Succeeding on DraftKings is about identifying which players are going to perform the best over the course of just a single game. For the most part, you don’t need to worry about injuries or how a player is going to produce down the line; you just need to be on point for one day.

Many players have taken that to mean that all that counts is the short-term; “What does it matter how all home teams have performed coming off of a bye?” some might question. “I just need to predict how this player will perform on this home team in this particular game.”

I’ve taken a different approach to daily fantasy sports in that I really do concentrate on long-term trends. I care how backup running backs perform as a whole and which position has historically been the best in the flex. Yes, every decision is unique, but I think there are really good reasons to emphasize long-term trends in daily fantasy sports.

So why does it make sense to study aggregate data over long periods of time when analyzing just a single player in an individual game? Here are four reasons.

 

Variance

 

A huge part of winning at daily fantasy sports, maybe even the most significant part, is understanding when results are just short-term variance (noise) and when there’s really something there (a signal).

Well, in many sports, there’s a whole lot of short-term variance that humans aren’t built to properly recognize. It’s really, really difficult to know when a trend is “really” a trend and when it’s just dumb luck.

Take injury-proneness, for example. Some players might be more injury-prone than others, which we’d expect to be the case, but it’s extremely difficult to identify which players are injury-prone in a way that’s predictive. Maybe we can do it after the fact, but otherwise it’s just extremely challenging to separate past injuries from randomness.

In the case of injuries, we’re better off going with the long-term data. I’ve found that running backs who are tall with a low body-mass index are significantly more likely to get injured than short, thick running backs, for example.

We see similar phenomena across all of sports, from “hot streaks” to matchup strength. We can assess the past with accuracy, but usually not in a way that aids in making better predictions.

 

Uncertainty

 

Built into the notion of variance is uncertainty. I think the biggest mistake that daily fantasy players make is failing to account for their own fallibility. Sometimes, you’re just going to be wrong. If you don’t factor that uncertainty into your projections and lineups, you’re going to leave money on the table.

Think about the flip of a coin. There’s a whole lot of uncertainty about the results of an individual flip. Because of that uncertainty, you’re best off going with the long-term rates.

Now, if you had a significant enough sample size of past coin flips that you could conclude that the coin is biased in some way, then maybe we’d have something. But if you’re trying to use short-term past results of something as random as coin-flipping to predict future results, you’re going to be in trouble.

On DraftKings, I’ve found that tight ends have historically been really smart flex plays for GPPs, but not-so-great flex options in cash games. If you’re going to use a tight end in the flex in a head-to-head league, you need to establish a higher level of confidence in that pick since the data suggests tight ends as a whole are too volatile to use as flex plays in cash games over the long run.

It’s not that you can’t ever do it, but rather that you need to account for uncertainty in your projections. If you have a running back rated just behind a tight end as the top value, it might make sense to side with the running back as your flex play, a move that you wouldn’t make if you ignored long-term trends and treated your projections as flawless.

 

Baseline

 

In the coin flip example, we have a baseline with which to work; we know heads and tails come up 50 percent of the time over the long run.

When we analyze long-term trends, we’re trying to establish that baseline so that we can make better short-term decisions. We can use Bayesian inference, for example, to estimate the probability of a certain event occurring as we gain more and more new information, but that’s not possible without some sort of initial baseline with which to work.

What’s an example of a baseline? In the intro, I said I care how backup running backs as a whole perform once they take over as the starter, for example. Well here’s a look at fantasy points for starting running backs versus backups (who become starters due to injury) from 2009 to 2013.

Backup RB 4

 

The backups have outplayed the original starters! There are a lot of reasons for this, most notably that running backs are very dependent on their teammates for production and that NFL teams are generally pretty bad at assessing running back talent, but it’s certainly useful information to know.

I think the data shows that there’s good reason to start with the assumption that a backup will produce the same as the original RB1 once he moves into the starting lineup. Then we can try to refine that prediction by asking relevant questions. Is there good reason to believe the backup is much worse than the starter? Will the backup see the same workload that the starter did?

We can’t ask these questions without an original baseline with which to work, and it would be foolish to use a handful of recent numbers, stats that are likely misleading, as the baseline.

 

Equity

 

All of these reasons for emphasizing long-term data add up to what I believe is the most important aspect of an “evergreen” approach to daily fantasy sports research: it helps build what I call intellectual equity.

That is, when I perform a study on backup running backs or the optimal flex position, that information has a much longer shelf-life for me than trying to figure out how Joe Haden will perform on A.J. Green. The latter type of analysis is certainly useful and it shouldn’t be replaced, but it becomes irrelevant almost immediately.

Think about what you learned in college or high school. How much of the trivial shit do you still know? Any idea when Napoleon stormed the Bastille? How about the length of the Mississippi River? Could you point to Belarus on a map? Did you even know Belarus is a country?

The reason that college is valuable (for some) isn’t because of the insignificant stuff you learn, but rather because you learn how to learn. The ability to problem-solve and rationalize is far more valuable than knowing a stupid fact.

In a sense, a scientific approach to daily fantasy sports that emphasizes overarching philosophies and data is very much akin to buying a house. When you buy a house, you build up equity in it. This concept of studying long-term trends is similar in that you’re building equity in the form of process-oriented knowledge that you can call upon at a later date.

A “Week 10 QB Plays” article (which I wrote this week, by the way) is like paying rent in that you’re trading in something of value to you (your time, in this case) for something that has immediate use, but no long-term value. That’s not to say that it’s bad, since sometimes renting is very necessary and preferable to buying, but it’s not going to help you with the timeless, big-picture concepts that create the foundation of true understanding.

The point isn’t to forgo reading time-sensitive content, but rather to be aware that the path to dominance is through “building equity” in developing your process.

Focus on understanding the why over the what.