Wildcard Watch Simulation: Week 14 – Cowboys Edition

It’s Dallas week in DC. As we prepare for Sunday’s showdown with the current division leaders, a fair number of Washington fans are hopeful of seeing an upset. Dallas coach, Mike McCarthy has said he is confident of a win. That’s just the kind of statement that motivates an underdog, like Taylor Heinicke’s team. If TH4 and the revamped defense can keep the momentum going, it’s not that hard to imagine a Washington win at FedEx this weekend.

What a difference four weeks make. At Week 8, heading into the bye, it seemed to most fans that the wheels had truly come off. Outwardly, Ron Rivera didn’t seem to have any answers, but he must have known something. The team that came out of the bye and upset the Bucs bore little resemblance to the one we had watched the week before. Instead of playing out of position, the defense was suddenly playing as a team. Where the offense had been committing mistakes and missing opportunities, suddenly it was finding ways to win.

While it might have seemed laughable to talk about the 2-6 Football Team making the playoffs before the bye week, following Sunday’s road win in Vegas, they have strengthened their hold on the last wildcard seed and are currently riding the longest winning streak in the NFC.

But a playoff finish is still far from guaranteed. Entering Week 14, the 6-6 Football Team finds itself in a tight race for the last two NFC wildcard playoff berths with three other six-win teams, San Francisco (6-6), Philadelphia (6-7), and Minnesota (6-7). And right behind them a trailing pack of Carolina, Atlanta and New Orleans remain in striking range at 5-7.

Swept up in the excitement of the WFT’s Lazarus-like resurrection, and taking the lead from the ever faithful Bill-in-Bangkok, I have committed to tracking the WFT’s race for a playoff berth, for as long as they are still in it. Rather than trying to compete with Bill’s fact and logic-based Wildcard Watch series (Week 12, Week 13, Week 14), which would be a fool’s errand, I will have fun with the wildcard race by doing a weekly projection which incorporates a significant element of chance, in an attempt to simulate the kinds of surprises which keep the NFL season interesting from week to week.

Last week I debuted a simple projection model in which I set each team’s probability of winning its matchup based on its win-loss record over the last five games and those of its opponent. Let’s see how last week’s projections stood up against the actual game results.

---

Week 13 Model Predictions

In Week 13, the model predicted a dream run to the division championship by the Football Team, with a win over the Raiders continuing a winning streak that would extend all the way to Week 16 and include sweeps of the Eagles and Cowboys. The model called the Vegas game correctly, as well as Detroit’s highly improbable upset of the Vikings, which dealt a major blow to one the WFT’s wildcard competitors.

That’s a promising start. Let’s have a look at how it did overall. The following chart shows the predicted win-loss outcomes, with correctly called winners highlighted in blue and incorrect calls in red.

The model got 7 out of 14 predictions right, which is about what you’d expect by simply tossing a coin. That is a shockingly poor performance. I am embarrassed to put a product like that on the pages of Hogs Haven. Something needs to be done.

Last week, my editor, James Dorsett commented on the draft article, asking why I chosen to base the predictions on W-L records when it has been shown that point differential is a better predictor of team performance. I didn’t have time to think that through, because my sims were all suited up and ready to take the field. However, given my model’s pathetic performance in Week 13, it’s time to bring in some competition for my analytics team.

During the week, I brought in James’ guys for a try out. They fed the week eight through twelve game results through their point-differential-based model and produced the following results:

That’s more like it. The point differential model accurately predicted ten out of fourteen Week 13 game results. Unlike Ron Rivera, I don’t hesitate to make midseason adjustments when it’s clear my model’s not getting it done. I’ve released my analytics team and signed James’ crew for the remainder of the season. Fortunately, analytics guys don’t have to go through any COVID protocols before they start. They can work from home.

---

Point Differential Model

Let’s see how the new point differential model works.

The previous model embodied two basic principles:

1. You are what your record says you are – meaning that the best measure of how likely a team is to win its next matchup is based on how it fared in its previous games.

2. But you are really only as good as your last five games – recognizing that teams’ fortunes wax and wane throughout the season, I limited the analysis to the last five games, in an attempt to have the model reflect how teams are playing right now.

I’m going to stick with those basic principles and just switch around the way I measure teams’ records, from the binary metric wins and losses, to a continuous variable, point differential (team’s total points minus opponents’ total points).

