Over the past four to five years, the general DFS public has latched onto the technique of stacking as a means of manipulating variance, and for good reason. Take a second and read that hook again, paying particular attention to the portions in italics. Why did I highlight “general DFS public” and “manipulating variance”?
We’ve recently seen a bleed over in stacking from DFS to other formats of fantasy sports, like Best Ball and redraft. Are people simply taking a technique that works in one format and applying it to others without understanding why we stack in the first place? My answer is a resounding “yes.”
The remainder of this piece will explore variance, define stacking, analyze why we stack, pit theory versus reality, and formulate conclusions on stacking in Best Ball and redraft.
In statistics, variance is simply the square of the standard deviation or a measure of the distance of individual data points in a data set from the mean. To better understand its application to fantasy football, we must first hold a sound conceptual understanding of ranges of outcomes. A range of outcomes in statistics defines a maximum and minimum value for a given event and associates probabilities with each individual outcome, with the data set resembling a bell curve. The below graphic should help to visualize this concept.
But the variance we see in sporting events, which involve imperfect humans (the players, the coaches, the referees, the umpires, etc.), spans countless other applications that go far beyond the simple statistical interpretation. Things like injuries, busted coverages, officiating, weather, or even how points are scored on a fantasy roster induce additional variance, or deviations from the mean.
I want to spend some time really hammering down our understanding of variance in fantasy sports because so much of what we do as top-level fantasy players revolves around it. In the NBA, the lowest variance sport, player outcomes are congregated about the mean at a higher rate. Visually, this would create a tall and skinny bell curve. Conversely, in the NFL and MLB, where single actions generate larger amounts of fantasy points (like a touchdown or home run), player outcomes and probability chance of occurrence are much more spread out. Visually, this would create a short and flat bell curve, one with greater outliers. The statistical explanation for this finding can be explained by examining the way in which players score points. In the NBA, a single action during the game can generate a maximum of 3.5 points (a made three-pointer), while players also have more raw opportunities to contribute fantasy points to your roster throughout the game (points, rebounds, steals, blocks, threes, etc.). In the NFL, a single action during the game can generate a maximum of six points (a touchdown), while the deviation in ways to score points and the value that those actions generate is greater.
That entire discussion only explored the act of scoring fantasy points. When also considering the other means of introducing variance into sport (injuries, busted coverages, officiating, weather, etc.), we can begin to understand why we would want to manage (decrease) or leverage (increase) variance depending on multiple inputs like contest size, the number of entries allowed per person, payout tables, the strength of competition, and the sport itself.
A solid percentage of the DFS field utilizes these principles and concepts when building lineups without even knowing it by referencing 70%, 80%, and 90% outcomes in projections. Most fantasy players understand that we want to manage (mitigate or decrease) variance in cash games and confined leagues and leverage (increase) variance in large field GPPs, season-long contests, and Best Ball tournaments. But that is the limit of much of the field’s knowledge and application of variance. We’ll explore some higher-level applications of variance in the coming sections (and the coming articles, podcasts, and courses!).
The definition of a stack changes slightly depending on the sport. Put most simply, a stack involves a grouping of players from the same team. In practice, a stack is a quarterback paired with a minimum of two of his pass-catchers. A quarterback paired with one of his pass-catchers is simply a correlation. A game stack correlation would be a full three-member stack from one team, brought back with a pass-catcher or running back from the other side.
The short answer is we stack as a means of manipulating variance in our favor. We do this for two primary reasons: we either want to manage, or mitigate, variance or we want to amplify, or leverage, variance. In cash games, small field GPPs, and confined leagues, where we have to beat a smaller number of people for a smaller prize, we typically want to manage variance. We get no additional money if we beat second place by 30 points or by 0.01 points. Conversely, in large field contests and large field GPPs, the payouts are extremely top-heavy, and we need to be placing ourselves in a position to win the whole contest when we get things right. To do so, we need to be willing and able to accept a larger amount of variance by leveraging, or increasing, our potential range of outcomes.
In theory, we should be looking to manage variance as much as possible in confined leagues, cash games, and lower-entry GPPs and leverage variance as much as possible in large field contests and large field GPPs. What we see most commonly around the space is people taking this to the extremes, unwilling to accept any semblance of variance in the former and accepting far too much variance in the latter.
And since stacking is simply a means of manipulating variance, it has different applications for varying sports and contests. In a sport like MLB, where variance is inherently high due to the nature of hitting a round object with another round object, stacking is a way to capture chunk points on a team that projects highly on the night. The theory would dictate we simply play eight hitters that all hit home runs, but that is extremely difficult to do in practice as the chance for a single player to hit a home run in an MLB game is around 4-6%. That would mean the chances of picking eight individuals that hit a home run on the same slate are astronomically poor.
