The new ELO-based ranking system

[airwalker]

@MrRoboto Mr. Roboto, this is excellent work! It’s actually quite perfect timing, I created my own spreadsheet a few days ago for Revised players on TripleA. The one thing I can’t seem to wrap my head around is setting an ELO and then updating it without creating a circular reference. If you could help this I would be eternally grateful! My spreadsheet is here: https://bit.ly/revisedstats

Thanks!

[gamerman01]

The dwarves are digging deep, all the way to 9/30/17 so far…

Watch as the life-time ELO rankings take shape…
Or wait until all is entered and edit checks done

[gamerman01]

@airwalker

I just saw that your post was queued for approval, and approved it

[airwalker]

@gamerman01 Thank you

[oysteilo]

From the first page:

“A win by the current #1 against the current last place will award only meager 4 points for the winner and -2 for the loser.
However, the last player would receive a whopping 136 for a win and #1 would suffer -87 for that loss!”

Maybe this has been adressed before and. I am just curious why this is fair.

In chess, the winner and the loser gain/lose the same number of points whereas here the winner always gain more than the loser lose. Why is it this way? This means that everyones rating will increase over time given the number of players are constant? How is this fair to new players?

I am sure I am missing something!

[oysteilo]

I say everyones rating will increase. I mean “everyone” or most player.

[gamerman01]

The loser’s ELO drops more than the winner’s gains if the winner’s ELO is higher than the loser’s

I think you can access the “results” tab
Type in a player’s name in the yellow box and I think you’ll be able to see

It’ll be more efficient if @MrRoboto answers the rest

[MrRoboto]

We have 4489 games finished now.

The Average rating is 1486,77
The Median rating is 1459.

So as you can see, everyones rating hasn’t increased and a brand new player with 0-0 is even better than average ;-)

[Arthur Bomber Harris]

Something is wrong if the average player is worse than the new player since this has been incorrect from personal experience.

I looked through the data to see the apparent ELO of new players and they have been averaging around 1360 meaning they beat people ranked above this number the same frequency as they lost to people ranked below this number in their first game. I excluded the first few years of data as everyone was fresh in the League. Not the most rigorous mathematical calculation and I am sure you could be more accurate in the numbers.

Unfortunately this is a bit of a recursive calculation if you wanted to do this properly; adding in new people at ELO=1360 will lower the ELO of the entire community meaning you have to again adjust the ELO of newbies. In the end you probably will get around 1330 as the best estimate of a fresh person joining the League, but I would be interested if you could do a more thorough evaluation.

[MrRoboto]

You are correct of course.

I meant someone who joins the league is 1500, before finishing a game and therefore on paper better than average.
But most people seem to start with losses. I could get the correct data for that (and might find out that I am wrong with that hypothesis) but frankly am too lazy so a rough estimate is looking at people who currently have exactly very few games finished.

Out of 37 players who have completed a single game, only 8 have won that single game while 29 have lost it.

Out of 16 players who have completed exactly two games, NONE has won both and only 6 of them went 1-1 while 10 have 0-2.

Out of 20 players who have completed exactly three games, two have won all 3, 6 players have won 2 out of 3, 4 went 1-2 and 8 out of 15 went 0-3

So we have 71 players with 1-3 completed games and only 14 out of 71 have a rating of 1500 or higher.

I suppose it’s safe to say that new players tend to be worse than average - which shouldn’t be surprising.

[Arthur Bomber Harris]

Do you want to adjust the starting value of ELO down a bit in your spreadsheet so there isn’t incentive to pick on noobs to increase ranking? Somewhere in the mid-1300s would be more appropriate?

I doubt there are many stellar players who are joining the League and will dominate veterans during their initial matches. Gameplay in this forum is so much higher than the casual gamer who thinks OOB is actually balanced sans bid.

[MrRoboto]

This wouldn’t have any effect besides everyones rating going down.
In fact, it took me merely 5 seconds to change the starting ELO to 1300 and the result is:

Average rating is 1286
Median rating is 1259.

As you can see, the average / median is exactly 200 lower than before…

[MrRoboto]

Besides, we WANT to encourage players to take on fresh new recruits joining this community.

We even discussed a flat bonus for playing against a new player (with new being defined as having a maximum of 3 completed games), but ultimately decided against it.

I don’t have an issue with someone who tries “farming” new players. Yes, chances are these new players are overrated at 1500 and should on average be much lower, but remember: You DO risk facing an undiscovered elite player who could be a 2000 in disguise. Losing to that strong player would also give lots of negative rating since he / she is massively underrated at 1500 as well.

