View Single Post
Old 11-30-2012, 09:35 AM   #16
Join Date: Jul 2009
Posts: 822

Originally Posted by jmnk View Post
how is it flawed? What is so hard to understand that beating players that are no better than you will not make you move up?
If your dynamic rating is 3.89 (so you are 4.0 player) and you beat a player with dynamic rating of 3.52 (so he is also 4.0) by a score of 6:4 7:6 - how is that supposed to mean you should move up? By NTPR formula, which is based on old and proven ELO rating designed originally for chess (and now used very widely in competition where you can't possibly play --all-- of the opponents: tennis, soccer, on-line games, etc), the expected result is likely around 6:2, 6:2 - so you did worse than expected, which means your ranking went down despite the win.
Now have it happened 15 times over the course of the season and your dynamic ranking (and therefore NTPR ranking at the end of the season) will really go down and not up.

Now, if you want to argue that in tennis we should not care about games won/lost but only about whether someone actually won a match - you may have an argument. Such a system will have its own set of problems, mainly that you will have only two outcomes of any match (a win or a loss) vs. a set of outcomes (depending on games won/lost difference) and therefore there will be way less data for algorithm to crunch. Not that such a system is unheard of - Rogers tennis ranking in Canada is based purely on win/loss (and the 'weight' of the match, some matches are treated more importantly than others). FIFA ranking (which is actually very similar) uses a hybrid method - it cares about both pure win/loss, but also takes into account goal differential (which is equivalent to games differential in tennis).

All of those WTF-like cited cases are likely outliers anyway. The system overall is pretty accurate and serves the purpose. plus it is easy to say 'flawed system' - why don't you propose a better one? I'm sure it will have many edge cases where it will seem illogical (to an untrained eye of course).
Originally Posted by LuckyR View Post
Well, it flawed because, to use your (and the OP's) example the statistical chance that your example 3.89 player will meet only players of his or lower ranking 15 times in a row is less than 3 one thousanths of one percent.

In other words that reality exists only in strawman arguments like your post, not in the real world.
Let's see. Assuming you play in 4.0 league, so you are going to play only 4.0 NTPR players (in fact, due to the fact people are allowed to play up, you may in fact play someone below 4.0, but --never-- above 4.0). Again assuming that all 4.0 players are uniformly distributed - i.e there's equal chance of your opponent having 3.52 dynamic ranking as someone having 3.97 (in short, meeting a player of any dynamic rating in 3.50-3.99 range is equally likely). If you ranking is 3.89, you have 78% chance of meeting someone with a lower dynamic ranking. ((3.89-3.5)/0.5). The chance of that happening 15 times in a row is actually ~0.024 - which is a bit over 2%. Not sure where you got your "less than 3 one thousanths of one percent." from??
Note, that both of the above assumptions made the result --smaller-- that what would have happened in real life. Since you actually --can-- play someone with DR less than 3.5, and in any range there are likely more players in the lower/mid part of the range than in the upper range - so again in reality you are more likely to meet someone with a lower ranking.

Originally Posted by LuckyR View Post
You scoff in your second bolded statement at the idea of W/L record being meaningful.
I do not scoff at the idea of using W/L record as being meaningful. I'm sorry if it appears so. I'm saying that using games difference for casual USTA play is 'more meaningful' than using just W/L results.

Originally Posted by LuckyR View Post
My answer: if it works for Pros, why use a convoluted system for everyone else? If you lose in the first round of Wimbledon, it doesn't matter to your ATP points if it is to Fed or a Lucky Loser, you get the same credit.
Because top PROs play around 75-100 matches a year so there's way more data sample with just W/L ratio. Those that lose early have only 30-50 matches, the data sample is small, that's why it can be easily argued that PROs in 150 - 300 range are pretty much all the same level and they move so much with each result. Look at Janowicz - he had essentially one excellent week and few so-so results, and he moved from outside 100 to 26 spot. While I like him a lot - you really think he improved so much?
In addition in ATP tournaments, the points you get depend on the round you won. You know why - because underlying assumption is that the deeper you advance the better players you meet, so your win is worth more. Of course it does not work all the time - but on average, over the course of the year, it works fine.
But that will not work for USTA league - because players simple do not play that much. And there's no 'going deeper' in the tournament, for those that do not advance to post-season play (which is a vast majority) all the matches are worth the same.

Again - please propose a system and we can take records of few players from USTA and I'm sure we will be able to show you that your system is equally, if not more, illogical.
jmnk is offline   Reply With Quote