My theory is that the matchmaker finds a potential match by picking players around the same SR, then divides them across teams to make it as even as possible based on projected performance from individual players(MMR). This means that even if the match is made of 12 people at your SR, if you’re performing better than the average of the lobby, your remaining team members will total out to below the average of the lobby to even it out. If it were linear, this wouldn’t be a big problem.
But, given how much impact a single thrower or smurf can have, performing a bit above average can result in you being paired with more throwers, new players, and against more smurfs. This has a drastic impact to the actual result of the game, even if blizzard’s internal numbers don’t realize the impact of those players.
I go into more detail on this idea here, using objective points rather than speculation or anecdote:
Please note that this is just a theory, and while I feel I’ve done a decent job substantiating it, there is no more proof that it is true than there is that the inverse is true. Without Blizzard being open about it, we really can’t know for sure.