Writers of Pro Football Prospectus 2008

Most Recent FO Features

ThomasEar10.jpg

» 2013 Play-Action Defense

Are the best defenses against play action the best against regular passes too? How much impact does play action really have in an NFL game, and does it correlate from year to year?

16 Oct 2013

Scramble: The Power of Location

by Tom Gower and Mike Kurtz

Mike: I am reading a transcript of a Fangraphs chat from today. I mean, you end up with this: "I was made fun of by a homeless man for wearing an Ottawa jersey in Southern California." That is awesome and would never show up in any sane column.

Tom: And it now ends up on Twitter.

Mike: Ugh. Except that was the punchline from about 300 characters of setup because LANGUAGE.

Tom: Granted.

Mike: "Write about numbers with care. Don't overuse them. Avoid using numbers when you don't have to. When presenting numbers, especially complex ones, consider how they'll be read and pay some attention to formatting. Explain, concisely but coherently, any concepts that aren't immediately obvious." Good chat.

Tom: That sort of thing harder is when you're a stat site. I do not think it is useful to explain DVOA in every article on FO. I am, however, much more cautious on articles not on FO.

Mike: I think they have us pretty soundly beaten on number and bizarreness of stats, but yes.

Tom: Yes, we need more stats, especially bizarre ones. Unfortunately, I'm not coming up with a backronym stat for BELICHICK off the top of my head.

Mike: Haha.

Tom: Back ELusiveness In Cornerback ... I picked the wrong last name to start with. PAYTON is much better, maybe PAssing Yards To Open Numbers. Of course, for a stat like that, it would be better if we have more precise location data than we get in an NFL gamebook, where passes are listed "long" or "short" and "left," "middle," or "right."

Mike: The additions of which were heralded with much rejoicing. We really are in the Stone Age.

Tom: I've commented on this before, but my involvement with the old PBP transcription project has shown me we're ahead of where we were 20 years ago. I'm looking at a line from a 1989 box score right now, and it literally says "Moon passes incomplete to Jeffires."

Mike: Now I'm struggling to remember what came before the Stone Age. Whatever it was, it sure wasn't good.

Tom: I just know I probably wouldn't have survived long in it. Fortunately, we are starting to get more data on where on the football field things actually take place. See, for instance, these maps of where quarterbacks threw the ball in 2012.

Mike: That is a good first step, but it is definitely the first step.

Tom: Absolutely.

Mike: As the past 30 or so years of baseball analytics has shown us, once you scratch down a bit, you get more meaningful statistics, which lead to more statistics as your theory improves, and most importantly, your data collection becomes more sophisticated. It's hard to overstate the importance of PitchF/X, for instance

Tom: Which leads to the obvious question, what would the football equivalent of PitchF/X be, and what could we use it to learn?

Mike: Not only for the simple tools of identifying pitch type and location and speed, but the derived stats like O-contact and Z-contact that came once people had to think about how to use that data effectively.

Tom: Right. What could we learn if we had trajectory information on quarterback passes?

Mike: Hell, I was at PNC for the NL Wild Card game, and the scoreboard, after every pitch, not only had the pitch speed, but also the vertical and horizontal movement. Seriously, it was pretty awesome.

Tom: Nothing like what PitchF/X shows, of course, but we could figure out just how much some quarterbacks, like a young Michael Vick and perhaps Jake Locker, really were fastballing their short throws in there.

Mike: There are other things that our theoretical system could make available, also, like consistency of release position and velocity.

Tom: Yes. Another neat technological tracking thing would be the ability to watch from a player's perspective. We got a glimpse of that when Sam Bradford tried out Google Glass over the summer. (Yes, I know Chris Kluwe did so as well.)

Mike: I think the real issue with all of these is that football players are constantly moving. Very, very quickly.

Tom: That's one of the interesting things to consider. A pass is a singular event in a very dynamic play.

Mike: If the receiver and the quarterback were always in the same spot every time it would simply be a matter of positioning whatever fancy cameras we have at that spot and reaping the benefits.

Tom: You need full optical tracking, like you see in Europe for soccer matches.

Mike: So, how do those systems actually work? I've never really figured that out.

Tom: Technically? I don't actually know. I assume they have cameras and are able to identify and track players. But I never cared enough to learn the real details. Yes, football would be more complicated, both with the constant substitutions and players I think ending up in much closer physical proximity to other players than is normally the case in soccer.

Mike: To be fair, I just looked at the portion of STATS' site that purportedly explains how SportVU works, and it doesn't actually explain anything, either. There are cameras and algorithms and other buzzwords, and data happens. I'm guessing "magic."

Tom: I was looking at Opta and found the same thing.

