## Passing = Winning

Advanced NFL Stats owes its start to an old water cooler debate: What's more important, offense or defense? Running or passing? A few years ago, I still had some statistical software left over from grad school loaded on my laptop, so I thought, "Hey, maybe these are questions that can be definitively answered." I tried to answer those questions with one of my original posts three years ago, What Makes Teams Win. When I read my older stuff, I sometimes want to cringe, but not with that one. It holds up very well, and it's well worth revisiting for newer readers, this time with more data. In this post, I'll do just that, focusing on the relative importance of running and passing.

When I was little, my dad taught me the inanity of the 'running leads to winning' fallacy. We'd watch a game on Sunday, and invariably we'd hear the announcers talk about how a team always wins when their star RB got at least 25 carries or so. They'd wax poetic about the noble nature of pure, old-fashioned, run-it-up-the-gut football. My dad would say, "Yeah, by that logic, teams should start kneeling in the first quarter. Kneeling leads to winning, right?"

But even in today's modern game, coaches continue to cling nostalgically to the run. We don't need to infer this from their play calling because as soon as they leave coaching for the broadcast booth, they spell out their running-leads-to-winning philosophies in plain English. Just last season, Brian Billick, who was the offensive coordinator for the most prolific passing offense in history, told the audience of NFL Network's Playbook NFC that the real secret to the Saints' success was their running game. Right.

It's possible to directly test the relative importance of running and passing toward the ultimate goal of winning using regression. Regression is one of those things that sounds really complicated but really isn't. If you have a relationship between two variables, say net passing efficiency and team win totals, you can plot the relationship on a graph, then draw a line that fits the relationship best. And with a little algebra we can use an equation to estimate the relationship.

If we wanted to estimate a team's win total using only its net offensive pass efficiency, I'd use: wins = 2.4 * pass efficiency - 6.6. For example, if a team threw for 7.0 net yards per attempt, that would work out to: 2.4 * 7 - 6.6 = 10.2 or about 10 wins.

Let's do the same thing with running efficiency.

Looking at the relationship of running efficiency (yards per rush attempt), a few things stand out. First, compared to passing efficiency where the efficiencies typically range from about 4.5 to 7.5 net yards per attempt , the range of running efficiency is much tighter. It typically ranges only from about 3.5 to 5.0 yards per attempt. Second, the pattern formed by the relationship with wins is just a big blob. Compared to passing with its nice tight relationship, running's relationship with wins is weak. There may not be a relationship at all, except that there is a conspicuous absence of many teams with very strong running games and very low wins. To my eyes, this suggests that a good running game can save a bad team from notching only 3 or 4 wins, but it's not going to lead a team to the playoffs. Notice how few of the very best running teams exceed 8 wins, while the best passing teams win at least 8 games.

In fact, if you look at only the points at or above the 10-win line, where the playoff teams typically  reside, the graphs tell very different stories. Teams with 10 or more wins are no more likely to have strong running games than weak or even extremely weak running efficiency. In contrast, the teams with at least 10 wins are far more likely to have had above-average passing efficiency.

Comparing the two regression equations tells us a lot. For every yard of improvement in passing efficiency, a team can expect 2.4 additional wins. But for every yard of improvement in running efficiency, a team can only expect 1.1 additional wins.

When you have more than one variable that contributes to an outcome, its best to use multivariate regression. You can't visualize the regression model graphically, but you can produce an equation like the ones we see above. For the dependent variable, I'll use team win totals. For the estimator variables, I'll use efficiency stats for running, passing, turnovers, and penalties. This model is intended to be explanatory rather than predictive, so things like fumbles lost, which involve a large degree of randomness, are included.

There's one more thing I'm going to do before the regression itself. To truly compare the relative weights of the variables, I "normalized" them. This adjusts each variable so they're all on the same unit-less scale. Normalizing sets the mean of each variable to 0 and its standard deviation to 1. For example, a team that averages 4.8 yards per carry would be 1.5 standard deviations above average, and would have a normalized running efficiency of 1.5. Normalizing allows us to the answer the question, 'Which aspect of the game is it more important to excel in?"

The table below lists the result of the regression.  Each estimator variable is listed along with its coefficient--its relative weight in terms of its importance in winning.

