virgy, would you say that it's realistic that a player could after 200-300 games still be as good as the average player 200-300 rating point above him? I'm curious. We don't have the matchmaking algorithms in front of us to know what the margin of error ought to be . We don't know the possible rating spread that the alg will allow. I've been in games where my team's total rating was 350 points lower than the other team's total, with all 5 players well below all 5 of theirs.
My point is that even when all the math theory in the world is in front of us, there is still room for some statistical outliers. you reference large numbers making the statistics easier to trust, but I don't believe that 200-300 games with a hidden matchmaking algorithm is anywhere close to a large sample number. Especially considering the rating scale change after your first 10-20 games.
All of us who have played ranked will agree that a 100 point swing is not uncommon in a day. All it takes is 6-7 games in a row where your team is markedly worse than their own, which happens often enough. So you would have to agree that a 100 point difference in rating is not enough to say that a player is better or worse than the next, correct? Your certainty that the higher rated player between two players rated only 100 points apart is the best player might only be 55-60%. I'm speculating wildly here, because I don't have the algorithms in front of me.
Even at Shaya's number of games played, there is still a big 100-200 point window that represents a sliding scale of what his strength is. There's more certainty at higher ratings and with more games played, but the whole middle range of 1000-1300 is a big blurry mess.
Thing is, 100 point swings are common in a day both up
and down. For a game like this, with the amount of normal and ranked games played (wins and losses together is probably around 1000 for most, much higher for the Shaya's of the group), that sample size should be relatively accurate. There is variance in that you can't remain at the same position after a game, win or lose, but there are bands around your Elo you will likely remain at unless you become markedly better or worse over a short period of time (at which point the variance bands will just shift). Saying 'I'm a 1500 player because I can go up to there on good swings" is a bit erroneous, as that may be the proverbial lucky streak you get when in fact your baseline is 1350 or something like it.
Yes, we don't know their algorithms, but I feel some what safe in assuming they are decent at worst; rarely have I experienced a game where, even my team playing our best and on even accounts, have we completely dominated or been dominated. I've been in games where my pre-arranged 5s team averaged 100 below the other team, and we may have lost, but it wasn't entirely one sided. Because the increments used are relatively small (10-15 Elo per game over a spectrum from ~800 to ~2400), having a streak of 6 consecutive feeders on your team and none on theirs (which, assuming 15 Elo per loss) and assuming matchmaking is fair would be a little over 1.5% in terms of probability. That's assuming
everything goes against you and for the other team. From my experience, this rarely happens (general case is both teams have feeders, but they capitalize on yours more than you do on theirs). Queuing with a competent friend would reduce the chance of a feeder slightly
per game, but taken as a series would reduce the probability even further since the reduced term would be carried through each game.
But, to use essentially everyone as an example, people 'crawl' their way back up to their baseline. If you happen to get on a lucky streak and don't actually improve during it, you'll most likely slip back over time and barely recognize it. Thing is, we don't
perceive it like that and we develop a bias. If some one dies a lot, they are a feeder. Not necessarily the case; their top Elo could get matched against your worst in a solo lane if you don't talk about it first. When I play with Mogwai in arranged 5s, we win much more when he takes solo top than when he jungles or lanes, simply because it's taking advantage of that gap if it exists during that game. However, many people enter the same mentality of "I'm in Elo hell and everyone here sucks' and want to put the solo lanes on their own shoulders and chaos ensues.
Back to the stats; yes, I think Elo is fairly accurate. I feel a lot of people try to hold onto the notion that they are 100 below where they should be but, until you hover around that mark and not just touch it before falling back down, you aren't there. Most have played enough games to have a large enough sample to be very accurate. The largest margin for error comes from the fact people improve, but that should cause you to move upwards in Elo, not down. Techincally, with the influx of players and the advance of max Elo, staying at the same spot actually means if there are a majority of ranked players above your Elo, you are moving down in percentile.