Data scientist, physicist, and fantasy football champion

Week 2 QB Predictions

Week 1 is almost certainly a statistical outlier, with Brees, Brady, and Rodgers all outside the top 10 and the top 3 QBs as Smith, Stafford, and Bradford. That said, Model A was far and away the best model last week. I still like Model F (probably because it’s what I’m studying these days), but it was just buried in the middle of the pack with the rest of them.

Nothing to do but move forward and assume that everyone will regress to their statistical means. Which they will. That’s what statistics is all about.

In week 2 I’ll run the same models I did in Week 1. I still wouldn’t trust Model F yet; it’s kind of my minority report model right now. It said Brady would be #5 (instead of #1 or #2), and had Stafford fairly high on the list, but Model A guessed more of the top 15 QB correctly, and that was what really mattered last week. You’ll see in a sec that it’s still a little weird, probably because I don’t understand how to make a model using Partial Least Squares.

Let’s get moving:

ModelTypePlayerScoreOppScoreScoreOppScoreHomeAwayHomeAwayPlayerOppDataYears
ALMXXXXXX 2015-2017
BLMXXXXXXX2015-2017
CLMXXX 2015-2017
DGibbs SamplerXXX X2015-2017
EGibbs SamplerXXX XXX2015-2017
FPLSXXXXX X2015-2017

Model A:

.Model A-1.png


Terms:

Player
Score
OppScore
Score*OppScore
Home/Away
Home/Away*Player (Some players have a bit more of a problem playing at home or away. Looking at you, Roethlisberger…)
Data years: 2015 - 2017

Model B:

.Model B-1.png


Terms:

Player
Score
OppScore
Score*OppScore
Home/Away
Home/Away*Player
Opponent
Data years: 2015 - 2017

Model C:

.Model C-1.png


Terms:

Player
Score
OppScore
Data years: 2015 - 2017

Model D

.Model D-1.png


Terms:

Player
Score
OppScore
Opponent
Data years: 2015 - 2017

Bayesian (Gibbs sampler)

Model E:

.Model E-1.png


Terms:

Player
Score
OppScore
Score*OppScore
Home/Away
Home/Away*Player
Opponent
Data years: 2015 - 2017

Bayesian (Gibbs sampler)

Model F:

.Model F-1.png


Terms:

Player
Score
OppScore
Score*OppScore
Score^2
OppScore^2
Home/Away
Home Team (which stadium they’re in)
Opp
Team (some QBs play better under different teams)
day (Thu/Sun/Mon)
Data years: 2015 - 2017

Model details: Partial Least Squares fit using the caret package. I’m just throwing data science at the wall here and seeing what sticks. I need to work on variable selection, and if you’re reading this note past week 1 it’s because I still need to work on variable selection. For now, just consider Model F to be experimental.