 Normalized Variable Coefficient Constant 8.00 Off Passing 1.54 Def Passing -1.23 Off Running 0.44 Def Running -0.44 Penalty Rate -0.25 Off Int Rate -0.32 Def Int Rate 0.45 O Fum Lost Rate -0.43 D Fum Lost Rate 0.25

All variables were significant well beyond the p=0.05 standard, and the model accounts for over 82% of the variance in team win totals, meaning we can be fairly certain the estimated weights are roughly accurate.

Here's how you can interpret the results. The constant means we start at 8 wins for a completely average team. For every standard deviation above average in passing efficiency, a team will win an additional 1.54 games. And for every standard deviation above average in running efficiency, a team win an additional 0.44 games.

Right away we can see that passing efficiency is the most important aspect of performance. At first glance it's three times as important than running efficiency. But that's before we factor in interception rate, which when added on top of passing efficiency makes passing over four times more important.

Fumbles (and fumbles lost) occur just over twice as often on pass plays than on run plays, usually due to a sack. So it's not clear how to apportion fumbles in the run-pass comparison. But even if fumbles strictly happened on run plays, it still wouldn't make running more than half as important as passing.

On the defensive side of the ball, we see the same general relationship. Stopping the pass is several times more important than stopping the run.

Theoretically, running should be just as important as passing due to game theoretic considerations. That's what is loosely meant by the adage "the run sets up the pass" and vice-versa. But despite this well-worn cliche, coaches and coordinators simply overdo the "setting up" part by over-playing the run on both sides of the ball. It's no different than a boxer who jabs too much.

It's not that running doesn't matter. It's just that passing is far more important. Running successfully can be critically important near the goal line, where the short field makes passing very difficult. Running also helps ice games in which a team has a lead, but that implies you need to somehow gain the lead in the first place.

Occasionally we do hear analysts on TV and in the newspaper columns refer to the NFL as a "passing league," but it's often meant as a criticism intended to chide teams for running running often enough. It's puzzling that in a sport watched and dissected by so many people, we continue to hear calls for teams to stick with the running game.

If I were advising a general manager, I'd tell him to largely forget about the run. Get a RB who's good at picking up blitzes or catching the ball.  Never draft a RB in the first few rounds, and whatever you do, don't waste precious cap space (or payroll budget) on him. Get a quality QB at all costs. Assess your linemen on how well they pass block, and don't worry as much about their run blocking. Get lots of pass rushers on defense. Got a LB that's a great run stopper but can't play coverage? Trade him to some sucker team that cares that they only give up 3.8 yards per carry rather than 4.2 yards per carry. That's how you build a perennial playoff contender.

### 48 Responses to “Passing = Winning”

1. Anonymous says:

isn't there a problem by using efficiency numbers for this kind of comparison?

good teams more often find themselves in a situation where they are happy to run out the clock. so they run the ball up the middle for no gain to drain the clock and hence decrease their rushing efficiency.

bad teams on the other hand find themselves trailing early and often which forces them to throw a lot more to keep any chance alive for a comeback. with these desperation passes they decrease their passing efficiency.

this also works the other way on the defensive side of the ball. the good team with the big lead is willing to allow short and medium gains but wants to protect against the deep ball. this leads to an increased rushing efficiency for the bad team. and the bad teams priority on defense is to deny a first down which makes them vulnerable for the deep ball improving the good teams passing efficiency stats.

i think it would be interesting to see if the result is this clear if you only use first half numbers.

2. SportsGuy says:

You don't mention sample size here. What was it?

3. Dave says:

It strikes me as odd that the coefficients are not more symmetrical. For instance, why aren't the Off Passing and Def Passing simply equal with opposite signs? Every pass that is used for the Off Passing data series is also used for the Def Passing data series by their opponent. It seems strange that my team's 10 yard pass would increase my team's probability of winning more than it would decrease my opponent's probability of winning. I realize this isn't a single game win probability model, but all of the game outcomes are equal and opposite as well, so the logic should still hold.

Also, if you used data with ties, how did you account for those? 0.5 wins for each?

4. Brian Burke says:

.5 for wins.

Sample was all teams from '02-'09. n=256.

