Okay, great. Don't worry, it's still going to be referenced, but yes.

23:18Eric Yeah, good. But anyways, so the old score, the new score is equal to old score plus some waiting. So the waiting is based on the qualifier. So like say A wins in an upset. If it's just a friendly match, so I don't quite understand these terms, but they have games that are like friendlies.

23:39Thomas Yeah, it's like it's not related to a tournament at all.

23:43Eric And I think FIFA also has a waiting based on the margin. So it does try to account for like how big the margin of the game is because, I mean, soccer is very low scoring game, which is one of the reasons I was a little nervous about my rating system is like, I mean, if it's a 0-1 game, it's like one small away from ruining things, right?

24:10Thomas Yeah. Well, how was it a penalty? Did it go to penalties, which has a whole different implication? Another gap in the data, my friend.

24:18Eric But then I was thinking about that 0-1, like how is vibes any better? Like how do you possibly know that some random ball is going to get deflected and the goalie is going to miss it?

24:28Thomas Well, no, I look back and say, look, I remember the last three World Cups and I know that Mexico went to PKs with Germany and Germany won because their keeper stopped all of them. And that wasn't a fluke. So that's the same as Germany winning right out versus if it was the inverse and the team like snuck in and they won. And it was by the luck of the numbers of the gods that they won in PKs. Like, well, that team went deep. So in your model, they'd be like, oh, look, they got plus whatever. So you kind

25:01Eric of do this in your head. But

25:04Thomas you're

25:04Eric basing it more than just the outcome of the game, huh?

25:09Thomas What's the vibes? It's like, what's the read of the game? What's underneath the numbers? Because you want to do it at scale, which means you can't double click on everything or whatever, quadruple click. And I'm only going to reference lived memory for the most part where I can be like, oh, see that? And also like the vibes that are associated with it.

25:27Eric Yeah. But you just can't possibly know about all of the different games happening.

25:31Thomas It's not important because all that really matters is the players that are alive right now and the coach that's playing, not what the fuck happened in 1956. And so

25:39Eric I think it's the 1890 game. Germany was

25:43Thomas playing probably like with a literal like goat's bladder.

25:49Eric The result was zero one. It's important, Thomas. It's important.

25:55Thomas But weather, because different team styles played in rain versus not is like a huge factor.

26:02Eric Yeah, that's where my fancy model was going. Basically, I have the city of all the games and I was putting extracted variables. So what's the altitude? What's the weather? Stuff like this into new columns. I'd like three different like venue type variables that was then used to predict the outcome of the game.

26:29Thomas But again, you're saying it actually wasn't dramatically different enough to be worth all the effort and energy to crunch that data, which I do feel like is

26:36Eric what I've done a few percent. But at the end of the day, it's like when team and there's plenty of this when teams are so closely rated, it's like a coin flip. And there's already a lot of luck in soccer since it's such a low scoring game. I don't know if I should say luck, but there's a lot of like noise in the outcome of the games. Yeah,

26:55Thomas I think it's as if why crunch the data at all when you can just use your intuition? This is another word for vibration.

27:07Eric But then how would a non-sports guy be able to come and talk to a sports person about?

27:14Thomas Okay, so really the goal here is just for you to fit in. Is that

27:19Eric for me? I'm just trying to be cool, dude.

27:23Thomas We know

27:24Eric fit in.

27:24Thomas Yeah, we know.

27:26Eric Well, I think the other thing is I'm just showing how flexible this is. I can come into some sport that I know nothing about and I might have a chance to dethrone. How many years have you been a huge soccer fan?

27:39Thomas I've been a World Cup guy since 06.

27:42Eric Oh my God. How sad will it be if you lose to just someone that doesn't know what they're doing?

27:47Thomas Well, that's great. He's just crunched the numbers. I'm winning either way. Okay, so what you're trying to tell me.

27:56Eric It's the journey, not the destination. I feel like I'm skipping a few steps here.

28:02Thomas Well, again, the goal is for you to not even watch any of these games or enjoy them.

