By Kurt Bullard
It’s midway through August, so we’re now less than one month away from the beginning of the NFL season. Now, then, is about the time where people begin to scout out their potential fantasy football picks. There is a plethora of fantasy football projections and predictions on the web. Some of this forecasting relies on qualitative analysis more than I’d like, especially coming from writers who aren’t “football guys” (or even football guys’ guys). Overthinking fantasy football drafts tends to create more problems than it solves in my opinion, so focusing on a few principles may be more helpful in the war room than knowing every player’s story.
In this article, I’ll focus on how to look at wide receivers, specifically with respect to what previous stats to look at when drafting a wideout. In the analysis, I looked at the top 60 performing WRs from the 2014 season on a per-game basis to see how they performed in the next year. There were three top-60 WRs who didn’t play in 2015 (Nelson, Benjamin, and Cruz), so my analysis was reduced to the top 57 WRs from the 2014 season.
I then ran a multivariate stepwise regression on 2015 points per game totals against seven variables—yards per reception, non-red zone targets per game, red zone targets per game, non-red zone receptions per game, red-zone receptions per game, red zone touchdowns per game and non-red zone touchdowns per game.
The stepwise regression produced the following model:
It produces a very simple result: non-red zone targets per game are the stat that fantasy football players should look at first and foremost, while non-red zone touchdowns should also be of interest to fantasy owners everywhere. The stepwise regression receptions per game, yards per reception, and all red-zone stats not significant in predicting next year’s per-game performance. Not only does the model only have two variables, but it is relatively predictive as well, as it explained approximately 39% of the variation in WR performance last season.
I found it a bit odd that red zone stats were not predictive of next year’s performance—it seems that you’d want a player who consistently was getting looks in the red zone. However, after creating the following chart for the top 80 players of 2015 ranked by red zone targets to see how they performed the year before, it seems that, unless you’re a blue-chip red zone threat a la Dez or Gronk, red zone touchdowns seem rather variable from year-to-year.
Non red-zone touchdowns, however, are significant in the model. While the reason for their inclusion can only be met with my speculation as to why, it may signal a play-making ability that transcends defensive red zone schemes.
It is also noteworthy that the model does not include any negative coefficient—that is to say, there aren’t any “red-flag” statistics that one should look out for before drafting a wideout.
In the past, my “fantasy advice” has been anti-Rodney Ruxin; that is, keep it simple and don’t overthink. I’ve argued not to freak out about opposing defenses for QBs and RBs when selecting weekly lineups, and not to overreact to a player’s first game against a division opponent when they go up for a rematch. Today, I’ll do the same with WR performance: while there are many factors at play in predicting performance, one stands out amongst the others: targets.
What do the correlations between all of your independent variables look like? Could collinearity be a culprit in why some combinations of independent variables in your models with red zone stats do not perform as well as the chosen model?
This is a really awesome regression model. I am trying to do something like this through stata. I am just wondering if there is a larger data set for NFL players that has all the variables you chose or did you have to go to each player individually and enter the information.
You only tested a single season?