Ratings - please read before commenting - Archived

[computerrank]

The WHR current divisional ratings are objectively better than the old ratings. So Boxrec will stay with them.
The p4p ratings and the all-time ratings are just for fun. They will be tweaked later.

Q1: 'Objectively better' Is that in your assessment only?
Q2: Will you implement a toggle so we can at least have the option between these and the seemingly more popular, and in my assessment 'objectively' better, previous ratings?

Both already is posted:
- Q1: yes, better likelihood for all results and better prediction power
- Q2: a toggle with multiple options is possible and on my list, but there is no decision about priorities

[Manrae]

It has been a long time since I've been here and I actually semi-stopped following boxing for a little while, but, quarantine...

So I decided to take a trip back and ...my, my, how things have changed

I only have 3 questions:

1. Are the ratings next to the boxers representing their "pre" or "post" fight rating? I think that it's "post-fight" but I just want to be sure.

2. Have the annual ratings been updated to reflect the new rating system?

3. For the All Time Rankings, it seems like a boxer's peak is used as placement. However, I can't understand Marcel Thil's placement at 638.4 when he never peaked that high (189.2 - high). Can you explain it to me, please? Is the peak only used for Heavyweight boxers? If so, why is that?

[computerrank]

It has been a long time since I've been here and I actually semi-stopped following boxing for a little while, but, quarantine...

So I decided to take a trip back and ...my, my, how things have changed

I only have 3 questions:

1. Are the ratings next to the boxers representing their "pre" or "post" fight rating? I think that it's "post-fight" but I just want to be sure.

2. Have the annual ratings been updated to reflect the new rating system?

3. For the All Time Rankings, it seems like a boxer's peak is used as placement. However, I can't understand Marcel Thil's placement at 638.4 when he never peaked that high (189.2 - high). Can you explain it to me, please? Is the peak only used for Heavyweight boxers? If so, why is that?

- the ratings are at fight time - whole history ratings balance the rating at fight time regarding the result and all later and prior results with decreasing weight
- annual ratings from old system - not updated yet
- all-time ratings are peak ratings with division dependent weight factor - Heavyweight with weight 1
- for p4p ratings and all-time ratings these weight factors may be revised

[Manrae]

It has been a long time since I've been here and I actually semi-stopped following boxing for a little while, but, quarantine...

So I decided to take a trip back and ...my, my, how things have changed

I only have 3 questions:

1. Are the ratings next to the boxers representing their "pre" or "post" fight rating? I think that it's "post-fight" but I just want to be sure.

2. Have the annual ratings been updated to reflect the new rating system?

3. For the All Time Rankings, it seems like a boxer's peak is used as placement. However, I can't understand Marcel Thil's placement at 638.4 when he never peaked that high (189.2 - high). Can you explain it to me, please? Is the peak only used for Heavyweight boxers? If so, why is that?

- the ratings are at fight time - whole history ratings balance the rating at fight time regarding the result and all later and prior results with decreasing weight
- annual ratings from old system - not updated yet
- all-time ratings are peak ratings with division dependent weight factor - Heavyweight with weight 1
- for p4p ratings and all-time ratings these weight factors may be revised

I’m sorry but I still don’t quite understand what the rating is supposed to represent “at fight”. Like, depending on what a fighter does before & after a fight, can change their “at fight” rating? Even still, does that number reflect their value before or after the fight?

For example: Hagler 76.63 vs Leonard 74.21... are these numbers reflecting their pre or post bout value?

I’m just trying to understand why there are increases and decreases of the ratings of fighters, especially concerning activity/inactivity...

[computerrank]

- the ratings are at fight time - whole history ratings balance the rating at fight time regarding the result and all later and prior results with decreasing weight
- annual ratings from old system - not updated yet
- all-time ratings are peak ratings with division dependent weight factor - Heavyweight with weight 1
- for p4p ratings and all-time ratings these weight factors may be revised

I’m sorry but I still don’t quite understand what the rating is supposed to represent “at fight”. Like, depending on what a fighter does before & after a fight, can change their “at fight” rating? Even still, does that number reflect their value before or after the fight?

For example: Hagler 76.63 vs Leonard 74.21... are these numbers reflecting their pre or post bout value?

