≡ Menu

Is Arian Foster declining?

[House-keeping note: I’ve added the Salary Cap Calculator to the gray header tabs at the top of each page, so you can now easily get there no matter what page you’re on at Football Perspective.]

A quick look at Arian Foster‘s statistics over the last three years paints a picture of a player in decline:

2010*+ 16 327 1616 16 4.9 101.0 20.4 66 604 9.2 2 2220 18
2011* 13 278 1224 10 4.4 94.2 21.4 53 617 11.6 2 1841 12
2012* 16 351 1424 15 4.1 89.0 21.9 40 217 5.4 2 1641 17


Foster’s declined in rushing yards per game and yards per carry over the last two years, while his value in the receiving game fell off a cliff in 2012. One could reasonably conclude that Foster simply isn’t the same player he used to be, and that he could drop off even more in 2013.  But while the traditional statistics tell one story, what do the advanced metrics say?

Peterson and Foster trade secrets

Peterson and Foster trade secrets.

According to Football Outsiders, Foster had an outstanding 52% success rate in 2010, but he was successful on only 44% of his carries in 2011.  And while his yards per carry average dropped in 2012, his success rate increased to 47% in 2012. That prompted this Nate Dunleavy to argue that based on his success rate, Foster improved in 2012.

On the other hand, his DVOA did take a slight dip, from 2.3% above average in 2011 to -1.6% below average last year. There’s no positive way to spin Foster’s receiving numbers, either. His targets didn’t drop that significantly in 2012, but he produced a pitiful 217 yards. Some of the explanation for the low yardage totals could be that Matt Schaub was checking down to Foster more frequently last year, but that wouldn’t explain why Foster (1) rarely broke a tackle and made a big gain or (2) stopped being an intentional part of the passing game.

Brian Burke’s statistics paint a more pessimistic view. According to Burke, Foster’s Win Probability Added was 1.14 in 2010, 0.14 in 2011, and 0.02 in 2012; even worse, his expected points added has dropped from 52.5 to 6.9 to -23.7 over the same period. Burke’s definition of success rate includes targets, so according to Burke, Foster has declined by that measure as well. Burke views the majority of runs as unsuccessful, so he’s graded Foster’s success rate at 45.8% in 2010, 41.5% in 2011, and a dismal 37.8% (no doubt aided by his poor receiving numbers) in 2012.

But there is some evidence to indicate that Foster the player isn’t declining at all – after all, Burke and Schatz would be the first to tell you that the numbers for Foster are heavily influenced by factors outside of Foster’s control. Pro Football Focus actually argues that Foster has been improving each year. PFF graded Foster at +3.2 as a runner and +10.6 as a receiver in 2010, +6.6 as a runner and +5.6 in the passing game in 2011, and then +9.1 as a runner and -1.5 as a receiver last year. Here’s one explanation for those outlier numbers in the face of diminishing traditional and advanced statistics: PFF ranked the Texans as having the third best run-blocking offense in 2010, the 7th best in 2011, and the 13th best last year.

On the left side of the line, Duane Brown, Wade Smith, and Chris Myers have been constants over the last three years. But the right side has changed significantly since 2010 and 2011, when Mike Brisiel, Eric Winston, and Joel Dreessen blocked for Foster. At right guard, Houston went first with Antoine Caldwell and then Ben Jones, but neither were particularly effective last year. The dropoff at right tackle was even more severe, as Derek Newton wasn’t able to bring anywhere near as much to the run game as Winston. And while Dreesen wasn’t a huge part of the offense, he was a better blocker than James Casey or Garrett Graham.

So are Foster’s traditional numbers declining because the right side of the line has weakened? Not necessarily. The table below shows Foster’s yards per carry on all runs over the last three years, broken down by the direction of the rush attempt per the play-by-play description:

Yearleft endleft tackleleft guardup the middleright guardright tackleright end

Foster has actually run better to the right side all three years, so I’m not sure how much we can blame the change in personnel. More worrisome would be his struggles up the middle. On 125 carries behind one of the guards or up the middle in 2010, Foster gained 671 yards, for a 5.4 yards per carry average. That dropped to 4.3 in 2011 on 122 carries, and then an abysmal 3.3 average on 137 runs last year.

We can also break down what percentage of Foster’s runs came in which direction:

yearleft endleft tackleleft guardup the middleright guardright tackleright end

The Texans did call fewer runs to the right last year, and wisely are running more frequently behind their All-Pro left tackle. From the data, it’s tough to say whether the play-calling, Foster himself, or the lineman are responsible for the decline in production over the last couple of seasons. Part of it is undoubtedly regression to the mean, as expecting Foster and the Texans to continue to maul teams in the running game was probably unrealistic.

