| YouTube: | https://youtube.com/watch?v=1ejtdt78rCo |
| Previous: | The Deep Sea Sub Built By a Cereal Company |
| Next: | (None) |
Categories
Statistics
| View count: | 708 |
| Likes: | 81 |
| Comments: | 11 |
| Duration: | 11:16 |
| Uploaded: | 2026-11-16 |
| Last sync: | 2026-11-16 18:15 |
Citation
| Citation formatting is not guaranteed to be accurate. | |
| MLA Full: | "The Chess Rating System That Predicts Everything." YouTube, uploaded by SciShow, 16 November 2026, www.youtube.com/watch?v=1ejtdt78rCo. |
| MLA Inline: | (SciShow, 2026) |
| APA Full: | SciShow. (2026, November 16). The Chess Rating System That Predicts Everything [Video]. YouTube. https://youtube.com/watch?v=1ejtdt78rCo |
| APA Inline: | (SciShow, 2026) |
| Chicago Full: |
SciShow, "The Chess Rating System That Predicts Everything.", November 16, 2026, YouTube, 11:16, https://youtube.com/watch?v=1ejtdt78rCo. |
You might have heard of the Elo rating if you’re into chess. Or women’s football. Or competitive Pokémon. Or even… fabric inspection. So how can a rating system for chess be so universal? And, /mathematically/, what does it mean when your Elo is… ee-/low/? Let’s get into it.
Hosted by: Johnny Dickson (he/him)
----------
Support us for $8/month on Patreon and keep SciShow going!
https://www.patreon.com/scishow
Or support us directly: https://complexly.com/support
Join our SciShow email list to get the latest news and highlights:
https://mailchi.mp/scishow/email
----------
Huge thanks go to the following Patreon supporters for helping us keep SciShow free for everyone forever: ., Adam Brainard, Alex Hackman, Bethany Matthews, Blood Doctor Kelly, Cye Stoner, Anita, Anne Herrington, Friso, Jeremy Mattern, Garrett Galloway, Chris Curry, Chris Mackey, Chris Peters, David Johnston, Eric Jensen, J.V. Rosenbalm, Jaap Westera, Jacob Puthoff, Jason A Saslow, Jon Coffman, Joseph Ruf, Jp Lynch, Kevin Bealer, Kevin Knupp, Lyndsay Brown, Matt Curls, Piya Shedden, Shaji John, Steve Gums, Timos Gies, Toyas Dhake, yeyette
----------
Looking for SciShow elsewhere on the internet?
SciShow Tangents Podcast: https://scishow-tangents.simplecast.com/
TikTok: https://www.tiktok.com/@scishow
Instagram: http://instagram.com/thescishow
Facebook: http://www.facebook.com/scishow
Bluesky: https://bsky.app/profile/scishow.bsky.social
#SciShow #science #education #learning #complexly
----------
Sources: https://docs.google.com/document/d/e/2PACX-1vSvnZqAJclQt1hhsnO19HivbWlZbxG9czVBTIp5OyY1UGepKb1cGapbME4Z2ssaNwPUL8K3wMeojLKf/pub
Hosted by: Johnny Dickson (he/him)
----------
Support us for $8/month on Patreon and keep SciShow going!
https://www.patreon.com/scishow
Or support us directly: https://complexly.com/support
Join our SciShow email list to get the latest news and highlights:
https://mailchi.mp/scishow/email
----------
Huge thanks go to the following Patreon supporters for helping us keep SciShow free for everyone forever: ., Adam Brainard, Alex Hackman, Bethany Matthews, Blood Doctor Kelly, Cye Stoner, Anita, Anne Herrington, Friso, Jeremy Mattern, Garrett Galloway, Chris Curry, Chris Mackey, Chris Peters, David Johnston, Eric Jensen, J.V. Rosenbalm, Jaap Westera, Jacob Puthoff, Jason A Saslow, Jon Coffman, Joseph Ruf, Jp Lynch, Kevin Bealer, Kevin Knupp, Lyndsay Brown, Matt Curls, Piya Shedden, Shaji John, Steve Gums, Timos Gies, Toyas Dhake, yeyette
----------
Looking for SciShow elsewhere on the internet?
