Ratings - please read before commenting - Archived

Lennox · Post by **Lennox** » 14 May 2006, 13:46

Sorry Computerrank, but I do not understand.

At the moment with present Boxrec system it is:
15 common upsets
8 upsets unique to boxrec
5 upsets unique to ibo

23-20 you could say

Do you dispute?

If you have a new system perhaps when that is active it would be the time to compare. If you can produce a set of ratings that look more like 'the order of the best 100 fighters in each division at this moment in time' then I think you have a winner, if you can achieve 0% orediction upset then well done there will be no need to actually have the fight as we will already know the result.

JCS · Post by **JCS** » 14 May 2006, 14:04

Lennox wrote:
jcs83md wrote:still missing the point on this "study". Lennox you said prediction rate doesn't matter yet you're tracking upsets..
Yes, Jcs I said i'd look at both sets of ratings for a month, to see if IBO prediction rate is vastly inferior to Boxrec. Some have said that is the criteria for 'good ratings'. It would appear so far that Boxrec does not have superior prediction rate to IBO. In general upsets are common in both and the marginal ones that are unique to IBO or Boxrec are immaterial. Perhaps it will change over this next week.

I have and still maintain a good set of ratings have to look good, no one will take a set ratings seriously with Luis Collazo at number 51, or Virgil Hill at number 29, for that reason I maintain Boxrec ratings are the let down of this very fine site. I do not think anyone in their right mind would say Boxrec ratings are superior to IBO. Just because IBO does not publish its formula does not make it crap, Coca Cola do not publish their recipe after all.

I am more convinced 'Prediction method' is absolutely not how to produce a set of boxing ratings, they should mirror what has happpened rather than what might happen which is IBO philosophy. Luis Collazo at 51 and some novice who is 7-0 at number 50, you are joking.

You're like talking to a brick wall.. For the hundredth time.. the BoxRec rating system does not use prediction rate as its sole basis.... it too must look acceptable. However, what better metric is there to use when trying to figure out what version of a rating system to use?

You don't understand. You can run one rating system today and it'll look great, but a year later it may look like total trash... and it is simply due to patterns throughout time caused by fighters, decisions, etc.

JCS · Post by **JCS** » 14 May 2006, 14:07

Lennox wrote:Sorry Computerrank, but I do not understand.

At the moment with present Boxrec system it is:
15 common upsets
8 upsets unique to boxrec
5 upsets unique to ibo

23-20 you could say

Do you dispute?

If you have a new system perhaps when that is active it would be the time to compare. If you can produce a set of ratings that look more like 'the order of the best 100 fighters in each division at this moment in time' then I think you have a winner, if you can achieve 0% orediction upset then well done there will be no need to actually have the fight as we will already know the result.

I believe C-Rank is saying that his new system had 0 upsets (aside from common upsets) in the same timespan you were comparing.

Lennox · Post by **Lennox** » 14 May 2006, 14:45

jcs83md wrote:
Lennox wrote:
jcs83md wrote:still missing the point on this "study". Lennox you said prediction rate doesn't matter yet you're tracking upsets..
Yes, Jcs I said i'd look at both sets of ratings for a month, to see if IBO prediction rate is vastly inferior to Boxrec. Some have said that is the criteria for 'good ratings'. It would appear so far that Boxrec does not have superior prediction rate to IBO. In general upsets are common in both and the marginal ones that are unique to IBO or Boxrec are immaterial. Perhaps it will change over this next week.

I have and still maintain a good set of ratings have to look good, no one will take a set ratings seriously with Luis Collazo at number 51, or Virgil Hill at number 29, for that reason I maintain Boxrec ratings are the let down of this very fine site. I do not think anyone in their right mind would say Boxrec ratings are superior to IBO. Just because IBO does not publish its formula does not make it crap, Coca Cola do not publish their recipe after all.

