Fascinating article on 2019's close games - Must read

I'm not sure what, if anything, all this means. It *might* mean that we were better than our record this year, and might have some regression to the mean next year which would be helpful.

This is not a paid article - the tables in the article didn't paste well and it will read MUCH better on the original site at [fivethirtyeight.com]


Not All Close NFL Wins Were Actually Close
We used win probability to find coin-flip games — and flipped them.
By Ty Schalter

Filed under NFL

San Francisco 49ers v Seattle Seahawks
Running back Marshawn Lynch and the Seattle Seahawks came up just short on Sunday against San Francisco. OTTO GREULE JR. / GETTY IMAGES
In an alternate universe, the 2019 Seattle Seahawks won the NFC West division title on Sunday night when they gave the ball to power back Marshawn Lynch on the shadow of the goal line — instead of, as they did in this one, taking a 5-yard delay-of-game penalty and failing to punch it in from 5 yards out.

Instead of a 29-26 win, the Seahawks suffered a 26-21 defeat at the hands of the San Francisco 49ers. Of course, that’s the way of things in the NFL: A team that’s leading by as many 7 points is still just one play1 away from trailing.

So what if we could wave a magic wand and visit that reality? What would the NFL’s 2019 season have looked like if every game that finished within one score finished with the opposite result?

That was the question posed by Football Outsiders contributor Bryan Knowles after Week 16, and his answer — an alternate-universe season table with the results of all one-score games flipped — touched off a small firestorm on NFL Twitter. With the Dallas Cowboys at 12-32 going into the final weekend and the Seattle Seahawks sitting at 3-12,3 the extreme divergence of these results from reality proved a nice diversion for NFL observers tired of working out playoff scenarios.

“The outcome of one-score games tells us less than huge wins or huge losses,” Knowles argued. “One-score games can easily be decided by one unlucky bounce. So this is a list of unlucky teams, or maybe teams that need to do better in clutch moments or need better coaching.”

Lucky bounces are obviously unpredictable — and not only can teams not count on them, but a team that gets lucky in the win column one year is unlikely to get lucky again the year after. This idea, regression to the mean, is such a truism of NFL analysis that Football Outsiders included it in a series called “FO Basics” almost 10 years ago. But not all one-score games are determined by lucky bounces.

In Week 3, the New Orleans Saints traveled to Seattle and scored three first-half touchdowns against the Seahawks. As the Saints started the fourth quarter with a 27-7 lead, ESPN’s win-probability model gave them a 99.1 percent chance to win. It didn’t matter one bit that the Seahawks scored three garbage-time touchdowns (the last of which came as regulation expired). The final score was 33-27 — technically a “one-score game” — but luck had nothing to do with what was really a dominant, wire-to-wire Saints win.

On the other hand, after the Arizona Cardinals intercepted San Francisco 49ers quarterback Jimmy Garoppolo with 4:41 left in their Week 11 contest, the Cardinals took back over at their own 25-yard line with a 3-point lead and a 76.4 percent chance to win. But then they went three-and-out before surrendering two fumbles and two touchdowns in the game’s final minute — turning a 3-point edge into a 10-point loss in a handful of plays. But by the “one-score game” standard, this one didn’t count as close.

We have much better tools than the final score to describe what happened in an NFL game. Win probability can tell us whether games that seemed close after the fact were actually coin flips in the closing minutes. So I found the games in the 2019 NFL season that ESPN’s win-probability model gave a 60-40 (or narrower) split at any point with five or fewer minutes left in regulation.

Sixty-three of the NFL’s 256 games fell within this threshold, and some teams played a lot more of them than others. Here are the seven teams that gain the most by flipping all their losses or ties in coin-flip games to wins:

Detroit could have used a little luck
NFL teams by the most wins that could have been gained had their outcomes in 2019 coin-flip games* flipped

RECORD
TEAM ACTUAL IN COIN-FLIP GAMES ONLY IF ALL COIN-FLIPS WERE WON TOTAL WINS GAINED
Detroit Lions 3-12-1 1-5-1 9-7 +5.5
Indianapolis Colts 7-9 2-4 11-5 +4.0
Cincinnati Bengals 2-14 0-3 6-10 +4.0
Denver Broncos 7-9 2-4 11-5 +4.0
Los Angeles Rams 9-7 0-4 13-3 +4.0
Los Angeles Chargers 5-11 2-4 9-7 +4.0
Arizona Cardinals 5-10-1 1-3-1 9-7 +3.5
*Coin-flip games are those with a win probability no greater than 60 percent for either team at any point in the last five minutes.

SOURCE: ESPN

Had the Detroit Lions caught all the breaks instead of going 1-5-1 in coin-flip games, they would have finished 9-7 instead of 3-12-1. That’s a whopping 5.5-game improvement. Five other teams picked up four victories in our exercise, with the Indianapolis Colts, Denver Broncos flipping losing seasons into double-digit wins.

Notably, the Los Angeles Rams had the most coin-flip games without a win, turning a potential repeat of last year’s 13-3 record into a 9-7 campaign that didn’t even get them to the playoffs.

But which teams would have had the most to lose if Lady Luck had left them entirely?