5. Giles says:

Anonymous: It would be possible to discount any short runs designed to kill the clock by only using runs from the first 3 quarters or to ignore any plays in the last 5 minutes when the team in possession is winning.

My only question would be that if this were a real team then the defense would always know to line up in coverage for a pass play. Wasn't there an earlier post that talked about the game theory of play calling whereby you need to call Run enough to prevent the defense from always lining up against Pass?

6. Anonymous says:

"If I were advising a general manager, I'd tell him to largely forget about the run. Get a RB who's good at picking up blitzes or catching the ball. Never draft a RB in the first few rounds, and whatever you do, don't waste precious cap space (or payroll budget) on him. Get a quality QB at all costs. Assess your linemen on how well they pass block, and don't worry as much about their run blocking. Get lots of pass rushers on defense. Got a LB that's a great run stopper but can't play coverage? Trade him to some sucker team that cares that they only give up 3.8 yards per carry rather than 4.2 yards per carry. That's how you build a perennial playoff contender."

I agree in principle, but consider the following:

1) You can't be flat out terrible against the run. If you are then you end up with the situation the Patriots had against the Ravens in the wild card round last year. You have to at least have competence.

2) Good QBs don't grow on trees. Sometimes you can't get one even "at all cost". Teams don't let good QBs in their prime hit the free agent market. Sometimes having a game manager and playing good defense is the only option a team has to win.

3) I agree you shouldn't get RBs in the first round generally, but having a genuine play maker at the RB position is huge for a team. I'll take CJ, AP, or RR any day of the week.

4) As for linemen its all about the position they play. Tackles need to be focused on pass blocking but interior linemen should be good for both the run and the pass.

7. Isaiah Hunter says:

Perhaps the more interesting study would correlate running attempts to passing efficiency. This would answer the age-old question regarding running making it easier to pass.

8. zlionsfan says:

I don't know that having a star RB is necessarily as big as you might think ... either that or perhaps last year was not the season to use as an example. Baltimore needed to win a tiebreaker to squeak into the playoffs, and Tennessee didn't even have that chance. As for the Vikings, well, it could be argued that the improvement they made from 2008 had nothing to do with the running game. (And if you check out the WPA for the 2009 Vikings, it'll make that argument very clearly.)

As little as we may want to admit it, having a star RB simply isn't that important. A QB can do far more for a team than a RB can.

9. Anonymous says:

Success rate has always seemed to be a much better metric for run efficiency than raw yards/attempt. I don't know if you have a dataset to do so, but what does the second chart look like w/ success rate in the X, instead of yards/attempt?

10. Carson says:

Brian, I'm curious about a couple of things, as follows:

1. The Definition of "Quality QB" -- Is it possible to say with any accuracy what the traits of a quality QB are? Is it possible to weight them with any degree of accuracy, as you have with the relative worths of running v. passing?

I'd expect that some combination of decision-making ability and arm strength/accuracy would have to be considered here. Is there any way to quantify that, or is this largely the province of scouting?

2. The Value RB -- I wonder which RBs are providing the most value at the moment -- that is, which RBs' are doing the most blocking and pass catching for the fewest amount of dollars.

11. ASG says:

All this and yet the QB who has led the league in passing has yet to win the Super Bowl that year.

12. Dave says:

Adding to my previous comment about symmetry in coefficients: shouldn't there be another variable in addition to the Penalty Rate? Like maybe "Opponent Penalty Rate"? The variables are supposed to be things that the team itself can control, but to some degree things like pass interference and offsides are drawn penalties.

Perhaps Penalty Rate should be replaced by Offensive Penalty Rate Difference (Offense - Opponent Defense Penalty Rate)and Defensive Penalty Rate Difference (Defense - Opponent Offense Penalty Rate). This should clear up the symmetry problem I mention above.

Hey Brian are you using ANY/A or just NY/A? I'm guessing it's ANY/A to take into account sacks.

Anyways, using the formula wins = 2.4 * pass efficiency - 6.6, that would put that Jets 2010 season at a win expectancy of just 2 wins.

I wonder if their +7 wins over expected would be the highest of this era? I would have to guess it would be.

