Do we really want to score 3-0 wins differently from 3-2 wins?, For nerds only.

This user has no status.
This user has no status.
Well-Known Member
Jul 2017
1,788
859
2,984
We can. It would be that great a change to the RC calculations. In addition there is much more computing power available today than there was 10 years ago so computer power and memory are not a problem anymore.

On a recent thread I initially thought it would be a great idea to score 3-0 wins higher than 3-1 wins. I did the math, it took me only a few minutes to get the shocking result and more time to pretty print it.

In short scoring 3-0 wins differently than 3-2 wins is something that I don't think most TT players would want or are ready for.
Below are calculations of the probability of winning games for best 3 out of 5. No big deal yet. The shocker occurs when one player is much better than the other. I start with the assumption that the best player will win 11 out of 12 matches. That is approximate difference of 138 ratings points which isn't much. From the probability of winning the match I can calculate probability of winning each game. Then I calculate the probability for
each win loss combination. Notice all the probabilities must add up to 1 and the probability of winning 3-0, 3-1 and 3-2 must add up to the probability of winning the match. This is my sanity check.

There are two ways of looking at this.
1. One is to multiply the probability of winning a game * games played . This is relatively simple.
For example compare a 3-5 win to the expected wins 5*p.game=3.85. . In this case the winner did not does well as expected. Should the winner actually have his ratings points go up or stay the same when he didn't perform up to the expected level for the ratings difference?
Another example. Compare a 3-0 win to the expected wins. 3*p.game=2.31 . In this case the higher player did slightly better than expected. Maybe he deserves more points than what the current RC system would provide.

2. The second method involves adding up the probabilities.
A 3-2 win involves adding up the probabilities for 0-3, 1-3, 2-3 and finally the 3-2 win. The sum is less that 0.5 which means the winner didn't do as well as expected and should maybe lose points.

Now when the ratings are much much closer to being equal the distribution of win combinations is closer to symmetrical binomial function and would work out much closer to the way people would expect.

Look at the chart below. The calculations are not too difficult.

The question I have is does anybody really care? People usually oppose change. That is clear to see. People usually don't like 'better' if they can't see why it is better for them.

I just thought I would but this 3-0 and 3-2 issue to sleep forever.


Screen Shot 2018-01-16 at 11.10.59.jpg
 
Last edited:
  • Like
Reactions: thomas.pong
This user has no status.
This user has no status.
Member
Dec 2016
496
391
1,723
Heck no, I saw your post on the other thread and felt like commenting about it. Said thread is already going in all directions so I didn't, but as someone who always takes time to really get into a match I don't think that should be the case, without any need for crunching numbers or anything. There are any number of comeback artists around that could really be prejudiced by this, and in my view it would add unnecessary pressure as opposed to "mind games" (for example letting the opponent get comfortable and perhaps over-confident by taking it easy and observing during the first game, at times surprisingly effective against sanguine players who can become frustrated, looking at their bat and stuff when things turn around), taking more risks while "testing the waters" or any other strategic aspect in general.
A win is a win, whichever way you take to achieve it. This to me makes things over-complicated and puts too much emphasis on a factor that shouldn't really be one.
 
  • Like
Reactions: UpSideDownCarl
This user has no status.
This user has no status.
Well-Known Member
Dec 2015
1,057
1,181
2,719
Read 1 reviews
No we don't. Winner takes it all is one of the essences of sports, what makes it so beautiful. Also people always have different playstyles. You would have to look at everyone independently to figure out if he performed "bad" or "good" and even then it would be subjective.
 
says MIA
says MIA
Well-Known Member
Nov 2016
2,132
1,095
11,017
I'd say a win is a win, you should never lose points if you win. However, maybe you should win more points with a 3-0 than a 3-1 or 3-2. More importantly, I think a player should lose less points if he loses 2-3 or 1-3 than 0-3. especially against a much higher rated player, maybe the losing player's rating should even be adjusted slightly higher if losing 2-3 to a much higher rated player, but that could prove difficult.
 
says Shoo...nothing to see here. - zeio
says Shoo...nothing to see here. - zeio
Well-Known Member
Jan 2018
7,742
9,644
19,211
I'd suggest to ask David J. Marcus the author of RC to answer it. TT or any sports' rating system is already too complex, it doesn't need to be messier.
He has already. He explained in an old article(4. Using scores) why he chose not to take the scores(games and points) into account for RC.
 