Weekly model summary

RankABCDEF
1Ben RoethlisbergerDrew BreesAaron RodgersTom BradyTom BradyTom Brady
2Tom BradyTom BradyTom BradyDrew BreesDrew BreesAaron Rodgers
3Aaron RodgersBen RoethlisbergerRussell WilsonAaron RodgersBen RoethlisbergerDrew Brees
4Drew BreesAaron RodgersDrew BreesCam NewtonAaron RodgersMatt Ryan
5Cam NewtonJoe FlaccoCam NewtonRussell WilsonCam NewtonRussell Wilson
6Russell WilsonRussell WilsonDerek CarrMatt RyanEli ManningCam Newton
7Joe FlaccoCam NewtonMatt RyanBlake BortlesRussell WilsonBen Roethlisberger
8Matt RyanEli ManningBen RoethlisbergerDerek CarrJoe FlaccoDerek Carr
9Alex SmithMatt RyanKirk CousinsJameis WinstonMatt RyanEli Manning
10Eli ManningAlex SmithJameis WinstonEli ManningDerek CarrJameis Winston
11Marcus MariotaBlake BortlesBlake BortlesBen RoethlisbergerBlake BortlesKirk Cousins
12Derek CarrDerek CarrMarcus MariotaPhilip RiversAlex SmithDak Prescott
13Kirk CousinsKirk CousinsDak PrescottKirk CousinsJameis WinstonMarcus Mariota
14Blake BortlesJameis WinstonEli ManningJoe FlaccoKirk CousinsBlake Bortles
15Jameis WinstonPhilip RiversPhilip RiversMatthew StaffordPhilip RiversPhilip Rivers
16Philip RiversMarcus MariotaCarson PalmerAlex SmithMatthew StaffordMatthew Stafford
17Matthew StaffordMatthew StaffordMatthew StaffordMarcus MariotaMarcus MariotaCarson Palmer
18Carson PalmerDeShone KizerAlex SmithCarson PalmerCarson PalmerAlex Smith
19Tyrod TaylorBrian HoyerJoe FlaccoAndy DaltonAndy DaltonJoe Flacco
20Dak PrescottCarson PalmerAndy DaltonDak PrescottJay CutlerAndy Dalton
21DeShone KizerAndy DaltonTyrod TaylorSam BradfordBrian HoyerTyrod Taylor
22Brian HoyerTrevor SiemianSam BradfordTyrod TaylorTyrod TaylorSam Bradford
23Andy DaltonTyrod TaylorJay CutlerCarson WentzTrevor SiemianJay Cutler
24Jay CutlerJay CutlerCarson WentzJosh McCownSam BradfordCarson Wentz
25Trevor SiemianSam BradfordDeShone KizerJay CutlerJosh McCownTrevor Siemian
26Carson WentzJosh McCownBrian HoyerTrevor SiemianCarson WentzBrian Hoyer
27Sam BradfordCarson WentzTrevor SiemianBrian HoyerDak PrescottJared Goff
28Josh McCownJared GoffJosh McCownJared GoffJared GoffJosh McCown
29Jared GoffDak PrescottJared GoffDeShone KizerMike GlennonJacoby Brissett
30Jacoby BrissettJacoby BrissettJacoby BrissettDeshaun WatsonDeshaun WatsonDeShone Kizer
31Mike GlennonMike GlennonMike GlennonMike GlennonDeShone KizerMike Glennon
32Deshaun WatsonDeshaun WatsonDeshaun WatsonJacoby BrissettJacoby BrissettDeshaun Watson

 

Conclusions

Models A and F were the best last week, so let’s focus on them for now.

Wow, Roethlisberger is way up there this week at home against a medium matchup (MIN) but expected to get about 26 points. Brady, Ryan, and Wilson are also in high-scoring matchups this week. As always, if you have a top QB, play a top QB.

If you’re streaming this week, Models A likes Alex Smith. There’s no way you’ll get a repeat of last week, but with a predicted 26 points against a middle-of-the-road-vs-QBs Philadelphia defense he’s not looking like a bad pickup.

Model A also puts Winston, Bortles, and Flacco in the top 15. As streaming options go I’d play one of them if I were a little more desperate (or had planned on Luck returning by now). I doubt any of the remaining top 15 from Model A are available in your league.

A word of caution about Model F: I don’t understand how it worksm and I’m still learning about it. There’s a chance that it’s all wrong and I’ll have to just try again or completely scrap it at a later date. That said, it’s interesting that the estimated points for Model F take a similar structure to the results from Week 1. In week 1 there were 3 top QBs then about 15 guys all with 13-18 points. Model F thinks this week will be similar, with Brady and Rodgers as the stand-out top QBs and the next 17 QBs (from Brees to Flacco) to all get between 15 and 18 points. If you’re a Model F truther, it looks like all you have to do is avoid playing Dalton, Taylor, or Bradford and you’ll be good.

I’m choosing between Stafford and the waiver wire this week, and he’s way down there in both Model A and Model F. If I actually trust Alex Smith or Joe Flacco to be the #7 or #9 QB this week a la Model A, then it might be worth it to pick one up for the week. This definitely merits further thought on my part.

Unfortunately, I don’t have time to think. I have Kickers to model.

Week 2 K Predictions

Week 2 DEF Predictions