Warning: Cannot modify header information - headers already sent by (output started at /home/content/85/8979285/html/wp-includes/post-thumbnail-template.php:1) in /home/content/85/8979285/html/wp-content/plugins/wp-super-cache/wp-cache-phase2.php on line 62
dallas cowboys stat analysis | The DC Times

The DC Times

A New Way to Look at the Cowboys, NFL, and Fantasy Football


Why the Dallas Cowboys (and all NFL teams) should attempt WAY more two-point conversions in 2011

Jonathan Bales

One of my favorite topics on which to write is football theory–why teams make the decisions they do and if they could become more efficient in their choices.  In the past, I’ve completed studies detailing why teams should go for it more often on fourth down, run more on third down, throw more deep passes, run more frequently to the weak side, pass out of ‘running’ formations, be far more aggressive on 2nd and short, and so on.

It is my view that, if a coach was to implement a perfect understanding of advanced statistics and game theory into his in-game decisions, the potential impact would far outweigh that of any single player.  A “perfect” record of fourth down decisions, for example, could be “worth” a handful of points to a team in any individual game.  With all of the statistically inefficient choices NFL head coaches make each and every game, there is an enormous opportunity for less talented teams to outperform superior ones based solely on statistics.  Now is truly a golden age for the NFL in that teams are just beginning to crack the surface of advanced statistics that baseball uncovered years ago.

One such statistical issue about which I feel very strongly is the use of two-point conversions.  I talked about this issue at length during the 2010 season, particularly after the Cowboys’ Week Seven loss to the Giants and their Week 16 loss to the Cardinals.  From the New York post-game review:

Later, the Cowboys did score a touchdown to close within 12 points.  They decided to go for a two-point conversion, and Tirico immediately went off about “awful” the decision was.  I normally like Tirico, but he needs to stick to play-by-play and keep his nose out of matters of football theory.  On this topic, he was again as wrong as could be.

You’ll often hear announcers say it’s “too early to go for two.”  But what does that even mean?  How is it ever “too early?”  The decision to go for a two-point conversion should be based on a variety of factors, including the score, a coach’s confidence in his two-point play, and so on.  Actually, if the probability of Team X converting on a two-point attempt is 50.1 percent, they should almost always go for two.  The expected points of 1.002 is greater than that of an extra point (which can obviously only be as high as 1, even with 100 percent accuracy).

Thus, you’d only want to go for an extra point in non-normal game situations.  Suppose Team X scores a late touchdown to tie the game.  They’d clearly want to attempt the extra point to secure the win.  Going for two points would be quite disadvantageous in that scenario.  If football commentators knew the statistics and theory behind two-point attempts, perhaps they’d be saying “It’s too early to try the extra point.”

There are more reasons that Tirico was unjustified in his stance.  Down 12, the decision of whether or not to attempt a two-point try is indeed a “no-brainer,” but Tirico is on the wrong side of the debate.  If you go for two points and succeed, you’re down 10 points and now know that a touchdown and field goal will tie the game.  If you go for two and fail, you now know that you need two touchdowns to win.  If you kick the extra point, however, you might later kick a field goal that will turn out to be meaningless.

The idea that you want to “keep yourself in the game” by kicking an extra point is preposterous.  You actually want to determine what scores you’ll need as early as possible.  If you kick the extra point, then a field goal, you’re down eight points.  If you then score a touchdown and fail on the two-point attempt, you’re still another score away from winning the game.  The field goal attempt in between touchdowns becomes all but meaningless, and this is due solely to the fact that you didn’t attempt the two-point conversion as early as possible.  Failing the two-point try earlier, as I said above, provides you with the knowledge that you need two touchdowns to win.

Tirico and Jaws used the outcome of the game as justification for their view, but that’s wrong as well.  If you roll a six-sided die and bet even money on a specific number coming up, your bet is a dumb one regardless of the outcome of the roll.  The fact that you will win money one time out of six doesn’t justify the decision ex post facto.  When I listen to the Monday Night Football crew, I feel like I am betting that an even number will come up on my roll of the die–but all the commentators, I mean numbers, are odd.

Later in the season, I criticzed Jason Garrett for his failure to attempt a two-point conversion in the third quarter of a game:

  • Down 21-19 in the third quarter, Garrett decided to kick an extra point. Huge mistake. I’ve talked all season about why teams should try way, way more two-point conversions. Over the course of any given season, kickers make around 98 percent of extra points, while two-point conversions are successful around 48-49 percent of the time. While the expected points of extra points is higher (.98 x 1 is greater than .48 x 2), the difference isn’t great enough that it should overcome all game situations. For example, Garrett never would have kicked the extra point in the fourth quarter, as he doesn’t know if the Cowboys will score again.
  • Further, two-point conversions are only statistically inferior to extra points because coaches tend to call the wrong plays down by the goal line.  Over the last 20 seasons, rushing the ball has yielded a successful two-point conversion over 60 percent of the time.  Even if a team went for two points after nearly every score and rushed the ball each time, I doubt the success rate would jump below 50 percent (the break-even level at which two-point tries are statistically equivalent to extra points, assuming a 100 percent success rate on the latter).  Thus, extra points should actually only be attempted in very specific situations, such as a tied game in the fourth quarter.
  • On top of all of that, let’s not forget Buehler is about as erratic as kickers come.  His extra point success rate is nowhere near 98 percent (probably closer to 94 or so), meaning the Cowboys would only need to convert on 47 percent of two-point tries to yield the same expected points.  And if you’re correctly running the ball, what does it matter if Stephen McGee is at quarterback?
  • I assume Garrett attempted the extra point because he figured Dallas would score again anyway.  That’s faulty logic, however.  Even if we assume two-point conversions yield less expected points than extra points, and we take into account McGee’s presence in the lineup, the difference between a two-point try and extra point is still small enough that, for an extra point to be the right call, we’d have to assume there’s less than a one percent chance the Cowboys wouldn’t score again.  While it’s likely the offense was going to put more points on the board, it certainly wasn’t greater than 99 percent.
  • I updated live from the game last night on Twitter, and a few followers claimed that it was “too early to go for two and the chart says the extra point is the right call.”  While I appreciate everyone who took the time on Christmas to read my thoughts, that reasoning is simply incorrect.  What does it even mean to be “too early to go for two”?  While you certainly have less of an idea of the final score in the first quarter as compared to late in the game, you should always side with statistics.  If the numbers say attempting a two-point conversion is the right call (which they did for the Cowboys in the third quarter–and it wasn’t even close), then kicking an extra point is the risky move.  Further, NFL coaches are just tapping the surface of advanced statistics and game theory, meaning most of their “infallible” charts are dead wrong.  It’s Garrett’s job to give the team the highest probability of victory, and whether a decision seems “risky” or not to the public, it needs to be made.

