trackmyrating.com

This site is not accurate whatsoever in how it calculates ratings.

I lost 6-1, 6-1 to a guy who, according to the site, was ~4.10 heading into the match. Assuming that piece is accurate, that would be roughly a 3.70 match rating, not a 3.33. And he would not get a 4.38 rating for that match. His rating would either go up or down a teeny bit or stay the same.
 
This site is not accurate whatsoever in how it calculates ratings.

I lost 6-1, 6-1 to a guy who, according to the site, was ~4.10 heading into the match. Assuming that piece is accurate, that would be roughly a 3.70 match rating, not a 3.33. And he would not get a 4.38 rating for that match. His rating would either go up or down a teeny bit or stay the same.
Hey there. There are well over 20 million match rows on the site and in aggregate its pretty accurate historically. There are some cases where a particular match is off either because the record of the player is thin or there is a particular bug. If there's one that you think is off you can always send me a bug report through the site while logged and I'm happy to look into it.
 
Hey there. There are well over 20 million match rows on the site and in aggregate its pretty accurate historically. There are some cases where a particular match is off either because the record of the player is thin or there is a particular bug. If there's one that you think is off you can always send me a bug report through the site while logged and I'm happy to look into it.
I don’t think it’s a bug when two people are about 0.50 apart dynamically heading into a match, the better guy wins 6-1, 6-1, and then gets a match rating 1.00 higher than the lower ranked player. At that score, there is no justification for a difference in match rating greater than 0.50. Tbh, it should probably be 0.40, but I could see 0.45.
 
I had a chance to look at the site. I like how responsive it is, and I generally like the interface. It would be nice to have some functionality to search for matches by participant, and to break down match history and some of the stats by league year, ratings period year, or calendar year (in New England, we start "league years" early, so for example we're already playing 2027 leagues now). The font is a little small and low contrast in some areas for us older league players. :-)

On the match rating side, I think there's some dampening that needs to happen at the tails of the expected outcome. I'm a USTA 3.5 and have some 3.5 matches where I allegedly played like a 4.30, 4.40, and even 5.18, which is very flattering but really that's so far outside the general match level that it would be impossible to judge a skill mismatch like that accurately.

By the way, the 5.18 match is a match in which my partner was playing his first USTA match as a self-rated 3.5, and the system seems to have assumed he started with a 3.25 rating. He had only been playing for about a year but he's a fantastic athlete; it turns out that he should be rated much higher. Almost all of his individual match ratings in your system are in the 4.0 to 4.50 range, but because of that starting anchor you only have him as a 3.70. I'd modify your method to use a player's first several matches against players with stable ratings to estimate the new player's rating, then go back and assign final match ratings that influence the other participants' ratings only after there's a reasonably stable estimate for all of the participants. Matches in which there are participants whose ratings aren't stabilized yet shouldn't be used for ratings calculations. To handle S-rates that end up not playing much, you could just limit how far you go back and recalculate matches -- some matches will never count because proper ratings can't be estimated.
Thanks for the suggestions @time_fly . Self-rates are tricky, especially if they are far outside the bounds of the norm like your friend. If you can kick me a note while signed in using the feedback form I'll be able to look at the matches you're talking about specifically to see if its just that he plays significantly higher than the average of 3.5's or if there is something else going on. At this point there are thousands of people using the site and many are asking for features or asking questions about particular features so I won't be able to dig into it unless I know which specific match you're talking about. Thanks again for the suggestion, its helpful.
 
Thanks for the suggestions @time_fly . Self-rates are tricky, especially if they are far outside the bounds of the norm like your friend. If you can kick me a note while signed in using the feedback form I'll be able to look at the matches you're talking about specifically to see if its just that he plays significantly higher than the average of 3.5's or if there is something else going on. At this point there are thousands of people using the site and many are asking for features or asking questions about particular features so I won't be able to dig into it unless I know which specific match you're talking about. Thanks again for the suggestion, its helpful.
Have you noticed any interesting patterns for traffic to your site, such as a flurry of activity Thursday night before nationals?
 
@silverwyvern4 it's live, as of the update on the 28th. Fall matches in the early-start leagues now count toward this year's estimate, the year they were played in.

I re-ran the same check as last time on a fresh random 400 active players. On the 21st, 79 of 343 had a headline stuck behind newer matches. Now none of 353 do.

Thanks again for the flag.
 
