Value Based Drafting: Part II

The following article is a guest post by Frank DuPont, author of Game Plan: A Radical Approach to Decision Making in the NFL. You can also follow Frank on Twitter.

Earlier in the week we started to look at Value Based Drafting by covering some basics about what VBD is and how you can create a baseline that will allow you to compare players at different positions.  If you haven't read that piece, it's probably pretty important that you go read it now before going any further in this article.  In this article I'm going to expand into using VBD, which also requires a pretty healthy discussion of forecasting.

First though, some readers emailed me to ask how I accounted for the Flex position in coming up with the baselines.  I've actually updated the original post to explain, but it's worth going over again.  Forgetting about Flex for a minute, the baselines for RB and WR end up being RB35 and WR29.  That gets both positions to 408 games.  You can tell from those numbers that it takes more players to fill out the RB slots because RBs are less reliable in terms of how many games they play.  Then when you add in the Flex requirement, it ends up being filled mostly with WRs, but you go to WR44 and RB44 to fill out the Flex position.  It's pure coincidence that it happens to be the same number for both positions.  But RB44 and WR44 end up being the baseline for those positions when Flex is accounted for because it takes all of those players to fill out the 408 games for each position, along with the 204 games for the Flex.

A Word About Predicting the Future

After you've established your baseline, you have to come up with some projections in order to rank your players.  This is at least as important as coming up with the baseline.  Ok, it's actually probably a lot more important than coming up with the baseline so I'm going to spend a little time talking about forecasting.

There are three important things that we should never forget when we're talking about projections.  First, we're essentially trying to predict the future.  That's an inherently difficult task.  Don't be tricked into thinking that the recent past could have easily been predicted just because we can look back on it today and talk ourselves into believing that it was obvious all along.  Prediction is a difficult game. 

The second thing to keep in mind is that we should plan for reversion to the mean.  Remember Peyton Manning's 2004 season when he threw for 49 touchdowns?  He threw 28 the next year.  That's really close to being half of the 2004 total.  Remember Chris Johnson's 2000 yard season in 2009?  He ran for 1300 yards the next year.  After Randy Moss' 23 touchdown year he caught 11 the next year.  The point is, reversion to the mean happens and a lot of the time we won't be able to see the causes of the mean reversion ahead of time.  This might sound obvious, but we're going to be faced with a similar problem this year when we look at the draft board and see about 5 QBs coming off of what were all essentially record breaking years.

The last thing to keep in mind regarding projections is that oftentimes the biggest fantasy seasons from the year before were really just guys who managed to stay healthy all year.  If we take the guys who managed to stay healthy all of last year and pencil them in for a few missed games this year, and then take the guys who missed a few games last year and pencil them in for a little better luck in the injury department, those small changes will have a pretty dramatic impact on our projections.  Here's a hypothetical that might make this more concrete for you.  Imagine if Tom Brady had lost Rob Gronkowski for a few games last year and then imagine that Wes Welker had played most of the year at 70% health.  Now imagine that Brady missed a game due to injury and then played through a few others with that same injury.  What would his numbers have looked like?  What I've just described is pretty close to what actually did happen to Tony Romo last year.  The details might have been a little off, but I'm sure you can see that luck (or randomness) had a pretty big impact on the difference between Brady and Romo last year.  When we create our projections we have to acknowledge that luck and randomness play a part in football.

Similarity Based Forecasts

One of my favorite ways to project players is by using a similarity method.  I take a player like Cam Newton, look at players who had seasons similar to Newton's 2011 campaign (Daunte Culpepper for instance), and then just look at what those players did the following year.  This is a method that is similar to the way that Pandora chooses music, except that instead of looking for attributes like vocals and percussion that songs might have in common, I'm looking for things like yards per attempt, touchdowns, and rushing yards that can be used to demonstrate QB similarity.  I also compare players on things like height, weight and age.  A similarity based comparison does one critical thing in creating forecasts and that is account for some reversion to the mean.  But because we can actually see players who were similar to the players we're studying, similarity comparisons have a powerful psychological effect as well.  They make the idea of mean reversion real because we can see actual examples.

These things are always easier to visualize, so let's do that.  Below is a table that contains a list of receivers who had similar seasons to Jordy Nelson's 2011 season.

