Reign of Gaming

Sigh, let's go step by step.

Let's presume I'm maintaining a solid 55% winrate in some division (Silver IV).  Sadly, I've lost my division advancement twice.  Nonetheless, I hold my 55% winrate.  My elo has risen (due to more wins than losses) much faster than my division.  That may be fine for me, but it turns out that this is currently expected behavior.  In fact, the average elo in my division (Silver IV) is no longer where it was when the system was devised (say around 1300 or so).  Over time, the average elo increases, this is the definition of getting harder.  It causes pooling in the lowest division (if the average elo in Silver IV is inflated to 1500 then 1200-1500 elo players will be pushed down into Silver V, making the variance and range of skill there dramatically different than the next tier.  This behavior is likely not intended.

Simply put, a system that holds players in a division despite their maintaining an above average winrate has negative drag.

1. Actually I've been demoted after losing exactly 2 games post promotion.  And according to the people I asked, 2 consecutive losses was the number needed to be demoted.  It could be wrong, of course, but that's what I've been told by the people I talked to (nothing official from Riot as far as I know).

2.Actually it's nothing like saying that.  It's like saying in the old system, you have a 55% winrate but will be unable to gain elo over time.  Of course the real-world distribution isn't like that, many will advance, but more than half of the people with that winrate will fail to advance in any given series.  This is a large issue for an elo-based or elo-like system.  It causes systemic drag that puts people below the level where they win 50% of their games (your ideal elo resting spot).

3.  The problem of divisions getting harder, as expicitly stated, applies to all BUT the bottom bracket.  As players with above 50% winrates fail to advance, they pool up.  This means your division (say Silver III) is now full of people who used to be statistically better than 50% winrate...thus the division has gotten harder.  In fact, if they then maintain a 50%+ winrate, they're improving as they're now beating a better field.

4. I know a ton about elo and other systems, but that's not really important here.  The issue at hand is that the merging of elo and divion-based play result in over-arching behaviors that can be seen as negative to the community (they are just features of the interaction between the new division series and  point levels).

5. The average elo is unimportant.  It's fairly important that elo systems are determinsitic.  This means that, given all my opponents have the same elo (just for the purposes of modeling), all that matters is total wins - total losses.  I don't know if this is the case in LoL and would love to get some data.  For instance, if one starts at 1200 elo, then loses 10 games and then wins 10 games and ending around 1200ish. Whereas another 1200 elo player wins 10 games and then loses 10 games and ends up significantly higher or lower (there will be some disparity as different games are worth different numbers of points), then there is likely something odd going on in the system.  This may or may not be a problem.  We know that Riot can't use a simple elo system due to the team nature of the game, no worries, but if the division series structures are distorting ratings on a large scale, this can result in a really broken system.

Sigh, look at the math above, that's exactly what it shows.  It shows that, on average, a 55% winrate only has a 45.8% chance of moving out of a division.  That's essentially the whole issue at the moment.  I'm glad that your wins ordered well for you (note: many 55% winrate players will advance, just that on average they will not).  That said, your experience points to another thing I've been curious about:  if you really held a 55% winrate across 5 divisions, then either you improved exactly at the rate of advancement, or the difficulty across divisions isn't scaling correctly.  In a functioning elo system, you should have something like a 60% or even higher winrate in Gold V that then tapers as you approach your true level.  If players are maintaining a constant winrate across divisions, something is fundamentally broken with the system (the elo being used is unable to differentiate skill well, or the game is too high variance for elo to be indicative).  Given that plat players I play with tend to crush me, and silver players lose, I don't believe that elo is broken.  Data I would love to see, is the division/elo difference between a sample of players who have overall 50% winrates, but in different orders.  If you see a drastic difference between 18 losses followed by 18 wins and 18 wins followed by 18 losses, there is likely something skewing the system (18 is arbitrary).

Look at the math.  The whole point is that this isn't true.  A 55% winrate will not climb.  In fact, over time it will fall.  A 60% winrate should climb, but over time the same division will become relatively harder, making it impossible for the players who belong there to remain.  In the current system, being better than the range you're in does not cause you to climb.  Being MUCH better does, for sure, but a 55% player currently is not climbing, and that's a huge issue for systemic stability.  

Quote from darkChozo »

I believe you need to do a bit more statistical analysis in order to support your conclusion. Saying that a promotion series is harder to win than to lose isn't that helpful; we already know that that's the case (3/5 > 50%, basically), and while getting the exact probabilities for any given promotion series is useful, it doesn't really support that the system is broken as a whole. [just to cover myself, saying that a person with an above-average win rate can still have a >50% prob of losing a promotion series is just an extension of saying that the promotion series is harder to win than lose]

The proper way to do this would be to try to calculate the probability that a given player with X win rate will be promoted in Y games; or stated the other way, finding the average number of games that it takes a player with a X win rate to advance. You'd be looking to see the average time it would take for an above average player to advance out of a division, and also for the probability that a below-average player would be promoted due to luck (kinda sorta a false positive rate). I'd imagine that Riot did something similar when they designed the league system, probably with some metrics for promotion time and false positive rate.

