dunkr4790
02-27-2011, 11:37 PM
Figured I needed to post this before I got too busy and forgot about it. Mods, if this is in the wrong section, move it, although I'd like it to get some views.
Spreadsheet found here: https://spreadsheets.google.com/ccc?key=0Aga14RVhJgKGdEEyQlprVzd4NnkzMHd0c0lSVms0N Xc&hl=en&authkey=CPCOu7IP
Table of Contents
1. Overview
2. Explanation
3. Analysis/Problems
4. What’s Next?
Overview
Basically, WAR (Wins Above Replacement) attempts to objectively determine the value a player brings to his team in all aspects of the game. Baseball has a version (which I basically copied) on many different sites, such as Fangraphs (http://www.fangraphs.com). In my version, I use all the stats available to me (K, A, D, T, games played) to attempt to do something similar.
Explanation
Using the leaderboards provided by MLG throughout the season, I compiled all the stats over the course of the season (stats are on the main page here, but you probably noticed that already with the very well labeled box on the right side of the screen that has a nice pretty picture). Then, I took the sum of all the stats of anyone who had played 30 games last season (to prevent small sample sizes from influencing the sums). Using objective time as the base, I created a point system where 1 point is defined as 1 second of time. Next I found the ratio of kills to time, using this ratio to establish the number of points a kill or death is worth (kills are positive, deaths are negative). Since assists are only registered after a player is killed, I weighted those equally to kills.
Using a simple Excel formula, I calculated the total number of points each player had; then found the number of points per game. Because these values are so similar (at least per game), it makes more sense to compare them to a set value. Although comparing everyone to the average is possible, it makes more sense to compare someone to a player that could fill-in for them. This is where the replacement part comes into play. Defining replacement level as 80% as good as an average player (more on this later), I subtracted the number of points a replacement level player would produce from all the point-per-game numbers, to get points above replacement per game (PAR/G). Since this entire system is based around objective time, and 250 seconds of time wins an objective game, I divided this PAR/G number by 250 to get wins above replacement per game (WAR/G). I then used that number to find total WAR, as well as an expected WAR value over 20 games and 100 games. This procedure was done for all three seasons of Halo 3.
Analysis/Problems
Altogether, looking at the 2010 data paints an interesting picture. Obviously, the two top teams most of last year, Final Boss and Instinct, are well represented, with Roy, Cloud and Lunchbox in the top 5 as well as all of FB in the top 12 among players with 50 games played (and would be higher if Pistola’s TD numbers aren’t included). Because this is supposed to measure overall value, it would make sense that the teams that win the most games would have high WAR numbers. However, some players who had poor placings (e.g. Naded) were very well represented in WAR, and some players who had good placings (e.g. Enable) had a lower WAR than I would have expected. Overall, the players on top of the WAR leaderboard are very complete players, which I expected.
Obviously, this isn’t a perfect stat, and improvements are definitely possible. First, replacement level needs to be defined better. Replacement level in baseball is so well defined that simulations of a team full of replacement level players would lead to a winning percentage of around 30%. Looking at the winning percentage of Semi-Pro teams against Pro teams could lead to a better number, but the leaderboards from last year had inaccuracies (Totals wins and losses aren’t equal) and discrepancies (some teams had one more win or loss than the brackets would show), not to mention that most of the time since ’09, only low-seeded pro teams have played amateur teams. To further complicate things, if you want to use winning percentages for the 2008 season for the other seasons (because top pro teams actually played semi-pro teams), I haven’t been able to find brackets for that season (I’m not sure if any brackets can be found now…). Maybe some simulation is possible, but I can’t figure out how to do that.
Second, there aren’t any flag stats. There isn’t much I can do about that one besides ask, but those would help tremendously. Third, currently the war values are normalized such that an average player each year would have a WAR/100 of 2.8. Although this does a good job of comparing a player's performance between seasons, as 2008 was generally a more favorable environment than the 2010 season, but I'm not sure it should be kept. Finally, there’s the whole point that baseball has less teamwork than Halo, and that’s right. Trying to bring teamwork into the equation would bring in way too much gray area, and would make WAR too subjective, which is what I was trying to avoid (not to mention I would have no idea how to implement it).
What’s Next?
Currently, there are a lot of variables in the system mentioned above. I looked at the numbers for the 2009 season, separating the stats in slayer games and objective games, and finding separate WAR numbers for slayer games (SWAR) and objective games (OWAR). Because objective games have on the whole more opportunities to earn points because of objective time, the points earned in objective games are multiplied by a constant to make sure that objective games aren’t being weighted too much in the overall WAR. Once all that is done, you get a nice ugly mess shown on sheet 4.
Questions, suggestions, improvements, etc. are greatly appreciated. Sorry about the wall'o'text, but I feel there's a lot I needed to explain here.
tl;dr summary/rankings (player name/WAR/100)
2010
1. Cloud / 6.8
2. Naded / 5.6
3. Roy / 5.5
4. Ace / 5.14
5. Lunchbox / 5.07
2009
Heinz / 6.5
Ogre2 / 5.6
Legend / 5.3 (12th in split version)
Ace / 5.2
Victory X / 4.9
2009 Split
Heinz / 6.5
Ogre2 / 6.3
Cloud / 5.1
Ace / 4.94
Snipedown / 4.89
2008
Legit / 7.7
CallMeGod / 6.3 (21st in split version)
Ogre2 / 5.9
ElamiteWarrior / 5.71
SwisHaa / 5.70
2008 Split
Legit / 6.4
Ogre2 / 5.2
Poweezy / 5.1
ElamiteWarrior / 4.9
Hysteria / 4.4 (15th in combined version)
Spreadsheet found here: https://spreadsheets.google.com/ccc?key=0Aga14RVhJgKGdEEyQlprVzd4NnkzMHd0c0lSVms0N Xc&hl=en&authkey=CPCOu7IP
Table of Contents
1. Overview
2. Explanation
3. Analysis/Problems
4. What’s Next?
Overview
Basically, WAR (Wins Above Replacement) attempts to objectively determine the value a player brings to his team in all aspects of the game. Baseball has a version (which I basically copied) on many different sites, such as Fangraphs (http://www.fangraphs.com). In my version, I use all the stats available to me (K, A, D, T, games played) to attempt to do something similar.
