About the model
What kind of model is this?
The model is a general additive model, with an additional smoothing algorithm applied. I'm not going to bore you: it's draft day. If you want to read up on building predictive models and, in particular, general additive models and smoothing algorithms, I highly recommend Elements of Statistical Learning.
What the model does not do:
- The mock draft model does not take team needs into consideration. This is because team needs should, presumably, already be incorporated in each individual mock draft used in the model. While a human may think they could use the model percentages and leverage human information on team needs to create a better model, it hasn't worked in practice for me. (Last year I tried to beat the model by tweaking the model output a few percentage points here and there based on needs and "gut"; I ended up getting beat pretty handily by the computer model. In particular, the Matt Kalil / Trent Richardson swap cost me nearly 80% in percentage loss when the Vikings and Browns swapped those draft spots; I should have listened to the model when it said that there was a finite possibility that those guys would go in slots other than those in which they were commonly mocked.)
- The mock draft does not leverage combine information or school attended information. I made an attempt to incorporate the information last year, but found that it added absolutely nothing to the model. Again, mock drafters have probably already considered combine statistics in their ranking of individual players; it appears that they don't over or under weigh the combine data, or else there'd be a way for me to include it. Believe me, I've tried!
While I'm not going to divulge the exact variables that are used in the model or the exact model algorithms, I will provide a basic framework for what types of variables are used:
- Average mocked position, with additional weight given to better mock drafters (and negative weights given to a few!)
- A categorical variable separating players into quarterbacks, other offensive players, and defensive players.
- A variable that takes into consideration the degree to which there is one specific consensus slot in which a player may be drafted. In other words: if everyone mocks Ryan Tannehill 8th, the model knows something's up with Ryan Tannehill and slot 8, specifically.
- Variance of mocked position
- Minimum mocked position
At different ranges, the variables used are/may be weighted differently or removed altogether.
The Top 5 Picks
2013 is an anomoly as far as drafts are concerned: no deal has been reached between the Chiefs and a player, while there are several different players mocked in this slot. As such, the data predicts an incredibly volatile top 5 this draft, which should make for a great viewing experience. Let's take a look at the projected top five pick odds:
|1st||2nd||3rd||4th||5th||Top 5 Total|
A few things that jumped out to me:
- Despite the uncertainty between mock drafters at pick 1, Joeckel remains a 2:1 favorite to go at the first pick. The model doesn't see anyone but Joeckel or Fisher going here more than 0.5% of the time.
- Geno Smith projected in the top 5 at just 3%. This is the result of a few things: none of the best mock drafters have Geno Smith in their top 5; Mel Kiper didn't even have Geno Smith in the first round; "reaching" for a quarterback in the top 5 would be relatively unprecedented; high variance players this early in the draft are poor bets to go at their highest mock position.
- Lane Johnson, someone many of us hope falls to pick 11, is off the board by pick 5 38% of the time. The odds of him lasting to us don't look too good. (More on this later).
Who's left at pick 11?
Great question, I'm glad you asked. The graph that follows shows the odds of any given player being chosen after that draft slot. Basically, whatever the value is at 10, that's the odds the player is still draftable at pick 11. Here's how this looks for the notable offensive linemen:
Doesn't look good if you're hoping the Chargers end up with one of three "elite" tackles in this draft. However, the odds are good that one of Fluker, Warmack, or Cooper will be available.
The same chart for WR and DBs:
For those of you yearning for some help in the secondary or for another weapon for Philip, that chart is certainly good news.
The interesting storyline: where will Te'o and Barkley go?
Two of the more interesting storylines of this draft, in my opinion, are those of Te'o and Matt Barkley. Te'o for obvious off-the-field questions and a poor showing in the BCS title, and Barkley for an up-and-down collegiate career and general doubt on whether he can play at the next level (despite the multitude of praise heaped his way over his lengthy career at USC).
Here's the distribution representing the odds of the player being selected at any of these selections. Te'o has a more bell-type curve, while Barkley's is a classic "quarterbacks mocked at the end of the first/early second tend to jump up" distribution with a very small blip in the early teens.