No bump up with 100% winning record?

[gmatheis]

Originally Posted by robert

This guy went 13-0 in singles this year. 7 of them are local playoff and Sectional. But no bump in the year end rating?

Also he bagled the opponents in Sectional matches. How is it possible?

link to tennislink of this happening ??

[jmnk]

Originally Posted by beernutz

I don't think NTRP is entirely based on ELO since ELO only considers win/loss/tie and doesn't use any type of game differential like number of moves whereas NTRP is just the opposite. I don't think most people here who have issues with the NTRP are arguing that it should not incorporate score and dynamic rating differentials into the formula but that it should also incorporate win/loss into it. Hybrid systems which use both exist so it is not beyond the realm of possibility to think that the USTA could also include it.

you are almost right. however the principle of the Elo system is not necessarily just win/loss/tie - it is the concept of 'expected result' and the notion of the weight of the match (sort of how important the match is). You can make this 'expected result' to be anything that makes sense for particular competition. in chess, as you correctly pointed out, it is win/loss/tie. In soccer (FIFA) it is win/loss/tie by a specific number of goals, and matches at World Cup have obviously more meaning than a friendly warm up match. In Canada Rogers rating for junior tennis it is only strictly win/loss, weighted depending on the rank of the tournament. In USTA apparently it was decided that it makes more sense to make an 'expected result' to be a particular game difference during the match.
I agree that including win/loss into formula would make sense. Although it can be easily argued that it already sort of is - in the majority of results, when you win more games, you win a match. I would not be surprised if USTA algorithm does take win/loss into account, in addition to pure game differential. But even if it doesn't, on average, using just game differential is good enough. Gaming the system is equally easy either way - so that discussion does not belong here.

I get the impression that most folks that bring up those edge cases where "a player such and such went 11:0 and he is not bumped. Another went just 5:6 and he got bumped" just need to realize that above all the level of competition is what really counts. If a player goes 11:0, with the average score of 6:1 6:1, but all of his opponents are at the very bottom of a given range, or perhaps some of them even playing up - that is not that significant of an achievement. It just does not show that a player is bump-up-worthy. And that is precisely what the ranking algorithm provides - a more objective picture of one's level, rather than looking at win/loss ratio only.

[jmnk]

Originally Posted by LuckyR

Wow, to quote Cool Hand Luke "what we've got here is failure to communicate". I apologize for for being extremely difficult to understand.

A couple of background things:

1- I don't personally believe that how well any player plays the game of tennis can be quantified to three significant digits. Sure the ratings exist, but IMO the science of using rating numbers to predict outcome of matchplay is too poor to justify that level of implied confidence. In my experience the standard deviation of quality of play is so wide that 2 significant digits is suspect, if you don't believe me, ask Lukas Rosol...

2- Regardless of the numerical system, there are players whose numerical rating does not correlate to their quality of play, on purpose, that is they give false information to manipulate the rating. That difference I call Sandbagging and can result in playing matches against better competition who have lower ratings. I was trying, unsuccessfully, to explain that nuance in my last post, first paragraph.

3- I agree that no system will "prevent" sandbagging. As long as people are willing to throw matches completely that will be possible. But this system creates sandbaggers. Using your scenario, the player who is 3.99, will likely always play someone lower rated than they are and thus will go 15-0, as you cite. The system is completely unequipped to deal with the 3.99 who improves their quality of play, yet never "plays up" to 4.5 (as a 4.0). Never gets to demonstrate their improvement against better competition and never gets moved up.

4- You are right that a straight W/L system will incorrectly bump people up who just randomly happened to play a bunch of players at the low end of the scale, you noted that a 3.89 (in the absence of sandbagging) would do so about 2% of the time in a 15 match season. OK you got me, you keep that guy at 4.0, I move him up (incorrectly BTW 2% of the time, your number) to 4.5 Well that 2% of the time in my system what happens? He gets killed playing 4.25s as a 3.89. Next season, he's a 4.0 Self correcting system. At least as a consolation, in a W/L system, no one wonders "why am I playing the 15-0 guy from last year?"

5- The unaddressed issue is the opposite scenario: what about the 4.11 (playing 4.5) who goes 0-15 for the exact same reason the 3.89 goes 15-0? Does that guy get a break? Nope.

6- How better to acknowledge that you do in fact know what a simple mathematical computation is more clearly than by stating: "As you know..."?

7- When you beat someone or get beat 6-0, 6-0 do you think: "I think we were probably mismatched" or "logical score between correctly rated players"? Of course both are possible (what the USTA says), but in your opinion which is more likely?