Thanks

14. Chase says:

Insightful stuff as always, Brian. But as I've stressed several times on the PFR site, I don't feel comfortable using YPC to measure rushing efficiency. YPC is *highly* sensitive to long gains, which are both random and loosely correlated to a "good running game." The '60s Packers and the '90s Cowboys weren't about big gains. I just think that the statistical mean does a much, much better job measuring passing games than rushing games.

15. Brian Burke says:

Great comments. I think I'll need to merge my play-by-play database and my team-level stat database and test out some of the suggestions here.

16. Ian says:

"2) Good QBs don't grow on trees. Sometimes you can't get one even "at all cost"."

Great QBs don't grow on trees, no. But good ones are always available. What you need is a really good pass blocking offensive line. When you see a Brady or a Manning picking a defense apart, what most people don't see is that the pass rush never lays a finger on them. Any QB can look good when they've got all day to pick their throw. Teams with QBs with horrible stats tend to be the ones that can't provide any decent protection for them.

17. Anonymous says:

@IAN

By the same token though QB's make the line look better by making quick decisions or adjusting protection, etc.

Example: Warner in ARI. That line wasn't good.

"What you need is a really good pass blocking offensive line. When you see a Brady or a Manning picking a defense apart, what most people don't see is that the pass rush never lays a finger on them."

I feel strongly that a great offensive line is vastly overrated. Being a former lineman myself, it's not that I don't appreciate what they do. But a majority of sacks can actually be attributed to the QB - not the O line.

Everyone kept whining that David Carr might be a better QB if only he weren't getting massacred behind that horrible Texans line. David Carr leaves, and magically the Texans' total of sacks allowed decreases. Meanwhile, David Carr moves on to other teams and every time he gets in the game, he routinely gets sacked.

Being a Jags fan, I am painfully aware of Peyton Manning's prowess. The next time you see him on TV, count the number of seconds it takes him to release the ball. It's insane. You could put him behind the local high school offensive line and he would only take a few extra sacks. He knows how to either make the throw or throw the ball away.

19. Ian says:

I'm not saying there's no difference between a poor QB and a great one. You do get the "Peytons" who only need a second or two to make their throw. If you get one of those QBs then you count your blessings.

But given how expensive it is to get one of these great QBs, it seems like you'd be better picking up a good QB and buying them an extra half second or so per play with a great O-line.

@Ian

True.

21. Jim Glass says:

Anyways, using the formula wins = 2.4 * pass efficiency - 6.6, that would put that Jets 2010 season at a win expectancy of just 2 wins.

Have to remember the defense too.

That said, the Jets last year had the #1 D overall, #1 D against the pass, and a well above-average running O ... and still went only 8-7 in the regular season, not counting the game the Colts decided not to play for the last 20 minutes. Which shows the drag one player, the 30th rated passer, Sanchez, was on the entire team, even though most Jets fans are happily oblivious to it thanks to the great season he carried the team to.

Objectively, that great season consisted of the Colts effectively throwing that game to move the Jets from out of to into the playoffs, then beating a very mediocre Cincy team, and then winning one good game on the road against the Chargers.

All this is by way confirming the importance of the O passing game. With all the hype the Jets are generating this year, Sanchez had better pick it up a lot or there is going to be a lot of disappointment at some point.

This is not a knock on the Jets or Jets fans. I'm a Jets fan who's old enough to remember Browning Nagle as the great promising hope of the future.

22. Anonymous says:

I'd be curious to see what it looks like 20 years ago, if your numbers stay close or if running was more important back then. Obviously, the league has done a lot in recent memory to help the passing game, since the QB is the "star".

23. Anonymous says:

Also, what Chase said has some bearings. There isn't as good of a way to measure the run, thus the uncorrelated blob that the run game stats showed.

24. Unknown says:

Quality QB... running back who excels at picking up the blitz... great pass rush... perennial playoff contender!

Did you consult with Bill Polian on this article? Because you just described the Colts perfectly!

Joseph Addai gets a lot of flack for being a sub-par running back, but he pass-blocks better than most in the league, which is why he's still the starter in Indy.

25. Anonymous says:

LOL makes me feel pretty good about the Broncos. They (I can't say we lol I belong to Denver Dojo not the Denver Broncos) have improved upon an elite secondary. Elvis leaving the house has sucked, but won't cost more than one game.

