I built my first set of projections in the summer of 2012. They were awful. Why? I wasn’t thinking about each team holistically and accounting for dependent relationships. Since then, I’ve been building tools to help me do a better job. In a couple of days, I’ll be releasing a tool that will streamline the process for those of you that have never does so but would like to build your own set of projections. My goal was to simplify the process and make it less overwhelming.
Why Build This Tool?
One of the most common questions that I see on Twitter, hear asked on podcasts and receive emails about is ‘how to do projections’. While I don’t believe that there is a single way to correctly do projections, I am certain that there a number of ways to do them incorrectly. For example, if a set of projections places every wide receiver on the Steelers with 10 or more touchdowns and 1,000 plus yards but Ben Roethlisberger is only projected for 2,750 yards and 28 touchdowns, something has gone terribly wrong.
Start at the Top, Work Your Way Down
It makes sense that all of the yards and touchdowns accumulated by Steelers WRs, within the passing game, would aggregate up to Roethlisberger’s passing production. Of course, this assumes that we are projecting Roethlisberger to play the entire season. For our purposes, let’s forget about the caveats and nuance. Ultimately, the point that I’m trying to make is that in order for a set of projections to be done correctly, we need to ensure that all stats which are tied together (for example, receiving and passing yards and passing attempts and targets) are in agreement. For this reason it’s best to take a top-down approach.
First, you should come up with an estimate of the total plays that you expect an offense to run. From here, you will then divide this number between rushing and passing attempts. If you wanted to get really fancy, you’d subtract out sacks (the version of the tool I’ve put together to help you with the process does not get this granular. However, the Projection Machine at RotoViz.com does. It also includes a couple of other neat features. So I’d definitely recommend checking it out). I like to look at an offenses play-split from the prior five or so seasons and consider it in relation to the team’s win/loss record. I’ll then check in on the team’s Vegas Line and come up with a guesstimate for the current year.
Once you have a projected number of offensive plays and have split between rushing and passing attempts, you should then allocate a percentage of each to the team’s offensive players. In general, I will allocate a percentage of a team’s rushing attempts to its starting quarterback, RB1, and RB2. All remaining rushing attempts will be allocated to the team’s RB3+. Likewise, I then allocate the team’s passing attempts to its WR1, WR2, WR3, RB1, RB2, RB3+, TE1 and TE2. All remaining targets are then allocated to the team’s WR4+. This process ensures that we are dividing the available workload in a way that, at least mathematically, makes sense.
With a number of rushing attempts and passing targets now established for each player, you can project efficiency stats such as yards per carry for each player. These efficiency stats will be multiplied by the workload of the player they belong to in order to arrive at his total output which can be used to calculate fantasy points. It’s important to recognize that efficiency stats can vary widely (even for the same player) from year to year. For this reason, you’ll want to make sure that if you’re looking at a player’s prior season, it wasn’t an outlier. Further, you should be hesitant to project players at rates that are far above league average. Exceptions for players like Dez Bryant and Eric Decker, that historically produce very high touchdown rates, can be made but prudence is recommended.
So, projections are awesome right? WRONG!!! I am really anti-projection. This is not say that they aren’t useful, important and necessary but they are often used incorrectly. The final product often gets dumped on us with little context. We’re presented with a single number. One that is sometimes presented as the end all and be all of a player’s worth for the coming season. Even the best sets of projections include big misses and are miles away from perfect. There are just too many variables in play. So we shouldn’t be viewing a player as this single projection. Plus, when we sit down to look at a set of projections, we don’t know if their creator was going for best, mid or worst case scenarios. Were they aiming for conservatism, optimism or trying to be different for the sake of being different. More importantly, were they even being consistent when doing so?
Player A may be projected for 20 more points than Player B but does this really mean much? Depending on a number of variables it might not. This is why you should never draft strictly off of projections and why tiers have so much more utility.
Enter A Range of Outcomes
Rather, we should view a player as a range of outcomes. This is why there is a place for projections. We can use them to get a sense of what a player’s output might look like in a good, great or bad season. By tweaking a couple of key variables, we can get a sense of the things that could limit a player’s potential or realize that there is more room for him in an offense than we previously thought.
However, if we don’t take the time to explore these ranges, we are really limiting the usefulness of the projection. In my opinion, the process of putting the projection together and understanding the key components that will drive a player’s outcome are the fruits of your projection labors; not so much the final numbers.
For the above reasons, I think it’s important to work out a high, mid and low projection for each player. Once you’ve explored the possibilities of what the coming season could look like for a player, you’ll have a better idea of how to value him and understand that he can’t be encapsulated into a single projection. I’d really like to expand on the above but I think I’ve driven home my point enough.
Moral of the story, don’t get married to projections. They are not gospel and should not be treated as such. If you are going to use the tool please exercise caution and keep the above in mind!!!