Ultimately it boils down to this: what is more likely to happen and what is more likely to kill the fun factor of league play: A) a system that thinks it can predict outcomes by rating and only moves players relatively if they behave unexpectedly and beat a "better" player, regardless of how many "lesser" players they smear the court with or B) a system that acknowledges that matchplay is unpredictable for numerous reasons (I agree cheating is one, but a rare one) and figures that the quality of opponents will even out in the end, so a 15-0 guy is likely better than his competition and moves him up?

Either is reasonable but I feel, obviously, that B is less likely to mislabel players, especially in the long haul, as it is self correcting

It appears that you think that a system where --only-- win/loss is considered; regardless of the score, and regardless of the relative strength of of players involved, is better than the system where at least level of competition is taken into consideration.

Well, while I respect your opinion, the facts, the past experience, and the data from competition results from many sports suggest otherwise. I'm not sure how else I can present my case - but perhaps you could read up on logic behind FIFA, chess, or Rogers ranking.

[Counter]

Here in Holland, we have a dynamic rating system that I tend to think is very good, and that combines the following three features:

(1) For any given match, the effects on players’ dynamic ratings are determined ONLY by which player wins the match; scoreline is irrelevant.

(2) However, the strength of the opposition in any given match IS taken into account. Basically, the extent to which your dynamic improves as a result of a match win depends upon the dynamic rating held by your opponent. Thus, to take an extreme example, if you manage to beat an opponent who has a far superior dynamic rating, then relatively speaking your dynamic rating improves a lot (and of course your opponent’s dynamic rating worsens a lot). And conversely, if you beat an opponent whose dynamic rating is far inferior to your own rating, then neither your own nor your opponent’s dynamic rating changes at all.

(3) The system is completely transparent and therefore predictable. That is to say, at any given moment you know your own dynamic rating, as well as your opponents’. Therefore, once you know who your next opponent is, you can easily calculate how winning – or losing – your next match will affect your dynamic rating.

Moreover, the results in this rating system actually make sense: it concurs with intuitions and reasonable expectations.

Let me give one numerical example.
Dynamic ratings basically vary between 1.000 and 9.000 (although slight overflows on both ends are possible), where 1.000 represents the BEST players and 9.000 the WORST players. And basically each player is given a ‘match result’ for each match he plays during the current season, and the dynamic rating of a player at any point during the season is simply the average over this year’s match results so far. (At the beginning of the year, each player gets six fictitious match results equal to his end year rating of the previous season; these fictitious match results are then erased one by one as they are replaced by real match results obtained in the current season).

Now let us assume player A has a dynamic rating of 5.2, whereas his opponent B has a dynamic rating of 5.8 (i.e., A is supposed to be the somewhat stronger player). The ‘match result’ for each player is then calculated as follows:

• The winner gets his opponent’s dynamic rating MINUS one.
• The loser gets his opponent’s dynamic rating PLUS one.

So in our example, if player A wins he gets a match result of 4.8, and the losing player B get a match result of 6.2. If B wins, however, then A’s match result is 6.8 and B’s match result is 4.2.
Note that in this example, as always (except for the caveat noted below), the winner’s dynamic rating improves as a result of winning, and the loser’s rating worsens. But also note that in the case of a “surprise result”, i.e. the player B with the inferior rating manages to win, the effects on both players’ ratings are greater as compared to the “expected result”.

This always works, except if the difference between the two players’ ratings is greater than 1, AND the better-rated player wins the match (as expected). In that scenario we cannot calculate the match results in the usual way, as it would imply that the winner’s rating would worsen and the loser’s rating would improve. But in that case the solution is simple: neither player’s rating changes! In other words, in such a case, neither player is assigned a ‘match result’ for that match.

[dizzlmcwizzl]

I really believe that if we all knew what exactly the formula was that the USTA uses and had the level of transparency you describe many more players would be happy with it. Even though I suspect that the USTA system is much more convoluted.

Unfortunately, I have come to realize that USTA has created a National championship structure that rewards players who manipulate the system, so this level of transparency is problematic.

At this point I wish that all USTA team tennis would become local leagues and that the ratings system became merely a way to get like ability players together on the courts rather than an attempt to normalize players across the states.

[NLBwell]

If the section is bumping a ton of people down (which it appears some may be doing) the undefeated player may not be the best around next year.

[Winning Ugly]

Originally Posted by robert

This guy went 13-0 in singles this year. 7 of them are local playoff and Sectional. But no bump in the year end rating?

