Every year, the NFL trots out a slate of preseason games that eventually (or immediately) turn into tests to see how well you remember Day 3 of the draft or the scout team rosters. So the conventional wisdom has long held that preseason games mean nothing, that they have nothing to teach us about the impending and all-important regular season.
Similarly, individual player performances — at least on the box score level — don’t show any real connection to how they will play when the games count. You don’t have to go back beyond last season to find a perfect example of a summer red herring — Ja'Marr Chase’s preseason struggles didn’t matter once Week 1 came around. In recent seasons, as Los Angeles Rams coach Sean McVay and members of his coaching tree have held key starters out of preseason action entirely, reasons why the preseason logically should not be predictive have piled up.
Maybe those trends will eventually render the league’s August appetizers completely useless, but some recent research has identified one notable way that taking the preseason into account can inform our opinions about how well teams will play going forward — and in many cases put you one step ahead of the Vegas oddsmakers.
NFL preseason results can be predictive — in one key way
Kostya Medvedovsky started out looking at NFL preseason data thinking that more data is almost always better, and that it rarely makes sense to discard information about a player’s or team’s performance when trying to forecast what they’ll do next. See, Medvedovsky is the creator of the DARKO NBA projection system. And in calibrating his projections for basketball, he found that including similarly ignored preseason numbers improved the model.
“In particular, in basketball, you’ve been able to beat Vegas season win totals by just betting on the best preseason teams and betting against the worst teams,” Medvedovsky told Yahoo Sports this week, noting research has also shown predictive qualities for MLB spring training.
There are major differences in the level of data available, and in player usage, between the NBA and NFL. But when Medvedovsky tested his theory about the NFL preseason, he found a material effect. His model determined that, from 2001 to 2019, about 13% of an NFL team’s projection stemmed from the first three preseason games, and the other 87% from existing expectations — something like the Vegas win total projection.
“The difference between a great preseason and a neutral preseason was historically worth about 1.5 wins at the high end,” Medvedovsky said. “That’s pretty big in the context of a 16-game season.”
Even when factoring in preseason results only moved the projection by 0.5 wins, the new projection came down on the correct side of the Vegas line 63% of the time.
In the chart below, you can see how bettors would have fared if they used Medvedovsky’s projections incorporating preseason to choose a side on the standard Vegas line for win totals between 2001 and 2019.
Here’s how the study worked: Importantly, Medvedovsky used point differential — not just wins and losses — from every team’s preseason games from 2001 to 2019. He ran the numbers using both every preseason game and just the initial three — until 2021, the NFL preseason included four games — and found that using the first three was more meaningful.
He combined those results with the existing Vegas preseason win-loss totals and essentially solved for how much weight preseason results carried in the most accurate projections. If preseason games truly told us nothing, it wouldn’t have been anywhere close to the 13% that he found.
A key point to note: Saying preseason results are “predictive” doesn’t mean using only preseason results to project the season. And it doesn’t mean it works every time — the 0-16 Lions of 2008 famously went 4-0 in preseason. It means that adding the proper dose of preseason information to what we already know about teams can more accurately hint at how their seasons might turn out, especially when a particularly strong or weak preseason departs from expectations.
As for why exactly this holds true, it’s a hard nut to crack. It’s possible that testing the depth of talent is still a significant factor in a sport with so many injuries. Or it could be some subtle show of coaching or infrastructure competence. The plays run by the third-stringers are still the plays.
“In short, more data, and more recent data is almost always helpful,” Medvedovsky said. “And it would have been surprising if that wasn’t true of NFL preseason data.”
Which NFL teams could beat 2022 Vegas win totals?
With the 2022 preseason all wrapped up, I wanted to apply Medvedovsky’s findings and see what it could tell us about this year if the pattern holds true. I plugged the 2022 results into a simple version of his formula (using Football Outsiders’ calculation for Pythagorean records) and looked for which teams had their projections moved the most.
That list starts with some supposed bottom feeders who may be at least a little better than you think, but also includes some potential contenders. These are the teams whose projections would move by 0.5 wins or more based on Medvedovsky’s 2001-2019 model.
Houston Texans: The Texans were the only team whose projection moved by more than 1 win. Where their win total over/under is widely pegged at 4.5, the model incorporating preseason results gets them close to 6 wins.
Atlanta Falcons: Atlanta’s strong preseason — perhaps aided by testing out multiple viable quarterbacks and facing a weak slate — also added nearly a win to its projection, making it more of a 5.5-win team than 4.5.
Pittsburgh Steelers: You might count this as a more talented Falcons situation. The Steelers’ rotation of Mitch Trubisky, Mason Rudolph and top pick Kenny Pickett probably gave them an edge at quarterback, but there are always going to be explanations. History says teams that move the needle this much in preseason have wound up proving the projection right against Vegas 63% of the time.
Other teams that fit this bill by the numbers include the Chicago Bears and New York Jets. Three others also make the cut — the Las Vegas Raiders, Carolina Panthers and Miami Dolphins — but you could reasonably counter that they played one or more games against skeleton crews of “sit all the starters” teams.
There does appear to be more reason to put stock in the teams who improved their outlook than those who flailed in preseason. Why? When The Athletic’s Ben Baldwin did his own study to confirm Medvedovsky’s findings, he found a relationship between preseason and regular season point differentials. But you can clearly see that it’s a stronger correlation on the positive side.
I found this surprising so played around with myself and confirmed it. Picture shows relationship between preseason point differential and regular season, but there's also still a relationship conditional on team over/unders prior to season https://t.co/BScd0FOP5n pic.twitter.com/Fx5tb4aYWM
— Computer Cowboy (@benbbaldwin) April 15, 2022
Medvedovsky didn’t distinguish between good and bad preseason swings in his NFL study, but said he wouldn’t find it surprising if poor showings by good teams were less telling. He said in the NBA, that same scenario was materially less predictive.
“Teams with nothing left to prove may not try as hard,” he said.
Unsurprisingly, some teams that made a point of sitting all or most of their starters rated poorly: The Los Angeles Chargers, Tampa Bay Buccaneers and Philadelphia Eagles would all lose at least 0.5 wins off their projection, but there are obvious contextual reasons to think that may not be worth acting on. Others near the bottom — the Jacksonville Jaguars and Minnesota Vikings — might have fewer mitigating factors. It may also be worth noting that the Rams didn’t fare particularly poorly in this model despite sitting virtually all key players.
So Super Bowl contenders are Super Bowl contenders. Losing big in the preseason shouldn’t sway you as long as their loaded rosters remain intact. But if you understand the scale of the information, you can believe your eyes a little bit on some other preseason surprises.