I’m just trying to understand why there are increases and decreases of the ratings of fighters, especially concerning activity/inactivity...

at fight means at fight date after the fight

[jerd16]

Q1: 'Objectively better' Is that in your assessment only?
Q2: Will you implement a toggle so we can at least have the option between these and the seemingly more popular, and in my assessment 'objectively' better, previous ratings?

Both already is posted:
- Q1: yes, better likelihood for all results and better prediction power
- Q2: a toggle with multiple options is possible and on my list, but there is no decision about priorities

Well, a toggle I think would satisy everyone. It allows for users a preference and to easily distinguish between a rating at that time (with all we knew at that time) and rating with hindsight (what we know now).

My last question: is there any definable scaling between the previous set of ratings and these (apologies if my statistical terminology is not correct). I.e If the average fighter had 5 points in the previous ratings, under the new formula what would the average fighter's rating be? And does this scale/ratio apply consistently?

[computerrank]

Both already is posted:
- Q1: yes, better likelihood for all results and better prediction power
- Q2: a toggle with multiple options is possible and on my list, but there is no decision about priorities

Well, a toggle I think would satisy everyone. It allows for users a preference and to easily distinguish between a rating at that time (with all we knew at that time) and rating with hindsight (what we know now).

My last question: is there any definable scaling between the previous set of ratings and these (apologies if my statistical terminology is not correct). I.e If the average fighter had 5 points in the previous ratings, under the new formula what would the average fighter's rating be? And does this scale/ratio apply consistently?

Well, the star ratings in both systems rely on the same structure. 5 stars are best 100 active boxers, 4 stars,are best 300 active boxers, 3 stars are best 900 active boxers....
I will look and give this comparision here ...
But the ranks especially in the lowers ranks are NOT the same for each boxer.

The new system doesn't need some previously needed ad hoc parameters
- no additional points from nowhere
- no point reductions for inactivity and missing opponent quality
- a very natural way of getting the superiority of a boxer by the the raw ratings. Boxer a with rating A, boxer b with rating B. Winning probability of boxer a is A/(A+B).

The new system is kind of very simple and efficient.

[Manrae]

I’m sorry but I still don’t quite understand what the rating is supposed to represent “at fight”. Like, depending on what a fighter does before & after a fight, can change their “at fight” rating? Even still, does that number reflect their value before or after the fight?

For example: Hagler 76.63 vs Leonard 74.21... are these numbers reflecting their pre or post bout value?

I’m just trying to understand why there are increases and decreases of the ratings of fighters, especially concerning activity/inactivity...

at fight means at fight date after the fight

Ok, so post-fight like I presumed, thanks for clearing that up.

[JCS]

at fight means at fight date after the fight

Ok, so post-fight like I presumed, thanks for clearing that up.

That's a bit strange...

So after Jermain Taylor beat Bernard Hopkins twice.. Hopkins still had nearly triple Taylor's rating? Or is it displaying that way now, because we know what happened after?

[computerrank]

Ok, so post-fight like I presumed, thanks for clearing that up.

That's a bit strange...

So after Jermain Taylor beat Bernard Hopkins twice.. Hopkins still had nearly triple Taylor's rating? Or is it displaying that way now, because we know what happened after?

The ratings at bout time more represent the development of the career than the performance at this bout. So Taylor peaked at time of 2nd Hopkins bout, while Hopkins peaked at time of his Trinidad bout, 4 years before.

[Manrae]

Ok, I promise this is my final question

Here it states:
Bout Star Ratings

All bouts are rated with 0 to 5 stars.

Men

5 stars = both opponents have 68.9 rating points at least ~ best 100 boxers
https://boxrec.com/media/index.php/BoxR ... escription

Yet the website's bout ratings don't reflect this.
Ex) Leonard v Hearns 2, both well under 68.9
https://boxrec.com/en/proboxer/269

Is this an error or am I missing something?

[computerrank]

Ok, I promise this is my final question

Here it states:
Bout Star Ratings

All bouts are rated with 0 to 5 stars.