Also he bagled the opponents in Sectional matches. How is it possible?

Look -- if he is bageling people in SECTIONALS, then he should be good enough to move up. Period. That sort of beat down could earn you a visual DQ.

But, here's the thing. I have it on good authority that the computer rejects 6-0, 6-0 results! (Look it up, someone who knows). Anyway, this guy's bagel win may not even have counted -- did he have any others?. And, as others have said, if his district is a bit weak and he won against a lot of lower-in-level folks, then his dynamic rating could have fallen short.

However, if this guy goes undefeated again in 2013 and does not get bumped I'd eat a brick. Just tell him if he's up 6-0, 5-0 to drop a game or two. Seriously -- I was told this was why I did not get bumped a year ago...a double bagel win against a decent opponent never counted.

[schmke]

Originally Posted by Winning Ugly

Look -- if he is bageling people in SECTIONALS, then he should be good enough to move up. Period. That sort of beat down could earn you a visual DQ.

But, here's the thing. I have it on good authority that the computer rejects 6-0, 6-0 results! (Look it up, someone who knows). Anyway, this guy's bagel win may not even have counted -- did he have any others?. And, as others have said, if his district is a bit weak and he won against a lot of lower-in-level folks, then his dynamic rating could have fallen short.

However, if this guy goes undefeated again in 2013 and does not get bumped I'd eat a brick. Just tell him if he's up 6-0, 5-0 to drop a game or two. Seriously -- I was told this was why I did not get bumped a year ago...a double bagel win against a decent opponent never counted.

If the USTA is still dropping double-bagel matches, that, IMHO, is a big flaw in the system. I think the reasoning for dropping them is they don't tell you specifically how much better player A is than player B, only that player A is "at least" X better. And using X as the value could either hurt the winning player or help the losing player when it shouldn't.

However, if the expected result between the players was closer than 6-0,6-0, the above bad situation won't occur and it seems more correct to at least adjust the players apart (winning player rating go up, losing player rating go down) based on the result rather than completely ignoring it.

[beernutz]

Originally Posted by schmke

If the USTA is still dropping double-bagel matches, that, IMHO, is a big flaw in the system. I think the reasoning for dropping them is they don't tell you specifically how much better player A is than player B, only that player A is "at least" X better. And using X as the value could either hurt the winning player or help the losing player when it shouldn't.

However, if the expected result between the players was closer than 6-0,6-0, the above bad situation won't occur and it seems more correct to at least adjust the players apart (winning player rating go up, losing player rating go down) based on the result rather than completely ignoring it.

I agree. The rating system has to be able to account for 6-0, 6-1 and 6-1, 6-1 wins so rather than just ignoring 6-0, 6-0 wins just treat them as a 6-0, 6-1 win plus the difference between a 6-0, 6-1 and a 6-1, 6-1 win. That may not be exactly right in all cases but neither is any other part of this system. The argument that the USTA uses that not all 6-0, 6-0 wins are created equal is imo a stupid reason for ignoring them entirely (if that is in fact what is really happening).

[gmatheis]

Originally Posted by Winning Ugly

Look -- if he is bageling people in SECTIONALS, then he should be good enough to move up. Period. That sort of beat down could earn you a visual DQ.

But, here's the thing. I have it on good authority that the computer rejects 6-0, 6-0 results! (Look it up, someone who knows). Anyway, this guy's bagel win may not even have counted -- did he have any others?. And, as others have said, if his district is a bit weak and he won against a lot of lower-in-level folks, then his dynamic rating could have fallen short.

However, if this guy goes undefeated again in 2013 and does not get bumped I'd eat a brick. Just tell him if he's up 6-0, 5-0 to drop a game or two. Seriously -- I was told this was why I did not get bumped a year ago...a double bagel win against a decent opponent never counted.

Yeah, and I have it on good authority that the moon is made of green cheese!

I have heard exaactly the opposite, that 6-0 does count towards your DNTRP.

Not to mention we still have no link to the tennislink of this supposed undefeated person.

All we have is you telling us about it.

[dizzlmcwizzl]

I will say that you used to be able to find on almost every section website concerning ratings that double bagel match scores were thrown out of the dynamic rating. Almost like it was part of a formal USTA statement.

However, now the references are increasing harder to find. There is at least one section still with a reference to double bagel scores but it may not be an updated site.

We have had a couple folks claiming inside knowledge on here saying that they used to not count and now they do. It is hard to know what to believe, but if I had to wager I would put my money on the belief that the USTA now incorporates double bagels into our dynamic ratings.

