Thursday, Sep 5th
Monday, Sep 9th

Understanding Projection Fragility


It’s Saturday night and you’re putting rosters together for Sunday. You fire up your favorite projection system and you see a bunch of guys projected super close to each other. How do you start to determine who you like more or less out of a group of similarly-projected players?

There are a lot of factors that go into this, of course. Ownership, correlation with the rest of your roster, maybe even just a gut feeling of who you like. One area that I don’t see talked about much, however, is the variance inherent in each player’s projection (i.e. how robust or fragile a projection is). Two players might both be projected with a median outcome of 12 points, but does that mean they’re projected identically? In most cases, the answer is no. 

Slot vs Perimeter

If we think through this, it makes a lot of sense. Imagine our two players with a projected median score of 12 points. One is a high-volume, low-ADoT slot receiver, while the other is a fast perimeter wideout who projects to only see 3-5 targets per game but can take one to the house from anywhere. This is a pretty easy example; the perimeter wideout here has the more fragile projection. He could very easily end up with anywhere from zero to two catches and hardly any fantasy points or he could take one to the house from 80 yards out, add on another catch or two, and be at 20+ fantasy points. The slot receiver is highly unlikely to get to 100+ yards, but he’s also highly unlikely to result in a complete dud that craters a roster. The perimeter wideout’s projection is more fragile, while the slot receiver’s projection is less fragile (or, “robust”). For the math folks in here, both players might have the same mean or median projection but their standard deviations can be significantly different. 

Unclear Workload

Another slightly trickier example is a situation with an unclear workload. Imagine a team’s starting running back gets hurt and we don’t have a lot of clarity on who’s going to take over. A projection system will likely split the workload between the two potential options, resulting in fairly mediocre projections for both of them. But in reality, that’s rarely how it works out; one guy will get more of the work, either due to a coach’s plan or due to just getting hot and being given more work as a result. In this case, the projection of both of these running backs is extremely fragile, in that we know a median outcome (from a projected workload perspective, at least) is a highly unlikely outcome. Odds are we don’t look back at the end of the day and see that the two running backs had the same number of carries; one of them is probably going to see the lion’s share of the work.

How to Utilize

So how do we use this? A simplistic way of thinking would be something like “play low fragility guys in cash games and high fragility guys in tournaments.” That isn’t entirely incorrect (you do generally want to build a cash roster with a relatively narrow range of outcomes), but it’s incomplete. Whether or not you want to add a higher or lower fragility play to a given roster depends on the rest of the roster and on the type of contest in which the roster is entered. In a large-field GPP where you know you need ceiling performances across the board, I would tend to lean towards more fragility (especially at low ownership). In a smaller-field tournament or in cash games, where you don’t need a huge score to win and it’s more about having a solid top-to-bottom roster, a lower fragility play may be a better fit. You also want to consider the roster as a whole: are you taking a lot of risks elsewhere? Maybe a low fragility play for that last spot is the right decision. But if you’re already using several robust plays, that roster could probably benefit from some additional fragility. 

Another good rule of thumb that I use (and you’ve probably seen me say this more than once if you’re in the OWS Discord; which you should be!) is “play high-variance plays at low ownership, not at high ownership.” Now, this is just my personal playstyle, but when a player with a fragile projection becomes a central point of chalk on a slate, I’m generally going to avoid that play. That is not because I think he’s a bad play, but because with a high-fragility projection, his likelihood of putting up a completely roster-cratering score is relatively high, and I’m happy to let a large chunk of other rosters on the slate take that risk (of course, he could also bury me; like Will Fuller’s 50+ Draftkings point game a couple of years ago when he was the highest owned receiver on the slate!). But, we have to take risks in order to win. When a player has a very robust projection (think Christian McCaffrey in a great matchup), I’m less likely to worry about ownership there, because a player with a very robust projection is unlikely to put up a score that will kill a roster. This is, incidentally, where the old adage of “eat chalk at running back, be contrarian at receiver and tight end” comes from; running back projections are much more robust than receiver projections because running backs generate scoring from a much larger volume of touches than receivers do. 

The important takeaways here:

  • Players projected similarly can, in fact, be wildly different; just because their median projection is the same does not mean their range of outcomes is!
  • Fragility is not a good or bad thing. It’s just a thing. Learn to work with it and use it to your benefit. 
  • Think about fragility across a whole roster, not just in terms of one player. A good roster has the right amount of fragility across all of its plays for the contest in which you’re entering it. 

Learn from X!

<< Optimizer Training >>