Later, Garrett issued a statement on his decision to kick the extra point (below).  I responded with this:

According to Jason Garrett, he didn’t go for two points when down 21-19 in the third quarter of Saturday night’s game because “What happens when you start making those decisions is sometimes you get a little hasty and say, ‘OK, if we get two here that will tie us up.’ But typically, what happens when you have another quarter to play, there are a couple more scores and the whole thing kind of plays itself out a little bit.”

Although I’d wager that the majority of NFL coaches agree with Garrett’s assessment, it is the wrong one.  I hate to be so blunt about it (secretly I love it), but he’s just dead wrong.  Garrett points out that there will typically be more points scored after the third quarter, which is correct, but somewhat irrelevant.

First of all, as I’ve already pointed out, two-point conversions may not even yield less expected points than extra points.  If that’s the case (which would be a virtual certainty if teams ran the ball more on two-point attempts), then going for two points should be the status quo, with an extra point only being attempted in specific game situations (such as tied late in the contest).

Even if extra points are generally statistically superior to two-point tries, however, Garrett still made the wrong decision.  While I agree with his notion that more points were likely to be scored, that fact is far from certain.  Actually, for an extra point to be the right decision in that scenario, we would have to assume that the chances of neither team scoring again was small enough that it wouldn’t account for the disparity between the expected points of an extra point (about .98) and a two-point attempt (.96 at worst).

As it turns out, Garrett would have to assume either that the chances of neither team scoring again were below one percent or that the offense’s chances of converting on their two-point try were closer to 25 percent than 50 percent.  Anyone believe either scenario to be the case?

Me neither.

I used a lot of old material here because I feel like actual game situations are the most effective way to state my case.  We can see real-world situations in which forgoing a two-point try has dire consequences for a club.

Ultimately, I believe an NFL team could secure a significant number of “extra” points by attempting two-point conversions after the majority of touchdowns.  As I stated above, NFL teams convert two-point attempts about 60 percent of the time when they run the football.  That number would certainly decline with increased rushing attempts, but the efficiency of two-point pass attempts would subsequently increase.  I see no reason why an NFL team (particularly one with an offense as potent as that of Dallas) wouldn’t be able to convert a minimum of 55 percent of two-point tries if they focused on improving their efforts.

Assuming the Cowboys score three touchdowns per game in 2011 (they scored 43 in 2009 and 46 last season), the “increase” in expected points would be 6.24 (assuming 97 percent accuracy on extra points, which is a stretch if David Buehler is still the kicker).  That might not sound like a lot, but there’s a solid chance those “extra” points would result in another win for the ‘Boys.  Assume a 60 percent conversion rate and that number jumps to 11.04.  Not too shabby for a philosophical decision that would require relatively little practice time.


Jason Garrett Tipping Plays Via Formation: ‘Double Tight Strong’ Usage in 2010

Jonathan Bales

Note that my results also include "Double Tight I," which is the same as above with the fullback lined up directly behind the quarterback.

In my study on the Cowboys’ 2009 usage of ‘Double Tight Strong’ (left), I noted that the Cowboys ran a strong side dive 71.6 percent of the time they lined up in the formation (83 of 116 plays), including 85.7 percent of the time when they motioned into it (42 of 29 plays).

Defensive coordinators clearly caught on to this trend, as the Cowboys’ yards-per-rush on the strong side dives decreased from 7.8 over the first five weeks of the season to just 4.4 over the rest of the year (including only 3.2 against all teams but the Raiders).  Thus, the opposition was fully aware of this trend of Garrett’s coming into the 2010 season.  The Cowboys’ efficiency on “Double Tight Strong/I” plays is representative of that.

The Cowboys lined up in the formation 81 times in 2010 (35 fewer than 2009, at least), running the ball 82.7 percent of the time.   Of those runs, 52 (77.6 percent) were strong side dives.  The overall strong side dive rate (including passes) was 64.2 percent–down from 71.6 in 2009–but still way, way too high.  Once again, when Dallas motioned into the formation, the rate of strong side dives increased (to 72.7 percent of all plays).

Unlike 2009, however, the Cowboys did not find success on these strong side dives at any point during the season.  The ‘Boys averaged just 2.15 yards-per-carry on the 52 strong side dives in 2010.  On all other runs (almost all to the weak side), the Cowboys averaged 4.87 yards-per-rush.  In 2009, Dallas also found far more success when running weak side out of “Double Tight Strong”–averaging 6.7 yards-per-rush–indicating that defenders truly have been keying in on the strong side dive.

Of course, Garrett loves to use this formation in short-yardage situations, so could this be the culprit for the lack of yardage?  Not really, as the average yards-to-go on “Double Tight Strong” plays was 5.94–lower than the overall rate, but not by an incredibly large margin.  Actually, 37 of the plays from the formation came with exactly 10 yards-to-go.  That’s 45.7 percent.  An additional 15 of the plays came with 5+ yards-to-go, meaning 64.2 percent of the plays came in situations that were clearly not short-yardage.

And it wasn’t as if the Cowboys were thriving on the short-yardage plays either.  Of the 29 plays from “Double Tight Strong” with four or less yards-to-go (and nearly all of them were with exactly one yard-to-go), the Cowboys converted a first down or touchdown just 13 times.  That’s only a 44.8 percent conversion rate on very short-yardage plays. Kind of sick.

One might argue that some predictability can be good if utilized correctly.  The 64.2 percent strong side dive rate might be less detrimental to an offense, for example, if they use playaction passes to take some shots downfield on the other plays.  Thus, an offense could “concede” a strong side dive or two (or 52, apparently) to set up big plays in the passing game.

That sounds great in theory, but Garrett didn’t call many “high-upside” plays out of the formation at all.  Actually, the average distance of the Cowboys’ passes from “Double Tight Strong” was just nine yards.

