This is the 142nd installment of The Labyrinthian, a series dedicated to exploring random fields of knowledge in order to give you unordinary theoretical, philosophical, strategic, and/or often rambling guidance on daily fantasy sports. Consult the introductory piece to the series for further explanation.
A couple of weeks ago I wrote a piece about how to build dominant FantasyLabs trends. Nothing’s official yet, but I’ve been told that I’m likely to win the Pulitzer, so that’s exciting.
Anyway, this piece is a short follow-up on that one.
Straight Copy/Paste, Homie
Per usual, I’m about to quote myself. Here’s an excerpt from my Pulitzer-winning article:
How NOT to Build a Trend
Let’s say that on a team implied for 4.5 runs there’s a No. 6 hitter who costs $3,500 on DraftKings and has a high recent batted ball distance of 250 feet. How do you build this trend? You go to the Trends tool, you see that on average the 128,779 batters in our database have a -0.01 DraftKings Plus/Minus, and you say, “This guy is going to crush that number.” And then you start adding filters:
- Runs: You set 4.5 runs as a minimum with an uncapped ceiling. The Plus/Minus jumps to +0.45.
- Lineup Order: You select batters hitting first through sixth. Now the Plus/Minus is +0.73.
- Salary: You set $3,500 as the maximum with no minimum floor. The Plus/Minus climbs to +1.18.
- Recent Batted Ball Distance: You set 250 feet as a minimum with an uncapped ceiling. This is the money shot.
Once you regain consciousness, you check the numbers again just to make sure the trend is actually correct. Yep, it is:
And then you squeeze this player into as many rosters as you can with our Lineup Builder — because a player with a 29 percent Upside Rating must be used a lot, right? — and then . . . you wonder why you’re not getting the results you want.
There are two main reasons:
- You’re assuming that players who match for such trends will hit or exceed the average. You’re not taking into account the certainty that within the cohorts of past players are many performances that fall short of salary-based expectations.
- You’re building unrepresentative threshold trends. The cohorts of past players aren’t highly comparable to the players around whom you’re constructing these trends in the first place.
Basically, when you build these trends you’re engaging in intellectual dishonesty. You’re asleep and dreaming on a bed of bullsh*t.
I want to think more about the dangers of threshold trends in this piece.
What is a threshold trend? It’s one that is similar to the trend above. Instead of using ranges (an implied run total of 4.0 to 5.0 runs, for instance), a threshold trend uses boundless minimums and maximums. As long as a player is above or below a certain number for a particular criterion, he satisfies the requirements for inclusion.
The primary argument for threshold trends is that, even if they’re not highly representative, they nevertheless provide a sense of possibility. They show the potential a player possesses if he has an outlier performance. They’re useful for identifying prospective Black Swans. In this perspective, it doesn’t matter much if threshold trends paint portraits characterized by accuracy and precision. What matters is that, like Impressionist or Cubist artists, they present something similar enough to the subject in order to be recognized as a representation. Per this logic, the matching cohorts of threshold trends don’t need to be highly comparable to the subjects who inspire the trends. Rather, they just need to be similar enough.
In theory, this logic is fine. In actuality, the logic (I believe) probably leads DFS players astray in two primary ways:
- With threshold trends, it’s easy to overestimate the ceiling a player has.
- With threshold trends, it’s easy to overestimate a player’s odds of hitting his ceiling.
Even when DFS players know the trends they’re using are skewed, they still act as if they’re perfect — as if Pablo Picasso’s representations of people are in fact how people actually look. Or, more precisely, as if any given person could inspire a Picasso painting.
The Tails of Thresholds
When you create a threshold trend around (for instance) a No. 6 hitter with an implied team total of 4.5 runs, salary of $3,500, and recent batted ball distance of 250 feet, the cohort of matching players in the aggregate looks not too similar to the batter who inspires the trend.
This is hypothetical but probably fairly accurate: Let’s say that the cohort of comparable players on average looks like a No. 4 hitter with an implied team total of 5.1 runs, salary of $2,900, and recent batted ball distance of 276 feet. Realistically, how similar do you think that player is to one who bats lower in the order, is on a team implied to score fewer runs, costs more, and hits the ball not as far?
Because threshold trends are unbounded, a substantial number of matching players have outlier tails in at least one of the criteria — and those tails can often be the source of elevated Upside Ratings. Because of how these trends are created (with the target players’ metrical limits serving as the thresholds) the subjects of research cannot be similar to the cohort players with outlier criteria tails. Basically, tournament upside lies on the periphery, and the target players who inspire threshold trends are always located at the thresholds and thus are as far away from the positive periphery as any matching player can be.
As a result, the target players of threshold trends usually lack the total upside of the cohorts who match for the trends.
In most cases, don’t use the Labs Tools to build threshold trends. Even if you know they’re unrepresentative, you’re still at risk of acting as if they’re more representative than they are.
And I just read that Soundgarden’s Chris Cornell is dead. Without him Pearl Jam might never have existed and the Seattle music scene would’ve been entirely different. Without Soundgarden, it’s possible that Nirvana, Alice in Chains, and Pearl Jam would’ve been bands unknown to anyone outside of the Pacific Northwest. Like a great leadoff hitter in an MLB lineup, Soundgarden made the bands around them better and likelier to have success, and Cornell did the same with his bandmates.
Cornell was the American Robert Plant who didn’t need a guitarist to help him write songs. Of all the Seattle singers, he was the best of the breed — and he was the one who survived the ’90s. Every song he wrote after Soundgarden first broke up in 1997 has basically been a gift to the world. We’ve been playing with house money for the last 20 years.
Cornell was the Picasso of grunge.
So long, Black Hole Sun.
The Labyrinthian: 2017.47, 142