Cimarb, on 26 May 2014 - 04:39 PM, said:
Also, Elo really needs to be determined by more than JUST win/loss. WLR is a great stat, but it has little to nothing to do with an individual's contribution to a team. I won't go into detail, as I'm sure you have heard all sorts of arguments about this, but determining an individual's value based off of a team-based stat is odd.
Elo is determined by wins and losses, but not by win loss ratio. This is a critically important point, and it is why there are no stats for determining Elo. Elo is not calculated by stats at all, and can't be. The whole system is based on calculating the likelihood of your team winning the match by using past results, and then adjusting scores if the calculation ends up being incorrect. It's an adaptive system that does work reasonably well (no, not perfectly, but there is no perfect system).
You can't incorporate stats into it and still have it work. You'd need an entirely different system for that.
Now, here's the thing. Everyone wants to think other stats will better reflect your skill, but ultimately that's wrong in this context.
The
only thing that matters is: Did you contribute positively to your teams victory?
How can we determine that? How much damage you did? Damage done is irrelevant: I can score 600+ damage in a battle and contribute nothing noteable to my teams' success, and I can score 200 damage, get 3 kills, and be a linchpin of the team's success. In many cases, in fact, high damage games are a result of poor play, not good play.
Assists? Again, that's not a reflection of how well you did. I'd rather have a player focus down a target or two rather than splatter a little damage over multiple foes uselessly.
Kills? Still meaningless. You're usually a better contributor to your team if you switch from a crippled target to a threatening one, giving the kill to whoever's nearest when mopping up time comes around. Also, while "kill stealing" isn't a thing (you SHOULD be focussing fire, after all), likewise if you push that HGN to a red CT then someone else finishes him, it's not a reflection on your performance.
KDR? As kills isn't particularly relevant, neither are deaths. If you play a Squirrel light/medium, and do it well, you'll likely end up with a rather poor KDR but could be strongly contributing to your team's success.
And then, what about all the immeasurable factors? Maybe you're a terrible shot, but a decent leader skilled at organizing PUG teams? Just being a good communicator can have a massive impact on your teams likelyhood to succeed.
See, here's why Elo is so commonly used. On a match-by-match basis, it's generally useless. However, when provided with many matches, it's the best system to actually gauge your real contribution to a match. No matter what people like to say, they are always 1/12th of the match. Sometimes, that's not enough to make a difference that changes the outcome (either because your team would win or lose regardless of what you did, but sometimes it is. As the "auto-win" and "auto-lose" are equally likely (yeah, everyone likes to think they lose more because of terrible players, but that's just stupid confirmation bias working - the other team has the same factors at play), statistically speaking over many matches then the differentiating factor is
you.
Thus, Elo is the only system that can account for the immeasurable factors.
Again, it's not perfect. It fails primarily in two ways:
1) Small sample size. Without lots of matches, individual matches have too much weight. This is why, incidentally, we don't have per-chassis Elo scores - the more scores we have, the more random they are. We could have a better overall score if we only have one Elo rating rather than 4, but there's a strong argument to be made that play between weight classes is very different, and players can be very good at one and terrible at another (I'm terrible at lights, excellent in assaults, for example). Have to pick your poison with this.
2) Player pool availability. This cannot be fixed. The hard truth is that even with a seemingly large pool of players playing concurrently, at the extremes of the scale there will be very, very few players searching for a match at any given time. As such, if you're very high or very low Elo, you
will get a larger spread in your matches. It's simply unavoidable. The only solution to this is extending matchmaking time, but that obviously has it's own failings. I've shown math for this elsewhere - even with 50000 concurrent players(which would be indicative of 6 digits worth of players playing daily, a very good number), having 300 of them searching for a launch at any given time is extremely generous, and of that only a tiny portion will be high or low rated.
Any other rating system will fail in these same ways as well. If the pool of players at your skill level actively searching for a match at exactly the same time as you is low, then you're going to either fail at searching, have really long waits, or have more random matches. There's simply no other options there.
Designing a matchmaking system is hard, and it can't work perfectly unless you've got both an
extremely large player pool
and a vast sample size. MWO will never have the former (it'll always be a niche game), and the later only comes with time.