Welcome to the third installment of “The Wrong Read,” an article series that reflects on various RotoViz podcast episodes. You can get a sense of what this series is all about by reading the introduction to the introductory piece. Or just keep reading below.
Week 1 Overreactions
Week 1 of 2017 has come and gone. We learned a lot. Or did we? On the latest episode of RotoViz Radio, host Pat Kerrane and guest George Kritikos discussed how we ought to react to all the new information we got from Week 1. Specifically, how much do we need to adjust our priors based on one game in 2017? Are the Patriots a defense we can exploit? Is Tarik Cohen going to lead the Bears in targets? Is Buck Allen the next Danny Woodhead? Is Kareem Hunt the best RB in the league? Is the Vikings offense for real? Is Andy Dalton droppable?1
The last two questions especially interest me: I drafted a ton of Dalton in best ball, and—despite drafting a ton of Stefon Diggs, a half ton of Kyle Rudolph, a quarter ton of Adam Thielen, a few kilos of Jerick McKinnon, and a single share of Dalvin Cook—I drafted exactly zero Sam Bradford. It really makes no sense in retrospect, considering that I was apparently otherwise bullish on the Vikings’ passing offense. What can I say?—I guess Bradford was never that high on my list of available players when my turn came to draft. Anyway, that’s not what I want to write about.
On the podcast, George talked briefly about the need to react quickly to new information, not least because you need to put in waiver claims for players who could potentially be league winners. If you don’t think it’s possible Cohen could actually lead the Bears in targets, go look at their WR depth chart. Cohen might be their most skilled ball carrier, in which case the Bears should do everything they can to get the ball in his hands. Then again, he might also be a one-game wonder, and come crashing down to earth in Week 2. But that’s not what I want to write about either.
Priors and Uncertainty
Of course, there is another reason to react quickly: the information we get from Week 1 is probably the best information we have at the moment. Our priors are based on a certain assumption of continuity from one season to the next. But given all the player moves, personnel changes, and injuries that happened during the offseason, there’s not much continuity to be found. Add to this the fact that each team plays a substantially different schedule than they did last year, and it’s no wonder football is a hard sport to predict.
Priors, of course, are always corrigible, which is to say they’re always wrong, but with the implication that there’s hope for improvement. In Bayesian statistics the term “prior” refers specifically to a prior probability distribution. Think of a prior in this sense as our belief, before we see any evidence, about how likely an event is to occur. We assign a certain probability to an event before undertaking whatever analysis is necessary to determine its true likelihood–and that assigned probability is our prior. The prior is a starting point, but you expect to revise it after examining the evidence.
Consider projecting how rookies will perform in the NFL. You can look at their college stats, or at similar prospects from past draft classes, and both will enable you to make educated guesses about how a prospect might perform in the NFL. But you’ve never seen this rookie play in an NFL game against other NFL players, so any projections you make have little actual evidential basis. You expect to revise them once the player sees real NFL action.
Projecting players on a new team involves the same sort of uncertainty. You’ve never seen Terrelle Pryor catch passes from Kirk Cousins while competing for targets with Jamison Crowder and Jordan Reed. Although you have more information on which to base your prior, there’s still a lot you don’t know. There’s a lot that simply can’t be known in advance—a lot of factors we are blind to.
Projecting players in more stable situations in a new season is still subject to these same limitations, only to a smaller degree. In other words, the reason it’s a good idea to react quickly to new information early in the season is because you are definitely wrong about the things you thought going into the season. It’s a good idea to revise your beliefs quickly.
I think I was wrong about Bradford. I believed his completion percentage was a fluke, and that he would end up returning to his pre-Minnesota form, when he was a mediocre QB for the Eagles, and before that a below average QB for the Rams. The problem with priors, like projections based on past results, is that they so often turn out to be very wrong:
The next season Tom Brady threw for 4,806 yards and 50 touchdowns. The only difference between Brady in 2007 and Bradford in 2017 is that Brady had never thrown 50 touchdowns before.2
Bradford Will Throw 50 Touchdowns – 2017 Edition
My hyperbolic title is an homage to a question that has intrigued RotoViz writers for years. I’m trying to react more quickly to Week 1, so let’s see whether Bradford’s new pace can finally get him there. What would it take for Bradford to end the season with 50 touchdowns?
He’s already got three, so he’d need to average 3.13 TDs per game over the rest of the season. He’d need 3.53 per game from here on out to break Peyton Manning‘s all-time record of 55. We can make use of the Projection Machine to figure out how he does this.
Here are my team-level assumptions to make this work.
As you can see, the Vikings haven’t come near these marks in their recent history. But what I’m talking about is unprecedented, so that’s not really a surprise. And besides, the Patriots have surpassed these numbers multiple times in the last decade. So it’s not that unprecedented.
What we’re looking for is 700 pass attempts. Three teams have reached that mark in the last five years. Two different teams were around 680 last season. This is a real possibility. Those 1,162 plays are a lot—it would have been the most last year, but not the most ever. The Patriots ran 1,191 in 2012, and the Lions ran 1,160 that same season. 1,162 plays is definitely ambitious, but also within reach.
It gets trickier with with the efficiency metrics. Putting the top three targets—Diggs, Thielen, and Rudolph—at the 75th percentile in terms of touchdown rate doesn’t get us that close. These guys have got to go well beyond having career years.
If we assume Rudolph beats his career high TD rate of 0.095 by getting to the 10 percent mark, and give him last year’s target share of about 0.23, that gets us part of the way there. If we assume Diggs and Thielen split the 2007 Randy Moss role, and therefore split his TD rate of almost 0.15 by each getting 0.075, that gets Bradford to 44 TDs.
For this to work we’ve got to assume that Diggs and Thielen are each a little bit better than half a Randy Moss. So we’ll bump each of them up to the second highest touchdown rate a Vikings’ receiver has achieved over the last fifteen years, which is about 0.085.
If we also assume Bradford funnels passes their way (and why wouldn’t he when they’re such efficient touchdown scorers?) then we’ve got it! Bradford will throw 50 touchdowns in 2017—if everything I just put into the Projection Machine happens.
As expected, it also gives Diggs and Theilen some amazing seasons:
And Rudolph probably ends the season in the top five . . . among WRs.
If we assume further that Diggs and Thielen draw so much coverage away from each other that they can be even more efficient in the end zone, we can bump their TD rates up to 10 percent each. That gets Bradford to 55.6 TDs, which we round up to 56 for a new NFL record.
You should have drafted more of this offense. But you couldn’t have known that until Week 1.