I am more convinced 'Prediction method' is absolutely not how to produce a set of boxing ratings, they should mirror what has happpened rather than what might happen which is IBO philosophy. Luis Collazo at 51 and some novice who is 7-0 at number 50, you are joking.
You're like talking to a brick wall.. For the hundredth time.. the BoxRec rating system does not use prediction rate as its sole basis.... it too must look acceptable. However, what better metric is there to use when trying to figure out what version of a rating system to use?

You don't understand. You can run one rating system today and it'll look great, but a year later it may look like total trash... and it is simply due to patterns throughout time caused by fighters, decisions, etc.

JCS no it is you not understanding. Computer rank said he proved that Boxrec was better than IBO, and GermanB and johnshep said it was a fact. So far it is not a fact. Whatever formula Boxrec uses, it is not good, but perhaps the new one is better. My trial will continue for one more week. I would like to see your top 100 in heavyweights, of its good I will say so. I am exactly not like talking to the wall, but the rest rating system is the one the compares to mosts opinion, if that can be done by a computer then good. You have never said which rating system you think is better, current Boxrec or IBO, which is it?

JCS · Post by **JCS** » 14 May 2006, 14:52

Lennox wrote:
jcs83md wrote:
Lennox wrote: Yes, Jcs I said i'd look at both sets of ratings for a month, to see if IBO prediction rate is vastly inferior to Boxrec. Some have said that is the criteria for 'good ratings'. It would appear so far that Boxrec does not have superior prediction rate to IBO. In general upsets are common in both and the marginal ones that are unique to IBO or Boxrec are immaterial. Perhaps it will change over this next week.

I have and still maintain a good set of ratings have to look good, no one will take a set ratings seriously with Luis Collazo at number 51, or Virgil Hill at number 29, for that reason I maintain Boxrec ratings are the let down of this very fine site. I do not think anyone in their right mind would say Boxrec ratings are superior to IBO. Just because IBO does not publish its formula does not make it crap, Coca Cola do not publish their recipe after all.

I am more convinced 'Prediction method' is absolutely not how to produce a set of boxing ratings, they should mirror what has happpened rather than what might happen which is IBO philosophy. Luis Collazo at 51 and some novice who is 7-0 at number 50, you are joking.
You're like talking to a brick wall.. For the hundredth time.. the BoxRec rating system does not use prediction rate as its sole basis.... it too must look acceptable. However, what better metric is there to use when trying to figure out what version of a rating system to use?

You don't understand. You can run one rating system today and it'll look great, but a year later it may look like total trash... and it is simply due to patterns throughout time caused by fighters, decisions, etc.
JCS no it is you not understanding. Computer rank said he proved that Boxrec was better than IBO, and GermanB and johnshep said it was a fact. So far it is not a fact. Whatever formula Boxrec uses, it is not good, but perhaps the new one is better. My trial will continue for one more week. I would like to see your top 100 in heavyweights, of its good I will say so. I am exactly not like talking to the wall, but the rest rating system is the one the compares to mosts opinion, if that can be done by a computer then good. You have never said which rating system you think is better, current Boxrec or IBO, which is it?

You keep claiming prediction rate is not the best way to make a rating system though... but wouldn't a better prediction rate most likely cut down on these upsets you speak of?

C-Rank's new system had 0 uncommon upsets in the same span as your analysis.. but I bet you won't like the look of the ratings..

So where is the happy medium??? Do you purposely cut down on the prediction rate to make you happy?? Do you do so to make me happy, or Martin happy, or JohnShep happy?? Because, in order to alter the rating system for the approval of public opinion, you will lower the prediction rate. Where is the "sweet science" in that? If you do alter the system simply for the onlooker's perception, what happens in 1 year? Do you change it again so the public approves??

How do you draw the line? I tell you what you do.. You take the best prediction rate possible and post it, because a computer will never ever, reach the level of subjectivity and intuition that an experienced person's ratings will possess. Let the people bitch and whine all they want.. but what more can you expect from a computerized system?

