If you came up to me and handed me $100 to bet on this Sunday’s Super Bowl, how would I bet it? Before I answer this fun and completely hypothetical question, I want to say something that is anything but fun or hypothetical.
Gambling is dangerous. As radio host Colin Cowherd repeatedly says on his show on ESPN Radio, you should never gamble more than you can afford to lose. The fear of becoming addicting to gambling is one of the biggest reasons that I’ve only set foot in a casino a handful of times since turning 21. It’s fun to win and the hope that the next hand will turn around a spell of bad luck can be the easiest lie we ever tell ourselves.
Please think long and hard before placing any money on any sporting event. If you have second thoughts about the wisdom of making that bet, you probably shouldn’t be placing it. On that note, I consider myself more than a casual NFL fan but far less than an expert. I’m looking to have some fun with a few of the many available prop bets associated with the Super Bowl; I’m not looking to make money. If you use my suggestions to make bets, you are assuming all of the risk as I am not claiming to know anything more than your typical NFL fan. To reinforce the fact that this is completely a thought exercise, I will be placing bets in simoleons, the currency of Sim City.
PROP BETS
The Super Bowl very clearly shows that the United States is simply obsessed with two things; the Super Bowl itself and gambling. Before going any further, I’d suggest reading the following piece by Bill Barnwell on Grantland.com about a few various prop bets associated with this Super Bowl.
http://www.grantland.com/story/_/id/8889349/bill-barnwell-prop-bets
There is one section that is especially important in this piece that I’m just going to quote here because before reading Barnwell’s piece, I had no idea what a line of +700 meant.
This is Barnwell’s explanation of what it means when the line of there being overtime is +700 and the line on there not being overtime is -1000:
“The "+700" figure next to "Yes" means that a bettor would receive $700 back in profit if they bet $100 on the event occurring before it actually happened. If you held this winning ticket with $100 and brought it to the counter, you would be handed back $800 (the $700 profit plus your initial $100 bet). The "-1000" figure next to "No" replaces the positive sign at the beginning with a negative sign; it indicates that you have to bet the dollar amount in question to win $100. In this case, if you wanted to win $100 betting against the possibility that there would be overtime, you would need to bet $1000. If there's no overtime, you would win $1100 (the $100 profit plus your initial $1000 bet).”
In other words, if you’re betting on a line with a negative number, you’re betting on an outcome that has a better than 50/50 chance of happening. In this context, it makes complete sense. NFL games rarely go to overtime so betting on there NOT being overtime and being correct only yields 10% on your bet.
MY BETS
In a lot of ways, I view the thought of determining which bets to make a lot like I would view a roulette table before the ball is rolled. It seems logical to me to bet smaller amounts on some longshot outcomes while betting larger amounts on more predictable outcomes to hopefully balance out the missed longshots and keep me above even. My goal is making these picks is not to maximize my potential outcome; it’s to maximize my outcome while minimizing my exposure to long odds bets.
Obviously, the list of bets that Barnwell puts forth isn’t comprehensive but I’m going to pick from that list and the list at one other website I found. First, the favorites:
MENTIONS OF THE “HARBAUGH BOWL, “HARBOWL”, OR “SUPERBAUGH”
OVER 2.5 (-110)
This seems like a complete no-brainer to me. Somehow there are people out there that think the broadcasters are going to resist using one of these catchy terms once or twice at most? While it might not be Phil Simms or Jim Nantz that goes here, you have to remember that the broadcast team includes the sideline reporters and the crew doing the halftime show. My bet is 20 simoleons on the over.
WILL THERE BE A SUCCESSFUL 2 POINT CONVERSION?
NO (-500)
In the first 46 Super Bowls, teams are 6-13 when going for two and those attempts have come in 8 different Super Bowls. However, for these purposes it must be considered that the San Diego Chargers were 2-2 in Super Bowl XXIX. Therefore, there has only been five Super Bowls with a successful two point conversion, or 10.9%.
