well good luck with that

 

Great idea. I've toyed with the idea too, but it's too much work.

Please post the results when you're done!

 

Well, calculating points is not to much work, but maintaining the whole database could prove to be demanding.
In the meantime, I would like to discuss some technical/mathematical issues regarding calculations, as well as some practical matter.
For instance, maintaining "career" ranking list, 52-week list, and even year-to-day (ATP/WTA race to championship) list?

 

There were some attempts before, but that was only an one time effort...
http://tennisplanet.wordpress.com/2...2010/06/20/chess-elo-rating-system-adapted-to-men’s-tennis-from-rafafan-thanks/

At that time, ranking of players was (top 10, at least):
1 Roger,Federer,SUI 2007.7 2890.2
2 John,McEnroe,USA 1985.3 2835.6
3 Ivan,Lendl,USA 1990.1 2814.3
4 Rafael,Nadal,ESP 2009.4 2799.6
5 Bjorn,Borg,SWE 1980.5 2771.4
6 Jimmy,Connors,USA 1980.1 2665.1
7 Novak,Djokovic,SRB 2009.9 2619.4
8 Pete,Sampras,USA 1997.2 2609.6
9 Andy,Murray,GBR 2010.1 2584.2
10 Rod,Laver,AUS 1975.2 2567.8
...

The first number is year.month when the peak points were achieved, the second is the actual peak rating points.
Compared to chess players:
1. Kasparov, Gary……………… g RUS 2851 2851
2. Anand, Viswanathan………….. g IND 2771 2769
3. Kramnik, Vladimir…………… g RUS 2760 2758
4. Shirov, Alexei……………… g ESP 2734 2751
5. Morozevich, Alexander……….. g RUS 2758 2748
6. Leko, Peter………………… g HUN 2701 2725
7. Adams, Michael……………… g ENG 2708 2715
8. Ivanchuk, Vassily…………… g UKR 2702 2709
9. Bareev, Evgeny……………… g RUS 2698 2709
10. Topalov, Veselin……………. g BUL 2690 2702

 

The first number is year.month when the peak points were achieved, the second is the actual peak rating points.
Compared to chess players:
1. Kasparov, Gary……………… g RUS 2851 2851
2. Anand, Viswanathan………….. g IND 2771 2769
3. Kramnik, Vladimir…………… g RUS 2760 2758
4. Shirov, Alexei……………… g ESP 2734 2751
5. Morozevich, Alexander……….. g RUS 2758 2748
6. Leko, Peter………………… g HUN 2701 2725
7. Adams, Michael……………… g ENG 2708 2715
8. Ivanchuk, Vassily…………… g UKR 2702 2709
9. Bareev, Evgeny……………… g RUS 2698 2709
10. Topalov, Veselin……………. g BUL 2690 2702

Great idea, really, but what kind of list is the one above? They are not present numbers (Magnus Carlsen tops with 2843) and they are not historical highs either, except for Kasparov. Anand and Kramnik has both been above 2800. Ivanchuk has been over 2780, at least.

This is the present live chess list:
http://www.2700chess.com/
and the last officially publicized one:
http://ratings.fide.com/top.phtml?list=men

 

I had an error in formulas, some values were permuted. I corrected it, now calculations work as suggested. Thanks Litotes!

 

Amazingly, today I was thinking of doing the same thing myself.
My thinking is to treat each set separately, so a 2:1 is like a result of 0.666, but the value is doubled by the number of sets * K.
K is the constant of the system. which will be decided by what gives the best predictability to the rankings.

 

One possible concern in this methodology is would it cause an "inflation" of points over the time.

If we take a match result in sets, then a question is how to treat results in women's tennis, since all matches are played in "best of 2 sets". Should they be treated completely separately, or use criteria for male tennis? That is, treat win in 2 sets as 1, win 1 set = 0.5, or 2 sets = 0.66. The second approach would lead to significantly lower maximal rating for female players, and I think that is not fair.

There is also a matter of selecting a proper K constant. For my calculations I initially put this constant to 24, but there are some systems in chess which use different values for K depending of strength range, for instance higher K for rating below 2000, and lower K above 2000. K constant effectively determines the maximum difference in points that can result as an outcome of a single match.

 

I was also thinking of using the Glicko2 ranking system, which is an improvement of ELO, and creating different set of rankings for each surface, otherwise the rankings will be biased by the surface played lately...

 

I looked up the Glicko system a bit... If the Elo ranking list is not used as an official ranking, for instance as a base for tournament seeding, then using Glicko is a bit of an overkill, don't you think?!

 

One interesting thing I discovered...
Points in Elo system correlate well to ATP/WTA points as:
<Elo points> ~ 300*loge(<ATP/WTA points>)

 

Here is a good KRACH based ranking, developed by a statistican named Jeff Sackmann:
http://tennisabstract.com/jrank/atp.html
(the owner of the blog I gave before)

 

I've tried in the past to experiment with different ranking algorithms, but I will be developing my own when I get the time. Just for fun.