You can see that even while some of these players might differ from Nelson in one area or another, as a composite, they are very close to Nelson.  Then I look at what these players did in the year after they were similar to Nelson.  Those results are shown in the following table (with some names missing for guys who didn't record any stats in the following year).

The following article is a guest post by Frank DuPont, author of Game Plan: A Radical Approach to Decision Making in the NFL. You can also follow Frank on Twitter.

Earlier in the week we started to look at Value Based Drafting by covering some basics about what VBD is and how you can create a baseline that will allow you to compare players at different positions.  If you haven't read that piece, it's probably pretty important that you go read it now before going any further in this article.  In this article I'm going to expand into using VBD, which also requires a pretty healthy discussion of forecasting.

First though, some readers emailed me to ask how I accounted for the Flex position in coming up with the baselines.  I've actually updated the original post to explain, but it's worth going over again.  Forgetting about Flex for a minute, the baselines for RB and WR end up being RB35 and WR29.  That gets both positions to 408 games.  You can tell from those numbers that it takes more players to fill out the RB slots because RBs are less reliable in terms of how many games they play.  Then when you add in the Flex requirement, it ends up being filled mostly with WRs, but you go to WR44 and RB44 to fill out the Flex position.  It's pure coincidence that it happens to be the same number for both positions.  But RB44 and WR44 end up being the baseline for those positions when Flex is accounted for because it takes all of those players to fill out the 408 games for each position, along with the 204 games for the Flex.

A Word About Predicting the Future

After you've established your baseline, you have to come up with some projections in order to rank your players.  This is at least as important as coming up with the baseline.  Ok, it's actually probably a lot more important than coming up with the baseline so I'm going to spend a little time talking about forecasting.

There are three important things that we should never forget when we're talking about projections.  First, we're essentially trying to predict the future.  That's an inherently difficult task.  Don't be tricked into thinking that the recent past could have easily been predicted just because we can look back on it today and talk ourselves into believing that it was obvious all along.  Prediction is a difficult game. 

The second thing to keep in mind is that we should plan for reversion to the mean.  Remember Peyton Manning's 2004 season when he threw for 49 touchdowns?  He threw 28 the next year.  That's really close to being half of the 2004 total.  Remember Chris Johnson's 2000 yard season in 2009?  He ran for 1300 yards the next year.  After Randy Moss' 23 touchdown year he caught 11 the next year.  The point is, reversion to the mean happens and a lot of the time we won't be able to see the causes of the mean reversion ahead of time.  This might sound obvious, but we're going to be faced with a similar problem this year when we look at the draft board and see about 5 QBs coming off of what were all essentially record breaking years.

The last thing to keep in mind regarding projections is that oftentimes the biggest fantasy seasons from the year before were really just guys who managed to stay healthy all year.  If we take the guys who managed to stay healthy all of last year and pencil them in for a few missed games this year, and then take the guys who missed a few games last year and pencil them in for a little better luck in the injury department, those small changes will have a pretty dramatic impact on our projections.  Here's a hypothetical that might make this more concrete for you.  Imagine if Tom Brady had lost Rob Gronkowski for a few games last year and then imagine that Wes Welker had played most of the year at 70% health.  Now imagine that Brady missed a game due to injury and then played through a few others with that same injury.  What would his numbers have looked like?  What I've just described is pretty close to what actually did happen to Tony Romo last year.  The details might have been a little off, but I'm sure you can see that luck (or randomness) had a pretty big impact on the difference between Brady and Romo last year.  When we create our projections we have to acknowledge that luck and randomness play a part in football.

Similarity Based Forecasts

One of my favorite ways to project players is by using a similarity method.  I take a player like Cam Newton, look at players who had seasons similar to Newton's 2011 campaign (Daunte Culpepper for instance), and then just look at what those players did the following year.  This is a method that is similar to the way that Pandora chooses music, except that instead of looking for attributes like vocals and percussion that songs might have in common, I'm looking for things like yards per attempt, touchdowns, and rushing yards that can be used to demonstrate QB similarity.  I also compare players on things like height, weight and age.  A similarity based comparison does one critical thing in creating forecasts and that is account for some reversion to the mean.  But because we can actually see players who were similar to the players we're studying, similarity comparisons have a powerful psychological effect as well.  They make the idea of mean reversion real because we can see actual examples.