I'd like to see a prediction based upon the idea that Kyle Orton will have to spread it out to many recievers. Does the spreading of the ball have any significance on win percentage?

i.e. Orton forced it to Marshall way too often last year, as he never had that good of a reciever to throw to. On the other hand, Marshall didn't catch too many touchdowns....not impressive for Coach "Brady and Moss setting records" McDaniels.

Posting as anonymous coward, but I am JudoJohn

26. James says:

Wonderful stuff Brian
Three suggestions

1. Should there be an interaction term in your regression equation to see if having a good rush and a good passing game leads to more wins than simply adding the two as if they dont effect each other. Similarly are one dimensional teams (good pass poor run or vice versa) better or worse then simply adding their individual coefficents together. Do you also need an interaction with defense which might reflect the need for teams with poor defenses to pass a lot and get garbage yards at the end of the game. This could answer the "pass sets up the run and vice versa" game theory question.

2. Why not use your expected points and/or your WPA metrics to see what the average on a pass play is vs a run play you can also add in special teams play (maybe this is what you hinted at above). This would also fully account for interceptions and the higher fumble rate on passing plays. I think this could close the gap as I think plays like a 2yd rush on 3rd and 1 are positive plays that teams want to be able to do regularly but is penalised by average efficiency. Similarly running down the clock in the 4th quarter with a lead is not optimum from an efficiency or even an expected points view point but is for WPA.

Alternatively all this might not effect your main result that passing is still more beneficial than rushing which would be really fascinating. I think this would show that defenses are not focusing on the pass enough from a game theory point of view. Therefore MY advice to the GM to stock up on pass rushers and DBs with the draft picks he has saved from not picking a blue chip .

3. Can you also assign penalties to run plays or pass plays which are linked to skill (or lack thereof) e.g holding, off/def pass interfeence intentional grounding, although some will be non-specific or unassignable (offside personal fouls.) Or do you treat a 10 yard holding penalty on a pass play as a type of sack.

James

27. Ian says:

Brian,

I had a very quick and dirty look at some of the points mentioned above (in particular, the 'running into the line to eat the clock' effect of good passing teams that have a lead). There does seem to be a slight negative correlation between pass efficiency and rushing efficiency, which we would expect if there was a killing the clock effect.

I wonder whether there's any value in looking at offensive efficiency stats for teams when tied/trailing and when ahead (from the PBP data), and whether there's any significant difference between the two.

28. Martin says:

You have a number of insightful observations here. I'm glad to see someone challenging the wisdom of announcers about the run. Too few people question what they hear. I'm also excited to see data analysis used to back your claims.

Passing skill is more influential because the variance in passing outcomes is larger. Very, very few runners are capable of turning no gain into a first down on a regular basis. Those that could are in the Hall of Fame. By contrast, a skilled passer will turn an incompletion into a first down several times each game. Woody Hayes was correct about the pass, but wrong about the implications of his famous statement. Passing is critical precisely because it's a high risk, high variance strategy.

You bring up game theory, but your conclusion does not account for it. In equilibrium, the expected payoff from a running play is equivalent to that of a passing play. If your running game stinks, then your expected payoff from throwing the ball will also stink. It would be accurate to say "invest fewer resources in the run and rush defense". "Largely forget about the run" is far too strong a statement that is not justified by the analysis.

Your conclusions are limited to the regular season. You are maximizing the wrong function (regular season wins) in your analysis. Fans want teams to to maximize playoff wins, subject to the condition of making the playoffs. Ignoring rush defense is a recipe for playoff disaster, as the playoffs will be loaded with strong rushing teams that can exploit a weakness in rush defense.

Penalties only matter at the margin, so it's unsurprising that they are not decisive in games between mismatched teams. Your current data pool is polluted with games played by inferior teams. If you look at the cases where penalties should matter, their influence probably will re-emerge.

29. Unknown says:

Martin--you're spot on about what fans want, it's why Jets fans are so optimistic about this season. I'd love to see BB test another one of FO's mantras, that by and large playoff games are won in non-repeatable ways, because the teams are in general so good that clock management and game-day coaching generally, as well as luck, the kicking game, etc.

