Guild Wars 2

Question: is there something the matchmaker can learn from combat statistics and personal score that it could use to tweak upward the MMR of the people stuck in low-MMR teams?

They would have to keep this secret to avoid people stacking their stats a certain way.

What kinds of stats in a match identify the carriers from the carried?

As a start, maybe a carrier has few deaths, even with lots of damage to himself, and a larger number amount of caps.

This is tough so I would expect a lot of iterations and discussion before we could get to the heart of what a carrier would look like strictly on statistics from a battle.

Carrying is mostly related to decision making / timing etc. Things that are impossible to capture in stats alone.

Well I know how I would do it if I were in charge. Based on my background computer science, I would back gather hundreds of matches where there was an known carrier. And I would use a machine learning algorithm to find patterns that are not immediately obvious.

Sometimes these patterns don’t make obvious sense, but you will find them always with the carrier and so you would look for people with that sort of pattern.

Id rather go the other way around about it…
DONT adjust that one persons mmr with best death/kill and dmg dealt/received ratios.
Just add them, weigh them, and that one person that gets best doesnt get downrated.

OR they could just base MM on divisions/pip alone……………………………………………

Locuz.2651:

Carrying is mostly related to decision making / timing etc. Things that are impossible to capture in stats alone.

Nailed it

Cynz.9437:

OR they could just base MM on divisions/pip alone……………………………………………

This is what should have been done.

I am not so sure that it is so decision-oriented that patterns would not emerge from seeing the data from games that had a carrier. Decisions have consequences, and the consequences should show up in the data in some way. And when you see the pattern, it might not make immediately obvious sense to you what was happening in the game that generated that data pattern. But it would be there.

This is what machine-based learning does today for many things as they track what they call “big data”.

It might be as simple as carriers win all their fights and are the only person on the team who tends to do so at a high degree. They take very high damage on a non-bunker class but never die once. Or something like that.

The first problem with this entire thread is the idea that “carriers” even exist.
The concept of whether you’re “carrying” is entirely based on the context of the match.
A average player on a team of rabbit ranks vs some slightly less than average players is obviously going to be “carrying”
That same player on a team of proleague players is no longer “carrying” and is in fact “getting carried”

The second major issue is what being a “carrier” really means. Carriers are if you boil it down, just players who are better at winning than other players. Which means if you want to find carriers you want a system that can evaluate how good a player is at winning, which leads to the glicko and ELO systems that already exist.

Crinn.7864:

The first problem with this entire thread is the idea that “carriers” even exist.
The concept of whether you’re “carrying” is entirely based on the context of the match.
A average player on a team of rabbit ranks vs some slightly less than average players is obviously going to be “carrying”
That same player on a team of proleague players is no longer “carrying” and is in fact “getting carried”

The second major issue is what being a “carrier” really means. Carriers are if you boil it down, just players who are better at winning than other players. Which means if you want to find carriers you want a system that can evaluate how good a player is at winning, which leads to the glicko and ELO systems that already exist.

Carriers are good at making the TEAM win, that they are assigned to. They make enough of a difference somehow to erase the skill difference between the teams, even if their MMR is similar to their teammates.

Laserbolt.6731:

Carriers are good at making the TEAM win, that they are assigned to. They make enough of a difference somehow to erase the skill difference between the teams, even if their MMR is similar to their teammates.

…and their win rate will reflect this. Hence glicko and ELO systems being standards for most ranking systems.

But clearly there’s a problem because people’s MMR is putting them among people that they have to carry. If MMR was accurate, they wouldn’t be with people that had to carry.

Since your MMR is determined from the teams you are placed on, it’s kind of circular.

Here’s the rub… a lot of players think of themselves as carriers, when in truth they are only slightly better than their teammates.

Now, dont get all angry and read on

How many of you took the time after respawn to just stop and wait 20 seconds for a teammate to also respawn, instead of mindlessly running where you think you are needed.

How many of you take the time between matches to reflect on what you could have done better – not your team… you. Me for example, I record my matches… and I usually re-watch them and keep a handful where I made certain mistakes or just got way outplayed. Just for analyzing, what exactly went wrong.

How many of you actually really know their main profession’s in and outside… and adjust traits to teamcomp. I seldomly see tempests running stone heart when facing a team with 2 DH / revs / thieves and max one necro. Instead, I guess… they leave it on diamond skin.

How many of you are really able to admit they kittened up when it happens. I mean, it happens, no big deal.

Thing is… when you soloqueue, you often get way too many play-makers that have their own theories on what works best. Just look at the threads here on this forum. Push far or not, play sides or not. Rush lord before 350 or not.

Carrying also means supporting a bad decision of your team and trying to make the best out of it. So they rushed lord just after they wiped the whole team into base… its a stupid mistake, lots of people make it… but rageing in teamchat at them wont help in this situation, it only makes it worse.
On the other hand… slowing respawners down with cc and/or kiting them away (yes, a lot of players are that stupid) … thats actually really helping your team. Probably saving the day.
And here’s the thing. This kind of behaviour will NEVER show up positively in any statistic. Instead, the other teammembers making that stupid mistake will look good on paper.

So, and about that mmr atm… I took some time to re-think that reddit AMA sentence about mmr not being properly down-adjusted when loosing against a higher mmr team. It might actually really be the source of evil for those really good players trapped in mmr hell atm. Because it equalizes a bit below average mmr with a bit above average mmr. Et voila, all of a sudden you get players on your team that – before this error occured – wouldnt have been matched with you.

/end wall of text

personal score and top stats are too narrowly defined to singly indicate a carrier, and the rating system would have to have intelligence behind it to pick out a carrier from an aggregate of each stat and score.

Yasi,

Good observations. And yeah, 4000 matches and I still run right out of base! How do people learn this stuff? Probably from watching twitch, right?

insanemaniac.2456:

personal score and top stats are too narrowly defined to singly indicate a carrier, and the rating system would have to have intelligence behind it to pick out a carrier from an aggregate of each stat and score.

Could be. Especially in light of the things Yasi talked about

Cynz.9437:

OR they could just base MM on divisions/pip alone……………………………………………

I agree with this, even though I have been lucky to be placed in mostly winning teams. I feel like the current system suffers from a sort of confirmation bias: it thinks you are one skill level, and then puts you in teams with others of that skill level, meaning you either win big b/c it thinks you are better than average for the division or lose a ton b/c it thinks you are worse. It not only makes uneven matches while the ladder stabilizes, but also discourages those stuck in “MMR Hell” b/c they keep getting stomped over and over. Seeding purely by division means that bad players WILL get carried up to ruby (or maybe only sapphire), but they should stabilize there (where there is no division/pip protection) as they will even out wins/losses, as the system seems designed to do. Better players will progress up more slowly, as they get stuck with much worse players sometimes in lower divisions, but they WILL continue up the ladder as their wins will outweigh their losses due to their ability to carry against worse players.

To add to that, they need to work out a way to pre-seed divisions a little bit:

1. Have a series of play-in games to determine which tier you start in.

OR

2. Everyone in Legendary+ from the prior season starts in sapphire, everyone from diamond starts in emerald, everyone else starts in amber. This could be dangerous in that those high-ranking players could get stuck with long wait times to start a season, but that would happen eventually as they move out from the main pack.