About a month ago, I wrote an article looking at the draft value for running backs. My approach was to demonstrate a value over replacement (VoR) philosophy. For those unfamiliar with the concept of VoR, it is the idea that some positions are harder to replace than others. It isn’t a measure of value specifically, but rather a measure of the drop-off in performance relative to the next available player.
In that article, I showed that some common metrics of RB performance do not drop-off substantially based on the player’s draft position. In other words, delaying an RB selection in the draft may not have a high cost in performance. But is the same true for other positions?
I’ve stated previously that there is not a purely analytical way to identify which position to draft and when, as it varies by team need, relative draft strength of position, and the individual skill set of the player being considered. However, an analytical approach can provide a relative baseline ranking of VoR between positions to help form a strategy around those other, more critical variables.
The Colts are in need of pass rushing and have a #3 pick available (at least for now). So, is it wise for the team to spend that pick to fill that need?
First a qualification: pass rusher isn’t a specific position in the NFL and so it won’t appear in any of the NFL stats or draft data. As such, my analysis will simply be comprised of drafted players listed as defensive ends by Pro Football Reference between 2000 - 2016. This means I am going to be excluding some players who are considered NFL pass rushers and including others who are not, but the numbers should make for a good sample(1).
DEs are about 13% of all 1st round picks, which along with WRs leads all positions(2). So teams are not shy about picking DEs early. To estimate the position’s long term value, I created a performance decay curve using the average AV(3) earned by DEs each year after the draft. For comparison, I have shown the same curve for drafted RBs.
This comes with the standard disclaimer that AV is not the best stat for any given position but when comparing across disparate positions, like this, it’s pretty useful. This graph shows that DEs “hold” their value a relatively long time, especially when compared to the decline in the RB curve.
The above chart shows the DE decay curve broken out by draft round. It is no surprise that there is a drop-off between 1st and 2nd rounds, but notice how dramatic that drop is, especially in relation to the 2nd through 4th rounds, which look like the same curves past year 5.
So is this it? Is this the VoR curve I was looking for? Of course not. Like I would actually stop after just 2 graphs.
One problem is that I don’t want to use AV as my performance metric. There is not a lot of freely available data for defensive lineman, but I was able to extract 5 metrics that are all important for a pass rusher: sacks, QB Hits, tackles for loss, forced fumbles and tackles on 3rd or 4th down that prevents a 1st down conversion(4).
These curves shows all of the stats on a per game basis, revealing a trend of decreased production for later draft rounds. But this graph is a poor data visualization choice as the volumes vary dramatically between the stats and therefore the relative slopes for the lower volume trend lines are flattened.
To fix this, I converted the curves into normalized numbers. Specifically, I used z-scores, but if you don’t know what that means, it’s not important. Just think of it as re-scaling the numbers to comparable amounts.
That’s better. You can see that the trends are very similar between all metrics. Okay, so is this a VoR curve?
This graph is just measuring game volume and I already knew that starters (early round picks) would have more game volume than back-ups (later round picks). This merly confirms that.
What I need is to know is how well do those those back-ups perform relative to the starters. I want quality not quantity. To do that, I have to even the field by converting the data to efficiency metrics using snap counts(5).
These are the normalized curves on a per snap basis and through the first 6 rounds, the trend is obvious. Backups don’t play as well as the starters.
Now this may be a “well, duh” moment for you, but if you read my previous article the “ah-ha” from that was that the running back curves did not display a performance drop-off. On average, RBs drafted in later rounds did just as well as the early rounders in performance efficiency stats like YPC and explosive run rate. So, this “duh” stat is actually a new result.
As a final step, I am going to combine all of these trend lines into a single curve by simply averaging them. If I wanted, I could use weighting factors to give more value to one metric over another, but without research into what those specific values should be, I’m happy with equal weightings. As a comparison, I am going to do the same for RBs using the previously mentioned stats.
And there we have it. This suggests high VoR for DEs as the drop-off is significant between draft rounds. Comparatively, the same cannot be said for RBs where, until the crazy spikes in rounds 5 & 6, the curve is relatively flat(6).
What this has shown is that there is an “urgency” to finding a pass rusher in the draft. The opportunity cost of not selecting early is much higher than a position like RB.
Now this isn’t a proof that the Colts must select a pass rusher at #3. As I stated before, there is no purely analytical way to do that. But it does provide support to make that pick if the other variables fall in place.
1) There are 407 draftees between 2000-2017 with a designation of DE. Those players translate to 1,910 seasons on an NFL roster. Of those, 1,545 are designated with a position of DE (1,475) or OLB(70), which is about 81%. Conversely, there were 4,178 draftees with a non-DE designation. Those players translate to 17,684 seasons on an NFL roster. Of those, 514 are designated with a position of DE (300) or OLB(214), which is about 3%.
2) Since 2000, defensive back is the is the most popular first round pick but that is not a single position rather a merging of cornerbacks and safeties.
3) AV data extracted from Pro Football Reference roster data
4) I had available, but specifically excluded, standard tackles as I’m simply not a big fan of that stat.
5) The snap count data was extracted from Football Outsiders and limited to 2012 - 2016
6) The shape of the curves is driven by the stats selected. Choose different stats and get different curves. I applied the same methodologies to alternate stats and found similar results. So the shape of the curves will change with different stats but not dramatically.