Monthly Archives: August 2014

Regression Red Flags

Posted on by
6

The season is just one week away, and I haven’t been able to do anywhere near as much as I’d like to regarding a season preview.  However, I’ll try to remedy that this week, starting with a lightning-round style rundown of potential areas for regression.  Before I get there though, take a minute to download your copy of the 2014 Eagles Almanac here.  Just $10 for a PDF, and you’ll find a lot of great articles to get you ready for the season.

Now, the important things.

Before last season I spent a lot of time explaining why the Eagles would likely rebound strongly from the 4-12 season.  I projected the team to win 9 games and challenge for the division title when most pundits had them relegated to 5-6 wins.  What did I know that they didn’t?  Regression factors.

It won’t be news to any readers here, but there is a large degree of luck involved in nearly every aspect of the game of football.  A few specific areas, in fact, have a significant impact on the game and are almost entirely random.  That means we can get a lot of information about this year’s expected performance by looking at last year’s statistics and combing for outliers.  For example, the 2012 Eagles recovered just 35% of all fumbles.  Fumble recovery is almost all luck, and we’d expect a team to recover close to 50%.  That means, holding all other factors equal, the Eagles were likely to improve last year because they were very likely to recover a greater percentage of fumbles.  For the 2013 season, the Eagles actually recovered 43.75% of all fumbles.  That’s still less than we’d expect, but it’s a sizable improvement from 2012.

Now that I’ve explained it, let’s take a look at a handful of specific areas for which we’d expect mean-regression.

Pythagorean Expectation

This isn’t really a “mean-regression” candidate, but it’s vital to point out.  Last season the Eagles scored 442 points and allowed 382 points.  That performance would lead us to expect a Win/Loss record of 9.4-6.6 (from Pro-football-reference.com).  The upshot is that the Eagles record was slightly better than they “deserved”.  A difference of .6 wins isn’t big, but the direction is important.  The Eagles weren’t quite as good as their record from last year suggests.

Fumbles and Fumble Recovery Rates

I covered this a bit in the example above, but this is pretty low-hanging fruit as far as statistics with the potential for mean-regression.  Last season, the Eagles recovered roughly 44% of all fumbles.  They fumbled the ball 1.1 times per game.  Both of those numbers are good news for Eagles fans.  1.1 fumbles per game is close to average, and a 44% recovery rate suggests the Eagles are more likely to be better this year than worse.

However, there is one potential flag.  Opponents fumbled the ball 1.7 times per game when playing the Eagles.  That placed them tied for 1st in the league (with 4 other teams). Now, if you remember this post from last year, you’ll know that there is no significant persistence in forced fumbles from year to year.  Here is the chart:

Therefore, we can’t say that the Eagles were “good” at forcing fumbles last year and will be again this year.   Instead, we can expect that other teams will NOT fumble the ball as often as they did last year.  That’s bad.

However, 1.7 fumbles per game is not an outrageous amount.  So while there will probably be some regression, it won’t be huge and it might also be balanced out with positive regression in the overall recovery rate.

Fumbles verdict:  Neutral

Injuries

I covered this a couple of weeks ago, but it’s important enough to repeat.  According to Football Outsiders, the Eagles were 2nd in the league last year in Adjusted Games Lost (behind Kansas City).  That means they were healthier than every other team, as measured by this statistic.  Here’s the good news:  AGL might persist.  I looked at this stat in last season’s run-up, and found a correlation value of .30.  So, at least within the data I had, a good performance in AGL one year DOES suggest an increased likelihood of a good performance the following year.  Also, we have Chip Kelly’s “sports science” regime to consider.  It’s certainly logical to believe improved nutrition and fitness will lead to fewer injuries.  This is actually a really important aspect of the Eagles season to track.  If Chip Kelly really can keep his team healthy to a greater extent than other teams, it will be a big advantage for the Eagles going forwards.

Overall, we still have to expect some regression.  The Eagles will probably not finish in the top 2 again in AGL.  Still, with the modest persistence in AGL and Chip Kelly’s focus, I don’t expect as much regression as we might otherwise assume.

Injuries verdict: Slight negative

Interceptions

I won’t spend much time on this one, because I cover it in the Almanac and have addressed it previously.  Basically, Nick Foles was so good last year at avoiding interceptions that it’s nearly impossible for him to duplicate that performance.  However, Foles college and rookie stats do suggest an ability to avoid interceptions.  Hence, I expect Foles to throw interceptions at a higher rate, but to still rank among the best in the league in that category.  All other things equal, though, this is a negative regression indicator for the team.

Interceptions verdict: Negative

Field Position

This one is a bit under-the-radar.  Net Starting Field position is a byproduct of both turnovers and special teams, making it largely random from year-to-year.  Here is a persistence chart showing 5 years of data from Football Outsiders (2008-2012):

The correlation value is 0.14.  That’s large enough to note, but shows there’s a very large degree of variance from year-to-year.

