Ratings - please read before commenting - Archived

[champion571]

I still don't understand, how this system was created or at the very least how the rating are staged, I mean come on, you have an
active 2-time former super welterweight world champion - Travis Simms rated #56 at middleweight, I personally think this is a bias system!

But, you have guy's like james kirkland who haven't fought over a year, who never fought a top 10 rated fighter, rated in the top 10, something is clearly wrong here!

[Asterix]

I still don't understand, how this system was created or at the very least how the rating are staged, I mean come on, you have an
active 2-time former super welterweight world champion - Travis Simms rated #56 at middleweight, I personally think this is a bias system!

But, you have guy's like james kirkland who haven't fought over a year, who never fought a top 10 rated fighter, rated in the top 10, something is clearly wrong here!

James Kirkland last fought LESS than a year ago. As soon as it is over a year, he will automatically be regarded as inactive.

As for it being biased, go and have a look at the calculation. It's all completely open.

http://boxrec.com/media/index.php/BoxRe ... escription

[Asterix]

I'm just curious how Patrick Majewski W 12 (KO 7) - L 0 (KO 0) - D 0 got to be rated 46th in the World?

[computerrank]

I'm just curious how Patrick Majewski W 12 (KO 7) - L 0 (KO 0) - D 0 got to be rated 46th in the World?

Code: Select all

date      |division            |boxer               |opponent            |rs|dec|box_0>box_1|opp_0>opp_1
----------|--------------------|--------------------|--------------------|--|---|-----|-----|-----|-----
2006-09-20|Middleweight        |Majewski            |Peacock             |W |TKO|    0|    0|    0|    0
2006-12-14|Middleweight        |Majewski            |Dunham              |W |TKO|    0|    6|    1|    1
2007-02-16|Middleweight        |Majewski            |Cordova             |W |UD |    6|   11|    3|    3
2007-06-02|                    |Majewski            |Irwin               |W |TKO|   11|   11|    0|    0
2007-08-31|Middleweight        |Majewski            |Williams            |W |UD |   11|   11|    2|    2
2007-11-10|Middleweight        |Majewski            |Collins             |W |KO |   11|   68|   38|   22
2008-02-15|Middleweight        |Majewski            |Paz                 |W |TKO|   69|   76|   29|   23
2008-05-09|Super Middleweight  |Majewski            |Espinal             |W |UD |   66|   69|   44|   43
2008-11-25|Middleweight        |Majewski            |Rivera              |W |RTD|   79|  125|   78|   49
2009-04-24|Middleweight        |Majewski            |Lubash              |W |TKO|  123|  153|   89|   60
2009-06-06|Middleweight        |Majewski            |Mundy               |W |UD |  151|  168|   85|   69
2010-02-06|Super Middleweight  |Majewski            |Pietrantonio        |W |UD |  172|  175|   42|   41

[jrafusesr]

how do i get a hold of someone that can change a mistake on my record

[Asterix]

how do i get a hold of someone that can change a mistake on my record

Please post it in here: http://boxrec.com/forum/viewforum.php?f=1 :)

[computerrank]

new options for listing bouts ... you will get the ratings of the opponents before and after the bout ... you will be able to follow the career of every boxer in more detail now ...

try to check the buttons ON or GRAPHICS in the bout options ...

[conan_the_cribber]

new options for listing bouts ... you will get the ratings of the opponents before and after the bout ... you will be able to follow the career of every boxer in more detail now ...

try to check the buttons ON or GRAPHICS in the bout options ...

First of all, a great addition to the site. Now it will be possible to understand a fighter's rating and see how it went up and down due to his results. Thanks for implementing it.

However, I think the implementation has got some bugs. Have a look at Roy Jones career. When I had the graphics button on, I noticed the following
- Roy's rating bar drops before the Clinton fight. There appears to be no reason for this.
- Roy's rating bar appears to lengthen before the Omar Sheika fight, again for no reason.
- I tried this with various screen widths and the result was the same.

