One thing that I have put a lot of thought into over the last few years is coming up with a system to "penalize" players that are inconsistent fantasy scorers from week to week. In other words, we would rather have a guy that scores 10 points a week than a guy who scores 5 one week, and 15 the next.
There are two reasons for this: first, we want a team that is going to score consistently across the season, and especially into the post season. We want a team that has a chance to win every week, not one that gets blown out one week and destroys the competition the next. Also, it is my belief that players that are more consistent from week to week will also be more consistent year to year, and thus will be lower risks.
But how do we we quantify this an put it into our projections?
I've started down this path by including standard deviation in all my projections. The idea is to subtract some portion of the standard deviation from the players' projections, so that the more consistent players will rise to the top of the projections. But how much do we subtract?
One of the beauties of the standard deviation is that it grows more slowly than the variances in a player's performance. Consider a player that scores 8 fantasy points every week. He finishes the season with 128 points and a standard deviation of 0. A player that scores 8 points every week, but 9 in the last game of the year, winds up with 129 points and a standard deviation of 0.25. So if we subtract standard deviation from the total, we get 128.75. The second player is punished for not being totally consistent, but he still grades out better than the first player, as he should.
Now consider a player who scores 16 points for 8 weeks, but 0 points for the other 8 weeks. He also scores 128 points on the season, but has a standard deviation of 8.3. Subtracting the standard deviation from the total results in an adjusted total of 119.74. That puts him lower than the others, but still above a player that scores 7 points each and every week. Subtracting double the standard deviation would put him below the 7 points a week player; however, given some the standard deviations we are seeing in our projections, that would make some players projections go all the down to, or even below, zero.
So, for the first pass of projections, I will be subtracting 1 standard deviation from all projections, and we will see what the results are.
Subscribe to:
Post Comments (Atom)
2 comments:
This is a very interesting concept. Just out of curiosity, when you subtracted each players standard deviation from their projected point totals for the year, what were the standard deviations based on? Were they based on the prior year or years? Did you calculate these yourself or are historical standard deviations by player and year available somewhere?
Good question. For my 2008 projections, I used the game-to-game deviations from the 2007 season, since that is all that was in my database at the time. For 2009, my plan is to use standard deviations from the last 3 years (2006, 2007, and 2008), if I ever get off my butt and put it all together. Of course, this deviations will have to be tweaked for players that have changed roles in the past years - for instance, I am sure that Micheal Turner had a much higher standard deviation when he was in San Diego as a backup than he had this year in Atlanta, so I would probably only use his 2008 standard deviation for his 2009 projections.
Post a Comment