Our Game Level Similarity Projection (GLSP) Apps are weekly projection apps that rely on historical matchups in order to create a projection for the upcoming one.
In fantasy football it’s common to hear start/sit analysis phrased in the following way: In Adrian Peterson’s last four matchups against the Lions he’s averaged 100 yards rushing and a touchdown.
That’s great, but it’s tough to make a projection based on just four matchups. But GLSP makes a similar projection by just extending that analysis to include other running backs similar to AP, and other defenses similar to the Lions. It looks for the 20 most similar recent matchups and then creates a low, median, and high projection.
The matchups are found by taking the player’s average stat line over the games that you specify. Those averages are crucial for the way that similar players are found. Let’s take a look at an actual projection to see how they work. In Week 1, Russell Wilson will face off against the Green Bay defense. GLSP will use Russell Wilson’s stats, Green Bay’s stats, and the game’s point spread and over/under to find 20 similar matchups.
Here are Wilson’s stats over his last 16 games.
GLSP then finds game logs from other QBs similar to Wilson and other defenses similar to GB. Here are the 20 contributing matchups.
You can see that essentially all of the other QBs are running QBs or involved in read option offenses. The defenses were all similar to Green Bay’s 2013 defense. In fact Green Bay shows up twice in the list of similar matchups. If you think about it, this is an extremely intuitive way to look at the issue. In fact if you were researching the matchup you might look at how GB had played similar QBs, and how Wilson had played similar defenses. GLSP just automates the search for you.
GLSP then offers a range of potential outcomes based on those matchups.
The App gives you a few different ways to test the reasonableness of the projection by modifying the games that go into the player’s averages. For instance, if a player has seen a recent bump in usage, like Keenan Allen did in 2013 after Week 4, then you might want to throw out the games that occurred before that bump in usage as they will make Allen seem like a lower production player. Or, if a QB leaves a game after attempting just one pass, you may not want that game included in their averages.
When you modify the contributing games, the primary change you’re making is to the kind of player that the similarity search finds. For instance, if you throw out games where a certain number of targets weren’t compiled, you’re not specifying a minimum of targets for the upcoming week. You’re just modifying the kind of player that is unearthed in the similarity search.
The modifications that you make to the player’s average stat line through selection of the contributing game log should be done in an attempt to get as close as possible to an estimate of the “true” kind of player that the subject player is. To clarify, using the above Allen example, it might not make sense to adjust the game log immediately after the bump in usage. You should probably require more evidence that a player really is a 70 yards per game player.
GLSP is an app that will allow you to gather evidence for purposes of setting lineups. It should help you answer questions like whether you should start your WR3 that has a good matchup compared to your WR2 who has a poor matchup. You can do that by looking at the results of past matchups.
One other note is that when the season starts your best bet is to start players in the order that you drafted them. Defenses change enough season to season that expecting a defense to be average is a reasonable assumption. Only in an instance where you know that a defense was not close to average the prior year, and you have a good reason to expect them to be similar quality this year should you really take matchups into account during Week 1. Otherwise just starting players generally in the order you drafted them makes the most sense. Additionally, as the season goes on we’ll be incorporating more current season data into the defensive stat lines for GLSP.
GLSP is just a lineup setting tool. But like any tool, it matters how you use it. You shouldn’t just look at the output from the app to see what the projection is saying. You should look at the player averages and see if those averages accurately describe the player. If they don’t, then there are ways for you to adjust the averages by selecting the game sample that the averages are computed from. Another way to think about the app is that it’s not correct to say “GLSP says” but it is correct to say “Based on the changes I made, GLSP says” as having a thinking operator for the app is the best way to get to a meaningful projection.