News:

Login  |  Register

Author Topic: Advanced Stats  (Read 5155 times)

Jwarrier

  • Posts too much
  • *****
  • Posts: 1491
  • Karma: +12/-40
  • REG
    • View Profile
Advanced Stats
« on: June 25, 2016, 09:27:54 AM »
I'm making this thread to get some input regarding the advanced WDL stats I have been collecting, mostly the RATing formula.  For the first time since I released RAT I am considering a complete overhaul to the formula.  The reason?  After four complete seasons with stats collected trends could be seen and I started to feel that offense was not getting enough bonus in the formula.  There are some basic issues with creating a rating formula with doom ctf stats that lead to it being very difficult to find a fair comparison across positions. 

The first issue is how my data is collected.  Every round perspective is collected individually.  This means there is a record for every player in every round of every game.  If a game lasts 3 rounds that would mean there are 18 records collected about that game.  The second data related issue is the stats are collected only by player stats and not team stats.  This makes it difficult to rate a defender due to the fact you can only go by FRAGS, DAMAGE, K/D, and DEFENSES.  The stats have no way to look and see how many captures were allowed or not allowed, only how many captures were made by each individual player.

The next issue was one Ralphis, Stab, and myself explored when we were discussing the first version of RAT.  We initially agreed that offense was the most important stat and should receive a high bonus in the stats.  After plugging it in the trends had the offensive players rated dramatically over defenders so I decided to tone it down.  When you compare and average offense player to an average defense player usually a defender will have a higher overall average rating.  Why?  When you play offense you have some games where you might only score one capture, and some you might be a complete boss and score 5.  This will dramatically increase/decrease your average RAT.  If you have a game of 17 RAT because it was a defensive standoff and the next one with 32 RAT you finish with a 24.5 RAT.  A defender in both games will average very similar stats regardless of the result so they are likely to have a 28 RAT and a 24 RAT to get an average of 26 RAT which even if they lost the game (because of the offense player's brilliance) they were statistically RATed better. 

This is the next issue.  To make up for an offense player who outplayed a defender but was rated lower I originally added a win bonus.  5 Rating for a win, 2.5 for a draw and 0 for a loss.  Suddenly the offense player had a 29.5 RAT and the defender had the 26 RAT.  Everything looked good and the balance was there!  Wrong.  After 4 complete seasons the top 10 player list consisted of 6 defenders and only 4 offense.  "Well that isn't bad right? After all like you said a defender will average a more consistent rating compared to an offense player."  The issue I see is all 4 of those offense players had ABOVE AVERAGE defensive stats.  This means they were hybrid offense players and that meant your traditional runners like Ralphis, Stab, Turska were receiving lower than expected RATings.  These trends made me rethink my concepts and start tinkering with the formula.



Current Version
WDLSTATSv4 OFFICIAL

Rating formula: oRAT / dRAT
[Efficiency / 10] + [Frags / 100] + [(1 + (K/D Ratio)/10)) * Damage / 1000] + [Defenses / 10] + [Power Pickups / 20] + [2 * Captures] + [Pickup Captures] + [Assists / 4] + [(Touches + Ptouches) * Capture Percentage] + [Touches / 10] + [PTouches / 10] + [Result * 1]

Key changes:   Comparing offense to defense.
  • Frags: Defensive stat.  On average a +/- 0.2 RAT point.  Added to increase value of dRAT due to decreased DMG + K/D impact
  • DMG: Defensive stat.  On average a +/- 1 RAT point.  Removed the K/D stand alone RAT bonus and added it as a multiplier to the DMG.  A 200% K/D + 5000 DMG is equal to (1.2 * 5) instead of (2 + 5)
  • POW: Offensive stat.  On average a +/- 0.5 RAT point.  Increased the value of this stat by 5.  POW lead to captures and control of the POW should lead to increased RAT.
  • AST: Offensive stat.  On average a +/- 0.1 RAT point.  Decreased the value of this stat by 2.  AST are mostly incorrectly awarded and should not have dramatic affect on RAT.
  • PTouches: Offensive stat.  On average a +/- 1 RAT point.  New stat I added after doing the math and seeing the average PTouch converts ~33% of the time to a score.
  • RES: Decreased value of WIN BONUS by 5.  REMOVED COMPLETELY.  A winning player should not be so highly favored in RATing over a losing player.  Let's leave the RAT to the actual stats.
  • oRAT/dRAT: Added a separate RAT that shows the impact of the stats to the player's RAT based on the offense or defense stats of that player.  Allows you to see player position trends easier.

