generic screen name wrote:Its totally bias to undefeated boxers, which is why alot of prospects who have a minor and/or crappy belt are up in the rankings. Look at the lb for lb ranking, Isaac Hlatshwayo is ahead of Jose Luis Castillo. Isaac Hlatshwayo!!!!!
#1 at Heavy is going to bounce around until someone gets a solid foothold on it and fights enough to retain it. There's just no dominant force there.
Fair enough, but Calvin Brock? That is ridiculous the guy hasn't even won an eliminator and is fighting for continental titles. He's top 10 but never anywhere near number 1. These rankings blow!!
#1 at Heavy is going to bounce around until someone gets a solid foothold on it and fights enough to retain it. There's just no dominant force there.
Fair enough, but Calvin Brock? That is ridiculous the guy hasn't even won an eliminator and is fighting for continental titles. He's top 10 but never anywhere near number 1. These rankings blow!!
They sure do blow if you're comparing them to all the other typical ranking schemes out there.. but its a power/performance rating.. not a who beat who.
JCS83MD wrote:
#1 at Heavy is going to bounce around until someone gets a solid foothold on it and fights enough to retain it. There's just no dominant force there.
Fair enough, but Calvin Brock? That is ridiculous the guy hasn't even won an eliminator and is fighting for continental titles. He's top 10 but never anywhere near number 1. These rankings blow!!
They sure do blow if you're comparing them to all the other typical ranking schemes out there.. but its a power/performance rating.. not a who beat who.
I know that. But going by whats been said about these rankings it predicts he'd beat Wlad and Valuev and all those below - all because he beat Ibragimov, Lawrence, Bostice and McCline (I wont mention Kenny Craven).
nickd wrote:
Fair enough, but Calvin Brock? That is ridiculous the guy hasn't even won an eliminator and is fighting for continental titles. He's top 10 but never anywhere near number 1. These rankings blow!!
They sure do blow if you're comparing them to all the other typical ranking schemes out there.. but its a power/performance rating.. not a who beat who.
I know that. But going by whats been said about these rankings it predicts he'd beat Wlad and Valuev and all those below - all because he beat Ibragimov, Lawrence, Bostice and McCline (I wont mention Kenny Craven).
This thing is seriously flawed.
I invite you to do a better job. It doesn't guarantee he'd beat them.. it just favors him.
Right this second actually.. it'd favor Klitschko, Valuev if the fight was in Germany
Monty, been a year or so, but I thought I'd drop in and see what's up. Looks like you got a couple of jihad programmers in here. I wish I could win lotto and have enough time to fix up the boxrec rankings. I admire you for keeping on fighting against this travesty. And it really is a travesty. To have all the information that you need to make the perfect ratings system and then to f+ck it up so bad is a travesty. These rankings suck.
To Mr Boxrec Statisti or whomever it may concern. I think you need some help and need to step back and take a deep breath and think about what you are doing. Here are some comments from a statistician (me) regarding my negative comments above.
a) Every statistical problem has a domain space. Heard of that? b) Every heard of a plausability check, I'll explain that too.
Domain space
-----------------
The domain space dictates what statistical solutions you can apply and what results you can recommend. For example, comparing two drugs in a medical trials. Although drug A is perhaps far better than the other in most cases, in some random cases it kills people. Drug B may have only mild success, but it is safe. There is no point recommending Drug A although it is statistically better. This decision is because of the "domain space" i.e. any random risk of killing people is bad medically.
The domain space of boxing dictates certain things. Here are two biggies.
a) An up and coming fighter can beat a thousand stiffs and should never ever be the number one ranked fighter. There is always the test of fire required and many a Hector Camacho Jnr type fighter has failed this test.
b) The results of direct comparisions between two fighters are the ultimate guide to relationship between two boxers. If A beats B, then A at that moment in time, then A is better than B. If B was an elite fighter and A's win was an upset, then that probably means, that B was fading.
Plausability checks.
---------------------
Plausability checks are used to do a quick common sense analysis of your result. This is particularly useful in complex algorithms, as is here the case. Here's an example.
