NFL Week 16: The Wentz vs. Foles Debate

Last week was quite the week for the model, to say the least. It all began with an atrocious 0-4 start for -17.163 units before rebounding in the Sunday afternoon games with two wins. This then set us up for the Monday Night Football game, which was covered in-depth in last week’s newsletter. When it was all said and done and the dust settled,  the model’s largest play of the year, CAR +6 for 8.8 units, cashed and brought Week 15 in the green with a 3-4 finish for +0.84u. For the year, the model is now 69-38-4 for +102.27 units and a 23.93% ROI. That includes an 11-6 record for +20.30 units since its introduction to the Bet It Up newsletter.

Before proceeding, I just wanted to highlight a few things from the 0-4 start to Week 15. Of those four losses, three were by a combined five points and two of them (NYJ and GB) were by a point or less. For reference, there were a total of two model losses by a point or less in the first fourteen weeks and of the season (104 plays). But even if we ignore that (which we probably should) the most important thing to remember about betting a model is that method of profiting is to accumulate edges over a relatively large sample. If we believe the model does have an edge, then even losing a bet or a stretch of bets should be okay in our frame of mind since we are aware that each placed bet accumulates positive expected value.

Through Week 15, the model has accumulated 415.8 points of disagreement across 111 plays (or 3.7 points per play). From that accumulated disagreement, the model has generated 102.27 units of profit. That means that simply the action of placing a bet on a model play has generated 0.92 units of expected profit this season, regardless of the outcome of the bet. Another way of looking at it is that every time someone chooses to not “tail” a model play due to their own disagreement or hesitation, they are losing 0.25 units of expected profit per point/unit of the play. Last week, I got plenty of people telling me they were passing on the CAR +6 play because “It’s the Saints” or something similarly arbitrary. Using the information I just highlighted, those people passed on nearly 2.2 units of expected profit on the spot by simply choosing to not “tail” and place a bet on CAR +6, regardless of the outcome of the game. If that doesn’t highlight the inefficiency of choosing to not follow any given model play, I’m not sure what can.

For this week I wanted to highlight a play that I’ve already shared on Twitter that has been met with some hesitation, which is obviously fitting with what we just discussed in regards to the value of model betting. This week we will focus on the Week 16 matchup between the Houston Texans (10-4) and the Philadelphia Eagles (7-7), and take a deeper look at the entire Wentz vs. Foles debate, which has now reared its head for the second consecutive season. In particular, I’d like to refute the idea that the Eagles are better with Foles at quarterback. I’ll do this by using actual comments I’ve received this week.

“I don’t know. Nick Foles > Carson Wentz.”

The two things that drive this sentiment are that the two most memorable things about Foles for public bettors and fans are A) his postseason run and Super Bowl win last year and B) the Eagles win this past Sunday against the Rams. It’s easy to forget that Foles was near irreprehensible in the first two games of this season and that he was very mediocre in the three starts he had in 2017 leading into that postseason. Below is a chart showing various stats and metrics comparing Wentz and Foles, using data only from their starts and I even included Foles’ postseason data to try and tip the scale in his favor. Note that Foles edges Wentz in only seven out of forty total splits shown here. All data was obtained from Pro Football Reference, ProFootballFocus, and Football Outsiders.

Talent-wise I believe the change with Foles stepping in at QB is important to note. He throws to the receivers much more often than Wentz did, who gave a lot of attention to the TEs.”

“I think the fact that he focuses on different receivers throws out some defending [preparation] for the majority of Eagles’ footage. [There is] different play-calling with a less mobile QB, which also changes gameplans.”

Both of these comments came from the same comment thread, so I’m combining them here. The overarching argument here is that the Eagles with Foles offer something different (and better) due to a supposed change in target share. But this is simply untrue as Alshon Jeffery is the only pass catcher who has seen a >5% change in their target share when going from Wentz to Foles in the last two seasons. But that’s nullified by the fact that Jeffery’s splits shows a whopping three game sample size with Foles, with one of those games (W17 2017) having Jeffery only in on 31% of offensive snaps when he is usually in on ~93-95%. The Eagles opted to rest many of their starters during that game as they had nothing to play for, having locked up the 1 seed regardless of the outcome of that game. Agholor’s smaller jump in target share (+3.7%) is largely driven by from Week 1 and 2 of this year in which he had 22 targets, which is an astounding 26.2% of his total targets on the season. His volume from those two games can be explained by Alshon Jeffery’s absence as well as having Mike Wallace as their WR2 that game, who has zero receptions on three targets this season.

Lets step away from the whole Wentz vs. Foles debate for a second and just do a quick thought experiment. Just last week, the Eagles were 14 point underdogs for their game in Los Angeles against the Rams. This week, they are now -2 favorites at home against the Houston Texans. Nothing talent-wise has changed across those three teams. If we use a flat 3 point home field factor for this thought experiment, that means the Eagles would be +4 if this week’s game was in Houston (-2 -> +1 when moved to a neutral field, +1 -> +4 when moved to Houston). Comparing that to last week’s +14 spread @ LAR, these lines are implying that the Rams are approximately ten points stronger than the Texans. That means that in a game between the two, the Texans would either be 13 point underdogs in Los Angeles or 7 point underdogs at home. I don’t know about you, but I would take either of those lines in a heartbeat. So that should say to you that either A) last week’s line was an overreaction, B) this week’s line is an overreaction, or C) both are an overreaction.

What did the model think? Last week, the model made the PHI @ LAR game PHI +7.1 after Foles was declared the starting quarterback. This was obviously a pretty sizeable 6.9 point disagreement from the PHI +14 line that moved from +9.5 as a result of the news (Reminder: spreads of +7 or higher have been removed from model consideration as of Week 10 which is why PHI +14 was not a play). So if the public and the market saw Foles as having a significantly negative impact on the Eagles strength, why is it so high on him this week? The obvious answer is that beating the Rams on the road in primetime has drawn a lot of attention. And what have we learned in the past? Recency. Bias. Generates. Value. And the model thinks that that is the case this week. For the Texans / Eagles game, the model makes the game PHI +0.3, so a line of PHI -2 means there is 2.3 points of disagreement. This makes HOU +2 -104 a model play, risking 2.403 to win 2.3 units. This is obviously not a massive disagreement, but value is value and we’re here to accumulate any and all significant model edges.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.