First of all - great job on your site!

I have a question regarding how the dynamic rating is calculated. In one of the posts you mentored that USTA most likely takes last match 'rating' and averages it with previous 3 _dynamic ratings_, and the result is the current dynamic rating. That makes sense to me. I would assume that your calculation uses the same logic.

But then in other posts you seem to imply that you take last match rating and you average that with previous three _match ratings_. And the result is the current dynamic rating. And that is why the most recent match that is above expectation can still lower the dynamic rating because the match that is 5 matches _old_ and that was a really good result no longer really is taken into consideration.

Could you elaborate on what the logic is?

also, is there any trick I need to do to have my matches updated more often? My accounts again not showing matches from past two weeks or so.
 
First of all - great job on your site!

I have a question regarding how the dynamic rating is calculated. In one of the posts you mentored that USTA most likely takes last match 'rating' and averages it with previous 3 _dynamic ratings_, and the result is the current dynamic rating. That makes sense to me. I would assume that your calculation uses the same logic.

But then in other posts you seem to imply that you take last match rating and you average that with previous three _match ratings_. And the result is the current dynamic rating. And that is why the most recent match that is above expectation can still lower the dynamic rating because the match that is 5 matches _old_ and that was a really good result no longer really is taken into consideration.

Could you elaborate on what the logic is?

also, is there any trick I need to do to have my matches updated more often? My accounts again not showing matches from past two weeks or so.
Hi @jmnk thanks,

It's the first one. The way USTA has described it in an older FAQ of theirs your new dynamic rating is the match rating you just earned averaged with up to your previous three dynamic ratings, not your previous match ratings. That difference makes a difference because each of those dynamic ratings already has everything before it baked in. So an old great result never drops out cleanly the way I made it sound. It just fades.

Here's how a match above expectation can still nudge you down. If you've been climbing, the older ratings in that average sit below where you are now, and the average pulls toward them. Say your last three were 3.40, 3.44 and 3.50, and you post a 3.54. The new rating is (3.54 + 3.40 + 3.44 + 3.50) / 4 = 3.47, which is below the 3.50 you walked on court with, even though you beat it. It works the same way in reverse on the way down.

Coming from Elo, the difference is that Elo adds a step to your current rating, so beating expectation can only move you up. This is a moving average of your recent ratings, so where you were a few matches ago still gets a vote.

On my side, the structure is the same, but USTA knows everybody's exact dynamic rating and I don't. All I can see are the match results and the year-end ratings they publish, so I estimate everyone's dynamic from the results and keep checking those estimates against how the year-end ratings actually come out. The numbers on the site are that approximation.

If you send me a note while logged in I'll be able to see if something funny is going on with your matches. Shouldn't be missing two weeks.

Todd
 

Toddero​

Thanks for the nice site. BTW in my section tri-level does count toward ntrp end of year calculations. I am not sure if they are included in the normal dynamic rating or if something else happens.
@Moon Shooter sorry for the slow reply. As you said tri-level counts toward your year-end rating, and your tri-level results have been in your estimate all along.

The year-end rating and the running dynamic are two different things, and a league can be in one and out of the other. Tri-level isn't one of the leagues the *******'s own league Q&A lists as producing dynamic values, so a tri-level match counts toward the number USTA publishes in December without moving your dynamic along the way.

Looking into your post turned up two things I had wrong though so appreciate the flag. Thanks!
 
It will be interesting to see how this site does at predicting EOY.

I’m a 4.08 on here but a 4.31 on TR.
For sure. Our models estimate differently but are fairly close historically. There are definitely outliers and yours is a pretty significant gap (so is mine incidentally). Either way I’ll learn and adapt the model in December. My model does tend to skew a little more conservative (I estimate promotions more accurately whereas they estimate demotions more accurately) but I’m ultimately aiming for accuracy not being conservative. I reworked the model so many times that I was getting diminishing returns so it’ll be good to ground truth it in December.
 
For sure. Our models estimate differently but are fairly close historically. There are definitely outliers and yours is a pretty significant gap (so is mine incidentally). Either way I’ll learn and adapt the model in December. My model does tend to skew a little more conservative (I estimate promotions more accurately whereas they estimate demotions more accurately) but I’m ultimately aiming for accuracy not being conservative. I reworked the model so many times that I was getting diminishing returns so it’ll be good to ground truth it in December.
Great work. It’s fun having another source of data to compare.
 
