BoxRec

computerrank wrote:... the new ratings will be the Boxrec Current Ratings

... All Time Ratings and P4P Ratings derived from them

... no other ratings/rankings needed

... they will be object to further improvement

Wow! I can go back to having a life now. I'll miss you guys, but I doubt you'll miss me.

Martin, I still haven't found the problem in the example I posted and there were some other questions as well. There might be some room still where I can help, despite the ratings being already brilliant.

conan

best light heavy Tarver??? 3rd Jones????????

Adamek and Erdei far behind them on points???????

Boxrec is an excellent historic database, but the illogical and unprofessional changes in the ratings make it less and less serious...

Alain wrote:best light heavy Tarver??? 3rd Jones????????

Adamek and Erdei far behind them on points???????

Boxrec is an excellent historic database, but the illogical and unprofessional changes in the ratings make it less and less serious...

Well, if he was still active, then the best light heavy would be Hopkins. Until this loss, Tarver was clearly the number 1 at light heavy. The question remains how far does he fall by losing to a p4p top 5 in Hopkins. Not very much.

The ratings reflect the career performance of the fighters. Adamek looks great, but till now he has only beaten Paul Briggs in a life and death battle and the wooden Thomas Ulrich. Erdei has beaten the inconsistant Julio Gonzales and the wooden Thomas Ulrich. For either boxer, that's nothing compared to Tarver's victories over then p4p #1 Jones, unified light heavy champ Johnson, Harding, Montel Griffen etc.

Anyhow, this is all very specific. What about your opinion in general.

conan

conan_the_cribber wrote:

Alain wrote:best light heavy Tarver??? 3rd Jones????????

Adamek and Erdei far behind them on points???????

Boxrec is an excellent historic database, but the illogical and unprofessional changes in the ratings make it less and less serious...

Well, if he was still active, then the best light heavy would be Hopkins. Until this loss, Tarver was clearly the number 1 at light heavy. The question remains how far does he fall by losing to a p4p top 5 in Hopkins. Not very much.

The ratings reflect the career performance of the fighters. Adamek looks great, but till now he has only beaten Paul Briggs in a life and death battle and the wooden Thomas Ulrich. Erdei has beaten the inconsistant Julio Gonzales and the wooden Thomas Ulrich. For either boxer, that's nothing compared to Tarver's victories over then p4p #1 Jones, unified light heavy champ Johnson, Harding, Montel Griffen etc.

Anyhow, this is all very specific. What about your opinion in general.

conan

How can you call Tarver as the best current champ? he does not even have a belt, and has three losses during the last year years... I know that the current boxing world is not only about belts but more and more about names fighting media money but does it mean that Tarver should have 500 more points on the ratings? by the way, a the last months saw Chad Dawson leading the LH list... how can you argue for him????? this is ridicolous...

especially if it is a computerized rating system and not a subjective "media" list. Jones also lost badly 3 out of his last four games... and he overcomes fighters with winning streaks?????? come on, this is highly illogical from a computerized list....

Alain wrote:best light heavy Tarver??? 3rd Jones????????

Adamek and Erdei far behind them on points???????

Boxrec is an excellent historic database, but the illogical and unprofessional changes in the ratings make it less and less serious...

Alain,

Adamek and Erdei never reached the level of Jones Jr. Tarver and Johnson did by their successful bouts.

So Adamek and Erdei have to defeat one of them - this is boxing.

Nothing more behind ...
Martin

... here is a log of all bouts of the top 5 in the light heavyweight ratings since 2000-01-01.

I think, you will be able to follow the log. You see the ratings before and after the bout for both boxers - the values are slightly different (based latest not published version) - but the ranking is still valid.

It is very consequent and logical from any boxing perspective.

JCS83MD wrote:RECOMMENDATION

Ok well here's one

Most people would say Caballero should be ranked ahead of Ponce De Leon

So we'll use Ponce de Leon in this example


Limit Ponce De Leon to the post-fight rating of Caballero until De Leon beats a boxer who is that rating or higher.. or perhaps we can be liberal and say a fighter within 5% of that post-fight rating.

Thoughts on this Martin?

