Elo Ratings - Division 1

DubStat

Ohio wrestling analytics
Hello Ohio wrestling community,

I am excited to share with you a new Elo rating system that I have created specifically for Ohio High School wrestling. As many of you know, Elo ratings are mathematical models commonly used in sports to measure the skill level of competitors or teams.

Here's how it works: Each wrestler will begin with a base rating, determined by their past performance and other pertinent factors. After every match, the rating will be updated based on the outcome of the match, the caliber of their opponent, and the level of dominance they achieved (pin, tech, major, dec). To ensure the accuracy of our results, I'm using data from JakesWrestling.com, which contains a majority of matches across the state. However, please note that some matches might be missing, such as dual meets and tournaments not uploaded to common wrestling websites like TrackWrestling or Baumspage.

If you find this to be a valuable resource, I plan on running these ratings weekly beginning next season.

Your feedback and suggestions are welcome as I work to refine this system and make it even better.

I've attached the full end of year ratings for all weight classes in Division 1 - other divisions to follow shortly.
 

Attachments

  • D1 - Elo Ratings - DubStat.xlsx
    69.7 KB · Views: 441
Last edited:
 
Jakes has a ton of matches missing so not sure that this will be all that accurate.

One thing I noticed is that for my son's team, a wrestler is rated lower than his backup who he pinned in wrestle offs.
 
Jakes has a ton of matches missing so not sure that this will be all that accurate.

One thing I noticed is that for my son's team, a wrestler is rated lower than his backup who he pinned in wrestle offs.
Jakes has a good amount of data but there are some parts of the state that don’t have as many results as others. I’m hoping we can encourage people to send more tournament and dual meet results to Jake throughout the year, which I believe will benefit Ohio high school wrestling in general. I’m hoping these ratings will be a catalyst for getting more complete results.

If you’d like to dm me the kids on your sons team, I can take a look and see why the algorithm is preferring the backup.
 
Jakes has a good amount of data but there are some parts of the state that don’t have as many results as others. I’m hoping we can encourage people to send more tournament and dual meet results to Jake throughout the year, which I believe will benefit Ohio high school wrestling in general. I’m hoping these ratings will be a catalyst for getting more complete results.

If you’d like to dm me the kids on your sons team, I can take a look and see why the algorithm is preferring the backup.
Jake's is a great tool but there are alot of results statewide that don't get reported in. That includes most dual meets and some of the smaller tournaments. I looked at his website a bunch over the years for some reference but it definitely doesn't tell the whole story. Some districts are tougher than others and so there are kids that would've made state had they been in another district so some kids that finished on the podium at state or were qualifiers are behind someone who didn't qualify in your metrics.
 
It's not just duals that don't go to jakes. I believe that tournaments on Flo aren't captured there as well.

There are tournaments not in there that were on track as well.
 
One weight class I noticed that was hard to understand the rankings was D2 175, a state qualifier was ranked 34. How does a kid that makes the top 16 come in at 34? I get that there were probably other kids better from other districts, but 18 others doesn't make sense.
 
Hello Ohio wrestling community,

I am excited to share with you a new Elo rating system that I have created specifically for Ohio High School wrestling. As many of you know, Elo ratings are mathematical models commonly used in sports to measure the skill level of competitors or teams.

Here's how it works: Each wrestler will begin with a base rating, determined by their past performance and other pertinent factors. After every match, the rating will be updated based on the outcome of the match, the caliber of their opponent, and the level of dominance they achieved (pin, tech, major, dec). To ensure the accuracy of our results, I'm using data from JakesWrestling.com, which contains a majority of matches across the state. However, please note that some matches might be missing, such as dual meets and tournaments not uploaded to common wrestling websites like TrackWrestling or Baumspage.

If you find this to be a valuable resource, I plan on running these ratings weekly beginning next season.

Your feedback and suggestions are welcome as I work to refine this system and make it even better.

