The NFL started its regular season this weekend, and against the cacophony of overreactions, you hear the occasional analyst say something along the lines of "Week One doesn't show us who these teams will become" or "Teams are still shaking off the rust."
This feels correct-- the teams simply haven't played together for very long, and many of them feature new players, even new coaches. It's also not hard to think of examples of Week One results that were far from indicative of longer-term success: just last year, the Bengals dropped their season opener before making it to the AFC Championship Game, while the Bears beat the formidable 49ers in Week One before limping to the worst record in the NFL.
With this in mind, let's take a look at the data to see how much Week One matters! Our method here will be pretty simple: we're going to look at the correlation between a team's record in a given week and its overall season record. Collecting data from 2010 to 2022 (an arbitrary but sizable sample), here are the correlations between records within each week and overall season records:
Here's the same table sorted by strength of correlation:
Three things stand out to me, in ascending order of interesting-ness:
1) The correlation is positive (and significant) for every single week. This isn't shocking: the better team tends to win the game, irrespective of week. Also, if a team wins in a given week, that win will be part of its overall record, so it stands to reason that winning that particular week will be positively correlated with overall record.
2) Week 18 is at the bottom. At first blush, this makes sense since that's the final week of the season, so many games are irrelevant and many starters are resting. HOWEVER, for many of the seasons in this dataset, the NFL had a 17-week season. But Week 17 has a much stronger correlation with overall record, even if we just subset to years with 17-week seasons. This means that either this is just noisy OR there's something about the 18 week season that leads to the last week being much less relevant. For instance, if over recent years, teams are more cautious about playing starters in the final week, or if the longer season leads to teams clinching playoff spots earlier, we might see a weakened Week 18 correlation. I'm open to your theories because this baffles me a bit.
3) THE BIG TAKEAWAY: it does seem like the earlier weeks are less predictive! The correlations are still positive, which means that you'd still bet on the teams that win early games to end up being better, but 5 of the 6 least correlated weeks are in the first 6 weeks of the season.
Just for good measure, let's re-run the analysis, but this time let's look at margin of victory or defeat in each week instead of just result (we're still sorting by correlation here):
Week 18 now floats toward the top, which makes me think it's just a bit noisy (especially since there are fewer Week 18 results in this dataset on account of there were no Week 18 games until 2021). But the bigger story is that the weaker early-season correlations hold!
So rejoice, fans whose teams struggle in the early going! You have my permission to keep dreaming a little longer!
I myself would take heart, but oddly, the Bears didn't even play yesterday. Weird!