That is the final answer, IMO

computerrank · Post by **computerrank** » 14 May 2006, 15:03

jcs83md wrote:
Lennox wrote:Sorry Computerrank, but I do not understand.

At the moment with present Boxrec system it is:
15 common upsets
8 upsets unique to boxrec
5 upsets unique to ibo

23-20 you could say

Do you dispute?

If you have a new system perhaps when that is active it would be the time to compare. If you can produce a set of ratings that look more like 'the order of the best 100 fighters in each division at this moment in time' then I think you have a winner, if you can achieve 0% orediction upset then well done there will be no need to actually have the fight as we will already know the result.
I believe C-Rank is saying that his new system had 0 upsets (aside from common upsets) in the same timespan you were comparing.

... exactly ...

Lennox · Post by **Lennox** » 14 May 2006, 15:03

I think there will always be upsets and prediction rate counts for little, as I have said lots of times I am more intrested in a quality set of ratings thats puts the boxers in an order of mert, the ones nearer the top being the deserving ones to challenge for titles. We know bright prospects will beat older fighters that are past their peak, but the bright prospects should be ranked lower than the proven fighter until such time as the results has been proved, that in most cases would be an upset, long may it continue.
You never answered my last question. Better ratings IBO or Boxrec (current)?

Lennox · Post by **Lennox** » 17 May 2006, 05:28

One more upset in Boxrec

* Adrian Diaconu W12 Andre Thysse, Diaconu 34 IBO 50 BR
Thysse 44 IBO 44 BR

Common upsets 15
Upsets only to Boxrec 9
Upsets only to IBO 5

computerrank · Post by **computerrank** » 17 May 2006, 07:12

Lennox wrote:One more upset in Boxrec

* Adrian Diaconu W12 Andre Thysse, Diaconu 34 IBO 50 BR
Thysse 44 IBO 44 BR

Common upsets 15
Upsets only to Boxrec 9
Upsets only to IBO 5

In new BoxRec Ratings before the bout: Diaconu #6 v. Thysse #69

So all-in-all still:

Common upsets 7
Upsets only to Boxrec 0
Upsets only to IBO 12

Lennox · Post by **Lennox** » 17 May 2006, 08:52

No Computerrank. Play the game fair and show some grace.

I am doing it by the boxrec ratings as posted.
When you post your new ratings we can conduct a new study.

In the meantime IBO ratings show BETTER prediction than Boxrec.

JCS · Post by **JCS** » 17 May 2006, 08:54

LOL.. Martin I think its time to implement the new system

BTW Lennox.. I've seen it work.. I don't think he's lying about that.

computerrank · Post by **computerrank** » 17 May 2006, 09:07

The new ratings could be launched at once, I learned.

I asked for some time to do some last tests - the ratings shouldn't be changed too often in a short time.

JCS · Post by **JCS** » 17 May 2006, 09:23

computerrank wrote:The new ratings could be launched at once, I learned.

I asked for some time to do some last tests - the ratings shouldn't be changed too often in a short time.

Don't leave us hanging too long!

Lennox · Post by **Lennox** » 17 May 2006, 09:35

Well I look forward to seeing the new ones. Remember though you have been explaining how good present boxrec ratngs are, and that they are proven twice as good as IBO, in my test they were not better at all, marginally worse in fact, now you seem to doubt the quality of present Boxrec, no one will actually say IBO ratings are better than present Boxrec, I keep asking the question, your only reply was "it was proven more accurate".

To me the look of the ratings is very important, they have to subjectively look like the order of the best fighters, I am less concerned with prediction quality, if both can be achieved then that would be great.

JCS · Post by **JCS** » 17 May 2006, 09:37

Lennox wrote:Well I look forward to seeing the new ones. Remember though you have been explaining how good present boxrec ratngs are, and that they are proven twice as good as IBO, in my test they were not better at all, marginally worse in fact, now you seem to doubt the quality of present Boxrec, no one will actually say IBO ratings are better than present Boxrec, I keep asking the question, your only reply was "it was proven more accurate".

