Ratings - please read before commenting - Archived

[Asterix]

The Home advantage looks like it's changed things massively. Valuev down to 8, Povetkin down to 14.

Does the home advantage only come into play if it's a points decision, or does it also affect knockouts?

[computerrank]

The Home advantage looks like it's changed things massively. Valuev down to 8, Povetkin down to 14.

Does the home advantage only come into play if it's a points decision, or does it also affect knockouts?

It also affects stopped bouts.

For opponents with pre-bout ratings of 1:1 the effective result ratio is

2.3:1 for stopped bouts and

3:1 for decisions on points

both in favour of the boxer with his promoter in charge.

[JCS]

The Home advantage looks like it's changed things massively. Valuev down to 8, Povetkin down to 14.

Does the home advantage only come into play if it's a points decision, or does it also affect knockouts?

It also affects stopped bouts.

For opponents with pre-bout ratings of 1:1 the effective result ratio is

2.3:1 for stopped bouts and

3:1 for decisions on points

both in favour of the boxer with his promoter in charge.

WOW. That is huge.. much more than I expected. In the past I thought we tested with 15-20% extra....

[computerrank]

The Home advantage looks like it's changed things massively. Valuev down to 8, Povetkin down to 14.

Does the home advantage only come into play if it's a points decision, or does it also affect knockouts?

It also affects stopped bouts.

For opponents with pre-bout ratings of 1:1 the effective result ratio is

2.3:1 for stopped bouts and

3:1 for decisions on points

both in favour of the boxer with his promoter in charge.

WOW. That is huge.. much more than I expected. In the past I thought we tested with 15-20% extra....

... we had 2:1.

[JCS]

Martin... The "Journal of Sports Science" results you are referring to.. is that from the article entitled, Do judges influence Hometown advantage?

If so, I am curious to know why the ratios used above are so vast.

According to the article, the probability of a home win for equally ranked competitors are:

0.57 - KO
0.66 - TKO
0.74 - PTS

Shouldn't the ratios be more like.. 7:4 for KO, 2:1 for TKO, 3:1 for points?

[computerrank]

Martin... The "Journal of Sports Science" results you are referring to.. is that from the article entitled, Do judges influence Hometown advantage?

If so, I am curious to know why the ratios used above are so vast.

According to the article, the probability of a home win for equally ranked competitors are:

0.57 - KO
0.66 - TKO
0.74 - PTS

Shouldn't the ratios be more like.. 7:4 for KO, 2:1 for TKO, 3:1 for points?

Well done ... I verified the findings with an extended worldwide base of bouts including opponents with both more than 50 pre-bout rating points (mature level) for 1970 and later.

... and found the values noted:

.75 (3:1) for decisions on points and .70 (2.3:1) for stopped bouts.

[JCS]

A question Martin.

Fighter A has 890 rating points.
Fighter B has 885 rating points.

The fight is on Fighter A's home turf.

Fighter A wins by a narrow split decision (115-113,111-117,115-113)

What would the resulting points be? Furthermore, what would the resulting points be if it were a draw?

[Grey]

Fabio Moli moved up into the top 25. The world weeps.

[JCS]

Fabio Moli moved up into the top 25. The world weeps.

Monte Barrett moved above Valuev. I weep.

[Asterix]

Maybe the new "home advantage" feature shouldn't be quite so influential. ;;-)

[computerrank]

Bug in the new ratings, caused by multiple promoters per show.

The bug causes erratic changes back and forth in the ratings of boxers involved in multiple promoter shows with given home advantage.

The fix will need until Friday this week.

[computerrank]

A question Martin.

Fighter A has 890 rating points.
Fighter B has 885 rating points.

The fight is on Fighter A's home turf.