28:06Eric Why would I enjoy a game when I can just crunch numbers instead?

28:12Thomas Yeah, that is your game. That is your knockout round is creating the fucking elo for 32 teams. So what you're telling me here, Eric, is you can toggle the rating based off of each competition and some variables involved with them if it's upset.

28:37Eric Yeah, exactly. So I would go through the historical matches. I said I had like 50,000 roughly. And then I use these kind of, you know, I had a slightly tweaked version. Yeah, so anyways, we can go through each game and the beauty is we just figure out how many points trade between the two teams and we can make that happen. And I just keep a running track for all the teams, you know, at the end of the day. But, you know, my goal was the 32 teams that were coming in. Yeah,

29:07Thomas I'm with you. This one I have less problems with acknowledging the already stated issues I have with your faulty dataset.

29:17Eric It has a good interpretation. It naturally kind of self-corrects quite well. So that's why

29:24Thomas you're showing me is controlling for vibes.

29:35Eric Okay. So now, you know, I did hit a snag and I'll tell you that. And you kind of brought this up earlier is a lot of these games, like these teams don't play each other. They're like all over the place. So I went and calculated the connectivity graph of all of these different great games. So the connectivity graph is you can imagine like Cape Verde. So they have low connectivity with a lot of the other teams.

30:03Thomas Meaning like

30:04Eric they might have played a team like X. X might not have even played the big dogs. They might have played Y and finally Y had played the big dogs, right? So that's like what we call connectivity. So I went and calculated the connectivity and I tried to just drive to the different regions because a lot of these like regional areas, like they'll play teams close

30:25Thomas to each other. They're all based on the regions in the world.

30:29Eric Oh, that's why it's got such strong diagonal. Yeah,

30:32Thomas 100%. So, yeah, UEFA's Europe, CONMEBOL is South America. CONCACAF is like Caribbean and USAFC. I think it must be African. I don't know what CAF for OFC is, but one's got to be. This

30:45Eric doesn't even have a lot of games. So I don't know

30:47Thomas what this is. It might be Oceania or something. I'm glad to see you just did every game since 2010. You finally came to some reason here, which is good. It honestly really is for these regions because there's so many more countries that have been added, I'm sure, again since 1890. Okay. But yeah, that so if I... Which is the

31:08Eric US, Canada? Is there a North American region?

31:11Thomas That CONCACAF, the third row is North America. North America and the Caribbean. It's like the Confederation, Central America, Caribbean. I don't know.

31:21Eric Got it. So obviously they play a lot of games with each other. And then they have decent connectivity with CONMEBOL.

31:28Thomas South America, which is South America, which makes sense.

31:31Eric And then

31:31Thomas with Europe, which also makes sense.

31:34Eric Got it. And then these others, it's tails down. So, you know, the main thing is like when you have these teams that have low connectivity and they're doing very well, it's hard to rate them against each other. Which is one of the downsides of the ELO rating systems.

31:50Thomas In math, that's what we call quite incestuous data. Quite incestuous connectivity chart is what a lot of people...

31:57Eric I guess that's

31:58Thomas your term

31:59Eric if you want to be edgy.

32:01Thomas Your words, not mine.

32:04Eric Anyways,

32:05Thomas I

32:05Eric ended up throwing in a little extra number here and there to correct for some of this regional bias.

32:13Thomas And here's my question. Is the ELO score only based off the teams that are competing in this World Cup?

32:23Eric No, it's not. It's everything but four. It's everyone. And so as example, say you have like an island and I say island in a connectivity sense. Like a region that doesn't play other

32:37Thomas teams. I've heard you say it a long time. Keep going. I love it.

32:42Eric An island that doesn't play other teams. It's

32:45Thomas also a literal island because that's Oceania. Which is quite fun.

32:50Eric So say there is like three or four games played between the different regions. But they're not the team I care about. I might calculate a region correction factor. And then apply it to all people in that region. So it just helps spread some of this region strength into some of these teams that might kind of play other teams at all.

33:15Thomas But how can you determine strength? Like what do you mean because that region has teams with higher ELO scores?

