USTTA rating v.s. expected win-rate

This user has no status.
This user has no status.
Active Member
Oct 2012
891
510
1,733
Read 1 reviews
There’s always a better graph.

And I’m also a number oriented person, that’s why I have a science and an engineering degree, but I know when to not be number oriented, that’s the key, life isn’t always about numbers.
Same here got my MS in mechanical engineering. and have been practicing engineering for more than 35 years (humm why should I disclose this...?) . Life is not always but sometimes is about numbers.
 
This user has no status.
This user has no status.
Member
Dec 2018
413
447
1,453
Read 1 reviews
David Marcus' Rating Central system is much better.
Agreed, though for better accuracy a Glicko system like Ratings Central needs more data than an Elo system. Might not make much difference outside table tennis hotbeds. Could be why USATT doesn't use it, or maybe it's just politics.
 
This user has no status.
This user has no status.
Well-Known Member
Jul 2017
1,772
856
2,945
I thought the curve was based on real data?
It should be. The theoretical ratings difference vs percent win graph in OldUser's link looks like it is for chess.
As and engineer, would you trust an equation that failed if the input is just a little more or little less than the "normal" range. You know their is always noise. You get some credit of doubting why the probability is 1 with a ratings difference of 250.

No one one this forum knows how to calculate the curves for the ratings correctly. To prove that I do I will send you a PM that shows how it done and no one else know what they are talking about. I will send another link showing an example that a mechanical engineer should understand.
 
This user has no status.
This user has no status.
Well-Known Member
Jan 2019
1,119
723
2,225
Read 2 reviews
It should be. The theoretical ratings difference vs percent win graph in OldUser's link looks like it is for chess.
As and engineer, would you trust an equation that failed if the input is just a little more or little less than the "normal" range. You know their is always noise. You get some credit of doubting why the probability is 1 with a ratings difference of 250.

No one one this forum knows how to calculate the curves for the ratings correctly. To prove that I do I will send you a PM that shows how it done and no one else know what they are talking about. I will send another link showing an example that a mechanical engineer should understand.
Ohh so secretive , you joined the free masons or what? Btw, you’ve got an error there in your post, it’s obvious though so you should know what’s wrong.
 
  • Like
Reactions: hclnnkhg
This user has no status.
This user has no status.
Active Member
Oct 2012
891
510
1,733
Read 1 reviews
Now I think (I hope) I understand both hclnnkhg and brokenball's points. hclnnkhg was just trying to estimate probability of winning/losing with a simple equation while brokenball wants a more accurate equation (sigmoid) that can match real data. However, there is no real data (for example: results of win/lose of 1000 rated players playing multiple matches in USTTA database) here for comparison to tell which equation is better to match.
It is clear that the quadratic function is not good if the range is beyond 250. For me only, I'd not use it to tell my friends that I will beat you guys 100% for sure since my rating is 250 pts above :) . But I may still use it to have a fun discussion with some players who are 100 or 50 pts higher/lower than me. Life is short, folks.
Thanks brokenball for the video. It has a lot of info that I need more time to digest but I get your point.
 
This user has no status.
This user has no status.
Well-Known Member
Jul 2017
1,772
856
2,945
Now I think (I hope) I understand both hclnnkhg and brokenball's points. hclnnkhg was just trying to estimate probability of winning/losing with a simple equation while brokenball wants a more accurate equation (sigmoid) that can match real data. However, there is no real data (for example: results of win/lose of 1000 rated players playing multiple matches in USTTA database) here for comparison to tell which equation is better to match.
The sigmoid is the better match because no matter which ratings difference you use, the result is a probability that goes from not quite -1 to not quite +1. This confirms your belief that the probability shouldn't be and isn't 1 at 250 although it is pretty close.

It is clear that the quadratic function is not good if the range is beyond 250.
Yes, yes, yes but I took a lot of sh!t for just saying it was obvious.

For me only, I'd not use it to tell my friends that I will beat you guys 100% for sure since my rating is 250 pts above :) . But I may still use it to have a fun discussion with some players who are 100 or 50 pts higher/lower than me. Life is short, folks.
The pdf I sent you was only pages 12-13 out of 29. The rest has all sort of caluations.
Yes. it is best to discus with friends. This forum is a hostile and very woke crowd.

