Writers of Pro Football Prospectus 2008

Most Recent FO Features

BarrettJT16.jpg

» SDA: Rivalry Showdowns

Rivalry week has significant conference and Playoff ramifications. Should Alabama, Mississippi State, Oregon, or Florida State be worried about getting upset by their rivals?

26 Aug 2005

Dropping Some Knowledge on Seattle

Guest Column by Kevin Pelton

Living in Seattle, it's been impossible to follow the NFL the last two seasons without hearing about the Seahawks' problems with dropped passes. And hearing some more. And a little more after that.

The Seahawks infamously finished second in the NFL in drops in 2003, a surprise for a talented unit that boasted Darrell Jackson (who signed a lucrative new deal after the season), former top-ten pick Koren Robinson, and Football Outsiders favorite Bobby Engram. The problems were written off to focus and confidence issues and a host of other excuses.

But 2004 came and went, and the problem continued. Jackson tied for second in the league in drops, and Robinson would have topped him had he not been so busy getting suspended. The situation culminated in the usually reliable Engram dropping a pass in the end zone that would have sent the Seahawks' Wild Card playoff loss against the St. Louis Rams into overtime.

The jokes came fast and furious, with Tuesday Morning Quarterback replacing "Blue Man Group" as the Seahawks' nickname with first "Dropped Passes Group" and later "Folgers -- good to the last drop."

Lost in all the attention paid to the drops was the fact that the Seattle secondary allowed St. Louis to throw for 313 yards and a pair of touchdowns in the playoff loss.

Was too much made of the Seahawks drops? It seems a fair question to ask. By Football Outsiders' advanced metrics, the Seahawks finished with the NFL's third-best offense in 2003 and were 13th even in a down 2004, so it's not as though the drops rendered the offense ineffective.

The problem with focusing on drops in a statistical sense is that they are merely a visible failure on the part of the receiver. By contrast, no stats are kept on times the receiver misjudges the ball and doesn't catch it or doesn't get open in the first place.

A rough analogy I've long had in my head compares drops to strikeouts for hitters in baseball. Conventional wisdom long held that hitters who struck out frequently were not valuable, but statistical analysis has demonstrated that strikeouts are not substantially different than groundouts or flyouts.

What do the numbers say in football? With the help of the FO data, I set out to investigate.

The first step is to evaluate exactly who drops the ball most frequently. Drops aren't an official stat, but they're tracked by STATS, Inc. (Of course, there's no way to get the data without clicking on every single wide receiver's individual page, and drops are not listed for tight ends and running backs unless they are in the league's top 20 overall.) Their 2004 leaderboard (for wide receivers only) looks like this:


2004 Drops Leaders
Player Team Drops
Chad Johnson CIN 14
Darrell Jackson SEA 11
Donald Driver GB 11
Az-Zahir Hakim DET 11
Koren Robinson SEA 10
Keary Colbert CAR 10
Jimmy Smith JAX 10

Bad Seahawks memories flooding back ... Anyways, this is a pretty misleading when you consider Chad Johnson was targeted for 169 passes, Az Hakim just 57. A better way to evaluate drops is in terms of percentage of targeted passes. Doing so gives us the benefit of better being able to sort out the best players. I've calculated "drop percentage" for every receiver with at least 50 targeted passes last season, and here are the best and worst in terms of hanging on to the ball:


Highest Drop Percentage, Top 5
Player Team Target Drop Drop %
Az-Zahir Hakim DET 57 11 17.5%
Koren Robinson SEA 67 10 14.9%
Darius Watts DEN 53 6 11.3%
Antonio Bryant CLE* 72 8 11.1%
Keary Colbert CAR 92 10 10.9%
*Bryant's stats are from Cleveland only


Lowest Drop Percentage, Top 5
Player Team Target Drop Drop %
Deion Branch NE 51 0 0.0%
David Givens NE 106 1 0.9%
Torry Holt STL 137 2 1.5%
Jabar Gaffney HOU 68 1 1.5%
Michael Clayton TB 122 2 1.6%

In a result that is no shock for those not named Matt Millen, Hakim was the league's most frequent dropper last season. The much-maligned Robinson ranks second, followed by a pair of players who were not on the first list.

At the other end, Deion Branch -- who also caught 68.6% of passes thrown to him last season to rank fifth in the league -- is certifiably the NFL's most sure-handed player.

