By Quinn Miller
How ‘all doing, you Quinnian Studs?
We’re taking a brief step back from the specifics of the case study we’re working on this week in order to take a gander at the statistical analysis relevant to fantasy football. I’ve put together my best shot at explaining the basics of statistical analysis geared toward dynasty league fantasy domination. I hope you enjoy!
The statistical analysis necessary to fantasy football success isn’t rocket science; Jeff will testify to this: if I can do it, it can’t require too much intellectual heavy lifting. That said, the manner in which the statistical analysis becomes manageable is through the collecting and consequent winnowing of the available data into a smaller, more targeted set of data. For instance, I use average, not total, points, because it allows me to modify the outliers, and thus to take into account numbers and occasions that are unlikely to reoccur such as games that end early due to injuries, games where a player’s output is compromised by the fact that he rushed back from an injury, or games in which a player doesn’t score at all.
My goal is to compare players’ midpoint production with their average production of all players in order to determine how consistently each player produces. Midpoint production is similar to average production, only more targeted in that it is the calculation of his average production after removing the player’s top 2 highest scoring weeks, the bottom 2 lowest scoring weeks, and any weeks in which the player missed part of the game or did not score at all. Midpoint average is, in some ways, a better predictor of player performance than actual average simply because it excludes aberrant production from consideration and thus protects one’s calculations from games that skew the numbers. That said, it should be kept in mind that midpoint production is a data point meant to speak along with, and not necessarily in place of, average production.
At this point, a hard hard-core stat guy would compare the standard deviation and variance for each player, but, as Jeff will tell you, I don’t have the cerebral weaponry to do that, and I’m only hard-core. Instead, I look at the scores for each player (16 weeks minus the 2 outliers and any other games that should be excluded from calculation), and modify the outliers to reflect my own evaluation of the player’s talent and opportunity going forward; if I think three weeks of scoring should be excluded on the high end from my calculations of the midpoint average, then I do that, and compare the number I come up with as an additional data point to the average. To quote my brother, “That way I am not ignoring the production entirely, just accounting for it.”
Another way to handle the outliers is to adjust them instead of removing them. If player A averages 15 points per game, and has two games of 20 point production, one game of 25 point production, and one game of 50 point production, you change the 50 to a 25 which will still be at the top end of his point production, but won’t skew the average as much. Whichever way you go, your goal is should be to be consistent. Measuring a player’s consistency is precisely what you’re trying to do when you’re doing statistical analysis in fantasy football.
You’re probably noticing that while the analysis I’m propounding is rooted in math, I’m introducing my own biases in ways that depart from purely rationally methods. There are more complicated methods of statistical analysis that can be done in place of the biases I’ve introduced, but because A) I’m not mathematically competent to perform or explain them, and B) the vicissitudes and larks of the myriad and various fantasy football gods would immediately nullify any advantage I might gain by can attaining a greater correlation. Even the Almighty Quinn can’t predict the future. He can only notice trends in hindsight.
Here’s the Cliff Notes outline of my process:
1. Compile data
2. Rank players by average ppg
3. Remove stats from weeks where a player left the game early, and weeks where it appears a player returned from an injury prematurely.
4. Note trends for each player (e.g., became a starter in Week 6), and modify stats to reflect that, for example, by removing weeks 1-5.
5. Note any 0′s and their cause. Is it predictable? Likely happen again?
6. Sort each player’s weeks from lowest to highest total points scored.
7. Calculate average of the remaining weeks minus the two lowest and two highest scoring weeks. (Weeks with 0 points scored can be included amongst the two lowest, but not injured or non-starting weeks.)
8. Compare this “midpoint average” to the “Actual average” and see which players are affected
NOT AFFECTED: The averages are similar.
MIDPOINT LOWER: The player’s average is skewed high due to some outliers, such as an amazing game. While a 20 catch, 370 yard, 4 TD week is an awesome week, your WR is unlikely to do it more than once, and as such, including those numbers in your analysis isn’t going to help you predict their success week in and week out.
NOTE: Midpoint averages are usually lower due to the fact that they involve removing the positive outliers. This is due to 0 being the lowest possible score (indicating a low score of the “average ppg” below average) whereas there is no cutoff on the positive end, so a player can score more than “average ppg” above average.
