In 2017, the Carolina Panthers went 11-5, thanks largely to a 7-1 record in games decided by 7 or fewer points. This means that in non-close games, Carolina went 4-4.
In 2018, Carolina was even better in non-close games, going 5-2. But the Panthers didn’t replicate their close-game success; in fact, the Panthers luck swung wildly in a different direction, going just 2-7 in close games. This has actually been a trend of the Ron Rivera / Cam Newton Panthers. The duo got off to a very rocky start, going 1-5 (0.167) in close games in their first year together in 2011, and then 1-7 (0.125) in close games in 2012. After that season, I wrote that Carolina looked like a team on the rise that would probably start experiencing good luck soon. That turned out to be one of my most accurate predictions, as in close games, Carolina went 5-2 (0.714) in 2013, 4-2-1 (0.643) in 2014, and then 6-1 (0.857) in 2015. The Panthers then swung the other way, going 2-6 (0.250) in 2016, then up to 7-1 (0.875) in 2017, and back down to 2-7 (0.222) in 2018.
The Panthers yo-yo performance in close games is not that unusual; for the most part, winning percentage in close games is not sustainable. How did I measure that?
1) I looked at the records in close games (7 or fewer points margin of victor) and records in non-close games for every team since 1990.
2) Because of small sample sizes, I then calculated a projected winning percentage in close games based on that team’s winning percentage in non-close games after adding 11 games of .500 play to that record. So if a team went 2-0 in non-close games, they’d be expected to have a 0.577 winning percentage (7.5 wins out of 13 games) in close games.
3) I then calculate how many wins above expectation in close games each team had. By this measure, the 2012 Colts exceeded expectations the most. That year, Indianapolis went 2-4 in non-close games, so we would expect them to have a 0.441 winning percentage in close games (7.5 wins out of 17 games). Indianapolis played in 10 close games (expected to win 4.41 games) and actually won 9 of them, giving the Colts 4.59 more close games wins than expected. The next year, Indianapolis won 1.70 more close games than expected, so this regressed to the mean quite a bit.
The correlation coefficient between wins above expectation in Year N and in Year N+1 is just 0.04, making it practically zero. The graph below shows all teams since 1990; their wins above expectation in Year N is on the X-Axis, and wins above expectation in Year N+1 is on the Y-Axis. As you can tell, it’s essentially a random scatterplot of data:

And here is the raw data:
Finally, here is the 2018 data: The Cowboys had 2 non-close wins last year in 6 games, for a 0.333 winning percentage in non-close games. But in close games, Dallas won 8 out of 10 games, a 0.800 winning percentage. We would have expected Dallas to win about 44.1% of their non-close games, so by going 8-2, Dallas finished 3.59 wins above expectation, most in the league last year.