So for those arguing MatchScore sorting will push groups up, here's some graphical representation of why it won't be as effective as WLR.
This is actual season 46 data pulled from Jarl's database. All 34581 players.
Jarl's list doesn't give information about groups, because that's not recorded. However Jarl's list does give a player's unit tag or lack thereof.
Assumption: Players without a tag are more likely to play solo and players with a unit tag are more likely to play in groups. Not 100% but it's the closest to accurate group information we can get. If anything, any deviation from this assumption will push both averages closer together.
I have processed the plot as if it was a normal distribution, but there's some skewing, especially for WLR. I'm keeping it simple and it's close enough to see the trends.
X axis is WLR
Global average is 1.017
Approx Solo(no tag) average is 0.998
Approx Group(w/tag) average is 1.342
Box depicted represents the standard deviation. Approx 67% of players lie within that box.
Y axis is avgMS
Global average is 218.4
Approx Solo(no tag) average is 215.8
Approx Group(w/tag) average is 263.4
Box depicted represents the standard deviation. Approx 67% of players lie within that box.
Approximately 45% of players lie within the combined XY box.
The red lines are the global average MS and global average WLR. Where they cross is the average player.
Both yellow marks show the average Solo(no tag) and average Group(w/tag) player.
Facts:
The correlation between Match Score and WLR is 0.656. Which I would call moderate.
Correlation between KDR and WLR is 0.697.
Correlation between KPM and WLR is 0.670.
There is no correlation between Mech weight class and WLR.
The correlation between Jarl's adjusted MS and WLR is weak at 0.420.
Analysis:
So we see that outliers tend to stagnate in Match score, but still skyrocket to the right on WLR
. Even when the Matchscore is in the 150s, which would suggest "poor skills". However the stats show that those "poor skills" are more likely to win a game than some players in the 300s.
Interesting trend is that Group(w/ tag) has a more pronounced effect on
WLR than on
Matchscore (any surprises there?)
Also, we can see there are much more lower scoring players with high WLR than low WLR players with high score.
Stomps:
Number of stomps are
directly related to how skilled players are at winning a game, not at scoring high.
To explain, we know about 35% of games are stomps. Regardless if you are on the giving or receiving end.
So 35% of losses are losses to stomps, 35% of wins are wins with stomps.
This is related to WLR directly here's how:
0.0 WLR:
35% stomp losses
0% stomp wins
0.5 WLR:
23.3% stomp losses
11.7% stomp wins
1.0 WLR:
17.5% stomp losses
17.5% stomp wins
2.0 WLR:
11.7% stomp losses
23% stomp wins
The total of stomps is always 35% of games. HOWEVER, WLR is a direct indication of how likely you will suffer a stomp, which decreases the more WLR goes up, and how likely you are to stomp people.
Once you team up similar WLR other than 1.0 together, you get an increased probability of causing/suffering a stomp. In reality, the numbers are probably not evenly distributed, and chance of stomps is probably lower around the 1.0 WLR mark and higher at the other ends, but linear approximation still gives the idea.
If you team up players, statistics are doing their weird thing. By teaming up the 2.0 WLR with the 0.5WLR vs 2 players with 1.0WLR, you get about an equal chance to see a stomp from either side. However by making teams with average 1.5 and 0.7 WLR each, you get a 43% chance of seeing a stomp.
Stomp chance is a
function of team combined WLR. Matchscore is moderately correlated to stomps.
I might get around making correlations to Matchscore later, but that's less useful. I don't claim being a stats guru so please point it out if I made a mistake.
Have fun discussing stats!
Edited by Cluster Fox, 26 June 2020 - 02:00 PM.
6
Kosomok, on 24 June 2020 - 08:09 AM, said:
