Dynasty, in Theory: Consistency is a Myth

In a system where the typical margin of victory is 30 points or more, the difference between a "consistent" running back scoring 10 points and an "inconsistent" running back scoring 5 in a given week is negligible, but those 30-point spike weeks from the inconsistent back suddenly look a lot more useful.

When analysts model consistency under real-world conditions rather than relying on just-so hypotheticals, they find a negligible effect. For instance, when a poster on Footballguys' message board ran a Monte Carlo simulation on one of his leagues comparing a hypothetical receiver who scored 10 points every week (10ppg average) to a different receiver who scored 5.25 points twelve times and 30 points three times (10.2ppg average), he found both receivers produced an identical winning percentage.

In another simulation, an anonymous author found that if they filled an entire team with consistent players and a second team with inconsistent players, the second team needed to average just 0.5 more points per game (not per player, but for the team as a whole) to achieve the same record as the first.

Because they use intentionally exaggerated examples of "consistent" and "inconsistent" players, studies like these set an upper limit on the potential value of consistency. That upper limit is somewhere around 3 points per season for a maximally consistent player or 8 points per season for an entire maximally consistent team-- just one extra 20-yard reception or one extra 20-yard touchdown run over an entire year.

(This also takes as a given that we can even identify consistent players in advance, which is overly generous. Historically, there is virtually no correlation between a player's consistency from one season to the next.)

Players Don't Exist In a Vacuum

Comparing players one-on-one is dangerous because fantasy football is not a game of one-to-one comparisons. Every consistent or inconsistent player will have six or more teammates around him. 

I was provided an example after the 2021 season. Joe Mixon and James Conner ranked 4th and 5th in points scored but were very inconsistent. Aaron Jones and Josh Jacobs ranked 10th and 11th but were much more consistent from week to week. Someone told me they would prefer the Jones/Jacobs duo to the Mixon/Conner duo, which prompted me to check the box scores.

Individually, Mixon and Conner were inconsistent. Together? Mixon's big weeks happened to coincide with Conner's duds and vice versa. As a result, they outscored the Jones/Jacobs duo in eleven out of eighteen weeks. 

Your Team Is A Portfolio

The key insight here is that when you bundle high-variance assets together, the resulting portfolio has a much lower variance. A single boom/bust player might only boom 25% of the time, but if your lineup features eight boom/bust players, invariably, several players will boom in any given week to offset all the others who busted.

(Hey, if Forbes can write about fantasy football, then Footballguys can write about Modern Portfolio Theory.)

Today, I wanted to test this by seeing if we can find more Mixon/Conner pairs where the duo was more consistent than either back alone.

To do this, we need to define consistency. This is a stickier wicket than it might first seem-- many of the common definitions prove unsuitable for the task. Some choose to define consistency by how often a player hits some sort of minimum threshold-- the two most common ones I've seen are 10 points or "a Top 12 / Top 24 / Top 36 weekly finish".

The problem is that these tend to be less a measure of consistency and more a measure of points scored. (A running back who tops 10 points in fourteen different weeks scored at least 140 points for the season, after all.) They also tend to be fairly arbitrary; by the former measure, a back who scored 9 points in every game would be "less consistent" than a back who scored 11 points four times and zero points every other week.

Another approach is to calculate the standard deviation of a player's weekly scores. This is better-- after all, the whole purpose of standard deviation is to quantify how much a value varies from observation to observation. But standard deviation tends to be proportional to the mean, so this will find that all of the highest-scoring players are also the "most inconsistent".

(For a quick illustration of this, imagine a league with fairly typical scoring settings-- 1 point per 10 yards, 6 points per touchdown, 1 point per reception. Now imagine a running back who averages 15 points per game with a standard deviation of 7 points. If you take that league and double all point values-- giving 1 point per 5 yards, 12 points per touchdown, and 2 points per reception-- the same running back will now average 30 points per game with a standard deviation of 14 points. The standard deviation doubled in size despite the underlying performances remaining unchanged.)

A Suitable Measure

Fortunately, there's a simple fix for this: if we divide a sample's standard deviation by its mean, we get its "coefficient of variation", or CV. A player's CV represents the percentage of a player's average performance that tends to vary from week to week. Players with lower CVs are "more consistent" than players with higher CVs.

With this in hand, I have taken all running backs who have played in at least 15 games this season, sorted them by fantasy points scored in their first fifteen games, and calculated the standard deviation and coefficient of variance for the Top 20 backs. (Why fifteen games? Because an odd number ensures in any comparison, one back will "win" more weeks than the other.)

To make the comparisons cleaner and eliminate any issues with injuries or byes, I am lining comparisons up by game number for the player. As an example, "Game 5" for Bijan Robinson came in Week 5, while for Jahmyr Gibbs it came in Week 6 because the Lions had a Week 5 bye. Here's the full data set for those who would like to play along at home.

We'll describe any back with a CV at least 0.05 points lower as "more consistent", any back with a CV at least 0.05 points higher as "more inconsistent", and any back with a CV within +/- 0.04 points as "equally consistent". From here, we just need to find pairs where a more consistent back outscored a more inconsistent back in at least 8 of 15 weeks despite averaging fewer points per game over the full sample.

Finding Our Huckleberries