Talk Tennis

Talk Tennis (http://tt.tennis-warehouse.com/index.php)
-   Adult League & Tournament Talk (http://tt.tennis-warehouse.com/forumdisplay.php?f=35)
-   -   Question about USTA rating algorithm (http://tt.tennis-warehouse.com/showthread.php?t=465319)

storypeddler 06-01-2013 09:40 PM

Question about USTA rating algorithm
 
Let me see if I understand how this works, and if I'm wrong, would someone please correct me?

If my doubles partner and I play another team that, according to the computer, we are supposed to beat 6-3, 6-3. and we do, in fact, beat them 6-3, 6-3, then we should not expect our dynamic rating to go up from that result, nor their dynamic rating to go down? The result was the expected result, so the numbers all remain as they were going in?

Would it make any difference if you played a half dozen matches against different opponents, and in each case, won by the expected margin? No change in dynamic rating still?

andfor 06-01-2013 11:29 PM

Does anyone actually know the algorithm?

What happens when your opponents play new opponents next week?

Don't take this the wrong way, but what's the point trying to figure out the unknown?

dizzlmcwizzl 06-02-2013 02:36 AM

That is our understanding yes. If you keep winning (or losing) by the expected margin your rating will not change.

Furthermore, if you only play with the same partner exclusively then your relative dynamic ratings difference will remain the same after each match.

What I mean by this is that if you are a 3.60 and he is a 3.70 at the start of the season and you only play with each other then by the end of the season his dynamic rating will still be 0.1 higher than yours.... no matter what happens during the season.

Further - further more, the USTA ratings system is supposed to be a zero sum calculation .... meaning if you and your partner do better than expected and after a match each of your ratings went up 0.05 points then the ratings of each of your opponents would go down by the exact same margin.

schmke 06-02-2013 08:27 AM

The USTA does not disclose the exact algorithm, but does offer quite a bit of information on how it works in various documents and FAQs that you can find on the USTA web-site(s) or by Googling around.

And dizzy is generally correct about always playing with the same partner preserving the gap between the partners no matter what the scores are. But if one partner plays even one match by themselves or with a different partner, this isn't true as that would likely change their rating and one of the past 3 ratings that is part of the average is now different between the partners so the gap between them would change.

To andfor's question, what your opponents from this week do the following week has no impact on your rating unless you happen to play them again the future or they become a benchmark player which would mean year-end benchmark calculations will factor in for your match against them at the end of the year.

See my blog at http://computerratings.blogspot.com/search/label/tennis for some posts on this subject and a description of the Estimated Dynamic NTRP reports I can generate.

tennixpl 06-07-2013 10:57 AM

ranting but in college football when number 2 loses to number 1 team why does the number 2 team lose ranking, this confirms the algorithm, its only when you lose as number 2 to number 3 should you be downgraded to some mystic level.

its more amazing that the computer says you should beat them 6-3, 6-3. while it sounds dynamic i doubt it really it is it is probably just a fancy highly complex algorithm that is still static in nature, hey statisticians need jobs too.

its all a guess and roll with the system, or get a stats degree and offer your services for free to the UTSA :)

schmke 06-07-2013 11:49 AM

Quote:

Originally Posted by tennixpl (Post 7486608)
ranting but in college football when number 2 loses to number 1 team why does the number 2 team lose ranking, this confirms the algorithm, its only when you lose as number 2 to number 3 should you be downgraded to some mystic level.

This is an issue with pollsters than a properly written rating system. Some, but not all rating systems for football, mine included, look at the expected result and move a team according to how they do in relation to that, not necessarily just at win/loss. With such a system it is possible to lose a game and have one's rating increase, and corresponding one can also win a game and have the rating decrease.

Quote:

Originally Posted by tennixpl (Post 7486608)
its more amazing that the computer says you should beat them 6-3, 6-3. while it sounds dynamic i doubt it really it is it is probably just a fancy highly complex algorithm that is still static in nature, hey statisticians need jobs too.

its all a guess and roll with the system, or get a stats degree and offer your services for free to the UTSA :)

The NTRP rating system is indeed dynamic. Not sure what you are saying regarding it being complex but static in nature. The high level algorithm used is described in a variety of USTA documents so it isn't a real secret. They don't disclose many of the details of course so it isn't reasonably possible to exactly replicate their calculations, but I've come up with pretty good estimates. See the link in my other post for more info.

