Photo: 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 touchdown rate, and Christian McCaffrey is one of the high outliers in touchdown rate, then Christian McCaffrey goes into Group A, and may the fantasy gods show mercy on my predictions.

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 my predictions from 2019 and their final results, here's the list from 2018, and here's the 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 talked about how the ability to convert yards into touchdowns was most certainly a skill, but it was a skill that operated within a fairly narrow and clearly-defined range, and any values outside of that range were probably just random noise and therefore due to regress. I predicted that high-yardage, low-touchdown receivers would outscore low-yardage, high-touchdown receivers going forward.

In Week 5, I talked about how historical patterns suggested we had just reached the informational tipping point, the time when performance to this point in the season carried as much predictive power as ADP. In general, I predicted that players whose early performance differed substantially from their ADP would tend to move toward a point between their early performance and their draft position, but no specific prediction was made.

In Week 6, I talked about simple ways to tell whether a statistic was especially likely to regress or not. No specific prediction was made.

In Week 7, I speculated that kickers were people, too, and lamented the fact that I'd never discussed them in this column before. To remedy that, I identified teams that were scoring "too many" field goals relative to touchdowns and "too many" touchdowns relative to field goals and predicted that scoring mix would regress and kickers from the latter teams would outperform kickers from the former going forward.

In Week 8, I noted that more-granular measures of performance tended to be more stable than less-granular measures and predicted that teams with a great point differential would win more games going forward than teams with an identical record, but substantially worse point differential.

In Week 9, I talked about the interesting role regression to the mean plays in dynasty, where the mere fact that a player is likely to regress sends signals that that player is probably quite good and worth rostering long-term, anyway. No specific prediction was made.

In Week 10, I explained why Group B's lead in these predictions tended to get smaller the longer each prediction ran and showed how a small edge over a huge sample could easily be more impressive than a huge edge over a small sample. No specific prediction was made.

In Week 11, I wrote that yards per pass attempt was an example of a statistic that was significantly less prone to regression, and for the first time I bet against it regressing.

In Week 12, I talked about "on pace" stats and how many of the players who wound up setting records were not the players who were "on pace" to do so.

In Week 13, I came up with a list of players who were getting hot just in time for the playoffs... and then explained why they probably weren't getting hot just in time for the playoffs, predicting that they'd cool off back to their normal production level going forward.

In Week 14, I offered the cold comfort that if you lose in the fantasy playoffs, the odds were never in your favor, anyway.

In Week 15, I made our last prediction of the year, once again looking at yard-to-touchdown ratios for touchdown regression. I also noted that if we cut the duration of our prediction in half, we could make up for it by doubling the size of our prediction to offset.

In Week 16 I charted how incoming talent tended to ebb and flow over time, which led to positions trending older or younger in a process which was itself prone to regression.

Today we close the books on our last two predictions of the season. Our "hot" players essentially regressed as predicted; they averaged 11.28 points per game from Weeks 1-12 but 15.89 points per game from Weeks 9-12; over Weeks 13-16 that fell back to 10.89 points per game, right in line with their full-season averages.

Now, some of the players in the sample did not, in fact, appear to regress. T.Y. Hilton averaged 7.98 points over the full season but 10.90 over the preceding four weeks. In his case, it seemed that this hot stretch was a harbinger for things to come as Rivers seemingly rediscovered the receiver and Hilton averaged 17.68 the rest of the way. But this is why we make rigorous predictions; four weeks ago, why would one have preferred T.Y. Hilton to, say, Marques Valdez-Scantling, who averaged 9.10 points over the full season and 12.68 over his last four weeks? Valdez-Scantling averaged 5.03 points per game over the last four weeks, including 0 (or fewer!) yards in three out of the four games.

This is why we use group averages; while individual outcomes are noisy, when you average them out you can clearly see the trend emerge.

Similarly, one could look at Davante Adams scoring three touchdowns last week and ask whatever happened to that touchdown regression he was due for? Adams and Mike Evans were both in the "negative regression" bucket and put up 30+ points in the championship game; they were league-winners. Jeff Wilson and Nelson Agholor were listed as candidates for negative regression and they both scored 20+. Adam Thielen, JuJu Smith-Schuster, and Jimmy Graham scored 15+. Where was the regression we were promised?