Ultimately, this formation will continue to haunt Dallas until 1) the strong side dive rate decreases dramatically or 2) Garrett utilizes the predictability from the formation to set up big pass plays.  Garrett has improved in a number of areas as a play-caller over the past few years, but focusing on improving “Double Tight Strong” calls (or scrapping it from the playbook altogether) should be high on his list of priorities.


Cowboys 2010 Pass Rates from Specific Personnel Groupings

Jonathan Bales

Game theory dictates that offensive coordinators should increase (or decrease) the frequency of any particular event (runs, passes, draws, counters, playaction, etc.) until its efficiency and overall production are maximized.  For example, in my analysis of the Cowboys’ 2010 weak side runs, I noted that the Cowboys have been significantly more successful on weak side runs than strong side runs or runs from balanced formations.

Jason Garrett adjusted accordingly in 2010, running to the weak side on 22.8 percent of all runs–up from 19.5 percent the prior season.  Still, the ‘Boys could benefit from an even higher rate of weak side runs, as the 4.72 yards-per-rush number when running to the weak side was significantly greater than that of strong side or balanced runs.  But how does Garrett uncover the “perfect” weak side run ratio?  I addressed that problem in the past:

Game theory suggests Garrett should slowly increase the number of weak side runs until the average yards-per-carry is maximized.

But how will we know when that number is reached?  The answer would be simple if we assumed those people drawing up plays to try to stop the Cowboys offense–the opposing defensive coordinators–were perfectly rational.   In that scenario, the defense would call an increasing number of weak side blitzes until they minimized the overall yards-per-carry.   They would in effect create their own Nash equilibrium.

Of course, defensive coordinators do not always call plays in a rational manner. Their knowledge is not unlimited, and so sometimes they may call too many weak side blitzes or not enough.  Perhaps sometimes the number of weak side blitzes they dial up has no correlation at all with the offenses’s weak side running rates or successes.

Garrett’s job, then, must be to take into account the thoughts and tendencies of defensive coordinators (perhaps easier said than done), and then adjust his play-calling accordingly.   If Team X calls an inordinate amount of weak side blitzes, for example, then the Cowboys own Nash equilibrium will be shifted to include more strong side runs (and vice versa).

Thus, play-calling (or effective play-calling anyway) is not simply about knowing your own players.   It is about successfully predicting the calls of defensive coordinators by knowing their tendencies.   This may sound extremely difficult (and it is from the standpoint of one individual play), but aberrations tend to flatten out over the course of a game in such a way that, despite not knowing individual play calls, a team can assume a “regression to the mean” of sorts where a team’s overall tendencies will always eventually shine through.

For Garrett, this means being one step ahead of the game.  Instead of simply knowing what you want to do, you have to know what your opponent thinks you are going to do, then adjust accordingly.  When playing against a really stealthy coach, you may have to know what he thinks that you think that he thinks about what play you are going to call.

If you are an offensive coordinator and have called three straight weak side runs in a row, for example, your natural inclination may be to deviate from this tendency.  You might do this in an effort to “mix it up.”   But game theory suggests you should take into account the opposition’s thoughts before making a decision.

Perhaps you know that he is thinking that you are thinking that he will call a weak side blitz to combat your recent success. Knowing this, you would assume he may call a strong side blitz (or none at all), and you would be correct.   Thus, despite three straight weak side runs, the best play call is yet another weak side run.

Being “unpredictable” isn’t about changing play calls just for the sake of changing the play, but about adjusting your tendencies according to your opposition’s tendencies to create an environment where potential success will be maximized.

This same sort of method can be used as the rational behind a plethora of play-calling alterations.  One such change we could potentially see from Garrett is the run/pass ratio from specific personnel groupings.  In the chart below, you can see Garrett’s 2010 pass rates based on personnel, along with the relation to his 2009 rates.

In much the same way that weak side runs can be optimal for an offense, so too can passing the ball out of “untraditional” personnel groupings (or, on the other hand, running the ball from pass-heavy personnel packages).  There’s a reason the ‘Boys have found a ton of success when passing out of “running” formations (and with “running” personnel).

The passing success of the Cowboys out “running” formations is equivalent to the success teams have when running the ball on 3rd down.  There is nothing inherently efficient about running the ball in these situations.  Rather, the success comes from your opponent’s expectations.

Similarly, passing out of “running” formations isn’t an inherently superior strategy to passing with four wide receivers on the field.  Instead, it works because of the defense.

Think of it like this. . .let’s say passing the ball out of a four-receiver set receives a hypothetical score of 80 points (this total is arbitrary and independent of a defense).  Passing the ball out of a double-tight formation, on the other hand, is intrinsically worth just 60 points.

So, why would a team choose the latter scenario–a “sub-optimal” strategy?  Because the strategy is only “sub-optimal” in theory.  In practice, the defense makes substitutions to be able to effectively defend each formation.  To counter the run against the double-tight formation, they knowingly decrease their ability to thwart the pass.

Thus, they may receive a pass defense score of 75 against a four-receiver set, but just 50 against double-tight.  In that case, passing the ball out of double-tight yields a 10 point advantage for the offense, compared to just a five point advantage when throwing the ball out of the “passing” formation.

When analyzing Garrett’s personnel-based play-calls, we see that he is generally improving.  When the Cowboys implement two tight ends, two wide receivers and a running back, they are generally a balanced team, passing the ball 58.6 percent of the time.  This is down from a 71.9 percent pass rate in 2009.

Garrett is also calling more passes from run-oriented personnel packages (such as two tight ends, one receiver and two running backs), and less passes from pass-oriented personnel groupings.  The only exception is the one tight end/four receiver package, which the Cowboys implemented only 25 times all season.

I’d still love to see the Cowboys run the ball more in three-receiver sets and pass more out of 2 TE, WR, 2 RB (one of those “2 RB” is usually a fullback, by the way).  If Garrett finds a way to efficiently run the ball without a fullback on the field and continue to throw the ball well out of two-tight end looks, the Cowboys will take huge strides in becoming a much more unpredictable, and potent, offensive football team.

Dallas Cowboys Times is on Twitter.

Subscribe to our free e-mail updates.


Tony Romo Versus Blitz, Perceived Blitz in 2010

Jonathan Bales

It is quite obvious that Tony Romo’s improvisation skills are vital to the success of the Cowboys’ offense.  He has used his quick feet and athleticism to make the offensive line look above average in pass protection–or at least superior to reality–for years.