30. Anonymous says:

I think a good way to back the "passing is more important than winning" argument is to use that sample that has been used alot for the passing EP studies; 1st and 3rd quarters, score within 10 points, 1st downs. To me this data set represents when teams truly have the option to choose their play type. Using the PBP data Brian posted from 2002 to 2009 I plotted each teams 1st down run percentage (in the said sample) with that teams winning percentage. There were 32 teams multiplied by 8 seasons, so i had 256 data points. The linear fit for the graph shows a slope of 0.4472 upward (Winning percentage = .4472 * 1stdown running percentage + 31.126). The R2 value was .0246, which is really low but still statistically significant to at least the p = .025 level (the tables only went up to sample size of 120). I would be curious if anybody could duplicate my results.

Note that, if this is true, it doesn't depend on how effective the given team is at passing, by simply choosing the pass there is a higher expected win total. Even though teams with good passing systems will pass more often, teams with good running systems will also run more often. To me this graph says that the teams that choose to pass more win more games than the teams that choose to run more

Also, Brian, did that you ever get that site running where we can post our analysis?

Thanks, Andy

31. Jeff Clarke says:

"Ignoring rush defense is a recipe for playoff disaster, as the playoffs will be loaded with strong rushing teams that can exploit a weakness in rush defense."

I'm not sure this is true. I think one of the basic points this article tries to make is that the good teams (ie the playoff teams) are the ones that can successfully throw the ball. The "playoffs are all about rushing" theme has always struck me as mostly cliche and not at all based in fact.

I admit I haven't looked at this in depth (perhaps Brian or someone else will) but I doubt that the relative differences in stopping the run vs stopping the pass is any different in the playoffs.

32. James says:

Brian
To try and address one of the questions I posed yesterday. I have played around with your expected points tables for each down and distance and I realise that the yards needed for a first down are twice as valuable as the yards gained after a first down at least on first and second downs. Does this make the lower variance of rushing more valuable despite the lower average yards gained?

For example if you look at the effect of different yards gained on first down by graphing the expected points of 2 and 10 from the 50, 2 and 9 from the 49, 2 and 8 from the 48, etc the expected points rise at 0.14 points per yard until you reach 9 yards but once you make the first down the expected points rise at the usual 0.07 points per yard (until you reach the 10 yard line). The same is true for the results of second down plays. (Third down and fourth down DONT show this pattern, if you assume the team will kick/punt on 4th if they fail on third or work out the cost of the turnover on downs if they fail on 4th).

Should teams rush more on first and second down but pass on third down?

As a different way to look at this at first and 10 on the 50 yard line a 5 yard gain gives second and 5 on the 45 which is worth 2.19 points. But if alternatively you attempted a 15 yard pass which is completed 50% of the time this gives an expected points of the average of first down on the 35 (a completion) or 2nd and 10 on the 50 (incomplete) the expected points for these two alternatives are 2.96 and 1.46 which averages at 2.21. So a guaranteed 5 yard gain on first down is worth almost the same as a 50% chance of 15 yards!! This makes play efficiency highly dependent on down and distance as on first and second down "required" yards are much more valuable than "excess" yards.

Obviously there is no such thing as guaranteed yards but given the lower variance of rushing I think this makes it more valuable from an EP perspective than a similar efficiency passing which might explain some of the difference between the two "efficiencies" just looking at yards.

Also can I echo the question about any progress a site to post our elegant analyses /dump our crazed rantings made possible by your pbp data?

James

33. willkoky says:

I agree with everyone who wants to see junk time eliminated and explore if a good run game sets up a good pass game.
I don't know if I agree with the guy who wanted to use the leverage of 3rd and 1 because I have not seen it proven that a team can be really good at 3rd and 1 conversions. They could be lucky or just good at running in general. Converting 3rd and 1 may not be a repeatable skill.

34. James says:

@James - 5 yards per carry is a lot! Only 11 players with more than 80 carries had 5 y/c or more last year, and Rice/McGahee and Williams/Stewart play for the same team.

As a whole, last year only one team had over 5 y/c (the Titans of course) and only 9 teams since 2000 have managed it. You might recognize some of the associated names - Vick and the Falcons twice, Peterson in '07 and Culpepper's Vikes in '02, Chris Johnson last year, Priest Holmes in '02, LT in '03, Ahmad Green in '03 and the '08 Giants.

