The Stats Room: Week 3 QB Projections

Week 1 was a mess with quarterbacks showing inconsistent play with the average projection being off by 6.5 points. This week, the values are down to 4.7 points off. The projections sort of felt like projections. Here are the results from week two.

Where I missed is on the extreme downside (Hoyer, Kizer and Glennon). I am not sure if it is an early season anomaly but for the first two weeks, I haven't gone low enough on some guys. Maybe I shouldn't worry about them because most won't be starting except in super flex leagues. I'm not sure.

Due to some delays with the wide receivers, I couldn't get my projections up before the Thursday night game. My higher than standard projections paid off with Hoyer scoring 26 points.

Besides the downside players, the big misses evened. While I underestimated Brady, the difference was offset by almost nailing McCown's total (15.7 vs 14.9).

I thought I might have a possible answer to my high projections. I only use the weekly point spread and previous season data for the projections estimates (for now). Additionally, the model uses linear regression. I thought maybe at the change at the low and/or high extremes isn't linear.

To test the theory, I went back and found the average quarterback points and pass attempts for the number of points Vegas expected the team to score. Here are the results (min. 10 games at each point total).

While the r-square improved a small bit with a logarithmic best fit line, it's not much (less the 0.01). It's not the improvement I was looking for.

The major factor that might be missing is opposing defensive talent. I know if a team is going to score fewer points; I just don't know if it will be on the ground or the air. Incorporating defensive projections might solve this problem. This possible change will just have to wait a week or two.

Without any quarterback corrections, here are the week projections

Note: No projections for the quarterbacks in the Minnesota-Tampa game with no line and starters not set.

While I have a better understanding of quarterbacks, I will finally move onto wide receivers. Compared to quarterbacks, their projections are easier. Basically, just targets and the basic catch rate.

Note: Running back projections are a complete mess. Injuries and fickle coaches make projecting them nearly impossible. I will eventually get to them, but the level of error with them is off the charts. The talent among backs is constant so, for now, concentrate on potential touches.

The two biggest wide receiver issues I'm dealing with are rookies and incorporating in-season data. My projection database doesn't update in-season, so I eventually linked up all the in-season and pre-season target estimates. The key word is tried. I hope I got everyone, but I am not 100 percent sure. Let me know if I missed any non-rookie.

For wide receivers, I will use the projections I previously published for the base talent rates.

With my limited time this week, I just found the weight for first two weeks of targets and the previous season's targets. For this week, the previous season targets per game get weighted at 45 percent while the in-season numbers get the other 55 percent.

Another receiver with a higher RotoWire projection is Brandin Cooks (7.7 pts vs. 11.0 pts). The overall numbers aren't that far off but the ranks are significant (20th vs. 7th). The biggest difference comes down to the projection for Brady and the entire New England offense production being high. Again, I need to see if defenses can be projected and if those projections are useful for predicting offense beyond just the spread values.

Two weeks down, 15 more to go to get these basic projections streamlined. The biggest takeaway I have found is what doesn't work (a lot) and the level of unpredictability. My results have been close, but I know I can do better. See everyone next week.