MFL10 Optimization Part 2: Solving the Onesie Positions
In part one of this project, we followed the evolution of a fictional 2016 MFL10 draft from an inefficient mess into a league full of optimized rosters. That first step provided us with some general guidelines for roster construction. Now we dig a bit deeper, in search of more detailed and more actionable conclusions. We’ve already learned, for example, that we want to roster either two or three quarterbacks. But when is two better than three? The following analysis addresses these questions for the “onesie” positions: Quarterback, Tight End and Defense.
For these positional breakdowns, I have expanded the scope and depth of our optimization data. While in the initial analysis, observations were based on the results of 200 “rounds” of optimization and used only 2016 ADP and player data; going forward we will refer to the results of three more thoroughly optimized drafts. Below are the positional allocations for optimized drafts using 2014, 2015 and 2016 data, with each having gone through 1,000 rounds of trades.
These tables, and the picks underlying them, provide us with a roster construction roadmap. Let’s see what they tell us about each position.
There are 36 teams between the three optimized drafts in our sample, and all of them landed on either two or three QBs for their team’s optimal allocation. Here are the splits by year and overall:
Overall, two QBs is a bit more popular than three, but that varies year to year, and there is not one clear choice. So how do we know when we need to use that valuable roster spot on a third QB? The answer lies in the strength of our first two picks at the position. The tables below show the ranks of each team’s QB, grouped by whether they took two or three.
Wouldn’t it be nice if there were just some rule that every team were following to decide whether two or three was the best choice at QB? It’s never that simple, BUT with a little squinting, I can get us pretty close. The few exceptions make it a little more difficult to see, but based only on those tables, we can define a QB “rule” that 33 out of the 36 teams implicitly followed in building their perfect rosters. If a strategy is good enough for 92 percent of all-knowing robot drafters, it’s good enough for me.
Quarterback Decision Tree:
Easy, right!? Well, no. It’s not actually as easy as checking a couple of boxes, because unlike our robot test subjects, we are not all-knowing. The highest-scoring QBs are valuable – my fake teams fought over them. But, nobody knows for sure who this year’s top-six or top-16 QB will be. It would not take you long to find a handful of articles on this very site that explain how bad we humans are at projecting year-end results.
That said, I believe this simple decision tree is a great guide. My advice is gauge how good you feel about your own rankings before deciding you are strong enough to stick with two QB. If you are not willing to optimize (i.e. draft only two QBs) on the assumption that you have a top-six producer in a guy like Aaron Rodgers, Andrew Luck or Tom Brady, I suggest you simply avoid drafting those high-cost players entirely. Knowing we probably need both players to finish in the top sixteen, two-QB is a dangerous play for the late-round QB approach. When you get to those mid-teen rounds, ask yourself the questions above, and if you don’t land on a confident “YES,” it’s probably time to grab that third QB.
There is greater dispersion among the optimized TE allocations than I expected when I started this exercise. Rosters ranged from one to five TEs across the three drafts. You’ll find the distribution below.
Half of the final rosters held two TEs, with most of the remaining teams going with either three or four. Let’s address the outliers first, as I will not recommend drafting only one TE or going with five, despite both options sneaking onto our board.
The team that stopped at one TE from the 2014 draft used its first-round pick on Rob Gronkowski. Gronkowski played every game that year and produced 266 fantasy points, making it his best season since 2011. Does this mean you should go with just one TE if you expect a healthy Gronk? My short answer is no, having personally traveled halfway down that analytical rabbit hole, but that’s a story for another time. If you require some evidence, note that the team with Gronkowski in 2015, when Gronk scored nearly as many points (256) as he did in 2014, went with two TEs. It’s not a worthwhile risk.
The team with five TEs was able to piece together a complete season from late round players who missed several games, or otherwise only produced one or two useful scores. Given 1,000 tries, that team made the jigsaw puzzle fit – it’s not something you or I would have any chance of achieving on purpose. I didn’t want to ignore these three results completely, but having given them a look, I think it is safe to focus on the juicy center of our distribution.
Two, Three or even Four?
In the first article of this series I stated that “deviating from 2-3 at QB, TE or DEF is almost definitely not a good idea.” While I still feel that way about QB and defense, I can no longer dismiss the idea of four TEs after seeing 25 percent of the new optimized rosters going with four (The More You Know Rainbow dot gif). So we have three viable options – how do we decide?
Begrudgingly, I accept four TEs as an option, but I still see only one critical decision to make: whether to stop at two. Each of the nine four-TE teams did not take their fourth until the final round of the draft. Allocating that final pick to TE, while a net positive, was not a game changing decision for these teams.
Again, I’ve taken the makeup of each team’s TE group and looked for a potential “rule” imbedded in the distribution. The following criteria could have been used by 29 of 36 teams (81 percent) to determine TE allocation.
Tight End Decision Tree:
As was the case with the QB decision tree, this playbook is neither perfect nor iron-clad. It is again dependent on the strength of your convictions. If, for example, you believe there is a clear tier break after the first five TEs, but your rankings get hazy thereafter, then you might want to set your first decision point at TE5 instead of TE6. Either way, picking an early TE and then only ending up with two fits RotoDoc’s two-TE1 strategy.
I remain skeptical that adding a fourth TE is a winning move. I might consider a fourth TE more seriously under these circumstances: it’s the final round of the draft, I think it’s unlikely that I have a top-10 TE, and I have not used six picks between QB and defense, but feel good about my strength at both positions.
The QB and TE strategies described above depend on having some understanding of a top tier at the position. Applying that methodology gets tricky with defense, where ADP has been a very poor predictor of final rank relative to QB and TE. The things in which I am most confident with respect to drafting defenses are that we should draft between 2-3 of them, and we should save those picks for the last few rounds of the draft. With all of this data on hand, though, we must be able to do better than that. Here is the distribution of defense allocations across our three optimized drafts:
I’ll note that the one team that landed four defenses also happened to be the lowest scoring in its draft, but otherwise I am going to focus entirely on the question of two vs. three. The optimization favored three defenses by a ratio of just over 3:2, and exhibited almost no relationship between the rank of a team’s first defense and the number of defenses that ended up on its roster. So what drives some teams to go with two instead of three?
Part of the problem with understanding defenses is structural – there are only 32 NFL teams, which means it is impossible for every drafter to roster three defenses. The fact that all 32 defenses get drafted in each simulation suggests to me that we might see more three-defense teams if it were possible. I have made the argument for three defenses in the past, and these results do not hurt that cause.
My advice continues to be to target three defenses in the last three or four rounds of the draft, with the exception being when you have already locked yourself into taking both three QB and three TEs – try to keep your total allocation to the onsies between 7-8.
As I’ve continued to work on the roster construction problem, I have grown more confident that there is not one optimal positional allocation. Roster optimization is a fluid process, and our goal is to go into each draft with a map containing as many routes to success as possible. As you progress through your drafts, remember to ask yourself the questions above, and put yourself in the best position to win. Decisions may get a little more complicated in the next installment, when we tackle running back and wide receiver. Until then, good luck!