What You See is All There Is: Recency Bias in Fantasy Football Injury Talk

It’s not a stretch to say that 95% of fantasy football talk takes the form of “Player X is risky”  or “Player Y is safe.”  But the problem with the talk of safety and risk is that our brains aren’t very good at doing things like assessing safety and risk.  We’re susceptible to something called recency bias (also known as the availability heuristic).  It’s fairly easy to illustrate recency bias with examples from fantasy football.

Prior to the 2011 season the most common analysis of Detroit Lions quarterback Matt Stafford focused on his injury proneness.  Stafford had missed 18 of his first 32 NFL games due to injuries.  The quarterback was then drafted in the 9th round of fantasy drafts last year due to the fear that it wasn’t possible for him to stay healthy.  But after one season of health Stafford is now going in the 2nd round of drafts.  The talk of injury proneness is gone.  A reasonable question to ask is which of the following is most likely to be correct as to Stafford’s injury proneness?

  1. Stafford has been cured of his injury proneness.
  2. Stafford is still injury prone but just happened to string together a season of consecutive starts without becoming injured.
  3. Our collective perception of Stafford as injury prone was simply wrong to begin with.

It doesn’t matter which is correct because they’re all problematic for the discussion of injury proneness.  Both the second and third possibilities are essentially acknowledgments that we don’t have the ability to perceive injury proneness in a way that will be helpful in forecasting the future.  The first possibility, that injury proneness could be cured, is essentially an acknowledgment that it doesn’t even exist because if injury proneness is anything, it is a difference in physiology.

The psychologist Daniel Kahneman has an idea which he calls “What you see is all there is” (or WYSIATI).  Kahneman has shown that humans often make errors in judgments because we only consider the information that we’ve personally observed.  But often the evidence that we’ve observed isn’t all the evidence that there is.  Kahneman’s concept of WYSIATI is thick in discussions of injury proneness.  One of the most common things that happens in discussions of injury proneness is that only a player’s NFL injuries are taken into account.  Fantasy football owners who don’t follow college football haven’t seen the player’s college injuries and thus the player’s college injuries may as well not exist.  During Adrian Peterson’s career at Oklahoma, he missed time in four games due to a high ankle sprain and then missed seven games with a broken collarbone.  Peterson’s college injuries aren’t much different than San Diego Chargers running back Ryan Mathews’ injuries that have resulted in Mathews being painted as injury prone.  So was Peterson injury prone in college, then cured of his injury proneness in the NFL, and then became afflicted again last year when he tore his ACL?  If Peterson’s history of injuries in college didn’t dictate his injuries during his next 4 NFL seasons, why would we think that Mathews’ past injuries would tell us anything about his future tendency to become injured?

Let me throw out a hypothetical situation now so that I can open up your mind to the idea that our perception of injury proneness might simply be our minds trying to see a pattern where none exists (or at least where the real pattern is too complicated to be perceived).

Let’s pretend that I am going to take 32 running backs in the NFL and simply assign injuries at random.  I’m going to do it in a lottery process.  I’m actually going to do it with a little help from Excel’s random number generator.  So I’m basically going to just draw a number out of a hat and that number will be between 1 and 32.  Then I’m going to put the number back into the hat, shake it up, then draw again.  Each time I draw a number it will be like I’m assigning an injury to a running back.  Here are the results of this random number game drawing the numbers between 1 and 32 (you can actually replicate this experiment using the RANDBETWEEN() function).

Number Times Drawn Out of Hat Number Times Drawn Out of Hat
#1 2 #17 1
#2 0 #18 0
#3 0 #19 1
#4 2 #20 0
#5 3 #21 0
#6 3 #22 0
#7 3 #23 0
#8 2 #24 1
#9 0 #25 1
#10 1 #26 0
#11 1 #27 0
#12 1 #28 1
#13 2 #29 1
#14 0 #30 0
#15 1 #31 3
#16 1 #32 1

 

This is based on 32 drawings and you can see that each number was not drawn the same number of times.  Numbers 5-7 were all drawn 3 times.  Several numbers weren’t drawn at all.  Now imagine that instead of random numbers that came out of Excel, we were talking about injuries.  Numbers 5-7 would be considered injury prone, while #26 and #27 would be said to have a talent for staying on the field.  The key is that this is a random process and yet it’s given us results that don’t look random.  These results would be easy to look at and think that they mean something even though we know that they don’t mean anything.

To be clear, I’m sure that humans differ in important ways when it comes to the probability of injuries occurring.  I’m sure that they differ in how much stress their body parts are able to endure.  I’m sure they differ in recovery times.  I’m sure they differ in their ability to move in a way that avoids injury.  But the problem is that we know that a good amount of injuries are random.  We also know that they are low frequency events.  So can our human brains be relied upon to understand the effect of partially random, low frequency events when we know how prone we are to recency bias?

The problem is this:  If a player has had a lot of recent injuries then our expectation is that they will have a lot more.  If a player hasn’t had many recent injuries then our expectation is that they will remain healthy.  That’s as complicated as our thinking process gets and I don’t think it’s particularly enlightened or anything that we should be putting a lot of stock in.

  • JacksonBlackson

    Excellent.

  • TL

    how about accounting for the fact once a player sustains an injury, they are more likely to sustain a similar injury since that muscle/joint/ligament is likely not as strong as it was before? (granted this doesn’t apply to all injuries)

    • jack_sprat2

      Anyone care to take the other side of my bet, 100 : 100, that the injuries for which that’s most true are concussions?

    • FantasyDouche

      You can probably tell from my article that my first goal was to dismiss the idea of injury proneness in the way it’s currently talked about, where injury proneness explains both muscle strains and broken bones (Mathews). But also I tried to be careful to not say that it doesn’t exist. Only that our perceptions of it are off base. I think what you’re saying is more than reasonable.

      • Ted Pochinko

        As usual, love your line of thinking with this and I completely agree. Stafford is the prime example of a guy who’s had injury problems his whole career and one full season the general public forgets about it.

        My question is how do you account for this when putting a value on a player? I know we can’t disregard the injury prone tag all together, but how much stock do you put into it?

        • FantasyDouche

          Ted- Thanks for the comment. The rule I’ve created for myself is that if I draft a guy, he has to be healthy on the day I draft him. After that, I can’t really control it. Typically you get an “injury prone” discount anyway, so it’s not like you ever have to pay equal value for the “injury prone” guy.

          • Corporatemundo

            How do you feel about Mathews? He was being as drafted as high as 4th overall, and now can be easily be had at a 1 round discount.

  • Tony B

    Great article. People don’t understand variance in most aspects of life, another example, and exploitable in drafts.

  • Ghostfacenahmean

    Taleb could’ve written a chapter about this in Fooled by Randomness. It’s another of the countless examples of the human psyche’s miscalculation of risk vs reward. People are much more afraid of losing than they are excited about winning, and in order for that precept to hold true, we don’t have to deny that risk exists, we simply must observe that we have skewed significantly from expected value. This is also an excellent example of recency bias and filtering, not dissimilar to the ‘hot hand’ phenomenon fans often buy into during basketball games etc. While there is evidence that in some fixed difficulty, fixed motion sports, e.g. bowling it definitely is possible to develop a hot hand, in most variable difficulty sports (baseball, basketball) there is no evidence that a player who has,again e.g., made his previous 3 fg’s is more likely to make a 4th despite widespread belief that he indeed will.

    • Ghostfacenahmean

      Whoops, just noticed receny bias is in the title. Game theory ftw. Love fantasy douche.

    • FantasyDouche

      Awesome comment. Thx for reading.