These things are always easier to visualize, so let's do that.  Below is a table that contains a list of receivers who had similar seasons to Jordy Nelson's 2011 season.

You can see that even while some of these players might differ from Nelson in one area or another, as a composite, they are very close to Nelson.  Then I look at what these players did in the year after they were similar to Nelson.  Those results are shown in the following table (with some names missing for guys who didn't record any stats in the following year).

Jordy Nelson 2011 Similar Receivers – Year 2

Overall, the group performed extremely well on the year after they were similar to Nelson 2011.  But one glaring difference is that their touchdowns were down fairly dramatically.  When we're thinking of drafting Nelson this year, it might be helpful to remember that he could be in for the same sort of mean reversion that affected Dwayne Bowe and Miles Austin in recent years.  They were still decent receivers, they just didn't put in back to back years of really high touchdown totals.  This is the kind of information that using similarity based comparisons in our forecasts yields.

I've spent a lot of time talking about forecasting because it's at least as important as creating a VBD baseline.  But let's move on to actually applying those forecasts so that we have some actionable fantasy intelligence.

I went ahead and ran projections for the top 44 WRs and RBs, along with the top 15 QBs and TEs.  Then I matched up those projections with ADP to see what ranges of positions offered relative value.  Here are the broad stroke takeaways from that exercise:

• My similarity based projections don't assign a lot of value to the top QBs this year.  They don't project for a high enough value over the baseline 15th QB, in order to justify their draft position.  Instead, the highest relative value for QB is in the QB7 range.  I could actually see a draft strategy that focuses on trying to pick up Tony Romo in that range in lieu of spending an early pick on one of last year's top QBs.  Romo actually shares a number of traits with the top tier of QBs, including a completion percentage over 65%.  But with Romo you get the "bad luck" discount based on the numerous injuries that affected the Cowboys' offense last year. Another reasonable strategy would be to try to get Michael Vick in this range as he was actually essentially equal to Cam Newton last year except for rushing touchdowns.
• Most of the top players at a position do not offer excess value because they are projected highly, and also cost a high pick.  Perhaps the only exception is Rob Gronkowski, who I have recently called a reasonable approximation of Calvin Johnson (Gronk is also a few years younger).  This is an area where changing the TE baseline from TE8 to TE 15 is meaningful.  That means that Gronk is worth relatively more.
• On a relative basis, the range of WRs that offer the highest value compared to where they're being drafted are between WR10 and WR20.  Wide receiver is a position that I think will be fairly flat this year, which is to say that there won't be a huge difference between the very top WRs and the starter level WRs.  The NFL is now a "spread it around" league.  But at the WR position, the WR10-WR20 range does appear to offer the most relative value.  This is the range of WRs that includes Julio Jones, Jordy Nelson, Victor Cruz, Demaryius Thomas and Hakeem Nicks.  That's a pretty attractive group and I could see bypassing the WRs in the top 10 and taking two guys from the WR10-WR20 group.
• Despite the relative devaluing of the RB position, my similarity projections still favor going RB early.  This is not so much of a vote of confidence for the early RBs.  Outside of the top four guys, there are a lot of question marks.  This is a comment on how terrifying the guys at the back end of the RB position are.  Remember that we know that the approximate demand for RBs is equal to about the top 44 players at that position.  The guys who are currently going in the RB35-RB44 range are guys who if they got lucky and someone got injured, might still be in timeshare backfields.  It might be scary to go RB early this year, but it might be a lot scarier to wait.
• You now have the baselines needed to calculate your own values based on whatever projections you might favor.  I also realize that some people play in leagues that don't follow the standard scoring that my baselines were created from.  I'll try to work on putting together a spreadsheet or a form that will allow people to calculate baselines for leagues that might have different roster requirements.  When I have that done I'll update this post with a link and also send it out through Twitter.

If you're still awake after going through a lot of material on the topic of Value Based Drafting, I do think it's important to note that a fantasy draft shouldn't be boiled down to one number.  Ideally you want to use the information that VBD gives you to go through various alternatives for your draft.  You want it to tell you what the relative costs of waiting on a position are.  If you decide that you're in love with Aaron Rodgers, you can still consult VBD to see what kind of values will be available at the other positions later.  Use Value Based Drafting as another tool in your arsenal.  Don't completely farm out your fantasy draft to one idea.