This is what I noticed when switching to the ratings On mode.
- I expected to see what I usually see when u print out a rating history, namely that the after fight rating is the before fight rating of the next fight (excepting of course division changes). However the numbers are all over the place.
- Roy Jones lose 700 pts between beating Glen Kelley and fighting Clinton Woods. Something is wrong there.
- After reading the numbers, I realised that the ratings numbers don't match the bars. Jones is currently 300 odds and was 2100 odd. Yet the bars are about 2/3 as long. This would only make sense if the bars where logarithmic or something.

conan

[computerrank]

@conan_the_cribber:

...

However, I think the implementation has got some bugs. Have a look at Roy Jones career. When I had the graphics button on, I noticed the following
- Roy's rating bar drops before the Clinton fight. There appears to be no reason for this.
- Roy's rating bar appears to lengthen before the Omar Sheika fight, again for no reason.
- I tried this with various screen widths and the result was the same.

-> no error
- the ratings change due to the daily adaption to division performance level = mean of #8 ... #12 ratings per division cluster set to 650
- the ratings change due to the daily adaption to missing opponent quality in the last 18 months

This is what I noticed when switching to the ratings On mode.
- I expected to see what I usually see when u print out a rating history, namely that the after fight rating is the before fight rating of the next fight (excepting of course division changes). However the numbers are all over the place.
- Roy Jones lose 700 pts between beating Glen Kelley and fighting Clinton Woods. Something is wrong there.

-> Jones best opponent in recent 18 months was Gonzales with 503 points. But had Jones 2100+ points. So he lost a lot of points for missing opponent quality.

- After reading the numbers, I realised that the ratings numbers don't match the bars. Jones is currently 300 odds and was 2100 odd. Yet the bars are about 2/3 as long. This would only make sense if the bars where logarithmic or something.

-> The scale is lograthmic as the rating is based on ratios.

- 1st mark = 1 point (or less)
- 2nd mark = 10 points
- 3rd mark = 100 points
- 4th mark = 1,000 points
- 5th mark = 10,000 points

[conan_the_cribber]

Haven't got enough time to debate this.

Obviously, the scale should be reduced to better indicate the difference in rating thru the career. 10000 is way too high. I looked at Sugar Ray and Willie Pep and Pep got to 4500 or so. I doubt there's anyone over 6000 and there never will be.

I don't know how the minus criteria for "credible" opponenet is calculated. But smashing 700 points off Jones's record is clearly an "over the top" penalty. He fought Gonzales who was top 10 in everyone's books and Kelly, a borderline top 10 at the time. I think the problem is, if you have an EXTREMELY high rating, then you are going to find it hard to keep matching up with people in your division, who are equally high. I think you need to introduce some sort of logarithmic, when comparing the quality of opponents.

To give you a comparison. Jones lost 710 pts because of the 18month thingy, yet lost only 750 pts, when he got KO'd by Tarver. You're effectively saying, rating wise, that these two events are equal. That sounds very wrong.

conan

[computerrank]

@conan

...

Obviously, the scale should be reduced to better indicate the difference in rating thru the career. 10000 is way too high. I looked at Sugar Ray and Willie Pep and Pep got to 4500 or so. I doubt there's anyone over 6000 and there never will be.

-> the highest rating ever was 7,500 points - the scale is logarithmic - so log 10,000 = 4.0 and log 7,500 = 3.87 = 97% of the scale

I don't know how the minus criteria for "credible" opponenet is calculated. But smashing 700 points off Jones's record is clearly an "over the top" penalty. He fought Gonzales who was top 10 in everyone's books and Kelly, a borderline top 10 at the time. I think the problem is, if you have an EXTREMELY high rating, then you are going to find it hard to keep matching up with people in your division, who are equally high. I think you need to introduce some sort of logarithmic, when comparing the quality of opponents.

