Leveraging the Wisdom of Crowds for Fantasy Football

WARNING: This has nothing to do with national security, but is nonetheless awesome.

This evening I will be participating in that great annual tradition which marks the transition from Summer to Fall: the fantasy football draft.

A large part of having a successful fantasy football draft is being able to adjudicate the value of a player more accurately than your opponent. This year, I decided to take the “wisdom of the crowds” approach to evaluating players by analyzing large amounts of mock draft data from FantasyFootbalCalculator.com. This website is a great repository of draft data, which may be viewed as a proxy for on how the “market” of fantasy managers value different players in aggregate.

Using a short R script I collected a large sample of mock drafts with the same number of managers as my league, then crunched some very simple statistics on the data to assist me in my own draft strategy. Specifically, I collected the most recent 2,000 completed mock drafts with the same number of managers as my league, and computed the following statistics:

• Mean draft position
• Median draft position
• Standard deviation of draft position
• Median absolute deviation of draft position
• Percent times player was drafted

While very simple, these statistics can provide interesting insight into how different players are being judged. For example, the top ten players are very stable over time, however, there are slight changes in the variance of draft position for these players that are telling. Consider the results for Maurice Jones-Drew and Ray Rice in my most recent data pull. These players are consistently ranked third and fourth in overall draft position, but the standard deviation for Jones-Drew in aggregate is slightly higher. These differences in variance can be interpreted as the market having slightly less confidence that Jones-Drew’s true value is the third overall pick than it does Rice is the fourth.

Another interesting insight that can be drawn from this data is identifying the players that the market has the most difficultly evaluating. Because the distributions in these data are rarely single-peaked, I use the median absolute deviation (MAD) of draft position to identify the players the market has most inconsistently ranked. Below, I have plotted players’ median draft position against their MAD, and labeled those players whose MAD is in the 95th percentile.

This chart is interesting for several reasons. First, the smoothed fit line shows a gradual decrease in the confidence of a player’s market valuation (depicted by the upward slope) the later a player is drafted, which peaks around the 110th player drafted and then confidence goes back up in later rounds. This makes sense, as players drafted early are those the market has a more confidence valuation for, then in later rounds that confidence goes down until the final rounds where late draft picks are more consistent. Also, as of this morning it appears Braylon Edwards is the player most difficult for managers to evaluate; perhaps this is evidence of a “Hard Knocks Effect.”

There is much more that could be done with the data, and I won’t give away all my secrets here—I have to some something for comparative advantage. As an added bonus, however, the R code also pulls individual player performance data from AdvancedNFLStats.com and outputs them as CSV files. Now, get out there and own your league with data!