It's worth noting that Johnson and Jackson, one-two in the league in total drops, ranked 10th and 16th in the NFL, respectively, in drop percentage.

As a first cut at looking at how important drop percentage is, I compared it to players' catch percentage to see how well it reflects all passes not caught by the receiver, whether "catchable" in STATS' opinion or not. The correlation coefficient between the drop percentage and catch percentage is -.229, indicating that players who drop the ball more frequently tend to catch fewer passes. The relationship, however, is not particularly strong. Recall that a correlation coefficient near zero indicates no correlation, while a value close to 1 or -1 indicates strong correlation.

Perhaps this makes more sense graphically, so take a look at the chart at right which compares drop percentage and catch percentage.

On the other hand, if you look at the top and bottom five lists, the players who drop the ball least often average a 63.2% catch percentage, as opposed to 53.7% for the frequent droppers. That's a pretty significant difference.

Presumably, players like Hakim and Robinson are playing despite questionable hands because of the difficulty their speed presents for opposing defenses. Because of this, it makes sense to look at how well drop percentage correlates with overall receiver performance, as measured by DVOA. (DVOA explained here.) In this case, the correlation coefficient is even lower (-0.055), which indicates almost no relationship whatsoever.

The corresponding chart shows how weak the relationship is; the receivers are distributed in something approximating the circle familiar to anyone who's taken a statistics class as an example of variables with no relationship.

It's also interesting to note the regression line on the chart, which indicates even if there was a relationship, it wouldn't mean much. The difference in projected DVOA between Hakim and Branch, based on their drops, is less than 10%. (Oddly, both players were dramatically better than their projection. That means something, but I'm not sure what.)

The last test is to look at drops at the team level. The Seahawks (7.1%) did finish second behind the Jacksonville Jaguars (8.2%) in team drop percentage last season. Why didn't the Jaguars' drops get more attention, you ask? Probably because the team wasn't that good -- but also because nearly a quarter of Jacksonville's drops came from running back Fred Taylor, who's not expected to be as sure-handed as a receiver. (Taylor's drop rate would, to go on a tangent of a tangent, rank second only to Hakim were he a wide receiver.)

The NFL's top three passing teams -- Indianapolis, New England, and Minnesota -- ranked first, second and fifth in lowest team drop percentage. Between them, however, were Tampa Bay (1.7% passing DVOA) and Arizona (-22.4%), while fourth-rated Kansas City was fifth in highest team drop percentage. The league's worst passing team, Chicago, was above-average in avoiding drops.

Again, there's a relationship, but it's a weak one (-0.225).

After looking at the data, I don't think my original analogy is accurate. There is clearly some harm in dropped passes. A better baseball analogy -- at the risk of forcing it -- appears to be errors. Like errors in baseball, dropped passes are harmful, but they're relatively infrequent, making them less important than other factors (as range tends to be more important in baseball). Also, like errors, drops tend to be more visible than these other mistakes, making them stand out to the hecklers in the stands and causing them to be overrated.

Lastly, what of my Seahawks this season? The statistics indicate improvement is likely. Seattle replaced the biggest offender, Robinson. Combined, the Seahawks' two new receivers (Jerome Pathon and Joe Jurevicius) and Engram, who will be asked to step up, dropped only four passes in 156 targets last season. (Jurevicius did not have a single drop in 37 targets.) If all goes as planned, by midseason we should be back to TMQ having only the Seahawks' garish monochrome home jerseys to pick on.

Kevin Pelton is a basketball writer by trade who is employed by the Seattle SuperSonics. His work appears on SUPERSONICS.COM and storm.wnba.com. He frequently dropped passes in flag football at the University of Washington. If you are interested in writing a guest column for Football Outsiders, something that takes an unconventional look at the NFL, drop us a line at info-at-footballoutsiders.com.

Seattle fans: Make sure you also check out our thoughts on your team in Four Downs: NFC West from earlier this week, and listen to FO editor-in-chief Aaron Schatz tonight at 8:05pm Pacific on Sports Radio 950 KJR.

Posted by: Guest on 26 Aug 2005

31 comments, Last at 05 Sep 2005, 12:48am by Kevin Pelton

Comments

1
by Rachel Wacholder (not verified) :: Fri, 08/26/2005 - 3:49pm

The Seahawks should have signed me.

I will lay out for every last ball...

