This article is part of our The Stats Room series.
For evaluating last week's projections, we will use the total number of points in which the standard RotoWire and my projections were off. Additionally, I just took the average of all the projected values (15.8 points) and saw how far they were off.
When I posted the projections last week, I inserted the projected RotoWire points but later discovered it uses 0.05 points per pass yard. With most owners using 0.04 points per yard passing, I had to recalculated the values.
For reference, I had two systems for estimating pass attempts, the Vegas line and previous season attempts. I found the previous season attempts to be most accurate. I use a final hybrid model that uses the Vegas line value if the quarterback threw very little or none in the previous season.
I compared the average amount each value was off from the actual results and the standard deviation for the differences. The total will show an overall accuracy while the standard deviation will show grouped up are the final differences.
For example, if one system was off zero points for 16 quarterbacks and 14 for the other
For evaluating last week's projections, we will use the total number of points in which the standard RotoWire and my projections were off. Additionally, I just took the average of all the projected values (15.8 points) and saw how far they were off.
When I posted the projections last week, I inserted the projected RotoWire points but later discovered it uses 0.05 points per pass yard. With most owners using 0.04 points per yard passing, I had to recalculated the values.
For reference, I had two systems for estimating pass attempts, the Vegas line and previous season attempts. I found the previous season attempts to be most accurate. I use a final hybrid model that uses the Vegas line value if the quarterback threw very little or none in the previous season.
I compared the average amount each value was off from the actual results and the standard deviation for the differences. The total will show an overall accuracy while the standard deviation will show grouped up are the final differences.
For example, if one system was off zero points for 16 quarterbacks and 14 for the other 16, they averaged points with a standard deviation of 7.1. Now, if another system was off by seven points for everyone one, it still averaged seven points but had a standard deviation of zero. It was more consistent with its projections.
Note: Rookie DeShone Kizer was not projected last week.
Here are the results from Week 1.
PLAYER | VEGAS LINE | PREV PASS ATT | HYBRID | ROTO WIRE | AVG | ACTUAL | VEGAS LINE | PREV PASS ATT | HYBRID | ROTO WIRE | AVG |
Aaron Rodgers | 19.6 | 20.3 | 20.3 | 21.8 | 15.8 | 16.5 | 3.1 | 3.7 | 3.7 | 5.3 | 0.7 |
Alex Smith | 18.2 | 17.2 | 17.2 | 12.0 | 15.8 | 31.0 | 12.8 | 13.8 | 13.8 | 19.0 | 15.2 |
Andy Dalton | 17.4 | 17.5 | 17.5 | 16.3 | 15.8 | -3.0 | 20.4 | 20.5 | 20.5 | 19.3 | 18.8 |
B.Roethlisberger | 16.4 | 16.5 | 16.5 | 17.6 | 15.8 | 15.7 | 0.7 | 0.8 | 0.8 | 1.9 | 0.1 |
Blake Bortles | 15.4 | 17.5 | 17.5 | 11.5 | 15.8 | 10.3 | 5.1 | 7.2 | 7.2 | 1.2 | 5.5 |
Brian Hoyer | 15.9 | 15.0 | 15.0 | 11.4 | 15.8 | 3.2 | 12.6 | 11.8 | 11.8 | 8.2 | 12.6 |