Let's say I do a marketing survey and use some new stat tools and come up with a statistic that 98% of people in Wyoming prefer Coke to Pepsi. My first task as a statistician is not just to blindly accept this result because the data supports it, but rather to look at with "common sense" eyes. I doubt very much that 98% prefer Coke to Pepsi, certainly the sales figures dont support it. If the statistical formula is right, and the data is right, then what was the problem. In this case, it was the fact that Coke carried out the survey themselves and the people asked were Coke employees or relatives or friends of Coke employees. Fortunately because I used a plausability check, I refused to publish the crap findings that 98% of people prefer Coke.
And now the big question for you Mr Boxrec Statistician. What is your plausability check? Please think about this for 30 seconds before continuing to read. Please. Think.
Think, what is my plausability check?
.....think.....
.....think some more......
....think hard, please, I'm not trying to be condiscending, how do you validate the results of your complex algorithm......
....think....
I will tell you now. I hope you thought about it though. The plausability check that you have is, the feedback from boxing fans and the message that you are getting is that the rankings are very poor. If you set up a thread and say, who is the best Light Heavyweight. I am prepared to say that in an honest poll 98% of the people will not vote for Chad Friggen Dawson. Especially if Jones, Hopkins, Woods, Tarver, Briggs, Erdei etc are in the poll. If you know boxing, then you know that what I say is true. It is completely implausible to have him as champ. Ditto for Calvin Brock at Heavy. Your results fail the single most valid plausability check that you have available.
I will tell you what you are doing wrong now. You are using a single "predictability" statistic as your plausability check. That is exactly the same as the 98% people prefer Coke story. The algorithm is right, the data is right, but the results suck. You have to dig deeper than just a single statistic to prove the virtues of your algorithm. By not digging deeper or accomodating the realities of the boxing domain space, you fail the real plausability check i.e. informed opinion.
Predictability, has a roll in validating the generated rankings, but not an exclusive one. You need to incorporate the pecularities of the boxing domain space. Only then can you come up with a plausible set of rankings. It is entirely conceivable, that with a reduction of 2 or 3% in predictability, that you generate a list that finds 90% acceptance amongst the boxing fans. If you persist in something that is implausable, then sooner or later the boxrec ratings will have no weight in the free press at all, they certainly have already lost that credability with the boxrec forum.
Using a statistic to justify another statistic is a really rocky path. Please take a second to reconsider your line of thinking. Before I go, here's another little statistical anecdote. Ever heard of the Coke causes polio statistic? Before they had a vaccine for polio, it was noted that the number of new cases of polio in the USA grew proportionally to the amount of Coke consumed. The results were truly eerie, it was almost a perfect fit, too much to be a coincidence. Was Coke one of the causes of polio? Certainly the statistics proved this. However again a plausability check saved the day. Common sense dictates that a sugar based drink doesn't cause polio. The simple reason that the stats fitted so good, was that polio was more common in summer, and that the consumption of Coke also went up in summer.
With that as a farewll I advise you, don't blindly use statistics, otherwise you will produce implausible and nonsensical results.
Be sure, we are very aware of your arguements - I read your message very carefully.
We know, it would cost "only" a few percents to make the ratings much more acceptable to most.
We decided not to do this, and to publish a pure performance rating, optimized for best prediciton only. We are not marketing our ratings, we publish the best performing ratings available. It is not our goal to be acceptable for the traditional exspectations, we know. This is not our domain.
Regarding your statement, that boxers never should be in the tops without defeating opponents near the same level:
- We studied this aspect very deeply
- We compared the top boxers' ratings and the level of their opponents' ratings and found (by comparing the pre-bout ratings and the results):
- Even for a boxer, only boxing and defeating opponents 300 and 400 points lower - the probability of
-- winning or losing to an opponent of his own rating level is about 50%,
-- is more than 50% for opponents, rated only a little bit lower
-- and of course less than 50% for opponents, rated only a little bit higher
-- so the ratings are quite consistent in this sense
This simply means, the current ratings are quite correct - and it means that these boxers have quite a good chance to defeat opponents of definite higher level than they might have chosen before - as soon as they step up.
This also means - and I understand this - we all would prefer for sure, to see every boxer box at his approriate level. But unluckily they don't.
