Similarity based projections are probably the most powerful thing that I use to draft my fantasy team that most fantasy owners just don’t have at their fingertips. Similarity based projections address shortfalls in other projection systems that might have a tough time accounting for variables that aren’t linear (age for instance) or perhaps variables that might interact with each other (like receiver size and touchdown totals). Instead of saying that a receiving yard is worth 0.7 receiving yards when projecting the following season’s fantasy results, as a regression formula might tell you, I just look at how a group of similar players performed and make an assumption that my subject player might fit within the range of how similar receivers performed. Similarity scores were the reason that I had Vincent Jackson on every single one of my fantasy teams in 2012.
Generally speaking, similarity scores based on a dataset that includes the entire previous fantasy season are preferable to similarity scores based on a dataset that has been filtered in some way (like 2nd half only for instance). But in some cases there are legitimate reasons for filtering the dataset. For instance, when we forecast Michael Crabtree for next year, is the relevant dataset his entire season performance? Or might the relevant dataset be his performance when Colin Kaepernick was starting? If we have a player that was injured and left a game after being targeted just once, like Jordy Nelson, would it be reasonable to throw that game out of the dataset so that it doesn’t drag down the player’s per game numbers?
I have a module below that will generate 15 similar player-seasons for any wide receiver that you choose. Not only that, but you can also throw games out if you want to. If you want to see what Michael Crabtree’s averages look like based just on the time that Kaepernick started, uncheck all of the boxes before the Chicago game. When you check and uncheck boxes, not only will the player’s averages change, but the 15 similar players might change as well. The key thing that I look at when I run similarity projections is how the comparable group fared in the year following the season that they were similar to the player I’m looking at. You can see those results by clicking on the tab that says “Season N+1”. Lastly, if you want to see a plot that shows the distribution of year-over-year change in fantasy point scoring, you can click on the “Plots” tab.
Maybe you have a few guys that you’re thinking about keeping for next year and you’re not sure? Check them all out, adjust the relevant dataset and then look at the N+1 results for the Comparable Players.
*Note that the stats should be self-explanatory, except for perhaps FPOP, which is simply a fantasy efficiency metric. Think of it like Points/Target adjusted for field position.