[Phil]

Reading the description of how the power rankings are determined, it does not take into account the level of the opponents (like Lehman's do) nor the level of the tournament. It would appear sectionals (or even club games if they are reported) would have the same gravity as NABC's.

[qwery_hi]

Phil, on Aug 8 2010, 08:19 AM, said:

Reading the description of how the power rankings are determined, it does not take into account the level of the opponents (like Lehman's do) nor the level of the tournament. It would appear sectionals (or even club games if they are reported) would have the same gravity as NABC's.

Incorrect.

From the explanation - "Your game percentage (G%) for each and every game you play is divided into 3 parts. Your contribution/Power Rating (PR). Your partners contribution/Power Rating (Part) Your opponents contribution/ Power Rating (DOD). The sum of the opponents sitting your way, N/S if you are sitting N/S, is averaged and the amount above or below the average pair (50) is your degree of difficulty (DOD).
PR = G% + DOD - Part or PR + Part - DOD = G%"

Since the opponents PR at Nationals will be more, the DOD for Nationals will be more.
So a same percentage game at the NABC's will result in a higher PR.

[NickRW]

Yeah. As rating systems go, it is probably about as good as you can get.

If I've read between the lines correctly, and if the EBU ever actually gets around to doing it, would be very similar - except that it would be a rolling average of your last 1000 or so boards.

Nick

[Mbodell]

JanM, on Aug 7 2010, 08:46 PM, said:

What exactly is this list? There are a lot of good and not-so-good players on it, and there are some numbers that I don't understand. Apparently it has something to do with matchpoint performance? How do they choose whether to include someone on the list?

This is a quite good rating system for individual performance (and partnership performance, if you look at other links) in matchpointed games.

It includes all nationals, regionals, sectionals (not STAC), and a small number of clubs (get your clubs to submit their results to get a better rating for everyone).

To be rated as an individual player you have to play 12 rated games with other rated players, similarly partnerships can be rated playing at least 12 games (only matchpointed games count) over the past 2 years (only the past 2 years count). People without this many games are still rated (and you can find their ratings by looking at individual events or what not), but the author of the site feels the ratings have too little data and are thus suspect and hence doesn't list them in the top X players list.

The scale is such that the average player is a 25. Each point is 1 extra percentage. If a 26 player and a 27 player played together in an average field their expected score would be 53% (26+27). If the same players played together on the first day of the blue ribbons (field strength around +9, iirc) they'd be expected to score 44% (26+27-9). If the same players played together in one of my local clubs 299er games (field strength around -10) they'd be expected to score 63% (26+27+10). The field strength is just the average partnership strength. So if you had a field that was made up of nothing but people rated 26 the field strength would be 2 (26+26 = 52, 52-50 = 2 field strength of 2).

Meckstroth's rating of 35.96 in 101 games (now that New Orleans has been added) means that if he was playing with an average player in an average field he'd expect to score 60.96%.

You can also look at the top partnership ratings (the average partnership is 50, as two 25 rated people form a 50 partnership): pair ratings

Here Ramer-Kwiecien slightly edge Meckwell for the top matchpoint partnership with at least 12 rated games. The scores above 69 mean that they'd be expected to score that 69% in an average field. In a field like day 1 of the blue ribbons they'd be expected to score 60 (subtract the strength of field 9).

In this list there are two numbers the PairR, which is the rating they'd achieved as a pair, and the PR which is the sum of the two player's power rating (assuming each player has a power rating from playing enough games with enough different partners). You can see, for example, that Meckwell play slightly better together than they do with other partners. From their "raw" individual results they'd be expected to be "only" 68.74 but they are 69.17 instead.

Note also that Meckstroth is a member of the 2nd rated pair (Meckwell) and the 5th rated pair (Meckstroth and Perry Johnson).

In addition to how impressive Meckstroth is, note that the Diamond team has all 3 of its pairs in the top 28 rated matchpoint partnerships. Gittleman-Moss are 10th with 66.62, Diamond-Platnick are 19th with 65.96, and Greco-Hampson are 28th with 65.65. And this is only matchpoint results, so the pairs get no credit for any results in team games like, say, winning the spingold.

[awm]

The problem with this rating system is that it's designed in a way that enforces linearity on a lot of the relationships. There are pretty easy examples to show that this isn't true.

For example, Elianna and I have a rating of around 59. When we play in our local club game, our average score is in the low 60s. When we play on the first day of a national pairs, our average score is in the low 50s. The "degree of difficulty" should account for the differential.

All very well so far. Meckstroth and Rodwell have a rating of around 69. It's quite believable (though impressive) that their average score on the first day of national pairs is in the low 60s. However, I don't believe that their average score in our local club game would be in the low 70s. It's almost impossible to have an expected score which is this high against even a marginally competent field, simply because there will be some flat boards and some boards where you get "fixed" by the opponents doing something good that the field can't manage.

To give an even more extreme example, suppose for some reason Meckstroth and Rodwell played in a very weak game (like a 299er pairs). Would their expected score be almost 90%? I really doubt it; sheer randomness will cause them to score only in the 70s probably...

So my point is that the effect of strength-of-field is not actually linear. The net effect on the rating system is that people who consistently play in very difficult rated events only (and don't play regularly in rated club games for example) will tend to have higher ratings than people who at least occasionally play in weak fields.

