Reflections on IDP Projections – Introducing the Model

Firstly, an apology.  I’ve been terribly quiet recently for longer than I anticipated – pretty much since the draft.  It had nothing to do with not wanting to share thoughts on the IDP world, and everything to do with drowning under data.  I’ve just finished my first ever attempt to project a full year’s worth of IDPs and I wanted to share how I went about it.

As everyone who reads this will already know, there is more information on offensive fantasy football than almost anyone can reasonably consume.  There are dozens of good websites, even more podcasts, TV shows and a world of social media that all have their own heavily-researched and well thought-out opinions.  Sites like RotoViz are heavy on the data side, and others like Pro Football Focus are all about gauging on-field performance. 

The IDP world, however, is relatively light on information.  As a best guess, roughly 10-15% of fantasy football leagues include IDPs, so there is a much smaller audience.  Consequently there is a much smaller pool of people working in the space.  I’d estimate there are only 30 or so writers who dedicate significant time to IDP.  So there is less information as a whole and more supposition.  Some of that, of course, is spectacular and wonderful and fascinating and helpful.

My own personal belief is that there is much more accuracy to be squeezed out of IDP.  This project is my own humble attempt to be as accurate as I can be.  Rather than me putting together rankings based on who I think is going to be “good” or “bad”, I want to have a sensible go at estimating exactly how every relevant IDP will produce and then building rankings based on that.  I’m not stupid enough to think I will be wonderfully accurate but I do think a proper process will help me eradicate some of my personal prejudices and incorrect assumptions.

[am4show have=’g1;’ guest_error=’sub_message’ user_error=’sub_message’ ]

Going through past data, it strikes me that IDPs can vary wildly in their performance.  We all know that contract years, new coaches, experience, age and usage drastically change how players play and produce.  Coaches, on the other hand, are relatively conservative.  Like the scouts in the Moneyball movie, they like to think they know what they know and are happy to continue with what’s got them this far.  I’m certainly not going to pretend I know better, but this does mean predicting performance and production can be more accurate by understanding coaches than players.  This means that my model is built primarily around how coaches’ schemes utilise different roles.  My primary contention is that a certain defensive coordinator will use his strong safety (or nose tackle, or ‘will’ linebacker, etc.) in a certain way pretty consistently.  He’ll adjust depending on the quality and traits of the players available to him, but not vary the basic scheme that much.

The Method

The first thing I did was build a database of all current defensive coordinators in the NFL.  For each of them, I broke down what that coordinator has managed to produce by position over the last several years.  That means primarily: how many snaps, tackles, sacks, interceptions etc. has each relevant player on the roster managed?  As suspected, it turns out there is a fair amount of predictability across years here.  Even for coaches who have switched teams and used entirely different personnel, we mostly see clear tendencies.

In the case of new coordinators, I’ve made assumptions based on which coaching tree they come from.  So for example Matt Burke in Miami has never coordinated before, but comes from the Vance Joseph/Mike Zimmer/Marvin Lewis tree.  I’ve assumed he will use a similar defense to them – at least until I’m proven wrong.

That was a mammoth task and it’s taken a while, but it gave me a database of roughly 640 individual roles (each NFL team has about 20 IDPs who see significant playing time per season).  What I did next was make my best estimates for depth charts.  Obviously plenty of things will change, and I’ll be flat-out wrong quite a lot, but I needed to start somewhere. I’ll adjust as we get more information.

The reason I say “depth charts” and not starting line-ups is because the terms ‘starter’ and ‘backup’ are not nearly accurate enough.  Some schemes keep their starters on the field as much as possible (linebackers for Gregg Williams are a good example) and some like to rotate to keep players fresh (Williams does this for his defensive linemen).  What this means in effect is that for some teams being the nominal “starter” is a big deal, whilst for others it’s only a slight advantage over the backup.

This is perhaps the biggest issue right now.  Some teams are very clear, some are mostly set and some are all over the place.  Good luck anyone trying to predict the 49ers defensive line rotation at this stage.  But again, I’m not going for 100 per cent accuracy at the moment – I’m trying to set a sensible framework which I can use to set a big board and rankings for the season.

After I put all that data in (with a lot of help from the rest of the IDP team here at DLF), I had a big list of data telling me what I think approximately 700 NFL players going to produce.  Academically, that’s interesting stuff, but it doesn’t help much with IDP leagues.  So I simply built a calculation enabling me to update projections based on league settings.  Regardless of whether your league awards two points or six points for a sack, a given player still has ten sacks in my model.  I can very quickly change scoring and have a custom ranking built for individual leagues. 

There are a few other things going on that I’m sure you’re thinking. 

What about injuries?

The model factors some injuries in.  Projections include the fact that all NFL teams have injuries every season, and therefore a sensible amount of missed time is inherent.  That will hold up across the model as a whole.  But it doesn’t account for major injuries to specific players.  If your starting linebacker goes down with a broken pelvis in week one, I’ll be miles out. 

What about trends?

Some payers and schemes are clearly trending up or down.  Cam Wake is getting rapidly older (just like me). Jalen Ramsey clearly developed through his rookie season.  I’m working on a way to make some subtle alterations here to reflect things like this but at the moment it’s not included.

What about player quality?

Some players transcend scheme.  Although Gregg Williams likes to rotate his linemen, Aaron Donald has been so good he had to stay on the field.  Ndamokung Suh stays on the field because the Dolphins are desperate to try and get value from his ridiculous contract.  I’m trying to include this too, but some instances exist in the model where stars with new coaches are not accurately reflected.

Which brings me to where I am now – my first ever projection model built from the ground up.  I don’t think I’m right on everything.  There will be some embarrassing, glaring errors in the system.  I’m sure you’ll have your own thoughts on how it can be improved.  But I’m happy.

I’ll be sharing my work over the summer by looking at individual teams and what I think will happen player by player.


tom kislingbury