This was originally posted at DGDB&D about
a year two years ago, and it was posted here last year. I have updated and tweaked it (again) for re-posting here. Consider this one Draft Theory III, and we’ll just all hope that it’s more like Toy Story 3 and less like The Godfather Pt. 3.
Classic Definitions of "Reach." There exists a general consensus when discussing the draft that teams must avoid "reaching" for a player. Loosely, I think we can define "reach" in this classic sense as
Selection of X before pick Y1, where Y1 is the upper limit of a player’s likely draft range assigned by ranking all available players against one another. (i.e. The player "should" go between picks 15-25, thus a pick above 15 is a classic reach.)
Obviously, I suppose, within this traditional framework, there is some gray area — there are varying degrees of "reach," such that the team taking our hypothetical 15-25 player at 14 is not near as "incorrect" in their selection as the team taking him at, say, 8.
Implicit in this classic definition of reach is the definition of a "good" pick, defined as:
Selection of X between picks Y1 and Y2, where Y1 and Y2 are the top/bottom limits of the player’s likely draft range, with a selection made after Y2 defined as a "value" pick.
Problems with the Classic Definition. On its face, the classic equation seems valid enough. Given enough rankings, one could conceivably compile them and arrive at an upper and lower limit for each player. The problem lies in the underlying determination of a player’s value. More specifically, the theory assumes some sort of fixed value for each player that allows them to be ranked in a meaningful way; it assumes that the ranking accounts for all (or nearly all) of the variables so that one can accurately compare players of two different positions, from two different teams, who played under myriad different conditions, and come up with a determination of which player is "better."
Such an assumption is inherently false — it has to be — because the data required to accurately compare two players makes your model too narrow to compare many others. This phenomenon is not unique to football, however.
Captain, there seems to be something approaching. I believe...yes...it’s a strained analogy!
There is a theory in particle physics known as the Heisenberg Uncertainty Principle, which states that certain physical properties (position and momentum) of a given particle cannot both have precise measurements at the same time — the more specific you get with one of the measurements, the more general you must be with the other, because defining one necessarily excludes, limits, or ignores the other. While not perfectly analogous, I think you can apply a similar idea to the ranking or relative draft value of players — the more you try to pin down exactly where a player should be drafted in relation to other players, the fewer players you can actually rank (and vice versa: the more players you try to rank, the less accurate you can be with their specific relative values). This is due to the overwhelming number of variables that you have to consider if you want to arrive at your best specific comparison.
Here’s what I mean: I can tell you with a fair amount of certainty that A.J. Green is a better wide receiver than Torrey Smith because I can account for a large number of variables vis- à -vis two guys who play the same position at the same collegiate level. I can look at all sorts of physical measurements, game film against any common opponents, game film in general to get an idea of route running/hands/attitude, interviews with each, college stats, and pretty much any other metric you can conceive. With that much information available, I can reach a fairly solid conclusion. And, if I want to extrapolate a little further, I can compare two players, decide which one is better in the abstract, and then factor in team-specific variables — offensive style, personnel, coaching philosophy or preference, etc. --- to decide which one is better for any given team. [Author’s note: unless one of your team-specific variables is "prefer the less-talented player always," the answer here will still be "A.J. Green."]
Now, contrast that with what I can factor in if I want to compare all the players in the draft. Physical measurements don’t work very well because you can’t directly compare a DT and a WR, so you are left comparing each player with a hypothetical archetype of a DT or WR, then comparing how each stacked up in their respective comparisons. Game film against common opponents isn’t overly handy, except in position-specific comparisons (i.e. which OT is better?). You’re ultimately left with very few variables that are universally applicable in your comparison, and you can’t even begin to factor in team-specific variables beyond the first few picks due to the number of "if Team A takes X, then Team B will look at..." scenarios that build upon one another like a choose-your-own-adventure book.