JCS83MD wrote:

JCS83MD wrote:RECOMMENDATION

Ok well here's one

Most people would say Caballero should be ranked ahead of Ponce De Leon

So we'll use Ponce de Leon in this example


Limit Ponce De Leon to the post-fight rating of Caballero until De Leon beats a boxer who is that rating or higher.. or perhaps we can be liberal and say a fighter within 5% of that post-fight rating.

Thoughts on this Martin?

... not applicable, as the ranking is fixed after the bouts only until the next bout took place.

Then the rating is open to improvement, when defeating appropriate opponents.

The requested strict ranking is much more unjust to the boxers - also remember the weak prediciton rate of 62%.

Why cut the improvements against valid opponents, when there is a rule in place - the opponent must be inside 25% of the boxer's rating in order to get an improvement.

Well if we were still concerned with prediction rate we'd still have the old system on.. or so I thought.

I also feel that 18 months is a long time to wait before penalizing inactivity.
Instead of a 33% decrease in 18 months, I'm curious to see what a 25% decrease in 15 months would look like.

conan_the_cribber wrote:

computerrank wrote:

conan_the_cribber wrote:
- the winner gets the 10 points anyway
- the winner gets a percentage of the difference of his losers own rating minus 25% of his own rating as long as the loser was inside 25% of his own rating
- percentage is 20% for close win (SD, MD, close win on points)
- percentage is 40% else
-- A (1000) KO B (500):
--- new_A=1000+0.4*(500-1000/4)=1100
--- new_B = 400
- the loser losing the same amount of points


Still getting my head around this.

(500-1000*25%) = 250
40% of 250 = 100
new ranking = 1000 + 100 = 1100.

But if it was two fighters of a 1000, then it would be

(0-1000*25%) = -250 <<<****fixed****
40% of -250 = -100
new ranking = 1000 - 100 = 900.

So the winner moves backwards? I think some other formula is used.

I think my examples with Margarito is invalid. 25% of 1000 is 250, so a fighter with 1000 can fight anyone above 250. Am I right?

conan

... try again ... find your fault ...

I found my fault, but not yours.

Well you didn't make it easy for me did you Martin? The 500 is the losers points and the difference between the fighters. So looking now at the example and changing it slightly, so that it's 1200 vs 1000

(1000-1200*25%) = 700
40% of 700 = 280
new ranking_A = 1200 + 280 = 1480.
new_ranking_B = 1000 - 280 = 720.

Now that it seems to make sense, it means, that if Adamek beats Briggs he'll be on 739, ahead of Jones. If Erdei beat Jones, he wood move ahead of Johnson. But if Johnson beats Woods, then he will still be narowly behind Tarver. A shame that, but what can you do, Tarver legitimately got so many points from his wins.

conan

@conan

... you found the solution, congrats ...

Gee thanks. Here's an idea. Now that the formula for working out the before and after rankings is so easy, why not put a little javascript "Ratings Calculator" on the ratings page, so that anybody can go and find out what the new rankings would be, if they wanted to.

conan

conan_the_cribber wrote:Gee thanks. Here's an idea. Now that the formula for working out the before and after rankings is so easy, why not put a little javascript "Ratings Calculator" on the ratings page, so that anybody can go and find out what the new rankings would be, if they wanted to.

conan

... not bad ...

The more I look around at the rankings, the more I like them. I like that the launching system is gone, but we also don't have the problem that used to be in place where fighters started with a certain number of points. In that old ranking system, you'd have a guy with a 0-1 record ahead of a guy with a 1-5 record only because he hadn't yet had enough time to lose his points. Having all fighters who haven't proven anything at zero points is better (although, even if they are considered to be not ranked amongst themselves they should be considered within the total ranking - heavyweight rankings show 'rank/876' when there are 1098 heavyweights in the database).

The launching system seemed designed to hold back David Rodriguez, who is apparently using his career to create a statistical outlier. I think under the predictive system, he would have jumped very, very high as soon as he beat a launched fighter (like Wallace McDaniel ). Now, he is ranked appropriately at 139, well below his IBO ranking of 82 (which does not even include his last win). Brian Minto is 61 on IBO, but 26 on Boxrec, so there are definitely some big differences. David Diaz is 10, Palyani is 28. Thank goodness - that is much, much better. Palyani is still a lot higher than on the IBO rating (57), which I think is appropriate.

