We Should Stop Using Per-Target Numbers to Predict Future Fantasy Football Success
We don't always ask the right questions when reading fantasy football analysis. We're fed information -- oftentimes amazingly well thought out information -- but what's presented isn't always as valuable as it might seem.
We don't always ask the obvious question of Why?
Take this article that I wrote on five players whose stock took a hit since the NFL's free agency period started. In it, Kenny Britt was featured. And when analyzing Britt, one of the statistics that I threw up in a table was yards per target.
If you read the article -- or if you're looking at it quickly right now -- did you ask yourself...Why?
Describing versus Predicting
We all know there's a difference between what's already happened and what will happen. Our language even dictates this: Tony Romo was a good quarterback, whereas Jared Goff will be a good quarterback (he won't).
In a way, this is no different than what we see with data. A touchdown-to-interception ratio may tell us what's happened in the past, but does that same metric tell us what's going to happen in the future? Not necessarily.
There's nothing wrong with using touchdown-to-interception ratio to describe the way a quarterback played. I'd argue there are plenty of better ways to do it, but no one is wrong by saying Tom Brady threw 28 touchdowns to 2 interceptions for a 28-to-2 touchdown-to-interception ratio. And I wasn't wrong by saying Kenny Britt had the best yards per target rate among relevant Rams wide receivers this past year.
If I were to pretend that that number meant something for 2017, though, then I'd be wrong.
That's why you have to ask Why?
Some statistics are just more predictive than others. That's part of the reason we created and have our expected points model, Net Expected Points. It's a more accurate representation of value when compared to traditional statistics, so we're able to utilize it in a forward-thinking, predictive way.
This is particularly true at the wide receiver position, since so much of a player's production relies on the cohesiveness of so many pieces coming together. While a quarterback, for instance, is able to control so much of a game, a wide receiver relies heavily on good passes from a quarterback, strong teammates to take coverage away, and much more.
In turn, there's a lot of variance created among traditional wide receiver metrics. What a receiver does one year within a statistic might not tell us a whole lot about what he's going to do the next.
Especially with rate stats.
The chart below shows a list of wide receiver statistics and the r-squared values -- "a number that indicates the proportion of the variance in the dependent variable that is predictable from the independent variable(s)" -- of these metrics year over year, dating back over the last five seasons. To put this another way, it's looking at how sticky these statistics are from one year to the next among all receivers over the last five seasons. The closer the number is to zero, the less worthwhile.
|Yards per Target||0.03|
|Yards per Reception||0.01|
|Yards After the Catch||0.49|
|Yards After the Catch per Catch||0.03|
|Fantasy Points per Target||0.01|
Volume is everything.
Plenty of analyses like this have been completed before, and many have drawn a similar conclusion: efficiency really doesn't matter a whole lot. That is, efficiency doesn't matter a whole lot when you're trying to predict how a particular player is going to perform moving forward.
Why is that? Well, here's a shameless plug, but Matt Harmon talked about it a bit during Episode 2 of The Late-Round Podcast, which you can find below (34-minute mark).
Harmon said, "A target can be an indicator of quality. It might not necessarily be something that should be a denominator in an equation for a player [evaluation], because you're inherently welcoming in other variables...in terms of What's the quality of targets they're getting? What's the quality of quarterback play? Where are they being targeted?"
The reason these rate stats don't correlate all that well from one year to the next is due to a compounding number of variables. A wide receiver can be used differently in Year N to Year N+1, a quarterback may play better or worse, and so on. That will change the quality of targets, which will change the denominator in the fraction. Thus, altering the stickiness of the metric.
Even when you take the wide receivers over the last five years and remove the guys who saw a 50-plus target difference from one year to the next â€” that is, to tighten things up and show only players who saw similar volume from one year to the next â€” things remain fairly similar. We naturally see better correlation between volume-driven statistics, but per-target metrics still aren't useful.
|Yards per Target||0.04|
|Yards per Reception||0.01|
|Yards After the Catch||0.70|
|Yards After the Catch per Catch||0.02|
|Fantasy Points per Target||0.02|
This doesn't make per-target metrics completely worthless. They can still describe a situation that's already occurred. But if you -- anyone -- are using them in standalone form to help predict what's going to happen next season, remember to ask yourself the question...