Cam Newton | 19.6 | 18.5 | 18.5 | 18.2 | 15.8 | 13.1 | 6.5 | 5.4 | 5.4 | 5.0 | 2.7 |
Carson Palmer | 15.8 | 17.0 | 17.0 | 16.8 | 15.8 | 9.0 | 6.9 | 8.0 | 8.0 | 7.8 | 6.8 |
Carson Wentz | 15.6 | 16.4 | 16.4 | 14.5 | 15.8 | 18.9 | 3.2 | 2.5 | 2.5 | 4.3 | 3.1 |
Dak Prescott | 19.0 | 15.9 | 15.9 | 16.5 | 15.8 | 17.1 | 1.9 | 1.2 | 1.2 | 0.7 | 1.3 |
Derek Carr | 17.3 | 16.8 | 16.8 | 17.3 | 15.8 | 18.7 | 1.4 | 1.9 | 1.9 | 1.4 | 2.9 |
Drew Brees | 16.6 | 18.9 | 18.9 | 17.4 | 15.8 | 15.6 | 1.0 | 3.3 | 3.3 | 1.7 | 0.2 |
Eli Manning | 15.4 | 16.1 | 16.1 | 16.7 | 15.8 | 6.7 | 8.7 | 9.4 | 9.4 | 10.0 | 9.1 |
Jared Goff | 14.9 | 12.8 | 12.8 | 11.9 | 15.8 | 16.4 | 1.5 | 3.7 | 3.7 | 4.5 | 0.6 |
Joe Flacco | 15.2 | 17.5 | 17.5 | 13.9 | 15.8 | 6.8 | 8.3 | 10.7 | 10.7 | 7.0 | 9.0 |
Josh McCown | 14.7 | 14.9 | 14.9 | 4.9 | 15.8 | 9.5 | 5.3 | 5.4 | 5.4 | 4.6 | 6.3 |
Kirk Cousins | 17.7 | 18.3 | 18.3 | 18.6 | 15.8 | 10.6 | 7.1 | 7.7 | 7.7 | 8.0 | 5.2 |
Marcus Mariota | 19.4 | 16.0 | 16.0 | 17.9 | 15.8 | 18.8 | 0.6 | 2.9 | 2.9 | 1.0 | 3.0 |
Matt Ryan | 19.2 | 17.7 | 17.7 | 19.9 | 15.8 | 17.9 | 1.2 | 0.3 | 0.3 | 2.0 | 2.1 |
Matthew Stafford | 17.7 | 17.6 | 17.6 | 17.2 | 15.8 | 27.1 | 9.4 | 9.5 | 9.5 | 9.9 | 11.3 |
Mike Glennon | 17.0 | 7.8 | 17.0 | 12.5 | 15.8 | 12.5 | 4.5 | 4.7 | 4.5 | 0.0 | 3.3 |
Philip Rivers | 15.8 | 16.4 | 16.4 | 14.5 | 15.8 | 17.7 | 1.9 | 1.2 | 1.2 | 3.2 | 1.9 |
Russell Wilson | 21.1 | 19.8 | 19.8 | 17.4 | 15.8 | 8.3 | 12.8 | 11.4 | 11.4 | 9.1 | 7.5 |
Sam Bradford | 16.8 | 17.0 | 17.0 | 14.1 | 15.8 | 25.5 | 8.7 | 8.6 | 8.6 | 11.4 | 9.7 |
Scott Tolzien | 17.4 | 10.7 | 10.7 | 7.3 | 15.8 | 1.3 | 16.1 | 9.3 | 9.3 | 6.0 | 14.5 |
Tom Brady | 17.3 | 17.9 | 17.9 | 22.6 | 15.8 | 10.7 | 6.6 | 7.2 | 7.2 | 11.9 | 5.1 |
Tom Savage | 16.2 | 12.6 | 12.6 | 7.7 | 15.8 | -1.5 | 17.7 | 14.1 | 14.1 | 9.2 | 17.3 |
Trevor Siemian | 15.6 | 15.5 | 15.5 | 8.9 | 15.8 | 22.7 | 7.1 | 7.2 | 7.2 | 13.8 | 6.9 |
Tyrod Taylor | 19.8 | 17.7 | 17.7 | 16.0 | 15.8 | 18.8 | 1.0 | 1.1 | 1.1 | 2.8 | 3.0 |
Average | 17.2 | 16.3 | 16.6 | 14.9 | 15.8 | 13.7 | 6.7 | 6.7 | 6.7 | 6.6 | 6.4 |
Standard Dev | 1.7 | 2.6 | 2.1 | 4.2 | 0.0 | 8.1 | 5.4 | 4.8 | 4.8 | 5.1 | 5.3 |
Overall, it was close across the board with all my misses being on the down side (Dalton, Flacco, Savage, Wilson and Hoyer). Alex Smith was the only quarter back to be 10 points over his projection while five were 10 below. Historically, the average was 15.1 points, so I was going to be a bit high.
The main takeaway I found, no matter the projection system, was that most quarterbacks fell into a range of outcomes of +/- 7 points. While the starting at a higher projected value is better, the weekly variance is huge and unpredictable.
One issue I had with the first week was an inability to project rookie quarterbacks. I found the historic composite value for all rookies and used it as a projection. I tried to incorporate draft position to see if the higher the pick, the better the performance. I found no correlation.
While I was doing the quarterback values, I did the same for wide receivers and running backs. Again, I found no correlation with draft pick and production. I plan on examining these values during the offseason to digger deeper for signs of talent. For now, here are the rate stats I will use.