33:23Eric Well like as example, so say this small region. They have a big team Brazil. They go and play other regions. But then in the island we see Brazil play these other teams. We can kind of figure out how like Cape Verde might fare against some of the other teams by looking at how Brazil fares against them.

33:46Thomas I think that's a huge problem in the data.

33:51Eric Yeah. So you don't think I should correct for region bias?

33:56Thomas No, I think you're conflating. I understand why. And the reason you're doing is there's a gap in the data. Which is an inherent problem with trying to use data for everything. Especially now we'll cover this 48 teams. Which means there's more teams from these underrepresented regions. So that's dunking into my initial bucket of the problem. I think you're conflating too hard. I think you're over correcting. If you in fact did use this. Because I don't think that's an actual good representation. Because it's too removed. It's two steps removed. There's so many different. There's too much conflating going on. And I think you're doing it to cover the gaps in the data set. And that again is the inherent problem data. This is a big... Or I won't say big. I won't be hyperbolic. I think this adds cumulatively to the problems with your model here, Eric. And your data.

34:49Eric Yeah, I mean sometimes you have to kind of cover up some holes in your data set. And work with what you got, right? And at the end of the day, these small teams, we don't have any ratings against them. So we only know how they do in their friendlies. But we know how their friendlies do against other regions.

35:09Thomas I get it. I mean, I respect it. Because it's like, you want to just use vibes here. But you're locked into your math. And so you're...

35:18Eric Well, yeah. And since it's a zero-sum game, a lot of the strength happens in the region. So a lot of the strength tuning is happening there. But this more crude region to region stuff is not happening as much. And

35:34Thomas that's a totally... And it's more pronounced in this World Cup than it ever has been literally in the other World Cup. Which again, makes me so much more... I'm at like 92.5% confidence I'm going to beat you. Especially after the debacle of the slide.

35:50Eric Yeah,

35:51Thomas it's very

35:51Eric complicated. I

35:52Thomas can keep track of 32 teams. 48. It's somewhere in between those two. So

35:57Eric you admit that at some point there's just too much to comprehend. I

36:02Thomas would have been more blasé in my application of strategy to each determination of who goes through. But now that I'm saying that, I could have just predicted which teams would have moved through. Because it wouldn't be predicting every game. It would have just been like who is into that round of 32. So actually I strike that. No, I still wouldn't fucking nailed it. So connectivity is a problem. You band-aid it by doing, my friend likes your friend, so I think we might be friends.

36:34Eric Yeah, that's kind of right.

36:39Okay,

36:43Eric so yeah, we had this one small tweak. And then I was like, okay, maybe a single number. It

36:53Thomas always goes this way. We have a lot

36:55Eric more relevant data here. We've got goals. And goals are very natural. So we could have two ratings. One attack, one defense. And we can imagine the goal for one side is based on their attack and the opponent's defense, right? So we have a very natural way to apply these two ratings to the outcomes that are tracked in all of our historical data set, right?

37:22Thomas The two outcomes being points scored and points allowed. Yeah, exactly. So

37:26Eric this is way better of a signal than who wins. Because it's like win, loss, or draw, right? You don't click a little bit more.

37:36Thomas Yeah, because

37:37Eric it's like zero one. It's like this is a very common outcome. And it doesn't

37:43Thomas tell you the whole story.

37:45Eric And you can adjust your update factors and oh, it's worth 30 points or 42. But at the end of the day, if we can apply them to their own ratings based on attack and defense, a lot more powerful, right?

37:57Thomas OK, I like it. I'm with you on this. You're finally improving something here. And you

38:01Eric know what? I took that and scratched it. I don't even need these fancy models.

38:07Thomas You fucking got me so good. Well, I don't

38:09Eric even need these models, all right? I think this would be a good model. When I went and calculated it, it was a slight bump up. But I don't like getting too fancy with this stuff. You fell

38:23Thomas back on elegance.

38:24Eric Yeah, I fell back on elegance. And I might do that again because there's something beautiful in the simplicity of just having a single number per team.