On the issues which are being brought up:

1) Should fighter A always be ahead of fighter B if he beats him? After a lot of thought, I think the answer is yes. The problem arises in the next iteration - when fighter A loses to fighter C, C takes all of those points A won against B. Most of the time this is okay, but when a 3-63 Euro opponent upsets some protected prospect, then the fighter who gets the first chance at him will be a major beneficiary. It would be nice if there was a way to adjust points based on the following results. If the 3-63 fighter goes out and loses his next 6 fights, then the points earned against him in those fights would be diminished in some way. Alternatively, you could perhaps create an adjustment for a result against a fighter who then goes on to win several fights in a row. Perhaps Omar Nino Romero would have been more highly ranked if his early career win over Jorge Arce had been slightly, continuously bumped up in value as Arce began moving up the rankings. Same, say, with Rodney Jones and his early career win over Antonio Margarito.

2) The biggest perceived problem with these rankings is the grip on the top of fighters like Ruiz and Jones. The old system's big benefits to young, active prospects did correctly forecast rising youth in some cases. For instance, Sharmba Mitchell was ahead of Paul Williams in every ranking set-up, except the old predictive method, even though Williams was the favorite in the fight. I don't think that means that the predictive ranking was good, but it does perhaps show a missing element in the new rankings.

It would be very interesting to see a regression analysis done to determine the peak age of boxers. This was successfully done with baseball (peak age was found to be about 27) and I think with so many boxing records in the database you could probably find a significant statistic of when fighters start to decline. Of course, it depends on weight class and the number of rounds fought as well - but I would love to see a way to measure 'decay'. This might be a downward adjustment for a loss by a veteran figher, since losses past a certain age or experience level are usually the signifier that the fighter is on a downward trend. If a significant statistic could be gleaned, it might help to correctly downgrade fighters who are shot or approaching the end of their careers, and fix perceived problems like Sharmba Mitchell.

Anyway, great rankings Martin. I am glad we have just one set, and it is this one.

emile,

great post, thanks.

I will walk thorougly through your messages. But it will take some time, I will be off in my holidays soon - and back midth of September.

Thanks again and have a good time
Martin

Maybe there's room for an automatic deduction for two consecutive bad losses, or three consecutive losses.

As I think a little more - if you could just statistically determine peak age and start minor deductions over every 'x' time period, I think it might work without a complicated algorthym. As long as the deduction was small enough, older fighters like Bernard Hopkins who kept winning would maintain their place at the top - but as soon as they started losing and dropping back down to the level of their peers, they might see a ranking penalty. Take Calzaghe. Currently, he has a massive 600 point advantage over his nearest competitor. Even with a slow age deduction for 34 years, he would easily remain number one - but if he was to lose to Kessler, say, and be brought back some, that age deficit might start to make it easier for a younger fighter like Andrade or Bute to catch him in the rankings. This might be especially helpful in the unusual case of the light heavy division, where you have three 37 year-olds hogging the top of the rankings by beating each other. If all of them are seeing age deductions, then losing, even to each other, will bring them down to the younger fighters more quickly - and in fact there would be a multiplier effect since all of them would have diminished rankings, so their wins over each other would be less valuable. Any of them, however, could restore their place at the top with a win over a top younger fighter, as it should be.

Even if the peak age was normalized across all weights, it might still work. Since most heavyweights are over 30, they would all get penalized to some degree and not effect the rankings much, except to give a bonus to fighters who had the same accomlishments at a much younger age. Most minimumweights might not be affected, but their careers tend to be short anyway, so the basic rule of just having to beat the guy above you would stand.