Because of this inherent inaccuracy, these general draft rankings, at best, give you a big picture idea of how to group players — that is, they give you a roadmap where you can say that these 40 or so players are all roughly equal in terms of their likelihood of NFL success, and any 32 of them could go in the first, etc. I would wager that, with very few exceptions, you usually can’t get any closer than +/- 10 as a margin of error, though I haven’t tested that part of the hypothesis.
Which brings me back to my initial point: defining which picks are or are not a "reach" based on how they match up with a small range of possible pick values is flawed from the start because it assumes a ranking accuracy that simply doesn’t exist. Strangely enough, this problem of assuming accuracy in the rankings is exacerbated in any true "best player available" approach. Without accounting for team-specific variables, there’s no way to say that the rankings represent the "best" choices for a specific team. The SLB who is supposed to go around 15 might not be near as good a fit for your team as the kid who is supposed to go around 40, but the reliance on expert rankings such as these when determining BPA (and the reluctance to appear to be reaching) means that a team will more likely than not take the former kid.
Which means, ironically, they didn't get the best player available because they were trying to take the "best" player available.
Trading down as it relates to classically-defined "reach." An offshoot of defining "reach" as we have above is that teams talk about trading down so that they are choosing between the Y1 and Y2 range for the player they are targeting. The problem with this, as should be obvious, is that the Y1 and Y2 are arbitrarily defined, meaning that even trading down doesn’t guarantee that you aren’t "reaching" in some sense of the word. Because value can only be defined in this context after the player has played in the NFL, a pick at any position today could be seen in retrospect as a reach or a steal. While such hindsight evaluation is pretty much useless going forward, it serves to illustrate the inexact nature of the rankings.
Which, of course, isn’t to say that trading down is pointless. If you are targeting a specific player, and you know you can trade down, gain some extra draft picks, and still get your player (within N% degree of certainty, N approaching 100% in this case), then trading down is a fantastic strategic move. But that’s only if you can still get your targeted player. Assuming a correct evaluation on your part to determine which player best suits your team, getting that player at 15 (or wherever) has more intrinsic value than getting an extra fifth-round pick (or whatever) and missing your targeted player.
The other role trading down serves, which remains somewhat independent of this analysis, is in a situation where a team isn’t enamored with any specific player, but rather has four or five that they consider equal in terms of value to the team, so they trade down with the idea of taking whichever player remains on the board when their turn comes up. This is still a valid strategy, assuming a team is not making their decision solely on whether they like or don’t like a player in the initial draft slot based on which players are "supposed" to go there.
Redefining. Before I get to my revised draft theory, we need to rework our "reach" equation to account for the uncertainty discussed above. In previous years’ versions of this theory, we have redefined it to be:
Selection of X at pick Y when X is N% likely to have been available at pick Z, where Y is your current round selection, Z is your selection in any subsequent round, and N is a percentage based on team-specific needs (including organizational risk aversion).
Now, as killtacular noted last year, too much reliance on organizational risk aversion here can result in absurd scenarios where teams who are not 100% sure that a fifth-round-quality player will still be around in the third round might take him in the second round. Clearly, even under our new theory, that would have to be considered some kind of reach.
Thus, in version 3.0, we are taking it a step further and defining "reach" as:
Selection of X at pick Y when X is N% likely to have been available at pick Z, where Y is your current round selection, Z is your selection in any subsequent round, and N is a percentage based on team-specific needs and accounting for organizational risk aversion, perceived depth at a particular position in the current draft, and historical drafting patterns (i.e. teams rarely take safeties in the first round, etc.).
Granted, this is a slightly less elegant way to define it, but I think it gets us closer to finding an optimal N% that becomes near-universally applicable. Just as with a classic definition of "reach," this newest definition impliedly carries a definition for a "good" pick.
Selection of X at pick Y when X is more than N% likely to have taken before pick Z, where N% varies depending on team-specific needs; accounts for organizational risk aversion, perceived depth at a particular position in the current draft, and historical drafting patterns; and considers whether X is seen to fill an immediate starting need (such that N% is appreciably lower if X is likely to be an immediate starter)
Again, bulkier, but more accurate.