38:36OK,

38:41Eric so now

38:41Thomas I want

38:42Eric to discuss the luck aspect of this thing. So I want to try to, and I think I've got two slides on this, and I want to try to link that.

38:51Thomas Thanks, I was hoping there'd be more slides.

38:55Eric Yeah, so like a lot of the, so say we have like a team that has, they're equally rated, right? We'd expect it to basically be a coin flip.

39:09Thomas And I'm going to put a reality onto this. That would be like a Mexico-England who are playing in an hour and a half. Most people, maybe England's a little bit up, but they're playing in Mexico, so we'd call it a 50-50 maybe.

39:23Eric OK, and for a game to draw would be right along this line. So we're starting to look at goals now. I kind of prompted it with the last slide as like, maybe we use goals as our metric to tune the system. I didn't end up doing it, but I like to show this because it does show some interesting things here. So this line is the draw. So, you know, sometimes it's 4-4, sometimes it's 0-0. You do see like these things are bigger based on how

39:50Thomas many games. Lower scoring, yeah.

39:53Eric So this bottom one, this is the favorite goals, and this is the underdog's goals. So this is just raw stats based on the rating system that I was using. It's not like FIFA's rating system. It's the one that I was calculating at the beginning of the show. And it's also not accounting for how big the rating gap is. It's literally just a straight favorite. So if you're five points up, you're still a favorite,

40:16Thomas technically.

40:18Eric So in the next slide, I think I start to add some nuance to this. You know, the thing that I like to see is just like, it's not very stark. You know, and maybe it's because that I have small favorites in here. You know, maybe I should have excluded them, but like favorites get upset at a lot in soccer. And in particular, it's like there's a ton of games that's like 1-1, 0-1, 2-0. It's a low scoring game, which means like some small event affected the whole 90 minute game.

40:51Thomas Penalties, a red card. Yeah, like somebody getting an injury, their best player getting injured. God,

40:57Eric you know, that's why it's fun. And it's also like, you know, like I do feel like the gut has a chance here.

41:07Thomas Because you're living in a binary world that doesn't explain the outcome. The outcome is how

41:13Eric I would describe it. There's a small signal and a lot of noise, you know?

41:18Thomas Yeah, yeah, yeah, yeah, absolutely. Of course, yeah. Okay, so

41:22Eric the

41:24Thomas favorite team, you're not as likely to be successful as one would think if you're the favorite team.

41:32Eric You still see like a strong signal down here. You know, like the favorite has a lot stronger, like they have a lot better chance versus you look at this, you don't see them putting up big scores and stuff. So there is some good signal in here, but it's like, yeah, it's noisy. They only happen every four years, right?

41:51Thomas Sorry dude, you chose math, man.

42:01Eric We got to parse this one a bit because it gets a little complicated here because we're now thinking about the ratings. So you see the ratings here on the bottom, the ELO gap, 100 gap, 200 gap, 300 gap, 400, 500, 600 plus. So the gaps on the bottom and then the margin, the goal margins on the left. So zero is a draw. So this line across the middle is all the draw games, right?

42:27Thomas A single plot on this scatter plot, like the one you have highlighted, the Saudi Arabia, Argentina, that dot means Argentina was up 400.

42:42Eric Yeah, they'd be up 450.

42:44Thomas That's just the difference.

42:46Eric Yeah, the difference, yeah.

42:48Thomas Got it. So Argentina was 2,200 and Saudi Arabia was 1,800 or something like that.

42:52Eric That's why the rating system and the logistics curve is so beautiful. All that matters is the rating difference.

42:58Thomas Okay, all right. Yeah, I got you.

43:00Eric Yeah, so is it upset since it's below the line and it's red? And Saudi Arabia won,

43:07Thomas but they

43:07Eric weren't the favorite.

43:09Thomas Oh no, it's just goals. Okay, all right. I'm with you.

