Saturday, July 26, 2014

Riot changed the matchmaking algorithm to address the blue side advantage - but also introduced a bug that place high ranking players on the purple side close to 70% of the time

Previously I made a bunch of posts describing the blue side advantage in League of Legends and how it is affected purely by camera angle. Apparently, purely because of how the camera is set up in the game, blue side wins 55% of the time. This was in response to several posts on reddit complaining about the issue.

In response to this, Riot seems to have changed something that dramatically reduced the blue side win rate from 55% to around 52%. Here's a plot of the day-by-day blue side win rate in North America Solo Queue around early June.


We see that the blue side win rate was around 55% until June 3rd to June 4th, where it sharply dipped to around 52%. This has been the case until the time this blog is written (July 24th).

How did this happen? What changed on June 4th that could have accounted for this? There was a patch on that day, but there was nothing to address the camera angle. As it turned out, Riot changed the matchmaking algorithm to put the stronger players on the purple side.

What evidence is there? Well, it turns out that after Riot's changes on June 4th, high elo players (i.e. Diamond and Challenger players) have a higher chance to play on purple side than on blue side. Prior to the change, a Challenger player has roughly 50% chance to play on blue and purple side. After Riot made the changes to put better players on the purple side, Challenger players now have ~58% chance to play on the purple side.


Making matters worse, the higher you are on ladder, the more likely you will be playing on the purple side. Take these players for example:

Vasilli, Sneaky (main and his smurf), Popstar Adrian, Meteos, and Mor

I have 340 games of these players on record since June 4th. Out of these 340 games, 68% of them had these top players playing on the purple side. I have 580 games of them prior to June 4th and they played on purple 49% of the time.

I think there are two consequences of this.

1. Riot essentially acknowledged that blue side has an advantage. Previously Riot was fairly mute on the blue side advantage, in particular to how it affected competitive play. This stealth change to matchmaking confirms that Riot does believe blue side has an advantage now. What consequences does this have for the competitive scene?
2. Riot most likely never had higher MMR teams on the purple side prior to June 4th, 2014. I often hear that Riot always put the slightly better players on the purple team. If this were the case, then prior to June 4th we should also see challenger players playing on purple more often - but unfortunately this was not the case and challenger players used to have 50% chance to play on blue or purple.

Saturday, July 19, 2014

Doom Bot statistics!

I got 50k Doom Bots games over the last 2 days on NA*. Here are some statistics below:


Average Win Rate for Players: 50.98%


Average Win Rate per League Tier**:




10 Most Popular Doom Bot Choices for Players
Name Win Rate Popularity
Ezreal 51.17% 18.66%
Lucian 56.60% 17.96%
Yasuo 53.72% 17.72%
Caitlyn 54.56% 16.70%
Ziggs 62.61% 14.77%
Master Yi 48.93% 14.69%
Jinx 51.35% 14.21%
Lux 57.36% 13.33%
Lee Sin 54.72% 10.66%
Tristana 50.03% 10.55%

10 Most Successful Doom Bot Slayers (with at least 1000 game records)

Name Win Rate Popularity
Ziggs 62.61% 14.77%
Kog'Maw 60.32% 4.11%
Galio 58.12% 4.05%
Lux 57.36% 13.33%
Jarvan IV 57.28% 2.71%
Riven 57.02% 6.83%
Lucian 56.60% 17.96%
Vel'Koz 56.07% 5.79%
Xerath 55.94% 5.79%
Talon 55.14% 3.73%

10 Least Successful Doom Bot Slayers (with at least 1000 games records)

Name Win Rate Popularity
Twitch 42.23% 2.77%
Garen 42.93% 2.83%
LeBlanc 43.44% 3.04%
Kassadin 44.00% 3.63%
Veigar 44.39% 3.80%
Cho'Gath 44.45% 2.29%
Fiddlesticks 44.56% 2.14%
Shaco 45.57% 2.15%
Ashe 45.58% 4.68%
Dr. Mundo 45.83% 2.55%

* "Doom Bot" refers to Doom Bot of any level (1, 2, or 5 bombs). I cannot distinguish between the levels. You can blame Riot for that. Unfortunately there isn't much I can do about it. lolking.com has the same problem.