SciShow Tangents Podcast: https://scishow-tangents.simplecast.com/
TikTok: https://www.tiktok.com/@scishow
Instagram: http://instagram.com/thescishow
Facebook: http://www.facebook.com/scishow
Bluesky: https://bsky.app/profile/scishow.bsky.social
#SciShow #science #education #learning #complexly
----------
Sources: https://docs.google.com/document/d/e/2PACX-1vSvnZqAJclQt1hhsnO19HivbWlZbxG9czVBTIp5OyY1UGepKb1cGapbME4Z2ssaNwPUL8K3wMeojLKf/pub
You might have heard of the Elo rating if you’re into chess.
Or women’s football. Or competitive Pokémon.
Or even… fabric inspection? This ubiquitous mathematical formula shows up in tons of competitive environments. It’s named after its creator, physicist Arpad Elo, a chess master aiming to develop a fair method of ranking chess players.
But it’s spread so far beyond just chess… it can practically tell you how likely you are to win anything at all. Not only that, but it’s broken containment. Far beyond competition, the Elo method can be applied to nearly anything with a head-to-head win/lose outcome.
Meaning it’s found its way into studies about fabric. And marine mammals. And more.
So how can a rating system for chess be so universal? And, mathematically, what does it mean when your Elo is… ee-low? This is how you can use the Elo rating to determine if your good at practically anything.
My name is Johnny, and this is SciShow. [♪INTRO] To understand how we got the modern Elo system, let’s start at the beginning. Chess was first played in the seventh century CE. But the story of Elo ratings starts way more recently, in the 1960’s.
Before 1960, chess players did have a rating system, called the Harkness system. But the community had some issues with it. For one thing, it ended up unevenly rewarding losses and hiding previous wins.
Elo proposed an improvement on this system, starting with a discussion of what a rating needs to actually do. The main point of a rating system is to provide a numerical way to compare or rank players. To do this, it must reward wins, not reward losses, and reflect the likelihood of a win between two players.
Because every player with an Elo rating will not play against every other player, there’s a bit of probability involved in computing this sort of value. Elo incorporated this in the form of a statistical basis for his scores. His initial method was based on the assumption that player “strength”, or likelihood to win, followed a normal distribution.
This is the classic “bell curve”, where most people fall near the middle. Elo used the differences in wins and losses to determine a new score for players after every game. Basically, the more you won, the more your rating would go up.
The equation computed a new rating after each game or tournament by adding two values to a player’s old rating. Here’s what you need to do to calculate your new rating. The first thing you’d add to the old rating is a term representing the player’s performance in the most recent event.
This is the difference between their wins and losses, multiplied by a special parameter meant to scale how much each event should affect a chess player’s rating. The second term you’d add is the sum of the differences between the player’s original rating and ALL their opponents’ original ratings. Hidden in this equation is some secret statistics, in the form of the two numbers modifying each term.
These come from Elo’s original plan for the rating: to make sure it could predict the likelihood that a player would win. These two scaling factors were chosen to approximate a fit of the equation to the curve representing the likelihood of a player winning, as a function of their and their opponent’s Elo scores. So if a player rated 1600 beat a player rated 1800, their updated scores using this equation would get slightly closer together, reflecting the fact that the 1600-rated player is a little more likely to win now, given new information from that game.
If you win more, number goes up, if you lose more, number goes down. And they’re scaled to reflect real-life winning and losing probabilities. Seems sensible, right?
Well, Elo wasn’t satisfied. He published yet another improvement to the rating system in the magazine Chess Life in 1967. Though it was present in the initial version, the new version really highlighted the use of probability theory at its base.
Unlike the previous version, which used an approximation of the probability of winning, in this one Elo computed the probabilities himself and put them into a table for everyone to use. Armed with this table, Elo presented a new version of the equation, more accurate to the probability distribution at its core: In this version, there’s only one term we’re adding to the player’s old rating, but it packs a real punch. This term includes the difference between the player’s actual wins and expected wins.