Last season, the Eagles Net Field Position was 0.93.  That means the offense, on average, started with the ball nearly one yard farther than the other team’s offense.   0.93 was good for 11th in the league last season.  For reference, the 2012 Eagles had a Net Field position value of -6.67 yards, worse than any team from last season.  Kansas City led the league last year with a value of +9.61 yards (which is ridiculous and will definitely regress).

I’ve already explained why the Eagles will likely be worse in the turnover department this year than last.  Thus, we should expect field position to be a bit worse.  However, we also have to account for special teams.  This is a bit qualitative, but it’s clear the Eagles roster is deeper this year than it was last year.  That should result in a stronger STs unit.  I don’t have much confidence in that “analysis”, though.  It sounds reasonable, but STs performance is very unpredictable.

Of course, there’s another major piece to this puzzle that I still haven’t mentioned….Alex Henery.  More accurately, the absence of Alex Henery.

According to Football Outsiders, the Eagles had the 2nd worst performance on kickoffs last season, and Henery had the lowest gross kickoff value in the league.  I have no idea how well Parkey will play, but at the very least, he has a stronger leg then Henery.  Again, that should result in better kickoff performance.  Last season the Eagles recorded touchbacks on just 40% of their kicks, which ranked 24th in the league.

So where does that leave us?  Well we should expect some negative regression due to worse performance in turnovers.  Conversely, special teams can be expected to improve a little, at least in the areas that most directly effect field position.

Field Position Verdict:  Neutral

Wrapping it up

There are other areas to explore, but these are the primary ones.  I know I haven’t said anything groundbreaking here, and this analysis isn’t nearly as fun as it was last year (when there were a LOT of areas with very large positive regression expectations).  However, this is, in general, very good news for Eagles fans.

It says, basically, that last year was not a fluke.  The Eagles improved greatly from 2012 to 2013, and whenever you see a 6-win jump in one year, you should look carefully for luck-driven performance.  That’s not the case here.  The Eagles didn’t win games last year because of anything unsustainable, with the notable exception of Foles interception rate.

Thus, we shouldn’t expect and significant negative regression.   The fill-side, of course, is that the team isn’t likely to get a natural boost from any luck-driven areas.  That means improvement, if it comes, will have to depend more heavily on the actual skill of the players.

Final note: remember that there’s no guarantee the Eagles will finish where we expect them to in any luck-driven category.  Just because the “natural” recovery rate for fumbles is 50% doesn’t mean the team will hit it.  As explained above, teams can and do deviate significantly from the mean every year.  Just because the Eagles weren’t particularly lucky last year doesn’t mean they won’t be unlucky this year.  It’s just not our expectation.

2014 Risk Factors: Injuries

Posted on by
8

As you all know, we should be thinking about this season in terms of an expected performance distribution.  There are a range of outcomes for the Eagles this season, which varying probabilities for each related to how good/bad the team is.  Today, I want to first talk conceptually about the distribution shape.  Then I’ll move into the main topic: Injuries.

I’m going to assume everyone knows the basics of a Normal Distribution (Bell Curve).  I’ve used it often enough here that it shouldn’t be unknown.  The relevant question is: how do NFL team performance distributions compare?  Using the Normal curve as a baseline allows us to logic our way through certain adjustments, leading us to a better mental model for understanding ex-ante team expectations.

I’m primarily concerned with two dimensions: kurtosis and skew.  Skew is relatively self-evident, and more important for our topic today.  It relates to the symmetry of the distribution and the existence of outliers to either side.  Kurtosis isn’t as well known.  It also relates to the shape of the curve, but concerns the degree of peakedness vs. heavy tails.  In other words, kurtosis tells us how much data is located in the center of the distribution (or, conversely, NOT near the center).  Here’s a visual example:

Screen Shot 2014-08-07 at 10.51.33 AM

Now, to the good stuff.

What do we think the shapes of NFL performance distributions are?

I don’t have any data (yet), so we’re operating conceptually (as usual).  Let’s start with Kurtosis, because it’s relatively straightforward and not as important for out topic today.  In generally, I think NFL distributions are fairly Platykurtic.  There is a LOT of luck in the NFL.  That means a team’s “true” performance level is less likely to actually manifest than if there was little luck.  That means ANY projection we make is fairly uncertain.  As a result, it’s not enough to just say “expect 9 wins”.  Any projection of value will also include an expected range, or at least some explanation of downside/upside outcomes.