** Challenger players won 76% of the time. But I only got 76 challenger players/games on record, so the sample size is awfully small.

Why Riot Won't Implement Features to "Fix" 4v5 problems in LoL

Recently I have seen a lot of reddit posts regarding "fixing" 4v5 in LoL. This usually includes a mixture of earlier surrender, easier surrender, and less LP loss. It is my opinion that Riot will never implement these kind of features because:

1. The chance of someone AFK the entire game is actually extremely small (1 in 1000 per game per player), so it's a rare event not worth addressing.
2. Implementing such features will make the AFK problem worse because the players then would give up more easily.

First let's take a look at what are the chances a player will actually AFK for an entire game. To be clear, I am talking about a player who AFK from the beginning of the game and does nothing all game, not someone who plays the game a little bit, then toxicity kicks in and decides to not play anymore (that's an entirely different beast of its own). This means this player will not have any CS or items when the game ends. To this end, I found out that a player has only a 0.1% chance to AFK an entire ranked game over a 11 million sample size over the last 3 months.

What is 0.1% chance? Just to give you some perspective, the chance of a player performing a pentakill in a game is 0.13%. This also implies there is a 0.4% chance that such AFK player will be on your team, or an average of 1 in 250 games. This is an extremely rare event.

Now you may argue that I am talking about "AFKing" in a narrow sense - the player must end the game with 0 CS and 0 items. This is a valid argument, but it also brings to my next point - that a player who is AFK at the beginning of the game almost always come back, so if Riot implements an early/easier surrender mechanics it's actually going to make the AFK problem worse.

Probably all players know that 4v5 is a uphill battle - and this is the correct assumption considering an AFK player has only about a 10% chance to winning the game - so it's hot hard to imagine most players want to give up when faced with such situation. However, if surrender is earlier, the problem is made worse because the AFK player will have even shorter time window to rejoin the game and make the difference, so the likelihood of a player being AFK will actually increase.

The same can be said about less LP loss. A lot of 4v5 games end really quickly (less than 10 minutes), because the side with 4 players simply decides to give up. If LP loss is even less with an AFK player on the side, it will make this phenomenon even more frequent. Again, this is unfair for the player who may still be trying to connect.

Finally, I want to mention that accidents can happen and a player may not be able to join the game immediately due to a variety of reasons. Even high ranking challenger players may occasionally fail to join a game. When a player did not join at the beginning of the game, the said player should be given plenty of opportunity to rejoin the game.

Attached: the list of 11006 players who went AFK, with their league rank and whether or not they won.

https://gist.githubusercontent.com/anonymous/fddc87553acaedd5c1e0/raw/52f201b0eca5252cc2bf82ec105f905468f2d3f4/gistfile1.txt

Wednesday, June 4, 2014

Some Data on One For All Mirror Mode (popularity, average game length per champion)

Thank god I didn't play Karthus in mirror mode.... Zzzz.....

Champion Popularity, One For All Mirror Mode


Average Game Length, most popular 50 champions


Average Game Length


These figures are made using over 80k One For All Mirror Mode games collected over the last a few days.

Minor rant: I was hoping to be able to mine more One For All games, but with the Riot's server lagging and the lack of popularity of this mode, I was only able to mine so much. Alas. 

Saturday, May 31, 2014

Blue Side Advantage Part 2

I feel that there needs to be a continuation on the topic of blue side advantage due to two things readers brought up:

1. Whether or not the blue side advantage is affected by players' skills; e.g. perhaps Diamond/Challenger players are less likely to be affected by this advantage, and by extension, in the competitive scene the blue side advantage is minimal.
2. Whether or not this effect depends on the champion you play. I hope the previous post convinces you that the blue side advantage is mostly due to camera angle (since One For All Mirror Mode removes all other factors); so there is a possibility that champions with long range skillshots are affected by blue side advantage more than champions with short range.