The expected win rate was taken from Elo’s win probability table. To scale this win discrepancy, the equation introduces a new parameter: K. This parameter is meant to represent the impact of a game or tournament’s winning rate based on the player’s experience.
Elo suggested that this should be different for new versus seasoned chess players, since new players’ ratings were likely to change faster as they learned. This parameter is a little odd, though, since it’s hard to really quantify how much a game should count toward a player’s rating. Elo used a variety of numbers for different levels, lowering K for more experienced players.
A common standard assignment of K was 32, which gave ratings similar to the earlier approximation system. One big benefit of these changes was that it generalized the rating system for games where someone played against an opponent with a way, way different score. The earlier system was what we call a local approximation.
For most common match-ups, it worked fine, but if a beginner went up against a master, it would stop working the way it was supposed to. Using a pre-made table makes ratings for this matchup possible to compute, however unlikely it may be. The system was further improved later by slightly changing one of the base assumptions: that everyone’s performance followed the normal distribution.
This assumption wasn’t too far off. They ended up using the logistic distribution, which makes the probabilities on the tails slightly higher. That reflects the fact that very low-rated beginner players and very high-rated expert players have a higher chance of success than the normal distribution indicates.
These changes get us to the modern version of the formula, the definitive how-good-you-are-at-chess score of today. But they also make the formula usable for other applications. Not only is it still used in chess, it’s also been used in other games, like
Magic: The Gathering and even in sports forecasting. Elo is everywhere these days: It’s at the chess game, it’s at the boxing match, it’s at the combination chess game and boxing match. Yes, chess boxing is a real thing that exists. For all of these new applications, there are often adaptations made to the system to better fit the use case.
In women’s football, for example, an adapted Elo is used that accounts for margin of victory and other factors that apply in sports, like home field advantage. It seems like Elo rating has a really high Elo rating at, well, being a rating. There are still some issues with it, though.
One is that the choice of K needs to be specified for every new application, and the logic behind it can seem a little arbitrary. In chess, Elo’s argument was that newer players will probably start with a lower score, but improve faster with each new game, since there’s just more room to improve. In that case, a higher K for newer players that decreases over time makes sense.
In a sport like football, though, it would be much harder to determine an experience level for the whole team to base K off of. Plus, experience is hardly the only factor affecting the outcome in most team sports. Another is that all the bells and whistles being added for new applications can reduce the simplicity of the original model, which was a huge point in its favor.
But most concerning to the pro card- and video- gamers of today: in competitive spaces for games like
Magic: The Gathering, players can reach an Elo rating they’re happy with and just stop playing to rest on their “e-laurels.” This keeps artificially high scores around for players who may not have competed in a long time. Games like Magic have implemented adjustments and replacements because of this. Despite these concerns, Elo remains a super useful method, and not just in sports and games. Over time, it has grown and evolved through the work of other scientists, just like it had while Elo himself was still making edits and improvements.
At its core, the Elo rating system is for comparing performance of two things where there is a binary outcome, like winning versus losing. But winning and losing a game isn’t the only sort of binary outcome there is. In a way, you can frame almost anything as a binary outcome.
And scientists have gotten pretty creative with this. Like, comparing better or worse products for quality assurance? They have an Elo for that. Specifically, for inspecting fabrics.
Inspecting fabrics for defects like holes or marks can be done manually and take forever. Or you can automate it! But how can we tell the computer to decide what kind of defect, if any, is present in an image of fabric?
Computer vision works well for fabrics that aren’t highly patterned, but can slip up when trying to differentiate between polka dots and a hole without a ton of data. That’s where Elo comes in. Researchers trained a model to use Elo ratings to keep track of how many defects, and what kind, a sample of fabric had based on images.
Certain defects were encoded as “strong players”, with others encoded as “weak players”, so as the model inspected more images from a sample, it built up a total score for a piece of fabric. Comparisons of different defects were then represented by their Elo score. A hole would be counted as a “weak” player compared to a lighter-looking defect, so sections of fabric that repeatedly lose to others can be identified as possibly having a hole.
