Through the early stages of the 2019 season, I’ve become increasingly interested in DraftKings “Showdown” contests. For those unfamiliar, these are contests that involve a player pool from only a single game, and have no positional restrictions like the classic DK roster format. I feel that there are a couple factors that might make the Showdown contests ones we want to attack as part of our contest portfolio.
Because the Showdown slates are a newer format, there is reason to believe there might be less pricing efficiency (which is good news for sharp players).
All primetime games have Showdown contests - there is reason to believe that the contestant pool might not be as sharp for primetime games, which tend to attract more casual/recreational players.
The format is new and flexible - I feel that we haven’t reached a consensus on what positions and strategies are optimal for Showdown play, and the positional flexibility of the format contributes to an even less clear optimal approach.
With the last point on my mind, I set out to do some statistical research into what positional strategies tend to produce winning lineups.
The data: We didn’t have access to a ton of historical data, I was only able to get contest data from Weeks 2 & 3 primetime Showdown contests. This article focuses on Showdown cash games, particularly the single entry $25 double up.
The methodology: We want to treat cash games as contests where we can achieve one of two discrete outcomes: win or lose. It doesn’t matter how much or little we win or lose by, the outcome of our cash contests are binary. As such, a logistic regression is the most sensible approach for modeling factors that contribute to winning cash lineups.
Below are the coefficients of a logistic regression model that regresses cash success/failure on a number of variables that describe every contestants’ lineup. I’m not going to go into details on how to interpret logistic regression coefficients, if it is something you’re interested in I suggest doing some internet research on interpreting logistic regression coefficients, interaction variables, and the method for converting logistic coefficients to odds and probabilities.
I’ll do my best to summarize what I’m seeing in the quantitative results from above.
In neutral game environment (in regards to spread and game total, assuming 45 points is a neutral total), it seems like even exposure to teams is desirable, perhaps even skewing by one player to the underdog. Increasingly favorite-heavy lineups correlated with reduced win probability (“Stack-5-1” to “Stack-2-4”), however, there is strong positive correlation between the number of favorite WR/TEs stacked with the favorite QB.
TE is the only position with strong positive correlation between win probability and positional captainship. This was striking to me. Further data will be needed to more confidently validate this finding, but it is an interesting finding to consider when building your Showdown lineup. All other positions have positive but insignificant win probability correlation. This means that they are directionally better captain plays than D/ST, which doesn’t appear in the coefficients because it is assumed to be the factor baseline.
After TE, QB and WR appear to be closest to significance. My conclusion here is that in general it is desirable to seek out captains who gain fantasy value through the air, either as pass-throwers as pass-catchers.
Captaining the kicker is non-significant, but it’s coefficient is negative, suggesting that captaining kicker is even worse than captaining D/ST. Long story short: it may seem tricky or cute, but don’t captain a kicker.
In neutral game environments, it actually can be advantageous to captain the underdog.
In neutral game environments, the increasing numbers of QBs, Ks, WRs, and TEs in your lineup tend to produce higher win probabilities (in that order). It seems like neutral game environments don’t help running backs or defenses help you win.
As spread increases, we typically want to increase our exposure to the favored team.
Increasing spreads make RBs more desirable as captains, QBs less desirable captains. However, we’re talking about pretty big spreads to reach a tipping point - something approaching double digits.
As the spread increases, we want to consider captaining a player from the favored team more seriously. This is a pretty responsive trend, where the tipping point occurs at pretty pedestrian spreads.
As spread increases, we want to consider rostering fewer QBs and more running backs and defenses. This isn’t a very sensitive variable, again we’re talking about spreads near or above double digits.
A pretty noteworthy trend is that with rising game totals, we want to more seriously consider captaining skill positions, avoiding captaining QB or kicker or defense. The notion of captaining the favored team becomes increasingly desireable.
As game total rises, having some exposure to an underdog stack becomes increasingly viable.
It’s worth noting that none of these findings should be deployed in a vacuum. As is the case with DFS roster construction, we must consider the tradeoff from one approach to another. It is also worth noting that this analysis doesn’t considers players as, well, players. Obviously some running backs function more heavily in the passing game. Some tight ends don’t function in the pass game at all, we shouldn’t be captaining these TEs because the model says “let’s captain a tight end.”
I wanted to walk through my most recent Showdown cash lineup (which actually ended up a few DKP under the cash line) and how I used some of these model results to inform my construction. At the end of the day, these are results produced from a small sample, and the fact that they don’t consider players and their relative prices and values does leave it open to criticism. So I’ll leave it up to you as to how heavily you want to incorporate this approach into your Showdown lineup construction.
With a spread of 4 favoring the Packers, I knew I wanted to play either 4 or 5 Packers vs. 2 or 1 Eagles. And given an above average point total (and I think potential to go considerably over), I knew I didn’t want to captain Rodgers, but rather wanted to captain one of his pass catchers while still rostering Rodgers. I toyed with the idea of captaining Mercedes Lewis for the sake of sticking to the “captain the tight end results”. I moved off of that idea pretty quickly because, well, when you have viable skill players like Adams, Jones, and MVS, captaining the backup tight end seems silly. However, if you had captained other tight ends Jimmy Graham (the thought never crossed my mind) or Dallas Goedert (the thought briefly crossed my mind), well, you were in pretty good shape.
After moving away from Mercedes Lewis, I knew I would want to captain a GB pass-catcher. My captain decision really came down to MVS vs. Davante Adams. I chose MVS over Adams for the following reasons.
In a game that didn’t have a massive spread but not a completely neutral script expectation, and a relatively high point total, I was in favor of trying to avoid rostering defenses or kickers. As our model suggests, increasing game totals typically favor skill players first, and pass-throwers second. And for defenses to become viable, we would need a pretty skewed game script.
As totals increase, it is increasingly valuable to have some exposure to an underdog stack.
So I know I want a GB receiver at captain. I know I want to pair him with Rodgers. I know I won’t be seeking savings from a D/ST or K. And I think I would also like to play a secondary stack with Wentz and a Philly pass-catcher.
In order to fit in Wentz, Rodgers, and Adams without dumpster diving at a skill position or rostering a kicker of defense - I felt compelled to play MVS. I didn’t hate this decision, I think per-$ he and Adams couldn’t be projected that differently.
While I couldn’t captain a tight end, I wanted to roster one, and given I was committed to Adams and MVS, I felt it had to be on the PHI side. Enter Dallas Goedert who I thought was priced very affordably - happy that worked out.
My last roster spot to fill was some GB skill player around $5K. I could’ve gone with Geronimo Allison, but I did feel that given a reasonable spread of GB -4, having some exposure at GB running back could be desirable, as running backs trend up in high-spread games. Between this trend and what I thought was a really good price, Jamaal Williams presented himself as the ideal play. I felt the play was good, avoiding the injury bug is tough to do with much certainty, hope Williams is back on the field sooner rather than later.
So that’s about it. That’s where my head is at with these Showdown games at the moment. I plan to continuously update this model, and adjust my strategy accordingly and will see what can be done to productionalize this or a similar model so that ASA subscribers who are less familiar with logistic regression can still use the quantitative power that is behind said model. We’ve got another fun Showdown slate coming up this Sunday - Dallas vs. New Orleans - hope everyone crushes out there.