Tekadept, on 14 April 2014 - 06:10 AM, said:
I understand it perfectly in relation to it being ranking an individual based on individual performance. You are failing to grasp what I am trying to say, what I'm trying to say is it *WONT IF* you happen to luck out on a large portion of those 100 matches and get a good team made up of good teammates/premades who carry your team to victory. It wont go *DOWN* as much as it really *should*.
What you are hitting upon is an interesting question when ELO is applied to groups. Fundamentally, the issue you are getting at is that if someone is literally the worst player ever in a 1v1, and just suicides, they will lose every drop. Any time you win, or lose, it is because you are better (or played better) or worse than your one opponent. However, in a 12v12 this will not be the case. This is because the other 11 players on your team might just carry your sorry rear to victory over the enemy 12. It doesn't matter how good the matchmaker is, sometimes this will happen.
Based on my experiences with discos/afks/etc. in the PUG queue, my impression is that having one player on your team missing (so an 11v12 in the enemy favor) gives your side a moderate but not insurmountable disadvantage. With team skill being equal, you might win 30-40% of those matches. Which means even if you are literally so bad that an AFK is your equal you'll still win roughly 30% of your matches in the long run.
This same principle applies at less obvious levels when you are doing badly (made a mistake positioning, got instagibbed), but your team still wins, or when you're doing well (taking little damage, hitting many good shots), but your team collapses around you, putting you in an unsalvageable situation. In both cases you win or lose based not purely on you own skill or performance, but based on how the team around you did. Still, we can all agree that there are at least some matches where how well you perform as an individual (whether by making a good play that starts a landslide, or just consistently wearing down a key opponent) will completely flip a game from a loss to a win, or vice versa? That means there are some matches where your skill is able to shine through, and others (barring your true skill being WAY higher than your ELO) that are basically flips of the coin.
The principle in question? Noise. In statistics, this is random fluctuation (wins or losses that were not deserved) that serves to (hopefully only partially) cloak the signal (true skill) being measured. There are countless papers and even books written on this one subject, but the long takeaway is that given the right statistical tools, you are able to reduce the impact of noise on your measurements dramatically - it just requires more data (in this case, more matches). Unfortunately, this means that the matchmaker needs a lot of matches to settle someone's ELO correctly. Likewise, no matter how perfect the matchmaker, it can never change the fact that sometimes no matter how well or poorly one individual does, the impact of the other 23 'Mechs' play on the field will overwhelm that (you having a good game does not overwhelm the fact that all of Bravo lance is currently black out drunk).
As a side note, the reason that kills/match score are a very poor substitute for a win/loss based skill ranking is that it rewards a very specific style of play, regardless of performance (assassins and carries, compared to tanks and supports, to use MOBA roles which are less clearly defined in MWO). I too can cower in the back, wait for things to get bloodied up, shoot something dead, then power down in a corner. Doesn't mean I'm helping my team to victory as much as I could. If you can put up great numbers, but your team is still losing (again remembering due to noise, you will still lose
some matches even when you do things well), think about what you could do to help not just your own stats, but your teammate's ability to stay alive or deal damage. If you aren't consistently winning, you obviously aren't as much better than your opponents/teammates as you think you are. The best players will WIN 80+% of their matches on your account, I promise.
Finally, enough theory talk about ELO! Open up a separate thread for this. I want more questions/answers/suggestions for Karl!