Mixed Doubles

[travlerajm]

USTA rarely counts any mixed results at all. UTR made many inexplicably bad decisions, but it is still the best the public has for predicting results in tennis - including mixed.

I don’t think you can say it’s good at predicting mixed results when it excludes the majority of matches for some players for an illogical reason, and the exclusion invalidates the rating based on the rest of the matches due to the biased sample.

 

[TennisOTM]

UTR is in a good position to do some research into doubles ratings calculations for scenarios where partners have a wider rating gap. It seems clear that the standard way of taking a straight average to represent the team strength does not work well in those cases. UTR must have realized that it doesn't work well, but their solution to just ignore some matches, with a hard cutoff, appears to cause other problems. Surely they have enough data to help figure out a method that works better.

 

[travlerajm]

UTR is in a good position to do some research into doubles ratings calculations for scenarios where partners have a wider rating gap. It seems clear that the standard way of taking a straight average to represent the team strength does not work well in those cases. UTR must have realized that it doesn't work well, but their solution to just ignore some matches, with a hard cutoff, appears to cause other problems. Surely they have enough data to help figure out a method that works better.

Absolutely.

The case of a UTR 8 + 3 vs. UTR 6 + 5 combo is very common in 8.0 mixed, yet the algo excludes these.

As a high 4.5M in mixed who usually wins in 8.0, I was finding that roughly 75% of my mixed matches were getting excluded by the UTR algorithm, which is silly.

It’s probably true that an 8/3 combo will beat a 6/5 combo 70% of the time, but that should be easy to build the adjusted expectation into the algo.

 

[FuzzyYellowBalls]

As a high 4.5M in mixed

Impressive

 

[travlerajm]

Impressive

At least TR acknowledges mixed results.

 

[FuzzyYellowBalls]

At least TR acknowledges mixed results.

Naturally you love TR , sticky ratings.

 

[travlerajm]

Naturally you love TR , sticky ratings.

My current rating puzzle is this:

I played my first usta singles match in a long time, about 6 years. Not a performance I’m that proud of — squeaking by against a 4.0 Amazon employee from Delhi — but that’s beside the point.

Point is that I beat him and won more games than him, but my match rating on WTN is worse than my opponent’s established rating. This doesn’t give me much confidence in the WTN algo.

 

[TennisOTM]

My current rating puzzle is this:

I played my first usta singles match in a long time, about 6 years. Not a performance I’m that proud of — squeaking by against a 4.0 Amazon employee from Delhi — but that’s beside the point.

Point is that I beat him and won more games than him, but my match rating on WTN is worse than my opponent’s established rating. This doesn’t give me much confidence in the WTN algo.

It's either a) averaging your recent performance with the ones from 6 years ago, if they're on there, or b) averaging your performance with some made-up guess of what your initial rating was.

I've seen b) happen and it's absolute madness. A player's rating will get anchored to the inital value that was based on nothing, so that they have 3-4 wins against players who are all still rated better than them.

 

[travlerajm]

It's either a) averaging your recent performance with the ones from 6 years ago, if they're on there, or b) averaging your performance with some made-up guess of what your initial rating was.

I've seen b) happen and it's absolute madness. A player's rating will get anchored to the inital value that was based on nothing, so that they have 3-4 wins against players who are all still rated better than them.

Upon closer inspection, it’s because when I clicked on the match, WTN shows only the pre-match ratings.

It had assigned me with an very low pre-match rating (presumably by starting with my 4.5 high ntrp singles rating from 6 years ago, then extrapolating an age curve from my mid 40s into my my early 50s, and assuming age-associated rating decay).

My pre-match WTN rating was lower than my 4.0 ntrp opponent’s established WTN rating, but then my post-match rating leapfrogged his. So it’s at least logical.

 

[ptrbwmf2026]

Am I wrong for thinking that the loss my partner and I had in mixed last Sunday was a fluke?

My partner held serve every time (5 for 5), and I held 3 out of 4 times. And we won the 2nd set 6-0.

I would imagine that’s a 95% win rate, but I could be wrong.

 

[TennisOTM]

Am I wrong for thinking that the loss my partner and I had in mixed last Sunday was a fluke?

My partner held serve every time (5 for 5), and I held 3 out of 4 times. And we won the 2nd set 6-0.

I would imagine that’s a 95% win rate, but I could be wrong.

