Scrabble for Beginnerds 101: Ratings, The Elo System

1 2 3 4 5
  • You read that right, it isn't a typo. This topic is for all of us Beginnerds! ;)

    First, you need a basic understanding of what Scrabble Beta bases their ratings upon. That would be the Elo Rating System. Arpad Elo created the system as a way to evenly match opponents in chess games. Extrapolations have been used for various other competitive games, including Scrabble.

    Just because you win a game, doesn't mean your rating should or will increase. If you are playing a weaker opponent who does well against you, you may lose rating points even though you won the game. The expected outcome is for the stronger player to win. If the stronger player has a weaker showing, the player who loses may even get a couple points added to his/her rating while taking a loss. Even if you win 5 games and only lose one game, your score may still drop depending on the strength of your opponents.

    What does this all mean? If you are obsessed with keeping your rating, like some players seem to be, play against players with a higher rating and don't worry about players with lower ratings than your own. Winning against players with a rating 100-200 points below your own will get you a W, but only add a point or two to your rating for the effort. Win a game against a player with a rating 100-200 points above your own, not only will you get a W, but you'll get more points added to your rating for he effort. The converse applies. Lose to a player with a higher rating, you won't lose much on your rating.

    The classroom is now open for discussion! ;)

    Like this post to subscribe to the topic.
  • Dude... you are a rock star. Thanks for the class. How quick and easy was that to learn about! Appreciate it!

  • Cory, I can't seem to process the gory details of that article right now. So maybe you can just clarify how a "weak showing" can affect your rating even if you win. Does that imply that the point spread is factored in? I always thought the point spread was irrelevant.

  • Thanks for the info Cory...I was just wondering why I am unable to see my friends stats anymore...when I first started in October I was able to see global and my friends stats but now I can only see global. I am definitely not obsessed with the stat thing but like to check it out once in a while to see how my opponents stand and how much of a stomping I expect to take...all in good fun though:-)

  • It depends on how exactly the rating system is implemented here. In general, the rating difference between you and your opponent corresponds to a probability of winning, with easy-to-remember points as follows:

    +400 = 91%

    +200 = 75%

    +120 = 67%

    equal = 50%

    -120 = 33%

    -200 = 25%

    -400 = 9%

    After a game, the rating system needs to figure out how much your rating changed. So, it subtracts the expected result (the probability above) from the result you got (1 for a win, 0.5 for a tie, 0 for a loss). Then it multiplies that difference by a constant K (seems to be 30 for this app), which basically determines how wildly ratings fluctuate. So, the formula is:

    new rating = old rating + [actual - expected] * K

    So, if you beat someone about 120 points below you, you can expect to gain about 10 rating points. Obviously, since your expected probability of winning will never be more than 1, you can never lose rating points by winning, though at some point the gain approaches zero. However...

    Under the old USCF system (which may be the current one since I haven't played tournament chess in many years), there was a provisional rating calculated on your first twenty tournament games. Then, your rating was averaged according to the ratings of your opponents. If you won a game, it factored your opponent's rating +400 into the average and if you lost it factored in your opponent's rating -400. So, if you were still in the provisional stage and outranked your opponent by > 400, you would actually see your rating drop even if you won.

    I haven't seen any compelling evidence that ratings on here go through a provisional stage, so winning shouldn't ever lower your rating. I also don't think the point spread in winning or losing should affect how your rating changes (and in regular tournament play it doesn't) but of course who knows exactly how the rating system has been implemented here?

  • Thank, Cory! This is the answer I was looking for. You da sailor!

  • Now, we must play.

  • i disagree with you cory, i am at 1800s

    and i play with my girlfriend often where she is at 1100s

    (+/- 700)

    and EVERY time i win, i earn two or three points

    folks, you will never lose points if you win

    you will never gain points if you lose

    win is win

    lose is lose

  • This is true but I am still glad to lose to someone that teaches me the finer points of Scrabble. Despite the ratings I still gain some knowledge. Thanks for the information!!

  • Dude. Awesome. Thanks!

  • This has been really interesting reading. Thanks for all the info.

    The only thing that makes me crazy about the ratings is when I fell below the 1200 point. That meant that all my friends who had started a game but never completed it, or those who played one game and never again were above me. I know it shouldn't really matter, but that just bugged me.

