Our current method of offsetting for pre-mades is pretty effective despite the anecdotal cases. However, we contracted Dr. Sam Danziger (samdanziger.com), a guy who has a Ph.D in computational biology, to take a look at our matchmaking system and see if he could find a better way to identify "team strength".

That really is the magic number -- If you know how strong a team will be, it's easy to match them fairly. A team could be a premade, or a bunch of solos you throw together, or whatever.

The general results of using this improved method can be seen in these two graphs:
http://beta.leagueoflegends.com/boar...=1&d=125511370
http://beta.leagueoflegends.com/boar...&d=12551137099

Without going into too much trade secret detail, since this stuff isn't easy, a few things were determined:

1) As we all know, blue team has a bit of an advantage right now. We are going to do some steps to reduce this advantage, but we will be matching around this detail for the time being.
http://beta.leagueoflegends.com/boar...1&d=1255113709
http://beta.leagueoflegends.com/boar...1&d=1255113709

2) 5 man pre-mades have a significant advantage, while partial premades have less. We of course already knew that, but we have a better corrective method than before that we will implement soon.
http://beta.leagueoflegends.com/boar...1&d=1255113709
http://beta.leagueoflegends.com/boar...1&d=1255113709

3) We have a better way to deal with the "carry" and "feeder" effect, and found a way to account for this better.
http://beta.leagueoflegends.com/boar...1&d=1255113709
http://beta.leagueoflegends.com/boar...1&d=1255113709

We hope to implement most of these changes soon -- probably within a month to 6 weeks.

This is pretty cool to see you guys putting so much effort into making Matchaking work.

Kudos - I look forward to implementation.

cool story bro, and fancy graphs you got!

fifth...?

Interesting... I'm curious as to what these graphs are based upon though? Obviously these "After Correction" graphs aren't real data, since the fixes haven't been implemented yet. So... what are the graphs based upon?

Also, could we get any more specifics on WHAT the actual changes will be?

Time will tell the graphis look great but in actions is what matters

the post fix data in fact is real data. they took the past data and reran through the new model to find the new predicted values and compared with actuals. looking at those graphs, they do seem a lot tighter, and hopefully real world results will reflect this as well.

look forward to seeing the changes soon.

but lastly, data can be transformed in a lot of ways to fit any model you would like. i just hope that these were the right changes.

kudos for letting us see this type of work!

Its not "real" i mean they use real data but its a simulation of what would of happened. YES it is more than likely going to be very close to that out come in live but you never know. Like I said time will tell if this system will work. I HOPE IT DOES!