-> the requirement for sufficient opponent quality already is logarithmic
- best opponent better than 50% of own rating -> no loss of points
- best opponent only 0.25 of own rating -> loss of 25%
- no bout at all in this time -> loss of 50%
- Gonzales as his best opponent was around #12 - this means Jones could be expected to have had a better opponent within 18 months - otherwise this excellent rating was no longer justified - this is the idea of this rule - no sitting on old merits

To give you a comparison. Jones lost 710 pts because of the 18month thingy, yet lost only 750 pts, when he got KO'd by Tarver. You're effectively saying, rating wise, that these two events are equal. That sounds very wrong.

-> I don't think so.
- Jones lost 32% due to missing opponent quality - best opponent only 23% of his own rating
- Jones lost 39% due to his loss against Tarver, then at 46% of his own rating

Long periods of missing opponent quality (or even inactivity) often are a significant indicator for lowered abilities - as proved here by the Tarver loss

Code: Select all

2001-07-28|Light Heavyweight   |Jones Jr            |Gonzalez            |W |UD | 2168| 2168|  504|  504
2002-02-02|Light Heavyweight   |Jones Jr            |Kelly               |W |KO | 2191| 2191|  428|  428
2002-09-07|Light Heavyweight   |Jones Jr            |Woods               |W |TKO| 1480| 1712|  890|  659
2003-03-01|Heavyweight         |Jones Jr            |Ruiz                |W |UD |  907| 1146|  766|  502
2003-11-08|Light Heavyweight   |Jones Jr            |Tarver              |W |MD | 2203| 2148|  936|  991
2004-05-15|Light Heavyweight   |Jones Jr            |Tarver              |L |TKO| 2082| 1264|  961| 1859

[conan_the_cribber]

@conan

...

Obviously, the scale should be reduced to better indicate the difference in rating thru the career. 10000 is way too high. I looked at Sugar Ray and Willie Pep and Pep got to 4500 or so. I doubt there's anyone over 6000 and there never will be.

-> the highest rating ever was 7,500 points - the scale is logarithmic - so log 10,000 = 4.0 and log 7,500 = 3.87 = 97% of the scale

I don't know how the minus criteria for "credible" opponenet is calculated. But smashing 700 points off Jones's record is clearly an "over the top" penalty. He fought Gonzales who was top 10 in everyone's books and Kelly, a borderline top 10 at the time. I think the problem is, if you have an EXTREMELY high rating, then you are going to find it hard to keep matching up with people in your division, who are equally high. I think you need to introduce some sort of logarithmic, when comparing the quality of opponents.

-> the requirement for sufficient opponent quality already is logarithmic
- best opponent better than 50% of own rating -> no loss of points
- best opponent only 0.25 of own rating -> loss of 25%
- no bout at all in this time -> loss of 50%
- Gonzales as his best opponent was around #12 - this means Jones could be expected to have had a better opponent within 18 months - otherwise this excellent rating was no longer justified - this is the idea of this rule - no sitting on old merits

To give you a comparison. Jones lost 710 pts because of the 18month thingy, yet lost only 750 pts, when he got KO'd by Tarver. You're effectively saying, rating wise, that these two events are equal. That sounds very wrong.

-> I don't think so.
- Jones lost 32% due to missing opponent quality - best opponent only 23% of his own rating
- Jones lost 39% due to his loss against Tarver, then at 46% of his own rating

Long periods of missing opponent quality (or even inactivity) often are a significant indicator for lowered abilities - as proved here by the Tarver loss

Code: Select all

2001-07-28|Light Heavyweight   |Jones Jr            |Gonzalez            |W |UD | 2168| 2168|  504|  504
2002-02-02|Light Heavyweight   |Jones Jr            |Kelly               |W |KO | 2191| 2191|  428|  428
2002-09-07|Light Heavyweight   |Jones Jr            |Woods               |W |TKO| 1480| 1712|  890|  659
2003-03-01|Heavyweight         |Jones Jr            |Ruiz                |W |UD |  907| 1146|  766|  502
2003-11-08|Light Heavyweight   |Jones Jr            |Tarver              |W |MD | 2203| 2148|  936|  991
2004-05-15|Light Heavyweight   |Jones Jr            |Tarver              |L |TKO| 2082| 1264|  961| 1859

As always Martin, always happy to have a friendly debate with you.