The vast majority of Romo’s “schoolyard” plays–the ones where he jukes and dodges defenders, all the while keeping his eyes downfield in search of the big play–have come on blitzes.  Not only are there more defenders for Romo to elude (and thus less in coverage), but the quarterback is also underrated in his ability to diagnose defenses and promptly hit the open receiver.

Most of Romo’s reads get made before the snap.  How often do you see the play clock tick down to just one or two seconds before the Cowboys snap the ball?  This is because the team uses every available second to call the play(s), diagnose the defense, and make the necessary adjustments.

As I looked into my database of Cowboys’ 2010 offensive snaps, I noticed a trend that seemed to confirm these ideas.  I track not only when a defense blitzes, but also when they show a blitz pre-snap.  Most of Romo’s mistakes over the past few years have seem to come in two situations:

  1. When defenses don’t blitz and sit back in zone coverage, forcing Romo to make accurate throws, and
  2. When defenses show blitz pre-snap but back into a safe coverage

In the chart below, you can see that Romo was incredible against the blitz this past season.  His 6:1 touchdown-to-interception ratio on blitzes is far better than the 5:6 ratio he displayed against “regular” defenses.  Romo is particularly outstanding when he knows a blitz is on the way, recording a ridiculous 136.7 passer rating in these situations.  His 9.45 adjusted yards-per-attempt is ridiculous.  (AYPA subtracts sack yardage and 45 yards per interception–the number of yards, on average, each interception is “worth” in terms of a team’s win probability.  Thus, AYPA is an awesome tool for assessing a quarterback’s value against the blitz).

When teams did not blitz Romo in 2010, however, he was slightly below average.  His passer rating his historically always been lowest when a defense shows blitz but then backs off, and that was again the case in 2010 (71.3 rating).  Romo’s 2.04 AYPA in such situations tells the whole story.

I think Romo’s failures stem from the importance he places on pre-snap reads.  When defenses show a blitz but then don’t come, Romo’s original read is usually taken away.  He can then sometimes panic, and although I truly believe Romo is a tremendous talent and a Championship-level quarterback, he does not possess incredible accuracy.  He makes a lot of his plays by buying extra time to allow receivers to become wide open.

This would explain why he still does well when teams do not show blitz but then end up coming after him.  What he sees post-snap may differ from his pre-snap reads, but he possesses not only a quick release, but also the athleticism to make good things happen that may not have been designed in the original play.

Overall, it seems clear Romo performs much better when he “knows” whether or not a blitz is coming.  When teams do not blitz, his passer rating is 15.3 points higher and his AYPA 3.53 yards better when teams do not show it as opposed to feigning a blitz.  When defenses do send extra defenders, Romo’s passer rating is 1.58 times as high and his AYPA nearly four yards superior if the defense “shows” it as opposed to disguising their intentions.

So you want to stop Romo?  Year in and year out, it had been proven to not blitz him often, but feel free to act as if you will.  When you do blitz, you better disguise that as well.


Dallas Cowboys 2010 Rushing/Passing Efficiency By Down

Jonathan Bales

I’ve talked before about why I believe the Cowboys should throw more often on first down, particularly out of running formations.  Despite the league-wide transition to an emphasis on throwing the football, defenses still tend to primarily defend the run on first down.

Well, I sorted through my 2010 play database today to determine the Cowboys’ efficiency on first down passes.  I quickly realized the numbers were relatively useless without a comparison to statistics on other downs, so I calculated those as well.  Then, I postulated that an even stronger down-to-down comparison of passing statistics would be accomplished by noting the team’s rushing efficiency too.  The result of all of this is below:

Note: I did not count QB spikes or kneel downs, and sacks/QB rushes are counted into the passing totals.

A few notes:

  • The Cowboys’ completion percentage remains relatively steady, regardless of the down.  I was really surprised to see just a 4.4 percent difference between first and second down passing.
  • You might think the Cowboys would run more on second down than first, but that’s actually not the case.  Nearly two-thirds of second down plays have been passes.
  • As expected, third down passing efficiency trumps that on first and second down.  I would speculate this is due to game situations–defenses don’t mind yielding a 10-yard gain on 3rd and 15.  Still, 8.01 yards-per-attempt is tremendous for any down.
  • The low rushing efficiency on third down stems primarily from 3rd and short situations.  Running on 3rd and 4+ is actually quite successful.
  • The greatest disparity between rushing and passing efficiency comes on third down (passing is 2.36 times as efficient), followed by first down (1.80 times as efficient), and then second down (1.56 times as efficient).  You might ask, “Why not just pass the ball every play?”  Well, aside from the fact that defenses would quickly adjust, running the ball also yields a higher percentage of positive plays–there are no incomplete passes.  A 3rd and 1 run is almost always superior to a pass for this reason.

There are a lot of other conclusions that can be drawn here.  I’d love to hear what some of the DC Times regulars think about this data.


From the Archives: Effectiveness of Tony Romo’s Audibles

By Jonathan Bales

I recently headed into my film study database to decipher Tony Romo’s effectiveness when he audibles out of the original play call. Sometimes Romo will actually call an entirely new play at the line of scrimmage, while other times he will simply signal for the team to check into the second play which was called in the huddle (the team often calls two plays in the huddle, planning to run the first unless Romo checks out).

The latter scenario is marked by a phrase many of you have probably heard the Dallas’ quarterback yelling on television, “Kill, Kill, Kill!”  When you hear this, Romo sees something in the defense that makes him believe the first play called in the huddle will be unsuccessful.   The second play, which is the one run after the “Kill” call, is generally dissimilar to the original call to combat whatever problem Romo noticed.

Note that the Cowboys really have very little incentive to use “dummy” audibles.  Nearly every check they made in 2009 (75 of 79) was a “Kill” call.  These calls give the defense little strategic advantage, as they do not know either play which was called in the huddle.  Thus, because there’s no important information being conveyed at the line of scrimmage, there’s not really anything to cover up with a “dummy” call.

I tracked every play during which Romo called an audible all season, and here are the results:

As you can see, Romo checked out of a play 79 times in 2009, which equates to 4.9 per game and 8.0 percent of all plays.   The Cowboys were much more successful when Romo audibled into a run, which he did 59.5 percent of the time he checked.  They averaged 5.8 yards-per-rush on all run audibles, a full yard better than their season average.  They were not as successful on pass audibles, however, averaging a half yard less per attempt than their season average.