I've attached the full end of year ratings for all weight classes in Division 1 - other divisions to follow shortly.
Appreciate all the effort you put into this. I was wondering if the computer is putting more weight on Ohio matches versus national level tournaments such as Powerde? for example D1 113. How does the computer put Rohr ahead of Brown who majored him at States 12-4?
 
Appreciate all the effort you put into this. I was wondering if the computer is putting more weight on Ohio matches versus national level tournaments such as Powerde? for example D1 113. How does the computer put Rohr ahead of Brown who majored him at States 12-4?
Yes, one of the things that I’ll have to somehow handle differently is the out of state matches. Right now the algorithm doesn’t recognize the skill level of the national level guys which hurts the St Ed’s guys the most. They are consistently being under rated due to their out of state schedule. Good catch
 
Yes, one of the things that I’ll have to somehow handle differently is the out of state matches. Right now the algorithm doesn’t recognize the skill level of the national level guys which hurts the St Ed’s guys the most. They are consistently being under rated due to their out of state schedule. Good catch
Yes this is the issue with going analytical. it doesnt weigh all of the criteria appropriately like what we can determine based off the eye test. The stats are cool and a lot of work to formulate. but this is more precise than accurate, a way to fix the criteria and incorporate it into the algorithm would be to reward and weigh the national level competition differently than the OSHAA matches.

Like, if a wrestlers wrestles 20x over the summer and goes 15-5. you can add priority to the 15 wins and each win is .2 = to 3.
Even a wrestler who wrestles 40x over the summer and goes 25-15, 25*.2= 5. this doesn't qualify the level of competition but does account for who has wrestled the most over the summer and presumably would have an advantage going into the OHSAA season. ADD this to the final result and you will get more accurate data field because it encompasses more tangible value.,. just an example and numbers off of my head easy to follow for others.

You are definitely on the right track just need to pick and choose criteria that can vastly separate the field yet be inclusive to where all kids can have the chance. Summer wrestling is just like extra credit, available to all athletes but the ones who take advantage of it, see the results on the Mat.
 
Hello Ohio wrestling community,

I am excited to share with you a new Elo rating system that I have created specifically for Ohio High School wrestling. As many of you know, Elo ratings are mathematical models commonly used in sports to measure the skill level of competitors or teams.

Here's how it works: Each wrestler will begin with a base rating, determined by their past performance and other pertinent factors. After every match, the rating will be updated based on the outcome of the match, the caliber of their opponent, and the level of dominance they achieved (pin, tech, major, dec). To ensure the accuracy of our results, I'm using data from JakesWrestling.com, which contains a majority of matches across the state. However, please note that some matches might be missing, such as dual meets and tournaments not uploaded to common wrestling websites like TrackWrestling or Baumspage.

If you find this to be a valuable resource, I plan on running these ratings weekly beginning next season.

Your feedback and suggestions are welcome as I work to refine this system and make it even better.

I've attached the full end of year ratings for all weight classes in Division 1 - other divisions to follow shortly.
Great job. I am actually kind of an expert in rating systems, and I must say I am surprised that the end results are this accurate. You may or may not know but one of the biggest drawbacks to using the ELO system for a single wrestling season is the lack of total matches. In chess, you get rated over the course of your whole career, so we are talking about hundreds of matches. And in the early parts of your career you are considered "unrated" because your lack of matches makes your initial 1500 rating very unreliable. Something you might know that will help you is that the ELO system can be modified in a bunch of different ways - maybe some of which can be helpful, if you have the data. For example, instead of awarding 1 point for a win and 0 points for a loss, you could have a range of outcomes based on the result detail of each match. Examples could be 1 point for a pin, 0.9 for a techfall, 0.8 for a major, and 0.7 for a decision. That may sound strange to people not familiar with how the ELO system works, but if you have the data, that might be a more accurate way of doing it since actual results might more closely match the calculated "Expected Result" ultimately giving you more stable numbers. Anyway I'm glad someone actually did the work of plugging this stuff into the formula. Pretty awesome.
 