  • Richie

    I’d be curious to see how common it is for a RB who averages 75+ yards/game to have his yards/carry decline while his success rate improves.

    eg; Is it fluky for a guy to have a worse average but an improved success rate? Or the other way around?

    I think success rate is a cool stat, but I just haven’t seen in context enough to get an idea of what it really means. I’d like to see it shown as a graphic during NFL broadcasts. “Today, Arian Foster has rushed for 115 yards, and has been successful on 74% of his runs.”

    • Richie

      (I wish this site gave me a grace period to edit my posts.)

      We are so conditioned to think of football performance in fantasy football terms (mainly: counting stats). But long before Football Outsiders, I used to see players have a great game in counting stats, but it felt hollow to me, because the guy rushed for 175 yards, but had 2 long runs and a bunch of 1-yard gains in the middle of the line of scrimmage.

      • Chase Stuart

        My guess is that success rate and YPC are only moderately tied together. It never dawned on me until this very moment, but they’re probably like completion percentage and Y/A.

        • George

          Just a random thought, with a large enough data set and a certain level of criteria (e.g. say over 100 attempts and over 4 YPC and maybe 5 rushing TD’s) I’m sure you could probably get some value out of running a regression from this and seeing what you come up with (hopefully a reasonable R2 value etc.). Or even chuck a set of variables in, maybe and see what you come out with (I’m trying to think of reasonable and related things that you could put in – I suppose p values would sort this out etc. but you know where I’m coming from, e.g. if you are using YPC you may as well take yards and carries out).

          • Chase Stuart

            I don’t think you could get very accurate results on a regression (although that doesn’t mean one shouldn’t be run just to prove that to be true!).

            Again, I think it would be like putting in 200 pass attempts and over 7.0 Y/A and seeing if you could predict completion percentage based on Y/A. You couldn’t. The one factor you could add would be rushing first downs, but that’s not widely available on the individual level, either.

            • George

              Fair point – I hadn’t seen the pass attempts, y/a one but I know where you are coming from. Sometimes having the data and what would seem like sensible variables doesn’t mean you will get an answer (and you end up almost trying to fit the data to solve the problem, rather than seeing what it actually tells you which defeats the object). I’m not sure how the success metric is measured (I’m assuming it is gaining positive yards) but given the variables you’d almost need a constant (like yards gained from 1st and 10, between the red zones) to eliminate other issues (such as down and distance and potential scoring opportunities etc.). Great work as usual though.

              • JeremyDe

                This is from Brian Burke’s advancednflstats.com page.

                Run success – “A success would be: On 1st down–a gain of 4 or more yards; on 2nd down–a gain that at least halved the distance to go; and on 3rd down–a conversion for a new set of downs.”

                • George

                  Thanks. I found how they calculate it on Football Outsiders and the good thing is they allow for down and distance as well, and also accounts for score in the fourth quarter (e.g. dependent on what is on the line – but I suppose then you have the issue of playoff games). This would pretty much remove some of the bias that just “a did the run make a gain” stat would have.

                  They pretty much run with the same values as Burke (40% of yards on first down, 60% on second down, 100% on 3rd down – so 4 yards on 1st and 10 = 1 yard on 3rd and 1). Those switch to 50/65/100 if down by 7 in the fourth, or 30/50/100 if up by any amount in the fourth (to allow for running the clock down).

                  Basically anything over 50% is regarded as good so you could argue Foster is regressing to the mean as Chase does. Another way of looking at it would be to suggest that the Texans play calling has improved (or become more varied) and Schaub has improved perhaps (or maybe what his receivers are contributing improved) which means that maybe Foster isn’t such a big part of their offense anymore resulting in the reduced DVOA perhaps? That could also be caused I guessed by the decrease in the strength of the run blocking (as Chase mentions), and teams lining up more against the run (as they had doubts against Schaub to beat them with his arm – drop in DVOA and QB Rank as evidence of that or maybe that they were too reliant on Foster).

      • Chase Stuart

        And regarding your grace period point, if there’s a way to do that on wordpress, I’m happy to implement it.

        • Richie

          I don’t know. I know some blogs have it. Not sure if it’s a wordpress option.

    • James

      There’s some significant debate over what constitutes a success. Brian Burke at ANS would tell you it’s a play where the offense’s chances of scoring that drive have increased, which is something like 6+ yards on 1st and 10, 80% of the distance on 2nd down, and a conversion on third down. Meanwhile, FO somewhat arbitrarily decided it was 4+ yards on 1st, 50% of the distance on 2nd, and a conversion on third. Since most runs only average 3-4 yards, you will have significant differences in success rate between the two sites.