Men

5 stars = both opponents have 68.9 rating points at least ~ best 100 boxers
https://boxrec.com/media/index.php/BoxR ... escription

Yet the website's bout ratings don't reflect this.
Ex) Leonard v Hearns 2, both well under 68.9
https://boxrec.com/en/proboxer/269

Is this an error or am I missing something?

The stars are correct, but the explanation seems not to be complete.
The point values for the star ratings are converted in relation to the weight division in favor to the the lower weight divisions. Heavyweight is straight with factor 1, but Super Middleweight = (240 lb/168 lbs) * (240 lb/168 lbs) * (240 lb/168 lbs) = 4.17. so the 5 star limit there is 68.9 / 4.17 = 16.5

[mike1989]

please write the formulas for calculating the current rating and the all-time rating? with specific examples ???

[computerrank]

please write the formulas for calculating the current rating and the all-time rating? with specific examples ???

It is quite simple for the all-time rating. It is the career peak rating corrected for the weight division.

It is too complicated now for the current ratings, as a boxer's rating at bout time now depends on all other results and ratings at bout time of all boxers connected. With decreasing influence with growing time difference between the bouts and with decreasing influence with increasing number of bouts that happened in the meantime. There isn't a simple formula, I could give you.

If you are interested in details of the method, look here:
https://www.remi-coulom.fr/WHR/WHR.pdf

[marcianofan]

Well, the star ratings in both systems rely on the same structure. 5 stars are best 100 active boxers, 4 stars,are best 300 active boxers, 3 stars are best 900 active boxers....
I will look and give this comparision here ...
But the ranks especially in the lowers ranks are NOT the same for each boxer.

The new system doesn't need some previously needed ad hoc parameters
- no additional points from nowhere
- no point reductions for inactivity and missing opponent quality
- a very natural way of getting the superiority of a boxer by the the raw ratings. Boxer a with rating A, boxer b with rating B. Winning probability of boxer a is A/(A+B).

The new system is kind of very simple and efficient.

Call me crazy, but I think additional points from nowhere is a good thing.

For instance:
Fighter A and Fighter B are both debuting. Under the old system, they would (correctly, in my view) both have 0.000. Fighter A wins. Fighter B loses. Fighter A gets some minimal amount of points for his victory, while Fighter B gets none. Fighter B remains at 0.000 because he's done nothing positive yet in his career. Fighter A, at 1-0, is ideally at the lower reaches of the rankings among fighters who have done something positive (let's say 0.001), but at least he's appropriately separated himself from Fighter B, who has as low a rating as possible.

Points from nowhere simply show separation between fighters based on their results. They aren't a problem to be avoided, I don't think. Any problem that may have existed based on the points from nowhere was appropriately remedied by your fairly ingenious connectability bonus system, which basically kept the fighters who get all their points from "nowhere" on the outside looking in when it comes to serious rankings.

[Manrae]

Ok, I promise this is my final question

Here it states:
Bout Star Ratings

All bouts are rated with 0 to 5 stars.

Men

5 stars = both opponents have 68.9 rating points at least ~ best 100 boxers
https://boxrec.com/media/index.php/BoxR ... escription

Yet the website's bout ratings don't reflect this.
Ex) Leonard v Hearns 2, both well under 68.9
https://boxrec.com/en/proboxer/269

Is this an error or am I missing something?

The stars are correct, but the explanation seems not to be complete.
The point values for the star ratings are converted in relation to the weight division in favor to the the lower weight divisions. Heavyweight is straight with factor 1, but Super Middleweight = (240 lb/168 lbs) * (240 lb/168 lbs) * (240 lb/168 lbs) = 4.17. so the 5 star limit there is 68.9 / 4.17 = 16.5

I see, so the wiki needs a bit of updating, thanks a lot!

[SportsRatings]

- the ratings are at fight time - whole history ratings balance the rating at fight time regarding the result and all later and prior results with decreasing weight

This is what concerns me about WHR...by definition they're at their least accurate "right now".

What I mean is, they look great predictively when they know what's going to happen, and give a good ebb and flow of a fighter's career. But for *right now*, they lack the information they need, which is, the future results.

Am I getting this wrong? The WHR can tell you an accurate rating for a fighter's present state, but it can't tell you that right at the moment, it needs a few years to see what the present state's ratings should have been. The more hindsight correction there is, the worst the ratings really were "in real time."

