Sunday, 11 January 2015

Introduction

Here is my attempt to set up an up to date, ELO rating system for ranking ATP and WTA level Singles Tennis players.

I started this off by putting every player at 1500, and started off by using the Grand Slam tournaments from last year to calibrate the system. After each round the winner and loser of a match get a new ranking based on their result and the difference between the ranking of the players. Stronger players gain less points for winning, while weaker players gain more points. The average rating for all players is maintained at 1500.

As I am calculating this all manually and not automating any of it, there may be human error in some of the ratings, my ratings will be simplified and like any ranking system, there are some limitations to this system. But I think this system does a good job overall of ranking players, as the top players get to the top, and I think most of the ATP/WTA regulars are approximately in the correct order.

First, what I believe are limitations to the rankings.
  1. All new players to the ATP and WTA tour will start at 1500. As 1/2 of all players lose 1 match and thus lose rating points, this average level is above the median ranking, and so new players are going to be over-ranked, and will likely stay over-ranked as they may end up playing only 1 top level match in a year. (2016 update: I now have a system that only adds players to my rankings after they've played a certain amount, so this limitation has been dealt with)
  2. The player at the bottom of the rankings won't be the weakest tennis players in top level tournaments, just the ones who lost ATP/WTA matches most often.
  3. Right now, I believe that the top players are under-rated, due to the small number of tournaments entered in to the system so far. This should fix itself by the Australian Open. (Mid-season update: It took until the middle of the season for this to happen)
  4. Injured and inactive players maintain their ranking, even if they never come back to play again. It's impossible to tell what level an injured player comes back to, so their rankings are likely to be inaccurate for a while.
Now, for my limitations on the technical side.
  1. My ELO calculations are rounded to the whole number after each match. This loses a bit of accuracy but over time should make little difference.
  2. Strength of victory is not accounted for. I do not know of an easy way to give more points to those who win strongly without creating a system that ends up negating the disadvantage that top players have when they are drawn (and beat) obviously weaker players.
  3. The K-factor is 20 for all matches. I could use a different K-factor for Grand Slams than I do for regular tour matches, but I am not convinced that the increased gain and losses for players when doing this will improve the rankings of these players.
I expect to publish Women's ratings every Sunday, with Men's ratings some time later (as Men's matches are often scheduled after Women's matches).

6 comments:

  1. Great initiative.. Will look forward to tracking the elo ratings

    ReplyDelete
  2. Great initiative.. Will look forward to tracking the elo ratings

    ReplyDelete
  3. Do you count Challenger tournaments? How far back does it go? Also saw another site.

    http://tennisabstract.com/reports/atp_elo_ratings.html

    ReplyDelete
    Replies
    1. Hi. I'm not tracking challenger tournaments, as doing do would require me to work out how much they're worth compared to the main tour, and it would at least double both my weekly work and the number of players on my already large database.

      I started these ratings at the start of 2015, but they weren't very accurate back then. I'd estimate that ratings from the end of 2015 are reasonably accurate for estimating player strength.

      I've noticed that site myself, as a stats nerd, I'm all for anyone who wants to make their own tennis statistics and to share the with the world.

      Delete
  4. Hey, thanks for publishing this and staying after it. I'm glad you do both the men's and women's tour, as I'm a fan of both.

    If you need help automating the calculations, or ingesting a bunch of historical data, follow up with me and I'm happy to lend a hand in writing some code (as long as you're happy to keep plugging in match results and posting the outcomes).

    Best of luck,
    Steve

    ReplyDelete
  5. Hey, thanks for publishing this and staying after it. I'm glad you do both the men's and women's tour, as I'm a fan of both.

    If you need help automating the calculations, or ingesting a bunch of historical data, follow up with me and I'm happy to lend a hand in writing some code (as long as you're happy to keep plugging in match results and posting the outcomes).

    Best of luck,
    Steve

    ReplyDelete