43:12Eric Yeah, this is goals. So anyone below the favorite lost, the above the favorite won. So it kind of maps the last slide, which is why I kind of used it to motivate this to kind of help. And you can see there is a lot more blue, but there is quite a bit in the red space. So again, we see the favorites not winning all the time. What we're doing now though, is it's kind of spread out. So you can see this rating difference, right? The first thing I wanted to point out is, do you see these like almost equal teams? By stats, the slight favorite loses 45% win rate.

43:52Thomas Okay. So yeah, the team is either evenly, but this is an evenly match. We're still saying.

43:59Eric If they're up like five points in that, you know,

44:02Thomas like 25 of a different,

44:06Eric yeah, zero to 50, maybe even zero to 50. They, if you're up and all

44:11Thomas your shit, man, that just right there alone. Cause that's wherever most people live. Anyway, it means any system you're turning, mommy, you're trying to run in this data means it's bullshit. So we can just, we can call it now or, you know, I'm good to hang out, but yeah, keep going.

44:28Eric But at the end of the day, it's like these is like on par teams. It's like, how do you rate an England, Mexico? You called those out as equals. What do you, how do you choose

44:37Thomas the, the intangibles? So they're playing in Azteca at the stadium, which is at like 7,000 feet. There's inclement weather. England has so much pressure and they have choked. And Harry Kane, who's your main score choked last time by missing a PK. So all vibes. And I think England's going to lose. So what

44:54Eric you're in the head of the coach and all of these soccer players, you really think you can get in their heads?

45:00Thomas I think I can get in their heads. And I think me thinking I'm in their heads gives me more of a solid foundation to make a prediction than data. That's not necessarily appropriate and basically says it's 50 50. Like if you're saying it's 50 50, what are you even going to say? Well, I'm interested to see actually what you said now that we're talking about. So in this case, it'd be like, obviously you're going to use indirect vibe reads to make a prediction. That's all you got. That's all you got.

45:31Eric Yeah. I mean, that would be a reasonable choice, but not what I made.

45:36Thomas You're locked contractually. You're locked in that you can't, but yeah.

45:41Eric That's a good slide. You do see like, once you get up to these bigger ratings, there's a lot fewer upsets. It's like,

45:47Thomas you

45:48Eric can feel pretty good about your decisions. But when I did my bracket, I was not feeling good about these teams that were very close, obviously. Is

45:58Thomas this all finals still since 1930 or World Cup matches?

46:02Eric Yeah, 964 games. So I think this is World Cup.

46:05Thomas That's all World Cup. All right. So you just said scatterplot. Most times it's 50 50 or inverse.

46:14Eric Um,

46:16Thomas and I see you signals real, just outnumbered by. And then the question obviously becomes in a World Cup, what's the average differential?

46:23Eric So, and you know, even 300 points, like you're a strong favorite, but you're still only winning three games out of four. It's still just like noisy. Disappointing.

46:34Thomas Fun. Other, you call it noisy. Other people call it fun, but yeah. This only strengthens my argument that if this is a dad you're working with, you need vibes. Actually, you need vibes to be able to interpret this data.

46:50Eric Yeah. That's why I'll be so shocking when I beat you.

46:57Eric All right. So and then some of these are, you know, this mentions the World Cup. I'm not sure if we actually get this one, but oh, actually the one that's down here, this might be

47:11Thomas seven. I bet you that was Germany beating Brazil seven one or what's.

47:17Eric Yeah, so differential

47:18Thomas would have been small. It would have been pretty close.

47:27Eric Look at that because you brought it up and I'm like, I saw that or I saw the stats of that game. I didn't actually see a game. Yeah.

47:36Thomas That's insane. You just read for 30 minutes. Pure a CSV column. Is he watching a game just like in the Matrix? You just scrolling a CSV spreadsheet. I

47:51Eric stared at this row for like 15 minutes. I was like, what a row.

47:55Thomas You're Neo. You're Neo. Yeah, that was that. God, how much does that cement my vibrate? God, you're so fucked, dude. You're so

48:06Eric fucked. Well, and this is such an interesting thing with the sport and the bracket tournament. It's not a multi game tournament. A lot of big sports things are multi game. Well, I'm thinking about NBA versus football. I think they do single game ones. Elimination. American football. But once you're in single elimination, it's like you can be the best, but you're not guaranteed to win. That's what I was trying to highlight

48:32Thomas with some of those. Any given Sunday, Eric, anything can happen.