Fighter A wins by a narrow split decision (115-113,111-117,115-113)

What would the resulting points be? Furthermore, what would the resulting points be if it were a draw?

current formula

r_A_new = r_A + 0.345*r_B/had_pts*cd + 0.345/(1+2*cd)(r_B/had_pts - r_A*had_pts)*cd
r_B_new = r_A - 0.345*r_B/had_pts*cd - 0.345/(1+2*cd)(r_B/had_pts - r_A*had_pts)*cd

a) SD

lim=0, cd= 1/(3*8)/3=0.01, pv=8/12=0.67 (8 rounds boxed), had_pts=1.41

r_A_new = r_A + 0.345*r_B/had_pts*cd*pv + 0.345/(1+2*cd)*(r_B/had_pts - r_A*had_pts)*pv

= 890 + 0.345*885/1.41*0.01*0.67 + 0.345/(1+0.02)*(885/1.41-890*1.41)*0.67

= 890 + 1 - 141 = 750

r_B_new = 1025


b) draw

lim=0, cd=0, pv=0.67, had_pts=1.41

r_A_new = 890 + 0.345*885/1.41*0*0.67 + 0.345/(1+2*0)*(885/1.41 -890*1.41)*0.67

= 890 + 0 - 145 = 745

r_B_new = 1030


c) UD by 2 score points difference per judge

lim=0, cd=2*3/8/1.5=0.5, pv=0.67, had_pts=1.41

r_A_new = 890 + 0.345*885/1.41*0.5*0.67 + 0.345/(1+2*0.5)*(885/1.41-890*1.41)*0.67

= 890 + 73 - 73 = 890

r_B_new = 885

[jujigatame]

What is going on with the rankings right now? They look messed up.

[jujigatame]

Are these new rankings meant to be permanent? I understand they may have been a reaction to Hopkins jumping over Calzaghe due to the Pavlik win, but there are really bad side effects in most divisions. Some examples:

HW: Monte Barrett at #7 over Valuev, Povetkin all the way down to #14.

WW: Isaac Hlatswayo at #4 over Clottey and Berto.

LW: Anthony Peterson at #8 over Casamayor, Katsidis, and David Diaz. Juan Diaz down to #5 below Romanov.

SFW: Soto at #10, 2 spots below Lorenzo, previous rankings handled this better IMO.

FW: Larios at #36 is way too low, even though I think he's totally washed up the results don't really support that low of a ranking. Piriyapinyo at #8 seems ridiculous given that he's beaten absolutely nobody worth mentioning. Solis seems pretty low at #16.

SBW: Caballero ahead of Marquez is questionable. How does Marquez have 1000 points less than Vasquez when their last fight was the closest decision humanly possible?

BW: Hasegawa should be a clear #1 in this division.

[JCS]

Juji.. the controversial change was the attempt to neutralize the home promoter advantage. In other words, the rankings you see now are supposed to reflect the quality of a fighter if they were to all fight on neutral ground w/ neutral judges.

I have to doubt the strength of the home factor and the integrity of the data its based on.

[jujigatame]

Dammit, wasn't this already attempted a few years ago, and it worked out horribly so it was dropped? Why go back to it? These rankings are ugly as sin compared to the previous ones, which really did not need fixing.

[computerrank]

Juji.. the controversial change was the attempt to neutralize the home promoter advantage. In other words, the rankings you see now are supposed to reflect the quality of a fighter if they were to all fight on neutral ground w/ neutral judges.

I have to doubt the strength of the home factor and the integrity of the data its based on.

... you might doubt everything, but not the integrity of the data from the Boxrec database.

- all bouts from 1970 and later (recent bouts)
- both opponents at minimum 50 points pre-bout rating (mature level)
- 1 opponent with home promoter advantage (at minimum 4th bout with promoter, opponent at maximum 2nd bout with promoter)
- the rest is simple statistics

decision on points
- rating ratio (home:non_home) - result ratio
1.0:1.5 - 2.0:1.0
1.0:1.0 - 3.0:1.0
1.5:1.0 - 5.0:1.0
2.0:1.0 - 7.0:1.0
3.0:1.0 - 10:1

stopped bouts
- rating ratio (home:non_home) - result ratio
1.0:1.5 - 1.5:1.0
1.0:1.0 - 2.3:1.0
1.5:1.0 - 5.0:1.0
2.0:1.0 - 7.0:1.0
3.0:1.0 - 10:1

[JCS]

Martin, I do not doubt the objectivity of the enhancement. However, it differs from the norm that thought the BoxRec ratings were re-configured to achieve.