  • Seb

    Thanks for the article, great read. But I am curious as to what conclusions you make of it, regarding both the relevance of these metrics and Arian Foster’s true production. Does that mean YPC is not a good measure of success for a RB? Or is success rate largely independent of a RB’s Expected Points Added? I would assume, perhaps naively, that counting targets only works to make Brian’s metric a tiny bit more accurate than Aaron’s. But then should we really consider that Foster had a negative contribution to his team’s offensive production in 2012? Seriously? One doesn’t have to be a Texans fan to realize Arian Foster, somehow, somewhere, offers more value than that, right? So in the end, what does that say about the metrics: are they just flawed, or simply misinterpreted, or could it be that external factors (linemen performance, but also run/pass balance, passing efficiency, strength of opposing defenses against the run, field and weather conditions, etc.) all have too big of a weight in Foster’s apparent production to make any of these so-called advanced metrics really relevant? Very curious to read your opinion on this. Thanks in advance.

    • Chase Stuart

      I’ll start by noting that if you post with a fake e-mail address you won’t get notified when people reply to your comments 🙂

      You raise some complicated questions.

      — I do not believe YPC is a good measure of success for running backs.

      — I think success rate is probably decently correlated with a RB’s EPA, but there will certainly be outliers. I would suspect it is more correlated with EPA than YPC.

      — I do not agree with anyone that says Foster had a negative contribution to the Texans’ offensive production in 2012. With respect to Burke, we simply disagree on the value of running games. I find myself pretty far on the continuum that the passing game trumps all, but Burke’s view (which he often writes about) is even more extreme. As for FO, I think grading RBs is tricky. I prefer to view carries as indicators of talent instead of opportunities used, the way people like Schatz and Burke do.

      This, however, is a very complicated issue. I tend to think players like Curtis Martin and Eddie George were very good players and valuable running backs, but my guess is they would not fare very well in EPA or FO (I’m sure FO has data on both Martin and George, but I am not aware of what it says).

  • Seb

    Thanks for the reply. I do know it’s a complicated issue, that’s why I was asking you for guidance! But I also believe that challenging the relevance of these metrics is key to their success and our gaining a better, deeper understanding of the game and individual performances. I understand you only find marginal value in EPA and just slightly more in FO’s.

    As for the email address, it is perfectly valid, thanks for asking. More seriously, you can check this link – I do have to add that it dates back to 2005… 😉 http://lifehacker.com/144397/instant-disposable-gmail-addresses

    • Chase Stuart

      That just blew my mind.

      I had a long series of offline debates with Doug Drinen and Jason Lisk on how to grade running backs. I never quite put pen to paper exactly how I feel (or perhaps I never quite got comfortable deciding how I feel).

      I do know the argument centers around something like this:

      — The 1998 Jets were 5th in points, 4th in yards, and 3rd in first downs. They were by any measure a really good offense.
      — Curtis Martin was one of, if not the, best players on the 1998 Jets offense.
      — Curtis Martin rushed for just 3.5 YPC that year.
      –The Jets, with one of the best offenses in the league, put the ball in Curtis Martin’s hands on 39% of their plays. If you want to look at Martin’s YPC and say he was bad that year, I find it hard to reconcile that fact with the way the Jets used him.

      When a team leans so heavily on a player, I find that much more persuasive than his rate statistics.

      • Sunrise089


        Obviously football stats are complex. Obviously there’s tons of room for analysis on the margin. But…my sense is we need to give the simplest valid stats a lot of weight. The very simplest football ‘stat’ is counting, but since we’re not an in-game ‘analyst’ we can at least proceed to rate stats. And…a rate stat of 3.5ypc is terrible. Sure there might be mitigating factors, but they have to be pretty darn strong for me to overlook the ability to get multiple extra yards per play by not forcing the ball to the rb.

        Talking about how Martin was the Jet’s best player isn’t persuasive at all, to me at least. He might have been great, but it also looks like he was very overused. Likewise ‘team trust’ doesn’t tell me much when, in in-game terms, most NFL staffs aren’t too strong on the strategy front.

        It feels like most sports stat fans get that NBA points per game is a worse tool than even something simple like shooting percentage, or MLB home runs are offset if your OBP is .250. Running backs with a mix of low YPCs, negative EPA or WPA, and low success rate…pretty much seem like the same thing.

        I know there are exceptions – your work on receiver usage and Marshall’s value this year. Or Jordan’s shooting percentage versus his teammates’. Those are exceptions though for players operating at a very high level. In the martin case…I guess I’d be curious if you’d defend a WR with 150 catches for 750 total yards or a QB with 700 pass attempts for 3,000 yards on the ‘team confidence in them’ metric. If not I respectively wonder if at least a PART of your position is due to the same attraction (subconscious?) to counting stats and old school football that the mainstream guys love above all else.