MIDPOINT HIGHER: The player’s average is either consistently good or was skewed low due to some outliers, like a 0.
You should acquire (and keep) guys who with a higher midpoint average than their actual average for two reasons:
1. To Avoid the Ocho-Sucko Effect
Players with a higher midpoint average are more consistently good than players who have giant games. Chad Ochocinco is a prime example: he’d have a season in three games, and then do his best imitation of a Hoover for twelve or thirteen games. His statistical average would be higher than Player X, who was consistently good without being great, but on a weekly basis, Player X helped you win more.
2. They’re the Diamonds in the Rough
Players with a higher midpoint average frequently have a lower apparent value than their actual market value due to a 0 or other outlier reducing their production. This is true for players who score 0 or especially for players who were injured or didn’t become starters immediately. The perception of their production is biased, and consequently, they’re often available below market value.
So where are we?
Well, the basic idea of fantasy football is that the guy with the best team wins, and my premise is that the guy with the best team is the guy who’s best at evaluating players. Of course, after you evaluate a player, you still have to acquire the player. The above process is a method of player evaluation based on basic arithmetic, which, when combined with an understanding of norms, regression to the mean, marginal benefit, and marginal cost, can help to dictate when to buy a player and when to sell him.
Marginal benefit is the increase in points resulting from upgrading a player. The marginal cost is the cost in trade chips (i.e., picks, players, and/or $$$) I have to give to upgrade the player. Generally speaking, the marginal benefit of a stud over a consistent veteran is not worth the marginal cost of acquiring the stud, and as such, one is best served by keeping his depth and improving his team as a whole rather than acquiring the stud at the cost depth, picks, and/or cap-space. Of course, there are always exceptions:
1. Shallow leagues
If you play in a ten-team league and/or a league in which there are no IDP’s, then studs are a must.
2. Your roster is rife with inconsistent veterans
If you’re relying on Carolina Steve Smith and Shonn Greene to get you by, then the consistency of a Rodney White or an Alfred Morris is worth the premium.
3. You’ve got nowhere else to go
If your stable of RBs consists of the third-string backups to Arian Foster, Adrian Peterson, and Doug Martin, then you should probably make a trade.
As a general rule, try to minimize your marginal cost. Instead of trading for studs (high marginal cost), trade for top prospects or future draft picks likely to be in the top 5 that will enable you to buy a stud. Trading an aging veteran for a 1st round pick from a team that you think will lose is always a good gambit. You’ve probably noticed that very few people have the capacity to be honest with themselves, and consequently, it shouldn’t come as a surprise that not many owners are aware when their team is bad. It’s not unheard of for a team to trade a pick that turns out to be a top 3 pick for a player that they would never have traded a top 3 pick for to begin with, simply because the owner couldn’t evaluate his team objectively.
It’s important to do statistical analysis, but it’s equally important, in the context of a dynasty league, not to get so married to it that you fail to adjust to the fact that fantasy football is a consumer market. Your outlook should be as geared toward economics as it is toward stats so that you can maximize the exchange you get when you buy or trade. As with any market, you achieve the maximum return on investment (ROI) when you buy low and sell high, so you should always try to keep your finger on the pulse of the sense of supply and demand operating in your league market. Generally speaking, studs are low in supply, and in high demand, and as such, have a high price attached to them. It’s always better to buy a prospect low than to buy a stud high. Moreover, if you keep your eye on the market, you’ll have the opportunity to sell your studs at the highest market value and still better your team, even if the stud continues to produce.
This last point is paramount: a stud’s production is of great value to your team, however, because people are irrational and the market is driven by people, there may come a time where your stud’s value on the market is inflated beyond his statistical value, and in such cases, trading the stud is better for your team than keeping them. You should always keep your eye out for such players. Trent Richardson is an excellent example: he’s a damn good player who was hyped so hard that there was almost no way he was going to produce the statistical equivalent of his trade value. One owner I know got Bradford, a 2013 1st, and Jimmy Graham for T. Rich and a 2nd.
What do you think, Readers? Give me a holler!