Point 1, quality of opponent.

- If you can, go find out how many fighters where ranked in Jones's division that were more than 50% of his ranking, at the time of his "punishment". I tip there was one, Michaelzwevski. That means Jones had to fight him or get starched in the ratings. Obviously, having a field of one fighter, who for lots of reasons may not fight you, makes it a pretty harsh punishment for not getting this fight made.

- Using linear percentages (25% and 50%) is not logarithmic, it is linear. Jones was an outlier, off the normal scale. You need to do something like a log of Jones's and his opponent's ratings, in order to truly find out whether he was skipping talent.

Point 2 amount of loss.

I think you are splitting hairs here. By your own calculations, Jones lost 32% by not fighting Michaelsweski, and 39% by getting KTFO by Tarver. Obviously these figures are very close to each other, and obviously, they are in no way proportional to how the average boxing fan would judge these factors.

Clearly fighting sub-standard fighters over a longer period should affect your ratings. But using the Tarver victory, is really a bit hollow, as Jones went up to heavyweight in between. I really don't think the 18 month phase where he fought Gonzales and Kelly had anything to do with showing that he lost after a dramatic weight gain and loss.

Anyhow the whole key is "what is sub standard". And I find it hard to quantify that if you are p4p champ.

NEW POINT

- I went back over Jones's record. There is actually a much longer perios where Jones doesn't fight anyone above 50% of his points. He should have lost the points around the Richard Frazer fight. The previous 50% fighter was Montell Griffin.

[conan_the_cribber]

But it is still a cool thing to have next to the bouts.

It shows what a travesty the Richard Frazier fight was 2164 vs 177

cheers

conan

[computerrank]

@conan

Point 1, quality of opponent.

- If you can, go find out how many fighters where ranked in Jones's division that were more than 50% of his ranking, at the time of his "punishment". I tip there was one, Michaelzwevski. That means Jones had to fight him or get starched in the ratings. Obviously, having a field of one fighter, who for lots of reasons may not fight you, makes it a pretty harsh punishment for not getting this fight made.

-> the scope of a super boxer like Jones on top of his career has to be expected to be more than the narrow divisional perspective
- anyway - he could have boxed Michalszewski at 1655, or Woods earlier at 1062
- or the cruisers Nelson and Gomez at 1030 (cruiserweight) = 1400 light heavyweight, or Jirov 925 (cruiserweight) = 1300 light heavyweight

- Using linear percentages (25% and 50%) is not logarithmic, it is linear. Jones was an outlier, off the normal scale. You need to do something like a log of Jones's and his opponent's ratings, in order to truly find out whether he was skipping talent.

-> sorry, using ratios (percentages) means using a logarithmic scale = same difference for same ratio

Point 2 amount of loss.

I think you are splitting hairs here. By your own calculations, Jones lost 32% by not fighting Michaelsweski, and 39% by getting KTFO by Tarver. Obviously these figures are very close to each other, and obviously, they are in no way proportional to how the average boxing fan would judge these factors.

-> I still think, especially for a top boxer, 18 month not boxing a matching opponent means losing up to 50% of the rating. If he had boxed
- a 1000 point boxer opponent at least, he had lost only 2%
- a 800 point opponent at least, he had lost only 13%
- a 600 point boxer at least, he had lost only 24% ...

Clearly fighting sub-standard fighters over a longer period should affect your ratings. But using the Tarver victory, is really a bit hollow, as Jones went up to heavyweight in between. I really don't think the 18 month phase where he fought Gonzales and Kelly had anything to do with showing that he lost after a dramatic weight gain and loss.

Anyhow the whole key is "what is sub standard". And I find it hard to quantify that if you are p4p champ.

