Welcome to Regression Alert, your weekly guide to using regression to predict the future with uncanny accuracy.
For those who are new to the feature, here's the deal: every week, I break down a topic related to regression to the mean. Some weeks, I'll explain what it is, how it works, why you hear so much about it, and how you can harness its power for yourself. In other weeks, I'll give practical examples of regression at work.
In weeks where I'm giving practical examples, I will select a metric to focus on. I'll rank all players in the league according to that metric and separate the top players into Group A and the bottom players into Group B. I will verify that the players in Group A have outscored the players in Group B to that point in the season. And then I will predict that, by the magic of regression, Group B will outscore Group A going forward.
Crucially, I don't get to pick my samples (other than choosing which metric to focus on). If I'm looking at receivers and Jaxon Smith-Njigba is one of the top performers in my sample, then Jaxon Smith-Njigba goes into Group A, and may the fantasy gods have mercy on my predictions.
And then, because predictions are meaningless without accountability, I track and report my results. Here's last year's season-ending recap, which covered the outcome of every prediction made in our nine-year history, giving our top-line record (51-17, a 75% hit rate) and lessons learned along the way.
Our Year to Date
In Week 2, I laid out the four primary goals of Regression Alert. The first was to persuade you that regression to the mean is a real, actionable force in fantasy football. The core of that effort is making specific, concrete predictions based on nothing but regression to the mean and tracking the results over time. Here's a list of every prediction we have made so far:
- I predicted that high-volume, low-ypc running backs would outrush lower-volume, higher-ypc backs over the next four weeks (Week 4).
- I predicted that high-yardage, low-touchdown receivers would outscore low-yardage, high-touchdown receivers (Week 6).
The Scorecard
| Statistic Being Tracked | Performance Before Prediction | Performance Since Prediction | Weeks Remaining |
|---|---|---|---|
| Yards per Carry | Group A averaged 29% more yards per game | Group A averaged 43% more yards per game | None (Loss) |
| Yards per Touchdown | Group A averaged 14% more fantasy points per game | Group B averages 22% more fantasy points per game | 2 |
I promised an in-depth postmortem on our yards-per-carry prediction, and we'll get to that below. Fortunately, our Yard-to-Touchdown prediction is going much more smoothly. Through two weeks, every "low-touchdown" receiver has reached the end zone (compared to just two of the "high-touchdown" receivers). At the time of the prediction, Group A had scored 35 touchdowns; they've recorded just 2 in the two weeks since. Group B, by contrast, has already doubled their season total (five touchdowns in the first five weeks, five more in the two weeks since).
Most dramatically, Group A has seen its yard-per-touchdown ratio balloon from 79.5 at the time of the prediction to 404 since. They're struggling to score touchdowns over the last two weeks nearly as badly as Group B was at the time of the prediction (a ratio of 438). Obviously, we should expect this to regress over the final two weeks, as well.
Does Yards per Carry Still Regress?
2026 is the tenth season for Regression Alert, and this was the 14th time we predicted regression in yards per carry. After beginning with a nine-attempt winning streak (spanning from 2017 through 2022), we have lost four of five predictions since the beginning of 2023. Is it possible that something has changed about this statistic that makes it a less reliable candidate for regression going forward?
Well, yes, it's possible. Many good measures lose their utility over time. Height used to be a much stronger predictor of quarterback performance than it is today, while rushing ability was much less useful for estimating career outcomes among signal callers. The league changes. One of the primary flaws with yards per carry is that it is only weakly correlated with player quality, and plausibly that correlation could be strengthening with time. Additionally, yards per carry was typically a safe target because it didn't tend to drive future volume; if that is no longer the case, the prediction becomes much weaker.
But other problems we identified for yards per carry remain—it is noisy and heavily driven by outliers. My instinct is that it is still a great regression candidate and our recent struggles have mostly just been bad luck. To test this, I want to look past the top-line record of our predictions and look at the underlying data. Here are two charts showing the performance of the predictions from 2017-2022 and from 2023-2026:
| Group A | YPG Before | YPG After | % Change | CPG Before | CPG After | % Change | YPC Before | YPC After | % Change |
|---|---|---|---|---|---|---|---|---|---|
| '17-'22 | 72.1 | 58.6 | -18.7% | 12.8 | 13.3 | 3.9% | 5.65 | 4.43 | -21.6% |
| '23-'26 | 72.2 | 69.9 | -3.2% | 13.1 | 14.5 | 10.7% | 5.50 | 4.83 | -12.2% |
| Group B | YPG Before | YPG After | % Change | CPG Before | CPG After | % Change | YPC Before | YPC After | % Change |
|---|---|---|---|---|---|---|---|---|---|
| '17-'22 | 62.9 | 70.6 | 12.2% | 16.3 | 15.7 | -3.7% | 3.86 | 4.49 | 16.3% |
| '23-'26 | 58.8 | 61.0 | 3.7% | 16.2 | 14.4 | -11.1% | 3.62 | 4.23 | 16.9% |
Group A's yard-per-game average went from declining by a lot to declining by a little. Group B's went from rising by a lot to rising by a little. This isn't really an interesting finding; obviously that's why the older predictions succeeded, while the newer ones failed. The useful data comes when we break that yards-per-game average into its two components: carries per game and yards per carry.
Group B continues to see its ypc average increase by about the same amount (as a percentage of its original YPC), while Group A's ypc is still declining, but by less, which accounts for a significant chunk of the change. Does this represent a meaningful trend? It's unlikely; if yards per carry were regressing less than it used to, that should show up in both samples. Since it only appears in Group A, it's more likely just noise.
The bigger issue is that low-ypc backs are being punished by a greater decline in workload, while high-ypc backs are seeing their workload increase significantly more. The prediction hinges on Group B getting more carries than Group A, but Group A has marginally out-carried Group B over the last four years (14.5 per game vs. 14.4 per game).
Add on a moderate increase in the starting margins (Group A's ypg average at the time of the prediction has remained flat, while Group B's has fallen from 62.9 to 58.8), and that's pretty much the story of the current cold streak. It's not so much that yards per carry is no longer regressing as it is that teams shifted workloads more strongly in response to ypc's signals. Is that shift meaningful and intentional, or is it so much random noise? That's the real question we have to watch going forward.
Does all of this mean yards per carry is no longer a good regression candidate? I don't think so. You have to remember that we split our sample in the most unfavorable manner possible. If you draw brackets around the worst shooting stretch of Steph Curry's career, he'll likewise look like a pretty poor shooter. The differences were always going to be exaggerated simply because of the methodology we chose. But it's worth paying attention to the changes in workload as we move forward to see if this is a stable trend or just a statistical artifact.
Do Quarterback Stats Regress?
