Ratings - please read before commenting - Archived

[JCS]

Mesi at #22 does "appear" better than having him at #13, due to his inactivity where he probably achieved nothing in the gym, a point which I brought up earlier.

Also, we over-penalized fighters moving up from lower weight divisions. Fixing this obviously helped fighters who moved up to Heavy.. and fighters who beat them.

[conan_the_cribber]

Mesi at #22 does "appear" better than having him at #13, due to his inactivity where he probably achieved nothing in the gym, a point which I brought up earlier.

Also, we over-penalized fighters moving up from lower weight divisions. Fixing this obviously helped fighters who moved up to Heavy.. and fighters who beat them.

Thanks for your answer. Although I presented my argument with a lot of force and a lot of theory, we have at least we found some commonality. You think Mesi looks to high at 13, ring rust and all. I think many fighters look in the wrong place, some almost inexplicitly so. If you and Martin and Matty are all happy with Mormeck at 13 at cruiser, then good luck to you. I will mention this on other threads, just to see if other like minded people are around, but I wont pester you with this no more. Like Matty suggested, it's quicker to write my own rankings.

Again, thanks for answering the question.

conan

[JCS]

Mesi at #22 does "appear" better than having him at #13, due to his inactivity where he probably achieved nothing in the gym, a point which I brought up earlier.

Also, we over-penalized fighters moving up from lower weight divisions. Fixing this obviously helped fighters who moved up to Heavy.. and fighters who beat them.

Thanks for your answer. Although I presented my argument with a lot of force and a lot of theory, we have at least we found some commonality. You think Mesi looks to high at 13, ring rust and all. I think many fighters look in the wrong place, some almost inexplicitly so. If you and Martin and Matty are all happy with Mormeck at 13 at cruiser, then good luck to you. I will mention this on other threads, just to see if other like minded people are around, but I wont pester you with this no more. Like Matty suggested, it's quicker to write my own rankings.

Again, thanks for answering the question.

conan

Mormeck's at 11

[mattyp151]

Mesi at #22 does "appear" better than having him at #13, due to his inactivity where he probably achieved nothing in the gym, a point which I brought up earlier.

Also, we over-penalized fighters moving up from lower weight divisions. Fixing this obviously helped fighters who moved up to Heavy.. and fighters who beat them.

Thanks for your answer. Although I presented my argument with a lot of force and a lot of theory, we have at least we found some commonality. You think Mesi looks to high at 13, ring rust and all. I think many fighters look in the wrong place, some almost inexplicitly so. If you and Martin and Matty are all happy with Mormeck at 13 at cruiser, then good luck to you. I will mention this on other threads, just to see if other like minded people are around, but I wont pester you with this no more. Like Matty suggested, it's quicker to write my own rankings.

Again, thanks for answering the question.

conan

It's not a matter of everything being right, it's just a matter of expectations. I just feel people want a perfect list in their eyes, but will never get one. No way is Pac #6 fighter in the world, he's #3 in my eyes in the lowest. However, the system says he's 6, so I accept it because it's the way the system is.

I could nit pick all I wanted until I got a list I liked, but all that does is guarantee 1 person agrees. I would lose my 100% agreeance as soon as I put Floyd #1 because I know a ton of people see Manny the #1 guy in boxing. So, no matter how objective or mathematically based you get, there will always be disagreements.

[conan_the_cribber]

It's not a matter of everything being right, it's just a matter of expectations. I just feel people want a perfect list in their eyes, but will never get one. No way is Pac #6 fighter in the world, he's #3 in my eyes in the lowest. However, the system says he's 6, so I accept it because it's the way the system is.

I could nit pick all I wanted until I got a list I liked, but all that does is guarantee 1 person agrees. I would lose my 100% agreeance as soon as I put Floyd #1 because I know a ton of people see Manny the #1 guy in boxing. So, no matter how objective or mathematically based you get, there will always be disagreements.

Well, like I previously posted, I dont really care about the p4p polls. Always controversial and an absolute bitch to automate. I think the results are quite good, for such an extremely difficult task.