@Moon Shooter sorry for the slow reply. As you said tri-level counts toward your year-end rating, and your tri-level results have been in your estimate all along.

The year-end rating and the running dynamic are two different things, and a league can be in one and out of the other. Tri-level isn't one of the leagues the *******'s own league Q&A lists as producing dynamic values, so a tri-level match counts toward the number USTA publishes in December without moving your dynamic along the way.

Looking into your post turned up two things I had wrong though so appreciate the flag. Thanks!
What is the end result difference between these matches counting in year end and them not effecting our dynamic rating? Is it essentially very close to effecting the dynamic rating just some matches are out of chronological order?
 
What is the end result difference between these matches counting in year end and them not effecting our dynamic rating? Is it essentially very close to effecting the dynamic rating just some matches are out of chronological order?
@Moon Shooter honest answer: nobody outside USTA knows exactly.

The ******* book says tri-level matches are part of generating your year-end rating, so they do count toward whether you move up or down in December. What it doesn't say is how. USTA doesn't publish the year-end calculation, so "the same math with some matches out of order" is a reasonable guess. They could also weight those matches differently at year-end and nothing public would tell us.

During the season the difference is simpler, at least in the *******. Tri-level isn't one of the leagues the ******* uses for dynamic ratings, so a tri-level result doesn't move the running number USTA uses mid-season. That part varies by section: Southern, for one, only says tri-level can't get you bumped up mid-season.

On the site I'm estimating the year-end, so your tri-level matches go into your number in the order you played them, same as your other league matches. If USTA treats them differently, December is when I'll get a hint, and I'll adjust the model. Hopefully that was moderately helpful. :)
 
@Toddero I see a number of missing mixed matches. I think the current mixed season is being indexed as 2027 and not showing up as a result for the current year. This is PNW NWW and SWW
@ItalianScallion you're right about the 2027. Your fall mixed matches are on your page in Match history by date, but the "How you rank at your level" table files them under 2027 because USTA names the fall mixed league as 2027. I'll move it so that table puts them with the year you played them.

If you don't see them in Match history either, check the "Only matches that count toward my USTA rating" switch above the list. Mixed doesn't count for you, so that switch hides it.

Your three newest, from the 6th and the 8th, are in and should show up by Wednesday morning.
 
"−0.021 better than expected — took 67% of games against 53% expected, played like a 4.48, 6-3, 6-3"

Can someone provide some clarity on this example?

The person took more games than expected yet it decreased the rating .021?
 
"−0.021 better than expected — took 67% of games against 53% expected, played like a 4.48, 6-3, 6-3"

Can someone provide some clarity on this example?

The person took more games than expected yet it decreased the rating .021?
It’s a garbage website designed to generate clicks.
 
Easy there. It's a free site that gives us more info. We shouldn't complain, let alone insult. Keep it to yourself.
UTR is the best evaluative tool out there IMO.

It gives a pretty accurate assessment of where people are, and then it’s up to you to reconcile what’s on paper with what you see with your eyes.

The biggest issue I see with track my rating is they have way too many people getting bumped down and not enough people getting bumped up.
 
UTR is the best evaluative tool out there IMO.

It gives a pretty accurate assessment of where people are, and then it’s up to you to reconcile what’s on paper with what you see with your eyes.

The biggest issue I see with track my rating is they have way too many people getting bumped down and not enough people getting bumped up.
I don't agree. UTR is all over the place because it includes mixed results.
 
UTR is the best, and also has issues with mixed - both can be true.

It’s a cool website, and I love the historical stats, but my bump chance changes randomly from 5-20% week to week. Also how your section does at nationals plays a role that no one knows how to factor in.
 
UTR is the best, and also has issues with mixed - both can be true.

It’s a cool website, and I love the historical stats, but my bump chance changes randomly from 5-20% week to week. Also how your section does at nationals plays a role that no one knows how to factor in.
It absolutely has issues with mixed. There are multiple guys who are rated 3.0 (they’ll be bumped to 3.5) who have UTRs in the low end of the 4.0 range.

I know it’s not accurate whatsoever.
 
It absolutely has issues with mixed. There are multiple guys who are rated 3.0 (they’ll be bumped to 3.5) who have UTRs in the low end of the 4.0 range.

