Punishing queues: A proposed solution

Hi all,

I have recently been thinking about those queues in which one gets matched with significantly worse teammates and loses, resulting in a massive drop of rating which feels both unfair and punishing. I would like to mention that what I don't know if what I'm about to say has already been mentioned, but the fact that something like my proposed solution hasn't been implemented makes me think that's not the case.

As most of you may know, rating systems in competitive games tend to be based on the Elo rating system or similar math that attempts to quantify the quality of a player. These systems attribute a "matchmaking rating" (MMR) to a player and when two players compete there is an expected outcome, that the better ranked player will win. If the expectation is fulfilled there is little to no change in the rating of the players, if the expectations are suberted there is a bigger change in changes of the players involved. These systems work better the more times a player has been through the process and eventually reach accuracy, to the dismay of those rated poorly. And a matchmaking system will attempt to put two simililarly rated players in order to make a "fair match" which essentially is a 50/50 chance for those involved.

When this system is applied to a team game, such as GW2 pvp, the numbers that are used to compute these calculations are the average MMR of the players. In an ideal scenareo in which there is a big enough population this is not a problem whatsoever. However, if at any moment there is a drop in the sample of players avaiable to the matchmaker math happens.

As many of you may have experienced at some point. A highly rated player gets matched with four low player against an average team. This means that the systems literally expects a single player to be able to compensate for the rest of their team against competent players. This results in a frustrating experience for this hypothetical player. However, in order to make things "fair", GW2 when rewarding rating at the end of the game, it will compute the MMR of the individual against the MMR of the team opposing. This means that if the highly rated player won, it will win little to no rating, as they should have won according to the matchmaker's expectation, if they lost, they will lose a massive amount of rating, as they should have won. Inversely, a low rated player matched with highly rated players getting carried will either win a lot of points or lose very few points.

It doesn't take half a brain to realize that this situation is particularly bad, as it de-incentivices playing at certain points of time and it burns out players, potentially reducing the population, further worsening the problem.

My proposed solution is an obscure and mystical mathematical tool few people seem to know about, the standard deviation! This amazing feat of mathematical trickery is able to compute how disimilar an element of a sample is to the rest of the elements of the sample. If the matchmaker computed the std of the MMR of the team and then weighted down the change in rating based on how different their mmr is relative to their team, these situations would, at least, be less punishing.

TL DRThe system right now does:dMMR_p1T1=F(MMR_p1T1,MMR_T2)

Where dMMR_p1T1 is the change in mmr the player1 in team1 will experience once the game ends. F(MMR_p1T1,MMR_T2) represents the computation the matchmaker makes based on the expectation of mmr of the player1 in team1 against the average mmr of team2.

Which is unfair when:(MMR_p1T1-MMR_T1)/stdMMR_T1>>1

Where stdMMR_T1 is the standard deviation of the MMR of team1.

If the system did:dMMR_p1T1=F(MMR_p1T1,MMR_T2)*(stdMMR_T1/(MMR_p1T1-MMR_T1))

The effect of these discrepancies would be reduced.

This would potentially slow down the convergence of the system towards the true value the matchmaker should attribute to the player, but the conditions under which this correction factor comes into play aren't situations which contribute to the system's accuracy anyway. Additionally, this correction factor should have an upper bound of 1, or maybe some other number, as the function doesn't have an upper limit:when (MMR_p1T1-MMR_T1)->0 lim((stdMMR_T1/(MMR_p1T1-MMR_T1))) =infBut that would be easy to implement.

Thank you all for your time

IDK what my rating is but I know I get stuck with some absolutely garbage teammates. Like people running some really weird PvE build or doing god knows what running around and trying to kill their way to the top (and then dying off point in the middle of nowhere). Is this game so complicated that people can't even figure out the objective? Sweet jolly St. Nick, you'd think people could figure it out but they can't. I want a FFA gamemode plz.

Glicko-2 works fine in GW2 and I don’t see any reason to change how ratings are calculated.

The biggest concern with this game is how matches are put together with their matchmaking. A change was made a while back that made for quicker queues at the expense of quality games. Not verbatim but they wanted players to not have to wait 30 minutes to play fair matches 95% of the time but chose to have them get quicker matches. This is why plat3+ get placed with gold players and games turn into a shit show of questionable gameplay.