This user has no status.
This user has no status.
Active Member
Nov 2015
546
820
1,989
He has already. He explained in an old article(4. Using scores) why he chose not to take the scores(games and points) into account for RC.


It's a good run down of arguments and reasoning why he (DM) thinks it is not very practical to use actual game scores into the match, most of which boils down to "data might be crap". Most of it probably applies only to errors in recording number of points scored in the game, e.g. I saw myself people turning match score of 8,-8,8,8 which tells me that most likely nobody bothered to write down actual games scores. In my experience actual match score of 3:0, 2:3 is usually recorded accurately, but may be I'm lucky in that.

I suspect, however, if you could consider ideal case of games played in a sportsmanlike way (no losing on purpose, goofing off etc.) and being recorded error-free, then taking into account match score (3:0 vs. 3:2) and may be even game score (11:0 vs. 18:16) should lead to more accurate estimates, based on the argument that current one is not taking advantage of all available data, and in my experience "the one with more data typically wins". Whether improvement in accuracy is actually worth extra complexity - well, it's another topic altogether.
 
  • Like
Reactions: thomas.pong
This user has no status.
This user has no status.
Well-Known Member
Jul 2017
1,788
859
2,984
David Marcus makes a good point about not using scores but here and in the other thread we were talking about games won and lost, not scores. If people wanted to score by games it is very possible. The mathematics/probabilities is not a problem. Computing power is not a problem. People don't need to see or understand the complications. It should be hidden from them.

The first reason I started this thread was I saw the possibility of better players losing points for not beating low players by a wide enough margin. It raises the importance of each game by a lot. I didn't think that would be accepted and it appears I was right.

The second reason was to show that the scoring would be much more complicated than saying winning 3-0 gets you all the points and 3-1 gets you 75% of the points. Hopefully this idea never comes up again.

The next reason is to keep people from trying to waste time "improving" RC.

Finally, ERT's thread had gone off the tracks.

I did learn something from reading the using scores link above. It is interesting that new players are supposed to be started with a rating of 1400 with a standard deviation of 450. I know for a fact that this has not be done where I play. Where I play the where started out with much lower scores and not so large a standard deviation. This brings the RC ratings of all the players down even though RC is not a zero sum rating system. It is still hard to get points if every one is rated 200-400 points too low.

I have always felt the RC ratings in my area were too low for a long time and now I know why. I could bitch but that would do me no good. I know my first coach would not play in RC events. He came from the NE with a regular RC rating. There were no points to win here because his RC rating was 200 points more than everybody else.

My point is that the RC rating system is really very good. Sometimes the people that run the tournaments and enter the data are not.
 
  • Like
Reactions: thomas.pong
says Shoo...nothing to see here. - zeio
says Shoo...nothing to see here. - zeio
Well-Known Member
Jan 2018
7,742
9,644
19,211
Here is the problem. Once you incorporate the games scored in the evaluation of ratings, you need to answer the question why not the points scored as well? Is the latter any less important, given the former is the direct consequence of the latter?

Consider the 3 semifinal encounters between Wang Hao and Ma Long in the WTTC in 2009, 2011, and 2013. The scores were as follows:

Games(Points), Total Points/Difference
2009 - 4:1(10, 8, -10, 9, 9), 55:48/7
2011 - 4:2(7, -9, -7, 10, 7, 9), 61:55/6
2013 - 4:2(9, 7, 3, -7, -12, 3), 63:47/16

The match in 2011 proved to be the closest, followed by the one in 2009, and lastly the one in 2013. The total points scored is more or less in line with that observation. However, the same game score of 4:2 in 2011 and 2013 didn't reflect the striking difference in intensity at all. ML led 2:1 in games before dropping the next 3 in 2011. WH led 3:0 in games before dropping the next 2, then ML gave up in the 6th in 2013. Would you rate both players the same way in both matches just because the game score happened to be the same?
 
  • Like
Reactions: thomas.pong

JST

This user has no status.

JST

This user has no status.
Member
Nov 2017
280
191
1,259
Btw. there are ELO ratings which count matches and games separately (so you have main ELO rating and then "games" ELO which might be higher or lower). While it doesn't mess up with the ranking as such (for people who are keen into these and count just victories) it gives good indication if you are playing in the league below or above your level and similar. And once you have these data you can start experimenting with mixed ranking as well (e.g. as 3rd stat) and if it gives better predictions then why not to switch to it in few years? But I guess this is very much NOGO in case of USATT or similar system with so many players, history, politics, bias...
 
