Ratings - please read before commenting - Archived

[Manrae]

It is the BoxRec winner above loser rule:
Jorge Armando Sanchez Hernandez defeated Mario Gutierrez
Mario Gutierrez defeated Daniel Valladares

pure WHR ratings:
Valladares 6.312 at Minimumweight
Gutierrez 0.460 at Flyweight
Hernandez 0.050 at Light Flyweight

BoxRec winner above loser rule makes
Valladares 4.796 at Minimumweight ( = 4.407 at Light Flyweight)
Gutierrez 4.037 at Flyweight ( = 4.502 at Light Flyweight )
Hernandez 5.801 at Light Flyweight

I understand that... but when a 4 round fighter, with only 4 fights (3-0-1) is ranked #15 in the world in a division over former world champions and such...

How does this not ring any alarm bells for you?

Please consider adjusting or getting rid of the rule

There is an option. The winner above loser rule only says, that the winners in such a queue have to be above the losers - somewhere inside their current maximum and minimum WHR ratings.

Currently in case of an upset the winner is set to 100% of the loser's WHR rating and the loser is set to 10% of the difference lower.So a big upset winner can get that high as here.

An option would be to set the upset winner to only to 70% of the difference and the upset loser to 10% of the difference lower.

Then Hernandez would only be at #61 at Light Flyweight and not #15 as now.

And Valladeres at the end of the queue would be dumped from #11 Minimumweight now to #39 Minimumweight then.

Thank you. I'm curious to see this change as soon as possible.

[margaret thatcher]

maybe tank can be used as a perfect test case to work out some kinks

if boxrec would have him at #1 in all but 1 division from 135-154, obviously there's a problem with how he's getting points

[computerrank]

maybe tank can be used as a perfect test case to work out some kinks

if boxrec would have him at #1 in all but 1 division from 135-154, obviously there's a problem with how he's getting points

With more weight on close rated bouts, as discussed above, he would be only close above Josh Taylor with 106 / 80 at Super Lightwweight. And at Welterweight it would be Spence / Crawford / Davis and Crawford with 205 /160 / 82. And Super Welterweight is a very weak division currently and the move is hypothetical anyway.

[computerrank]

maybe tank can be used as a perfect test case to work out some kinks

if boxrec would have him at #1 in all but 1 division from 135-154, obviously there's a problem with how he's getting points

With more weight on close rated bouts, as discussed above, he would be only close above Josh Taylor with 106 / 80 at Super Lightwweight. And at Welterweight it would be Spence / Crawford / Davis and Crawford with 205 /160 / 82. And Super Welterweight is a very weak division currently and the move is hypothetical anyway.

I think, a rating cap with at most 2x the best boxed opponent rating within the last 3 years, should be a solution.

See Welterweight and Super Lightweight on test:

[margaret thatcher]

certainly does a good job in those divisions

[JCS]

Over on the MMA side, I look at the past 3 years in division and take the max of:

3x rating of the best win or 1.33x rating of the best draw or .33x the rating of the best close loss (split/majority). Max penalty of 1/3rd.

Lots of threes :) :)

[jujigatame]

Isn't this all a bit of a hack to work around the main issue though? Like how is a guy building his way to 400 points at 140 when his biggest win is against a guy who has 88 points at 130?

[computerrank]

Isn't this all a bit of a hack to work around the main issue though? Like how is a guy building his way to 400 points at 140 when his biggest win is against a guy who has 88 points at 130?

There are 2 options to deal with that:

- the cap as additional BoxRec rule for the current ratings -while the basic WHR ratings for prediction are unchanged.
- no additional BoxRec rule , but more weight for bouts with close rated boxers and treat all UD as clear wins and only all MD and SD as close wins within the basic WHR ratings. That would drop winner prediction ratio from 84.36 to 84.08 percent. But would improve winner prediction ratio for bouts with close rated boxers from 75.56 to 77.22 percent, which is a lot.

Now I give you the resulting divisional top ratings for both options:

First the cap option

[computerrank]

... and now the more weight for bouts with close rated boxers option:

[Manrae]

Current rankings are better than both...

And I see the actual, objective discrepancies have been ironed out, thank you

[SportsRatings]

How do these options affect the all-time rankings? HW in particular, as those seem to change a lot with each adjustment

[computerrank]

How do these options affect the all-time rankings? HW in particular, as those seem to change a lot with each adjustment

With the ratings cap solution the all-time-ratings wontn't change at all, because the basic WHR ratings are not changed.

[SportsRatings]

How do these options affect the all-time rankings? HW in particular, as those seem to change a lot with each adjustment

With the ratings cap solution the all-time-ratings wontn't change at all, because the basic WHR ratings are not changed.