2
by Björn (not verified) :: Fri, 08/26/2005 - 4:06pm

But what about drops that occur in clutch situations? Just like errors in baseball, not all drops are created equally. I would say that Bobby Engram's TD drop stands out as one of the biggest drops of the year. Aside from that, third down drops are more important than drops on second and one. But I do realize that the only way to collect this data would be to watch tape of every game. And I swear, if I win the lottery, I will do so.

3
by Lance (not verified) :: Fri, 08/26/2005 - 4:27pm

"Why didn’t the Jaguars’ drops get more attention, you ask? Probably because the team wasn’t that good"

Didn't the Seahawks and the Jaguars have the same record?

4
by Chuck-O (not verified) :: Fri, 08/26/2005 - 4:30pm

I enjoyed the article. The error analogy makes a lot of sense intuitively. I’d be interested to see drop percentages across positions… I’d guess that running backs and tight ends fare better than you’d think against wide receivers in terms of drop percentage (of course there’s no accounting for more difficult routes or coverage by a linebacker vs. a CB). As a Giants fan, I’d like see where Jeremy Shockey ranks… I’ve read a lot of discussion on his drops last year. It’d be interesting to see where he actually stands.

5
by B (not verified) :: Fri, 08/26/2005 - 4:35pm

The Jags did have the same record as the Seahawks, but the strength of the Jags was the defense. The strength of the Seahawks was supposed to be the offense, so problems they had on offense got more attention. Also the Seahawks threw the ball a lot more, so they had a higher number of total drops, which attracts more national media attention.

6
by noahpoah (not verified) :: Fri, 08/26/2005 - 4:39pm

Oddly, both players were dramatically better than their projection. That means something, but I’m not sure what.

It looks to me like it just means that the regression model is a lousy fit. After all, r = -0.055, and R^2 = 0.003. This means that 0.3% of the variance in DVOA can be accounted for by drop percentage.

I imagine you could find another pair of wide receivers, one who drops lots, one who drops few, both of whom had much lower DVOA than this model projects.

7
by dman (not verified) :: Fri, 08/26/2005 - 4:55pm

I think a better stat would be percentage of passes caught targetting the player, not just drops. Of course that penalizes receivers with poor qbs but there are starters in the league with absolutely horrendous caught/targetted percentages compared to their teammates. Burress and Toomer come to mind. But I guess since they're on the same team they'll get to suck equally and blame it on eli manning.

8
by Aaron (not verified) :: Fri, 08/26/2005 - 5:03pm

Well, we do keep that stat if you look at our wr sheets (link).

9
by Bright Blue Shorts (not verified) :: Fri, 08/26/2005 - 5:47pm

In my mind this is one of those issues where stats do not show the whole picture. What about what they do to team morale?

Listening to the Raiders games in 2003, I realised after we lost Gannon and Garner how important it is for an offense to avoid turnovers. With them, in 2002 Gannon had thrown just 10 ints in 600+ attempts and Garner had touched the ball over 300 times without a fumble. Ok, occasionally the offense would get stopped and we'd punt but that would force the opposition to play the long field. We were never giving them an easy game. In 2003, when Tui/Mirer/Martin/Johnson started throwing ints, and Fargas dropped the ball on the 1-yd line at least twice; any hopes I had went out the window. We'd finally be putting a drive together then the ball would be turned over and all the good work would be undone. Or even a penalty committed.

I'd think it's difficult even for professional athletes to keep believing and avoid losing heart when they watch their teammates making an error that could so easily have been a positive.

BBS :)

10
by RowdyRoddyPiper (not verified) :: Fri, 08/26/2005 - 5:55pm

"It looks to me like it just means that the regression model is a lousy fit. "

To put it in a more constructive way...the meaning in that would be that dropped passes by themselves are not a good predictor of DVOA, they need to be considered with other variables.

I can think of two ways to investigate this. 1) Replace the drops with successful catches in the player's stats and recalculate their DVOA. This should account for Bjorn's (sorry, I can't get umlauts on this keyboard, if you could actually tell me what those are called in swedish I'd appreciate it) point that all drops are not created equal.

2) You could build a model for receiver performance and compare your adjusted R^2 when drop %age is in the model versus not in the model.

I think #1 is the more straightforward way to go about things, since DVOA is a pretty good metric of success and it is a number that you can calculate exactly.