And - in spite of this, the majority of these risk-minimizing boxers perform better than most would admit, as soon as they dare.
These ratings are not made to punish boxers - they show the reality of based exspectations.
The traditional ranking style systematically under-estimates the performance of such boxers and so over-estimates the performance of the old guys.
You say you consider all aspects carefully? Then why do you abandon common sense as one of your aspects? Your entire plausability is based solely around a single statistic.
Mathematically speaking, what you are doing is indefensible. You have a multi-variate equation consisting of at least 10 variables (ko pts, draw pts, time decay etc etc). Finding the optimal parameters for this is like finding a needle in a hay stack and you can't solve it theoretically. You might have 85 for a KO and 75 for a draw but the values that truely optimizes the prediction percentage are really 1678940 for a KO win and 1546779 for a draw.
Maybe if you use these values you will produce a list that corresponds to the human generated lists. Maybe with these values you will create a list where, in your words, the experienced fighters are not a tick overrated and and the up and comers are not underrated. But unless you have a spare CRAY computer, you are not going to be able to find this out.
All you are doing is dicking around with a few values each week. You can
never say, that you have the optimum rankings for predictability. IN THE
MEANTIME, YOU UNASHAMABLY DESTROY THE CREDABILITY OF THE BOXREC RATINGS ON THE MERE PRETENCE THAT YOU HAVE THE OPTIMUM PREDICTABILITY. Chad Dawson at 1 and Stipe Drews at 2, if that isn't enough evidence that the algorithm is so completely way off optimal, then god knows what is. Maybe RJJ and Sugar Ray Robinson have to end up at 1 and 2 before you look at the ratings, take a sniff, and decide that they stink.
I haven't got time to care and do the rankings myself. In fact I will only come here for a day or so a year. But on that day I will let you know that what you are doing is statistically, mathematically and boxing-wise completely unsound. I had actually hoped that someone else had taken it over, someone who doesn't have a jihad approach to the predicability statistic, but no luck. That's why the rankings have been getting progressively worse over time. Right now they are a laughing stock.
You say you want to create something new. But at some point you have to step back and say, "what I've created is a piece of shit compared to before". Dude, the emporer has no clothes on. The only difference that fairy tale and here is that it's not a single brave child telling the truth, but a whole town full of adults and children telling you the same thing. Pull out the ear plugs dude. You need a better plausability check. Simple.
Your simplistic argumentation is an inadequate defence for the critisms of the poll, but a perfect one for someone who wants to live in their little shell and ignore the outside world. It's so isolated from the real world that it's almost autistic.
Here's a rhetorical question for you. How bad do your ratings have to be before you won't publish them? If fighter A beats fighter B twice in a row in successive fights, then are you prepared ratings where B is still ahead of A? Are you prepared to publish ratings where punch drunk fighters like Evander Holyfield or Riddick Bowe end up in the top 10? Don't even bother answering this one, it's like asking a flat earther if the earth is round. I know the answer. If you're prepared to publish Chad Dawson at 1 and Stipe Drews at 2 then you are prepared to publish anything, so long as your precious little autistic statistic goes up 0.002% compared to the last iteration.
Even the claim about being the "best predictative ratings" is a complete farce. "Best" compared to what. You done any long term predictability comparisons between your rankings and Rings or Scott Shaeffers at boxingtalk? No of course not. Their rankings might smash yours for predictability. The only "best" you have is artificial "best set of numbers I used for my multivariate equation". The fact that the "best set of numbers" produced crap ratings, is tragically irrelevant to you.
What a sorry waste of a life making the "75% predictability ratings". All tied up in my little algorithm, tweaking a few numbers and getting a 0.00003% improvement. Who gives a toss about such a statistic. Wake up dude, no-one cares, but a whole lot of people care about producing something worthwhile, like ratings that a site can be proud of. I repeat, right now the rankings are a laughing stock.
To JC8Whatever,
your ignorance of the issues shows through in your reply to my first post and the way that you murdered the other poll I set up with an inane discussion about hung toenails. Stop following Martin blindly and think about what you are doing. Read your way thru the abuse directed at Martin here to understand the foolishness of your endeavour.
