Photo: Timothy T. Ludwig, USA Today Sports

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 dive into the topic of regression to the mean. Sometimes I'll explain what it really is, why you hear so much about it, and how you can harness its power for yourself. Sometimes I'll give some 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 the metric I'm focusing on is yards per target, and Antonio Brown is one of the high outliers in yards per target, then Antonio Brown goes into Group A and may the fantasy gods show mercy on my predictions. On a case-by-case basis, it's easy to find reasons why any given player is going to buck the trend and sustain production. So I constrain myself and remove my ability to rationalize on a case-by-case basis.

Most importantly, because predictions mean nothing without accountability, I track the results of my predictions over the course of the season and highlight when they prove correct and also when they prove incorrect. Here's a list of all my predictions from last year and how they fared. Here's a similar list from 2017.

The Scorecard

In Week 2, I opened with a primer on what regression to the mean was, how it worked, and how we would use it to our advantage. No specific prediction was made.

In Week 3, I dove into the reasons why yards per carry is almost entirely noise, shared some research to that effect, and predicted that the sample of backs with lots of carries but a poor per-carry average would outrush the sample with fewer carries but more yards per carry.

In Week 4, I explained why touchdowns follow yards, (but yards don't follow back), and predicted that the players with the fewest touchdowns per yard gained would outscore the players with the most touchdowns per yard gained going forward.

In Week 5, I talked about how preseason expectations still held as much predictive power as performance through four weeks. No specific prediction was made.

In Week 6, I talked about why quarterbacks tended to regress less than other positions but nevertheless predicted that Patrick Mahomes II would somehow manage to get even better and score ten touchdowns over the next four weeks.

Another Week 7, another yards-per-carry prediction in the books, another dominant victory for regression to the mean. This is now the fifth time that I've pitted the lowest-ypc backs in the NFL against the highest-ypc backs in the NFL, and this is the fifth time that the low-ypc backs outrushed the high-ypc backs going forward. The most notable part? More than half the time, Group B hasn't just had more rushing yards than Group A... it's had a higher yard per carry average, too.

League-wide, all running backs combined are averaging 4.28 yards per carry this season. Since the prediction, Group A backs have averaged 4.29 yards per carry and Group B backs have averaged 4.35 yards per carry. The best running backs in the league in yards per carry were completely average going forward. The worst running backs in the league in yards per carry were completely average going forward. I will probably continue to stress this until the day I die: yards per carry is not a thing.

You know what most definitely is a thing, though? Rushing volume. Remember, Group B was distinguished not just by averaging fewer yards per carry, but by averaging more carries. And that remained decisively true going forward. There were 25 games with at least 18 carries in our two samples, and Group B accounted for 21 of them. As a result, there were 21 games with at least 80 rushing yards, and Group B supplied 16 of them.

Meanwhile, with one week left in your Yard-to-Touchdown ratio, we're on pace for an equally dramatic finish. Our high-yardage, low-touchdown Group B receivers have gained more receiving yards per game in each of the last three weeks. Has Group A's relative advantage at scoring touchdowns allowed them to keep pace? Hardly. Group B receivers average 0.35 touchdowns per game while Group A averages 0.33.

While I certainly wish he could have given me a bit more breathing room, Patrick Mahomes II opened up his featured turn with three touchdowns, putting him on the right pace to live up to the predicted positive touchdown regression.

Do You Even Watch the Games?

Football is a product that evokes strong emotions. Nearly everyone has a rooting interest and a stake in what happens. Football is a sport that people care about passionately, which in turn makes football a sport worth caring about passionately.

Given this level of universal ownership, however, for many, one of the gravest sins someone can commit is speaking ill of "their" team or "their" players. Worse still to make the case using statistics. The inevitable outcry typically takes the same form: "Do you even watch the games?"

It's a rhetorical question, one meant to challenge authority and thereby weaken the argument, but since I can't leave a rhetorical question well enough alone, I wanted to answer it. Yes, unfortunately, I do watch the games.

I consider myself a storyteller at heart. I love the sport because the autumn wind is a raider, not because field goal percentages drop by 2% in winds above 10 miles per hour. I consider myself a student of history, but I couldn't tell you how cold it was during the 1967 championship game other than to say that the game is known as the "Ice Bowl", and the referees couldn't use their metal whistles (they wound up shouting to signal the end of each play), and the announcer (Frank Gifford) at one point mentioned he was going to take a bite of his coffee (because it had quite literally frozen). To me, the numbers surrounding the game are far less interesting than the story surrounding the game.