My answers to these two questions are:

1. Blue side advantage is affecting high level players (Diamond/Challenger) just as much as low level players (Bronze). 
2. There are some differences between champions and how they are affected by the blue side advantage. Short range champions are slightly less likely to be affected by the blue side advantage. Rumble, in particular, is affected by this so much that he wins 57% of the time on blue side and only 41% on purple side.


The data and the analysis:

1. Blue side advantage vs players skill.

For this, I pulled roughly 1.8 million games played in patch 4.6 and 4.7, all of which are ranked. The results are as below:

   Blue Side Win % Purple Side Win % Difference Total Win % Games Recorded
Challenger 59.75% 49.35% 10.41% 54.59% 17299
Diamond 56.53% 46.92% 9.61% 51.73% 1035553
Platinum 55.88% 46.32% 9.55% 51.11% 2756535
Gold 55.94% 45.98% 9.96% 50.96% 5656094
Silver 55.11% 44.44% 10.67% 49.77% 7744235
Bronze 53.10% 41.27% 11.83% 47.18% 2675229
First of all, if we only look at the chance of winning on blue side, it seems that at nearly 60%, Challenger players are even more heavily affected by the blue side advantage. However, a closer examination shows that Challenger players are so good that they win 55% of the time overall, so their blue side win rate is inflated. To counteract this, I take the difference of the blue side win rate and the purple win rate. This gives us a roughly 10% difference in win rate for blue side vs. purple side for all league players. We will be using this number to examine the size of the blue side advantage.

Also, although this chart does not show this, Challenger players have equal chance to play on blue side or on purple side. 

Therefore, we see that blue side advantage decreases very slightly as the players' skill increases. However, blue side still wins roughly 10% more often than the purple side - which is roughly the same as saying blue wins 55% of the time.

(Comment: sorry if you are a statistics buff. I chose to not use relative risk or odds ratio because I feel risk difference is far easier to interpret)

2. Blue side advantage vs champion

This time, instead of having one row per league tier, I plotted the table to have one row per champion. I only chose champions that had at least 50,000 games to avoid looking at the potentially misleading numbers that may be produced by champions that are rarely played (yes you, Urgot).

Similar to our analysis before, we are going to look at the difference in win rate between blue and purple side - because some champions are so strong that they win ~55% of the time overall, similar to Challenger players in ranked. I sorted this table so the champions with the greatest blue side advantage are on top, and the ones with least blue side advantage are on the botttom.

