The NFL season is long, unpredictable, and full of many twists and turns. We overreact after Week One, re-evaluate everything we thought we knew after Week Four, and by Week Eight we realize we still know as little as we did when the season began.

Consider this season, for example:

• After Week One’s typical high-flying victory, we were all penciling the Chiefs in for their fourth straight AFC title game. Now, they would settle for just making the playoffs at all.
• The Packers started the season with a 35-point home loss and have since won seven consecutive games.
• While Cleveland was a trendy Super Bowl selection through three games, today they sit 4–4, last in the AFC North, with locker room turmoil threatening to completely torpedo their once-promising season.
• Sam Darnold somehow resurrected and un-resurrected his career — in just 7 games!

You get the point. Through eight weeks of the NFL season, nobody really has a firm idea about where their favorite team will end up.

Point differential has long been a popular statistic used to judge the quality of a team. While wins and losses are ultimately all that matter, the difference between winning or losing a game can often be a single dropped pass, incorrectly called penalty, or a missed kick.

Wins and losses can be fluky, and over the course of eight weeks, it is pretty common to have a rather mediocre team eke out a couple of close victories and possess a somewhat inflated record. For that reason, point differential, or average margin of victory, can often better reflect how a team is really performing over the course of 60 minutes each week.

A 6–2 team with an average margin of victory of +2 points has clearly been squeaking out wins week after week. Given the randomness of one-score game outcomes in the NFL, this team is probably going to regress in the second half of the season and start to lose those same close games. Likewise, a 5–3 team with an average margin of victory of +11 points has shown far greater dominance over the first eight games of the season and can probably be trusted to win more regularly as the sample size grows larger.

For this article, I used Week Eight point differential data going back to the 1978 season for each team and plotted it against their end-of-season win totals and, using linear regression, built a predictive model to project the number of wins for each team in the NFL this season.

The following chart shows each teams’ point differential through 8 weeks on the x-axis and the number of wins they ended that season with on the y-axis:

The graph shows a pretty clear correlation between a team’s mid-season point differential and their end-of-season win total. In fact, the correlation coefficient between point differential (through 8 weeks) and total wins (through 17 weeks) is 0.8016, which, considering the range of factors that influence a team’s record (injuries, weather, strength-of-schedule, home vs. away, etc.) is quite high.

Using NumPy’s “polyfit”, I obtained the following equation for my model:

y = 0.0457 * x + 7.9598

I used this equation PLUS the difference in a team’s current wins (through Week Eight) and their expected wins (through Week Eight) to predict final records through 17 games.

With these parameters, the model predicted within ± 1.16 wins of a team’s actual win total, on average, dating back to the 1978 season.

Check it out below:

Projected vs. Observed Win Totals Since 1978

Now, by inputting each team’s point differential through Week Eight of the 2021 season, the model will predict, with reasonable accuracy, the projected win totals for every team.

2021 Projected Win Totals

You now have a fairly certain idea of where your team will end up this season, give or take 1.16 wins.

Whether your team is fighting for the top draft spot or top playoff seed, I hope this article brought you some optimism and reassurance. If it didn’t, outliers exist for a reason, so don’t give up hope!

Best of luck to your team (unless your team is fighting for an AFC wildcard spot), and go Colts.

Michael Black

View code in GitHub repository.