48:36Eric The last five rolled cups. It seems like the best one only won once.

48:43Thomas So this is wait, can you walk me through this? Because this is actually just Brazil number

48:47Eric one is Brazil. Yeah. And they got knocked out in the quarterfinals. Take a look at this, though. So see right here. This is my calculated win probability. And the first game I have them winning 87%. And the second game I have them winning 60%. And the coin flip the other way.

49:07Thomas Well, oh my God. Here's another thing you probably never even considered the luck of the draw. So the lowest percentage win they had was the quarterfinal. So you have to pass through a harder, tighter gate than to ultimately even get to the finish.

49:25Eric Yeah, I

49:26Thomas did. Because it's called the group of death. You don't even fucking know about the group of death, do you, Eric? I

49:31Eric recognized the bracket was very lopsided in that Argentina has an extremely easy path except for Cape Verde. Yeah, weirdly. To the finals compared to the opposites out of the bracket. So I did. And the model just naturally accounts for that. So I didn't have to do anything fancy because math just works

49:50Thomas well. You don't have

49:52Eric to be such a vibes guy. You think you're vibes. Oh, I know how to read a bracket. Congratulations.

49:58Thomas The group of death is more diluted now with 48 teams because it's less likely that multiple top tiered or high ELO score teams would be in the same group.

50:10Eric The same group. Because then someone might get knocked out because it's such a strong group. Well, that's

50:15Thomas typically a group of death. Like you'd have Spain and Colombia like teams that would normally make it to the quarterfinals, but might not because they're all very in the same group. So where did you land? So you were like, I see this.

50:31Eric Let me point out one more math thing real quick before we go. So I want to show what this box is showing. So you see how all these numbers are quite high. You have 89%, 81%, 88. Still, so when you have like multiple of these things that like one loss, like you're out, you have to multiply these numbers together. So the actual chance based on all these win probabilities and you get that as single elimination. So it really makes it quite dire for these good teams. Like it's hard to, and I mean, you see it up here, here, one in five favorites wins, right? But that's what this number is showing. And you can see it's roughly around that one in five. You know, my numbers are showing a little higher than that, but statistically, I don't think it is, right? It's not

51:17Thomas that crazy that the top team doesn't go all the way through. So GD hard to do it. So

51:25Eric one in five versus my model is saying it's like 30%. So I'm more like one in four, one in three. So actually based on this, it seems like the odds should start going the other way. And the statistically favorite might win this one. That's how it works, right?

51:41Thomas No,

51:42Eric the opposite

51:43Thomas of that.

51:44Eric Fair point. And

51:46Thomas this gets to a theme I may bring up in future episodes. When you get multi-layered systems, what the use and value of data and predictions becomes less and less valuable. If you give me, so you could predict around a 32, probably. Well, the group stage is crazy. You could probably predict around a 32 to 16 with like 80% accurate accuracy, but as it gets to fewer and fewer, I mean, that's what we just showed and said. If my goal is to predict who's going to win the world cup, I save money, use vibes.

52:34Eric So this is where I'm wondering, so do you try to pick the winner and then fill out your bracket based on

52:40Thomas that question? Well, you know what? In my head, I did. And this, when I explain my own, I can talk about it, but I have an extreme bias towards Argentina since I live there and

52:50Eric think

52:51Thomas Messi's the best. So I was like, Argentina is going to win. And so I did, that did eventually come through, but no, I went left to right.

53:00Eric Yeah. Well, cause then you do get the strength of the bracket going through it and stuff. And you think through each matchup one at a time, basically.

53:08Thomas Yeah. So

53:09Eric that's what I do. I just do it very fast based on numerical functions.

53:19Thomas Okay. Sorry. So now we're going to throw it all on the table, right?