Champion Blue Side Win % Purple Side Win % Difference Total Win % Games Recorded
Rumble 57.34% 41.71% 15.63% 49.54% 53239
Jayce 54.91% 43.35% 11.56% 49.14% 90014
Shaco 56.77% 45.23% 11.54% 50.92% 93684
Olaf 49.93% 38.43% 11.50% 44.14% 59337
Teemo 57.10% 45.72% 11.38% 51.34% 136910
Veigar 52.42% 41.09% 11.33% 46.71% 57354
Velkoz 56.24% 44.97% 11.27% 50.62% 79882
Brand 57.35% 46.10% 11.25% 51.76% 76920
Lux 56.18% 44.94% 11.24% 50.59% 224541
Malzahar 58.26% 47.06% 11.20% 52.60% 86484
Nidalee 55.58% 44.42% 11.16% 50.09% 392071
Draven 56.93% 45.78% 11.15% 51.46% 186644
Ahri 54.16% 43.02% 11.14% 48.61% 119034
Chogath 52.00% 40.88% 11.12% 46.42% 68512
Varus 55.42% 44.39% 11.03% 49.92% 140731
Lissandra 52.74% 41.75% 10.99% 47.18% 54384
Ziggs 58.70% 47.76% 10.94% 53.26% 256552
Volibear 55.23% 44.30% 10.93% 49.72% 119082
Amumu 58.87% 47.95% 10.93% 53.32% 141434
Syndra 54.74% 43.84% 10.91% 49.30% 150155
Elise 50.20% 39.30% 10.90% 44.68% 103848
Singed 56.71% 45.82% 10.89% 51.24% 59682
Talon 57.47% 46.59% 10.88% 51.96% 74872
Morgana 57.71% 46.84% 10.87% 52.23% 648008
Sivir 54.84% 43.98% 10.86% 49.26% 159978
Heimerdinger 58.39% 47.57% 10.82% 52.95% 70838
Hecarim 52.35% 41.54% 10.81% 46.91% 57400
Jinx 57.07% 46.29% 10.78% 51.73% 440826
Blitzcrank 56.18% 45.42% 10.76% 50.85% 248811
Aatrox 55.13% 44.41% 10.72% 49.69% 74615
Karma 53.86% 43.14% 10.71% 48.51% 239518
Caitlyn 55.53% 44.84% 10.69% 50.17% 709054
LeeSin 51.97% 41.29% 10.68% 46.62% 490862
Trundle 53.94% 43.26% 10.68% 48.55% 118351
DrMundo 51.78% 41.16% 10.62% 46.47% 87591
Rengar 54.32% 43.73% 10.60% 49.03% 97593
Nasus 55.41% 44.81% 10.59% 50.10% 233418
Shen 50.55% 39.96% 10.59% 45.29% 87704
TwistedFate 53.22% 42.64% 10.58% 47.93% 125955
XinZhao 57.88% 47.35% 10.53% 52.63% 318665
MissFortune 58.97% 48.44% 10.53% 53.68% 149226
Yasuo 52.39% 41.87% 10.52% 47.12% 242814
Annie 55.05% 44.54% 10.51% 49.75% 207493
Vi 54.28% 43.78% 10.50% 48.98% 258399
Darius 54.72% 44.22% 10.49% 49.43% 143617
Thresh 53.66% 43.18% 10.48% 48.53% 747150
Vayne 50.83% 40.35% 10.48% 45.59% 379347
Pantheon 54.80% 44.33% 10.47% 49.52% 274628
Graves 55.09% 44.62% 10.47% 49.81% 193827
Garen 53.60% 43.14% 10.46% 48.31% 122017
Diana 52.68% 42.22% 10.46% 47.43% 74122
Sona 57.63% 47.17% 10.46% 52.36% 154788
Ezreal 52.30% 41.87% 10.43% 47.10% 307483
Swain 57.20% 46.82% 10.39% 51.98% 61108
Quinn 54.77% 44.40% 10.37% 49.60% 64520
Udyr 55.57% 45.20% 10.37% 50.31% 335632
Shyvana 54.04% 43.68% 10.37% 48.84% 323633
Akali 55.84% 45.48% 10.36% 50.62% 152723
Gragas 52.31% 41.97% 10.34% 47.07% 79301
Zyra 58.70% 48.38% 10.32% 53.54% 105222
MasterYi 55.68% 45.37% 10.31% 51.36% 197507
Nami 56.18% 45.90% 10.29% 51.04% 156443
Leona 56.75% 46.47% 10.28% 51.63% 550002
Ryze 54.56% 44.30% 10.26% 49.39% 271738
Braum 58.22% 47.96% 10.26% 53.33% 93063
Leblanc 50.18% 39.92% 10.26% 45.11% 220880
Ashe 56.01% 45.77% 10.24% 50.89% 140299
Tryndamere 57.11% 46.87% 10.24% 51.97% 183790
Renekton 53.31% 43.10% 10.21% 48.22% 347800
Malphite 55.37% 45.17% 10.21% 50.18% 129486
Irelia 54.97% 44.78% 10.19% 49.88% 105519
FiddleSticks 57.72% 47.53% 10.19% 52.57% 119728
Corki 55.38% 45.23% 10.15% 50.27% 65902
Khazix 54.69% 44.55% 10.15% 49.65% 299274
Tristana 54.76% 44.61% 10.14% 49.62% 183186
Fizz 55.54% 45.42% 10.12% 50.43% 215390
Vladimir 52.22% 42.17% 10.05% 47.16% 74755
Katarina 56.57% 46.54% 10.03% 51.52% 180100
Jax 56.56% 46.54% 10.01% 51.61% 420357
Orianna 53.33% 43.39% 9.94% 48.37% 203849
Warwick 59.33% 49.40% 9.93% 54.32% 333202
Nocturne 53.00% 43.09% 9.90% 48.02% 245216
Lulu 54.31% 44.44% 9.87% 49.29% 193290
Twitch 55.07% 45.21% 9.86% 50.15% 337923
Janna 56.41% 46.59% 9.81% 51.42% 106945
Soraka 57.14% 47.36% 9.79% 52.22% 324054
MonkeyKing 57.40% 47.71% 9.69% 52.52% 269605
Lucian 56.07% 46.38% 9.69% 51.24% 612664
Zed 50.19% 40.55% 9.65% 45.37% 212769
JarvanIV 55.77% 46.14% 9.63% 50.94% 143599
Riven 53.96% 44.52% 9.44% 49.23% 215531
Evelynn 53.68% 44.25% 9.42% 48.90% 118944
Kassadin 55.96% 46.60% 9.36% 51.25% 121801
Taric 55.51% 46.21% 9.30% 50.77% 84396
Fiora 58.14% 48.86% 9.28% 53.48% 117399
Alistar 48.73% 39.62% 9.12% 44.04% 81495
Kayle 58.97% 50.21% 8.76% 54.55% 268477