If you won a set 6-0 then you likely won more games than your opponent but still lost the match. You'd be maybe statistically favored to beat that team, so yeah it's a bit flukey but it happens. Also possible, but less likely, is that your opponents punted the second set on purpose.

I've always found it hard to play the match tiebreak after winning the second set easily. After cruising through a whole set with minimal stress, suddenly I'm in thrust into the high-stress points of the tie-break. And my opponents likely start playing with much higher intensity to redeem themselves after the embarrassment of the second set thrashing. It can be a tough mental situation to battle through.

 

[Moon Shooter]

I don’t think you can say it’s good at predicting mixed results when it excludes the majority of matches for some players for an illogical reason, and the exclusion invalidates the rating based on the rest of the matches due to the biased sample.

It is unclear what sort of bias it would create. I agree UTR excludes too many games. (mainly due to the 12 month cut off) But when you consider TR and USTA they entirely exclude all mixed doubles from their main rating - unless the person only plays mixed doubles. And USTA and TR excludes all mixed doubles when the league ends in a .5 even from their mixed doubles rating. Talk about throwing out good data that would improve their rating!

At least TR acknowledges mixed results.

Not for me or anyone that has played a few same gender rated matches they don't. It is a completely separate rating side rating. In fact I think they ape USTA and don't even count mixed games if the league ends in .5. If the league ends in .5 then it is yet another separate rating. UTR is much more sensible.

 

[travlerajm]

It is unclear what sort of bias it would create.

Due to the rule that partners must be within 4.0 UTR points of each other for a match to count toward the rating, the consequence is that UTR counts all the 8.0 mixed matches that I lose, but excludes all the 8.0 mixed matches I win (on most days).

If my partner is a UTR 3.5, and I perform at a UTR 8 level, it doesn’t count because then we’re 4.5 apart. If I perform at a 7 level, it counts.

Only counting the losses is as biased as it gets.

And since it can’t create accurate mixed rating for many players, the rest of the ratings in the system are also compromised.

 

[Moon Shooter]

Due to the rule that partners must be within 4.0 UTR points of each other for a match to count toward the rating, the consequence is that UTR counts all the 8.0 mixed matches that I lose, but excludes all the 8.0 mixed matches I win (on most days).

I'm pretty sure the 4.0 point difference rule excludes all matches no matter who wins.

If my partner is a UTR 3.5, and I perform at a UTR 8 level, it doesn’t count because then we’re 4.5 apart. If I perform at a 7 level, it counts.

Only counting the losses is as biased as it gets.

And since it can’t create accurate mixed rating for many players, the rest of the ratings in the system are also compromised.

I don't think it is based on the match performance but rather your and her overall rating.

 

[travlerajm]

I'm pretty sure the 4.0 point difference rule excludes all matches no matter who wins.




I don't think it is based on the match performance but rather your and her overall rating.

You’re missing the point.

It excludes the match. So then my rating drops down. If my rating drops low enough, then the match counts!

So UTR does iterations over and over, with my rating gyrating up and down because it doesn’t know which matches to count and which yo exclude.

But the reliability % goes up if more matches are counted, and the lower my rating, the more matches are counted. So eventually the algorithm finds more confidence in my rating if it counts my poor performances and excludes my good ones.

If I win with a bad partner, it doesn’t count. But if I win with a good partner, it does!

In other words, it excludes only the best performances, and keeps the worst. This is the definition of bias.

 

[Moon Shooter]

I don’t think you can say it’s good at predicting mixed results when it excludes the majority of matches for some players for an illogical reason, and the exclusion invalidates the rating based on the rest of the matches due to the biased sample.

I said it is the best the public has for predicting mixed results. I say that based on my own (albeit anecdotal) experience and things like Tennis OTMs test results. I did not say it was good all of these rating systems make inexplicably bad decisions.

Look at your own results and add up what TR predicts and what UTR predicts will win. It is not a mystery.

 

[travlerajm]

Look at your own results and add up what TR predicts and what UTR predicts will win. It is not a mystery.

If I take my 2022 mixed season, TR correctly predicted that I should go 7-0 undefeated.

UTR predicted that I should go 0-7.

 

[Moon Shooter]

You’re missing the point.

It excludes the match. So then my rating drops down. If my rating drops low enough, then the match counts!

So UTR does iterations over and over, with my rating gyrating up and down because it doesn’t know which matches to count and which yo exclude.

