This has been a big season for Pythagoras and his wins. With teams like the Giants earlier in the year and the Vikings all year, who have records that significantly outpace their Pythagorean Wins (expected wins based on point differential), the discussion often becomes about luck vs skill.
Is winning close games luck or skill? Is it repeatable or should fans hear the footsteps of regression coming? Some of you don’t care. When the Giants were 6–1 you just pointed to the scoreboard. The Vikings are 11–3, and you can tell me to “f*ck off nerd, 2nd best record in the NFL, suck on that” and honestly, maybe that’s for better. You probably live a happier life than I do, as I’m always letting the shadow of regression steal my joy. But anyway, here is my opinion. Winning close games is mostly luck and you should not expect it to continue. Not in the playoffs and not next year. Can it happen? Sure. Pythagorean wins for the 2011 Giants suggest they should have had a losing record, and then they continued to win in the playoffs all the way to winning the Super Bowl with a regular season point differential below zero. I also emphasize “mostly” luck because I’m sure there are clutch players and clutch coaches who tend to consistently shine when the lights are the brightest, but I’m not focused on those outliers. I think the data largely shows that a team’s success or failure in close games over a stretch of a season is not repeatable and is random, which leads to a correction down the road.
To find out if close game records are random, I plotted teams’ records in games decided by 3 or fewer points (or OT) in a year (horizontal axis) and then their records in close games the next year (vertical axis) and this is what it looks like:
There is no correlation from one year to the next, suggesting almost complete randomness in close games. Just for illustration purposes, the fake data below shows what a strong correlation might look like.
Let’s look at teams Net Wins from 2021 (if this looks familiar, I’ve shared this chart before)
If we take the extremes from last year and compare what they have done this year, the change in their net wins in 3-point (or OT) games shouldn’t be a surprise. The Raiders should not have been expected to go 6–1 in close games in 2022, and the Seahawks weren’t going 0–5 in close games either. And of course, both teams had a correction. Here is what has happened to the teams with net wins +4 or better and -4 or worse in close games,
Team 2021–2022 shift in net wins
It shouldn’t be a surprise that the teams with the best record in 3-pt games did much worse this year, and the teams with the worst records last season fared much better.
- Teams with a +4 or better averaged a decline of net wins in close games of -4.5 this season
- Teams with a -4 or better averaged an increase of net wins in close games of +4.5 this season
What about going into the playoffs?
What happened to the teams who overperformed in close games, thus owning a record far greater than they would have if they split their close games 50/50?
- The number one seed, 13–4 Packers…one and done
- The number one seed, 12–5 Titans…one and done
- The 10–7 Raiders…one and done
- The 9–7–1 Steelers…one and done.
Combined these 4 teams were 44–23–1(.654) but they were all immediately disposed of in the playoffs going 0–4 including both number one seeds. The combined record of these four teams outside of 3-pt games was nothing exceptional (23–20), a winning percentage (.534) that you would kind of expect from a one-and-done type playoff team.
For these teams at the extreme ends of the spectrum, we have seen two things.
- An underperformance in the playoffs relative to their record.
- Regression in their record the next year.
Regression to the mean is an important concept that I try not to overlook. The gravitational pull of regression is more powerful than my positive thinking. When they were saying the Giants were the worse 2–0 team ever, or one of the worst 6–1 teams ever, I believed them. This isn’t to discredit teams who have shown a propensity to keep pulling wins out of their ass. It’s more a warning about what might come next.
Playoff examples are an extremely small sample size (N=4). In the off season, I’m going to look at ten years of data to see what that tells us. Also, my entire group of extremes (N=6) is a small sample size as well, but chart #1 is not. That lack of correlation comes from 9 years of data (288 team seasons).
I don’t know how many of these 3-point games consist of a garbage time scores taking the score from -10 to -3 with little time left. You might not consider those to be close games in the context of this analysis, which is fair. But I don’t have that data easily available.