But the close-fights option would intuitively boost Ali who fought a lot of high-level opponents in the 70s, while Klitschko was probably far ahead of his competition at most times and might be faded as a result.

Though it's hard to project how the intuitive plays out in actual run-time!

It does seem that the close-bouts solution boosted Joshua because his (wide-favorite) loss to Ruiz is downplayed, while Fury's (big underdog) upset of Klitschko would be scaled back too.

[computerrank]

How do these options affect the all-time rankings? HW in particular, as those seem to change a lot with each adjustment

With the ratings cap solution the all-time-ratings wontn't change at all, because the basic WHR ratings are not changed.

But the close-fights option would intuitively boost Ali who fought a lot of high-level opponents in the 70s, while Klitschko was probably far ahead of his competition at most times and might be faded as a result.

Though it's hard to project how the intuitive plays out in actual run-time!

It does seem that the close-bouts solution boosted Joshua because his (wide-favorite) loss to Ruiz is downplayed, while Fury's (big underdog) upset of Klitschko would be scaled back too.

With the close-fights solution the influence to all-time is not breaking. See all-time Heavyweight top 50:

[jujigatame]

The 2nd sounds better to me but the results of the 1st look the better of the two. Jermell Charlo dropping off is a commonality in both that I don't fully understand.

Testing out potential fixes is fine but do we fundamentally understand the cause of the issue? Like how does a fighter end up with such a huge point total that dwarfs his best competition?

[computerrank]

The 2nd sounds better to me but the results of the 1st look the better of the two. Jermell Charlo dropping off is a commonality in both that I don't fully understand.

Testing out potential fixes is fine but do we fundamentally understand the cause of the issue? Like how does a fighter end up with such a huge point total that dwarfs his best competition?

Jermell Carlo:
- cap version: pretty easy to understand, his best defeated opponent within last 3 years was only 10% of his rating
- close-rating version: same point, ratios 10% and below, only exception Harrison and that was a loss

The fundamental cause of improving the rating with inferior opponents clearly defeated is that the probability for a 100% win always is lower than 100%. So the winner still will get a bit.

[computerrank]

So far I would prefer the cap version.

With parameters:
- the BoxRec current rating has a cap
- the cap is the maximum of the caps calculated from each bout within the last 3 years
-- the bout cap is (1 + 2*cd)*opponent rating for won and drawn bouts; cd = clear decision factor; 0 = draw, 1 = clear win
-- the bout cap is (1-cd)*opponent rating for not clearly lost bouts
- if the cap is lower than 25% of the WHR rating, the cap is 25% of the WHR rating

This limits rising BoxRec current ratings by defeating inferior opponents.
This lets the basic WHR ratings with it's prediction power undisturbed.

[Manrae]

The 2nd sounds better to me but the results of the 1st look the better of the two. Jermell Charlo dropping off is a commonality in both that I don't fully understand.

Testing out potential fixes is fine but do we fundamentally understand the cause of the issue? Like how does a fighter end up with such a huge point total that dwarfs his best competition?

Jermell Carlo:
- cap version: pretty easy to understand, his best defeated opponent within last 3 years was only 10% of his rating
- close-rating version: same point, ratios 10% and below, only exception Harrison and that was a loss

The fundamental cause of improving the rating with inferior opponents clearly defeated is that the probability for a 100% win always is lower than 100%. So the winner still will get a bit.

Can you explain how Tim Tszyu would rank over Jermell Charlo at 154? That’s a LOT more egregious than Tank’s current rating…

[jujigatame]

The 2nd sounds better to me but the results of the 1st look the better of the two. Jermell Charlo dropping off is a commonality in both that I don't fully understand.

Testing out potential fixes is fine but do we fundamentally understand the cause of the issue? Like how does a fighter end up with such a huge point total that dwarfs his best competition?

Jermell Carlo:
- cap version: pretty easy to understand, his best defeated opponent within last 3 years was only 10% of his rating
- close-rating version: same point, ratios 10% and below, only exception Harrison and that was a loss

The fundamental cause of improving the rating with inferior opponents clearly defeated is that the probability for a 100% win always is lower than 100%. So the winner still will get a bit.

If this is the case then maybe the solution here is to reduce the value of all wins to some degree? If KOs and dominant decisions are valued at 1.0 and fighters with poor opposition get overrated because 1.0 is always greater than the ratio in their point differences (so they are always outperforming the expected result) then maybe the solution is to make the max win value less than 1.0.

[computerrank]

The 2nd sounds better to me but the results of the 1st look the better of the two. Jermell Charlo dropping off is a commonality in both that I don't fully understand.