To me the look of the ratings is very important, they have to subjectively look like the order of the best fighters, I am less concerned with prediction quality, if both can be achieved then that would be great.

Well your opinion of what "looks good" differs from other people's... so you'll never be satisfied there until you're on the board of rating directors at BoxRec eh? ;)

Lennox · Post by **Lennox** » 17 May 2006, 09:54

Well I think IBO ratings are good and I looked at Fightnews after you said and they are good, I think the WBC ratings are by and large pretty good, they just have an occasional bad one rated. If new boxrec ratings are good I will say so, but as you say it is opinion. This argument/discussion seemed to start and then you were agreeing with me firstly, then you have switched sides. I get the impression that you, JohnShep and Computerrank all think IBO are better than Boxrec but will not say, you three are clear that prediction rate is the best way to produce a set of objective ratings and my recent test has shown that Boxrec does not currently have the edge over IBO predicting, so now you are going to make a new formula, if it makes better ratings the debate has been worth it.

JCS · Post by **JCS** » 17 May 2006, 09:56

Lennox wrote:Well I think IBO ratings are good and I looked at Fightnews after you said and they are good, I think the WBC ratings are by and large pretty good, they just have an occasional bad one rated. If new boxrec ratings are good I will say so, but as you say it is opinion. This argument/discussion seemed to start and then you were agreeing with me firstly, then you have switched sides. I get the impression that you, JohnShep and Computerrank all think IBO are better than Boxrec but will not say, you three are clear that prediction rate is the best way to produce a set of objective ratings and my recent test has shown that Boxrec does not currently have the edge over IBO predicting, so now you are going to make a new formula, if it makes better ratings the debate has been worth it.

I'm just saying.. w/ computerized ratings - prediction rate is the way to go. Yes, I do think the Boxrec system could be improved and thats what I argued from the start. Martin is getting ready to publish what he's been working on. BoxRec WILL have the edge over IBO in prediction but I bet that you will HATE how the BoxRec ratings will look.

IBO Ratings have their flaws and aren't updated that often.. so its hard to say which one is better.

computerrank · Post by **computerrank** » 17 May 2006, 17:36

Lennox wrote:Well I think IBO ratings are good and I looked at Fightnews after you said and they are good, I think the WBC ratings are by and large pretty good, they just have an occasional bad one rated. If new boxrec ratings are good I will say so, but as you say it is opinion. This argument/discussion seemed to start and then you were agreeing with me firstly, then you have switched sides. I get the impression that you, JohnShep and Computerrank all think IBO are better than Boxrec but will not say, you three are clear that prediction rate is the best way to produce a set of objective ratings and my recent test has shown that Boxrec does not currently have the edge over IBO predicting, so now you are going to make a new formula, if it makes better ratings the debate has been worth it.

Lennox,

seriously, you cannot evaluate any ratings by a short test. In order to get significant results, you have to go up to large number of cases. This is pure statistics. Otherwise all is incidents.

If you go by upsets as rare results, your standard deviation is the square root of the upset number. Given the upset number is 10, the standard deviation is the square root is 3.1. Accepted significance is given at a difference of 3 standard deviations, which is 9.3.

So you can see, with an upset number of 10, nothing is approved.

By the way, my test, I think 3 years ago, went to an upset number of about 100. And the result was, that Boxrec's Ratings (their release at that time), had a significant cut. Second (significantly, were my Ratings (my release at that time). And 3rd, significantly, were the IBO Ratings (lists at that time). Then clearly behind some ABC Rankings.

BoxRec ratings were improved since then. IBO ratings maybe also.

If only I could get the complete rules of the IBO ratings, I easily could evaluate both ratings against the complete BoxRec result database. But I nowhere can find the IBO rules.

