I have a suggestion that would incentivize the players at the top of the leader boards to play more matches against themselves, and play less on alts. Use the lowest value of the 95% confidence interval vs. the players rating, (the bottom of the error bar on the graph). Note: decay is completely different than deviation and is not factored at all into the rating calculations.
Why this will discourage “sniping”
There are currently two ways deviation will increase. Playing lots of matches against players that are rated much lower than you, and going inactive for a period of time. The advantage of getting a high deviation when ladder position is based only on rating is you can get much bigger gains than your peers for a win. If you combine that with queue dodging and playing off hours you have good odds of getting substantial rating increases for little risk. In the example plot, there are two scenarios of players that started at a rating of 2100 and deviation of 70. The blue player played 10 matches with a 60% win rate against players with a rating of 1950 and deviation of 70. The red player played 10 matches with a 90% win rate after they went inactive for a week to let their deviation grow to 150. As you can see the current system favors the person that takes time off and plays against lesser opponents. However, if you look at the bottom of the error bar after 10 matches the blue player is ahead. In fact the red player has to win 8 matches in a row just to catch up to the blue player even though he has a higher overall rating.
Why this encourages more unique players at the top
When you are at the upper end of the rating curve you run out of players above your rating. If you play a large number of games at a 50% win rate you end up with a deviation of around 60. Once you get above a 50% win rate, and run out of players above you, your deviation will increase. That means when you lose, you will end up losing a lot more points than you gain for a win. The bigger the gap between your rating and the opponents, the bigger your deviation will become. I attached a table comparing different win percentages over 344 games played against an opponent with an 1850 rating. It shows an approximate win % you would need to sustain a given gap over your opponents. If you have a 300 point differential, you will need to be able to sustain an ~90% win rate and won’t be able to get your deviation below ~78. For a 125 point differential you need a 70% win rate with a min deviation of ~64. However, if you queue with people near your rating, you will be getting the same amount of points for a win or a loss, and your deviation will be right around 60.
With the loss of the short term benefits of high deviation, the top players are better of ensuring they are always queuing when other top players are on. That way they won’t lose as many points for a loss, and they will be able to keep a lower deviation.
Here is a dump of the current NA rankings. As you can see if every player in the top 10 had multiple alts, it would be very difficult for top tier players to get matches where their opponents had a rating close to theirs.
Rank Rating
1 2336
2 2223
5 2061
10 2008
25 1951
50 1865
150 1809
250 1794
TLDR: If you use (Rating – 2*Deviation) for the ladders vs. just rating and keep a high rate of increase for deviation, it will make rating “sniping” ineffective and will incentivize all players to play more matches. If you don’t increase deviation for inactivity, you have ELO where you can camp at the top once you get there.