I know it’s not accurate whatsoever.
I am a UTR 3 based on my mixed rating. My 3.5 ntrp female mixed partner has higher UTR than me (I’m 4.5c).
 
"−0.021 better than expected — took 67% of games against 53% expected, played like a 4.48, 6-3, 6-3"

Can someone provide some clarity on this example?

The person took more games than expected yet it decreased the rating .021?
Hi @Blade, it isn't a bug, though I can see why it reads like one.

The new rating isn't the old one plus a step for this match. It's this match averaged with the player's recent ratings. I looked up that match. They came in at 4.22, but that number came from one big win in October, and the ratings before it were mostly around 4.10. The 6-3, 6-3 was worth a 4.48, well above where they were, but averaged in with those 4.10s it comes out at 4.19, just under where they started. Keep playing like that and the average catches up.

There's a small "i" on rows where the rating and the result point opposite ways that explains this, but it's easy to miss. Hope that helps.
 
UTR is the best, and also has issues with mixed - both can be true.

It’s a cool website, and I love the historical stats, but my bump chance changes randomly from 5-20% week to week. Also how your section does at nationals plays a role that no one knows how to factor in.
I just did a pretty big update in the model so that’s what moved it. The previous model was built when I only had 7 million matches I was modeling. The database is significantly more robust now. I wanted a best guess before the new rating results come out in December.
 
It’s a garbage website designed to generate clicks.
It’s free, it’s free of adds, and the only reason it’s available is because someone generously shares their time and analyses.
UTR is the best evaluative tool out there IMO.
Very different product, almost entirely unrelated to dynamic ratings and EOY bumps, you have to pay for it to get 95% of the information.

Apples and oranges, dude.
 
I am a UTR 3 based on my mixed rating. My 3.5 ntrp female mixed partner has higher UTR than me (I’m 4.5c).
How many matches is UTR calculating from? This is most surely due to a lack of data rather than algorithm issue, even though mixed is going to have the most volatile results on your UTR since that's where we're going to see the biggest UTR deltas between players
 
How many matches is UTR calculating from? This is most surely due to a lack of data rather than algorithm issue, even though mixed is going to have the most volatile results on your UTR since that's where we're going to see the biggest UTR deltas between players
I have reported the UTR doubles algorithm bug before, starting when I noticed it 4 years ago.

That year, in 2022, I went undefeated 10-0 in usta 8.0 mixed (as a 4.5c, playing with several low 3.5c female partners), winning about 61% of games, meaning most matches were routine scorelines.

My UTR doubles rating end of that that year was about 1 UTR unit below my average opponent UTR rating (oponent UTR match rating for doubles = sum of individual opponent UTR ratings minus partner UTR rating).

This was obviously an impossible result, indicative of algorithm error. When I emailed UTR about this, I received a reply:

“The data team has reviewed your query, and determined that your UTR rating is correct.”
 
Last edited:
I just did a pretty big update in the model so that’s what moved it. The previous model was built when I only had 7 million matches I was modeling. The database is significantly more robust now. I wanted a best guess before the new rating results come out in December.

Good because my bump chance is lower now - 5%. I did lose in a 10 pt tiebreak to a self rate that got DQd to 5.0. Somehow that match moved my dynamic rating down -.111. I’ve always wondered how self rate matches are handled, do they re-adjust at end of year?
 
Good because my bump chance is lower now - 5%. I did lose in a 10 pt tiebreak to a self rate that got DQd to 5.0. Somehow that match moved my dynamic rating down -.111. I’ve always wondered how self rate matches are handled, do they re-adjust at end of year?
@jdawgg self-rates are the messiest part of this, so it's a good question. Especially when people are sandbagging or experiencing delusions of grandeur like I did when I got back into it.

USTA's own FAQ says a match against a self-rated player doesn't generate a dynamic rating until the self-rate has enough results against established players. By their rule, that match waits until the self-rate has shown what level they really are.

Mine doesn't wait. It scores the match right away, and for a brand-new self-rate all it has to go on is the level they entered at. Lose a tiebreak to someone it thinks is a mid 4.5 who is really a 5.0, and it reads as a bad loss. That's most likely where your -.111 came from.

It does catch up. I recompute the whole season with every update, so once the DQ shows up in their USTA rating, they start the season at the new level and that match gets scored again against it. If the DQ hasn't reached me yet, you should see that match cost you a lot less once it does.

