“Ranked Account Seeding: Formerly known as “New Account Seeding,” Ranked Account Seeding is built to more accurately place you in your first games of Ranked. In the past, we were using a fixed placement to start you somewhere towards the bottom half of the curve. While starting everyone at the same spot was "fair", this meant players actually at the skill level of the starting MMR were frequently matched with new players whose actual skill was much higher or lower. Since MMR is close to a bell curve, this meant matchmaking quality suffered right at the highest-populated skill level.
By seeding players based on gameplay information from other queues, we’re confident that players will be more accurately placed at the start of their climb. This will make it so others aren’t experiencing matches that seem imbalanced from the start. We’ll be rolling this out slowly over the next couple patches so we can monitor and respond to impacts quickly and carefully.”
On paper, this seems like a good idea, but the execution was highly flawed – and I have some statistics to back it up.
A few months ago (before I read these dev updates and learned of the new system), I started to notice that I was running into a high number of very unexperienced players. And by unexperienced I don’t just mean “feeding” or “griefing”. More so, lack of map awareness, not breaking 50 CS, no warding, high death rates, etc. Players who, against no fault of their own, were just very new to the game and were not very good yet.
Because of this, and my professional background in data science, I decided to do some research and take advantage of na.op.gg’s awesome website that tracks historical gameplay and statistics about those games. I went back in time and started logging all of these games in a spreadsheet, specifically looking for games that included new players on either side. It was while I was conducting this research that I discovered the dev updates, which actually provided a reason for this increase in new players.
I quickly came across a few major findings which I will highlight here before moving into the technical analysis:
1) The new Ranked Account Seeding system is HIGHLY overestimating the skill level of new players according to KDA, gold per game, CS per game, and warding per game.
2) The new Ranked Account Seeding system is leading to increased systematic randomness to winning or losing and poorer game quality according to game outcomes, game time, and new ranked players’ win rates.
3) While Riot did not officially announce which patch they were rolling out the new Ranked Account Seeding, it was clearly rolled out in patch 10.13 on June 24, 2020.
4) The new Ranked Account Seeding system unequally places new players in the high silver and low gold tiers.
Note: It has been brought to my attention that Riot may have postponed the full Ranked Account Seeding update. That does not however mean that parts of it are not being tested, or that other new systems are not being tested. In addition, data is data. Something happened around patch 10.13, whether it is the new Ranked Account Seeding system or something else.
For this research, I defined a new ranked player an individual who has not been ranked in any previous seasons AND has not finished their placement games for the current ranked season. Once again, I used
my na.op.gg page as the data source and used a sample of 150 solo queue ranked games. All of these games were played at an average tier level of Silver 3 – Gold 4. The data for this project is publically available in this
GoogleSheet. (Don’t worry moderators, I redacted the names of the new players.)
Let’s start with point 3 from above. How do I know this new system rolled out in patch 10.13? Mostly because that is when new ranked players started appearing in my games. Prior to patch 10.13, I have a sample size of 80 games. In those 81 games, I only found five occurrences of a new ranked player – which is a rate of 6.17%. My sample of 69 games played after patch 10.13, found an astonishing 26 occurrences of new ranked players – which is a rate of 37.68%.
This is clearly more than an anomaly as this is a relatively large sample size over multiple months’ worth of time. While there is nothing innately wrong with playing with new ranked players in 37.68% of high silver and low gold games, it becomes a problem if these players are at significantly different skill levels than the average high silver and low gold players, which I will discuss next.
Let’s move to point 1 from above, that the new system is overestimating the skill of these new players. The KDA of those new ranked players over my sample combined to 58/175/111, which is a KDA of 0.97. The average KDA of a Silver player is 2.4.
For those who are not KDA fans, here are some additional metrics. The average gold per player per game in silver is 10,603. In my sample of new ranked players, the average is 7,494. Average CS in silver is 132, whereas in my sample the average was 99.8. The average amount of wards placed in silver is around 10.5, in my sample 8.2. For those of you worried about an over-sampling of support players, the opposite occurred, only 4 of the 26 players played support (meaning that average gold and CS for new ranked players for the above data is skewed upwards).
Finally, let’s look at point 2, that the new Ranked Account Seeding system is leading to increased systematic randomness to winning or losing and poorer game quality at these tiers.
The craziest number I came across in this whole project is the following. In the 26 occurrences of new ranked player in my sample size, the team with that player won only twice. That is a 7.6% win rate. In addition, in one of those two instances, there was a new ranked player on both teams, so one of them HAD to win. Taking that game out of the analysis, the win rate drops to 4.2%. If I was running a regression looking for a correlation between chance of winning and having a new ranked player on your team, I would produce one hell of an r-squared value.
Outside of win rate, game time is another good metric to look at when examining the quality of a game. The average silver game time is 29.21, however the average game time in the sample or new ranked player games was 26.06.
Finally, I would argue that this new system is leading to a poorer experience for these new players themselves. While the average win rate across ranked play is obviously 50%, the win rate for the new ranked players in my sample (in ranked play) is 26.9%. It’s going to be hard to make an argument that new players are being placed in the right tier and are receiving positive game experiences when their win rate is extraordinarily, almost laughably, low.
What if some new players should be ranked higher? A simple logic and experience would tell you that newer players, even if they are naturally gifted at the game, on average, will be some of the worst players in the game due to lack of experience. In addition, there is no supporting evidence that “good” new players are getting stuck with “worse” players in low gold or high silver. Look at any of the metrics that I provided above.
Even if this does suck for new players, it shouldn’t matter to anyone else. If you are good enough you can carry one bad player. This is an argument I hear all the time and there is no logical backing to it. Outside of a new player virtually making the game a 4v5, it is a 4v5 PLUS a person on the team of 5 has a significant gold and item lead. This isn’t even taking into account if the new player is the only person carrying smite on your team, or was expected to be a hyper carry.
Finally, an argument could be made that I am the common factor in this sample set. While this is true, I would simply argue that I am probably not accounting for all of these new players having poor games. My honor level is normal and over this sample I had a 3.39 KDA. I would by all means consider myself an average silver/gold player.
Obviously, this new system isn’t working. Riot should be incentivized to fix it for a number of reasons. First and foremost, game quality is suffering because of it – especially for their newer player base, which should be an important focus for any gaming company. Not only is this hurting newer players, it is also hurting players in high silver and low gold who are now wasting their time in almost a third of their ranked games. My recommendation is that new players simply start their placement games in Iron and have to win to move up, possibly with an upper cap on being placed Gold 1 after your placement games.
I think it is an easy argument to be made, that the average person who has never played ranked before is probably an Iron or Bronze level player. For the extremely few who fall outside of that assumption, they have an opportunity to climb just like everyone else. In my opinion, it is far better to make that small population climb than to plague and ruin 37.68% of silver and gold games (for all parties involved) with brand new players in a feeble attempt to weed those few players out who should be in silver or gold right when they start playing ranked.
Source: Original link
© Post "Statistical Analysis on the New Ranked Account Seeding – An Epic Failure" for game League of Legends.
Top 10 Most Anticipated Video Games of 2020
2020 will have something to satisfy classic and modern gamers alike. To be eligible for the list, the game must be confirmed for 2020, or there should be good reason to expect its release in that year. Therefore, upcoming games with a mere announcement and no discernible release date will not be included.
Top 15 NEW Games of 2020 [FIRST HALF]
2020 has a ton to look forward to...in the video gaming world. Here are fifteen games we're looking forward to in the first half of 2020.