Maybe another way to state this is the following. If Elianna and I play in the same LM pairs field as Meckwell, they will score better than us with extremely high probability. However, if we both play in the same 299er game (ignore the fact that this isn't allowed) we are both going to score ridiculously high. I think it becomes pretty random (based on gifts from opponents, what the field does on boards, etc) whether Elianna and I will score better or Meckwell do. The point is that when you have a pair which is much better than the field as a whole they always do well, but exactly how well is somewhat random and you can't compare two really good pairs based on their performance against really bad fields. Similarly, if you take two beginner pairs and throw them into the blue ribbons, they will both do really badly. But I don't think you can conclude that the pair that has a 35% game is "clearly better" than the pair that had a 30% game.

[JLOGIC]

Agree with awm, I don't think the pairs rating is a good predictor on what someone would actually score, this should be obvious since no one will average 69 % in an average filed heh. But if you use it as a ranking system and not a predictor it still has value.

The individual one seems better since there are larger samples on people though.

[rogerclee]

Two comments:

1) Actually I think the best way to inflate your PR is to have a good partnership with a weaker player. I guess you could extend this to just say "have a good partnership with anyone"; iif the partnership is more than the sum of its parts, then this is obviously good for the PR (and your bridge in general).

2) The linearity is quite a problem when you are comparing the finals of the blues (DOD around 14) and a 299er game (DOD around -10). However I think for comparing something like a sectional pair game to the first day of the life master pairs (difference of about 7 DOD), it's reasonably accurate, or at least about as accurate as you're going to get.

[Phil]

qwery_hi, on Aug 8 2010, 12:23 PM, said:

Phil, on Aug 8 2010, 08:19 AM, said:

Reading the description of how the power rankings are determined, it does not take into account the level of the opponents (like Lehman's do) nor the level of the tournament. It would appear sectionals (or even club games if they are reported) would have the same gravity as NABC's.

Incorrect.

From the explanation - "Your game percentage (G%) for each and every game you play is divided into 3 parts. Your contribution/Power Rating (PR). Your partners contribution/Power Rating (Part) Your opponents contribution/ Power Rating (DOD). The sum of the opponents sitting your way, N/S if you are sitting N/S, is averaged and the amount above or below the average pair (50) is your degree of difficulty (DOD).
PR = G% + DOD - Part or PR + Part - DOD = G%"

Since the opponents PR at Nationals will be more, the DOD for Nationals will be more.
So a same percentage game at the NABC's will result in a higher PR.

It also says that unrated players in clubs are rated initially at 'dead average'. We all know this isn't the case.

It seems a player that wanted to start with a high rating can play in a club without rated players and send the results in.

[JLOGIC]

rogerclee, on Aug 8 2010, 07:00 PM, said:

1) Actually I think the best way to inflate your PR is to have a good partnership with a weaker player. I guess you could extend this to just say "have a good partnership with anyone"; iif the partnership is more than the sum of its parts, then this is obviously good for the PR (and your bridge in general).

Says the guy with a very high avg DOD, and high avg %age, who's rating is only brought down by having very good partners on average

[Siegmund]

Re DOD, and it only partially accounting for the different strengths of fields...

Quote

All very well so far. Meckstroth and Rodwell have a rating of around 69. It's quite believable (though impressive) that their average score on the first day of national pairs is in the low 60s. However, I don't believe that their average score in our local club game would be in the low 70s.

Yup.

There are some odd artifacts in the ratings because of all (or almost all, anyway) tournament games, but very few club games, being included.

When I first looked myself up in power ratings, I was rated between 30 and 31, based on my performance in sectionals and regionals (which seems about right - averaging high 50s in stratified open pairs games, DOD ranging from about -1 to +2, with partners of similar or slightly lesser ability.)
When my club started submiting information on all of our games, I plummeted to between 28 and 29 -- because my local club has a DOD of about -5 in the summer and -6 or 7 in the winter... and averaging "only" in the low 60s.

DOD might work correctly for strong to very strong fields. For very weak fields, it seems to be quite a problem.

Having more clubs submit their data will help some. But there will still be a disparity between people who only play in tournaments vs only play in clubs vs both.

[Siegmund]

Quote

1) Actually I think the best way to inflate your PR is to have a good partnership with a weaker player. I guess you could extend this to just say "have a good partnership with anyone"; iif the partnership is more than the sum of its parts, then this is obviously good for the PR (and your bridge in general).

I think so too. Especially if you play in less-than-top-notch fields.

Among my partnerships, my strongest regular partner and I have a pair rating well below the sum of our power ratings even though we understand each other well; it's easier for us, in a club game, to place above average with a competent beginner as our partners, than to place above 60 with each other.

I am less sure that extends to top fields, where having one 'weak link' all of whose mistakes will be punished severely by strong opponents might drag your percentage down farther.

[barmar]

TimG, on Aug 8 2010, 09:05 AM, said:

So, those players who always (or virtually always) play with the same partner won't appear on the list.

If you always play with the same partner, how is any system supposed to distinguish your contribution to the partnership's success (or failure) from your partner's?

I suppose if your partner is more promiscuous, it might be possible to generate a rating for him, and then "subtract" it out to get your rating (if he does worse or better with you, it's presumably your fault). But this won't be as reliable as looking at all your results with different partners.

[TimG]

barmar, on Aug 8 2010, 09:18 PM, said:

TimG, on Aug 8 2010, 09:05 AM, said:

So, those players who always (or virtually always) play with the same partner won't appear on the list.

If you always play with the same partner, how is any system supposed to distinguish your contribution to the partnership's success (or failure) from your partner's?

It's not. I did not suggest it should. Just giving a reason some players might not be listed.