Side note: it'd be rather hard to model exactly how a player would advance through the system because match performance also affects your MMR, which in turn affects both your win rate [inversely] and your LP gains [directly], but you could limit your issues by assuming that a player stays at about the same MMR throughout.

Second side note: Actually, that makes me think of a potential flaw in the system. If you have a much higher MMR than your league would suggest, does that mean that you're being matched against similarly-high MMR players for your promotion series? In that case, there may be an edge case where you have to play an above average number of games because you get unlucky in a promotion series, win a game and get a billion LP, get unlucky in a promotion series, etc. Might be interesting, though I rather doubt that everyone who complains about anything ranked-related is being affected by such a situation.

While I"m not looking at game distributions (which I mentioned in the post), that's not really an issue.  I'm not saying that promotion is harder to win than to lose, I'm saying that given consistently above average play currently can easily result in demotion rather than promotion.  Acerunner claims that's fine, but he's missing the part where this causes the quality of the divisions to skew drastically over time.  As 55-60% winrate players are unable to advance, their division fills with such players.  They are then unable to maintain that 55-60% in their--now harder--division.  This leads to a strong downward pull.  Even being an elitest, this just makes the system overall very poor as the divisions lose meaning.  If you need to be 400 elo above your current division to advance, that's a problem.  It makes division V very very toxic as it fills with a HUGE variation in skill (making matchmaking bad and unfun) and the other divisions excessively difficult.

I don't know a ton about the Riot algo or how it relates elo to points gained in a division.  They've made a bit of a mess for themselves trying to smash a division system (which usually forces specific teams/players to compete) ontop of an elo system (which doesn't care who you play against).  The addition of promotion series makes matters even worse.  

crossposting from my reddit post (http://www.reddit.com/r/leagueoflegends/comments/199ww9/why_the_current_division_system_is_strongly/)

Since the s3 division system was introduced I have had a number of my friends and acquaintances complain that the system feels off or bad. I decided to put that to the test.

Test 1: Division Series 

Need 2/3 games to advance

Need to not lose the next 2 games to be demoted

Data:

https://docs.google.com/spreadsheet/ccc?key=0AinJYmM3L-EFdDZ5TmlpVk45Y08zT3VOYzNIRXlCVEE#gid=0

 As you can see, a player with a 55% winrate is more likely to fail (45.8% chance to both win series and then not lose the next 2 games) than to successfully stick a division higher.

Why is this a real issue?

Problem 1: One needs a mean winrate around 60% to be more than 50% likely to advance. Thus, over time, players will actually become stuck in brackets where they maintain an over 50% winrate. This means that division become more competitive as players play more games, with the exception of the lowest division in your bracket.

Over time divisions above the bottom division will become increasingly full of players who are playing below their level, this forces players previously correctly placed into the division down. As players play more games, advancement becomes much harder because the quality in all divisions except the bottom one increases exponentially. This leaves the bottom division full of a HUGE range of skill levels.

Problem 2: An elo system is designed to reward consistent play and find true skill over a large sample. The current division system does exactly the opposite. Instead of rewarding consistent play, the system heavily punishes (remember, you start the new division at 0 points, and 2 consecutive losses there is the equivalent of losing a promotion series--2 losses-- and 50 rating --2.5 losses...or 4.5 games) losses, especially consecutive losses. Wins, on the other hand, are gated by division series, meaning that for every division you advance, at least 2 wins are not counted towards your advancement.

Thus, consistent play is a worse way to advance than going on a win streak (something else explicitly rewarded by the current system). Win streaks and loss streaks, statistically, will happen, but win streaks are rewarded less than loss streaks are punished currently.

Note: It is important to note that the distributions are not discussed above, only means. This is because it's easier to talk about means than the probability distributions with standard deviations. This is obviously a weakness, if there is interest in a real in-depth discussion of the subject of division series, I'm interested in doing it but need time and a display of interest.

Fixes: The primary culprits here are the system behavior around the division series. Winning the series (2-3 games played) results in 0 points in the new division, making it far easier to fall back (losing the next two games) than it was to advance (gain 100 points, then win 2/3). Losing the series (2-3 games played) costs the player 2-3 games + 2-3 WINS (2-3 for series, 2-3 WINS to recoup the 50 points). This could be smoothed by promotion placing the player at 50 points in the new division. This way, losing is still harsh (and the system may retain some negative bias) but winning is rewarded.

TLDR: The current division system is negatively reinforcing, leading to pooling of players in division V. This makes all other divisions artificially strong and distorts skill distributions.

I'm looking forward to it!  I've been helping to contribute some video tagging, and these guys are awesome :).

This was really useful to me over the weekend.  Really love the Voyboy and Wickd (when not watching IEM or amazing Naijin Sword).  Anyway, wanted to bump and say thanks again :).

I use it a good deal when I'm working on learning one specific champion.  I've found that watching hard matchups played by pros really helps.  Also, the little things like how they use skills to last hit and when they can last hit creeps is super helpful, at least for me.