In 2013, the Buffalo Bills ran the ball 545 times, more often than any team in the league, and had one of the league’s lowest pass rates at just 51.2 percent, despite the fact that they trailed late in many contests and were forced to throw the ball to catch up. In contrast, the Peyton-Manning-led Denver Broncos, despite leading most games late, aired it out on 60.2 percent of their offensive plays.
The difference in pass rate between the two clubs was partially due to the skill sets of their personnel, of course, but the numbers are also representative of what the two clubs wanted to accomplish on offense and how they felt about their chances of winning.
If we polled NFL fans and asked them if the goal of an offense is to score as many points as possible, my hunch is that almost all of them would say yes. As we saw with the Bills in 2013, though, the goal isn’t really to score as many points as possible. If it were, Buffalo would have passed the ball on nearly every play, just as everyone else would, because they’d be able to run more plays, rack up greater per-play efficiency, stop the clock on incompletions, and ultimately score more points.
Nope, the goal isn’t to score as many points as possible, but rather to do everything possible to create the biggest positive differential between your points and your opponent’s points, i.e. win the damn game. The Bills, a team that knew they were underdogs in most games, would be in trouble if they came out throwing against a team like Denver, even though tossing the ball on every play would likely maximize their projected points.
Even a team like the Broncos isn’t concerned with point-maximization in many scenarios. There’s a reason teams run the clock down at the end of games; it increases win probability without improving the odds of putting more points on the board.
Another quick example. I was playing MarioKart the other day (by ‘other day,’ I actually mean like, no lie, probably 15 years ago) and I was racing a friend to the finish line. I had a choice to drive straight to the finish line (and thus “optimize” my finishing time) or steer slightly to the right to grab a question mark thing. I chose the latter because I figured it would give me a weapon that would allow me to overcome the time lost from going to grab it; in effect, I traded in a bit of time (equivalent to projected fantasy points) in order to increase my edge over my opponent (improving my win probability).
Maximizing Projected Fantasy Points
I bring this up because I think sometimes daily fantasy players equate maximizing their chances of winning to always, in every scenario, “optimizing” a lineup to score as many points as possible. I don’t think that’s the case at all. Let me give you an example to display why.
I had a lot of success playing in daily fantasy baseball tournaments this year by purposely selecting “sub-optimal” (in a vacuum) lineups that I knew wouldn’t maximize my projected points. That might seem ridiculous, but I truly didn’t want to maximize my points in GPPs.
When I qualified for the 2014 DraftKings Fantasy Baseball Championship in the Bahamas, for example, I did so with a relatively low score of 172 points. That’s certainly above-average for a typical league, but well below-average for a winning GPP score.
I won that qualifier by stacking a potent offense (the Oakland A’s) against a really good pitcher in Madison Bumgarner. Even though that didn’t optimize my projected points in a vacuum, I was relatively confident that few users would be stacking Oakland that night. Loaded with power bats, I was willing to take a chance that Oakland would get hot and/or Bumgarner would have an off night.
In this scenario, I was willing to trade in some projected points (the Rockies at home probably would have been the best bet for long-term scoring) for lower usage.
This really all comes down to usage rates. When a player isn’t in a lot of lineups in a GPP, it’s really advantageous for you if you use him and he has a big day. I’m going to give you an extreme example to show why this is the case.
Let’s suppose that you’re deciding between two players: Player A and Player B. You have Player A projected at 20.0 fantasy points and Player B right behind him at 19.5 fantasy points. Now let’s pretend that we have perfect knowledge of each player’s usage in a certain GPP, and we know that every other user is going to roster Player A. Literally every single user.
If you decide to “optimize” your projection and roster Player A, he would have zero usable value to you; he could score 100 points or he could score 1 point and it just wouldn’t matter because everyone would be even. You couldn’t possibly gain an edge.