In the NFL, we stack for similar but conceptually different reasons – reasons that vary based on format and contest structure. For DFS and large field GPPs, we want to capitalize on one team outpacing projections, ideally on the top-scoring team on the slate. We’re looking to capture touchdowns and yardage outputs from the major players on that team, which compounds by stacking the quarterback with two of his pass-catchers. Side note to that – full stacks are overutilized in NFL DFS compared to the percentage chance of success, while correlation plays (a quarterback with one pass-catcher or wide receivers from opposing teams or a wide receiver from a large underdog and the running back from the favorite) are underutilized. I digress.
But in Best Ball, we’re now faced with a format that defines a weekly game played over and over again with the same roster. Let’s take the most extreme example in recent memory of stacking and see if we can’t apply it to new, unfound, and underutilized methodologies. I speak of course of Kirk Cousins, Adam Thielen, and Justin Jefferson. Together, a stack of Adam Thielen and Justin Jefferson provided 11 WR1 weeks, but there were only two weeks where both scored as WR2s or better and only one week where both scored as WR1s. Kirk Cousins returned eight QB1 weeks, extremely solid for his ADP. And to answer the question before it is asked, yes, those two were the top pass-catching pair of 2020. For comparison, Calvin Ridley and Julio Jones combined for seven WR1 weeks. All Dallas wide receivers combined for seven WR1 weeks and two weeks where two scored as WR1s, with two of those scored by Michael Gallup and one by Cedric Wilson. Even the 2020 stack darlings, the Los Angeles Rams, combined for a grand total of four WR1 weeks, with three of those coming with both Cooper Kupp and Robert Woods finishing as WR1s.
The point we are making with this exercise has to do with variance – both managing it and leveraging it. Compare those scoring outputs to the extreme example of the nuts first/second and second/third-round wide receiver pairing in 2020: Davante Adams and Calvin Ridley. Together the two combined for 12 WR1 weeks, four top overall WR scores, and three weeks where both scored as a WR1. Now, there are many more factors to consider, primarily ADP, but I’m trying to make this example as clean as possible. By the numbers, stacking actually reduces the overall upside of a roster in Best Ball, a format that places a high level of emphasis on spike week potential and consistency.
Let’s take one more example into account. The Dallas Cowboys were widely regarded as a top-two stack heading into 2020. But what happened when one of those outside variance contributors we talked about earlier reared its ugly head (Dak Prescott injury)? Yea, the offense was never the same. What about injuries to the offensive line (also Dallas, Philadelphia, Los Angeles Chargers are prime examples from 2020) or injuries to a defense that influence offensive pass volume (Chargers and New York Jets from 2020) or injuries to a backfield that affect offensive pass volume?
So, how do we stack smartly to maximize our upside without overexposing ourselves to one variance-driven act? Two of my favorite methods of stacking are:
1. Stack bad offenses (with low cost of acquisition, or ADP)
2. Stack teams’ wide receiver three and four, both as a means of generating consistent WR2 production.
This method resolves the additional risk of overexposure while not sacrificing upside. In the first example, the low opportunity cost of stacking bad offenses means anything more than six WR2 weeks from late picks from the same team is well worth the cost of admission. Teams that fit this criterion last season were the Miami Dolphins, the New York Jets, and the Jacksonville Jaguars. In the second example, stacking the WR3 and WR4 from high-powered offenses creates the possibility of four to six spike weeks in a fully healthy year while also creating the opportunity for extreme upside should the WR1 or WR2 on the team get injured. Again, I arrived at these conclusions through my understanding of variance and how to manage and leverage it (so important!). Now put both those ideas together. Last year, if you left a draft with Davante Adams, Calvin Ridley, and a Jets team correlation of Jamison Crowder and Braxton Berrios, you captured the maximum upside from your WR1 and WR2, while gaining an additional four weeks of WR1 scoring and four additional weeks of WR2 scoring from the Jets pair to fill a WR3 spot on your roster. Take that one step further and stack two bottom offenses with four picks in the last half of the draft and your roster now holds two WR1s and two WR2s from a combined scoring perspective. Don’t fall into the trap of weird roster constructions like “zero-RB, modified zero-RB, and robust WR.” Let the draft fall to you as it may and utilize the stacking and correlation tools and tricks we’ve discussed. You will be left with a much higher ROI at the end of the season, I can all but promise you! For more discussion on the subject, check out the recently released podcast on stacking (Best Ball Pod :: Part 2 w/Hilow & Pawel).
JM + Hilow Pods (Coming soon!!!)
4-Part Best Ball Pod Series with Pawel (Now Available on Podcast Networks!)
Best Ball Game Theory ($10 off with Promo Code: save10)
The Theory (Game, and Otherwise) of Roster Construction (!!!)
NFL Edge + DFS Interpretations(!)
Exploring Extremes Pod with Pawel(!)
Hilow and Xandamere Strategy Pod(!)