Its a good thing you are a statistician, because you have the "people skills" and the professionalism of a hermit stuck in his cube for hours on end. You simply cannot construct a paragraph without harrassing something or someone, a trait which is shared by some of this forum so if you do choose to post more than once a year, trust me, you'd fit in.
Its no doubt you make many friends with your interests which consist of "slaying beasts" and "defeating evil wizards". You no doubt, are surely one that is taken seriously in all avenues of life. Instead of reiterating your opinion in that they suck and what not, how about taking some time and creating a rating system.
By the way, what's the point of creating a computerized system that mocks "The Ring" and every other human list out there that is simply a near-mirror image of each other's?
The appearance of the current system is not a mature product but I assure you its predictability is far superior to previous system's on this site. How does it compare to other systems?? That I am not sure, but perhaps Martin can provide some intellect on that subject as I do believe he has compared systems in the past.
Conan, you have to realize your coke to pepsi theory is a bit dodgy in regards to relevance.
Yes, I understand you're basically comparing the satisfaction of one particular survey to the accuracy of the rankings system.
Put it this way:
You compare coke to pepsi, two products that 100% are direct competition and have competed against each other. Now, say you throw in Mountain Dew and Sprite. Two more products, no exactly competitors of each other other than both being soft drinks, no similar flavor, and you never see them compared. Now, how would you measure a mountain dew/sprite/coke/pepsi challenge, when 50% of the sample doesn't compare and perhaps never will? It's like comparing Erdei to Dawson, and they haven't fought yet, and it is a possibility that they never will.
Dawson, Adamek, Erdei, and Briggs. Two of the 4 have fought, one is just making his way on the scene, and may not have a big fan base, and the other has hardcore fans, but outside of them, no one really looks at them as a world beater. However, does fan base and following make you a better fighter? No. It just means you have more exposure, a better promoter, maybe a better managerial team that gets you winnable fights in front of that crowd that adores you. Compare Erdei to the 98%. Erdei fights in front of German fans who love him every fight. Of course if you poll those fans, you'll get a high rating for him. Ask them who Chad Dawson is, they probably won't have a fvcking clue at this time. However, people like myself believe all 3 other people in the example of 4 I brought up would beat Erdei.
In polls/taste test it's a matter of personal bias, ignorance, and outside circumstances. If my dad was the CEO at Coke, I'd be the biggest Coke drinker in the world. Just the same as if I was Hungarian, I'd think Erdei was THEE best LHW in the World....feel free to use that bias as I am American and so is Dawson, which brings to light the problem with polls showing the accuracy/effectiveness of the rating system here.
People let biases other than pure boxing ability get in the way of their decisions when it comes to absolutes. Take a look at me and JCS talking on your thread. I honestly wouldn't say who's better out of Dawson and Adamek, but I see those two as the best. Does that make the system wrong and bad for not having Adamek #1 or #2? No. It's a prediction based system and not a poll, and that's where you seem to miss the point.
1st post, very helpful and polite. 2nd post, very helpful and not polite. Not everything laced with poison, you must admit. BTW I hate dungeons and dragons and enjoy the time with my kids the most. For exactly that reason, I will not slave away at a rankings algorithm.
But my inactivity, does not justify your results, nor does it invalidate my arguments. Nor does the tone of my argument, invalidate the well constructed critisms therein. To do that is just as cheap as saying "I have my predicability statistic and that's all I need".
What you have to do is try to answer my rhetorical question. Just how bad do the rankings have to suck, before you wont publish them? If you say "never" then it is truly you, not I that are so locked up from the real world, playing dungeons and dragons in terms of fighting the evil algorithm and the low predicability witch of the future.
Go on give it a try. How bad would they have to suck?
1st post, very helpful and polite. 2nd post, very helpful and not polite. Not everything laced with poison, you must admit. BTW I hate dungeons and dragons and enjoy the time with my kids the most. For exactly that reason, I will not slave away at a rankings algorithm.
But my inactivity, does not justify your results, nor does it invalidate my arguments. Nor does the tone of my argument, invalidate the well constructed critisms therein. To do that is just as cheap as saying "I have my predicability statistic and that's all I need".