So for actually predicting future, unknown results, do we know how good WHR really is?

[computerrank]

I get it Computerrank and I respect the reasoning. I think I saw you adjust them so that the winner of the bout always jumps ahead of the loser for a year or something? If that’s correct it takes away something from the new system and no tweaking was necessary..the system should remain true to its design.

A lot of posters predictably prefer the old ratings where an upset results in a huge jump for the winner based on one fight but it might have been better to delay the change until both ratings were available...keeps everyone happy.

So you have one supporter at least

I didn' adjust the WHR ratings themselves. All parameters and resulting values are undisturbed. There is a correction over-rule for DISPLAY purpose only and for a t most 1 year after the bout to set the winner above the loser. And that is only for pro boxing, not for amateur boxing, whose ratings are calculated too now, and will be published soon.

[computerrank]

- the ratings are at fight time - whole history ratings balance the rating at fight time regarding the result and all later and prior results with decreasing weight

This is what concerns me about WHR...by definition they're at their least accurate "right now".

What I mean is, they look great predictively when they know what's going to happen, and give a good ebb and flow of a fighter's career. But for *right now*, they lack the information they need, which is, the future results.

Am I getting this wrong? The WHR can tell you an accurate rating for a fighter's present state, but it can't tell you that right at the moment, it needs a few years to see what the present state's ratings should have been. The more hindsight correction there is, the worst the ratings really were "in real time."

So for actually predicting future, unknown results, do we know how good WHR really is?

The WHR ratings winner prediction power for all bouts between connected boxers from 2009 to 2018, with only regarding the bouts BEFORE the respective bout date, is 83.08 percent - and with including 16 additional parameters it is 84.02 percent.

[Rocky-]

Ryota Murata's debut against Akio Shibata is a 4 star fight, is this because of his Olympic achievements?

And how do you explain a debutant who has defeated another debutant receiving a 3 star rating? Is this also because of some under the hood system based on amateur accomplishments?

[computerrank]

Ryota Murata's debut against Akio Shibata is a 4 star fight, is this because of his Olympic achievements?

And how do you explain a debutant who has defeated another debutant receiving a 3 star rating? Is this also because of some under the hood system based on amateur accomplishments?

No, it is because Shibita was at that level at that time. And because Murata confirmed this level with his following career.

[Rocky-]

Thanks.

Don't get me wrong, I agree that it was a 4 star fight so it's a welcome change.

There are fights that ruin fighters though and their careers being finished shouldn't negatively affect the ratings of their last pre-shot matches. For example Kassim Ouma's brutal slugfest against GGG, a great showing for which he paid a heavy price. Going by your logic he wasn't deserving of 3-4 stars yet the man was operating at a level above any version of Akio Shibata.

All in all great job!

[marcianofan]

This is what concerns me about WHR...by definition they're at their least accurate "right now".

What I mean is, they look great predictively when they know what's going to happen, and give a good ebb and flow of a fighter's career. But for *right now*, they lack the information they need, which is, the future results.

Am I getting this wrong? The WHR can tell you an accurate rating for a fighter's present state, but it can't tell you that right at the moment, it needs a few years to see what the present state's ratings should have been. The more hindsight correction there is, the worst the ratings really were "in real time."

So for actually predicting future, unknown results, do we know how good WHR really is?

The WHR ratings winner prediction power for all bouts between connected boxers from 2009 to 2018, with only regarding the bouts BEFORE the respective bout date, is 83.08 percent - and with including 16 additional parameters it is 84.02 percent.

I think what he's getting at is that that prediction rate is entirely for bouts that have already happened. In other words, you've tailored a system that makes sense of past results (since there have been few if any bouts since the system went live). But the percentage at which it will predict future bouts seems uncertain because it has yet to be tested live in that context.

Out of curiosity, what was the prediction rate for the pre-bout ratings of the last version of the old system, and how are draws factored in, if at all?

[Manrae]

These new ratings seem to have hindsight factored in... I actually favor these new ratings as well.

[margaret thatcher]

So you get a guy like Tyson Fury having more points in his career after beating Seferi following 2.5 years away than he did right after beating Klit? How does this work