One possible explanation for the lower productivity in the passing game after checks is that defenses are more prepared to defend the pass after an audible.  They may assume an audible by the opposing quarterback means he sees an opportunity for a big play, probably a pass, thus making them more likely to effectively defend the pass.

So should the effectiveness of the run force Romo to check into a run play more often than the current 59.5 percent rate?  Not necessarily.  That percentage is already rather high, and if it would increase to say, 70.0 percent, the defense would have a great indication as to the play-call.

Further, there are times when Romo is even more efficient with his audibles, both when they result in a run and a pass.  These times are when the opposing defense shows blitz (only lines up as if they will blitz, but does not necessarily blitz), and the results are below.

Of the 79 times that Romo checked on the season, the defense was showing blitz at the time on 30 of those plays.  The Cowboys gained 237 total yards on these checks, gaining 1.9 more yards-per-rush against the blitz than their season average, and 1.5 extra yards-per-pass.  The sample size of 30 plays may be low, but probably significant enough to conclude that Romo is more effective in making audible calls when he perceives blitz.

Also notice Romo checked to a pass 56.7 percent of the time during these situations, compared to just 40.5 percent overall.  This may be because Romo is more comfortable putting the pressure on himself to make a play when the offense is facing pressure.

While this higher pass-to-run ratio could also be influenced by the fact that teams are more likely to blitz during passing downs, it probably is not too much of a factor, as the Cowboys would already have a pass called and would be much less likely to even audible.

So why is Romo more efficient making checks when the defense shows blitz?  One explanation is that Romo is just more effective versus the blitz in general.  This could reveal some of the success, but it probably cannot account for the full two yard difference in average yards-per-pass.  Further, Romo’s ability versus the blitz does not explain the increase in rushing average when he checks after seeing blitz.

Thus, we must conclude that Romo is generally effective in making audibles, but much more so when he believes a blitz is coming. This does not necessarily mean that he should check out of more plays when he perceives blitz, but perhaps increasing the number of audibles during these situations until the yards-per-play average peaks may result in an even more effective Dallas Cowboys offense.


From the Archives: Should Jason Witten Block Less? A Statistical View

By Jonathan Bales

Cowboys at Broncos, October 4, 2009. There are just seconds left on the clock with Dallas at Denver’s two-yard line, down seven. The Cowboys have two plays left to score and are sure to try to get the ball to their top playmakers.  Right?

Right, if you think the Cowboys’ best option is Sam Hurd.  The Cowboys force two slants to Hurd on 3rd and 4th and Goal and lose the game.  Don’t get me wrong–Hurd is a great bench player and has improved vastly, but, with Roy Williams out of the game, the Cowboys needed to target their Pro Bowl tight end.  Surely he had to be open on just one of the two plays.

The yards-per-pass average was significantly higher with Witten in a route.

But Witten couldn’t be open, because he was left in to block on both the 3rd and 4th down plays.  These two plays in particular got me to thinking:  how effective are the Cowboys on pass plays when Jason Witten is out in a route versus when he stays in to block?  I tracked every play of the 2009 season, and the results are to the right.

Witten was in the lineup for 485 pass plays (this includes sacks and Romo scrambles). c Of these plays, he was out in a pass route 374 plays, or 77.1 percent 0f the time.  Intuitively, it seems as though Witten must be out in a route more than about three out of four pass plays, but this was not the case.

As you can see, the Cowboys were more successful by leaps and bounds when Witten went into a route.  They averaged a stout 9.3 yards-per-attempt during these situations, compared to just 7.4 yards-per-attempt on the 111 pass plays where Witten blocked.

But why is this the case?  Is it purely due to Witten’s supreme receiving skills, or could there be another reason?  If Witten was out in a route more often during pass-friendly situations, such as when an opposing defense is playing a prevent, for example, then these numbers might be a bit inflated.

This was not the case, however.  Often times the Cowboys would be in a formation called “Gun 3 Wide Pro” (pictured left) during these situations, where Witten lined up next to Romo in the Shotgun.   He would frequently stay in to block, and only sneak out into a route if the protection was sound. Thus, this does not seem to be a reason for the higher passing average when Witten was a receiving option.

Another possible reason for a decrease in average is that the Cowboys frequently would keep Witten in to block when trying to go for the big play.   Opposing defenses often key the tight end to decipher run or pass, and when linebackers and safeties see Witten staying in to block, they tend to sneak up toward the line of scrimmage. Dallas loves to run one and two-man routes during these times, attempting to sneak a receiver behind an over-aggressive defense.

So why would this cause a lower yards-per-pass average with Witten blocking?   Perhaps attempting a big play gives the Cowboys a shot at a quick score, yet ultimately lowers the average because of the low chance of hitting on such a play.

If this is the case, though, we would expect the percentage of big plays to be significantly higher when Witten remains in to block versus when he is in a route.  Again, I dove into the film, and here are the numbers:

The Cowboys had slightly more big plays when Witten was not in a route.

The percentage of 15+ yard plays in the two scenarios is virtually even, while the percentage of 30+ yard plays is slightly higher when Witten remains in to block. These are the sorts of number that might be expected, but not a significant enough difference to explain nearly a two-yard difference in yards-per-play average (9.3 vs. 7.4).

Yet another possible explanation for the difference is that, when Witten stays in to block, there is a good chance the Cowboys are expecting blitz.  Blitzes could force bad decisions and lower the average.  The problem with this idea, however, is that Romo is one of the most effective quarterbacks in league history against the blitz.

If opposing defenses blitz and the Cowboys have proper protection, which would most likely be the case when they leave a tight end in to block, Romo would generally pick them apart (meaning this cannot be a correct explanation for the decrease in average).

One last potential explanation of the greater success displayed with Jason Witten in a pass route is that these situations are just generally safer than when Witten remains in to block.  The Cowboys are more likely to not be blitzed during these times, and if they are, Romo has his favorite option available for a hot route.   The numbers, though, do not seem to support this theory either.