Thanks brokenball for the video. It has a lot of info that I need more time to digest but I get your point.
The one modeling the non-linear valve? That is complex. The USATT data to sigmoid is trivial by comparison. The cool thing is that you pause and scrub back and forth to go over the video a couple of times.
 
This user has no status.
This user has no status.
Well-Known Member
Jan 2019
1,119
723
2,225
Read 2 reviews
The sigmoid is the better match because no matter which ratings difference you use, the result is a probability that goes from not quite -1 to not quite +1. This confirms your belief that the probability shouldn't be and isn't 1 at 250 although it is pretty close.


Yes, yes, yes but I took a lot of sh!t for just saying it was obvious.


The pdf I sent you was only pages 12-13 out of 29. The rest has all sort of caluations.
Yes. it is best to discus with friends. This forum is a hostile and very woke crowd.


The one modeling the non-linear valve? That is complex. The USATT data to sigmoid is trivial by comparison. The cool thing is that you pause and scrub back and forth to go over the video a couple of times.
No you didn't take a lot of sh!t for saying it is obvious. You took a lot of sh!t because you said it was obvious while it's obviously not obvious to the OP, and refused to elaborate, in other words you took sh!t because you were just being rude, that's all. Also, obviously you're ignoring the post where you were being pointed out that you were obviously wrong, classic.
 
This user has no status.
This user has no status.
Well-Known Member
Jul 2017
1,772
856
2,945
No you didn't take a lot of sh!t for saying it is obvious. You took a lot of sh!t because you said it was obvious while it's obviously not obvious to the OP, and refused to elaborate, in other words you took sh!t because you were just being rude, that's all. Also, obviously you're ignoring the post where you were being pointed out that you were obviously wrong, classic.
I thought I was being nice by giving you a chance to show us your ability with numbers.
 
This user has no status.
This user has no status.
Well-Known Member
Jan 2019
1,119
723
2,225
Read 2 reviews
I thought I was being nice by giving you a chance to show us your ability with numbers.
Like I said way way back, I didn't come on a TT forum to do math. And you're clearly still not acknowledging where you were wrong in the post I quoted. I guess it's not obvious to you, just like what you said was obvious, but was not to the OP.
 
This user has no status.
This user has no status.
Well-Known Member
Nov 2020
1,071
788
4,032
Now I think (I hope) I understand both hclnnkhg and brokenball's points. hclnnkhg was just trying to estimate probability of winning/losing with a simple equation while brokenball wants a more accurate equation (sigmoid) that can match real data. However, there is no real data (for example: results of win/lose of 1000 rated players playing multiple matches in USTTA database) here for comparison to tell which equation is better to match.
It is clear that the quadratic function is not good if the range is beyond 250. For me only, I'd not use it to tell my friends that I will beat you guys 100% for sure since my rating is 250 pts above :) . But I may still use it to have a fun discussion with some players who are 100 or 50 pts higher/lower than me. Life is short, folks.

In the context of this thread, having an approximation is not necessary. Neither for the expected win-rate, nor for the real win-rate. Actually, it is a distraction. And having an equation for the approximation, is by extension, even less necessary. This is from my perspective, and I am confident it is shared by couple of others.

For BrokenBall and you, it is all about approximation. Well, why not. Good luck. I agree in 1 thing: life is short.
 
says Shoo...nothing to see here. - zeio
says Shoo...nothing to see here. - zeio
Well-Known Member
Jan 2018
7,413
9,454
18,650
Gotta make another exception because of latej.

At the end of the day, what matters is does the system work as intended?

If a ranking system, which is used for seeding purposes, is working as intended, then the ranking differences among the players in the later rounds of a tournament should theoretically not deviate that much.

An example is the WTT Feeder Fort Lauderdale 2022. Both the ITTF TTWR and Ratingscentral failed to predict the upsets by the eventual winner Amy Wang, but the former actually works as intended better, as 3 out of 4 QF matches (that 1 QF match is actually off by design) are between players closer in ranking than the latter indicates.

p.s. Some background info on the USATT rating chart. If what David Marcus claims is true, then this thread is trying to reverse-engineer the win probability (and look for a perfect fit) that was pulled out of thin air.

rZ8f96T.png
 
Last edited:
  • Like
Reactions: latej
Top