11
by Björn (not verified) :: Fri, 08/26/2005 - 6:23pm

I really don't know what they are called in swedish. When I was very young, I got in the habit of calling them "two little dots over the o," and I really have just been doing that ever since. Umlaut is plenty exotic enough for me.

12
by zach (not verified) :: Fri, 08/26/2005 - 7:01pm

you could just copy and paste it from his name, like this: öööööööööööööööööööööö

13
by Reinhard (not verified) :: Fri, 08/26/2005 - 9:00pm

Like when your keyboard is missing a letter or something...

I think the hurt of the dropped pass is when you lose that third down, which can make then make such a large dent in total offensive performance. (ie Chargers offensive resurgence effect in reverse)
Also, if that player had stayed in and blocked, or just ran straight at a safety or something, you figure he would have helped his team more than dropping.

14
by Colin (not verified) :: Fri, 08/26/2005 - 9:59pm

Sure, drops are a bad thing, but we have to be able to blame drops on the quarterback and on the offensive line too. That way we'll know if the receiver is hurting his team or if he is doing a good job in a bad situation. If the quarterback has good protection, and good mechanics, and good touch then a worse receiver should look better. So the colts and the pats don't drop passes. Well both their quarterbacks are extremely accurate touch passers who have amazing pass protection so they're never throwing off their back foot, etc. Maybe we should consider that Matt Hasselbeck throws a difficult to catch pass. A better metric for this would probably be found by a method similar to that guy who bills himself as the football scientist. Look at the film and see how many passes that were in the right area and weren't thrown overly hard were dropped.

15
by Larry (not verified) :: Sat, 08/27/2005 - 2:38am

I think two things could be investigated to answer Colin's questions without resorting to quiting our day jobs:

1) Do receiver drops correlate year-to-year for individual WRs?

2) Do drops correlate with QB, rather than with WR and to what extent.

3) Look at receivers switching teams and their drops.

The effects might be small, this could be like looking for clutch hitting, but with much smaller sample sizes. So, perhaps a wild goose chase. But #1 and #2 would be interesting, if just to confirm the CW, which, for all we know, could be wrong.

Also, I'd be very, very wary of correlating any stats to DVOA and looking for meaning. Mostly that will reveal how DVOA is calculated and little else, I think. It might reveal 'indicator' stats that are symptoms of some larger skill, like some believe strikeouts reveal a pitcher's ability to lower BA on balls in play, but I think in the case of drops that's unlikely. This doesn't mean DVOA isn't very, very valuable, just that, as a very highly derived number, trying to find correlations to it doesn't strike me as a useful excersize.

Some may disagree, and they should certainly air their arguments.

16
by Larry (not verified) :: Sat, 08/27/2005 - 2:39am

OK, two things became three things. Maybe proofreading would be good.

17
by fnor (not verified) :: Sat, 08/27/2005 - 5:07pm

I tend to agree with the down idea... drops on 3rd down are a pretty big setback. Drops within 5 yards of the LOS are pretty bad too. The former might help to judge reliability and the latter for... well... doing their job. Sure, there are some tough catches within 5 yards, but with the cushion WRs are getting nowadays, those should all be in the bag.

18
by Vern (not verified) :: Sat, 08/27/2005 - 6:42pm

One reason drops are of such focus, is that people think they are a product of poor concentration (as opposed to say, poor route running, which would happen because you're just not that good at it). Thus folks get on drops because they think it is a problem a player can correct "if they want to."

Personally, I doubt that. (Why should anyone be able to change their reflexes/brain/nervous system vs. say develop speedier legs.)

But it could be tested. Do drop percentages change dramatically over time for a player? Have players been able to improve their drop rates more dramatically than improve other aspects of their play, say DVOA, from year to year?

19
by Kevin Pelton (not verified) :: Sun, 08/28/2005 - 12:48am

With regards to the downs, I think there are two possible questions you could ask:

1. How much did this receiver hurt his team by dropping passes?
2. How frequently does this player drop the ball?

Well, in theory you could ask these questions. In our cold, hard reality, as has been pointed out, we don't have drops tied to particular plays.

Where I think drops by situation would probably be useful is at the team level, a la the third-down stats FO has found so important. Presumably there's no skill to dropping passes only on first and second down, so we'd expect teams with a high number of drops on third down to improve in this regard the following season.