What you have to do is try to answer my rhetorical question. Just how bad do the rankings have to suck, before you wont publish them? If you say "never" then it is truly you, not I that are so locked up from the real world, playing dungeons and dragons in terms of fighting the evil algorithm and the low predicability witch of the future.
Go on give it a try. How bad would they have to suck?
conan
Martin and John decide on the particular prediction stat to use. Right now it is considering fights with boxers who are "established". If that stat goes up, the rating set is considered better.. considering there are no programmatical errors that would screw up the counts.
Mattyp151 wrote:Conan, you have to realize your coke to pepsi theory is a bit dodgy in regards to relevance.
Yes, I understand you're basically comparing the satisfaction of one particular survey to the accuracy of the rankings system.
Put it this way:
You compare coke to pepsi, two products that 100% are direct competition and have competed against each other. Now, say you throw in Mountain Dew and Sprite. Two more products, no exactly competitors of each other other than both being soft drinks, no similar flavor, and you never see them compared. Now, how would you measure a mountain dew/sprite/coke/pepsi challenge, when 50% of the sample doesn't compare and perhaps never will? It's like comparing Erdei to Dawson, and they haven't fought yet, and it is a possibility that they never will.
Dawson, Adamek, Erdei, and Briggs. Two of the 4 have fought, one is just making his way on the scene, and may not have a big fan base, and the other has hardcore fans, but outside of them, no one really looks at them as a world beater. However, does fan base and following make you a better fighter? No. It just means you have more exposure, a better promoter, maybe a better managerial team that gets you winnable fights in front of that crowd that adores you. Compare Erdei to the 98%. Erdei fights in front of German fans who love him every fight. Of course if you poll those fans, you'll get a high rating for him. Ask them who Chad Dawson is, they probably won't have a fvcking clue at this time. However, people like myself believe all 3 other people in the example of 4 I brought up would beat Erdei.
In polls/taste test it's a matter of personal bias, ignorance, and outside circumstances. If my dad was the CEO at Coke, I'd be the biggest Coke drinker in the world. Just the same as if I was Hungarian, I'd think Erdei was THEE best LHW in the World....feel free to use that bias as I am American and so is Dawson, which brings to light the problem with polls showing the accuracy/effectiveness of the rating system here.
People let biases other than pure boxing ability get in the way of their decisions when it comes to absolutes. Take a look at me and JCS talking on your thread. I honestly wouldn't say who's better out of Dawson and Adamek, but I see those two as the best. Does that make the system wrong and bad for not having Adamek #1 or #2? No. It's a prediction based system and not a poll, and that's where you seem to miss the point.
Matty thanks for your mail.
In what context did I use Coke vs Pepsi? I used it to illustrate the importance of plausability checks in a statistical undertaking. That's all. The point at the end of day remains, if you do something complicated statistically, then you better well check it with plausability checks. In this case, your plausability check is "how many people would come up with Dawson #1 and Drews #2" (actually as an aside, Drews at #2 is even worse than Dawson at #1).
What you think Matty? I ask 10 ranking afficiandos to put down their rankings, how many choose Dawson 1 and Drews 2? How about I ask 100 boxing journalists, how many get the same 1, 2. Or if I ask 1000 hard core fans, how many get the same one two? I consider it likely, that not a single person would rank the same one, two as the algorithm here. That's a pretty miserable result for a plausability check.
And that is my point.
And JCS admitted the rest, there is no predictability comparison to Ring or boxingtalk ratings, so you might be producing the worst of the three for all you know. Certainly the plausibility checks would indicate this.
1st post, very helpful and polite. 2nd post, very helpful and not polite. Not everything laced with poison, you must admit. BTW I hate dungeons and dragons and enjoy the time with my kids the most. For exactly that reason, I will not slave away at a rankings algorithm.
But my inactivity, does not justify your results, nor does it invalidate my arguments. Nor does the tone of my argument, invalidate the well constructed critisms therein. To do that is just as cheap as saying "I have my predicability statistic and that's all I need".
What you have to do is try to answer my rhetorical question. Just how bad do the rankings have to suck, before you wont publish them? If you say "never" then it is truly you, not I that are so locked up from the real world, playing dungeons and dragons in terms of fighting the evil algorithm and the low predicability witch of the future.