But on the other side of the ledger, Stefon Diggs lived up to his expected positive touchdown regression, joining Adams and Evans in scoring 30+ points. And CeeDee Lamb, Brandin Cooks, Myles Gaskin, and Michael Gallup all topped 20 points, with 7 combined touchdowns in Week 16 after combining for just 12 touchdowns in the fifteen weeks prior. Indeed, the headliners in Group A obscure the fact that our "high-touchdown" players scored 12 times in 25 combined games, while our "low-touchdown" cohort scored... 13 times in 18 games. One more touchdown in seven fewer games.

At the time of our prediction, players in Group A were scoring one touchdown for every 77 yards from scrimmage while players in Group B were scoring one for every 277 yards. Since our prediction, Group A scored one touchdown for every 156 yards while Group B scored one for every 146. A "nose for the end zone" isn't really a thing, it's mostly just random noise; over a long timeline, touchdowns tend to follow yards.

Our Final Report Card

To wrap up the season, I wanted to look back not just at this year's predictions, but at every prediction over the last four years. Remember, I'm not picking individual players, I'm just identifying unstable statistics and predicting that the best and the worst players in those statistics will both regress towards the mean, no matter who those best and worst players might be.

Sometimes this feels a bit scary. Predicting that stars like Davante Adams and Tyreek Hill, in the middle of historically great seasons, are going to start falling off is an uncomfortable position. But looking back at our hit rate over time makes it a bit easier to swallow.

Top-line Record

• 2017: 6-2
• 2018: 5-1
• 2019: 7-2
• 2020: 6-1
• Overall: 24-6 (80%)

The Misses

2017 Passing Yards per Touchdown Part 1
2017 Passing Yards per Touchdown Part 2

In our first prediction, Group A was outscoring Group B by 13%. I picked a bad four-week span to make the prediction, as they outscored Group B by 17% over our prediction span, but over the full season that fell to just 3%; solid regression, but not enough to count the prediction as a win. When I repeated the prediction later in the season it once again went poorly. My takeaway from this experience was that quarterback yard-to-touchdown ratios were much more skill-based than running back or receiver ratios (an idea that's backed up by looking at the leaderboard in the statistic), so I've stopped making this prediction anymore.

2018 Yards per Target

Just like with the last miss, I tried to make a prediction out of a statistic that had a large skill element to it. Over the full season, Group A's edge fell from 16% to 7%, which was at least movement in the right direction, but not enough to qualify as a win. Once again, I've stopped trying to figure out clever ways to make this prediction work, because the skill signal is just too strong, which means the movement going forward tends to be far less dramatic and the prediction is a bit less reliable. (We did log one hit to offset this one miss before I discontinued the prediction.)

2019 Patrick Mahomes II Touchdown Regression

I knew going into this prediction that it wasn't a great bet; in fact, I preceded the prediction with 18 paragraphs and 4 charts detailing the three biggest issues with the prediction I was about to make, then compounded the issues by breaking best practices again to make a prediction about a single player rather than a large sample (where the ups and downs would have more chance to even out), and broke them a second time by specifically choosing my player rather than sticking with whoever happened to be most extreme in the statistic I was betting on regression. Then when the original prediction lost in part because Mahomes was injured during the sample, I doubled down when he returned from injury and ran it again; this prediction was responsible for both of my losses that year. Really just a disaster from start to finish with a pair of humbling and well-deserved losses to show for it.

2020 Point Differential vs. Record

I paired teams who had the same record despite wildly different point differentials and predicted that the teams that were winning by bigger margins would win more games going forward than the teams that were winning by smaller margins. Not only did that prediction not work out over the four-week sample, extending it out through the entire season didn't help any; our Group A teams have actually won one more game than our Group B teams since the prediction. The lesson I take away from this failure is... nothing. Sometimes predictions fail because I got greedy or made an ill-advised design choice. But sometimes we just get unlucky. In the future, I'd be happy to make this bet again.

The Hits

Here's the outcome of all of my "Yards per Carry" predictions over the years, with the average at the time of the prediction, the average in the four weeks after the prediction, and the total swing.

