Mizeur, on 15 July 2014 - 01:59 PM, said:
Your conclusion is correct. The rest of your argument is an incorrect explanation of how Elo works.
You're right that you can't assess a player's individual skill using Elo if that player plays exclusively as part of a static team. An individual player's Elo would be the same as the team's. However, if before every game every player was placed into a random lottery draft, with a sufficient number of games an individual's Elo would become correlated to their individual likelihood to win.
MWO's matchmaker works a lot like that. And it uses Elo ratings to put together teams with similar probabilities to win within parameters deemed acceptable by the devs.
You actually read my post completely oppositely. Elo can ONLY predict static teams, not dynamic ones.
What you're saying... Mathematically speaking, no, it can't work that way. It's an impossibility. You're making an assumption that all individuals on a team will be completely unaffected by the play of other players on a team. The assumption only works if all players on a given team are operating independently, and that their individual actions will have the same positive or negative impact on performance regardless of what the rest of the team is doing. That all players operate in their own "bubble." That is never the case in any team sport (and rest assured, multiplayer games are organized as sport).
If this were organized more like, say, the Olympics, where each individual on a team is competing individually, then yes, Elo could certainly apply to the individual as a component of team win/loss. However, any sport where teammates must work simultaneously and cooperatively toward a goal would not allow this to apply. Myriad scenarios come to mind, but something so simple as a relay races proves the point. It is perfectly possible, and even likely, for a relay team to lose a race if even one of four runners is not pulling his weight. Conversely, even the best runners in the world cannot overcome the deficit created by mediocre teammates. The same is true of any sport. Thus, you cannot use the result of the whole team to assign a predictive skill assessment for the individuals that make up the team. Continuing with the Olympics as an example, there are numerous examples of individuals on an Olympic relay team failing to place well as a part of that team, but medaling in individual events. Obviously that shouldn't happen if Elo is to be believed, and is why it doesn't apply.
Further, even in sports where team rosters are largely fixed from game to game, Elo loses it's predictive ability should the roster, for any reason, change. To use football, for example... should a player in even an often looked-over position get injured and be unable to play, the impact of his loss on the team can turn a winning team into a losing one. This extends well beyond the "star" positions. How often do we see teams forced to change even a single player have a drastic change in their win/loss percentages thereafter? And how often do we see teams make only a few roster changes in the off-season only to have a completely different win/loss percentage the following season? This flies in the face of what Elo tells us about those teams. Further, sports halls of fame are filled with amazing players with amazing accomplishments, deemed to be representantive of the highest levels of their sport... and yet, how many of those can be said to come from teams whose abilities match those of the individual. Rarely any. Again, it's why Elo doesn't apply. The nature of Elo makes certain assumptions about a player (be it an individual or team) that can't hold true in most team-based play.
The "self-correcting" feature of Elo is just a nice way of saying that Elo is perfectly good and predicting the likelihood of a win... unless it doesn't... in which case we change our minds.
Look at how win/loss and player skill is predicted in any organized sport in the world. Team win/loss predictions are nearly always based on assessments of individual player skill and their overall impact on the team, and not the other way around. How do we assign a relativistic skill assessment to a player and provide a prediction of team utility? By the rigorous collecting and comparison of any relevant data that expresses how a player's actions effect the team. Hundreds of stats are collected on every player in every game, and cross referenced and correlated to every player on their own team, the opposing team, and the league average - both in that game, and cumulatively over seasons and careers. It's done exactly the same way in many modern multiplayer games too. The total number of stats tracked by Battlefield games is right up there with anything collected for the NFL. The game knows what you're doing at any given minute, and exactly the impact it's having on the team. It knows who's helping and whose hurting. And that's such an easy thing to do once you start paying attention.
The scoring system in MWO is virtually non-existent. It doesn't even provide any points for working the objective in either Conquest or Assault modes (you get nothing for neutralizing or capturing objectives on Conquest, and only turret destruction is awarded as an individual score in Assault, not base capture). It tracks virtually nothing related to player performance. W/L, K/D, accuracy, etc are meaningless stats out of context, which is why games like Battlefield do not use them as the primary basis for player skill ratings.
The devs would do well to drop Elo altogether and implement a proper scoring system, and a skill system based on that scoring system. I guarantee that if they do, matches will be more fair, and more even. Until then, it'll remain largely random.
0