Orange 06-14-2013 10:01 AM

It is difficult to compare football ratings to tennis ratings, in part because a football team has so many more players than a doubles tennis team. If the starting quarterback is injured and misses a game, the game is played anyway. Someone who knows this can take it into account for rating purposes. If a tennis player has an injury that is bad enough for him to miss a game, that game is not played.

tennixpl 06-14-2013 10:54 AM

Orange- the point about football wasn't in respect to the way things are rated simply the concept of a number 2 losing to a number 1, shouldn't affect number 2 per se. they were supposed to lose and did so they might still be number 2 team, but in some rankings that means they drop several slots. schmke's point about the other non-numerical analysis that goes into those rankings make that happen, though in UTSA case i guess it doesn't .

as far as team sports go in baseball they do know how an absence will affect the team, at least they have quantified it statistically, NBA too. i don't know if football has caught up on those stats methods yet. a lot more moving parts in football.

schmke - i have looked at your page actually, it looks like a very intriguing approach. I'll look more when i get more time.
My point about dynamic is that if i have regularly been raising in the rating do they take that trend into affect in my next match, or simple a static comparison of past matches weighted a certain way, similarly with who i beat last time, is he hitting his predictions or below or above them, and if so does that get fed back into what "should" have been our score and change the dynamic where i am at any given week.

last week i played a guy and the score was supposed to be 7-6, 7-6 him. but i win 6-4, 6-4. now the next week his match he is supposed to win 6-2, 6-2 but he only wins 7-6, 6-4. bc hes injured. does the fact that a trend of this guy not doing as good as he "should" get fed back into my rating? that would be dynamic in my mind and i doubt that is the way it is done. Dynamic seems to mean the update it every match and go to a hundredth of a point. so one week i am a 3.15 and the next i am a 3.21. A truly dynamic system has a strong feedback loop

just a question- if the prediction was wrong about a particular match up one the first match does it take that into account if i play that exact same player again in match two? perhaps against a particular opponent we have that whole match up thing where he just can't cope with one thing i do while other players can?


i also hate secretive approaches bc it breeds contempt, i mean if people are going to game the system then they will game system, ppl can throw a match or ppl can throw games to make a rating what it is so, why does the UTSA hide it at all?

cneblett 06-14-2013 10:58 AM

I was in a captains meeting a couple of years ago and they provided a basic of show the calculation works.

If you play singles, your full Usta rating, ex 4.06 is compares to your opponents rating, ex 4.18. They consider every .06 of difference to equal s break of serve and compare actual result to this. So in the example, the result should be, 6-4,6-4. If you win 7-6,7-6, they take the result number of breaks against expected. In this case, the difference is .12. They will take this number and divide by 2, and adjust both individuals by this amount. So after match, your new rating is 4.12and your opponents is now 4.12.

In doubles they add your rating and partners rating together for your team and do same for opponents. They then do the same as for singles with expected score and result. Both partners will adjust by the same amount so the difference between you and your partner will be the same after match as before.

Now how the other items like benchmark, your actual full rating, anything with tourneys, mixed, etc in the mix.

Hope this helps.

schmke 06-14-2013 11:47 AM

Quote:

Originally Posted by cneblett (Post 7507205)
I was in a captains meeting a couple of years ago and they provided a basic of show the calculation works.

If you play singles, your full Usta rating, ex 4.06 is compares to your opponents rating, ex 4.18. They consider every .06 of difference to equal s break of serve and compare actual result to this. So in the example, the result should be, 6-4,6-4. If you win 7-6,7-6, they take the result number of breaks against expected. In this case, the difference is .12. They will take this number and divide by 2, and adjust both individuals by this amount. So after match, your new rating is 4.12and your opponents is now 4.12.

In doubles they add your rating and partners rating together for your team and do same for opponents. They then do the same as for singles with expected score and result. Both partners will adjust by the same amount so the difference between you and your partner will be the same after match as before.

Now how the other items like benchmark, your actual full rating, anything with tourneys, mixed, etc in the mix.

Hope this helps.

Yep, this is consistent with what you'll find in the various USTA documents and FAQs. However, the rating you are saying is your "new rating" is actually the rating for that match which is then averaged with prior dynamic ratings. Perhaps that is what you are referring to as your "actual full rating".

schmke 06-14-2013 11:55 AM

Quote:

Originally Posted by tennixpl (Post 7507195)
schmke - i have looked at your page actually, it looks like a very intriguing approach. I'll look more when i get more time.
My point about dynamic is that if i have regularly been raising in the rating do they take that trend into affect in my next match, or simple a static comparison of past matches weighted a certain way, similarly with who i beat last time, is he hitting his predictions or below or above them, and if so does that get fed back into what "should" have been our score and change the dynamic where i am at any given week.

last week i played a guy and the score was supposed to be 7-6, 7-6 him. but i win 6-4, 6-4. now the next week his match he is supposed to win 6-2, 6-2 but he only wins 7-6, 6-4. bc hes injured. does the fact that a trend of this guy not doing as good as he "should" get fed back into my rating? that would be dynamic in my mind and i doubt that is the way it is done. Dynamic seems to mean the update it every match and go to a hundredth of a point. so one week i am a 3.15 and the next i am a 3.21. A truly dynamic system has a strong feedback loop

Understand you now, and no, the dynamic rating isn't a "truly dynamic system" by your definition then. However, year-end ratings do have feedback loops at least for benchmark players for those that played them and these calculations probably do fit your definition. And not to drag us back to football but my football ratings do have a constant/iterative feedback loop.

