I’m willing to bet if you read this post, then you play some form of simulatory sports game. OOTP, Madden, 2K, etc–pretty popular. All sorts of fantasy sports, too.
As an avid fantasy football player, I know that there have been weeks where I have left huge amounts of points on my bench and wracked my head over it. But that’s the exception, not the norm.
Each week in fantasy football, when the week concludes, each team has both an actual points score and a potential points score. Actual points, which I’ll call PF from now on, are pretty obvious–it’s how many points that team scored that week. Potential points (PPF) are a little bit more complicated–it’s how many points that team would have scored had they started their best possible lineup that week. So when you leave 27 points on the bench and start a 2-point guy instead, you will have 25 added to your PPF.
The ratio of PF/PPF is called efficiency, and I’m trying to use this metric to create a descriptive statistic, aiming at using past efficiency results to describe a team’s most likely PF value for a given week.
Currently, I’m calling that statistic Projected Points For (PROPF). Another crappy name, right? Projected implies projecting the future. But oh well. That’s not the hill I’ll die on.
The equation I’m using for PROPF is PROPF=PPF*((ET+EA)/2). ET stands for a team’s historical efficiency value (to account for owner skill) and EA stands for average efficiency–meaning the average efficiency of hundreds of teams, over hundreds of weeks. Using these two values is a way to do two things: one, allow for the fact that historical owner average is a huge driving factor in terms of efficiency; two, allow for the fact that owners can always increase/decrease in skill, more than likely to the overall mean.
Through 7 weeks in a single league, I’ve got 84 data points for EA, and 7 data points for each team. As an overall average, PROPF has estimated that each team will, on average, outscore their PROPF value by between 1-2 points per week. It assumes a buttload of regression–the top teams in the league in terms of PF outscore their PROPF by an insane number–roughly 12-15 points/game. In the small sample size, there are no terrible teams in terms of PF-but the worst team (sadly, mine) in terms of PF is underscoring its PROPF by less than a point/game.
So I think I’m on to something here. But the fact remains that the owner who’s got a PF of 15 points/game over his PROPF has a crazy average efficiency of 90+%–which is 10% over the league average efficiency. And he doesn’t look to be slowing down after 7 weeks and 7 data points.
I really need to find SOME coefficients to throw in front of ET and EA to account for this guy. He’s singlehandedly destroying my model–which is a good thing because I need to change it based on this.
But what if I’m right and he should regress? Is there a way to prove my rightness in his overachievement?
This, much more so than eRPA, is tough for me. There are no current fantasy football statistics to compare this to. So I don’t know how to find those coefficients.
Food for thought. See you when I come up with a new stat or I can figure out a way to improve a current one.
Stats by Avery 