Do we actually know how TR and TLS calculate their ratings?

#1
I don’t mean how well they predict or track USTA ratings, but rather how are they calculated. For instance, I have been told that TR takes the starting dynamic rating averaged with the most recent 3 matches. But this doesn’t seem exactly correct. Any idea for TR or TLS?
 
#2
I noticed that your match rating is approximately your opponents rating plus percentage of games won minus 0.5. So if your opponent is a 4.25 and you win 6-3 6-3 then you won 66% of the games. Your match rating is 4.25+0.66-0.5 = 4.41. Or there abouts.
 
#3
I noticed that your match rating is approximately your opponents rating plus percentage of games won minus 0.5. So if your opponent is a 4.25 and you win 6-3 6-3 then you won 66% of the games. Your match rating is 4.25+0.66-0.5 = 4.41. Or there abouts.
I have no idea if you’re right or wrong.

But if that is the calculation, then your own rating makes no difference, right?
 
#4
I have no idea if you’re right or wrong.

But if that is the calculation, then your own rating makes no difference, right?
I agree that your own rating before the match definitely has an influence on it. I have been told that it is your rating before your third to last match averaged with your last three matches. This gets it close, but not quite. And it doesn’t address how a rating for an individual match is calculated.
 
#5
I agree that your own rating before the match definitely has an influence on it. I have been told that it is your rating before your third to last match averaged with your last three matches. This gets it close, but not quite. And it doesn’t address how a rating for an individual match is calculated.
I’m just speculating but from what i gather the usta has some algorithm for an “expected result” based on each player’s rating going into the match. I even think I’ve read in usta published material that a top of level player “should” beat a bottom of level player 0&0, so that’s some evidence to the “expected result” theory.

I would expect tls and tr probably use some kind of “expected result” in their match calculations as well. If that’s wrong, then it doesn’t matter if their dynamic rating calculation is correct or not; garbage in, garbage out. My guess is they have both the match rating and dynamic rating calculation incorrect.

@schmke could probably speak more intelligently on the subject than any other on these boards. Or maybe not because from everything i read (and he’s told me), his calculations more accurately predict the usta calculations than either tls or tr.
 
#6
I believe they throw a bunch of names in one hat and assign them to random ratings drawn from another hat.


Sent from my iPhone using Tapatalk
 
#9
I agree that your own rating before the match definitely has an influence on it. I have been told that it is your rating before your third to last match averaged with your last three matches. This gets it close, but not quite. And it doesn’t address how a rating for an individual match is calculated.
I started out by looking at self rates with only one match on record. They didn't have a before rating so its not in the calc. I wanted to look at cases with less unknowns.
 
#10
So I took about 50 singles results and plotted game percentage vs the difference between opponent and match rating and its almost a perfect linear fit. The only outliers were against opponents playing up. Also, the TR guy might not be using match ratings in the rating calculation or not averaging them directly. Here's some results from the website.

Match DNTRP
3.88 3.88
3.83 3.88
4.19 3.95
3.73 3.84
3.90 3.88
4.03 3.89

The last match rating was the same as both the previous and current rating. So that last match rating would have had to replace another 3.88 rating for it to not influence the dynamic rating. There was no 3.88 in the previous 6 matches and futher. So either the match rating is not used in the calc or the matches are not weighted equally.
 
Last edited:
#11
I noticed that your match rating is approximately your opponents rating plus percentage of games won minus 0.5. So if your opponent is a 4.25 and you win 6-3 6-3 then you won 66% of the games. Your match rating is 4.25+0.66-0.5 = 4.41. Or there abouts.
And a loss? If you lose 6-3 6-3 to a 4.25 would you subtract 0.66?


Sent from my iPhone using Tapatalk
 
#14
Yes, and you add 0.5. So instead of 4.25+0.66-0.5=4.41, it's 4.25-0.66+0.5=4.09
Not being sarcastic, but how were you smart enough to figure TR’s formula? You just guessed around with random people’s ratings?

Also they seem to they seem to trash everyone’s rating from last year. Like if they predicted a bump from a dNTRP or 4.03 and that person stayed a 4.0 they just started them at top or bottom of level depending on how you look at it


Sent from my iPhone using Tapatalk
 
#15
Not being sarcastic, but how were you smart enough to figure TR’s formula? You just guessed around with random people’s ratings?

Also they seem to they seem to trash everyone’s rating from last year. Like if they predicted a bump from a dNTRP or 4.03 and that person stayed a 4.0 they just started them at top or bottom of level depending on how you look at it


Sent from my iPhone using Tapatalk
I didn't. I'm pretty close on the match ratings but I still don't know how he's using them for the dNTRP estimate. And yes I grabbed like 20 of my last singles matches and tried different things until something lined up.

On your second point, even if TR had the actual formula used by the USTA, they would also need all the current dNTRP ratings to calculate forward. The only time they get actual confirmation of dNTRP is when someone moves up or down. So they can only adjust once a year.
 

J_R_B

Hall of Fame
#16
So I took about 50 singles results and plotted game percentage vs the difference between opponent and match rating and its almost a perfect linear fit. The only outliers were against opponents playing up. Also, the TR guy might not be using match ratings in the rating calculation or not averaging them directly. Here's some results from the website.

Match DNTRP
3.88 3.88
3.83 3.88
4.19 3.95
3.73 3.84
3.90 3.88
4.03 3.89

The last match rating was the same as both the previous and current rating. So that last match rating would have had to replace another 3.88 rating for it to not influence the dynamic rating. There was no 3.88 in the previous 6 matches and futher. So either the match rating is not used in the calc or the matches are not weighted equally.
When you see this repeating in TR, it usually means one of the players is self-rated in their first two matches and therefore has no rating, and the match is just used to set that player's rating and everyone else's is just brought forward from the existing value.

I have no idea if you’re right or wrong.

But if that is the calculation, then your own rating makes no difference, right?
This is correct. I'm not sure of the formula they use, but I have seen that they start their calculation with the opponent's rating. I saw this in a match that ended "tied" last year (i.e. 6-3, 2-6, 1-0 or something like that that evaluated to 9-9). Each player received the other's starting rating for the match instead of an identical rating, which I thought was weird. I also believe this is wrong. The rating should start by averaging the starting rating of everyone involved and then adjusting from there based on the score, so that if a match "ties", both sides should get the same rating, not their opponent's starting rating.
 
#18
When you see this repeating in TR, it usually means one of the players is self-rated in their first two matches and therefore has no rating, and the match is just used to set that player's rating and everyone else's is just brought forward from the existing value.
I wasn't looking at new players for that one. The fact that the match rating becomes the players first rating probably means that he actually uses that in the calc and not just as a FYI type thing.
 
#19
I like how TR is transparent about their ratings calculations - they send me emails with FAQ answers. Great tool for estimating level that it really only limited in cases of tiny sample size.

I like TLS too, but they seem to update less frequently.
 
Top