San Francisco and Seattle both got lucky
NFL teams by the most wins that could have been lost had their outcomes in 2019 coin-flip games* flipped

RECORD
TEAM ACTUAL IN COIN-FLIP GAMES ONLY IF ALL COIN-FLIPS WERE LOST TOTAL WINS LOST
San Francisco 49ers 13-3 7-3 6-10 -7
Seattle Seahawks 11-5 5-1 6-10 -5
Atlanta Falcons 7-9 3-0 4-12 -3
Baltimore Ravens 14-2 3-0 11-5 -3
Chicago Bears 8-8 3-2 5-11 -3
Miami Dolphins 5-11 3-1 2-14 -3
Pittsburgh Steelers 8-8 3-3 5-11 -3
*Coin-flip games are those with a win probability no greater than 60 percent for either team at any point in the last five minutes.

SOURCE: ESPN

The 49ers played a league-high 10 coin-flip games, and they went 7-3 in them. Before fans of rival teams use this nugget as a cudgel against them, they should realize that means the 49ers were also three coin-flip games away from going 16-0.

Oh, and the fan base most likely to want to knock the 49ers down a peg? Seahawks fans, whose team finished second to San Francisco in the NFC West. But they also finished second to San Francisco in coin-flip wins and had the same minimum possible record, 6-10.4
Now that we know what the possibilities are, let’s repeat Knowles’s exercise. Instead of figuring out what every team’s maximum and minimum possible record would be if all the coin-flip games finished one way or the other, let’s reverse the real-world outcomes of all 63 coin-flip games and see where the teams end up.

Here are the final 2019 NFL standings, but with the result of every FiveThirtyEight-defined “coin-flip game” flipped:

How different would the standings have looked?
2019 NFL standings if the outcomes in coin-flip games* had flipped

AFC East NFC East
TEAM COIN FLIPS CHANGE IN WINS NEW RECORD TEAM COIN FLIPS CHANGE IN WINS NEW RECORD
NE 3 +1 13-3 DAL 1 +1 9-7
BUF 3 -1 9-7 PHI 3 -1 8-8
NYJ 2 0 7-9 NYG 4 0 4-12
MIA 4 -2 3-13 WSH 3 +1 4-12
AFC North NFC North
TEAM COIN FLIPS CHANGE IN WINS NEW RECORD TEAM COIN FLIPS CHANGE IN WINS NEW RECORD
BAL 3 -3 11-5 GB 3 -1 12-4
PIT 6 0 8-8 MIN 4 0 10-6
CLE 3 +1 7-9 DET 7 +4 7-8-1
CIN 4 +3 5-11 CHI 5 -1 7-9
AFC South NFC South
TEAM COIN FLIPS CHANGE IN WINS NEW RECORD TEAM COIN FLIPS CHANGE IN WINS NEW RECORD
HOU 2 0 10-6 NO 3 -1 12-4
IND 6 +2 9-7 TB 5 +1 8-8
TEN 3 -1 8-8 CAR 1 +1 6-10
JAX 2 -2 4-12 ATL 3 -3 4-12
AFC West NFC West
TEAM COIN FLIPS CHANGE IN WINS NEW RECORD TEAM COIN FLIPS CHANGE IN WINS NEW RECORD
KC 3 -1 11-5 LAR 4 +4 13-3
DEN 6 +2 9-7 SF 10 -4 9-7
LAC 6 +2 7-9 ARI 5 +2 7-8-1
OAK 3 -1 6-10 SEA 6 -4 7-9
*Coin-flip games are those with a win probability no greater than 60 percent for either team at any point in the last five minutes. The only outcome we couldn’t flip was the Week 1 tie between Arizona and Detroit.

SOURCE: ESPN

All four AFC division winners stay the same, with the Buffalo Bills holding on to a wild-card spot, and the Denver Broncos taking the other berth.5
There are more shake-ups on the NFC side of our bracket: The Cowboys hold off the Philadelphia Eagles for the NFC East crown, the Rams cruise to the NFC West title over the 9-7 49ers, and the Seahawks drop to fourth place, behind the now-7-8-1 Cardinals. In fact, in this scenario, both the Lions and Cardinals use their Week 1 tie — the only coin-flip game we couldn’t flip — to climb up out of their respective basements.

But with the 49ers and Minnesota Vikings claiming the wild cards, the Cowboys/Eagles and Rams/Seahawks flip-flops would be the only changes in the NFC playoff field.

Comparing these numbers to Knowles’s original study, it’s clear that our figures are much less volatile. The “one-score” method implies that many, many more games are up for grabs than our method indicates. Knowles’ projection had the Cowboys picking up at least four wins instead of one, the Seahawks losing at least eight wins instead of four, etc.

ESPN’s Brian Burke once calculated that about 52.5 percent of NFL results are based on luck — but given that the better team will also get lucky half the time, the stronger team should win about 76 percent of games. Thus, 76 percent is the best success rate any mathematical model could hope for in predicting NFL results over the long run, according to Burke. We identified 63 of 256 regular-season games that effectively came down to a coin flip. That share? 24.6 percent — almost identical to the share of NFL games Burke found are typically determined by chance.

We can also double-check our findings against Football Outsiders’ main performance metric, Defense-adjusted Value Over Average. DVOA projects an estimated win total, and no team this year over- or underperformed their estimate by more than 3.2 wins. As no team since at least 1985 has underperformed their estimate by more than 3.3 wins, it would take a calamity for the eighth-ranked Seahawks, with 10.9 estimated wins, to go 3-13 or 4-12 as the “one-score game” method projects.

Regardless of how much more accurate this win-probability-based method is (or isn’t) than simply looking at final scores, we know this much is true: A team that lost a lot of close games this year isn’t likely to be so cursed next year, and a team that won a lot of close games this season won’t be able to count on being so lucky in 2020.