 

I'll be giving my version of Elo rating system a try starting with the Abu Dhabi tournament. Although this is considered an exhibition tournament, its competitiveness is real, and so IMO is the meaning of the results. I decided to start with a "race rank" - starting with clear table for the season for all players, since I don't have the resources (time, mostly) to implement the full historical Elo ranking. I'm going to include all really competitive results - ATP tournaments not lower then 250-series, Davis Cup singles matches, Grand Slams, World Team Cup singles matches, Hopman Cup, and other serious exhibition tournaments like Abu Dhabi (suggestions are welcome).

The method I'll be using:
- every player starts with a clean sheet, 0 points;
- since in Elo rating only the difference in points is significant, and negative values of points are allowed in calculation, I will not introduce any starting bias;
- so, players who loose more often then not, will go into negative domain, while those who win will go into positive domain;
- I will be using a fixed K-factor of 24;
- I will normalize all results to the "best of two sets", meaning - 2:0 win in sets will be normalized to 1:0, and 2:1 win will be normalized to 1:0.5;
- consequently, matches in "best of three sets" will be normalized as - win of 3:0 to 1.5:0, 3:2 to 1.5:1;
- that way, matches in "best of three sets" (Slams, DC) will have a higher weight - effectively as if the K-factor was 36;
- and lastly, season will end with World Tour Championships.

So much for the start, and I'll see where this leads in the process... different tables for different surfaces, other possibilities for evaluating results, etc... Elo system is really very flexible, anything that maps to 0-1 range can be used for calculations: serve and return percentages, games won, points won, you name it!
I'm not a statistician by profession, so bare with me...

 

You can't add exho results.

 

Yes I can, I can do whatever pleases me with my list!

 

Seriously, the only measure of value for me is are the players motivated enough for a real competition. I will not count exhibition matches whose sole purpose is entertaining spectators. On the other hand, there is no reason not to include matches from tournaments like Abu Dhabi.

 

I already posted ELO Ratings including historical results:
http://www.menstennisforums.com/showthread.php?t=211663
I also tested it to make very them very predictive.

Feel free to make your own version if you wish, though...

 

I already posted ELO Ratings including historical results:
http://www.menstennisforums.com/showthread.php?t=211663
I also tested it to make very them very predictive.

Feel free to make your own version if you wish, though...

Thanks, Amirbachar, I'm aware of your work, and respect it. But my approach is different, especially (as I understood) regarding which matches are accountable. And it does't hurt to have multiple views at the same thing.

As I said, I'm not a statistician, and I don't pretend to the practical usefulness of the data. This is purely for informational/entertaining purposes.

 

Elo ranking system - player "race" rank

The list is moved, and can be found in my last post.

 

Another issue surfaced in the meantime that I had to address, regarding how to treat results of match that ended with the retirement of one of the players. My aim with this ranking is to take into account only what is really played (which IMO is in the spirit of the Elo rating system), and my atomic unit is a set, so if a player retired without completing a set, in my system I will award that set to his opponent, regardless of what was the result in that set at the moment of retirement.

Examples:
Today Stepanek retired against Benneteau at 0-0, 15/40, so, for the purpose of Elo rating I will record this match as 1-0 for Benneteau.
Yesterday G. Zemlja retired against A. Falla at 6-4, 3-1 for Falla, so this match is recorded as 2-0 for Falla.
And, if for example a player won the first set, and then retired in the second, I will record this match as 1-1.
You get the picture...

 

Weekly update ready, after 3 more events - Auckland, Sydney and Kooyong exhibition.
http://www.menstennisforums.com/showpost.php?p=12637457&postcount=32

 

good!

 

Update after the AO.
http://www.menstennisforums.com/showpost.php?p=12637457&postcount=32

 

This update comes rather late, but Davis Cup results turned out to be a bit challenging, bringing out new bugs in my programs.
Included are the results from Davis Cup World Group matches, and Group I too. I didn't feel like going deeper then G1, since very few of the top 100 players may be found there. Required too much of my time, which I haven't had much these days.
http://www.menstennisforums.com/showpost.php?p=12637457&postcount=32

 

Update after Vina del Mar, Zagreb and Montpellier.
http://www.menstennisforums.com/showpost.php?p=12637457&postcount=32

 

I edit my post with an update, and the next day it is reverted to previous?!?
Then I have to edit it again!
And this is not the first time!

 

This update includes results from tournaments in Rotterdam, Sao Paulo and San Jose.
http://www.menstennisforums.com/showpost.php?p=12637457&postcount=32

 

Since I've decided not to use any bias value, interesting thing to notice is that so far the difference between the first and the last on the list has accumulated to nearly 260 points.
The best average points per match (of those who played at least 2 matches) has Novak Djokovic, 16 points per match.

 

An update after Marseille, Buenos Aires and Memphis.
David Ferrer replaced Novak Djokovic on the top spot, after having played 22 matches already this season.
http://www.menstennisforums.com/showpost.php?p=12637457&postcount=32