So I watch football for the same reason I write about football: because I love football. And this is unfortunate because, at least as far as this column is concerned, this probably makes me a worse analyst.

Let's back up and ask a question that I'm sure most of you have asked yourself several times already: what is the point of Regression Alert? A lot of the insight contained here can certainly be actionable. It can identify potential trade targets, for instance. But if Regression Alert were designed primarily as a trade target article it would have more predictions, less theory, and less emphasis on accountability.

Regression Alert clearly points out players who are bound to regress, but if "identify the players most likely to regress" were the primary purpose, I'd highlight some more interesting, complicated, and esoteric statistics— air yards, average depth of target, market share of receiving yards, touches per snap, expected touchdowns, yards after the catch per reception. These are all fantastic regression targets, but instead, I limit myself to the most basic box score stats so that anyone with internet access and the address for Pro Football Reference can follow along at home.

Past predictions have a pretty strong success rate, but I make clear that many of the choices I make in this column directly cut against my expected success rate. If I were so inclined, I could follow 40 players over eight weeks and probably have a 95% success rate, but sometimes it's just more interesting to make a 60/40 call (like last week's Mahomes prediction) or even roll the dice on a 40/60 call (last year's Todd Gurley prediction).

In my mind, the point of Regression Alert is merely to showcase regression to the mean, explain why it is a universal law of the universe, demonstrate how it works, and give examples of it in action. My goal is hopefully to make you believe in its power as much as I do. And every choice I make in this column— putting the primary focus on accountability, mixing predictions and theory in equal measure, refraining from cherry-picking my samples— is made with that goal in mind.

And with that goal in mind, the fact that I actually watch the games is a liability. Because as much as I try to present myself as a neutral bystander in this process, I am not. I get to select which metric I'm focusing on. If I don't like the results of that metric, I'm free to select another until I land on a prediction I'm comfortable making.

While I don't get to selectively include or exclude players in my sample beyond merely setting the cutoffs, setting the cutoffs alone gives me broad power to influence the sample. I can decide after looking at the yards per carry rankings whether I want to compare the top 10 to the bottom 10 or the top 20 to the bottom 20 based on which backs fall in the 11-20 range.

Sometimes I'll choose a more expansive comparison to deliberately try to get some better names into Group A to make the prediction more dramatic. Sometimes I'll put my thumb on the scale to get some better players in Group B to boost my success rate. But in all cases, the subtle choices I'm making to influence who gets included and who gets excluded are based on my belief about which players are good and which are bad, a belief that is largely shaped by my experiences watching the game.

And I have a lot of evidence from the past that my thumb on the scale isn't necessarily a good thing. Maybe I try to avoid a "good" player because I fear he's not going to regress as much, and then he does regress, and my prediction is less impressive than it could have been. Or maybe I try to sneak a "good" player into Group B and he underperforms. Often the players who carry Group B are the ones I would have wanted to exclude if I had the ability.

This isn't just me. Arif Hasan covers the Minnesota Vikings for The Athletic. Despite primarily doing deep-dive film breakdowns, he's always been one of the most analytics-friendly reporters in the league. He also happens to read this column, and we talked a bit about Dalvin Cook a bit after Week 3. At the time, Cook was averaging 6.6 yards per carry, and Arif was saying that of course regression to the mean is real, but is it possible that Dalvin Cook was just so good and his true mean was just so high that maybe it would leave him relatively unscathed?

I was sympathetic to the question. I'm always sympathetic to all of the pushback every time I say that yards per carry "isn't a thing". As a watcher of football, I had likewise been floored by the quality of Cook's play. Despite this, Cook has averaged just 4.1 yards per carry since our conversation.

Watching the games had led our intuitions astray. Not about Cook (who has been and will likely remain a phenomenal talent). But about the nature of yards per carry (which has been and will likely remain a random number generator). If yards per carry didn't look meaningful there wouldn't still be so much profit in betting against it.

This is the core conflict of regression to the mean. If you're watching the games, betting on regression never feels good. It's always a deeply unsettling feeling to bet on players who you think are probably bad to outperform players who you think are probably good. Yards per carry is a parasite that enters the bloodstream through the eyes. I don't think there's a way to watch football and not fall prey.

But again, this is why accountability is such a big part of this column. I can remember how bad past predictions felt and compare that to how successful past predictions have been, and it gives me the necessary push to keep making new predictions. But I can't help but think that this whole process would be a lot easier and possibly even more successful if only I didn't watch the games.