Explanation
Using the leaderboards provided by MLG throughout the season, I compiled all the stats over the course of the season (stats are on the main page here, but you probably noticed that already with the very well labeled box on the right side of the screen that has a nice pretty picture). Then, I took the sum of all the stats of anyone who had played 30 games last season (to prevent small sample sizes from influencing the sums). Using objective time as the base, I created a point system where 1 point is defined as 1 second of time. Next I found the ratio of kills to time, using this ratio to establish the number of points a kill or death is worth (kills are positive, deaths are negative). Since assists are only registered after a player is killed, I weighted those equally to kills.
Using a simple Excel formula, I calculated the total number of points each player had; then found the number of points per game. Because these values are so similar (at least per game), it makes more sense to compare them to a set value. Although comparing everyone to the average is possible, it makes more sense to compare someone to a player that could fill-in for them. This is where the replacement part comes into play. Defining replacement level as 80% as good as an average player (more on this later), I subtracted the number of points a replacement level player would produce from all the point-per-game numbers, to get points above replacement per game (PAR/G). Since this entire system is based around objective time, and 250 seconds of time wins an objective game, I divided this PAR/G number by 250 to get wins above replacement per game (WAR/G). I then used that number to find total WAR, as well as an expected WAR value over 20 games and 100 games. This procedure was done for all three seasons of Halo 3.
Analysis/Problems
Altogether, looking at the 2010 data paints an interesting picture. Obviously, the two top teams most of last year, Final Boss and Instinct, are well represented, with Roy, Cloud and Lunchbox in the top 5 as well as all of FB in the top 12 among players with 50 games played (and would be higher if Pistola’s TD numbers aren’t included). Because this is supposed to measure overall value, it would make sense that the teams that win the most games would have high WAR numbers. However, some players who had poor placings (e.g. Naded) were very well represented in WAR, and some players who had good placings (e.g. Enable) had a lower WAR than I would have expected. Overall, the players on top of the WAR leaderboard are very complete players, which I expected.
Obviously, this isn’t a perfect stat, and improvements are definitely possible. First, replacement level needs to be defined better. Replacement level in baseball is so well defined that simulations of a team full of replacement level players would lead to a winning percentage of around 30%. Looking at the winning percentage of Semi-Pro teams against Pro teams could lead to a better number, but the leaderboards from last year had inaccuracies (Totals wins and losses aren’t equal) and discrepancies (some teams had one more win or loss than the brackets would show), not to mention that most of the time since ’09, only low-seeded pro teams have played amateur teams. To further complicate things, if you want to use winning percentages for the 2008 season for the other seasons (because top pro teams actually played semi-pro teams), I haven’t been able to find brackets for that season (I’m not sure if any brackets can be found now…). Maybe some simulation is possible, but I can’t figure out how to do that.
Second, there aren’t any flag stats. There isn’t much I can do about that one besides ask, but those would help tremendously. Third, currently the war values are normalized such that an average player each year would have a WAR/100 of 2.8. Although this does a good job of comparing a player's performance between seasons, as 2008 was generally a more favorable environment than the 2010 season, but I'm not sure it should be kept. Finally, there’s the whole point that baseball has less teamwork than Halo, and that’s right. Trying to bring teamwork into the equation would bring in way too much gray area, and would make WAR too subjective, which is what I was trying to avoid (not to mention I would have no idea how to implement it).
What’s Next?
Currently, there are a lot of variables in the system mentioned above. I looked at the numbers for the 2009 season, separating the stats in slayer games and objective games, and finding separate WAR numbers for slayer games (SWAR) and objective games (OWAR). Because objective games have on the whole more opportunities to earn points because of objective time, the points earned in objective games are multiplied by a constant to make sure that objective games aren’t being weighted too much in the overall WAR. Once all that is done, you get a nice ugly mess shown on sheet 4.
Questions, suggestions, improvements, etc. are greatly appreciated. Sorry about the wall'o'text, but I feel there's a lot I needed to explain here.
tl;dr summary/rankings (player name/WAR/100)
2010
1. Cloud / 6.8
2. Naded / 5.6
3. Roy / 5.5
4. Ace / 5.14
5. Lunchbox / 5.07
2009
Heinz / 6.5
Ogre2 / 5.6
Legend / 5.3 (12th in split version)
Ace / 5.2
Victory X / 4.9
2009 Split
Heinz / 6.5
Ogre2 / 6.3
Cloud / 5.1
Ace / 4.94
Snipedown / 4.89
2008
Legit / 7.7
CallMeGod / 6.3 (21st in split version)
Ogre2 / 5.9
ElamiteWarrior / 5.71
SwisHaa / 5.70
2008 Split
Legit / 6.4
Ogre2 / 5.2
Poweezy / 5.1
ElamiteWarrior / 4.9
Hysteria / 4.4 (15th in combined version)