The most obvious change is that the RAT average dropped by 2.5 points due to the WIN BONUS factor being downsized.  The average defender stayed the same while the average offender increased by 0.5 RAT.

I'm looking for opinions on how the RAT is formulated.  Do you think the changes made in the BETA v4 are good or bad?  What would you like to see added?

Archived Version
WDLSTATSv3

Rating formula:
[Efficiency / 10] + [K/D Ratio] + [Damage / 1000] + [Defenses / 10] + [Power Pickups / 100] + [2 * Captures] + [Pickup Captures] + [Assists / 2] + [(Touches + Ptouches) * Capture Percentage] + [Touches / 10] + [Result * 5]

Key factors:   Comparing offense to defense.
  • Efficiency: Water's formula compares the stats of all players involved in a round.  Every player starts at 100 but based on your stats will gain or lose EFF to another player.  On average this means a +/- 1 RAT point
  • K/D Ratio: Defensive stat.  On average a +/- 0.4 RAT point.
  • DMG: Defensive stat.  On average a +/- 2 RAT point.
  • DEF: Defensive stat.  On average a +/- 1 RAT point.
  • POW: Offensive stat.  On average a +/- .001 RAT point.  VERY UNDERVALUED STAT
  • CAP: Offensive stat.  On average a +/- 2 RAT point.
  • PCAP: Offensive stat.  On average a +/- 1 RAT point.
  • AST: Offensive stat.  On average a +/- 0.5 RAT point.  OVERVALUED SINCE MOST ASSISTS ARE CREDITED TO THE PCAP PLAYER INSTEAD OF THE ASSISTER
  • CAP%: Offensive stat.  On average a +/- 2 RAT point.  This provides a bonus on how effective a runner was.
  • Touches: Offensive stat.  On average a +/- 2 RAT point.  This is the pressure stat.  A player gets a bonus for providing more pressure on the defense




« Last Edit: June 17, 2017, 08:59:22 AM by Jwarrier »

Rude

  • Hall of Famer
  • ****
  • Posts: 196
  • Karma: +18/-27
    • View Profile
Re: Advanced Stats
« Reply #1 on: June 25, 2016, 10:56:48 AM »
I would recommend removing the win/loss bonus entirely.  If you want pure performance stats, the result of the game shouldn't matter.  If a player lost a match 5-4 but capped all 4 flags and essentially carried; and the other team had only 1-2 cap per player- I would think the losing player had the best performance of the game

Jwarrier

  • Posts too much
  • *****
  • Posts: 1491
  • Karma: +12/-40
  • REG
    • View Profile
Re: Advanced Stats
« Reply #2 on: June 25, 2016, 11:39:49 AM »
Rude I have taken out the WIN BONUS completely for the time being while it is considered.

Jwarrier

  • Posts too much
  • *****
  • Posts: 1491
  • Karma: +12/-40
  • REG
    • View Profile
Re: Advanced Stats
« Reply #3 on: June 11, 2017, 02:11:11 PM »
Stats updated and current through season 7.

Jwarrier

  • Posts too much
  • *****
  • Posts: 1491
  • Karma: +12/-40
  • REG
    • View Profile
Re: Advanced Stats
« Reply #4 on: February 01, 2018, 07:36:41 PM »
Stats updated through season 8