NEW POINT

- I went back over Jones's record. There is actually a much longer perios where Jones doesn't fight anyone above 50% of his points. He should have lost the points around the Richard Frazer fight. The previous 50% fighter was Montell Griffin.

-> the rating dropped step by step caused by missing opponent quality

1999-01-09 Richard Frazier: dropped from 2232 to 2164 (opponent quality effective from 1088 to 787)
1999-06-05 Reggie Johnson: dropped from 2164 to 1979 (opponent quality effective from 787 to 643)

[conan_the_cribber]

So I guess the Klitschko brothers are screwed, because they won't fight each other, and so will never fight anyone within 50% of their rating, because they dominate the division.

[conan_the_cribber]

-> I still think, especially for a top boxer, 18 month not boxing a matching opponent means losing up to 50% of the rating. If he had boxed
- a 1000 point boxer opponent at least, he had lost only 2%
- a 800 point opponent at least, he had lost only 13%
- a 600 point boxer at least, he had lost only 24% ...


This contradicts what you said earlier about 50% and 25%. Care to explain the EXACT algorithm.


Thanks

conan

[computerrank]

-> I still think, especially for a top boxer, 18 month not boxing a matching opponent means losing up to 50% of the rating. If he had boxed
- a 1000 point boxer opponent at least, he had lost only 2%
- a 800 point opponent at least, he had lost only 13%
- a 600 point boxer at least, he had lost only 24% ...


This contradicts what you said earlier about 50% and 25%. Care to explain the EXACT algorithm.


Thanks

conan

I already posted the main points:

-> the requirement for sufficient opponent quality already is logarithmic
- best opponent better than 50% of own rating -> no loss of points
- best opponent only 0.25 of own rating -> loss of 25%
- no bout at all in this time -> loss of 50%

These were the edge marks - and the loss is continuous between.

$r_after=$r_before* (1-0.5*(1-$r_opp/$r_before}/0.5));

[conan_the_cribber]

-> I still think, especially for a top boxer, 18 month not boxing a matching opponent means losing up to 50% of the rating. If he had boxed
- a 1000 point boxer opponent at least, he had lost only 2%
- a 800 point opponent at least, he had lost only 13%
- a 600 point boxer at least, he had lost only 24% ...


This contradicts what you said earlier about 50% and 25%. Care to explain the EXACT algorithm.


Thanks

conan

I already posted the main points:

-> the requirement for sufficient opponent quality already is logarithmic
- best opponent better than 50% of own rating -> no loss of points
- best opponent only 0.25 of own rating -> loss of 25%
- no bout at all in this time -> loss of 50%

These were the edge marks - and the loss is continuous between.

$r_after=$r_before* (1-0.5*(1-$r_opp/$r_before}/0.5));

Jesus, that helps.

so lets say before is 1000pts and the opposition was only 400 (I assume that is a max opposition).

r_opp / r_before = 0.4
0.4/0.5 = 0.8 (80%)
so 1-0.8 = 0.2

So finnally we have 1000 * (1-0.5*(0.2))
= 1000 * (1-0.1)
= 1000 * 0.9
= 900.

But because Jones was more like 2000 and only fought against 500s then he lost

$r_after=$r_before* (1-0.5*(1-$r_opp/$r_before}/0.5));[/quote]

= 2000* (1 - 0.5*(1 - 500/2000/0.5) )
= 2000* (1 - 0.5*(1 - 0.25/0.5) )
= 2000* (1 - 0.5*(1 - 0.125) )
= 2000* (1- 0.5*(0.875))
= 2000 * (1-0.437)
= 2000 * (0.573)
= 1146

So you're telling me, that Roy Jones is half the fighter he used to be, because he is fighting stiffs over a period of 18 months????

That is one helf of a punishment given boxings politics.

[Asterix]

Awesome new feature. It's really interesting. Thanks Martin.

I've got a small query regarding Bernard Hopkins. After losing to Calzaghe he has 1434 points. Then he magically goes up to 1563 before fighting Kelly Pavlik. The same thing happens but to a much greater extent before the Ornelas fight, from 1940 to 2414 in between fights. What's happening?