Mike: Looking at the basketball section, they at least claim to have some interesting fine-grained data: 'Detailed player and team statistics, including speed/distance, shooting information, passing details, touch breakdowns (drive, elbow, post, etc.) and never seen before defensive statistics.' Leaving the marketing aside, if the system can differentiate between types of touches in a basketball game, it might be more sophisticated than I give it credit for.

Tom: I think the big thing will be the ability to track how players move. Where do safeties align, and how much do they move from place to place, say from deep to the box on run plays and from the box over a tight end to a deep corner in coverage on pass plays. Does an Earl Thomas really cover as much more ground relative to a lesser free safety as he seems to? We could start to quantify things like anticipation, noting when a player makes his break relative to when a receiver makes his break or a quarterback makes his throw. I'm probably imagining at this point -- these are complicated calculations and hard to measure.

Mike: Well, assuming that we have or will have magic cameras that can track individual body parts on individual players. That is the next question: what do we do with that data?

Tom: I think it depends on what kind of data we will be able to get.

Mike: Well, in this scenario we have all of the data. You touched on a few things. I actually think the biggest advantage might be automation. At the moment our charting project does not record things like which receivers are jammed or given free releases. Who is on or off the line, etc., because it is far too labor-intensive to catalogue for every play.

Tom: Right. A lot of this data is theoretically collectable right now. It's just extraordinarily time-consuming to do so and even harder to do with a great deal of precision.

Mike: I think really what we will see is that, much like baseball, football can be reliably broken down into two-man matchups when sufficient data is available

Tom: Uh...

I think individual plays will end up looking a lot like small matchups.

Mike: And individual plays are all we ever care about.

Tom: I think the question will still be disaggregating the specific matchups that are exploited in a particular play with all the other matchups going on on the field at the same time, and the matchups on the previous and subsequent plays.

Mike: Right, I think you're looking at it too narrowly.

Tom: Of course, that's not a new problem.

Mike: We're interested in all of those match-ups on every play, even the ones where the ball isn't involved.

Tom: All our stats are produced in the context of everything else. Tom Brady throwing to Kenbrell Thompkins, Danny Amendola, and Michael Hoomanawanui or Brady throwing to Wes Welker, Rob Gronkowski, and Aaron Hernandez? Which will lead to better numbers for Brady? This is just in my head because it came up in a radio appearance on Tuesday, and I'm not sure how much people not inculcated into the cult of FO think about these things.

Mike: Yes, but we can begin removing context with enough data. We have data on safeties when they become involved in a play either in coverage or as a tackler. We don't have the play before, a play-action they don't bite on and read the pass appropriately, leaving the quarterback with only a dumpoff.

Tom: Right, I think it'll be hard to disaggregate the context even with more data. Aside from maybe those teams that like to match up their corners basically every snap.

Mike: Well, that is in addition to every sample play-action we have from the league as a whole. And how every single safety reacted. And how those fake handoffs themselves differed.

Tom: This is starting to sound like a question that we won't answer by 2020 even if we start getting data next year.

Mike: I think you're selling all of this short.

Tom: I'm not trying to sell it short. I'm trying to think through it and not having as much success as I think I should have. I think it'll also take us a while to sort out the valuable information from uninteresting non-differences. Mark Cuban has alluded to some bad personnel decisions the Mavericks made based on some early usage of advanced statistics that ended up being the result of small sample size.

Mike: I think what I'm trying to say is that we will finally have enough data to provide real normalization.

Tom: Yes. I think it'll take a while to get from the mass of data to a real normalization teams can translate to good intelligence and the media can use to talk about games in a smarter way. Though of course we'll still get players talking about teams winning because they wanted it more, and there will be aspects of the game in which we still will not be able to prove them wrong.

Mike: It will absolutely take time. But I think we have a good guess at the future of football analytics: cameras recording every minute detail of every play and moving away from the play-by-play. It will take ages to get to the point where anyone except us cares. But, at some point, everyone will. And then it might be worth it!

Tom: And, just like today, the best team in the regular season will still not win the Super Bowl!

Loser League Update

Quarterback: "I met a quarterback with a passing arm named Smith." "What's the name of his other arm?" "Few Loser League points this week." Alex drew runner-up with eight points, while Geno Smith was low signal-caller of the week with 6 points on 209 passing yards and a pair of interceptions.

Running Back: As far as low scores go, 4 is a relatively high one. Plus, Fred Jackson, Rashard Mendenhall, and Ray Rice all may well have been in your fantasy football lineup this week.