The point of it all: The Revised Theory. Using our new metrics, I propose a shift in the way teams approach each pick. We will use 75 and 90 as our relevant N% numbers for illustrative purposes, both because I think those numbers are pretty close to optimal and because NFL teams are conservative enough in their drafting that lower numbers would risk making this too unrelatistic:
Team determines which starting positions could conceivably be filled by a draft pick, then ranks those positions in order of priority based on team philosophy, etc.
First round: Team looks at all players who could fill a starting slot and who are more than 75% likely to be taken before Team’s second round pick (for example, the Texans would look at all players more than 75% likely to be taken before pick 42, which would probably include the top 50-60 players as ranked by draftniks). Team ignores players at positions unlikely to be filled by a rookie except where the position is very shallow in this draft or where ignoring those positions leaves a team with too few options. Team then ranks the players according to its own needs (i.e. the team-specific variables and other stuff mentioned above), ignoring the projected draft slots for each as determined by media/talking heads.
Which is to say, we use the experts’ opinions to cull the initial group from which we pare down our list, but, when making our actual list, the experts’ specific rankings go out the window. When the Team’s first-round pick comes up, the first remaining player on the list is taken.
Second round: Team looks at all players more than 75% likely to be taken between its first-round pick (to account for any players who might inexplicably slide) and its third-round pick. Same ranking process and selection process as above.
[Note: It was asked last year why a team should not just look at the players actually available at the time of their second pick plus guys N% likely to be gone by their third pick? The answer is because we are doing this evaluation ahead of time, and we want to know, if player G unexpectedly slides and is still on the board, how does he compare with our list. Once the pick comes up on draft day, of course you are not going to consider the guys who are missing; the point here is to consider a certain number of them ahead of time so that you aren’t thrown for a loop on draft day, wondering "oh, man, should we take this RB that we didn’t account for instead?!"]
Third round: Team looks at all players more than likely (based on the same 75/90 percentages) to be taken between its second and fourth round picks. Same process as above.
The strategy shifts a bit for rounds 4-7, as the Team looks at those four rounds as two separate groups of people instead of four.
Fouth/Fifth rounds: Team looks at all players likely to be taken between its third-round pick and the end of round 5 (because guessing this group down to your specific pick is pointless, as there are too many variables, and many of these guys, especially by the end of the fifth, are fungible to a certain extent). Rank and pick as above, with an depth at key positions becoming a more and more important team-specific variable starting in Rd 5.
Sixth/Seventh rounds: Team looks at all players likely to be taken in the sixth or seventh as determined by a consensus of draft predictors (see…they aren’t totally useless!) as well as any players considered unlikely to be drafted, but who seem to fill a team-specific need. Rank and pick as above, with particular emphasis place on depth above other factors.
Obviously, this approach is an attempt to get teams to take a big-picture view of the draft. Despite the apparent complexity of our formulae, at its heart, the theory is a simple one: Nothing is a reach if you are taking the best player for your team that you would not be able to get later, assuming (of course) that teams do not behave irrationally with respect to any one variable when calculating N%. The likelihoods (N%) and even some of the cutoffs for player groups, especially in later rounds, are open to interpretation and argument.
The beauty of this theory, if I do say so myself, is in its simplicity. The only important thing is targeting the best player for YOU, Mel Kiper, et al, be damned. If you can trade down and still get that player, great! If not, you take the player rather than a blind trade down. In the end, the payoff should be much greater and much more consistent, as you are applying the things important to you to a broader group of players in each round. Basically, the theory removes artificial restrictions that are inherently incapable of being accurate anyway. I fail to see how that is a bad thing, provided your team is capable of accurately evaluating talent (which is a whole other post).
2010 Postscript (Revised): To A Deeelux Apartment In The Skyyyyyyyyy. When I originally wrote this post in 2009 --- /pours out a little liquor for DGDB&D --- I noted that it could conceivably work for trade-up scenarios, though I wasn't exactly sure how. Last year, in the extended director’s cut of this post, I attempted to work through it. There are a couple changes from last year’s analysis, however, most of which came from a conversation with DisplacedTexan about 45 seconds after I posted last year.