Meanwhile, Player B isn’t in any other rosters. Even though he’s the “worse” choice in a vacuum, you’d need him to outplay just a single player (Player A) to have an immediate edge over the entire field. And what are the chances that Player B (projected 0.5 points behind Player A) outscores the more popular player? Pretty good, right? Maybe 45 percent? So if you roster Player A, you have no edge over the field. If you roster Player B, however, you have a 45 percent chance to gain an edge over every other user.
Usage is of course never this extreme, but the idea is clear: the higher a player’s usage, the less usable value he’ll offer you. That doesn’t mean you shouldn’t ever roster high-usage players, but just that the value of hitting on one is much lower than nailing a low-usage pick.
Projected Points in GPPs
Now you can start to see this idea come together. The goal in tournaments isn’t to maximize projected points, but rather to maximize your win probability. Your goal in cash games is to maximize your win probability, too, but that generally goes hand-in-hand with maximizing your projected points.
One of the keys here is understanding not only the probability of your decisions leading to great results on the field, but also the probability of that decision helping you win. When I selected the A’s over the Rockies, for example, I traded in a small percentage in terms of the probability of my offense reaching a certain threshold of points in exchange for a higher probability of a beneficial outcome if my decision worked out. Because so few people had the A’s offense, I didn’t have to compete with a huge number of relevant lineups once Oakland went off, which in turn gave me more usable value than if I rostered the Rockies, a more popular offense, and they did well.
Basically, using low-usage players or stacks increases the number of “outs” you have in a tournament. In poker, players who are down always want “outs,” or ways they can win a hand if they know they’re down.
The same is true in GPPs. Have you ever been near the top of a field in a tournament and think you have a shot to win, only to realize someone ahead of you has the same player(s) left to play as you? It makes it mathematically impossible to catch up (unless you utilize DraftKings’ late-swap feature).
By going against the grain and emphasizing low usage whenever possible, you can avoid those situations as much as possible. In short, if you use a chalk lineup, you basically need to be perfect with your picks because you’ll be competing with so many other comparable lineups. If you go against the grain, your projected points and even your probability of hitting a really high ceiling will decline, but your odds of winning can increase because, if things don’t go as planned for other users, you’ll be in a position to benefit.
That last sentence is key, particularly the phrase “if things don’t go as planned.” Basically, we’re looking to create GPP lineups that are antifragile, i.e. they benefit from chaos. A contrarian lineup is one that’s built to take advantage of things going wrong for other users.
When you fade the Rockies when they’re playing at Coors or bypass Peyton Manning in a perceived easy matchup against the Bears, you’re basically saying, “Okay, this might not be optimal in a vacuum, but if things don’t work out as planned, I’m going to be in the best possible position.”
Note that antifragile lineups are inherently more volatile than others. Because you benefit so immensely when the crowd is wrong but also get hurt so badly when the crowd is correct (since you have an entire pool of users that will pass you), results tend to be extreme with a contrarian approach; Ricky Bobby would be proud, because you’re generally either first or last (or close to the extremes, anyway). I tested my MLB tournament results against a normal distribution and I indeed finished very close to the top or very close to last in way more leagues that you’d expect from chance alone. That’s a good thing in GPPs.
Also note that this antifragile, contrarian, low-usage approach is one that is smart only in certain league types, namely those that allow for a large edge over the field if the popular players don’t perform well, i.e. tournaments. An against-the-grain approach with low-usage players is not a sharp move for cash games; you can’t really benefit much from the crowd being wrong on a player or two when “the crowd” is just a single opponent, for example. In cash games, your goal really should be maximizing projected points, which will generally come by playing the best combination of values.
None of this is to say that you should always fade players you think will be high-usage in GPPs. The popular guys are popular for a reason; they’re generally really good values, i.e. they’ll score a high number of points relative to their cost. The key is understanding when and how to fade popular plays, which is just as much art as science.
Regardless of your approach, keep in mind that the goal when playing in tournaments is not to maximize points, but to maximize win probability, and those are two very distinct things.