Arguments on the Rookie of the Year seem to be favoring Robert Griffin III over Andrew Luck at this point in the season. Griffin's performance has been spectacular, certainly -- he's thrown for 2,660 yards and 17 touchdowns against just four interceptions. His 714 rushing yards on 105 carries translates to a league-leading 6.8 yards per carry, and his 6.5 AYPA is tied with Peyton Manning for third in the league.
Luck runs well, but nowhere near Griffin's standard -- he owns 216 yards on 44 rushes. Luck has thrown for over 900 more yards than Griffin and matched Grffin's 17 touchdowns. But his AYPA is a middling 5.1, 21st in the league and tied with Andy Dalton. Griffin has a 23-point EPA advantage and a 0.09 point per play advantage.
But for those who prefer to use more context-neutral stats -- as opposed to WPA, which has Luck second in the league at 4.32, over 1.5 wins better than Griffin -- there is one point in Luck's favor: the sheer volume of his output.
Washinton's' offense is obviously focused on Griffin, but the team uses the run liberally, rushing over 22 times per game (not counting Griffin's rushes). Alfred Morris carries the bulk of the load, with 1,106 yards (4.8 per carry) on the season. The Colts have rushed just 22 times per game themselves, but they've also run 117 more plays (just under 10 per game) more than Washington, and those extra plays are all directed through Luck and the passing game. All told, Luck has thrown 14 more passes per game than Griffin.
Basketball analysts have picked up on an essential point of the game: as usage rate -- the burden of the offense -- increases, efficiency tends to decrease. Steve Kerr was a more efficient scorer for the 1995 Chicago Bulls -- he shot 52.4 percent from three -- but there was no way he could take 22 shots per game with any efficiency. Jordan continued to excel even when confronted with the most difficult shots -- under duress, at the end of the shot clock, or both.
There's a similar relationship with quarterbacks. The idea that certain quarterbacks can only handle 20-to-25 throws per game with efficiency holds up in the data. Observe, the average yards per attempt given a certain attempt total (data includes all games since 2009):
The graph is smoothed out using three-attempt moving average (the R-Squared is from the raw data). With this smoothing, we can easily see the trend -- for roughly every 18 attempts per game, a quarterback tends to suffer about a half-yard efficiency drop.
However, this is only the typical efficiency progression with increased attempts. All quarterbacks will have at least a slightly different trend. Luck has been forced to throw over 50 passes on three different occasions and over 40 passes seven times. Griffin's top attempt total on the season is 39, although Griffin has ran roughly five times more per game than Luck.
Griffin, like most quarterbacks, had his biggest games on days in which he threw between 20 and 30 passes. Luck had some of his best games throwing over 35 passes and even over 45:
Griffin has lost a yard of efficiency roughly every six passes, whereas Luck has lost one just every 50. Both are a bit extreme, but both also are similar to other elite quarterbacks. Griffin's drop-off rate, for instance, is similar to those of Aaron Rodgers and Drew Brees, whereas Luck's is most similar to Brett Favre. You can try other quarterbacks (since 2009) by clicking here and checking the boxes on the right.
Luck's ability to keep his YPA above 6.5 as he goes over 50 passes is impressive. Only the league's best quarterbacks are typically allowed to sling the ball over 50 times per game. When we account for that survivor bias and look only at the middle 95 percent of games (from 23 attempts to 42 attemts), we find a one-yard efficiency drop for every 20 passes, not 37.
In that case, we'd expect a quarterback throwing 45 times per game to average 6.7 yards and one throwing 55 times per game to average 6.1 yards. Luck's high-pass games have ranged from 6.6 yards per attempt (55 attempts) to 9.0 yards per attempt (48 attempts). Griffin currently averages 8.2 yards per attempt on 27 passes per game. At 47 attempts, we should expect him to be at 7.2 yards per. At 50 and 55, the efficiency drops to 7.1 and 6.9 respectively. Luck has averaged 7.2 yards per attempt in his games over 45 attempts -- a touch better than we'd expect from Griffin so far.
The efficiency and raw productivity gap isn't completely closed through this lens -- quarterbacks throw roughly one extra interception per 33 passes, which only about halves the interception gap, something missed by YPA. YPA also doesn't capture Griffin's running ability. Accounting for the two-way threat and ball security Griffin provides, he would still be my current Rookie of the Year choice. Griffin has the ability to do enough with 20 to 30 passes to put a game out of reach, something only a few quarterbacks possess.