By eye-balling this chart, I think it's clear that champions that are long range and/or heavily skillshot dependent (e.g. Jayce Nidalee Olaf Lux Varus Ziggs) have greater blue side advantage; on the other hand, champions with short range and few skillshots (e.g. Eve Kassadin Fiora Alistar Kayle) do not have such heavy advantage. 

One particular champion on that chart is Rumble. Rumble currently has slightly below 50% win rate overall. However, Rumble actually wins 57% of the time on blue side, and only 41% on purple side. This is roughly a 15% difference for Rumble on blue and purple side, far more than all other champions which hover around 9% to 11%. I believe the reason here is that Rumble's ultimate is a hard-to-land skillshot that has very long range. Due to the camera angle, it is far easier to aim Rumble's ultimate from the blue side than from the purple side.

Some implications of this:

1. You can make a champion look a lot stronger just by playing that champion more often on the blue side. Suppose a champion became FotM and he is always first picked - making this champion to be always played on the blue side. Then this champion's win rate will increase by 5% just by the blue side advantage. So, for example, if everyone decides that Ashe is strong and always first pick her, she will win ~55% of the time instead of her normal ~50% win rate. making her look a lot stronger than she looks. This hasn't happened yet and no champion has achieved 100% blue side appearance on ranked solo queue - even pre-rework Kassadin "only had" a 52% chance to play on the blue side.
2. It is demonstrated that blue side wins 10% more than purple, so if you can somehow "dodge" all purple side games without penalty, you will be climbing up in rating like crazy. So, for example, I can see that a player intentionally troll at champion select when he knows he is playing on purple (at champion select you already know which side you will be playing on), hoping that someone else on his team will dodge. This will give him a better chance of playing on blue side instead of purple.
3. The competitive scene is affected by this as well. So far,  Riot has dodged the question by claiming that the sample size from LCS games is too small to see its actual effect, but I think it's pretty clear now that blue side has an advantage after three seasons of LCS. If we believe that competitive games, similar to solo queue, also have 55% win rate on blue side for two teams of comparable strength, then playing on the blue side in competitive LoL is roughly comparable to playing as White on chess (http://en.wikipedia.org/wiki/First-move_advantage_in_chess).