My beefs were solely to do with the rankings withn the weight classes. Unlike p4p matchups where you often have nothing to go e.g. Mayweather vs Valuev, for regular ratings you can use real life match ups and real life expectations. My beef is that the algorithm produces too many odd things in the regular ratings e.g. Mormeck at 11. I expect that to come from the problems I addressed above. But as you rightly suggested, it's a better use of my time to write my own algorithm first, before trying to change this one.

p4p. What a nightmare to write.

conan

[JCS]

Monty - If a boxer only fought fighters who had a rating of 300/400.. There would be a maximum that their rating could achieve.. I doubt they would get much higher than 1000 or so.

The issue of a big record padding tournament is an issue.. and one that has been partially solved.. and hopefully fully solved soon... There are isolated incidents of that in small countries around the world.

Boxers in this realm are considered unestablished. You will notice a great deal of these boxers (Their rating is less than 300). I believe Martin is making the ordeal of becoming "established" harder.

Furthermore, I think your comment about fighters not fighting those of equal ability like in other sports does hold water - and that is the reason for the wackiness but investing time into some kind of "hard limit" protection would take the focus off of total objectivity, which is pretty much what we have now.

The ratings right now are quite volatile.. and in some circumstances do look strange, but if you observe them for their purpose (predictability) they tend to work, very very well.

Before I started pushing for the predictive rating, I was working on a system similar to the old "traditional style" rating system, but it was somewhat of a hybrid between this one and the old one. The old one limited fighters too much and in turn, really hosed some older fighters. I softened the limit and such, but at the time, looks was all that mattered. My system benefitted prospects a little bit more than your conventional system and let "old" guys hang around a bit too long..
Shortly afterwards, the decision was made to goto predictability.. and the rest is history... Now you've got prospects and such taking over, but then again, the stats don't lie.

[computerrank]

@monty

thanks for your very interesting email.

I understand your basic arguement , protected boxers might rule the ratings by building their rating on their opponents padded records.

As Jason already stated, we try to avoid this problem by using a launching process for all boxers. Boxers only get a valid rating, after they passed this launching process. And fights with one or both opponents unlaunched will not disturb the statistics.

But the question is, how to define this quality gate.

Our gate is (next relaese will be published soon):

- a boxer has to defeat at least an opponent, who defeated 3 opponents, who defeated an opponent, who defeated 3 opponents, who defeated an opponent, who defeated 3 opponents, who defeated an opponent, who defeated 3 opponents, who defeated an opponent, who defeated 3 opponents, who defeated an opponent.
- or he has to defeat a launched opponent

This seems to exclude about 50% of all active boxers from being rated at the moment.

Now you asked:


... I wonder if there could be some sort of measure for "padded records" or "protected" fighters... ...

This is a key factor, I think too, we are chasing for times ...

What are your suggestions - might really help. Everyone is invited to propose ...

Looking forward to your response
Martin

[Lennox]

Yeah I meant IBO.

IBO vs Boxrec anyone?

I'm up for that as soon as JohnShep implements the latest revision which should be soon.. Don't want any changeover midway through and its already in the queue.

If its started you are 3-1 down!
Boxrec predicted 3 incorrect fights in Germany with Sidorenko, Virchis and Castilljo winning. IBO got 1 wrong, Virchis. Looks to me as if Boxrec will lose the fights nearer the top but will win the ones lower down when its the more the prospect v faded fighter.

[computerrank]

Yeah I meant IBO.

IBO vs Boxrec anyone?

I'm up for that as soon as JohnShep implements the latest revision which should be soon.. Don't want any changeover midway through and its already in the queue.

If its started you are 3-1 down!
Boxrec predicted 3 incorrect fights in Germany with Sidorenko, Virchis and Castilljo winning. IBO got 1 wrong, Virchis. Looks to me as if Boxrec will lose the fights nearer the top but will win the ones lower down when its the more the prospect v faded fighter.

Lennox,

well, this would be absolutely overwhelmingly convicing statisically , isn't it? ... based on some cases ... what about real statistics?