However, the race is closer than most efficiency metrics would suggest. Luck's ability to power the offense even with the burden of nearly 60 passes on his shoulders is tremendously valuable to his team, masking the poor running game. If Luck adds to his five game-winning drives and expands on his big WPA lead down the stretch, all while handling the bulk of the offense with even just slightly above-average efficiency, it could be an impressive enough performance to carry the argument at season's end.
Jack -
I'm curious if you also think that running the ball a lot is the key to winning football games.
Correlation doth not causation make. It seems plausible that teams which are having trouble passing the ball are going to make more pass attempts because they are more likely to be behind, and because an incomplete is more likely to lead to 3rd and long than a completed one.
Gehrig,
Running the ball EFFECTIVELY is a key to winning games. Don't mistake quantity for quality.
And again no Russell Wilson... :)
That's funny, I always though scoring more points than the other team was the key to winning games.
People used to argue that piling up rushing attempts was a good strategy for winning. They would back up this argument with lots of data showing how correlated a high single game rushing attempt total was to winning.
It's a less common argument these days, for reasons that Nate touched on.
I'd be interested to see these graphs for more QBs, specifically some of the other elite ones you named
Sampling bias is extremely prevalent in sports statistics. For example the 3rd down stats may be over representative by either good teams who are on the field more and thus face more first down, or bad teams that don't convert on 1st and 2nd so are faced with more 3rd downs. Without doing an analysis it's difficult to see the relationship between strength of offense and number f third downs. Regardless, unless you evenly sample each individual team an even amount of each particular down, you are going to oversample someone. Perhaps both and they will more or less cancel each other out.
The same is true with the pass attempts. Perhaps ONLY good passing teams tend to pass for over 40 times a game. Perhaps they only do it when they are behind, or perhaps only when in a shootout and the other team doesn't run the clock out. Perhaps they only pass that much when there are lots of INTs and QB isn't having a great game and has to go back and make up for it and confidence plays a role, or the opposing team can get more aggressive. It's difficult to determine if this represents "all" QBS and "all situations" evenly. Most likely not, but perhaps it may not matter, perhaps it all balances out and the "skew" happens enough everwhere that you end up with a distribution that would in fact represent the whole. A team like the Vikings or 49ers are simply not built to pass 35+ times a game. But if they did, perhaps because of the threat of Adrian Peterson, there wouldn't be much of a dropoff. Then again, if they did, it would probably ONLY be because Ponder is having a great game. You can't really say why stats are the way they are and as a result it's very possible that it doesn't apply to all teams. Certainly some conclusion based upon stats would seem more valuable than none and it's worth considering different concepts like "number of attempts" and try to model it the best you can. And you can only use the data that you have and it may not be easy at all times to come up with a "random" sample without reducing the sample size to the point that there isn't a valid conclusion anyways. It's always a balance, and only when you have an extremely large amount of data can you consider removing enough data everywhere so it represents an even and random sample of each situation/QB/etc
I personally favor balance to win games. The offensive lineman cannot play as well after the 60th pass attempt as they can on the 1st. I don't know if this has been proven statistically, and there are a lot of variables that may make to hard to prove. For example, if teams pass that often they are not the "typical" sample size and there is thus sampling bias.
However, certainly that is one element that may contribute to the passing efficiency declining with number of pass attempts increasing. Logically, it just makes sense though. Anyone who does any strength training and 'accentuates' the "negative" phase", taking more time on the way down on a bench or leg press or whatever than on the way up knows the amount of weight needs to be drastically reduced to do the same number of reps. Similarly, when a lineman has all the momentum and force going at him, the pressure is put on the weaker heel and ankle muscles on the "negative" phase (as the OL cannot drive forward on pass protection without risking missing the opponent), but when run blocking the pressure is put on the opponent and more evenly distributed among stronger muscles. OL are much more likely to fatigue passing the ball and if you ever talk with OL you know this.
So you need to run occasionally to keep OL fresh and run down clock. On the other hand, so what if you pass more and make more plays in a game, the opponent has to keep up with you and has to pass and even if your OL become less effective, by scoring more points early, your opponent to keep up, likely will have to as well, so perhaps this doesn't matter. However, there are other elements as well. If you only pass, you become predictable and opponent can play the pass. But of course pass is a higher yielding play. But passing also is a more variable play, and when you are down the strategy favors pass even more to the point you may become predictable, so a low variance strategy early to keep the amount of passing plays limited early and the game close so you can use the more effective pass play at more crucial moments before you give anything away and when you have the statistical edge, certainly has some merit to it. Clearly a balance is needed, but what kind of balance? How aggressive? That is a difficult question to answer and take EVERYTHING into account. You can come up with models, but there may be subtle effects that over the course of the game make enough of a difference to be a lot closer than people think.