In hindsight, and as I’m about to show, point differential conveys a lot more information about how good or bad a team is than wins and losses. Just consider, two teams with 3-2 records over their last five games. One team has three wins by two touchdowns or more and two losses that were within a field goal of going the other way. The other team has two blowout losses and three wins by three points or less. The two teams look identical by W-L record, but one team appears to be much better than the other based on point differential. Let’s hope that point differential does a better job of predicting outcomes than W-L record, as our tryout suggested it will.

To put these principles into practice, I calculated each team’s point differential over their last five games. Then I converted the raw point differentials into a metric which shows how good or bad each team is relative to the rest of the league, which I’ll call P_win, just like last week. P_win was calculated as follows:

P_win = (PtDiff_team – PtDiff_min)/(PtDiff_max – PtDiff_min)

Where PtDiff_team is any given team’s point differential, PtDiff_min is the lowest point differential across the league (Atlanta: -69), and PtDiff_max is the maximum point differential in the league (New England: +109).

P_win values range from 0 to 1, in proportion to how week or strong a team has been over its last five games. Atlanta is the weakest team by this measure, with point differential of -69 and P_win value of 0. New England is the strongest team over the last 5 games, with a point differential of +109, and a P_win value of 1. The WFT is currently ranked 11^th in the league, with a five-game point differential of +13, and a P_win value of 0.461.

After that, the model works the same as last week. Ahead of each matchup, the probability that one of the two teams will win is calculated as follows:

p(win) = Team 1 P_win/(Team 1 P_win + Team 2 P_win)

To see how this works, lets see how the model sets the WFT’s odds against Dallas. Washington enters the game with a point differential of 13, which equates to a of P_win 0.461. Think of that as meaning that the model gives the team a 46.1% chance of beating any random opponent, without knowing who the opponent is. Dallas has an advantage, with a point differential of 23, equating to a P_win of 0.517. The WFT’s probability of winning is calculated as follows:

p(win) = 0.461/(0.461 + 0.517) = 0.471

Once again, to get the model to work, I had to set a floor on P_win, but this time it only applies to a single team, Atlanta. I set the floor at 0.05 (5% chance of winning), which is slightly lower than the next lowest team’s actual P_win (Jacksonville, 0.051).

Then a dice roll is simulated using the random number generator in Excel, which produces decimal numbers ranging from 0 to 1. If the random number is less than or equal to p(win) then Team 1 wins. If it is higher, Team 2 wins.

Using this method, I was able to project the winners and losers of the Week 14 games. However, after that, I hit a limitation that changes the way the simulation works for the rest of the season. In last week’s model, I was able to feed the predicted wins and losses back into the model to project the outcomes of the following week’s games. This resulted in a “momentum effect”, by design, where a few accumulated wins could greatly increase a team’s chance of winning their next game, and a few losses could increase their chance of losing, resulting in runaway hot streaks and cold streaks.

While the point differential model appears to have an advantage at predicting next week’s wins and losses, there is no obvious way to predict next week’s point differentials, without making questionable assumptions. Therefore, to project outcomes of games from Weeks 15 through 18, I had to assume that teams have all plateaued where they are at Week 13 and their P_win values will remain constant for the rest of the season. The model will update next week, by adding point differential values from Week 14 games and dropping point differentials from Week 9. But for each set of weekly projections a team is assumed to be as good or bad as its last five games for the rest of the season.

---

League Ranking by Point Differential - Last Five Games

As I mentioned above, ranking teams by point differential, instead of W-L record, offers a different perspective on how the teams stack up. I thought the results were interesting enough that they deserve some attention before we get on to the Week 14 Projection.

Here is the ranking of NFL teams by point differential over their last five games:

Over the past five games, the New England Patriots have really separated themselves from the rest of the league. The next two teams might come as a bit of surprise to those who have not been following the AFC very closely. Indianapolis has been playing very well, and Miami has pulled, more or less, even with the resurgent Chiefs.

Of most interest to WFT fans, three NFC East teams are in the top 11, and ahead of three current division leaders. The WFT is currently the sixth ranked NFC team by this metric, which just happens to coincide with their NFC playoff seed. Meanwhile, three of the WFT’s current competitors in the NFC wildcard race, Carolina, New Orleans and Atlanta, have been playing pretty poorly of late.

While the competition for a wildcard spot from teams outside the division might not be that strong, Washington’s road to the playoffs runs through five straight NFC East matchups. The Football Team will probably need to win at least three of those games to make the playoffs. That will be an uphill battle, since it requires a minimum of two wins against teams ranked amongst the NFL’s best teams right now.