When Witten did not go in a route, the Cowboys gave up eight sacks on 111 plays (7.2 percent), but Romo threw zero interceptions.  When the Cowboys’ top tight end did not stay in to block, however, the Cowboys yielded 25 sacks  on 374 plays (6.7 percent) and Romo threw all nine of his interceptions.  This 0.5 percent difference in sacks is not statistically significant enough to conclude that there is any real difference, while the eye-popping difference in interceptions proves that putting Witten out in a route is not necessarily a safer option.

Once the yards-per-pass with Witten blocking vs. in a route reaches the Nash equilibrium, perhaps 85 or 90 percent, Dallas will maximize their overall yards-per-pass.

Thus, we must conclude that the 9.3 yards-per-attempt during pass plays in which Witten was in a route is actually due to his ability to get open and make plays.

So, how do all these numbers affect the Cowboys’ future play-calling? The greatest success rate would arise through a steady increase in the percentage of plays that Witten is in a route that only stops once the “Nash equilibrium” is reached. While I won’t bore you by going into great detail about this term, know that it is basically when the average yards-per-play for both scenarios (Witten staying in and going out) is maximized.  At this point, the Cowboys will attain the greatest overall yards-per-pass average.

Thus, the best solution for next season’s playcalling would be for Jason Garrett to increase the “Witten-in-route” percentage until the Nash equilibrium is reached, reminding Romo to continue to limit turnovers, particularly when Witten is not in to help secure protection.


10 Dallas Cowboys Players Who Will Break Out in 2010


By Jonathan Bales

Last season, a few breakout players helped turn around the Cowboys’ season, most notably Miles Austin.  Following a Week Four loss in Denver, the ‘Boys were reeling.  Starting receiver Roy Williams got injured, opening the door for Austin’s 10 catches for 250 yards and two touchdowns in an overtime win in Kansas City.  Who knows how the team’s season who have unfolded without Austin’s presence in K.C.?

This season, a variety of players are primed for breakout seasons.  In uncovering these players, I like to look at statistics which are not generally utilized, but still provide an idea of how efficiently a player can perform.  Defensive end Jason Hatcher, for example, racked up 17 quarterback pressures in 2009, yet totaled just one sack.  With his snap count figuring to increase, the rate at which Hatcher reaches the quarterback could put him in line for six, seven, even eight sacks this season.

Below are 10 Cowboys players who are ready to become household names.  Click on a player to see in-depth statistics detailing his ’09 performance (or a scouting report for rookies).

10.  Doug Free

A lot is riding on the shoulders of Free–perhaps more so than any “new” Cowboys starter.  Free will have to beat out newly-acquired Alex Barron to win the starting left tackle job.  On paper, his athleticism and quick feet make him a natural fit for the left side.

Prediction: If Free starts every game and can yield less than six sacks, he did his job.  Also expect him to give up in the vicinity of seven hits and 25 pressures.

9.  Alan Ball

I just posted an analysis of a potential Alan Ball/Michael Hamlin training camp battle.  Ball’s coverage ability should be enough for him to retain the job, but he needs to upgrade his tackling.  Like Free, the competition behind Ball should force him to be at his best.

Prediction: The Cowboys are in dire need of a play-making safety.  Will Ball be that guy?  Expect him to be in the neighborhood of 3-5 interceptions.

8.  Stephen Bowen

With starter Marcus Spears a restricted free agent, the Cowboys figure to increase Bowen’s reps as the season progresses.  There is even talk of Bowen starting over Spears immediately.

While I expect Spears to remain the starter, Bowen will come in on more and more running downs (only 99 of his 484 total snaps came against the run).

Prediction: Expect Bowen to secure close to 600 snaps this season, including about 175 against the run.

7.  Felix Jones

Okay, so Jones has basically already broken out–but only in terms of efficiency (yards-per-carry), not totals.  In his short career, Jones has yet to reach 1,000 total rushing yards and has tallied just seven all-purpose touchdowns.

To take his game to the next level, Jones needs to stay healthy enough to obtain the touches necessary for big-time production.  With offensive coordinator Jason Garrett supplying him the ball more (particularly on counters, tosses, and so on), he will put up big numbers in 2010.

Prediction: Expect Jones to get about 200 carries this season, surpass 1,000 rushing yards, and score approximately eight total touchdowns.  He could also approach 40 receptions.

6.  Sean Lee

I’ll be the first to admit I was hesitant about the Cowboys drafting Lee.  The ‘Boys had such a high grade on him (click here to see their draft board) that they couldn’t pass up the value.

Lee has been magnificent in camp and looks to even have a slight advantage over Jason Williams for the nickel linebacker job.  I detailed the Lee/Williams battle previously.

Prediction: Lee’s snaps will be limited with Bradie James and Keith Brooking ahead of him, but he could make a real impact as a nickel linebacker and special teams player.  He is an instant upgrade over Bobby Carpenter and any big plays he makes will be a bonus for Dallas.

5.  Orlando Scandrick

Scandrick was picked on quite a bit last season, allowing a 62.9 percent completion rate while getting thrown at on 13.9 percent of all snaps–one of the highest rates in the NFL.  Teams will continue to test Scandrick this season with studs Terence Newman and Mike Jenkins outside.

So why will Scandrick improve?  Well, this is more of a hunch than anything, but there seemed to be a disconnect between Scandrick’s statistics and the game film.  He was almost always in position to make a play on the football, yet, for whatever reason, just did not do so.

Prediction: If he can improve his ball skills in the offseason, Scandrick could come down with five interceptions in 2010.

4.  Brandon Williams

Williams missed the entire 2009 season after tearing his ACL.  In his absence, rookie Victor Butler stepped up and showed he can be counted on as a pass-rusher.  The only problem?  Butler isn’t particularly stout against the run (we gave him a D+).

It is essential for Dallas to find a quality backup behind starters Ware and Spencer, particularly with the duo combining to play 2205 snaps last season.  That number must drop, and Williams is probably better suited to handle the load than Butler.  Williams has reportedly been sensational in offseason activities.

Prediction: Last season, Butler got 125 snaps.  Williams could see as many as 200 this season, particularly on running downs.  Hopefully he can rack up a sack or two in the process.

3.  Martellus Bennett

Bennett has gained more notoriety for his off-field antics than his on-field play.  Still, Bennett’s 2009 season was actually slightly underrated because his blocking was so tremendous (we gave him a B+ in blocking and a B- overall grade).