Quote:

Originally Posted by tennixpl (Post 7507195)
just a question- if the prediction was wrong about a particular match up one the first match does it take that into account if i play that exact same player again in match two? perhaps against a particular opponent we have that whole match up thing where he just can't cope with one thing i do while other players can?

I don't believe there is any "match-up" component to the dynamic ratings. It simply generates a rating that reflects how you performed against your competition using the match scores the opponent(s) (and partners if applicable) rating at the time of the match.

cneblett 06-14-2013 12:02 PM

What we were told is that they keep your rating on a running full time. So there is not an averaging, they update it after each match played when the computer runs.

schmke 06-14-2013 12:16 PM

Quote:

Originally Posted by cneblett (Post 7507352)
What we were told is that they keep your rating on a running full time. So there is not an averaging, they update it after each match played when the computer runs.

And that update is done by doing calculations, one of which is performing an average.

bruce nissenbaum 06-14-2013 04:10 PM

As I learned from captain's meeting, DNTRP is essentially a simple arithmetic moving average of 4 consecutive matches: The latest calculated match rating is averaged with the previous 3. This lessens the impact of a single match and reduces volatility of the rating.

Nostradamus 06-14-2013 05:07 PM

I just wish you know where you are at all times. I don't think that will lead to cheating. like if you are at 4.8, I like to know that.

dizzlmcwizzl 06-14-2013 08:38 PM

Quote:

Originally Posted by Nostradamus (Post 7507888)
I just wish you know where you are at all times. I don't think that will lead to cheating. like if you are at 4.8, I like to know that.

I think most folks would like to know ... and only a few would use it for evil.

However, the USTA's ultimate goal is to collect as much in league fees as possible. I am not sure this change would drive players towards paying more league fees. Players would be ambivalent at best, manipulative at worst and players at the low end of the band might simply not get to play.

Nostradamus 06-14-2013 08:52 PM

Quote:

Originally Posted by dizzlmcwizzl (Post 7508166)
I think most folks would like to know ... and only a few would use it for evil.

However, the USTA's ultimate goal is to collect as much in league fees as possible. I am not sure this change would drive players towards paying more league fees. Players would be ambivalent at best, manipulative at worst and players at the low end of the band might simply not get to play.

What if USTA charged a fee every time you want to check your dynamic rating ? Lets just say, they charge $40 each time you want to check your rating during the season. That could be a money maker for USTA.:)

schmke 06-14-2013 09:14 PM

Quote:

Originally Posted by Nostradamus (Post 7508180)
What if USTA charged a fee every time you want to check your dynamic rating ? Lets just say, they charge $40 each time you want to check your rating during the season. That could be a money maker for USTA.:)

I'll do it for less :)

Bash and Crash 06-15-2013 08:37 PM

Quote:

Originally Posted by schmke (Post 7508198)
I'll do it for less :)

I know my team captains were happy with your results in sectionals. Glad to see my rating on your system is comfortably under 4.5 so no bump up yet.

OrangePower 06-15-2013 08:51 PM

Quote:

Originally Posted by Nostradamus (Post 7507888)
I just wish you know where you are at all times. I don't think that will lead to cheating. like if you are at 4.8, I like to know that.

I think USTA publishing a more precise rating would be a very bad idea.

It will lead to some cheating, by allowing people to more effectively "manage" their rating so that they don't get bumped up.

But that would not even be my main concern. More concerning is that people are going to view the rating as an absolute measure of ability, when in fact it is not nearly that precise at the micro level.

For example, imagine a captain puts Player A in the lineup for a crucial match instead of Player B, despite the fact that Player B has a 0.1 higher rating. The captain could very well think A is actually a better player, or a better matchup for the potential opponent. Now a good teammate would accept this, but as we know there are always squeaky wheels, and this would just give them more ammo.


All times are GMT -8. The time now is 01:50 AM.

Powered by vBulletin® Version 3.6.9
Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
© 2006 - Tennis Warehouse