I think one of the main advantages passing has over running is the number of options. When you run, the RB has the option of hitting the hole, continuing down the line of scrimmage, or cutting it back. A QB has the option of throwing it to any one of three to five receivers in addition to scrambling and throwing it away.

This is where James' comparison falls apart above - a QB isn't limited to just one 50/50 15-yard pass, if it isn't open he could check down and get a 4-8 yard, high-percentage completion (or bomb it deep, move around the pocket to stall, or scramble).

35. zlionsfan says:

especially given that the trend in modern offenses is toward the 4- to 8-yard pass. In 2008, according to the FO data, the mean distance to target was just over 6 yards past the line of scrimmage ... comparing a 15-yard pass to a 5-yard run is most likely weighting it very heavily in favor of a running play. The mean yardage gained on first and 10 was a bit under 4.6 yards, though, so it's not off on the running side by that much.

In a general sense, I would say teams should be passing more on every down. I think it's still necessary to "set up the pass" in terms of keeping the defense aware that you will run from time to time, but I think most OCs spend series after series setting up the pass even when the pass is already working just fine.

36. Ian says:

I've just been looking at playoff game probabilities, based on regular season efficiency stats, for the 2002-09 seasons. This gives 88 games to use for the regression.

The only significant variables are venue (which we'd expect, the better teams play at home in the playoffs) and, strangely, 'fumbles lost rate' (so teams that lost more fumbles in the regular season tended to lose in playoff games).

So to Martin's suggestion that ignoring rush defense would cost teams in the playoffs, it doesn't seem to be the case in any significant way. The best way to win in the playoffs is to get home-field advantage (and don't caugh up the ball). That means, win more regular season games. Hence Brian's suggestions to maximise regular season wins are the best way to also maximise chances of playoff wins.

37. Anonymous says:

It's been a while since my last statistics class, but if we were to add the absolute values of these coefficients, it seems like those related to offense are larger than those related to defense. Does that mean that offensive efficiency is more important than defensive efficiency?

38. Anonymous says:

I like where you're going with this, and it's nice to see stats applied to everyday questions like this.

As a researcher, I would recommend a couple of things to think about.

First, your current model appears to be estimating whether the average carries or passing yards predicts game wins. To answer your original question (is it better to run or pass?) you should run the model with attempted carries versus passes (possibly as a ratio). That will tell you whether the extent to which a team *attempts* to carry the ball (vs passing) leads to wins which is what I think you are more interested in.

Second, I realize that you are not trying to do a scientific study, but if you want to refine your approach, you should consult a statistician about how to account for the fact that your dependent variable is not typical. What I mean is that "number of wins" is (1) a count variable which has some unusual properties, and (2) each win is dependent on another team losing.

Best of luck with your investigation! I'm enjoying it.

39. James says:

zlionsfan and James

I never said that rushes average 5 yards or that passing is an option between a 15 yd completion or incomplete. I actually never even said that the 5 yard gain was a run.

I was trying to show that
a expected points can produce very different results to yards per play.
b expected points rewards low variance plays on 1st and 2nd down
c this fact could help reduce the apparent gap between passing and rushing.

Let me rephrase my example.
A team's gameplan dictates that on 1st and 10 on the 50 they will call either a screen play or an out route pass.

In a game they had 3 such opportunites and called 1 screen play and 2 out routes. This resulted in a 5 yard completion on the screen play and an incompletion and a 15 yard completion on the two out route plays.

One assistant coach analyses the results by passing efficiency and finds that the out route is more efficient than the screen (7.5yds vs 5yds)
Another assistant coach analyses it by expected points and finds that the two approaches produced almost equal results in the game at average of 2.20 expected points after the two types of plays.

James

40. Steve O says:

As much as you and other commentators (usually rightly) rip Colts president Bill Polian, he evidently has this figured out. Your description of the most effective team pretty much describes the Colts. Good passing, terrible rushing. Good passing D, terrible rushing D, and that's where their money and strategy are year after year.

41. Matt S. says:

I don't think "average" is the really important thing. Some teams pass as their bread-and-butter, but run their way out of short-yardage situations, or run the ball just to keep the defense from playing the pass exclusively. In this case, it's important to avoid runs for loss, but if you're getting a bad YPA, your run game could still be doing its job.