By the way, Sidorenko was set in front by Boxrec, you forgot for the home advantage of 125 points - IBO doesn't regard ...

Virchis and Vidoz were 6 points different, so it could go either way.

PER ASPERA AD ASTRA - not only a few words thrown into the crowd...

Martin

[JCS]

Yeah I meant IBO.

IBO vs Boxrec anyone?

I'm up for that as soon as JohnShep implements the latest revision which should be soon.. Don't want any changeover midway through and its already in the queue.

If its started you are 3-1 down!
Boxrec predicted 3 incorrect fights in Germany with Sidorenko, Virchis and Castilljo winning. IBO got 1 wrong, Virchis. Looks to me as if Boxrec will lose the fights nearer the top but will win the ones lower down when its the more the prospect v faded fighter.

You pick 4 fights in 1 night of boxing... You must be a scientists. Maybe not, but if you were, you'd know small sample sets are not the correct way to conduct an experiment.

How about tracking all fights with fighters who are both on IBO and BoxRec for a month or two... then post your results.

After scanning of the IBO ratings, I'm convinced theres some manual tweaking going on.

[conan_the_cribber]

@monty

thanks for your very interesting email.

I understand your basic arguement , protected boxers might rule the ratings by building their rating on their opponents padded records.

As Jason already stated, we try to avoid this problem by using a launching process for all boxers. Boxers only get a valid rating, after they passed this launching process. And fights with one or both opponents unlaunched will not disturb the statistics.

But the question is, how to define this quality gate.

Our gate is (next relaese will be published soon):

- a boxer has to defeat at least an opponent, who defeated 3 opponents, who defeated an opponent, who defeated 3 opponents, who defeated an opponent, who defeated 3 opponents, who defeated an opponent, who defeated 3 opponents, who defeated an opponent, who defeated 3 opponents, who defeated an opponent.
- or he has to defeat a launched opponent

This seems to exclude about 50% of all active boxers from being rated at the moment.

Now you asked:


... I wonder if there could be some sort of measure for "padded records" or "protected" fighters... ...

This is a key factor, I think too, we are chasing for times ...

What are your suggestions - might really help. Everyone is invited to propose ...

Looking forward to your response
Martin

Martin,

could you explain this a bit more. In particular.

a) what influence does 'launched' have on a fighter? Are unlaunched fighters not included in the ratings.

b) when does the 3 beat 3 beat 3 etc logic take effect. At the time of the bout or at is it run as an extra routinge at the end.

c) are their several different 'gates'? That would be my preferred concept to achieve the levels that I described earlier.

d) Martin, do I take this as an implicit acknowledgement that predictability is not everything and the appearance is worth something? Are you going to throw the 'gate' out if it reduces the predictability?

@Monty. Nice to see you posting again. Nice work picking out the 'level' statement in Martin's post. I implied that Martin knew what had to be done, but you did better and nailed it here in your posts.

@Lennox. The 3-1 statement was a cheap shot, but it would be fascinating to keep a tally in a separate thread. Hell I might even do that for a month or two, at least for prominent fights.

conan

[JCS]

The thing with introducing subjectivity is that you will still have people who think the ratings "don't look right". Then what you have is a battle between members and ratings people that CAN actually take place with BoxRec changing the ratings every other day. Debate is good, but in that battle, noone EVER wins. Now, we have a statistic to rely on rather than public opinion, which is a much more consisteny tally.

Is there a way to pad records with similar rated boxers to produce imaginary bouts in order to attempt to match the equality experienced in other sports? Do we need to modify the statistic we use? These are issues which I could see being discussed... but how do we do so w/o lowering the objectivity rating?

[JCS]

It must be possible to exclude subjectivity and still address this perceived problem.

I guess the IBO rankings are objective but do they necessarily prove that by predictability?

The whole thing is starting to boggle my mind! In ELO ratings a player can only progress if he does better than expected by some margin so if a player continually beats other players around his own rating he doesn't move up that much. Is this system not used in Boxrec? If it is, then surely this wouldn't happen?