Despite the myriad of offensive weapons in Dallas, Bennett still has an opportunity to improve this season.  His blocking ability should allow him to hold onto the No. 2 tight end job.  The Cowboys will probably run less two-tight end sets in 2010, but when they do, Bennett will be single-covered.

Prediction: 30 receptions and a couple of touchdowns is a reasonable expectation.  He should also gain more respect as a blocker as long as he can remain consistent in that department.

2.  Dez Bryant

There really isn’t much to say about Bryant.  Everyone knows he is talented.  The only person that can hold back Bryant is himself.

The key to his 2010 production will be if he can overtake Roy Williams for the starting job opposite Miles Austin.  If so, he figures to put up some big-time rookie numbers.

Prediction: I expect Williams to retain his starting gig.  Since we think Williams could have a bounce-back season, it will be hard for Bryant to rack up huge stats without a ton of reps.  If he does not start, 45 receptions for 600 yards and five TDs is still possible.  If he can overtake Williams, however, Bryant has the talent to put up 1,000+ yards and perhaps even double-digits TDs.

1.  Jason Hatcher

Hatcher is essentially the Cowboys’ No. 4 defensive end right now.  So why is he my No. 1 candidate for a breakout season?  First, like Bowen, he will gain more snaps in 2010.

Like I said above, Hatcher tallied 17 pressures last season, despite playing just 391 snaps.  This 4.40 percent pressure rate led all defensive ends.  If he can maintain that rate with increased reps, the 25 or so pressures he will acquire will put him in position for multiple sacks.

Prediction: Hatcher will increase his sack total dramatically in 2010.  Six sacks is a realistic goal–an impressive number for a 3-4 backup defensive end.


Are Cowboys’ Play-action Passes Too Predictable?

By Jonathan Bales

A few weeks ago, I published five wacky stats from our 2009 Cowboys Play Database.  The first stat in that post dealt with play-action passes:

  • The Cowboys ran only four (FOUR!) play-action passes all season with 1-4 yards-to-go.

The number of plays on the season in that range: 132. Thus, Dallas ran play-action on just 3.03% of plays in situations with just 1-4 yards to go for a first down (situations with a legitimate threat of a run). I wouldn’t call myself an offensive mastermind, but that just doesn’t seem efficient.

With 10 yards remaining, however, the Cowboys dialed up 54 play-action passes (40.90% of all play-action passes came on this ‘distance-to-go’), making it the most frequent ‘distance-to-go’ for all play-action passes (relative to the number of overall plays from that distance).

The peculiarity of these numbers pushed me to research the Cowboys’ 2009 play-action passes a bit more in-depth.  Before I continue, I must note that I made a mistake in that last post (above).  The Cowboys did run 54 play-action passes with exactly 10 yards-to-go, but that number represents 59.3 percent of the total play-action passes, not 40.9 percent.

Nonetheless, the Cowboys ran so few play-action passes in short yardage situations that they actually ran one more play-action pass (five total) with 20+ yards-to-go than with 1, 2, 3 or 4 yards-to-go.  Like I mentioned above, the offense ran a play-action pass on just four of 132 plays (3.03 percent) with 1-4 yards-to-go.  The Cowboys were in situations with 20+ yards-to-go 100 less times–32 total–yet still ran one more play-action pass (the 15.6 percent play-action pass rate in this range is five times that in the 1-4 yard range).

I could be wrong, but defenses seem a bit more likely to jump up on play-action when there is a legitimate threat of run (as opposed to 20+ yards-to-go).

Not only did the offense run only four play-action passes with 1-4 yards to go, but they also ran just 18 play-action passes with less than 10 yards left for a 1st down.  Thus, just 19.8 percent of play-action passes came with less than 10 yards-to-go.

As I pointed out in a previous study on play-action passes (which I highly recommend), the Cowboys were not particularly successful (or terrible) with their play-action passes last season.  They averaged just 0.2 more yards-per-attempt on them as compared to regular passes, but they also yielded more sacks. The situations in which the Cowboys ran play-action passes are likely a major factor in their mediocre numbers.

Of course, two other statistics regarding play-action passes contributed to the offense’s lukewarm success when implementing them, both of which I addressed before.  The first has to do with a lack of downfield pass attempts:

Of the 83 playaction passes, only four, FOUR, were attempts of 20 yards or more. That is 4.8 percent of all pass plays. In comparison, the Cowboys threw the ball downfield 20 yards or more on 46 of the other 467 attempts, or 9.9 percent of all passes.

**Note that the 83 play-action passes mentioned above and in the previous article are non-sack plays.  There were eight sacks on play-action passing plays, adding up to the 91 total play-action passes.

One of the major reasons the Cowboys only attempted a pass downfield on 4.8 percent of all play-action passes was because of the high rate of screen passes:

The most shocking statistic of all, however, is the dramatic increase of screen passes used during plays when the Cowboys showed playaction. According to our film study, Dallas ran screen passes on 33 of their 467 non-playaction passes (7.1 percent). That screen rate more than tripled on playaction passes to 22.9 percent (19 of 83 passes).

While the increased rate of screen attempts during playaction passes may or may not contribute to the relatively low overall playaction pass average, it surely did nothing to reverse the perception of Jason Garrett as a predictable playcaller.

This ‘predictable’ label is perpetuated by the high percentage of playaction passes which were thrown to the same area of the field. Of the 83 passes, 53, or 63.9 percent, were to the right side of the field (compared to just 37.0 percent on other passes).

Ultimately, we would rate the Cowboys’ 2009 play-action attack as average.  One would expect a higher yards-per-attempt on play-action passes (due to the situations in which they are generally run), but the Cowboys averaged just 0.2 yards more per pass on play-action passes as compared to all other pass attempts.

However, the Cowboys threw the ball downfield (20 yards or more) just four times on play-action passes.  Meanwhile, they attempted screen passes following a play-action look at over three times the normal rate.  These factors surely contributed to the relatively low play-action pass average.

One final note

The Cowboys did have success with play-action passes out of one formation in particular:  “Ace.”  I’ve spoken before about why the ‘Boys should throw more out of “Ace” and other “running” formations.

Of the 29 total plays out of “Ace” formation, 18 (62.1 percent) were play-action passes. This may be a bit high, but Dallas did average 10.3 yards-per-attempt on the plays.  Of course, everything seemed to work out of “Ace”–the Cowboys averaged 14.3 yards-per-attempt on non-playaction passes out of the formation.


