Week 3 Computer Rankings
Three weeks into the new age, I've got just enough data that I might as well pull back the curtain. Our old friend Estimated Win Percentile is back on the rankings table. Remember that there still is not nearly as much data behind this as in the past, but I always feel a lot better about three games then two.
For the rookies among us and as a refresher for the rest of us, Estimated Win Percentile is what I use to rank teams and make predictions. In theory, EWP is the chance that a given team would beat an average team at a neutral site. For each game I calculate the total points the team scored (offense + defense - home field) and then I assign that score a percentile rank in relation to all total scores over the past two years (usually a rolling 28 FLOF regular season games, but because of the new scoring I wiped the slate clean and am just using scores from this year). For each team I then find their median percentile score and their average percentile score over their last 4 games. Then I average their median and average together and that is their EWP.
Weekly predictions compare the two teams' EWPs and take the 3-point home field bonus into account (right now I estimate that 3 points is worth 5% EWP). I have a chart that predicts everyone's chance to win each game remaining on the schedule taking home field but not byes or any other player related data into account. That informs the predictions part of the table above.
Luck is the difference between how many games a team has won so far and how many the computer would have expected the team to have won so far based on that team's current EWP. So ICL's -.9 Luck is almost entirely about his week 2 defeat despite a very solid 110 total points. Bill has led the league in this metric two years in a row and historically has always been judged by the computer to be a lucky team. Last year Bill won 10 games despite a .346 EWP which the computer thought was worth 4.8 wins.
The color coding of the numbers groups scores into tiers by percentile rank. Each tier is 1/6th. So you would expect to see 2 scores from each tier every week. Obviously we rarely do, but that would be a perfectly average week. Red = top 6th. Blue = 2nd / 6th. Green is 3rd / 6th or just above median. Yellow is 4th / 6th or just below median. Brown(ish) is 5th / 6th. Black is bottom 6th. Roughly I consider Red and Blue good, Green and Yellow average, and Brown and Black are bad.
Predictions for Week 4
Computer was .500 again in week 3. Below, I've included the computer's predicted chance that its favored team will win. Lagers and Dynasty feature two of the computer's top ranked teams right now. Bill is favored by less than a percentage point.
Mimes @ Rednecks (86%)
Wolfpack @ Malleus (55%)
Drow (55%) @ Old No. 7
Bronies (54%) @ Camp Adventure
Lagers @ Dynasty (51%)
Bill (> 50%) @ Hamburglers
For the rookies among us and as a refresher for the rest of us, Estimated Win Percentile is what I use to rank teams and make predictions. In theory, EWP is the chance that a given team would beat an average team at a neutral site. For each game I calculate the total points the team scored (offense + defense - home field) and then I assign that score a percentile rank in relation to all total scores over the past two years (usually a rolling 28 FLOF regular season games, but because of the new scoring I wiped the slate clean and am just using scores from this year). For each team I then find their median percentile score and their average percentile score over their last 4 games. Then I average their median and average together and that is their EWP.
Weekly predictions compare the two teams' EWPs and take the 3-point home field bonus into account (right now I estimate that 3 points is worth 5% EWP). I have a chart that predicts everyone's chance to win each game remaining on the schedule taking home field but not byes or any other player related data into account. That informs the predictions part of the table above.
Luck is the difference between how many games a team has won so far and how many the computer would have expected the team to have won so far based on that team's current EWP. So ICL's -.9 Luck is almost entirely about his week 2 defeat despite a very solid 110 total points. Bill has led the league in this metric two years in a row and historically has always been judged by the computer to be a lucky team. Last year Bill won 10 games despite a .346 EWP which the computer thought was worth 4.8 wins.
The color coding of the numbers groups scores into tiers by percentile rank. Each tier is 1/6th. So you would expect to see 2 scores from each tier every week. Obviously we rarely do, but that would be a perfectly average week. Red = top 6th. Blue = 2nd / 6th. Green is 3rd / 6th or just above median. Yellow is 4th / 6th or just below median. Brown(ish) is 5th / 6th. Black is bottom 6th. Roughly I consider Red and Blue good, Green and Yellow average, and Brown and Black are bad.
Predictions for Week 4
Computer was .500 again in week 3. Below, I've included the computer's predicted chance that its favored team will win. Lagers and Dynasty feature two of the computer's top ranked teams right now. Bill is favored by less than a percentage point.
Mimes @ Rednecks (86%)
Wolfpack @ Malleus (55%)
Drow (55%) @ Old No. 7
Bronies (54%) @ Camp Adventure
Lagers @ Dynasty (51%)
Bill (> 50%) @ Hamburglers
0 Comments:
Post a Comment
Subscribe to Post Comments [Atom]
<< Home