Chess and Boxing are a whole different ball of wax.

[JCS]

Granted but the question is still valid I feel.

A similar system is used, but I do believe the most recent match/performance of a fighter is weighed more in this system than in the chess ELO system.

[conan_the_cribber]

I think that is quite sensible and I wonder then why the "padding" question should be a problem then. A boxer beats a whole host of second class fighters so he shouldn't move up a great deal. Nor should he be penalised.

There's probably something I'm missing regarding the answer to Monte's question.

Hi CWC,

the old routine used to work similar to the ELO principles. You got 85 for a KO win, but it was reduced by the difference in the ratings. I guess the reduction is not a much as it used to be.

conan

[Petu v.d. Pajm]

Well, these Georgian (country, not state) fighters are really learning how to manipulate the rankings. Fight weekly fights in your home country (with a big question whether they are fought or only results announced) and suddenly a 19-year-old kid without a recognisable victim is number 9 middleweight in the world ahead of f ex Javier Castillejo AND Felix Sturm, fresh Euro champ Amin Asikainen and loads of other worthy fighters...

How come, Martin?

[JCS]

Well, these Georgian (country, not state) fighters are really learning how to manipulate the rankings. Fight weekly fights in your home country (with a big question whether they are fought or only results announced) and suddenly a 19-year-old kid without a recognisable victim is number 9 middleweight in the world ahead of f ex Javier Castillejo AND Felix Sturm, fresh Euro champ Amin Asikainen and loads of other worthy fighters...

How come, Martin?

Launch mechanism is going to be hardened to fix this issue.

[conan_the_cribber]

Well, these Georgian (country, not state) fighters are really learning how to manipulate the rankings. Fight weekly fights in your home country (with a big question whether they are fought or only results announced) and suddenly a 19-year-old kid without a recognisable victim is number 9 middleweight in the world ahead of f ex Javier Castillejo AND Felix Sturm, fresh Euro champ Amin Asikainen and loads of other worthy fighters...

How come, Martin?

Ouch, that one hurts, doesn't it. Even with the 3 beats 3 algorithm, he gets thru I guess. That even makes Stipe Drews look good in comparison. For the record the guys name was Shalva Jomardashvili and he made it to number 9.

I think the promoters are on to you Martin.

conan

[JCS]

Well, these Georgian (country, not state) fighters are really learning how to manipulate the rankings. Fight weekly fights in your home country (with a big question whether they are fought or only results announced) and suddenly a 19-year-old kid without a recognisable victim is number 9 middleweight in the world ahead of f ex Javier Castillejo AND Felix Sturm, fresh Euro champ Amin Asikainen and loads of other worthy fighters...

How come, Martin?

Ouch, that one hurts, doesn't it. Even with the 3 beats 3 algorithm, he gets thru I guess. That even makes Stipe Drews look good in comparison. For the record the guys name was Shalva Jomardashvili and he made it to number 9.

I think the promoters are on to you Martin.

conan

I think in the revision, he's not even in the Top 400.. because he will be considered un-established.

[computerrank]

... he will be #517 ...

I proposed the revision for update, John is busy ...

[computerrank]

@conan

could you explain this a bit more. In particular.

a) what influence does 'launched' have on a fighter? Are unlaunched fighters not included in the ratings.

- They have no valid rating. A valid rating shows 300 rating points minimum.
- Their launch points are below 300 points.
- When you look at the Boxrec Ratings, all boxer showing 300 points minimum are rated. All boxers showing less than 300 points are not rated.

b) when does the 3 beat 3 beat 3 etc logic take effect. At the time of the bout or at is it run as an extra routinge at the end.

- at the time of the bout
- the launch points are changed with the result

c) are their several different 'gates'? That would be my preferred concept to achieve the levels that I described earlier.

- you step up by defeating unlaunched opponents with these achievements
- or you defeat a launched opponent
- so it is a procedure with many steps - the final gate is to get launched

d) Martin, do I take this as an implicit acknowledgement that predictability is not everything and the appearance is worth something? Are you going to throw the 'gate' out if it reduces the predictability?