Now let’s look at skew, because that’s the more relevant measure right now.  Perhaps the most important thing to note here is that performance for an NFL team, as I’ve defined it here (Wins), is bounded on both sides.  No matter how bad a team is, it can’t win fewer than 0 games.  No matter how good it is, it can’t win 16 games.  Hence, when we’re looking at expected performance in terms of wins, the potential for outliers is limited.  Taking the next step, that means the distributions almost certainly are skewed for every team, provided we accept one more assumption as true: it is possibly, at least in theory, for a team to achieve every possible outcome (0 – 16 wins), regardless of “true” ability or expectation.  The Seahawks will almost certainly will more than 0 games this year…but it’s possible.  Even if the odds of that outcome are extremely small, if they exist they must be present on the distribution curve.

Similarly, a bad team will almost certainly not win 16 games.  But an extraordinary run of luck (like opposing injuries) could, in theory, produce a very positive outcome, up to and including 16 wins.  Again, the odds are close to zero, but they exist.

Therefore, an expected performance distribution for the Seahawks might look like this, with the X-axis representing 0 – 16 wins as you move from left to right:

Screen Shot 2014-08-07 at 11.04.10 AM

That curve is negatively skewed, as are the curves for most “good” teams, for reasons I explained above.

Now that we’ve settled that, we need to think about the reasons a good team might end up in the left side of the curve.  Put differently, we know that the Eagles are a relatively “good” team.  While their curve isn’t nearly as skewed as the Seahawks’, I do believe it’s still negatively skewed.  Given that, we can start to think about WHY, beyond the theoretical reasons (bounded range of outcomes), a team’s left tail might exist/be significant.

The most obvious reason is injuries.

Injuries, especially those to star players, present the type of negative events that can result in a team finishing with an outcome towards the left side of the distribution.  Here’s the important part: the Eagles are particularly susceptible this year, hence the team’s left tail is likely a bit larger than usual.  That’s also a big reason why I’m keeping my expectations for the team’s win total in check.  Outliers to the left side or a fat left tail will pull the mean of the distribution down.  So if we’re just talking about average expected wins (there are certainly other ways to look at this), the Eagles “true” level is likely lower than many fans believe.

If Nick Foles goes down….. If LeSean McCoy goes down….. If anyone on the offensive line goes down….

The Eagles are currently heavily dependent on just a few players.  The defensive depth chart improved a bit this offseason, but the offense (largely responsible for the team’s performance las year) is still very brittle.  The problem is, that brittleness was not readily apparent last year, and therefore is likely to be under-appreciated this year.

Last season, the Eagles ranked 2nd overall in Adjusted Games Lost, a measure from Football Outsiders that quantifies the impact of injuries a team suffers over the course of a season.  Here’s how the site describes it:

With Football Outsiders’ Adjusted Games Lost (AGL) metric, we are able to quantify how much teams were affected by injuries based on two principles: (1) Injuries to starters, injury replacements and important situational reserves matter more than injuries to bench warmers; and (2) Injured players who do take the field are usually playing with reduced ability, which is why Adjusted Games Lost is based not strictly on whether the player is active for the game or not, but instead is based on the player’s listed status that week (IR/PUP, out, doubtful, questionable or probable).

In 2012, the Eagles ranked 18th overall.

Clearly, they were more effected by injuries in 2012 than they were last year.  Similarly, the fact that the Eagles ranked 2nd last year combined with the relatively uncertain (non-persistant) nature of injuries means we should expect some mean-reversion.  Basically, it’s likely the Eagles will be more negatively effected by injuries this year, relative to other teams, than they were last season.

Of course, that itself doesn’t tell us much.  We also need to know it injuries, as measured by AGL, actually affect performance (as measured by Wins).  Well here’s the scatterplot showing AGL and corresponding Wins from 2009-2013.

Screen Shot 2014-08-07 at 11.28.28 AM

As you can see, there’s good news and bad news.  The correlation value is -.185.  If the Eagles revert towards the mean (as I expect them to), they’ll be relatively worse off than last year.  However, the correlation is relatively weak, so the effect might not be catastrophic.

Anecdotally, though, I think there’s reason to be concerned, particularly because I don’t like the QB/RB Depth Chart.  An injury to a starter is bad (and shows up in AGL).  However, if the drop-off in talent to the next guy isn’t huge, the effect won’t be significant.  Unfortunately, the Eagles don’t have that luxury at QB.  Moreover, LeSean McCoy is SO good that it’s really impossible to keep the gap between him and the 2nd stringer small.

I should probably note here that I’m not trying to be overly pessimistic here.  However, if we want to create a reasonably accurate performance expectation, we need to look carefully for risk factors.

Injuries are always a major risk factor.  But in the Eagles’ case, I think the risk is atypically large this year.  That doesn’t mean they’ll occur, but it does mean or ex-ante projection needs to account for them.

There’s much more to say around this topic, and I want to present a new version of the Depth Chart Over Time that will make potential injury risk more obvious.  For now though, believe in the Eagles this year, but recognize that the existence of serious downside potential (negatively skewed) results in a mean win expectation that is lower than some might expect (I’ll get a number on it before the season starts).  We can talk about median expectations some other time….