Also, once you've maxxed out your passing game, focusing on your run game makes a lot of sense. After all, smart opponents will do the best they can to stop you from throwing (pass defense being Job Two and all). Being able to run can help to thwart that.

Or maybe it's just that my team had a killer passing game and bad rushing YPC last year, won 13 games on its passing offense and ability to convert or alternate on the ground, and got wiped out on a 50/50 game in the playoffs when they stopped the pass, but couldn't run, and couldn't stop the run. A good halfback, and they would have won that game going away. I agree that an air game matters more....

42. Bob Weber says:

As a Broncos fan, I have to disagree with Anonymous above. If McDaniels was such an offensive genius it might have been worthwhile to hold on to his pro-bowl passers and receivers as you so astutely put it, "at all costs".

I've always thought that the 'establishing the run' mantra was backwards. If you can get a good passing game going it opens up the running game. The passing games of Troy Aikman, Kurt Warner and John Elway did wonders for Emmit Smith, Marshall Faulk and Terrell Davis.

43. Anonymous says:

This analysis actually didn't tell me very much new. I thought that it was established in the 1970s that 7 yards per pass attempt was the golden rule of winning a football game. (I assume that this is highly correlated to the net passing efficiency measure.)

Passing efficiency means that you do it really well when you do it. The early 1970s Dolphins had great passing efficiency, but threw only 12 times a game.

The stats say tilt toward a better passing game, which I agree with completely, especially with today's rules that favor passing. But the great teams do both.

44. JMM says:

The role of game theory in this should not be lost. With all the rules tilting to the pass, the Steelers defense, as well as other good defenses, still focus on first stopping the run. This is because then they can focus on the pass.

Offenses need to be able to run the ball to keep the defenses "honest."

45. Borat says:

Dear Brain:

In my country we have problem and the problem is no dads.

my dad also taught me the inanity of the 'running leads to winning'. But in my country survivng is the same as winning. So when fight come he taught me "running leads to not dying, yes? He sais that only fools are warriors. That's why Allah gave us the US to fight our wars. Nice yes? Why should we fight when we have US as one of our alleys?

The US has the 2nd best Navy Air Corps and the 7th best oral prostitutes. And we thank you.

Borat

46. Anonymous says:

This analysis really ignores how much "good passing" is really a result of favorable down and distance -- specifically staying out of 3rd and long.

I've heard Peyton Manning specifically say that what he hopes to get out of his running game is "3rd and manageable" situations, because by his own admission it's just that much harder to pass on 3rd and long.

I also know that Bill Belichick has talked about how he places such an emphasis on stopping the run on 1st and 2nd down, because he knows that creating obvious passing situations gives his D an advantage.

47. Anonymous says:

To add to the above, Peyton Manning's career stats on 3rd and +6 show:

617/1046 (59.0%) 7600yards 36TDs and 44INTs = 75.5 QB Rating

Yet Peyton's career stats on 3rd and less than 6 show:

460/696 (66.1%) 4547yards 68TDs 20INTs = 105.0 QB Rating

I also eyeballed the splits for Brady, Rivers and Favre and the same trend exists. So running may not win the game but it's fairly unwise to suggest that a GM should just "forget" about running. Like you said, running is low variance so it probably helps to have a RB who can all but guarantee you 4 yards a carry if only to set up a "3rd and manageable," especially if there was an incomplete pass on 1st down.

Also, if a 3rd and long makes Peyton Manning a 75.5 rated QB -- then you'd better care about not giving up more than 3.8 YPC. So you might want to hold off on trading away those run stuffing LBs to one of those "sucker" teams...like the Patriots. ;)

48. Dave W. says:

Back in 1987, Barra and Ignatin (http://www.amazon.com/Football-Numbers-1987-Allen-Barra/dp/0133242609) identified yards per pass attempt as the single biggest factor in determining who won NFL football games, far more important than running efficiency. If I remember their "adjusted yards per attempt" formula correctly, it was (total passing yards - 40*interceptions)/attempts. They didn't give details of whatever regressions they had run, but it was pretty clear that they were using something like that to come up with their analysis. So this has been true for the NFL for at least the last 25 years, probably more.