This is a prefect example of how Not to use statistics. The author makes the flaw of 1) having a bias opinion - Luck should be considered more for MVR then 2) setting out to try and fit the data to support this premise.
In addition there is no mention of SOS!!. This is simply ridiculous & reason enough that this article should be removed from this otherwise excellent & reputable website. The single most important consideration in statistical comparison is SOS.
For example the gap btw the best passing D & worse is 2.6 NYPA!!. Given that by any objective measure
Indy is near the bottom - Brian has them 32nd & Washington is 9th & Seattle is 5th. These are objective facts. So we know from the beginning that Wilson & RG3 have had to face a much tougher road than Luck. When you combine all the other stats its clear Luck is clearly OVERrated & very lucky. & should not be considered at all in this conversation. Oh by the way Indy Pythag. exp. win % is a 5-8 record. IF Indy was 5-8 would we even be talking about lUCK? Of course not! This discussion is ONLY fueled by the fact Indy is 8-4 Which we know is a mirage & against the weakest of opposition. And, Indy's wins have come against teams 28-38 RG# vs teams 34-38 and Wilson against teams with a WINNING record!
I'm also curious about Wilson. As others have pointed out, he carries less of the burden than RGIII or Luck, but he's also played a much harder schedule, especially compared to Luck.
There's actually significant evidence that usage rate and shooting efficiency in basketball are NOT strongly correlated:
http://wagesofwins.com/2012/10/24/what-would-happen-if-they-shot-more/
Obviously, the Bulls couldn't just give more shots to Kerr - there is a limit to how many shots he can shoot efficiently. However, it's possible they could have worked to get him some more shots in a game, and it's likely they would have been more efficient for it.
However, Kerr and the Bulls are an outlier example, and anyway the guy taking the bulk of their shots was fairly efficient too. A more useful example might be, say, the Lakers of the last few years, who tended to be better when their bigs shot more and Kobe shot less. You absolutely can structure your offense to get the ball to Gasol in the post, and his efficiency there doesn't tend to decline when you give him more opportunities.
---
However, there is one enormous difference between Football and Basketball in this respect. Every player on the floor in the NBA can shoot the ball on a given trip down the floor. With rare exception, though, only one player on the in the NFL field will throw the ball on a given drive, and the alternative is running, which is somewhat less efficient. So, while an NBA player really should pass up bad shots and let teammates shoot instead, an NFL QB often has no choice but to take low-percentage throws when stuck in a bad offense.
This is a very long way of saying that that the point Nate made about that negative trend graph being correlation, not causation, is an excellent one. Prolific but poor passing performances when a team is behind are quite common.
So yeah, basically there is not much here that is very convincing. It's basically a long way of saying that Luck has thrown more.
Excellent post. Brian's colleagues at WaPo are already discussing RG3 as MVP, but remarkable as he is, it's not even clear he's the ROY.
Maybe you should post the P values and R^2 for that RG3 trendline.
Never mind, I think we know what it would show...
Luck has also had a league leading 6 INTs dropped (per ESPN). All of those drops were in games where the Colts came from behind in the 4th.
I really wish I had time to watch all of these rookie QBs. There are often little things that get lost in the stats.
I remember in the GB/SEA game, Wilson made a terrible mistake on the play before the Hail Mary. He threw the ball 10 yards short of the end zone to a guy who had no hope of getting OB. It's the type of play you get crucified for, but because he couldn't connect he was spared.
Glad to see some-one got it right, good job Nate.
This research only shows the nature of how football works.
Winning teams pass less and run more while losing teams pass more and run less.
Since winning teams pass for a higher ave per pass about 75% of the time it only stands to reason that passing often produces a smaller ave per pass because passing often is something losing teams do, and losing teams pass for smaller ave per pass 75% of the time.
To truly see if this trend has any merit one would need to see the QB who passes 50 times for a lower ave per pass, what was his ave per pass after 20 passes, then after 30 passes in the same game, chances are his ave per pass would be lower after 20 passes and 30 passes as it would be after 50 passes because the team is losing and it's losing because of the low ave per pass.
Excellent Tarr. Though one thing to add. Analysts have not picked up the idea that as usage rate increases, efficiency tends to increase. Nobody looked at individuals and saw the efficiency drop as they shot more.
Instead, advanced statisticians started talking about how the high volume shooters were overrated. The talk of efficiency being negatively correlated with attempts was horrible analyst reaction to being told that their sacred cows weren't great players.
"And again no Russell Wilson... :)"
Russell Wilson is behind both Griffin and Luck in every stat that matters: WPA, EPA, AYPC, etc etc.