Always wondered if anyone ever looked at trying something like this, but figured there was no central database with enough results to create a model that would make sense. This is fantastic, and would only be stronger with more results. Being interested in statistics, but not an ELO expert at all, how would this system rank wrestlers who were 18-9 and 24-12, wrestling/winning/losing against similarly-ranked opponents - i.e. is there any volume bias at all? Also wondering, how does the Jakes site aggregate the data? Do they all get reported by third parties, or does the admin actually spend time hunting all that down themselves on Track, Baums, etc.? Too bad most dual meets aren't reported, although looks like most/all of the top-5/10 school duals are in the db at a quick glance. Also assume wrestlers who bounced between classes aren't ranked any differently in the model - i.e. someone wrestles 106 40% of their matches, 113 for 60%, is in the 106 tab with an ELO of 1600, they would have the same rating if classified as a 113? Sorry for the questions, but this is pretty interesting to me.
 
Always wondered if anyone ever looked at trying something like this, but figured there was no central database with enough results to create a model that would make sense. This is fantastic, and would only be stronger with more results. Being interested in statistics, but not an ELO expert at all, how would this system rank wrestlers who were 18-9 and 24-12, wrestling/winning/losing against similarly-ranked opponents - i.e. is there any volume bias at all? Also wondering, how does the Jakes site aggregate the data? Do they all get reported by third parties, or does the admin actually spend time hunting all that down themselves on Track, Baums, etc.? Too bad most dual meets aren't reported, although looks like most/all of the top-5/10 school duals are in the db at a quick glance. Also assume wrestlers who bounced between classes aren't ranked any differently in the model - i.e. someone wrestles 106 40% of their matches, 113 for 60%, is in the 106 tab with an ELO of 1600, they would have the same rating if classified as a 113? Sorry for the questions, but this is pretty interesting to me.
You bring up an excellent point. There is an old database for boxing called boxrec.com that uses a very different formula for their rating system than ELO but shares one main similarity - it’s a zero-sum point system (meaning the points the winning fighter gets is the same points the loser loses), and one thing they do is apply an algorithm to the rating score of a boxer when they fight up in weight. I don’t know the exact formula but let’s say you are a 106 wrestler with a 1600 rating, and your coach tells you at a dual that you must wrestle up a match at 113. In the boxrec system, the 106 would have an algorithm applied to the 106 that would lower his rating for that match only. The next match if that 106 goes back to the 106 weight class, the algorithm is reversed and his rating is bumped back up. If a kid wrestled half the year at one weight class then bumps down permanently, the algorithm is applied just that one time. It’s not a huge bump, but the purpose is to eliminate any disadvantage a wrestler may get by being asked to bump up. And mathematically it’s sound - everyone knows that you have a slight advantage if you’re at a higher weight class if all else is equal, and the ELO and similar systems calculate a prediction of performance, then compares the probability with the actual performance, with any deviation requiring an update. As I mentioned in my previous post, modifying the ELO ratings to include result detail (DEC, MajDec, TechFall, Fall) would mean that the system would essentially calculate a prediction of result, meaning wrestler A with one rating should Major wrestler B based on his lower rating. If that result happens, then very little changes in each of the wrestler’s rating. But if wrestler A gets a tech fall, this would be seen as a deviation from the expected result, leading to an adjustment in the wrestlers ratings. With that said, you can see how bumping a kid to a different weight class SHOULD result in a change to their rating because maybe without it, the expected result would be a decision, but the actual result ends up a major due to the smaller kid being out muscled…

On a side note, years ago I wanted to apply a modified ELO system to youth wrestling where two different algorithms were available - one for change in weight class, and the other if a kid bumps up to wrestle in a different age group - which at the youth level is a much more stark disadvantage than just weight. The cool thing about the algorithm is that it doesn’t matter what the weight classes are. If a kid that wrestled at 55 pounds one week and 57 pounds the next week it applies every time, even though 2 pounds wouldn’t move the needle much.
 
Top