How are we to know that the promoter info from say.. 1984 on some unknown card is valid? Furthermore, we can't simply assume that all promoter home advantage is equal. This could vary wildly.

[jujigatame]

It's an interesting idea, but quite obviously it does not produce sensible results. The rankings are clearly much, much worse than before. They were great before, and now they are filled with bizarre anomalies. This is a major step backwards.

[JCS]

A question Martin.

Fighter A has 890 rating points.
Fighter B has 885 rating points.

The fight is on Fighter A's home turf.

Fighter A wins by a narrow split decision (115-113,111-117,115-113)

What would the resulting points be? Furthermore, what would the resulting points be if it were a draw?

current formula

r_A_new = r_A + 0.345*r_B/had_pts*cd + 0.345/(1+2*cd)(r_B/had_pts - r_A*had_pts)*cd
r_B_new = r_A - 0.345*r_B/had_pts*cd - 0.345/(1+2*cd)(r_B/had_pts - r_A*had_pts)*cd

a) SD

lim=0, cd= 1/(3*8)/3=0.01, pv=8/12=0.67 (8 rounds boxed), had_pts=1.41

r_A_new = r_A + 0.345*r_B/had_pts*cd*pv + 0.345/(1+2*cd)*(r_B/had_pts - r_A*had_pts)*pv

= 890 + 0.345*885/1.41*0.01*0.67 + 0.345/(1+0.02)*(885/1.41-890*1.41)*0.67

= 890 + 1 - 141 = 750

r_B_new = 1025

So even though A won, B is now above A?

[computerrank]

Martin, I do not doubt the objectivity of the enhancement. However, it differs from the norm that thought the BoxRec ratings were re-configured to achieve.

How are we to know that the promoter info from say.. 1984 on some unknown card is valid? Furthermore, we can't simply assume that all promoter home advantage is equal. This could vary wildly.

The promoter information is only added to shows from official sources or at least is double-source-based.

So all shows (bouts) in doubt don't show this item and so are processed neutrally.

At last the effect is only implemented at 50% of its found value.

This also complies with the achievement perspective, as home advantage boxers' virtual achievements are a bit reduced and their opponents' achievements are a bit enhanced.

[computerrank]

A question Martin.

Fighter A has 890 rating points.
Fighter B has 885 rating points.

The fight is on Fighter A's home turf.

Fighter A wins by a narrow split decision (115-113,111-117,115-113)

What would the resulting points be? Furthermore, what would the resulting points be if it were a draw?

current formula

r_A_new = r_A + 0.345*r_B/had_pts*cd + 0.345/(1+2*cd)(r_B/had_pts - r_A*had_pts)*cd
r_B_new = r_A - 0.345*r_B/had_pts*cd - 0.345/(1+2*cd)(r_B/had_pts - r_A*had_pts)*cd

a) SD

lim=0, cd= 1/(3*8)/3=0.01, pv=8/12=0.67 (8 rounds boxed), had_pts=1.41

r_A_new = r_A + 0.345*r_B/had_pts*cd*pv + 0.345/(1+2*cd)*(r_B/had_pts - r_A*had_pts)*pv

= 890 + 0.345*885/1.41*0.01*0.67 + 0.345/(1+0.02)*(885/1.41-890*1.41)*0.67

= 890 + 1 - 141 = 750

r_B_new = 1025

So even though A won, B is now above A?

... yes, as it currently is - for non-clear decisions - cd<1

[jujigatame]

You haven't addressed the core issue here, which is that the rankings are now filled with ridiculous, unjustifiable anomalies, and have gotten much worse by any objective measure.

[JCS]

You haven't addressed the core issue here, which is that the rankings are now filled with ridiculous, unjustifiable anomalies, and have gotten much worse by any objective measure.

Objectivity (Prediction statistic) has probably increased... otherwise, I can't see any reason for the change.

Martin, the two other changes (heavyweight division weight and inactivity handling)... Those changes alone, would they impact the rankings much?

[jujigatame]

I thought the predictivity had been abandoned as a metric since it was shown to not lead to a sensible set of rankings when maximized?

As it stands right now, this ranking system is badly broken.