By having pieces of fabric going head to head in competition, the computer can get an automated, quantitative answer to whether there’s a defect in a piece of fabric, saving someone hours of the day in their job. Elo is also used in education research. Don’t worry, it’s not for pitting students against each other.
Instead, it’s used to treat students answering questions as an adversarial exercise. It’s not student vs student or student vs teacher, but student vs problem. In modern education, with the internet and computers in the classroom, adaptive teaching is becoming more common.
That’s teaching by adjusting content and problem difficulty to a specific student’s needs. By treating a student answering a question as a game, and computing an Elo rating for both the student and the question after each so-called match, teachers can better understand which students struggle with which concepts. And as a bonus, they get a numerical score telling them which problems are too easy or too difficult.
This lets educators easily make changes to content as students learn. If applications in pedagogy aren’t your cup of tea, maybe you’d like to hear about how to use Elo ratings to determine reproductive success. Yes, really.
Well, for sea lions, anyway. See, in the animal kingdom, reproductive success is often based on competition. And as we know, this is where Elo ratings shine.
One paper used Elo ratings to quantify the dominance of male sea lions in mating behavior with other males when competing for females’ attention. Each time a pair of male sea lions interacted, the researchers recorded their behavior to identify a more territorial or aggressive “winner”, using these assignments to adjust Elo scores accordingly. They found that these Elo ratings were a pretty good reflection of overall social dominance in the sea lion territory.
And for these marine mammals, social dominance is a big factor in which males mate successfully. As much as we’d like to believe otherwise, in the human part of the animal kingdom, this still kind of applies. That’s why dating apps like Tinder have used the Elo system to guide matches as well.
But since we like to believe our dating life is a little more complex than sea lions’, dating apps have shifted away from Elo. But to this day, Elo is holding its own in the chess world. So still part of matching up kings, queens, and monarchs everywhere; just on the chessboard, instead of on dating apps.
In our competitive world, we’re always looking to figure out who’s good, better, and best at things, at stuff! While the math on these rankings can get pretty complicated, there’s some surprisingly simple and elegant methods out there that are widely used, like the Elo rating. If you want to compare fabric, sea lions, or your own ability to win at pretty much anything, Elo ratings are a solid place to start.
There’s a reason the method has spread far beyond its very specific origins. All the way to places like the chess-boxing ring, where I hear it’s a real knockout. [♪OUTRO]
Or women’s football. Or competitive Pokémon.
Or even… fabric inspection? This ubiquitous mathematical formula shows up in tons of competitive environments. It’s named after its creator, physicist Arpad Elo, a chess master aiming to develop a fair method of ranking chess players.
But it’s spread so far beyond just chess… it can practically tell you how likely you are to win anything at all. Not only that, but it’s broken containment. Far beyond competition, the Elo method can be applied to nearly anything with a head-to-head win/lose outcome.
Meaning it’s found its way into studies about fabric. And marine mammals. And more.
So how can a rating system for chess be so universal? And, mathematically, what does it mean when your Elo is… ee-low? This is how you can use the Elo rating to determine if your good at practically anything.
My name is Johnny, and this is SciShow. [♪INTRO] To understand how we got the modern Elo system, let’s start at the beginning. Chess was first played in the seventh century CE. But the story of Elo ratings starts way more recently, in the 1960’s.
Before 1960, chess players did have a rating system, called the Harkness system. But the community had some issues with it. For one thing, it ended up unevenly rewarding losses and hiding previous wins.
Elo proposed an improvement on this system, starting with a discussion of what a rating needs to actually do. The main point of a rating system is to provide a numerical way to compare or rank players. To do this, it must reward wins, not reward losses, and reflect the likelihood of a win between two players.
Because every player with an Elo rating will not play against every other player, there’s a bit of probability involved in computing this sort of value. Elo incorporated this in the form of a statistical basis for his scores. His initial method was based on the assumption that player “strength”, or likelihood to win, followed a normal distribution.