Dallas Cowboys 2009 Offensive Player Efficiency Comparisons

Last week, a reader suggested we perform a value-based statistical analysis (similar to our 2009 Player Grades) which could be used to determine the worth of one player over another.  For example, how much better would the Cowboys be if Felix Jones played every snap at running back (disregarding fatigue)?  How costly would an injury to Jason Witten be?  Essentially, how much does each player contribute to a win?

This task is easier said than done (and since it isn’t even particularly easily said, it sure isn’t easy to do).  As the reader points out, one would have to “normalize” the conditions outside of the player to determine his true worth.  This is rather easy to do (relatively speaking) in a sport like baseball where the circumstances are basically always the same.

In football, though, no two plays are ever really identical.  Statistical comparisons among players on different teams are rather pointless, as the nature of each player’s system plays an incredible role in his statistical capabilities.

Nonetheless, there have been some attempts to “normalize” outside factors and assign an objective value to players.  In fact, we are in the process of making such an attempt right now.  Until then, we wanted to take a look at the values of Cowboys players gathered by some other leading football statistics gurus (and compare them to our own 2009 Player Rankings).

One such source (and perhaps the most well-known) is Football Outsiders.  The primary FO statistic with which we are concerned is DVOA, or Defense-adjusted Value Over Average.  FO describes DVOA as “representing value, per play, over an average player at the same position in the same game situations.”

DVOA is an excellent statistic to use to compare with our own player rankings, as both represent efficiency, not overall value.  For example, Roy Williams had a greater overall value to the offense than Kevin Ogletree in 2009, but most would argue Ogletree was more efficient during his snaps.

Another source of efficiency-based value rankings is Advanced NFL Stats–a site we refer you to a lot.  Advanced NFL Stats implements a statistic called Expected Points Added.  We’ve spoken about ‘expected points’ in the past, and ANS talks about it here.

In short, EP (expected points) is the value of a certain situation in football.  EPA (expected points added) is the difference between one situation and another.  If the Cowboys have a 1st and 10 at their own 30-yard line, for example, the EP of that situation is +1.0 point, i.e. on average, they can expect one point from that drive.  If Miles Austin catches a pass for 50 yards, the Cowboys’ EP shoots up to +4.0 (the expected points of a 1st and 10 at the opponent’s 20-yard line).  Thus, the EPA for that play is +3.0.

We are concerned with EPA/play–the amount of expected points a player adds to his team’s point total per play.

A final source for efficiency-based values is Pro Football Focus.  PFF is different from FO and ANS in that they do not necessary use the outcomes of plays to formulate rankings.  Instead, they break down each play and assign values based on their interpretation of how well each player performed his job on that play.  You can read more about their methodology here.

It is important to note that all three sites use a value of “0” as a baseline for average play.  Players in the negative are worse than average, and players with positive values are better than average for DVOA, EPA, and PFF’s values.

Click to enlarge.

The chart above displays the rankings and values for the Cowboys offense from all three sites (Football Outsiders, Advanced NFL Stats, and Pro Football Focus), along with our own grades.  A few notes before we analyze the data:

  • NR=Not Rated (likely due to insufficient sample size)
  • The statistics circled in blue are a player’s highest rating; those in red are his lowest.
  • Comparisons among players at different positions are meaningless due to the nature of the data.


  • Only two players, Martellus Bennett and Flozell Adams, were unanimously voted as “below average.”

Shockingly, ANS rated Roy Williams as slightly above average.  We love Williams’ attitude right now, but we couldn’t disagree more about his 2009 play.

  • Tony Romo’s highest rating (from FO) put him at just 7th among all quarterbacks.  PFF had him all the way at 15th.

We had Romo rated as the 7th-best quarterback in the NFL in our 2010 Starting Quarterback Power Rankings.  We would have ranked his 2009 play, though, as top-five.

  • It’s unanimous: Felix Jones is one of the NFL’s most efficient running backs.  He was ranked 5th, 6th, and 9th, respectively.

Jones’ lowest grade would actually probably come from us.  He has a long way to go to prove he can hold up over an entire season, but as far as efficiency, he’s one of the league’s best.  We provided him a B+ in short-yardage running, an A in overall running, a B in receiving, and a B in pass protection.

  • Two out our three sources agree with us that Barber was about average last season.  FO ranked him as a top 15 back.  Meanwhile, Tashard Choice checked in with a higher efficiency rating than Jones from two of the sites.

We rated Barber as an average running back in 2009 (77.2 percent).  We were also very high on Choice, rating him just 2.5 percentage points behind Jones.  Choice would have ranked as one of the league’s top running backs on Football Outsiders and Advanced NFL Stats had he played more snaps.

  • Jason Witten was ranked all the way from the league’s top tight end to No. 11.

Witten was the No. 1 tight end in our NFL Tight End Rankings.  There’s simply no doubt about it.

  • Opinions on Deon Anderson varied from slightly below average to the league’s 6th-best fullback.

We tend to agree with the latter.  The Cowboys averaged nearly two yards more per rush with Anderson in the game (as compared to John Phillips) and .2 yards more per pass.  Click here to see our in-depth study on Anderson’s 2009 play.

  • Miles Austin has arrived.  He was rated from 5th to 9th.

We gave Austin the third-highest grade of any Cowboy due to his low 2.2 percent drop rate and incredible 10.4 yards-per-attempt.

  • Ratings of both Patrick Crayton and Roy Williams varied.

Two of the three sources had Crayton as a top 16 receiver (in terms of efficiency).  Williams wasn’t high on anyone’s list, but PFF had him ranked all the way down at No. 100.

  • PFF was the only site to rank individual linemen, but their ratings fell in line with ours.

We were a bit higher on Leonard Davis and Andre Gurode and slightly lower on Kyle Kosier (who they listed as the Cowboys’ top lineman last season).  We gave Davis and Gurode “A-” grades and Kosier a “B.”  All three linemen made our list of Dallas’ top 15 overall players last season.


There is obviously quite a bit of work left to be completed in the area of advanced football statistics, particularly objective efficiency rankings.  Still, the difficulty of the task is no reason to concede.  The more we learn which statistics contribute to a team’s success (and how much), the closer we will be to “normalizing” subjective factors in an attempt to acquire objective player ratings.

Up Next: 2009 Defensive Player Efficiency Comparisons