The Game Level Similarity Projections — or GLSP Apps — are one of the RotoViz tools that I find invaluable for making lineup decisions.
Unlike traditional projections, these apps give a range of outcomes for each player’s weekly performance. The apps report each player’s low (25th percentile), median (50th) and high (75th) projections for the week, based on how comparable players have performed against comparable defenses.
- 50 percent of the time (e.g. 75th percentile minus 25th percentile), the player is expected to produce a score between his low and high projection.
- 25 percent of the time, the player should under perform their low projection, or over perform their high projection.
- The median projection provides a benchmark, with even odds of the player producing more, or less, points.
- The projections are most valuable in relation to each other. In other words, two players might have similar medians, but one has a much higher ceiling than the other.
- A large range of outcomes can suggest volatility. That might be good or bad, depending on what your lineup needs are.
So how exactly do the apps work, and how well do they perform? I thought today would be a good time to take a look at that, since we’ve had a few games this week, and most of us either have today off or are only pretending to work. Grab a drumstick and read on.
Keep in mind that this is just a very small snapshot — not even a full week of games. Over the course of a full season I’d expect the apps to perform even better.
Not counting Indianapolis, who played with a backup, 4 of 5 quarterbacks played below their high projection,1 and 4 of 5 played above their low projection. Sam Bradford under performed his low projection, but only by two points. Going in to the week, you knew (unlike GLSP) that Bradford was likely going to be without his top WR, so an underwhelming performance wasn’t a surprise. Not much to say about Kirk Cousins, who wildly out performed his projections.
Of the eight fantasy-relevant running backs to play yesterday, six played below their high projection, and seven above their low projection. The one downside disappointment — Frank Gore — came with a pre-game red flag that GLSP wasn’t able to consider (no Andrew Luck).
Only Pierre Garcon, Sammie Coates and T.Y. Hilton came in below their low projection. Coates had injury concerns coming into the game, and Hilton was injured during the game. Four of the 17 exceeded their high projection, but Donte Moncrief likely benefited from Hilton’s injury.
There’s another lesson here that can be sometimes gleaned from GLSP. Note the Detroit WRs; GLSP correctly liked both Anquan Boldin and Golden Tate more than Marvin Jones. Even though Boldin and Tate both beat their high projections, and Tate came close to his, GLSP also correctly suggested that Detroit’s passing game was set to be underwhelming yesterday. Boldin was the only one to really help you, while neither Tate, Jones, nor Matthew Stafford helped you much yesterday.2 GLSP didn’t do as well with Washington’s WRs, but I do think it’s usually informative to compare WRs on the same team, especially when that team (like Green Bay or Arizona) has multiple fantasy relevant options. GLSP often gives a clue about which one is likely to over- or under-perform in a given week.
Two came in above their ceiling, and three (of 10 total) below their low projection. On the under performers, you probably should have expected a poor game from Jesse James, since Ladarius Green had recently been activated. That’s the type of recent information GLSP doesn’t know, but you do. That, combined with James low projected ceiling, hopefully kept you from starting him.
Let’s consider a couple examples. First, Jordan Reed and Jason Witten. Reed had a huge game, while Witten had a dud. But you’ll notice that both had a very large range of potential outcomes. In other words, their projections should have been considered volatile, and their outcomes certainly were. Again, GLSP isn’t going to always nail volatility, but I think it will more often than not, and it’s something to keep in mind.
Second, let’s compare Reed and Vernon Davis. Yes, Reed had a huge game, but we expect that from him. GLSP also liked Davis, giving him nearly a 15-point ceiling. He didn’t quite get there, but 12 points was still a very solid outing.
Hopefully this exercise gives you a little more insight into how GLSP works. It does a good job of identifying a range of outcomes for each player, especially when considered against other players. It’s not perfect, but it illuminates situations that are obfuscated by simple point projections. It’s a great tool for helping you make difficult start / sit decisions.