This is the classic “bell curve”, where most people fall near the middle. Elo used the differences in wins and losses to determine a new score for players after every game. Basically, the more you won, the more your rating would go up.
The equation computed a new rating after each game or tournament by adding two values to a player’s old rating. Here’s what you need to do to calculate your new rating. The first thing you’d add to the old rating is a term representing the player’s performance in the most recent event.
This is the difference between their wins and losses, multiplied by a special parameter meant to scale how much each event should affect a chess player’s rating. The second term you’d add is the sum of the differences between the player’s original rating and ALL their opponents’ original ratings. Hidden in this equation is some secret statistics, in the form of the two numbers modifying each term.
These come from Elo’s original plan for the rating: to make sure it could predict the likelihood that a player would win. These two scaling factors were chosen to approximate a fit of the equation to the curve representing the likelihood of a player winning, as a function of their and their opponent’s Elo scores. So if a player rated 1600 beat a player rated 1800, their updated scores using this equation would get slightly closer together, reflecting the fact that the 1600-rated player is a little more likely to win now, given new information from that game.
If you win more, number goes up, if you lose more, number goes down. And they’re scaled to reflect real-life winning and losing probabilities. Seems sensible, right?
Well, Elo wasn’t satisfied. He published yet another improvement to the rating system in the magazine Chess Life in 1967. Though it was present in the initial version, the new version really highlighted the use of probability theory at its base.
Unlike the previous version, which used an approximation of the probability of winning, in this one Elo computed the probabilities himself and put them into a table for everyone to use. Armed with this table, Elo presented a new version of the equation, more accurate to the probability distribution at its core: In this version, there’s only one term we’re adding to the player’s old rating, but it packs a real punch. This term includes the difference between the player’s actual wins and expected wins.
The expected win rate was taken from Elo’s win probability table. To scale this win discrepancy, the equation introduces a new parameter: K. This parameter is meant to represent the impact of a game or tournament’s winning rate based on the player’s experience.
Elo suggested that this should be different for new versus seasoned chess players, since new players’ ratings were likely to change faster as they learned. This parameter is a little odd, though, since it’s hard to really quantify how much a game should count toward a player’s rating. Elo used a variety of numbers for different levels, lowering K for more experienced players.
A common standard assignment of K was 32, which gave ratings similar to the earlier approximation system. One big benefit of these changes was that it generalized the rating system for games where someone played against an opponent with a way, way different score. The earlier system was what we call a local approximation.
For most common match-ups, it worked fine, but if a beginner went up against a master, it would stop working the way it was supposed to. Using a pre-made table makes ratings for this matchup possible to compute, however unlikely it may be. The system was further improved later by slightly changing one of the base assumptions: that everyone’s performance followed the normal distribution.
This assumption wasn’t too far off. They ended up using the logistic distribution, which makes the probabilities on the tails slightly higher. That reflects the fact that very low-rated beginner players and very high-rated expert players have a higher chance of success than the normal distribution indicates.
These changes get us to the modern version of the formula, the definitive how-good-you-are-at-chess score of today. But they also make the formula usable for other applications. Not only is it still used in chess, it’s also been used in other games, like
Magic: The Gathering and even in sports forecasting. Elo is everywhere these days: It’s at the chess game, it’s at the boxing match, it’s at the combination chess game and boxing match. Yes, chess boxing is a real thing that exists. For all of these new applications, there are often adaptations made to the system to better fit the use case.
In women’s football, for example, an adapted Elo is used that accounts for margin of victory and other factors that apply in sports, like home field advantage. It seems like Elo rating has a really high Elo rating at, well, being a rating. There are still some issues with it, though.
One is that the choice of K needs to be specified for every new application, and the logic behind it can seem a little arbitrary. In chess, Elo’s argument was that newer players will probably start with a lower score, but improve faster with each new game, since there’s just more room to improve. In that case, a higher K for newer players that decreases over time makes sense.
In a sport like football, though, it would be much harder to determine an experience level for the whole team to base K off of. Plus, experience is hardly the only factor affecting the outcome in most team sports. Another is that all the bells and whistles being added for new applications can reduce the simplicity of the original model, which was a huge point in its favor.