        • Chase Stuart

          Some really good comments, Sunrise. Let me take them in order.

          1) A 3.5 YPC average is very bad in the abstract. But wouldn’t mitigating factors include being the focal point of a top-five offense? I mean, if you asked Bill Parcells in 1998 what he thought of the job Curtis Martin was doing, I don’t think he’d say “well his YPC sucks, so he sucks.” I think a poor YPC average on a bad offense is a different animal than a poor YPC average on a good offense, or at least is in some circumstances.

          2) I don’t know how much I can agree with this, when the Jets were 12-1 in games started by Testaverde (FWIW, Martin had 45 carries for 116 yards in 2 of the 3 non-Testaverde games, and missed the other due to injury). It would be nice to see quarter-by-quarter splits for that season for Martin.

          Here are some splits:

          In ’98, Martin had 231 first half carries for 844 yards (3.7 YPC) and 2 TDs in first halves. If the second halves of games, he had 187 carries for 581 yards (3.1 YPC) and 9 TDs. He also had 25 carries in the last two minutes of halves for 64 yards (2.6) and no TDs.

          He had a 3.3 YPC average in the first quarter, 4.0 in the second, 3.3 in the third, and 2.9 in the 4th. He had a 2.8 YPC average on 48 carries in the 4th quarter when the game was within 7 points.

          3) I’m not going to comment on the other spots examples. I think a WR with 150 catches for 750 total yards would be odd, but it’s conceivable that such a player could be valuable if he had like a 90% catch rate. At that point, he’s basically a RB.

          A QB with 700 pass attempts for 3000 yards would not be good in any way. Passing and running are fundamentally different. If you pass 700 times it’s because your team is losing, so it doesn’t have much to do with confidence.

          4) As someone who thinks like an economist from time to time, a revealed preference for a player is a big indicator of talent for me. When it comes from a HOF coach on a very good time, that just adds to it. Ditto Fisher and Eddie George. It’s not like they were looking to replace these guys after their big workload seasons, either. It was part of the plan.

  • Seb

    Definitely agree. And I would think the same of Arian Foster, he might not be as great as he was in 2010, he’s still certainly very good to be such an integral part of the Texans offense. But that does not say whether he is or isn’t declining (could be declining and still be among the top 10 backs). Ultimately, deep down, would you say he is? Or is the answer simply that we have no real way to tell? 🙂

  • sn0mm1s

    I loathe FO’s success rate and, in general, the way they measure RB stats.

    Now, their system is proprietary, so there is no way for me to nitpick specifics because the entire system is a black box. But here is a possible reason that Foster’s success rate is higher this past year than the prior year – he played in more winning games. FO sort of made their name off of the debunking of the “establishing the run” mentality but they then lower the threshold for a successful run if you are winning late in the game (I don’t remember the specifics and I am pretty sure what they have in their almanac isn’t an accurate reflection of their formula).

    Burke’s methodology is much more convincing in my opinion (now if he could only separate rushing vs. receiving plays). That 4 Yard gain on 1st down that FO uses was based off the Hidden game of Football IIRC. Burke’s data shows that a RB needs 5+ yards on 1st down to make the likelihood of converting another set of down greater than they were the previous play (assuming 1st and 10). So, FO’s baseline doesn’t have much credibility. In fact, I believe that they said at one point they use 4 yards because 5 yards on 1st down doesn’t happen nearly as often so they thought it was a high benchmark. To me, that just illustrates how the running game isn’t that important and teams run far too much.

    The thing that bugs me the most is that many of their stats don’t even pass the eyeball test – especially regarding players that break off big plays like Adrian Peterson, Chris Johnson, Barry Sanders etc. etc. In 1997, FO’s stats imply that a replacement level RB for the Lions would’ve rushed for over 1600 yards. In fact, throughout Barry’s career their stats imply that a replacement level RB would be *very* good rushing for the Lions (and not good for the Broncos, Cowboys, and many other rush heavy teams). The funny thing is, anyone you put on the Broncos was effective and the year after Barry retired the Lions didn’t even match (as a team) what Barry ran for in 1998 (and their stats suggested a replacement would’ve run for *more* yardage than Barry in 1998).

    I can put up some other nonsensical results but I generally don’t look at FO’s stats as being good indicators of anything when it comes to individual players (for teams they are great).

  • Arian Foster is a talent player. He had perform excellent but now we have seen hie performance graph has decreased. It is a bad sigh for him. I wish he will improve soon.