Mister Blastman, on 12 August 2016 - 05:27 PM, said:
Yet here we have a statistical median for the mm at 1:1 and an average at 1.09 overall for 34k players.
If you have a 1.39 win/loss ratio you're pulling your weight!
News flash - for every person who wins, one person will lose. So the global average must equal a 1:1 ratio.
The important part is the percentage of people that are at or near a 1:1 ratio. A perfect match maker would have 100% of the people at a 1:1 ratio, people are always playing their equals.
A good match maker would follow the 80/20 rule, where only the top 10 and bottom 10 percent is not close to that 1:1 ratio.
Putting it on a 5 point scale it would look like this:
5 - Perfect - 0% outside of the 1:1 range
4 - Good - <25% outside of 1:1 standard deviation range (top and bottom 12.5%)
3 - Average - <50% outside of 1:1 (top and bottom 25%)
2 - Poor - >75% outside of 1:1
1 - Awful - 100% outside
Using lost dragons numbers:
Lostdragon, on 12 August 2016 - 08:50 AM, said:
I downloaded this data and did a quick couple of formulas. For people with at least 10 games (32049 players) 50.57% are at 1.0 or better. Only 5.79% are at 0.5 W/L or lower. About 22% are lower than .75 W/L.
The average W/L is 1.09 and median is 1.0.
So looks like all the crying about the MM can just be dismissed as it is indeed only whining based on skewed perceptions.
Bottom 22% are getting crap matches, likewise there would be >22% top scores that are getting matches below their skill level. The top percentage is larger because the average numbers are greater than 1:1 so there are more winners than losers, caused by eliminating people with less than ten matches, who must have more losses then wins.
So lets say that the top is 23%. That means for 45% of the players, the match maker is failing. That is on the poor side of average on the above scale. That's hardly something commendable, and likely could be done with a random match maker that didn't use any rating. Nearly half the players are getting crap matches.
I would bet real money that if we had the same set of numbers from the elo based match maker the results would be higher, nearer to good but defiantly on the good side of average and probably in the high 60% to 70% range.
Edited by MrJeffers, 13 August 2016 - 10:43 AM.