But the reliability % goes up if more matches are counted, and the lower my rating, the more matches are counted. So eventually the algorithm finds more confidence in my rating if it counts my poor performances and excludes my good ones.

If I win with a bad partner, it doesn’t count. But if I win with a good partner, it does!

In other words, it excludes only the best performances, and keeps the worst. This is the definition of bias.

The matches count if your rating is closer to your partners. If her rating goes up relative to yours then more matches count whether you win or lose them. But they may help or hurt your rating regardless of whether you won or lost. If you win or lose with a partner that is over 4.0 UTR points below your rating it doesn't count. Even if you win and even if you win more games then your opponents counting the match might hurt your rating. And even if you lose the match and lose more games then your opponent counting the match may improve your rating.

I think UTR has bigger problems then this.

 

[Moon Shooter]

If I take my 2022 mixed season, TR correctly predicted that I should go 7-0 undefeated.

UTR predicted that I should go 0-7.

So you are looking at your UTR now today as well as your partner and opponents utrs now today and using them to predict what should have happened 2 years ago?

 

[travlerajm]

So you are looking at your UTR now today as well as your partner and opponents utrs now today and using them to predict what should have happened 2 years ago?

No. During the 2022 season, I had a UTR power subscription, which allowed me to see the pretty charts of my games won % and my opponent rating.

My average opponent UTR rating (combined UTR of both partners) during that season was about 10.8.

My UTR during that season stayed constant at about 6.8, with my partners all having UTR average of about 3.2. So my team UTR anveraged about 10.0, and always lower than opponent UTR in all matches.

I went 7-0, winning 61% of games. So the math didn’t add up.

 

[travlerajm]

No. During the 2022 season, I had a UTR power subscription, which allowed me to see the pretty charts of my games won % and my opponent rating.

My average opponent UTR rating (combined UTR of both partners) during that season was about 10.8.

My UTR during that season stayed constant at about 6.8, with my partners all having UTR average of about 3.2. So my team UTR anveraged about 10.0, and always lower than opponent UTR in all matches.

I went 7-0, winning 61% of games. So the math didn’t add up.

I wrote UTR to point out this discrepancy, but the response was that “the data team has looked at your record and determined everything is correct.”

So they did not admit their error (at the time, all matches were counted, but there was a bug in the algo that omitted the partner adjustment). But 6 months later, my doubles UTR suddenly jumped from 6.8 to 8.8.

Then a few months later than that, my UTR starting cycling up and down between 7ish and 9ish, depending on which matches were counted.

 

[Moon Shooter]

No. During the 2022 season, I had a UTR power subscription, which allowed me to see the pretty charts of my games won % and my opponent rating.

My average opponent UTR rating (combined UTR of both partners) during that season was about 10.8.

My UTR during that season stayed constant at about 6.8, with my partners all having UTR average of about 3.2. So my team UTR anveraged about 10.0, and always lower than opponent UTR in all matches.

I went 7-0, winning 61% of games. So the math didn’t add up.

I wrote UTR to point out this discrepancy, but the response was that “the data team has looked at your record and determined everything is correct.”

So they did not admit their error (at the time, all matches were counted, but there was a bug in the algo that omitted the partner adjustment). But 6 months later, my doubles UTR suddenly jumped from 6.8 to 8.8.

Then a few months later than that, my UTR starting cycling up and down between 7ish and 9ish, depending on which matches were counted.

Yes that rule may have a bigger effect on your rating due to some odd things about your matches. I think it is a bad rule but I think they have some other rules that hurt their overall predictions even more. The 12 month rule, the ceiling and floor and the hard line must win a match for 2.0 difference to count at all rule I think will hurt their predictions more.


I think they likely have that rule because people are not expecting partners to be over 4 UTR points different. So a team might go a set before they realize just how huge the difference in players is. And that could throw the predictions when these guys play normal matches. I agree though, in mixed, opponents should be looking for that. Yet even in mixed I see players still hitting to the male player with 1.5 higher USTA rating then the female when they have a shot that could go to either player. Does this accurately reflect skill level in the vast majority of games where these huge gaps in ability will not exist or will it throw off the predictions for normal matches? I guess they think it will throw it off. You should play some doubles with other 8 UTR players and see how you do. If you are playing just as well then I agree it might be a mistake. But if you would play worse on average then they are likely making a good decision for the overall predictions in their system.