[gamerman01]

I have a very similar concern as ABH after entering a couple thousand games.
I really appreciate the analysis (that MrRoboto quickly did) on players with 1-3 games complete (we currently have it set so the sensitivity of the first 3 games is higher). That they significantly underperform.
Correct, the solution using this system is not to change the starting ELO number, and thanks for verifying that 100%, MrRoboto

I tend to think maybe it’s not too big a deal for us who keep playing, because you play a few more people who aren’t unknown and your rating starts to correct. I’ve been busy entering games (and I’m deep into 2012 when things were just starting up, so I’m super close), so I haven’t looked hard at how fast a player would correct after bottom feeding a little.

I wouldn’t be surprised if MrRoboto thinks of some adjustment, though one is not necessarily needed IMO

We do have a sensitivity scale in place that we haven’t yet really played with or discussed yet. That’s another discussion. Currently we’re most interested in getting a new player up to an appropriate ELO quickly (like within about 6 games) so that a good player would not be significantly under-rated when entering the playoffs in year 1. I suppose this relatively higher sensitivity in the first 3 games would also make a significantly below average player drop faster. So it’s hard to feast on the new guys given we don’t have too many new ones come in, in a given year, and if their rating quickly drops from game 1, 2, 3

Not my clearest writing, there, my apologies. I’m tired, but I wanted to respond

[Arthur Bomber Harris]

@MrRoboto I had a cutoff date for looking at noobs, grandfathering in people who started the earlier years with a higher elo rating at start since most of the talent had been here for such a long time.

Newer players who began more recently had a lower assigned starting ELO since we appear to be attracting less talented people now instead of hardcore G40 whizkids. It isn’t the fairest of systems but does match the reality of the situation.

[oysteilo]

I still struggle here.

These are the numbers at the bottom of the “data” spreadsheet and show how much rating has changed depending on prior rating and who actually won the game. For the five first entries the increase in rating and decrease in rating is the same. To me this makes sense and the relative change is given by difference in original rating and who won the game.

Some of the other entries seem more random (I am sure there is a reason though). In one of the games the winner gained 58 points whereas the loser lost 37 points. How does this relate to the first 5 entries? I think this is important to understand. It is not obvious to me. Please explain.

[oysteilo]

I am making different posts. Here is another one… It is mostly just an observation and not any criticism. Using myself as an example I am currently rated no 3 in the elo system for the OOB playoffs starting 01.01.2024 as I have finished the required 3 game cap.

In gamerman01*s spreadsheet I am currently tied for 8 th place and it is give or take if I will qualify for the top bracket or not

This tells me that my results in 2022 (and maybe in 2021) stil count considerably towards my current elo rating.

This might not be a problem, but it also may indicate a higher barrier for new players to qualify for the top bracket. I know there has been som discussions around how to place new players rapidly.

Maybe, it is appropiate that the designing crew give a highlight of how the elo system works both with new players vs old players and how this relate to the 3 game cap and prior years influence. Also see my post above about about the sum of gain/loss not always equals zero

Finally, again this is not negative thoughts. I just dont fully understand how everything comes together and would like to understand better before the change becomes active on 01.01.2024

[MrRoboto]

I will address everything tomorrow. Today my whole family comes over for christmas.

Speaking of christmas:

Merry christmas to everyone celebrating it! And happy holidays to everyone else.

[AndrewAAGamer]

I thought we agreed, in earlier discussions, that we would have one historic rating for fun and interest and one current rating for each new year? That way this year’s playoffs are only based on games played this year.

Using data from before the current playoff year makes the system more like golf or tennis were Players are rated and seeded accordingly based on a rolling average that dates back 2 years (Golf) or one year (Tennis). Except our rolling average would date back to whenever the player started to play. So good play over time is going to result in higher playoff seedings versus someone who had medium or bad play over time even though both players had the exact same playing experience in the current year.

I would prefer a system more like football, basketball, baseball and hockey that starts everyone off at zero for the new year. The worst team in the league and the best team in the league from the previous year all start over even in the new current year.

My two cents…

[gamerman01]

Very well said, happy to have 2 respected players weigh in.
I’m going to wait until MrRoboto can answer some of these (I don’t know why ELO changes are the same for victory and defeated after game 1)

Part of the answer is that at this point the sensitivity ratings were somewhat arbitrarily put out there by MrRoboto, and I somewhat arbitrarily changed them to see what would happen.
That is, on the data sheet to the right, we currently have 110, 90, 70, 50 at threshholds 3 games, 6 games, 10 games (3 and 6 of course have become accepted league threshholds, 3 for getting a firm tier, 6 for qualifying for BM)

The matter of a new player getting a reasonable seat at the playoff table after finishing minimum # of games for his very first year is fairly easy to address, I think.

You (plural) raised other points and every sentence is appreciated. I do not consider this post here a complete answer, but some thoughts in response to some of your thoughts that I think will help advance the discussion.