[Winning Ugly]

Originally Posted by gmatheis

Yeah, and I have it on good authority that the moon is made of green cheese!

I have heard exaactly the opposite, that 6-0 does count towards your DNTRP.

Not to mention we still have no link to the tennislink of this supposed undefeated person.

All we have is you telling us about it.

Here's some evidence of your green cheese, gmatheis: http://tt.tennis-warehouse.com/showthread.php?t=432853. And, there's more if you search. Further, I was not the OP with the undefeated guy story, which I also felt was somewhat hard to believe given his performance at sectionals (hard to find a poor player there, although I did play one this year in a sectional doubles match.)

But, looks like they are finally fixing it. It happened to me so I know. Granted, it must be a DOUBLE BAGEL. 6-0 sets have never been dropped.

[3fees]

Flawed system.

[gameboy]

Folks, this is pretty standard math we are talking about here. I understand many of you don't use that much math in your everyday life, but ignorance is not something to be proud of.

Originally Posted by LuckyR

1- I don't personally believe that how well any player plays the game of tennis can be quantified to three significant digits. Sure the ratings exist, but IMO the science of using rating numbers to predict outcome of matchplay is too poor to justify that level of implied confidence. In my experience the standard deviation of quality of play is so wide that 2 significant digits is suspect, if you don't believe me, ask Lukas Rosol...

Good, you don't have to believe anything. These are all based on facts, stats and numbers. There is nothing to believe.

You say that "science" is too poor. Can you give me some scientific background for that? This is pretty much the same formulas used for every other sports (including Vegas odds) and what Nate Silver used to predict 50 out of 50 state election results. Everybody who knows anything about statistics use it, so why is the "science" poor? I am sorry but "because I said so" is not an acceptable answer.

Originally Posted by LuckyR

2- Regardless of the numerical system, there are players whose numerical rating does not correlate to their quality of play, on purpose, that is they give false information to manipulate the rating. That difference I call Sandbagging and can result in playing matches against better competition who have lower ratings. I was trying, unsuccessfully, to explain that nuance in my last post, first paragraph.

Sandbagging happens with any stratified system. It is not unique to this system.

Originally Posted by LuckyR

3- I agree that no system will "prevent" sandbagging. As long as people are willing to throw matches completely that will be possible. But this system creates sandbaggers. Using your scenario, the player who is 3.99, will likely always play someone lower rated than they are and thus will go 15-0, as you cite. The system is completely unequipped to deal with the 3.99 who improves their quality of play, yet never "plays up" to 4.5 (as a 4.0). Never gets to demonstrate their improvement against better competition and never gets moved up.

You lost me here.

It is clear that you do not understand how this system works. A 3.99 player can ALWAYS move up based on the current system. All you have to do is do better than expected. If a 3.99 plays 3.75 (average player), the 3.99 is expected to win a competitive match (say, 6-3, 6-4). However, if the match is not competitive (say 6-1, 6-1), the 3.99 player will get bumped up.

You really should understand how the system works before criticizing it.

Originally Posted by LuckyR

4- You are right that a straight W/L system will incorrectly bump people up who just randomly happened to play a bunch of players at the low end of the scale, you noted that a 3.89 (in the absence of sandbagging) would do so about 2% of the time in a 15 match season. OK you got me, you keep that guy at 4.0, I move him up (incorrectly BTW 2% of the time, your number) to 4.5 Well that 2% of the time in my system what happens? He gets killed playing 4.25s as a 3.89. Next season, he's a 4.0 Self correcting system. At least as a consolation, in a W/L system, no one wonders "why am I playing the 15-0 guy from last year?"

The current system is also self correcting - if you get bumped up and you are not competitive, you will get bumped down. But this system is better since people at the very top/bottom of the level won't get yo-yo'ed up and down every year. Again, learn the system and you won't wonder "why am I playing the 15-0 guy from last year."

Originally Posted by LuckyR

5- The unaddressed issue is the opposite scenario: what about the 4.11 (playing 4.5) who goes 0-15 for the exact same reason the 3.89 goes 15-0? Does that guy get a break? Nope.

It is addressed. If you do worse than expected, you will get bumped down. If you are competitive in most of your matches, there is no need to bump down, just because you are unlucky and lose all of your matches.

Originally Posted by LuckyR

6- How better to acknowledge that you do in fact know what a simple mathematical computation is more clearly than by stating: "As you know..."?

????