[computerrank]

@conan

... a bit more maths and less politics ...

But because Jones was more like 2000 and only fought against 500s then he lost

$r_after=$r_before* (1-0.5*(1-$r_opp/$r_before}/0.5));[/quote]

= 2000* (1 - 0.5*(1 - 500/2000/0.5) )
= 2000* (1 - 0.5*(1 - 0.25/0.5) )
= 2000* (1 - 0.5*(1 - 0.5) )
= 2000* (1- 0.5*(0.5))
= 2000 * (1-0.25)
= 2000 * (0.75)
= 1500

For defeating a best opponent of only 25% of his rating in the relevant time period?
Still holding a level of 300% of his best defeated opponent in the relevant time period?
Why not?

[computerrank]

Awesome new feature. It's really interesting. Thanks Martin.

I've got a small query regarding Bernard Hopkins. After losing to Calzaghe he has 1434 points. Then he magically goes up to 1563 before fighting Kelly Pavlik. The same thing happens but to a much greater extent before the Ornelas fight, from 1940 to 2414 in between fights. What's happening?

- the ratings change due to the daily adaption to division performance level = mean of #8 ... #12 ratings per division cluster set to 650

- light heavyweight division cluster lost some top ranked boxers in this period and so the rating excellence of rest top ranked boxers grew

[Asterix]

Awesome new feature. It's really interesting. Thanks Martin.

I've got a small query regarding Bernard Hopkins. After losing to Calzaghe he has 1434 points. Then he magically goes up to 1563 before fighting Kelly Pavlik. The same thing happens but to a much greater extent before the Ornelas fight, from 1940 to 2414 in between fights. What's happening?

- the ratings change due to the daily adaption to division performance level = mean of #8 ... #12 ratings per division cluster set to 650

- light heavyweight division cluster lost some top ranked boxers in this period and so the rating excellence of rest top ranked boxers grew

Are you happy with Hopkins growing from 1940 to 2414? Doesn't that mean his P4P ranking is inflated?

[JCS]

Awesome new feature. It's really interesting. Thanks Martin.

I've got a small query regarding Bernard Hopkins. After losing to Calzaghe he has 1434 points. Then he magically goes up to 1563 before fighting Kelly Pavlik. The same thing happens but to a much greater extent before the Ornelas fight, from 1940 to 2414 in between fights. What's happening?

- the ratings change due to the daily adaption to division performance level = mean of #8 ... #12 ratings per division cluster set to 650

- light heavyweight division cluster lost some top ranked boxers in this period and so the rating excellence of rest top ranked boxers grew

Are you happy with Hopkins growing from 1940 to 2414? Doesn't that mean his P4P ranking is inflated?

That's the point. Its based on division dominance. He stayed the same, the division representation got weaker, so therefore his P4P rating went up.

[computerrank]

Awesome new feature. It's really interesting. Thanks Martin.

I've got a small query regarding Bernard Hopkins. After losing to Calzaghe he has 1434 points. Then he magically goes up to 1563 before fighting Kelly Pavlik. The same thing happens but to a much greater extent before the Ornelas fight, from 1940 to 2414 in between fights. What's happening?

- the ratings change due to the daily adaption to division performance level = mean of #8 ... #12 ratings per division cluster set to 650

- light heavyweight division cluster lost some top ranked boxers in this period and so the rating excellence of rest top ranked boxers grew

Are you happy with Hopkins growing from 1940 to 2414? Doesn't that mean his P4P ranking is inflated?

Inflation is not the point ... but I am not happy.

Maybe the rank didn't change much. Maybe he was #1 anyway ...

For p4p I need a factor to make current division ratings comparable - and decided to use the division excellence ... but this has side effects ...

The current p4p ratings are object to improvements for sure ...

[Asterix]

Even though Hopkins would probably be number 1 anyway, I don't think his rating should be 400 or so higher than Pacquiao.