  • This is a valuable discussion...I'm just adding this blurb to get it to the top of the board again.

  • Valuable but contradictory - I would love to see the developers post a better description of the scoring here than can be found in the rules so we could clear it up.

  • So is the range from 0 to 9999, when creating a game, a result of lazy programming?

    I think it highly unlikely to ever see this feature in this EA online version, but I'd be curious to see what the distribution of players is on Scrabble, assuming the mean to be around, oh I don't know, 1400.

    I'm wondering also if, when I do create a game, the computer does limit the potential players to the range I've specified.

  • I understand how the ratings are supposed to work, but I am not sure the facebook app is calculating them correctly. By default, new users start out at 1200. I have played a few games with coworkers when I just started playing here. I won. They lost. Their rating went UP. Mine is down in the toilet and shows no signs of going anywhere no matter whom I defeat or lose to.

  • Theoretically, there is no upper bound to the rating. In practice, it seems to max out at slightly more than 2000. When they started programming the game, there would have been no way to know where the maximum rating would be. They must have felt they had to pick some number for the upper bound on ratings. My guess is that they chose the 9999 rating as a safe upper bound, in that it is very very unlikely than anyone will ever come close to having a rating this high or higher, but it isn't stupidly higher than the actual maximum (in contrast to say, something like 1,000,000).

    When I don't put restrictions on ratings for public games, I often get players who lie outside of my favoured range. When I do put restrictions on the game, I always get players within my favoured range. It could be just a happy coincidence, but I'm pretty sure that it's not.

  • i am stuck in a game with a 'v' and i dont know how to end the game. how?

  • If you won every single game against your coworkers, you only played against them, they only played against you, and your ratings have gone down while theirs have gone up, then it something is wrong.

    The ratings don't just give a rough-and-ready rule for determining who is better than who. Assuming you and your opponent(s) have all played a large number of games, you can actually calculate the probability of winning or losing against a given player based on your ratings (I don't know the exact formula). If you're playing against a bunch of people, you can use the ratings calculate the the number of games you are expected to win. If you win more than this, your rating goes up. If you win less than this, it goes down. The bigger the difference in the expected number of games won, the bigger the change in rating (again, I don't know the formula).

    If you won against your coworkers but lost against other people on the same day, then it's possible you won fewer games than the ratings predicted and thus your rating would go down. Similarly, if your coworkers lost against you but won against other people on the same day, then it's possible they won more games than the ratings predicted and thus their ratings would go up.

  • First of all, you shouldn't post this question in a thread that has nothing to do with the question you're asking.

  • thats real wise to not even answer the question wise ass

  • Tooo many Stephen Sharps...skip three times each and the game ends. 6 total.

  • thanks so much!!!! feel free to invite me to play a game with you if you can figure out how! lol

  • Knew this question would come up today! Lance, where are you???

  • Why am I not surprised that Mr. Sharp is located in DC?

  • Why am I not surprised that Lance is located in RI?

    Heck, why am I not surprised that Lance is whining AGAIN?

  • yea, thanks rob i was wondering what that was about

  • Hey Cory thanks for taking your time to compile this information succinctly and putting it out there for all of us. Truly altruistic!

  • DOes forfeiting cost your rating to go down more than jsut losing outright? If I'm losing dreadfully to someone (who I really believe is word generating) -- does it matter if I forfeit early on while 100 pts down, or wait until game ends, where I worse case would be about 100 pts down? Same question in reverse --if I'm winning by a lot and someone forfeits (by force or on their own), does my overall stat increase more or less than if they played the game out??

    If anyone knows, I would greatly appreciate the info. Thanks!

  • There is no provision in this rating system for margin of victory. A win is a win.

  • Are you sure, Art? At least some of the descriptions I've read have included a complex formula that includes the difference in the final scores -- and nothing on the SCRABBLE Beta web site seems to clarify this. It says it's "based on" ELO, but not that it IS that formula.

    This doesn't really matter a great deal -- I just wish the folks at SCRABBLE Beta would publish clearly how the ratings are determined. Just geeky enough so I'd enjoy it.

1 2 3 4 5