Best regards
Martin

Lennox · Post by **Lennox** » 18 May 2006, 07:36

computerrank wrote:
Lennox wrote:Well I think IBO ratings are good and I looked at Fightnews after you said and they are good, I think the WBC ratings are by and large pretty good, they just have an occasional bad one rated. If new boxrec ratings are good I will say so, but as you say it is opinion. This argument/discussion seemed to start and then you were agreeing with me firstly, then you have switched sides. I get the impression that you, JohnShep and Computerrank all think IBO are better than Boxrec but will not say, you three are clear that prediction rate is the best way to produce a set of objective ratings and my recent test has shown that Boxrec does not currently have the edge over IBO predicting, so now you are going to make a new formula, if it makes better ratings the debate has been worth it.
Lennox,

seriously, you cannot evaluate any ratings by a short test. In order to get significant results, you have to go up to large number of cases. This is pure statistics. Otherwise all is incidents.

If you go by upsets as rare results, your standard deviation is the square root of the upset number. Given the upset number is 10, the standard deviation is the square root is 3.1. Accepted significance is given at a difference of 3 standard deviations, which is 9.3.

So you can see, with an upset number of 10, nothing is approved.

By the way, my test, I think 3 years ago, went to an upset number of about 100. And the result was, that Boxrec's Ratings (their release at that time), had a significant cut. Second (significantly, were my Ratings (my release at that time). And 3rd, significantly, were the IBO Ratings (lists at that time). Then clearly behind some ABC Rankings.

BoxRec ratings were improved since then. IBO ratings maybe also.

If only I could get the complete rules of the IBO ratings, I easily could evaluate both ratings against the complete BoxRec result database. But I nowhere can find the IBO rules.

Best regards
Martin

I agree all tests are better over longer time, at present the score is 24-20 in favour IBO, I found your orginal test it was 27-37 in Boxrec favour. I did expect boxrec to be better at predicting than IBO because it was how you designed the system. The IBO rules are in one of the old editions of IWBR, if you beat a fighter, you are ranked above him and when that fighter wins you move with that fighter, but as mentioned Kirk Johnson is still ranked above Javier Mora, so maybe they have changed the rule or the IBO rules are different to the old ones.

JCS · Post by **JCS** » 18 May 2006, 12:21

There's your new rating system... LIVE

Based purely on statistics from fights in the 2000s... how a computerized rating system should be.

Grey · Post by **Grey** » 18 May 2006, 12:32

Uhhh, Alexander Dimitrenko is #5 at heavyweight.

Yeah, because THAT MAKES SENSE.

Lennox · Post by **Lennox** » 18 May 2006, 12:42

JCS83MD wrote:There's your new rating system... LIVE

Based purely on statistics from fights in the 2000s... how a computerized rating system should be.

Crap ratings and exactly how not to prodice a computerized rating system.
http://www.iboboxing.com

JCS · Post by **JCS** » 18 May 2006, 12:42

Lennox wrote:
JCS83MD wrote:There's your new rating system... LIVE

Based purely on statistics from fights in the 2000s... how a computerized rating system should be.
Crap ratings and exactly how not to prodice a computerized rating system.
http://www.spam.com

computerrank · Post by **computerrank** » 18 May 2006, 12:43

Lennox wrote:
JCS83MD wrote:There's your new rating system... LIVE

Based purely on statistics from fights in the 2000s... how a computerized rating system should be.
Crap ratings and exactly how not to prodice a computerized rating system.
http://www.spam.com

Beat them ...

Lennox · Post by **Lennox** » 18 May 2006, 12:48

computerrank wrote:
Lennox wrote:
JCS83MD wrote:There's your new rating system... LIVE

Based purely on statistics from fights in the 2000s... how a computerized rating system should be.
Crap ratings and exactly how not to prodice a computerized rating system.
http://www.spam.com
Beat them ...

I think you have beaten yourselves.