Go on give it a try. How bad would they have to suck?
conan
Martin and John decide on the particular prediction stat to use. Right now it is considering fights with boxers who are "established". If that stat goes up, the rating set is considered better.. considering there are no programmatical errors that would screw up the counts.
So unless John chips in with some common sense, we could see Riddick Bowe in the top 10 again. Certainly it seems you and Martin would be willing to print it (as you did not too long ago).
1st post, very helpful and polite. 2nd post, very helpful and not polite. Not everything laced with poison, you must admit. BTW I hate dungeons and dragons and enjoy the time with my kids the most. For exactly that reason, I will not slave away at a rankings algorithm.
But my inactivity, does not justify your results, nor does it invalidate my arguments. Nor does the tone of my argument, invalidate the well constructed critisms therein. To do that is just as cheap as saying "I have my predicability statistic and that's all I need".
What you have to do is try to answer my rhetorical question. Just how bad do the rankings have to suck, before you wont publish them? If you say "never" then it is truly you, not I that are so locked up from the real world, playing dungeons and dragons in terms of fighting the evil algorithm and the low predicability witch of the future.
Go on give it a try. How bad would they have to suck?
conan
Martin and John decide on the particular prediction stat to use. Right now it is considering fights with boxers who are "established". If that stat goes up, the rating set is considered better.. considering there are no programmatical errors that would screw up the counts.
So unless John chips in with some common sense, we could see Riddick Bowe in the top 10 again. Certainly it seems you and Martin would be willing to print it (as you did not too long ago).
conan
I don't think Bowe is going to get into the Top 10
Sorry JCS, he was there (or at least top 15). And it was in the last year. There were several versions of the algorithm that put too much weight on retirees coming back.
And that is my point. Martin will publish anything at all, even Bowe in the top 10. There is no plausability check.
What about you? You still haven't said where you would draw the line.
conan_the_cribber wrote:Sorry JCS, he was there (or at least top 15). And it was in the last year. There were several versions of the algorithm that put too much weight on retirees coming back.
And that is my point. Martin will publish anything at all, even Bowe in the top 10. There is no plausability check.
What about you? You still haven't said where you would draw the line.
conan
I am not sure what was going on at that time.. I wasn't really active in the ratings discussions/helping out back then. I started contributing about 8 months ago.
That was probably back when the "Traditional ranking" system had major flaws with decay rates.
If Bowe made it into the Top 10 or 15, then I assure you there is a major problem with some programming that affects prediction counts.
BTW, we have a revision in the queue waiting to be posted by the site mod. Not sure when that will make it on live though.
conan_the_cribber wrote:Sorry JCS, he was there (or at least top 15). And it was in the last year. There were several versions of the algorithm that put too much weight on retirees coming back.
And that is my point. Martin will publish anything at all, even Bowe in the top 10. There is no plausability check.
What about you? You still haven't said where you would draw the line.
conan
I am not sure what was going on at that time.. I wasn't really active in the ratings discussions/helping out back then. I started contributing about 8 months ago.
That was probably back when the "Traditional ranking" system had major flaws with decay rates.
If Bowe made it into the Top 10 or 15, then I assure you there is a major problem with some programming that affects prediction counts.
BTW, we have a revision in the queue waiting to be posted by the site mod. Not sure when that will make it on live though.
Wrong. There was no programming mistake. There was no "traditional algorithm". There was just the same old algorithm as ever, just with the solution set for multivariable equation back then, it produced the very charming effect of Bowe in the top 10 (or 15). However, as always, because it produced the best "predictability", the farcical ratings were published.
And you still haven't answered the question. Assuming no programming mistakes, would you publish anything at all? Martin would/has.
Besides, what do you mean "major flaws". The last time I was in this debate, Martin was clamiing 80% predictability. Now it's only 75% or so. That would mean this is more flawed than before. And I beg you, what could be more flawed than Chad Dawson #1 and Stipe Drews #2. Maybe 1 in a 1000 boxing fans would choose that. Maybe 800 or so would have one of Adamek Tarver, Tarver Adamek, Johnson Adamek, Adamek Johson, Johnson Tarver, TArver Johson. But none would have Dawson/Drews. That is so flawed.