Wide Receiver: Donnie Avery finished last this week with -2 points thanks to a fumble, but the most interesting statline of the week belonged to Danny Amendola, who had precisely zero receiving yards yet avoided the penalty. Your Scrambler was curious about just how rare this feat was. It is less rare than you think, but generally done by running backs. Including only players who had no more than two rushing attempts in that game, Amendola is the first to achieve that feat since Early Doucet did it in 2011. Robert Woods and Cordarrelle Patterson each matched Amendola's 0 points.

Kicker: Sebastian Janikowski and the left hash do not get along this season. Whatever the reason, his 51-yarder ended up uncharacteristically short and probably a bit wide even had it had the distance. Combine that with a made extra point and you get -1 Loser League points.

For where your team came in this week, see the Loser League results page.

Awardings of Fame and Misfortune

KEEP CHOPPING WOOD: The ideal KCW winner is a play by a player both harmful and funny. We do the best we can, but we rarely find an ideal combination of both. Terrelle Pryor's back-foot duck under pressure that set up Kansas City's go-ahead score in Oakland's loss Sunday was certainly harmful, but it was just a bad play by an inexperienced quarterback. By contrast, Brandon Weeden's second interception was funny. It was also harmful to his team, coming on first down (not third like Pyror's), and pretty devastating to his team, coming in the fourth quarter of a one-score game. In short, it was everything that made for a great KCW winner.

MIKE MARTZ AWARD: Why do teams with really good quarterbacks tend to coach conservatively late in games? As profiled in Clutch Encounters this week, Sean Payton is the latest coach to fall into this trap. The combination of the prevent offense with the prevent defense gave Tom Brady three chances to drive for the game-winning score in the final four minutes, and that is two too many.

Also, if we wanted to go back to giving out the Colbert Award, there was a surfeit of suitable honorees this week. Bill Belichick's decision to go for it on fourth-and-6 is the most obvious candidate, but the Jaguars actually played like an underdog with a fake punt call in the first quarter in their own territory, the Broncos ran their own fake punt later in the game, Bruce Arians went for two down 22-20 in the middle of the third quarter, the Ravens, Bears, and Seahawks all went for it on fourth down inside the five in the first half, and Ron Rivera actually went for it twice on fourth down on the same drive in the first quarter. It was as though sense finally started breaking out in the NFL.

Greg Schiano Kneeldown Locks of the Week

Tom: Sorry, Mike. Green Bay defeated the Baltimore Ravens, but by a mere two points. Meanwhile, the Buffalo Bills recovered from a 24-10 deficit to force overtime, covering the nine-point spread. In a reversal of our season-long trend, I was right and you were wrong. I'm sure it won't happen again.

Mike: My shame cuts like a knife.

Tom: As a reminder, all lines are courtesy of Bovada and were accurate as of time of writing. All picks are made without reference to FO's Premium picks.

Mike: Ooo, a pick 'em. I love pick 'ems.

Tom: They're tantalizing, just because you don't have to worry about screwy "we're fine winning this game by less than the spread" results. I don't blame you after last week.

Mike: Indeed. This one is particularly appealing, because one team matches up very well against the other.

Tom: The Redskins don't have a defense, so any team this side of the Jaguars matches up well against them? (Yes, the Redskins are 26th in defensive DVOA after this week, so they kind of have one.)

Mike: Yes, the Bears have a tendency to overpursue, but Robert Griffin is not nearly as mobile right now as he was last year, and probably will be later this year. There is also that. That is a big factor. Even Jay Cutler can't screw up a game against the Redskins. Chicago Bears over Washington Redskins.

Tom: There is another game this week where one team is favored by a single point, but I am not touching the Steelers again, on either side. Instead, there is one line that really stands out. As Aaron noted in this week's DVOA column, the Buccaneers are the best 0-5 team in the 25 year history of our numbers. The Falcons are probably still a better team, even with Julio Jones out for the year and all their other problems on offense. Mike Glennon stinks, but of course Atlanta is 31st in our defensive ratings. The Falcons should be favored by more than the home edge, but by more than a touchdown is a too tempting for me. Plus, if Tampa can screw it up, I have a ready-made scapegoat in the least popular coach in recent NFL history, Greg Schiano. Tampa Bay Buccaneers +9 at Atlanta Falcons.

Scramble Mailbag

Bill:: I could really use some advice for a 12-team, Head-to-Head, 1.5 PPR League: Aaron Rodgers and Darren McFadden for Tom Brady and Marshawn Lynch. I would be getting Rodgers. As I see them, the pros are the Rodgers already had his bye, and I could start a returning Steven Jackson and a resurgent Stevan Ridley, while McFadden stays on the bench. The cons are that Lynch is running really well and Rodgers' receivers are hurting. Thank you!

Mike: How strange is this year that Brady is the part of that deal that gives me pause?

