Correlations in Fantasy Football

Last updated: July 27th, 2020
In a real life NFL game, fantasy points are related.
An easy example: a QB throws a touchdown to his WR1, and they both get points.
Correlation is just the tendency to move together. If you knew NOTHING about a GB game, but saw Aaron Rodgers put up 500 yards and four touchdowns, you'd expect Davante Adams to have also had a better game than average, right?
It's not for sure (maybe Davante got hurt early or Rodgers threw all his TDs to other guys), but it
to happen.

Correlation is measured from -1 to 1

The correlation between two pieces of data (Rodgers points, Davante Adams points) is usually measured on a scale from -1 to 1, where 1 means the two move perfectly together, -1 means the two move perfectly in opposite directions, and 0 means they aren't correlated at all.
The QB-WR1 relationship is one of the strongest in a football, at about 0.40. The only one stronger is the
correlation between a QB and the DST they're facing, which is about -0.60.

All players in a game are correlated to varying degrees

The QB-WR1 correlation is easy to understand. A more subtle example: QBs facing each other in the same game are positively correlated at about 0.20.
(not always, otherwise the correlation would be higher), teams will get in a shoot out, and both QBs have to air it out to keep up. Other times, the weather will be awful, and no one can pass the ball downfield.
While any correlation is a single number between two pieces of information, we can look at
correlations using a matrix, like this:
Correlations By Position
To see the correlation between any two positions, just look at ROW and COLUMN of the position you're interested in.
For example, to check the correlation between a WR1 and the opposing QB, we look at the WR1 row and follow it over till we see the OPP QB column. It's 0.09, which is small, but not completely independent.

Correlations + Distributions

The Fantasy Math model works by (1) taking thousands of draws of randomly generated data following the correlation rules above, then (2) feeding it those draws through the modeled
probability distributions
The result let's you factor in correlations when calculating win probabilities or making who do I start decisions. You just look at all the thousands of (correlated) simulations, and see how often you win or start whoever maximizes your chances.

How Correlations Impact Your Matchup

Imagine you have a close call — say Tyler Lockett vs Stefon Diggs — and your opponent is starting Russell Wilson.
The fact Wilson and Lockett's points are positively correlated might affect your optimal decision. Precisely HOW it affects your decision depends on the rest of your matchup:
if you're heavily favored, the correlation might mean you should start Lockett as a hedge against Wilson blowing up
if you're the underdog, maybe you need Wilson to underperform AND Diggs do really well to have any chance at all, which means Diggs maximizes your probability of winning
That's a simple scenario, but it can easiliy get more complicated. For example, maybe you're heavily favored (so correlations suggest Lockett), but you actually have Diggs as slightly better. Is an opportunity to hedge against Wilson worth starting someone with a lower expected points?
And what about any other correlations in your matchup?
The model takes all this into account. You don't have to think about it or weigh the trade offs subjectively, it's all factored in.

Model Out Soon

The full Fantasy Math model that lets you keep track of correlations across your entire matchup will be out soon. Enter your email below to hear when it's ready.
Get Notified!