In a previous blog post, we discussed the relationship between Vegas over/under lines and actual points scored.  We concluded that games are far more likely to be explosive, with high levels of scoring, when there is a high over/under line.  Week 1 was a perfect example of this.  The Colts vs Lions and Raiders vs Saints games, with lines of 51.5 and 50.5, respectively, scored 74 and 69 points, massive point totals that generated huge fantasy scores.  The Colts and Lions combined for 898 total yards, and Drew Brees hit Brandin Cooks for a 98 yard touchdown pass.  These scores were huge, but not unexpected given how high their lines were.

This week we are rolling out our Tout Rankings and Player Rankings for Week 1 of the Draftkings season. The first system to understand is our Tout Rankings. Leading up to the beginning of the NFL season, we gathered DraftKings projections from six of the leading projection sites. This list includes:

  • Numberfire.com
  • Swishanalytics.com
  • Fantasyinsiders.com
  • Fantasylabs.com
  • 4for4.com
  • Fantasyalarm.com

From there, we cleaned up the data and combined it with the actual Draftkings results from the week. If you would like access to this data set, it is available under the spreadsheets tab with a premium membership. Finally, we analyzed the projections using four different metrics that test accuracy. The data came out as follows:

4for4 fantasyalarm fantasyinsiders fantasylabs numberfire swishanalytics
MSE 42.167 49.891 32.212 55.080 35.993 39.737
MAE 4.6005 5.2462 3.7637 5.5322 3.9514 4.2675
ME 0.2637 0.1945 0.1730 0.8943 0.0464 0.14793
RS 0.6524 0.6651 0.6403 0.7346 0.6483 0.5963

The rows are labeled as follows:

  • MSE: Mean Squared Error
  • MAE: Mean Absolute Error
  • ME: Mean Error
  • RS: R Squared

From here, we can take a look to see the pros and cons of each projection system. Across the board, numberfire comes away with the best values. The low MSE and ME mean that the average margin of error is on average pretty low. The very low ME means that the projection system is balanced, with error high as common as error low. And finally, the decent R^2 is an overall good sign that the data trends the correct way.

Additionally, fantasyinsiders had very impressive numbers for MSE and MAE where they were the lowest of both sites in each category. While their projections tended a bit high, they were overall fairly low on their error and most of their projections were pretty close to the actual value. This is also supported by a decently high R^2 value.

Finally, one of the most interesting sites of the week was fantasylabs. Their projection model had the highest error across the board on MSE, ME, and MAE, meaning that they heavily overestimated their projections for most players. Interestingly enough though, they also had the highest R^2, a metric that is most accurate as it approaches 1. This means that while they overestimated values, they had a more correct trendline of how player performance stacked up and understood the relationship between the field. Also note that on the fantasylabs site, there is an option to toggle between different models. These do not change the projection values, just the importance of each value in the lineup construction.

Some things to notice about the data in general, the ME is always positive. This means that these sites tend to overpredict the value of the data. Also, there is not too much variability in the data overall, no site is either exceptionally good or exceptionally bad. Next week we will gather more data and talk about the strengths and weaknesses of each site through Week 2.

The Advanced Sports Analytics DFS Player Ranking system measures objective skill in making DFS picks.  Low-owned players who score well demonstrate skill – you saw something that few others did.  Low-owned players who score poorly demonstrate a lack of skill – you chose a player that was so obviously bad that few others picked him.  Highly owned players are generally the consensus – everyone thought along the same lines, and there was really no differentiating skill involved.

Advanced Sports Analytics has developed a formula to measure the skill in picking players.  While other systems benefit high volume players, awarding points for lineups that perform well, our system measures overall skill by penalizing players for their objectively bad plays.  Therefore, it isn’t possible to climb our rankings by simply entering the maximum number of lineups – its skill that counts.