53:23Eric All right. Yeah. Let's see what the bracket predicts here. So let's start on the France Spain half. We've talked about it a little bit, but I believe this is the stronger half. Do you agree?

53:33Thomas Yes, I do agree. They have Spain, Belgium, Morocco, Portugal, Croatia, and Germany and France. Yeah. I'd agree. This is, this is probably the more, this side with more skilled teams.

53:47Eric That's what it seemed like to me. Cause on the other side, it seems like Argentina had a relatively easy path. If you see this, Argentina is quite a bit far ahead and I don't think they're that much better team than everyone. I think that it comes from the bracket. Their half of the bracket is easier.

54:08Thomas I'd agree with that. But let's be super clear. The one that you are calling your own is this.

54:14Eric This is my bracket. This is my bracket. I messed around with a bunch. At the end of the day, I started getting a little cocky. I was like, I don't need this. I don't need this. And I chose the most simple and elegant model I could think of, which is the single value per team rating model.

54:31Thomas Higher ELO score is the team that's going to win.

54:33Eric Yeah. Even when it's like one, five point. We'll look

54:37Thomas at France Spain, 51 49.

54:40Eric Yeah.

54:41Thomas That's fucking perfect.

54:43Eric I mean, so you kind of see how absurd it is, but I still have some confidence against the pure vibe approach.

54:50Thomas Dude, Spain, franchise in the 38 62 is not right. That's just not an accurate represent. Do you think

54:57Eric it's more fair?

54:59Thomas Yeah, it's closer for sure.

55:01Eric Let me see

55:02Thomas the other side of the bracket.

55:04Eric Yeah. So you should point out, you should critique my bracket here. Cause I think you pointed out a good thing. Like on this, like you think this is way off.

55:13Thomas And when I

55:14Eric read news articles, people were saying France, France, France.

55:17Thomas Is that what you call them? News articles.

55:21Eric What do you call them dude?

55:25Thomas Was that just newsarticle.com or?

55:35Thomas So there's an underlying theme and maybe if we want to jump into my bracket, we can. There's like three themes that guide my vibes here. One FIFA is the most corrupt organization in the world. They will absolutely change outcomes to whatever makes them the most money or seems the best. So if there's star players, they want to see your star teams. I kind of kidding, but not. And then I was proven correct with the turnover on the USA, not the red card flow. Valigan now being able to play. So that's an underlying thing. Second is I do like an underdog, but I'm biased against Spain and Argentina since I live there and I know people there. And I, like you were saying, I am over indexing on like their passion for the sport and thinking there may be better. So that is true. And then three, there's a lot of European teams that are like older golden generation. And this was in the group stage. So this is immediate feedback cycle in the group stage. A lot of these European big dog teams were getting beat by smaller teams. And it's because it's broader. They're getting exposed to other teams they haven't seen before. And they're taking advantage of the World Cup model, where if you just sit and play defense, you can push a team to get to penalties. And the group stages didn't go that way because you could drop, but you can just time a better team out. You don't need to beat them. If you could just wait and sell penalties when it's because penalties

57:00Eric are more of a coin flip and they're holding the coin flip.

57:03Thomas Exactly. So it's not necessarily the best team. It's who can outlast. So those are three foundational topics here.

57:11Eric Those make a lot of sense.

57:13Thomas And I don't know if you can model that.

57:19Eric Yeah. So the story you were just telling, let me go to it. I had an interesting question. So did Brazil actually go out in the first round?

57:28Thomas They beat Japan.

57:30Eric So

57:30Thomas Japan and Norway was the game that just happened.

57:33Eric That's interesting.

57:34Thomas So I was already predicting Brazil not to go far, though.

57:38Eric Yeah, you were. You were. Norway ended up going.

57:43Thomas And now they're going forward to the quarters. Because they beat

57:47Eric Brazil, actually.