• Group A had a 60% lead, Group B had a 16% lead, +76% total swing
• Group A had a 25% lead, Group B had a 16% lead, +41% total swing
• Group A had a 24% lead, Group B had a 4% lead, +28% total swing
• Group A had a 9% lead, Group B had a 23% lead, +32% total swing
• Group A had a 20% lead, Group B had a 30% lead, +50% total swing
• Group A had a 22% lead, Group B had a 23% lead, +45% total swing
• Group A had a 3% lead, Group B had a 36% lead, +39% total swing

We can't directly compare the total swings since the sample sizes vary so much (a 30% swing over a large sample might be more impressive than a 50% swing over a small one), but this prediction has gone 7-0 for me over the years with a median swing from Group A to Group B of of 41% and a minimum swing of 28%. I've made a lot of jokes about yards per carry over the years. I've called it "pseudoscience" and said it's "not a thing" or even "maximally not a thing". Some people find these statements provocative, but they're not intended to provoke. Yards per carry genuinely is almost entirely noise, especially over the kinds of samples we're dealing with inside a single season. Here are the receipts.

Here's the outcome of all of my "Yard to Touchdown Ratio" predictions over the years. (Where necessary, I've reworked some of the predictions to adhere to our traditional "Group A vs. Group B" format. This is a purely cosmetic change for comparison; the underlying data remains untouched.)

• Group A had a 28% lead, Group B had a 1% lead, +29% total swing
• Group A had a 21% lead, Group B had an 8% lead, +29% total swing
• Group A had a 7% lead, Group B had a 20% lead, +27% total swing
• Group A had a 28% lead, Group B had a 23% lead, +51% total swing
• Group A had a 26% lead, Group B had a 4% lead, +30% total swing
• Group A had a 23% lead, Group B had a 47% lead, +70% total swing
• Group A had a 22% lead, Gorup B had a 23% lead, +45% total swing
• Group A had a 2% lead, Group B had a 40% lead, +42% total swing
• Group A had a 15% lead, Group B had an 11% lead, +26% total swing

This is my favorite prediction for a number of reasons. For starters, this entire column was inspired by a pair of articles I wrote on this ratio back in 2015. But mostly I love it because it's such an incredible, slam-dunk regression target that virtually no one pays any attention to. Statistically-minded writers have known about the issues with yards per carry for decades now, but when I started writing this column virtually none of the discussion of players who were scoring "too many" or "too few" touchdowns linked that judgment to their yardage profile. But as you can see, that's exactly the link we should be making; predictions that touchdowns will follow yards are 9-0 with a median swing of 30% and a minimum swing of 26%.

Here are the various other miscellaneous (successful) predictions from the past four seasons

• Group A had 16% more yards per target, Group B had 11% more yards per target, +27% total swing
• Group A had 17% fewer interceptions, Group B had 57% fewer interceptions, +74% total swing
• Group A had 13% fewer interceptions, Group B had 17% fewer interceptions, +30% total swing
• Group A had 20% more kicker points per game, Group B had 36% more kicker points per game, +56% total swing

And general regression predictions that didn't follow the typical "Group A vs. Group B" format, instead predicting unidirectional regression for a single group.

• "Extreme" offenses and defenses regressed 11% toward the league average performance, as predicted.
• Defenses regressed 12% more than offenses, as predicted.
• Group A averaged 14% more passing yards per game, Group A continued to average 28% more passing yards per game, as predicted.
• "Hot" players regressed 108% of the way back to their full-season averages, as predicted.

I really wish interceptions played a much larger role in fantasy football because they're every bit as regression-prone as yards per carry and yard-to-touchdown ratios and there'd be so many opportunities to profit off of that knowledge. Kicker field goal to extra point ratios also regress pretty reliably, a fact I take advantage of when modeling kickers for my weekly Rent-a-Kicker feature (but which I wanted to point out explicitly in this column at least once).

Anyway, the whole point of this column is to convince you that regression to the mean is real, it's implacable, and it's actionable with very little effort on our own part. Accountability is crucial to making that point, which is why I go to such great lengths to track and report my results. You don't have to take my word on the subject, you can go back and check my track record for yourself. You can see why I'm such a big believer in the power of regression, and hopefully, you become something of a believer yourself.

As always, I appreciate you reading along this season, and look forward to doing it all over again in 2021.