Originally Posted by LuckyR

7- When you beat someone or get beat 6-0, 6-0 do you think: "I think we were probably mismatched" or "logical score between correctly rated players"? Of course both are possible (what the USTA says), but in your opinion which is more likely?

It would be the exact same chance as someone playing an opponent full level or more above/below. Not sure what your point is.

Originally Posted by LuckyR

Ultimately it boils down to this: what is more likely to happen and what is more likely to kill the fun factor of league play: A) a system that thinks it can predict outcomes by rating and only moves players relatively if they behave unexpectedly and beat a "better" player, regardless of how many "lesser" players they smear the court with or B) a system that acknowledges that matchplay is unpredictable for numerous reasons (I agree cheating is one, but a rare one) and figures that the quality of opponents will even out in the end, so a 15-0 guy is likely better than his competition and moves him up?

Either is reasonable but I feel, obviously, that B is less likely to mislabel players, especially in the long haul, as it is self correcting

Both systems have been tried. The current one was adopted because there were more problems with the old one.

[ronray43]

Quote:
Originally Posted by LuckyR
4- You are right that a straight W/L system will incorrectly bump people up who just randomly happened to play a bunch of players at the low end of the scale, you noted that a 3.89 (in the absence of sandbagging) would do so about 2% of the time in a 15 match season. OK you got me, you keep that guy at 4.0, I move him up (incorrectly BTW 2% of the time, your number) to 4.5 Well that 2% of the time in my system what happens? He gets killed playing 4.25s as a 3.89. Next season, he's a 4.0 Self correcting system. At least as a consolation, in a W/L system, no one wonders "why am I playing the 15-0 guy from last year?"

Origianlly posted by gameboy
The current system is also self correcting - if you get bumped up and you are not competitive, you will get bumped down. But this system is better since people at the very top/bottom of the level won't get yo-yo'ed up and down every year. Again, learn the system and you won't wonder "why am I playing the 15-0 guy from last year."

Actually, those of us who are truly between ratings, but stuck at the bottom of the upper rating (what I call 3.75s, i.e., good 3.5s who are bad 4.0s) would love to get yo-yo'd every year . . . one year of getting killed, followed by a year of winning . . . followed by a year of getting killed, would be a whole lot better than three years of getting killed . . . . with a fourth on the way . . .

[LuckyR]

Originally Posted by ronray43

Actually, those of us who are truly between ratings, but stuck at the bottom of the upper rating (what I call 3.75s, i.e., good 3.5s who are bad 4.0s) would love to get yo-yo'd every year . . . one year of getting killed, followed by a year of winning . . . followed by a year of getting killed, would be a whole lot better than three years of getting killed . . . . with a fourth on the way . . .

Sorry Charlie...

[JLyon]

Originally Posted by ronray43

Quote:
Actually, those of us who are truly between ratings, but stuck at the bottom of the upper rating (what I call 3.75s, i.e., good 3.5s who are bad 4.0s) would love to get yo-yo'd every year . . . one year of getting killed, followed by a year of winning . . . followed by a year of getting killed, would be a whole lot better than three years of getting killed . . . . with a fourth on the way . . .

I feel the same way. My overall record just screams 5.0 TennisPlayer!!
http://tennislink.usta.com/Leagues/M...VxeQ%3d%3d&e=1

[goober]

Originally Posted by JLyon

I feel the same way. My overall record just screams 5.0 TennisPlayer!!
http://tennislink.usta.com/Leagues/M...VxeQ%3d%3d&e=1

Have you tried appealing down your rating? I bet if you did an autoappeal it would go through because you have to be pretty close to the 4.5 cutoff.

[JLyon]

Originally Posted by goober

Have you tried appealing down your rating? I bet if you did an autoappeal it would go through because you have to be pretty close to the 4.5 cutoff.

No dice already tried auto-appeal and lost again. Even sent a nice little rant to head of Southern Leagues. Think now with addition of 5.5's to 5.0 leagues I am stuck at 5.0 until I hit forty in 5 years

[schmke]

Originally Posted by JLyon

No dice already tried auto-appeal and lost again. Even sent a nice little rant to head of Southern Leagues. Think now with addition of 5.5's to 5.0 leagues I am stuck at 5.0 until I hit forty in 5 years

Your challenge is that you are in a league that only plays a handful of matches and there may be a number of self-rated players you play with or against each year. I didn't check each and every one, but if you play only 4-5 matches and 2 or 3 are with/against self-rated players, you may only have a match or two that counts and that either isn't enough to move you or is too few to calculate a new dynamic rating so your previous year rating just carries over.