---

Jerome Miron-USA TODAY Sports

Week 14 Projection

It is Dallas week in DC. The Football Team will take to FedEx Field on Sunday as slight underdogs to the visiting Cowboys according to this model. The two teams’ respective five-game point differentials of 13 (WFT) and 23 (DAL) give Dallas a 0.57 to 0.43 win-probability advantage over the Football Team.

The Football Team’s dice roll is 0.06, well below the 0.43 win probability. It is a blowout win for the Football Team. It doesn’t get much better than that. The complete Week 14 projection is as follows, with projected upsets marked by asterisks:

In addition to Washington’s win over Dallas, the model predicts three other upsets this week: LA Rams (8-4) over Arizona (10-2), San Francisco (6-6) over Cincinnati (7-5), Minnesota (5-7) over Pittsburgh (6-5-1).

With these projected results, the NFC playoff division leaders remain unchanged from Week 13, but the top three teams now have identical records. Using the ESPN NFL Playoff Machine, the NFC playoff seeding at the end of Week 14 is as follows:

1^st Seed - NFC North Champion Green Bay (10-3)

2^nd Seed - NFC South Champion Tampa (10-3)

3^rd Seed - NFC West Champion Arizona (10-3)

4^th Seed - NFC East Champion Dallas (8-5)

5^th Seed Wildcard - LA Rams (9-4)

6^th Seed Wildcard – Washington Football Team (7-6)

7^th Seed Wildcard - San Francisco (7-6)

The WFT and 49ers are being tailed by four teams with 6-7 records: Philadelphia, Minnesota, New Orleans and Carolina. Atlanta (5-8) appears to be close to elimination, and the Giants (4-9) are just playing for draft position.

---

Week 14 Season Projection

Like last week, I’m just going to focus on the NFC East results and NFC playoff picture and not bother showing the full details outside the division and conference. Let’s start with how the new model projected the NFC East Division race.

The new model’s projection seems a lot more realistic than last week’s version which had the WFT winning out to Week 17. This time the WFT splits its games with Dallas and Philadelphia and beats the woeful Giants, who finished 8-9 last time. This is more like a best-case scenario that’s actually plausible.

Dallas wins the division this time, and the WFT finishes the season tied with Philadelphia at 9-8, which should have both teams vying for a wildcard seed. Let’s look around the NFC to see how the wildcard competition played out. I’ll just save myself the effort this time and plug my projected results into the ESPN Playoff Machine.

Five of the seven final NFC playoff teams were who you’d expect, but there was one mild surprise and one absolute screamer:

1^st Seed - NFC North Champion Green Bay (14-3)

2^nd Seed - NFC West Champion Arizona (12-5)

3^rd Seed - NFC South Champion Tampa (12-5)

4^th Seed - NFC East Champion Dallas (10-7)

5^th Seed Wildcard - San Francisco (11-6)

6^th Seed Wildcard – Carolina (10-7)

7^th Seed Wildcard – Los Angeles (10-7)

Both San Francisco and Carolina won all their remaining games from Week 14 to finish 10-7 and, along with the 10-7 Rams, pushed the two NFC East 9-8 teams, Football Team and Philadelphia, out of the playoffs. San Francisco’s late season run seems a bit unlikely, but it’s not out of the question.

Carolina’s winning streak, on the other hand, was just blind luck, and seems extremely far-fetched. Their winning streak was a result of facing two very weak opponents according to this model (New Orleans, Pt Diff -50; Atlanta, Pt Diff -69, worst in NFL) and getting lucky dice rolls against their other heavily favored opponents (Buffalo, Pt Diff 39; Tampa Bay 2x, Pt Diff 21).

The other teams that had been in wildcard contention with Washington in Week 14 fell behind. Minnesota and New Orleans finished 8-9, while Atlanta fell to 5-12.

This time, using a projection based on recent point differential, the WFT just missed the playoffs. If it’s any consolation to disappointed fans, the WFT and Philadelphia had the best records of NFC teams that missed the playoffs. And Ron’s team finished with a winning record, which is something of a rarity in the Snyder era. Based on this result, they would be drafting somewhere between 16 and 19 in April, with the other 9-8 teams: Philadelphia, Indianapolis and Buffalo.

I think most fans would be disappointed if the WFT finished 9-8 and missed the playoffs. That is a remarkable turn-around from week 8, the prospect of a 9-8 finish would have seemed laughable to all but the few stalwarts. That being as it may, this projection shows that such a result is at least mathematically possible, even if it doesn’t seem very likely.

---

Closing Thought

WE WANT DALLAS!!! WE WANT DALLAS!!!

---