57:48Thomas They beat Brazil right there, yeah. So this Mexico England game, for example, that's the 50-50. And this is all vibes I'm reading. So that's played at La Azteca Stadium in Mexico City, which is that elevation. There's severe weather happening. The last time England played there was the most infamous event in World Cup history, El Mano de Dio, which is when Diego Maradona cheated, had a handball, and scored and beat England in the quarterfinals, which was the closest they've ever gotten. England has created football. They're the most thirsty and desperate for bringing the cup home. They say it's coming home. They choke all the time. Harry Kane, the leading scorer. I already said all these things. Yeah, I

58:28Eric see that. No clutch, no clutch.

58:31Thomas Exactly. Yeah, yeah, yeah. So

58:32Eric the model had Mexico winning 33%, and you chose them where the model clearly didn't.

58:38Thomas Wait, my model or your model?

58:40Eric My model. I didn't even have them beating Ecuador.

58:46Thomas Because Ecuador coming in was higher rent. The biggest flaw you have here is the lack of updated information.

58:54Eric The recency.

58:55Thomas You spread back too far. You're not fast forwarding up.

58:59Eric I should have had more aggressive weightings towards the group. The group stage is the best data I had, and I knew that, and I just didn't even take advantage of it.

59:09Thomas Yeah, I think for me, my prediction that Brazil would lose to Japan was me believing in Japan and knowing that Brazil was weak and Japan was playing well. And then same with South Africa. I thought that they were on a good roll.

59:27Eric So see, you do have a good story, but it's not as confident because it didn't. You're right, you're right. Okay, so anything that's highlighted or underscored with this red, I had different.

59:40Thomas So you had Australia beating Egypt? What?

59:45Eric That's what ended up happening.

59:48Thomas So and you know what? I saw Australia play live and they won, but they're all defense. They had no attack. It was just on the counter and most saw. So Egypt had never won. So this is all 100% vibes I'm reading on. Mosul is here to perform. People want to see him. And Australia was never going to go far because they can't score goals. They just locked down and parked the bus.

60:12Eric Yeah, it's a good story. Yeah,

60:14Thomas that's where your data is coming through.

60:17Eric You had some Mexico. Did they win?

60:20Thomas Yes, they're

60:21Eric still one. So you have them going farther. Egypt and Egypt won.

60:26Thomas They win. Yeah,

60:27Eric Japan. So that was one that I got right. And you got.

60:31Thomas Yeah, I got you. Okay, so I'm two for I got two ones.

60:35Eric How about South Africa?

60:37Thomas No, that's wrong. They did not go through.

60:39Eric That's the only one on that side.

60:41Thomas So we're two for two. Right now we're even. I got two hard ones, right? And you got two hard ones, right?

60:50Eric Yeah, I got to figure out which games to watch that to see which ones I really need to win. So that I look better than you. So on this side, we talked about this one. So you have Japan versus Norway. And that

61:08Thomas game already happened. Yeah, so that's you were more right than me for that one.

61:13Eric Because you didn't have Japan. But Brazil, you said lost.

61:18Thomas Yeah, yeah, you're ****. You haven't gone on the semis, dude.

61:22Eric I was like when Argentina was about to lose. I was like, Oh my God. And then I was like, I bet you have them doing very well, too. Okay,

61:36Thomas so I think we've locked in the outcomes of the round of 32.

61:40Eric Okay, so here's the points for how far we are based on round of 32. Thomas has 12 banked and he has a max possible of 74. And the math bracket also has 12 banked and max possible of 74.

61:59Thomas So we're calling this a tie.

62:01Eric Yeah, yeah, we're going to keep going until someone has a complete victory. Like there's no possible way. It might happen at the end, but it might happen before. You know, I think we both have... Who do you have in your findings?

62:14Thomas I have Argentina. I have Argentina and Spain. So

62:18Eric that's the big one. It does

62:20Thomas Spain

62:21Eric get through or France? Yeah, I

62:24Thomas have France playing Spain. Yes, and I have Spain going forward. You have France going forward. So that will be...

62:31Eric Yeah, if one of those get knocked out early, that's gonna call it for us. So we're at a dead heat. We're at a dead heat. Math

62:39Thomas versus Vibes. Cheers, my friend.

62:42Eric Cheers.

63:08Cheers.