  • Like
Reactions: thomas.pong
This user has no status.
This user has no status.
Well-Known Member
Jul 2017
1,788
859
2,984
Here is the problem. Once you incorporate the games scored in the evaluation of ratings, you need to answer the question why not the points scored as well?
Is the latter any less important, given the former is the direct consequence of the latter?
In theory you are right but David Marcus makes a good case against. I don't like the fact that each net ball or edge ball affects ratings.

The effect would be that people would fight for every point even if they know they have lost. This could be a good thing.

Consider the 3 semifinal encounters between Wang Hao and Ma Long in the WTTC in 2009, 2011, and 2013. The scores were as follows:

Games(Points), Total Points/Difference
2009 - 4:1(10, 8, -10, 9, 9), 55:48/7
2011 - 4:2(7, -9, -7, 10, 7, 9), 61:55/6
2013 - 4:2(9, 7, 3, -7, -12, 3), 63:47/16

The match in 2011 proved to be the closest, followed by the one in 2009, and lastly the one in 2013. The total points scored is more or less in line with that observation. However, the same game score of 4:2 in 2011 and 2013 didn't reflect the striking difference in intensity at all. ML led 2:1 in games before dropping the next 3 in 2011. WH led 3:0 in games before dropping the next 2, then ML gave up in the 6th in 2013. Would you rate both players the same way in both matches just because the game score happened to be the same?
Or it could be a bad thing.

Like I said in ERT's thread, I don't have a bone in this fight. I only want to make it clear what the consequence are of using games won and lost to adjust the scoring. Now you bring up points won or lost. There are consequences to that too.

I have tables for scoring by points too. Now what? Ratings adjustments by points won and lost would effectively eliminate the need to play games to 11. Just play to whomever gets to 33 points first or go back to winning 42 points ( 2/3 games to 21 ). People with weird styles would like that because they could get an early lead while the opponent adjusted. As it stands now you can afford the lose the first game without a penalty except being behind by a game in the match.

Again, I just want to show the consequences of taking games won or lost.
So far it seem most people are against using games won or lost to affect the ratings.

So the question zeio asks is if adjusting ratings by point won or lost is even better would you like the change?
My vote is I don't care.
What do others say about adjusting ratings by points won or lost?
 
says Shoo...nothing to see here. - zeio
says Shoo...nothing to see here. - zeio
Well-Known Member
Jan 2018
7,742
9,644
19,211
The rules of table tennis dictate that whoever is 1st to 11 points takes the game, and 1st to 3 or 4 games wins the match. Personally, I don't see the point of taking the games and points into account.


A scenario where the winner scored less total points than the loser has no bearing on the outcome when it comes to elimination stage.


XX and MJ in team final at the Rio Olympics:
2:3(-10, -9, 3, 7, -10), 51:45/6


But then, there is also a scenario where the games(and potentially points) scored come into play on the final position. It happens quite often in group stage for singles and team event.


ML and CCY in group stage at World Cup 2012:
0:4(-9, -10, -7, -9), 35:45/10


That loss effectively put ML's fate into CCY's hand. For him to advance as the 2nd in group, CCY needed to lose 3:4 to Crisan in his last group match, which would've created a 3-way tie among CCY, ML, and Crisan. As long as CCY lost 0:4, 1:4 or 2:4, ML would've been eliminated. CCY didn't even have to try and he could send ML home. But the impossible happened, CCY lost 3:4.


How would one rate that? If one wants to be the most accurate, as ERT claims. Do we have separate algorithms for groups and main draw?


In the end, such rating system should not only be able to accurately assess the strength of players, but also reflect the nature of the game(styles/matchup etc.) as Marcus pointed out. Using standard deviation to represent the range of player strength is about the closest one can be.
 
Last edited:
  • Like
Reactions: thomas.pong
This user has no status.
This user has no status.
Member
Jan 2018
5
2
7
No we don't. Winner takes it all is one of the essences of sports, what makes it so beautiful. Also people always have different playstyles. You would have to look at everyone independently to figure out if he performed "bad" or "good" and even then it would be subjective.

Also, some matches don’t need to be played to max. To crush 3-0 your opponent. You need to control your stamina, don’t hurt you body by swinging to your max power in order to win 3-0 and give the opponent no chance. Why go 100% if it’s not a opponent that makes you give your 100% effort. You can lose a point here or there or even a set. And still win convincingly. Going all out is for life or death final matches, and last rounds of knockouts. imo


Sent from my iPhone using Tapatalk
 
  • Like
Reactions: thomas.pong
Top