So, I think a great step forward would be to run regression to find peak age. I wish I knew how to do that :) Have a great vacation Martin.

emile wrote:As I think a little more - if you could just statistically determine peak age and start minor deductions over every 'x' time period, I think it might work without a complicated algorthym. As long as the deduction was small enough, older fighters like Bernard Hopkins who kept winning would maintain their place at the top - but as soon as they started losing and dropping back down to the level of their peers, they might see a ranking penalty. Take Calzaghe. Currently, he has a massive 600 point advantage over his nearest competitor. Even with a slow age deduction for 34 years, he would easily remain number one - but if he was to lose to Kessler, say, and be brought back some, that age deficit might start to make it easier for a younger fighter like Andrade or Bute to catch him in the rankings. This might be especially helpful in the unusual case of the light heavy division, where you have three 37 year-olds hogging the top of the rankings by beating each other. If all of them are seeing age deductions, then losing, even to each other, will bring them down to the younger fighters more quickly - and in fact there would be a multiplier effect since all of them would have diminished rankings, so their wins over each other would be less valuable. Any of them, however, could restore their place at the top with a win over a top younger fighter, as it should be.

Even if the peak age was normalized across all weights, it might still work. Since most heavyweights are over 30, they would all get penalized to some degree and not effect the rankings much, except to give a bonus to fighters who had the same accomlishments at a much younger age. Most minimumweights might not be affected, but their careers tend to be short anyway, so the basic rule of just having to beat the guy above you would stand.

So, I think a great step forward would be to run regression to find peak age. I wish I knew how to do that :) Have a great vacation Martin.

I did a little bit of testing there. It seems to vary by division and since all fighters are different.. you could very well base such a decay on total fights or career age as you could on real age.

Most sports players follow a similar path because they've had similar backgrounds and what not. Even there, some positions have players who fade faster than others (catcher for one). Baseball peak age of 29 (not 27) may apply in most scenarios.. but a lot of times it does not.

Well, we are talking about statistics here. We would expect to see some fighters who don't fit the mold, but when I look at the divisions, I see some pretty strong indicators. At almost every weight, there are older fighters in decline who are hanging on near the top of the of the ratings.

Everything I can find says peak age is 27 for baseball, and Baseball Prospectus even had an article saying there is no evidence its any different for catchers.

http://www.baseballprospectus.com/artic ... icleid=569
http://www.ncbi.nlm.nih.gov/entrez/quer ... t=Abstract
http://www.sabernomics.com/sabernomics/ ... -continued

I'd bet the peak age is lower than people think for boxers as well, since the one athletic gift that older boxers gain - strength - is countered by the need to move up in weight class. Experience is obviously very important, but I think the large number of old fighters at the top of the rankings is probably more due to matchmaking than anything else.

Obviously, some fighters will not fit an age line. Jose Luis Zertuche is still young in experience, and perhaps still on the rise at 33, and Fernando Vargas looks to be shot at 28. But a decay rating would have an interesting effect on Nando, even if it didn't affect his own rating. Vargas hasn't fought a fighter under 30 since he beat 29-year-old Jose Alfredo Flores in 2001 and lost to 29-year-old DLH in 2002. Since then, the ages of his opponents has been 35,33,36,34,34. An age effect would hit Vargas for his last four years of work, perhaps indicating his own dimished fighting level at a young age.

emile wrote:Well, we are talking about statistics here. We would expect to see some fighters who don't fit the mold, but when I look at the divisions, I see some pretty strong indicators. At almost every weight, there are older fighters in decline who are hanging on near the top of the of the ratings.

Everything I can find says peak age is 27 for baseball, and Baseball Prospectus even had an article saying there is no evidence its any different for catchers.

http://www.baseballprospectus.com/artic ... icleid=569
http://www.ncbi.nlm.nih.gov/entrez/quer ... t=Abstract
http://www.sabernomics.com/sabernomics/ ... -continued

I'd bet the peak age is lower than people think for boxers as well, since the one athletic gift that older boxers gain - strength - is countered by the need to move up in weight class. Experience is obviously very important, but I think the large number of old fighters at the top of the rankings is probably more due to matchmaking than anything else.

