TLS Website

Discussion in 'Adult League & Tournament Talk' started by dblsonly, Mar 21, 2013.

  1. TFC

    TFC New User

    Joined:
    Nov 5, 2013
    Messages:
    2
    Yes, only Montgomery County. And everything I said in my post applies only to the Montgomery Co. TLS ratings.


    Yes, I can see that. Is that analysis for the whole US? Is it meant to mimic USTA's NTRP system?


    I will, as soon as I post this.


    I didn't conclude that it's impossible to get bumped up by playing at level (keep in mind that my conclusion pertained only to TLS ratings, which may not be an accurate estimate of NTRP ratings). What I did say was that it appears that it is far easier to obtain an outlier high TLS rating by playing up a small number of matches than it is by playing at level a great number of matches.

    I noted that 14 out of the highest 15 rated players in the high normal range (3.31-3.38 ) had all played a lot of matches at level, had excellent W-L records, and were in my opinion among the most likely to get bumped up prior to seeing the TLS ratings. In contrast, the lowest 6 of the 9 players in the high outlier range (3.46-3.53) had all played very few matches, had played only at the 4.0 level, and their combined W-L record (W-L game record) was very poor (4-27), even considering that they were playing up.

    In summary, I believe that the computer expectation of, for example, a team of 3.75 players beating a team of 3.25 players by 6-0, 6-0 is unrealistic and biased towards helping the lower rated players. It is impossible for them to fall below the computer expectations in such a match. In contrast, there are many things that could cause them to exceed such low expectations, including having an especially good match, an injury on the other team, the other team having an off day, the other team being low on adrenaline due to the lack of competition, a few lucky bounces, or invalid ratings.
     
    #51
  2. schmke

    schmke Professional

    Joined:
    Jan 20, 2010
    Messages:
    1,369
    Yes, it includes players from across the whole U.S., although I've done analysis of more players in some sections than others. And yes, my Estimated Dynamic NTRP is meant to replicate the USTA's NTRP system and (IMHO) does it very accurately.

    And thanks for e-mailing me. I did take a look at the 0-2 player you mentioned and my numbers don't have his rating above 3.5 so wouldn't expect him to be bumped. However, there are a few observations.

    While he went 0-2 and only won 7 games, he was playing in an 8.0C league and doing so with another 3.5 as his partner. In both matches they played two 4.0s so they are expected to lose pretty badly and going 0-2, even winning just 7 games was pretty close to the expected result.

    But, having played only 2 matches, I don't believe his 2013 year-end rating will change from 3.5 as you need to play at least 3 matches to have a year-end rating calculated. So I wouldn't expect him to be bumped for this reason too.
     
    #52
  3. Maui19

    Maui19 Hall of Fame

    Joined:
    Sep 18, 2010
    Messages:
    1,659
    One thing that struck me looking at the TLS detailed stats is how hard it can be for a player to get bumped up without playing up. I looked at a bunch of players who I believe are at the top of their rating category (i.e. very good 3.5s) and it was surprisingly difficult for them to get a high DNTRP when playing at their level. When they played up, their DNTRP level was MUCH higher than when playing at level. The dampening effect of playing lower rated players is profound.
     
    #53
  4. schmke

    schmke Professional

    Joined:
    Jan 20, 2010
    Messages:
    1,369
    Yes, playing up gives more opportunity to improve your rating, and because of some of the factors discussed here doing so may result in a player being bumped up that shouldn't. The key thing is "leaked" games.

    Let's take a hypothetical player right at 4.0 or so and the theoretical 6-0,6-0 is expected between players exactly 1 level apart. Against an average 4.0 (rating 3.75) he should win 6-3,6-3 and against an average 4.5 (4.25) he should lose 6-3,6-3. Against the 4.0, if he "leaks" any games and it gets closer, his rating goes down. Against the 4.5 if his opponent "leaks" any games and it gets closer, his rating goes up.

    Something I haven't done but will put on my list is some analysis of players that play up and how different their match ratings are for their matches at the different levels. Look for something soon.

    --
    http://computerratings.blogspot.com/search/label/tennis
     
    #54
  5. Reddirt

    Reddirt New User

    Joined:
    Jul 29, 2013
    Messages:
    32
    Does anyone know why auto appeals are granted? What is the reasoning behind this? Doesn't the USTA "trust" their early start and end of year ratings? Doesn't this just fall into the hands of the sandbaggers who prefer to win than play at their actual level?
     
    #55
  6. schmke

    schmke Professional

    Joined:
    Jan 20, 2010
    Messages:
    1,369
    Any rating system is going to have some margin of error and I think the USTA is simply allowing for that and players that think they will be overwhelmed at the next level up to appeal down if they are within the 0.05 tolerance (0.1 for those 60 or older).

    Does this provide a loophole for sandbaggers? Probably, but there are some controls in place to avoid that, namely that benchmark players can't auto-appeal, and if you've played more than 5 matches you also can't appeal. This makes the auto-appeal available only to a subset of the players that haven't played that much and perhaps have a case that they got bumped up due to a fluke match or two.
     
    #56
  7. J_R_B

    J_R_B Hall of Fame

    Joined:
    Jan 27, 2010
    Messages:
    2,667
    Location:
    Newtown, PA
    The big one is that A rated players are subject to 3 strike DQ like S rated players, so if you really are supposed to be at the next level and you appeal down, theoretically, you'll be DQd anyway. They are definitely recognizing that a step function representing a continuous distribution in a population is going to be inaccurate at the step boundaries and allowing those people to appeal and play down if they are in fact right at the edge, but with the risk of DQ if they are doing it knowing that they should be playing up.
     
    #57
  8. mikeler

    mikeler G.O.A.T.

    Joined:
    Sep 26, 2008
    Messages:
    19,814
    Location:
    Central Florida
    The latest update for Florida is WAY too low for our team. Hopefully that turns into reality at year's end though. :)
     
    #58
  9. doubleshack

    doubleshack New User

    Joined:
    Dec 15, 2009
    Messages:
    84
    Looking at the ratings for our area, they seem to quite accurately reflect the results from league matches. There are people rated above me I can beat, and people below me that can beat me. However, these ratings are not a definition of your level of play, they are a representation of your results from league play. And viewing it as such, it strikes me as being very spot on.
     
    #59
  10. mikeler

    mikeler G.O.A.T.

    Joined:
    Sep 26, 2008
    Messages:
    19,814
    Location:
    Central Florida
    I just realized our area was still not updated since the summer even though it said Florida was updated in November.
     
    #60

Share This Page