Testing out potential fixes is fine but do we fundamentally understand the cause of the issue? Like how does a fighter end up with such a huge point total that dwarfs his best competition?

Jermell Carlo:
- cap version: pretty easy to understand, his best defeated opponent within last 3 years was only 10% of his rating
- close-rating version: same point, ratios 10% and below, only exception Harrison and that was a loss

The fundamental cause of improving the rating with inferior opponents clearly defeated is that the probability for a 100% win always is lower than 100%. So the winner still will get a bit.

Can you explain how Tim Tszyu would rank over Jermell Charlo at 154? That’s a LOT more egregious than Tank’s current rating…

Jermell Carlo with a best win against Rosario 18.04 within 3 years. And his loss against Harrison 19.39.
Tim Tszyu with a best win against Horn 26.02.

[computerrank]

The 2nd sounds better to me but the results of the 1st look the better of the two. Jermell Charlo dropping off is a commonality in both that I don't fully understand.

Testing out potential fixes is fine but do we fundamentally understand the cause of the issue? Like how does a fighter end up with such a huge point total that dwarfs his best competition?

Jermell Carlo:
- cap version: pretty easy to understand, his best defeated opponent within last 3 years was only 10% of his rating
- close-rating version: same point, ratios 10% and below, only exception Harrison and that was a loss

The fundamental cause of improving the rating with inferior opponents clearly defeated is that the probability for a 100% win always is lower than 100%. So the winner still will get a bit.

If this is the case then maybe the solution here is to reduce the value of all wins to some degree? If KOs and dominant decisions are valued at 1.0 and fighters with poor opposition get overrated because 1.0 is always greater than the ratio in their point differences (so they are always outperforming the expected result) then maybe the solution is to make the max win value less than 1.0.

This results in a loss of prediction power. Taking a value of 0.9 for a clear win reduces the winner prediction ratio from 84% to below 80%.

[Manrae]

Jermell Carlo:
- cap version: pretty easy to understand, his best defeated opponent within last 3 years was only 10% of his rating
- close-rating version: same point, ratios 10% and below, only exception Harrison and that was a loss

The fundamental cause of improving the rating with inferior opponents clearly defeated is that the probability for a 100% win always is lower than 100%. So the winner still will get a bit.

Can you explain how Tim Tszyu would rank over Jermell Charlo at 154? That’s a LOT more egregious than Tank’s current rating…

Jermell Carlo with a best win against Rosario 18.04 within 3 years. And his loss against Harrison 19.39.
Tim Tszyu with a best win against Horn 26.02.

So... a win over a controversial 1-win wonder past their prime rates over 2 KO wins over 2 champions

...you don't see the problem here?

[computerrank]

Can you explain how Tim Tszyu would rank over Jermell Charlo at 154? That’s a LOT more egregious than Tank’s current rating…

Jermell Carlo with a best win against Rosario 18.04 within 3 years. And his loss against Harrison 19.39.
Tim Tszyu with a best win against Horn 26.02.

So... a win over a controversial 1-win wonder past their prime rates over 2 KO wins over 2 champions

...you don't see the problem here?

Boxrec goes with official result and ratings, doesn't regard controversial and champions..
even more interesting now - Brian Carlos Castano would be #1 now due to his draw against Lara:

[Manrae]

Jermell Carlo with a best win against Rosario 18.04 within 3 years. And his loss against Harrison 19.39.
Tim Tszyu with a best win against Horn 26.02.

So... a win over a controversial 1-win wonder past their prime rates over 2 KO wins over 2 champions

...you don't see the problem here?

Boxrec goes with official result and ratings, doesn't regard controversial and champions..
even more interesting now - Brian Carlos Castano would be #1 now due to his draw against Lara:

So…you are willing to see Tank’s ranking as a catalyst for change, but can’t see the problem with Tim Tszyu over Jermell Charlo

[computerrank]

So…you are willing to see Tank’s ranking as a catalyst for change, but can’t see the problem with Tim Tszyu over Jermell Charlo

The discussion about Davis' ranking was a catalyst, yes, but not the reason to change anything. The discussion lead me into finding a lot of top boxers with a massive unbalance between WHR rating and best opponent rating in recent time.

So introducing a rating cap would make a good additional rule for the BoxRec ratings, I now think.

Meanwhile I generalized the cap rules a bit:
- look into all bouts in boxers profile - not only the last 3 years
- give all bouts between 3 year ago and 6 years ago a weight of 0.5 (for the last 3 years weight 1)
- give all bouts between 6 year ago and 9 years ago a weight of 0.25 ... etc

That changes the results - and also the Tszyu / Carlo case - see below.