On year end, USTA hasn't published exactly how a DQ'd self-rate's earlier matches are treated.

If you send this to me from your own player page while signed in (the Feedback button, bottom right), I can look at that exact match and tell you what happened with it with more precision.
 
@jdawgg self-rates are the messiest part of this, so it's a good question. Especially when people are sandbagging or experiencing delusions of grandeur like I did when I got back into it.

USTA's own FAQ says a match against a self-rated player doesn't generate a dynamic rating until the self-rate has enough results against established players. By their rule, that match waits until the self-rate has shown what level they really are.

Mine doesn't wait. It scores the match right away, and for a brand-new self-rate all it has to go on is the level they entered at. Lose a tiebreak to someone it thinks is a mid 4.5 who is really a 5.0, and it reads as a bad loss. That's most likely where your -.111 came from.

It does catch up. I recompute the whole season with every update, so once the DQ shows up in their USTA rating, they start the season at the new level and that match gets scored again against it. If the DQ hasn't reached me yet, you should see that match cost you a lot less once it does.

On year end, USTA hasn't published exactly how a DQ'd self-rate's earlier matches are treated.

If you send this to me from your own player page while signed in (the Feedback button, bottom right), I can look at that exact match and tell you what happened with it with more precision.
This is reasonable way to handle self-rate.

You could also consider assigning a confidence level to the match rating, so that you can include the match and not have to censor it, while also not letting mis-rated selfies screw up your data too much. So a match having a self-rate opponent with no data would be weighted lower than other match ratings at first.
 
@jdawgg self-rates are the messiest part of this, so it's a good question. Especially when people are sandbagging or experiencing delusions of grandeur like I did when I got back into it.

USTA's own FAQ says a match against a self-rated player doesn't generate a dynamic rating until the self-rate has enough results against established players. By their rule, that match waits until the self-rate has shown what level they really are.

Mine doesn't wait. It scores the match right away, and for a brand-new self-rate all it has to go on is the level they entered at. Lose a tiebreak to someone it thinks is a mid 4.5 who is really a 5.0, and it reads as a bad loss. That's most likely where your -.111 came from.

It does catch up. I recompute the whole season with every update, so once the DQ shows up in their USTA rating, they start the season at the new level and that match gets scored again against it. If the DQ hasn't reached me yet, you should see that match cost you a lot less once it does.

On year end, USTA hasn't published exactly how a DQ'd self-rate's earlier matches are treated.

If you send this to me from your own player page while signed in (the Feedback button, bottom right), I can look at that exact match and tell you what happened with it with more precision.
Sent the feedback, interested to know what happened, thanks.
 
UTR is the best evaluative tool out there IMO.

It gives a pretty accurate assessment of where people are, and then it’s up to you to reconcile what’s on paper with what you see with your eyes.

The biggest issue I see with track my rating is they have way too many people getting bumped down and not enough people getting bumped up.
UTR is kind of trash really for recreational players. It changes your rating and when you look up previous matches, they show the updated rating. It makes it worthless for tracking performance or improvement. I won't even get into some weird ratings shenanigans. For juniors or players playing a ton in tournaments, it's a different issue. But those players probably don't care about league play :)
 
UTR is kind of trash really for recreational players. It changes your rating and when you look up previous matches, they show the updated rating. It makes it worthless for tracking performance or improvement. I won't even get into some weird ratings shenanigans. For juniors or players playing a ton in tournaments, it's a different issue. But those players probably don't care about league play :)
My latest annoyance with UTR…

I was about utr 7.5 after playing my most recent full season of UTR singles flex league couple of years ago…

But the algo decays the rating for adult players upon inactivity. So now my singles rating is 5.8. Last time I started with a UTR singles league, I went thru two full seasons against UTR 6 players, winning all my matches 6-0, 6-0 or close to that. Once again all my opponents are UTR 6’s so I just wasted my money.
 
This is reasonable way to handle self-rate.

You could also consider assigning a confidence level to the match rating, so that you can include the match and not have to censor it, while also not letting mis-rated selfies screw up your data too much. So a match having a self-rate opponent with no data would be weighted lower than other match ratings at first.
That’s a good suggestion. Just so I understand what I think you’re saying…based on the confidence level I would assign to the s rated individual I would then assign a similar weight to the match. Low confidence = low weighting? If that’s what you’re suggesting I like it a lot and will start doing that. Great suggestion.
 
Back
Top