- the prediction rate is calculated only from launched boxers - as unlaunched boxers have not valid rating
- so the launching process is the gate to ensure a valid rating
- maximum prediction rate for launched boxers is still the only target and measure

Best regards
Martin

[computerrank]

I think that is quite sensible and I wonder then why the "padding" question should be a problem then. A boxer beats a whole host of second class fighters so he shouldn't move up a great deal. Nor should he be penalised.

There's probably something I'm missing regarding the answer to Monte's question.

Cobwebcat

The Boxrec system is of the same class as ELO.

ELO used an exponential function of rating difference for prediciton.

Today US Chess uses an logistic function of rating difference - a bit different - not much.

Newest studies show a linear function of rating difference depending on the special decision seems to be even better. This is used by Boxrec.

But overall the results are not very depending on the functional form of the predictor.

Overall we have a situation, where you can say a boxer will succeed with an probability of

- 50 % for a rating difference of 0
- 65 % for a rating difference of 100
- 75 % for a rating difference of 200
- 85 % for a rating difference of 300
- 90 % for a rating difference of 400

Best regards
Martin

[computerrank]

...

...


computerrank, if you could enlighten me on the different ways of measuring predictability you've used...and what you found about that, then I would be really interested. What I mean is using only elite boxers as a measure (with probably no regard to "lower" boxers at all) and defining and shaping the formulas around that? I think that could be how your basic premise could still be used and accepted by me and all those fanatics out there.

Thanks

Monty.



thanks for your response.

We measure the 2 prediciton rates for bouts since 2000-01-01:

- between launched boxers (some 30,000 bouts)
- between top boxers with more than 75% win rate (some 3,000 bouts)

from about 100,000 bouts in this period.

The predicition rate is about 78% for top bouts.

The prediciton rate is about 83% for bouts between launched boxers

We could observe that rule variations have an impact to both rates in the same way - sometimes not with same gain or loss of percentage - but same direction. So currently I think the rules apply for launched and top boxers in the same way.

I still hope to find a record pattern of boxers, we all could agree to be overprotected or padded, - and then restrict their rating to comply with their results without overrating them.

Basic idea: If their records are overrated, then their opponents' should be too. So how to handle down this network of overrated boxers - if so.

Best reagrds
Martin

[computerrank]

Dear all.

I implemented a strict ranking scheme - based on ABC common suggestions:

- only bouts in division are considered
- a boxer starts at the end of the list
- a winner gets the rank of the loser - only some ranks lower, if the gap was very high
- a boxer loses some ranks, when he has no valid wins in the last 18 months - valid means defeating an opponent ranked maximum 3 times higher than himself, a minimum gap of 30 ranks is allowed.
- a boxer is sorted out of the ranks, when he has no fight in the last 18 months
- his last ranked is registrated, and will be used to sort him in after a come-back - but he will lose some ranks, depening on time of inactivity

And this is the result for heavyweights:

1 Wladimir Klitschko
2 Chris Byrd
3 Hasim Rahman
4 Jameel McCline
5 Nikolay Valuev
6 Calvin Brock
7 John Ruiz
8 Zuri Lawrence
9 Serguei Lyakhovich
10 Lamon Brewster
11 Matt Skelton
12 Danny Williams
13 DaVarryl Williamson
14 James Toney
15 Oliver McCall
16 Fres Oquendo
17 Javier Mora
18 Shannon Briggs
19 Monte Barrett
20 Henry Akinwande

Any opinions?

Perhaps John could be convinced to show such traditional rankings additionaly to the current performance ratings?

So one ranking for the performance fanatics - one for the traditionalists.

I cannot hold back -

- prediciton rate of performance ratings 83 %
- prediciton rate of tratitional rankings 68 %

I look forward to your response...

Best regards
Martin

[JCS]

It'd be interesting to see both sets.. but I would NOT replace the current system w/ that particular ranking scheme.