The game of football, and with that our understanding of it, has evolved substantially over the past few years. One of the key insights in recent years has been that weak links truly are one of the most important aspects of team building. For example, in 2020, I wrote this article about defensive players in the passing game:
In short, the defensive backfield goes as the third, fourth, fifth and sixth-best pass defenders go, rather than the first or second. So, while it might be true in some circumstances that coverage is more important than pass rush is, it’s more of a statement about units than the acquisition of individual players (e.g. you should prefer an elite edge player to an elite coverage player, all else equal). This dynamic also exists along the offensive line:
How fragile are offensive lines in pass protection? | NFL News, Rankings and Statistics | PFF
In 2021 I showed, using offensive line data in the passing and running games, the mechanism for this insight. The success of a team hinges on whether or not the offensive line as a unit makes zero mistakes or one or more mistakes. The delta in EPA or success rate in the running game is larger when a mistake is made than it is in the passing game due to the strong-link nature of the quarterback in an offense, but in both cases much is done by making sure the worst player doesn’t make a mistake:
The value of perfectly blocked runs and passes | NFL News, Rankings and Statistics | PFF
With new and better data, my former research group at PFF produced similar results for pass defenses last summer:
The effect of perfectly covered plays on NFL offenses | NFL News, Rankings and Statistics | PFF
As you can see here, when a play is not perfectly covered, meaning that at least one defensive back was charted as having made a mistake, the ceiling for an offense from an expected points standpoint raises substantially, even amidst variance.
Using the Weak-Link Framework in Team Building
While it has been very insightful to rigorously pin down the weak-link nature of football over the past few years, in and of itself it doesn’t add much in the way of prescription to team building. Firstly, it’s really hard to draft average players, and average players – especially at premium positions – are not necessarily cheap in free agency (see Chiefs left tackle Orlando Brown). Secondly, the context in which a player is average matters quite a bit. One of the reasons there is skepticism in the public and private realm for third-party grading data is that it is difficult to account for the differing contexts of roles and assignments on a football field – so a cornerback that travels with number one wide receivers often has weaker ratings than the second corner that covers weaker players, even though there’s a very good chance the former player is performing his job more competently than the latter.
And that’s where we get to the crux of this article, whose inspiration was gathered by reading about a different sport, basketball, in the book Spaced Out, by Mike Prada. Prada explains that, over the past decade or so, basketball lineups have evolved to – while still including all of the necessary tasks like ball handling, rim protecting, and so on – require these traits from different positions. If the team’s best ball handler is actually the small forward, that’s probably fine if the point guard has spot-up ability without the ball in his hands.
This article isn’t about basketball, but it is about how to apply tasks within units, like the secondary, offensive line, and to a lesser extent the receiving corps, to minimize the likelihood of a single failure.
Formulation of the Problem
Consider a group of n players that each have a probability of succeeding on each individual play. Assuming independence, an assumption which has different levels of validity depending on the situation, the likelihood that the entire unit succeeds – e.g. the system doesn’t fail – is the product of all of the probabilities. Assuming that there is a fixed sum of all of the probabilities in the group – an assumption that is reasonable given the constraints of the salary cap – the probability that the system doesn’t fail is maximized when all of the individual probabilities are the exact same.
Thus, the problem of team building in weak-link systems is not simply, or even necessarily, one of team building, but rather one of role selection. In other words, the optimization problem is selecting from a universe of n-role groups the set of roles for which each player has an equal likelihood of success. Jalen Ramsey is likely to be his team’s best player at outside cornerback, but it’s probably not likely that his unit’s best chance to succeed is with him at outside cornerback. Having Ramsey play a more elaborate inside-outside role that is difficult for him to execute – but easier for him to execute than it is for other players on his team – is likely part of the optimal solution for his defense. The Chiefs leaving right tackle Mitchell Schwartz largely on an island during his All Pro years while giving more help to left tackle Eric Fisher is another example of applying this principle to unit building.
So, What?
So, we threw out a relatively popular mathematical fact about team building. How does that help us, exactly? While it’s probably not feasible (yet) to enumerate roles and build an algorithm to find the set of roles that maximizes a unit’s ability to be resilient in the face of opposition, the principle to follow is straightforward.
Be that as it may, it cuts against some of the impulses that teams have when building rosters. Firstly, teams want their heavy investments to perform well. It may be hard for a fan to accept that it’s ok for a star cornerback to have a higher passer rating against than the team’s third cornerback, or that a highly-drafted left tackle is giving up more pressure – even after adjusting for position – than the left guard that is next to him. Modern public charting data, while adding a ton of value to the equation, is likely not rich enough to tease out enough of the assignment to make a normalization correction to a grade or a statistic. There is great incentive for both teams and players to pad statistics rather than to optimize for unit success near the beginning and end of a contract, respectively. This article suggests that there is a team-building edge in avoiding doing so.
The game of football is incredibly rich and evolving. Coaches and analysts are finding ways to maximize team output everyday. In weak-link systems it’s very clear that their job – assigning roles to players based on their traits – is maybe as important as the player selection itself.