Switching from moving down to moving (on) up requires the introduction of two more variables. First, we need to quantify the difference in terms of impact that the targeted player would have over the player currently on our roster who is slated to be the starter at that position. Quantifying such things is undoubtedly hard and requires making some assumptions that you have no concrete basis for (i.e. "This kid is not going to blow out both ACLs during camp.") To normalize, we'll translate each players' perceived value into a number between 0 and 100, where 100 is the absolute best player possible at that position and 0 is David Carr. We'll call the difference between the two players "D."
The second variable is a team-specific measurement of what D is actually worth to your team based on scheme, coaching, etc. Think of it as marginal value in a football context --- we want to know how much that raw 45 point jump from Shaun Cody to Marcell Dareus is actually worth to us. We'll call this contextualized D "D+."
So, fancy new D+ variable in hand, what do we have for an overall theory? Recall that our equation for trading down was, more or less, N% chance to get your targeted player at your new, lower draft position, where N approaches (but never reaches) 100%. (I realize that I am not accounting for the possibility that the value of the trade package could be high enough that failing to get your targeted player would not be a huge deal, but that is another post for another day.)
To account for our new variable, the equation for a good trade-up scenario would then be:
Pick of player X made at pick Y in exchange for (pick Z + value), where:
D+ of Player X > (D+ of top player you were targeting at Z PLUS D+ of average player likely to have been taken at the later pick(s) if you gave any up);
Player X is N% likely to NOT be available at pick Z; and
No player N% likely to be available at pick Z offers a comparable D+.
That is, where the marginal benefit to your team is greater than what you have to give up to get the player, the player is highly unlikely (N% approaching 100%) to be available with your original pick, and you can't get similar benefits from a different player by standing pat, you're in a scenario where a trade up makes sense.
The hard part here is being realistic in calculating D and D+ and being able to determine the appropriate "value" to be included with Z. I feel like the old draft pick value chart is useless in this equation simply because of how much emphasis it places on first-round picks, but I haven't checked this to be sure. Perhaps this chart would provide a better starting point.
Concerns Specific to 2011 Only (We Hope). Thanks to the ongoing labor dispute and all its attendant issues, application of the above theory in the 2011 draft has a few extra wrinkles in need of ironing. First, you’ve got the fact that, per the NFL, draft-day trades can only be made with picks, not players. Meaning that you are forced to account for the hard-to-calculate additional lost D+ in any trade up, rather than trading away a player whose value you have a more concrete sense of. Additionally, also per the NFL, teams who trade picks from future drafts do so "at their own risk," because the uncertainty of everything means that the new deal (if one is reached) might change things such that the trade you made winds up being a bad deal for you. Roll the dice, muchacho!
Next, it appears highly unlikely that a deal will be reached that would allow free agency to begin prior to the draft, so teams could be forced to address the most pressing issues via the draft just in case there is no free agency this year or in case (as is more likely) the new labor deal changes the salary cap in such a way that a team can’t afford to fill the holes they’d planned on filling in free agency. Plus, just from a practical standpoint, it seems far riskier for a team with glaring needs to hope to win a bidding war in free agency rather than address the needs to whatever degree possible during the draft.
Conclusion. As in previous years, I consider this a work in progress. If you drill down to the very, very basic premise of this whole thing, it is simply a way to see the big picture going into a draft. Yes, it’s also an attempt to quantify things that rarely have numbers assigned, and, yes, it requires a thorough understanding of your organization and a strong ability to evaluate talent, both currently on your team and available in the draft, but I fail to see how it is the least bit detrimental to strive for such things.
Now, not that any of you has ever needed an excuse to point out flaws in anything, but I welcome such feedback on this post. Some of those insights were integral in drafting version 3.0, and they will be just as helpful in drafting 4.0.
And, with that, fire away with your jokes about how the April 2012 version is the next time I am likely to post.