POS | COMP% | Y/REC | TD/REC | INT/ATT | RUSH ATT/PASS ATT | RUSH Y/ATT | TD/RUSH | FUM/TOUCH |
QB | 56.5 | 11.3 | 0.057 | 0.034 | 0.114 | 4.2 | 0.045 | 0.009 |
WR | 53.8 | 13.2 | 0.078 | x | x | x | x | 0.018 |
RB | 72.4 | 7.9 | 0.025 | x | x | 4.1 | 0.007 | 0.008 |
Week 2 Projections
With the ability to project rookies, here are the projected numbers for Week 2. For the second week, I tried to figure out if the first week's pass attempts would help in projecting the second week. They really didn't. With no in-season adjustment, here are the Week 2 projections using the following scoring system.
• Passing Yards: 0.04
• Passing TD: 4
• Turnover: -2
• Rushing Yards: 0.1
• Rushing TD: 6
PLAYER | VEGAS LINE | PASS ATT | HYBRID | ROTOWIRE | DIFF |
Aaron Rodgers | 20.0 | 20.3 | 20.3 | 22.4 | -2.1 |
Alex Smith | 18.6 | 17.2 | 17.2 | 14.1 | 3.1 |
Andy Dalton | 16.5 | 17.5 | 17.5 | 16.9 | 0.6 |
Ben Roethlisberger | 16.4 | 16.5 | 16.5 | 16.9 | -0.3 |
Blake Bortles | 15.9 | 17.5 | 17.5 | 13.4 | 4.1 |
Brian Hoyer | 14.0 | 15.0 | 15.0 | 9.5 | 5.5 |
Jacoby Brissett | 16.6 | 10.9 | 16.6 | 11.6 | 5.0 |
Cam Newton | 18.4 | 18.5 | 18.5 | 19.8 | -1.3 |
Carson Palmer | 15.1 | 17.0 | 17.0 | 15.1 | 1.8 |
Carson Wentz | 15.4 | 16.4 | 16.4 | 14.5 | 1.9 |
Dak Prescott | 18.1 | 15.9 | 15.9 | 18.2 | -2.3 |
Derek Carr | 15.3 | 16.8 | 16.8 | 18.0 | -1.2 |
Deshaun Watson | 12.1 | x | 12.1 | 11.9 | 0.3 |
DeShone Kizer | 12.2 | x | 12.2 | 9.6 | 2.6 |
Drew Brees | 17.0 | 18.9 | 18.9 | 20.3 | -1.4 |
Eli Manning | 15.0 | 16.1 | 16.1 | 15.6 | 0.5 |
Jameis Winston | 15.5 | 16.1 | 16.1 | 18.3 | -2.3 |
Jared Goff | 14.9 | 12.8 | 12.8 | 12.1 | 0.7 |
Jay Cutler | 16.8 | 14.2 | 14.2 | 13.7 | 0.5 |
Joe Flacco | 14.5 | 17.5 | 17.5 | 14.7 | 2.8 |
Josh McCown | 14.0 | 14.9 | 14.9 | 4.5 | 10.3 |
Kirk Cousins | 17.3 | 18.3 | 18.3 | 18.5 | -0.2 |
Marcus Mariota | 18.0 | 16.0 | 16.0 | 18.9 | -2.9 |
Matt Ryan | 19.9 | 17.7 | 17.7 | 19.9 | -2.3 |
Matthew Stafford | 16.5 | 17.6 | 17.6 | 16.0 | 1.6 |
Mike Glennon | 15.9 | 7.8 | 15.9 | 11.4 | 4.5 |
Philip Rivers | 16.2 | 16.4 | 16.4 | 17.3 | -0.9 |
Russell Wilson | 19.4 | 19.8 | 19.8 | 20.6 | -0.8 |
Sam Bradford | 15.9 | 17.0 | 17.0 | 12.7 | 4.3 |
Tom Brady | 17.9 | 17.9 | 17.9 | 24.0 | -6.1 |
Trevor Siemian | 15.3 | 15.5 | 15.5 | 8.2 | 7.2 |
Tyrod Taylor | 19.1 | 17.7 | 17.7 | 15.2 | 2.5 |
Average | 16.5 | 15.4 | 1.1 |
The previous season pass attempt model was used for Jacoby Brissett and Mike Glennon (few 2016 attempts) and DeShone Kizer and Deshaun Watson (rookies). Understandably, the system hates rookies.
The biggest difference is my projections being on the high side (about one point on average) mainly because no one is projected to have a value less than 12 points. This is the one disparity I've seen each of the first two weeks. While sub-12 point weeks exist, they happen just as much to both good and bad quarterbacks.
That's it for this week and hope for some improvements for Week 3. For now, I am working on wide receiver, which are showing some promising signs when it comes to target projections.