Obviously, some fighters will not fit an age line. Jose Luis Zertuche is still young in experience, and perhaps still on the rise at 33, and Fernando Vargas looks to be shot at 28. But a decay rating would have an interesting effect on Nando, even if it didn't affect his own rating. Vargas hasn't fought a fighter under 30 since he beat 29-year-old Jose Alfredo Flores in 2001 and lost to 29-year-old DLH in 2002. Since then, the ages of his opponents has been 35,33,36,34,34. An age effect would hit Vargas for his last four years of work, perhaps indicating his own dimished fighting level at a young age.

The first link says late 20s, early 30s. The second link uses data from 1965 (people live longer now and are in much better condition in later years)


Please see in the third link in the conclusion:

"But, that number 29 keeps popping up as the peak age for players, and this runs against the conventional wisdom."

This supports 29 as the peak age of baseball players. After looking at a plethora of stats (something i've done quite a bit of) 28-30 seem to be a player's best age in this day in age.


Most boxers seemed to hit their peak in their late 20s.. but heavyweights seemed to be early-mid 30s

The first article does not say what you are saying (it says "Catchers do not improve or maintain their rates of production into their late 20's or early 30's"). The third article is by a guy whose own study suggests 29, but he does say that that is against the conventional wisdom of Bill James, Tangotiger, and others.

In any case, it is not settled. So perhaps determining a peak age for boxers would be impossible, given the additional factors such as weight classes and (as you pointed out) uneven career trajectories. I think there is some potential here though.

... I just got in ... I remember to have heared of an old guideline ... boxers come to their peak at about 5 years of career...

computerrank wrote:

Alain wrote:best light heavy Tarver??? 3rd Jones????????

Adamek and Erdei far behind them on points???????

Boxrec is an excellent historic database, but the illogical and unprofessional changes in the ratings make it less and less serious...

Alain,

Adamek and Erdei never reached the level of Jones Jr. Tarver and Johnson did by their successful bouts.

So Adamek and Erdei have to defeat one of them - this is boxing.

Nothing more behind ...
Martin

Hi Martin,

Agree, the US fighters had good winning streaks till 2003. Since then especially Jones and Tarver had suffered several losses and basically have fought each other. Although I also agree that Adamek and Erdei would have to beat them to prove to the World, but if the US fighters are rather fighting for media money and DO NOT GIVE A CHANCE to those who would deserve it then what can be done? This is boxing...

I still do not have a clue how Dawson could lead the list for several months...

Cobwebcat wrote:Yep that's another problem of the A must beat B sceanario. It's statistically silly but does "look right" and fits in with boxing ethos.

Poor Computerank must be distraught in his search for a system that makes statistical sense and is also universally accepted. Unfortunately the two rarely go hand in hand.

Whatever he comes up with people will always find problems. There isnt a perfect system and all of them have flaws. This one will be no different other than I expect most people will think it "Looks a bit better"

Now we've got Roy Jones 3rd, and Stipe Drews STILL ahead of Paul Briggs. Caballero STILL behind Ponce De Leon (although you could argue either way on that one)... Barrera ahead of Pacquiao. Most people would disagree with these situations and I figured the traditional aspects of this update would alleviate those problems.

JCS83MD wrote:

Cobwebcat wrote:Yep that's another problem of the A must beat B sceanario. It's statistically silly but does "look right" and fits in with boxing ethos.

Poor Computerank must be distraught in his search for a system that makes statistical sense and is also universally accepted. Unfortunately the two rarely go hand in hand.

Whatever he comes up with people will always find problems. There isnt a perfect system and all of them have flaws. This one will be no different other than I expect most people will think it "Looks a bit better"

Now we've got Roy Jones 3rd, and Stipe Drews STILL ahead of Paul Briggs. Caballero STILL behind Ponce De Leon (although you could argue either way on that one)... Barrera ahead of Pacquiao. Most people would disagree with these situations and I figured the traditional aspects of this update would alleviate those problems.

... yes it is a hybrid ...

but, I see it absolutely postitive:

- the ranking aspect is definitely stronger than before (you need to defeat close opponents to get up)
- but the ranking aspect is no absulute barrier - as would be with a pure traditional ranking - you cann pass by defeating this close opponents

I think, purism in not the solution - the very majority accepts the new solution ... and they know, why