Tom: Weird things happen in the NFL. We'll see about Danny Amendola's apparent concussion and Rob Gronkowski's questionable and ever-delayed return.

Mike: That said, Brady still has significant upside and Lynch is the real deal.

Tom: Yes, while Rodgers will be without Randall Cobb, who went on revocable IR, and James Jones. That diminishes his edge on Brady. Lynch is the real deal, while McFadden is clearly not.

Mike: I get the feeling that the dropoff from Rodgers to Brady will be much smaller than the gain moving from McFadden on your bench to Lynch as your RB1. Do not take the deal.

Tom: With Steven Jackson still not practicing on Tuesday and Ridley in a Patriots offense that will be week to week, you need the bellcow back in Lynch. As frustrating as Brady has been this season, that's just too much of a downgrade at running back for an upgrade at quarterback. No deal.

Posted by: Tom Gower on 16 Oct 2013

11 comments, Last at 18 Oct 2013, 11:41am by Mike Kurtz

Comments

1
by wr (not verified) :: Thu, 10/17/2013 - 2:37am

KCW should have gone to Rob Ryan for using a prevent defense.

2
by Guest789 :: Thu, 10/17/2013 - 2:48am

After he had already stopped Brady twice in the last 3 1/2 minutes, only to see his offense give the ball right back?

-----

“Treat a man as he is, and he will remain as he is. Treat a man as he could be, and he will become what he should be.”

3
by James-London :: Thu, 10/17/2013 - 6:43am

Can we see the individual player scores in the LL please? Atm all we see is the team total

Phil Simms is a Cretin.

4
by NateTG (not verified) :: Thu, 10/17/2013 - 9:00am

The trickiest part is probably working out which players are which. A 'worst case' scenario is that someone has to type in numbers for each substitution, but they can probably be OCRed pretty well.

Triangulating for position is relatively straightforward as long as the cameras have enough frame rate and resolution.

Football helmets tend to be roughly spherical, and single-colored which may make them well-suited for automated tracking.

AFAIK The raw SportVU basketball data only shows player position in 2D and ball position in 3D. I think all the other stuff is inferred from there. (There's some info about this in the rebounding paper that was presented at Sloan.)

Just like with the NBA data, it would surprise me a great deal if raw NFL SportVU info were made available to the public at large, since I expect that some relatively sensitive information could be mined from it.

6
by my name is my password (not verified) :: Thu, 10/17/2013 - 9:03am

Also inevitable for the teams to install biometric measuring devices to measure temp/heartbeat etc. for players, but that would assuradly (and properly) be tightly controlled.

5
by my name is my password (not verified) :: Thu, 10/17/2013 - 9:00am

SportVU or an equivilant will be installed at all fields, I think its inevitable. At the least, it will automate the aggregation of data that I imagine all the teams are doing manually (or paying a service to aggregate). The obvious issue, that this site is well aware of, is one of sample size - baseball and basketball have orders of magnitude more plays per year.

I don't know how we, as an amateur public, develop sabermetric-type information that would allow for an objective evaluation of NFL players, given the error bars inherent with the sample size. However, we could get information that describes the past, and do it in a better method than current metrics.

(for the uninitiated, SportVU puts cameras EVERYWHERE, and tracks players/ball to a minute degree, and the resulting (large) data file can then be analyzed. Search Grantland archives for a good primer on how it works in the NBA).

7
by Snarky Name Here (not verified) :: Thu, 10/17/2013 - 12:15pm

I seem to have cornered the market on pretty good receivers with lousy QBs (and Harry Douglas). Pick two; at least one must be a receiver.

Steve Smith (at STL), Stevie Johnson (at Min), Andre Johnson (at KC), Andre Ellington (at Sea), or Harry Douglas (at TB). I'm thinking the Steves.

8
by Mike Kurtz :: Thu, 10/17/2013 - 12:32pm

The Rams are actually worst in the league against No. 1 receivers, so Steve Smith, Esq. is in.

Your other two options are actually pretty good, as Minnesota and Kansas City are both slightly above-average against top receivers, but both of them have to deal with the aforementioned QB issues. Kansas City's defense is better than Minnesota's, so I think you're right going with the Steves.

10
by Snarky Name Here (not verified) :: Thu, 10/17/2013 - 1:22pm

Thanks! One more question - Titans at SF or Bengals at Det? I'm thinking Cincy, if Megatron is out and Titans otherwise.

12
by Mike Kurtz :: Fri, 10/18/2013 - 11:41am

Reports of the Niners' demise are greatly exaggerated. Take the Bengals and hope Megatron doesn't play.

9
by Walshmobile :: Thu, 10/17/2013 - 12:49pm

Thanks for adding it so we can see who our players are for Loser League. I had totally forgotten who most of my team was