20
by Nick (not verified) :: Sun, 08/28/2005 - 1:10am

To be fair there seem to be more drop for wideouts than flankers, which makes sense, but maybe it is not fair to compare the two. There seems to be a coorelation between yards per catch (presumably yards per target) and drops as well.

As I recall, the year Sterling Sharpe caught 106 passes he caught an extremely high proportion of those behind the line on WR screens. A drop on those would indicate a problem.

Not all drops are created equal.

21
by Ryan Mc (not verified) :: Sun, 08/28/2005 - 2:37pm

The article suggests that the problems Fred Taylor had with drops didn't receive attention because he isn't expected to be as surehanded as the WRs. I think this sounds unlikely. A more likely explanation is that his drops probably came on swings and/or screens which weren't obviously going to be big plays. Conversely, a WR dropping a perfectly thrown ball when he's open 40 yards downfield seems like a huge missed opportunity. Again, as others have stressed, not all drops are created equal.

The other interesting part of this fact, though, is it does call into question a team's strategy of playing conservative and throwing lots of short passes to the RBs if the guy is just going to drop the ball a lot of the time.

A final thought: in praise of the Pats once more. Although their receivers aren't big names, the numbers above do suggest that they've managed to pick guys who are solid and dependable. An example of the team focussing on substance rather than style.

22
by Robert (not verified) :: Sun, 08/28/2005 - 6:20pm

It seems like the drop% vs. catch% chart is misleading, because it is affected by the accuracy of the passer. Receivers with an accurate QB will come out looking like they catch more passes. It would be a pain, but I would like to see the same chart with a correction for QB accuracy.

Also, it occurs to me that a tangent of a tangent is just the original tangent...

23
by Kevin Pelton (not verified) :: Sun, 08/28/2005 - 6:33pm

Robert, i did do a measure of catch percentage relative to team completion percentage and expected to use it. After I saw the results, however, it didn't really appear to add any value. *shurg*

24
by Pat (not verified) :: Mon, 08/29/2005 - 6:09pm

The correlation coefficient between the drop percentage and catch percentage is -.229, indicating that players who drop the ball more frequently tend to catch fewer passes. The relationship, however, is not particularly strong.

Well, you might want to try adding in an estimate of the error as well. Obviously at a minimum, the error in the estimate is +/- 1 catch/drop. With Branch, for instance, you can't tell the difference between 1.5% and 0%, because they're both less than 1/50.

A 2% error versus a 1% error on your drops chart is quite large! The correlation would probably improve a bit (with... any program that takes errors into account, though I don't know if Excel does) with errors taken into account.

25
by Carl (not verified) :: Mon, 08/29/2005 - 7:26pm

"The corresponding chart shows how weak the relationship is; the receivers are distributed in something approximating the circle familiar to anyone who’s taken a statistics class as an example of variables with no relationship."

I was waiting for that other shoe to drop.

26
by Pat (not verified) :: Mon, 08/29/2005 - 8:15pm

Robert:

Traditional QB accuracy is biased as well, though, because not all passes are as easy to catch. We don't have "location of catch" information for stadia other than Buffalo (I think it's Buffalo) so correcting for QB accuracy is nigh-impossible.

27
by b-man (not verified) :: Wed, 08/31/2005 - 10:37am

FO gets a shout from patriots.com (see link).

28
by Kevin Pelton (not verified) :: Thu, 09/01/2005 - 2:07am

A quick postscript to note that newly-signed Hawks receiver Peter Warrick had seven drops in 122 targets in his last full season (2003), a worse than average 5.7% drop percentage.

29
by ChrisS (not verified) :: Thu, 09/01/2005 - 5:14pm

A very good article. I would be interested in seeing the complete underlying drop/thrown stats.
The effect of a drop is big because in 2004 57% of completions went for a first down and another 7% were TD's. I would expect both of these percentages to be even higher for WR's. Assuming on average you score a TD for every 8 first downs and PAT's are automatic, each drop wil lower the team's expected score by about 1 point.

30
by aeneas1 (not verified) :: Sun, 09/04/2005 - 7:05pm

dropped pass ratio should be calculated by dividing dropped passes by catchable passes (receptions plus dropped passes) not by dividing dropped passes by targets.

31
by Kevin Pelton (not verified) :: Mon, 09/05/2005 - 12:48am

That is more intuitive, isn't it? But note it doesn't really make a huge difference. The only change it would create in the top or bottom five would be Keary Colbert falling behind Watts and Bryant.