But most concerning to the pro card- and video- gamers of today: in competitive spaces for games like
Magic: The Gathering, players can reach an Elo rating they’re happy with and just stop playing to rest on their “e-laurels.” This keeps artificially high scores around for players who may not have competed in a long time. Games like Magic have implemented adjustments and replacements because of this. Despite these concerns, Elo remains a super useful method, and not just in sports and games. Over time, it has grown and evolved through the work of other scientists, just like it had while Elo himself was still making edits and improvements.
At its core, the Elo rating system is for comparing performance of two things where there is a binary outcome, like winning versus losing. But winning and losing a game isn’t the only sort of binary outcome there is. In a way, you can frame almost anything as a binary outcome.
And scientists have gotten pretty creative with this. Like, comparing better or worse products for quality assurance? They have an Elo for that. Specifically, for inspecting fabrics.
Inspecting fabrics for defects like holes or marks can be done manually and take forever. Or you can automate it! But how can we tell the computer to decide what kind of defect, if any, is present in an image of fabric?
Computer vision works well for fabrics that aren’t highly patterned, but can slip up when trying to differentiate between polka dots and a hole without a ton of data. That’s where Elo comes in. Researchers trained a model to use Elo ratings to keep track of how many defects, and what kind, a sample of fabric had based on images.
Certain defects were encoded as “strong players”, with others encoded as “weak players”, so as the model inspected more images from a sample, it built up a total score for a piece of fabric. Comparisons of different defects were then represented by their Elo score. A hole would be counted as a “weak” player compared to a lighter-looking defect, so sections of fabric that repeatedly lose to others can be identified as possibly having a hole.
By having pieces of fabric going head to head in competition, the computer can get an automated, quantitative answer to whether there’s a defect in a piece of fabric, saving someone hours of the day in their job. Elo is also used in education research. Don’t worry, it’s not for pitting students against each other.
Instead, it’s used to treat students answering questions as an adversarial exercise. It’s not student vs student or student vs teacher, but student vs problem. In modern education, with the internet and computers in the classroom, adaptive teaching is becoming more common.
That’s teaching by adjusting content and problem difficulty to a specific student’s needs. By treating a student answering a question as a game, and computing an Elo rating for both the student and the question after each so-called match, teachers can better understand which students struggle with which concepts. And as a bonus, they get a numerical score telling them which problems are too easy or too difficult.
This lets educators easily make changes to content as students learn. If applications in pedagogy aren’t your cup of tea, maybe you’d like to hear about how to use Elo ratings to determine reproductive success. Yes, really.
Well, for sea lions, anyway. See, in the animal kingdom, reproductive success is often based on competition. And as we know, this is where Elo ratings shine.
One paper used Elo ratings to quantify the dominance of male sea lions in mating behavior with other males when competing for females’ attention. Each time a pair of male sea lions interacted, the researchers recorded their behavior to identify a more territorial or aggressive “winner”, using these assignments to adjust Elo scores accordingly. They found that these Elo ratings were a pretty good reflection of overall social dominance in the sea lion territory.
And for these marine mammals, social dominance is a big factor in which males mate successfully. As much as we’d like to believe otherwise, in the human part of the animal kingdom, this still kind of applies. That’s why dating apps like Tinder have used the Elo system to guide matches as well.
But since we like to believe our dating life is a little more complex than sea lions’, dating apps have shifted away from Elo. But to this day, Elo is holding its own in the chess world. So still part of matching up kings, queens, and monarchs everywhere; just on the chessboard, instead of on dating apps.
In our competitive world, we’re always looking to figure out who’s good, better, and best at things, at stuff! While the math on these rankings can get pretty complicated, there’s some surprisingly simple and elegant methods out there that are widely used, like the Elo rating. If you want to compare fabric, sea lions, or your own ability to win at pretty much anything, Elo ratings are a solid place to start.
There’s a reason the method has spread far beyond its very specific origins. All the way to places like the chess-boxing ring, where I hear it’s a real knockout. [♪OUTRO]