A line of -500 implies that something has only an 83.3% chance of happening when, according to Super Bowl history, it actually has an 89.1% chance of happening. Statistically speaking, this bet feels like a very safe one. NFL coaches don’t like going for two unless they’re completely desperate and the game is basically over but the Super Bowl is the one venue where some coaches might be willing to put it on the line.
The only reason this bet doesn’t feel terribly secure is that all of the two point conversion attempts have occurred in the past 18 Super Bowls. If you use that as your sample size instead of all 46 Super Bowls, then the “yes” bet looks like a steal. This can be considered the “entertainment” portion of the post when you read along and watch me talk myself out of a particular bet…
They wouldn’t, would they? Two great running games, plus Colin Kaepernick… plus it looks like Jim Harbaugh is willing to take this kind of risk…
Oh, this is just icky… I’m putting 20 simoleons on No but this really doesn’t feel safe.
WILL THERE BE A DEFENSIVE OR SPECIAL TEAMS TOUCHDOWN?
NO (-200)
I’m just not feeling this one. I feel like the two quarterbacks are playing well enough right now to avoid a pick-6 (this is the point where I remember Kaepernick throwing one against Green Bay) and I don’t feel like either team has a particularly dynamic special teams to be able to bring one back.
I know that Baltimore gave up not one but two special teams TD’s against the Broncos but I saw a very different team in the second half against the Broncos and against the Patriots than I did in the first half in Denver. They’re playing very well and I just don’t see them giving up another one. Sometimes it feels like defensive and special teams TD’s come in bunches and I can’t help but think that for the 2012/13 playoffs, we’ve already seen the bunches.
On top of all that, -200 seems like a pretty favorable line. This seems like a fairly high reward for an outcome with a fairly high probability. Twenty simoleons on No.
WILL THERE BE A SUCCESSFUL 4TH DOWN CONVERSION?
YES (-250)
Between Jim Harbaugh’s aggressiveness (and his new QB) and the Ravens’ win-it-all-for-Ray mentality, I can’t imagine there not being at least one successful 4th down conversion. I know these are stout defenses but if one team gets down early, it would make this even more of a lock to come true. If there’s only one attempt the whole game, then I could see it going over to the “No” side, but if one team finds themselves in a situation where they need two scores in the final five minutes of regulation, there will definitely be a 4th down conversion. Fifteen simoleons on Yes.
THE UNDERDOGS
WILL THERE BE A SAFETY?
YES (+900)
I was looking up data on two point conversions and noticed there were quite a few safeties. In 46 Super Bowls, there have been 7, or one in 15.2% of the games. This line implies a 10% chance of something happening so I’ll be getting some pretty good value here.
I know that Barnwell points out that the two pass rushes are struggling coming into this game but their special teams are well coached (enough to pin the other team deep) and there are other ways to get a safety, such as in Super Bowl XLIII when the Steelers were called for holding in the endzone. It also happened last year when Tom Brady was called for intentional grounding in the endzone.
Crazy things can happen when a football bounces… five simoleons on Yes.
WHO WILL THE MVP BE?
COLIN KAEPERNICK (+175)
JOE FLACCO (+225)
I’m lumping these together for obvious reasons. In my opinion, these two have been the MVP’s of their teams’ respective Super Bowl runs and I see no reason why that should stop now. Many players will be important on both sides of the ball for both teams but nobody is more important than these two guys.
Let me put it another way; when I ponder all the ways in which both teams can win this game, there are few instances where I think a team can win with their quarterback playing poorly. Five simoleons on each.
WHAT COLOR WILL THE GATORADE (OR LIQUID) DUMPED ON THE WINNING COACH BE?
RED (+750)
This is an example of how ridiculous prop bets can be. Having said that, considering that just about every beverage maker (and especially those that specialize in sports drinks) have a flavor that is colored red, this feels like a steal to me. When you add in the fact that the 49ers primary color is red, it just reinforces the feeling to me that this is way too high of a line. Five simoleons on red.
THERE WILL BE NO TOUCHDOWNS
YES (+20,000)
This is the longshot of all longshots and it becomes even more so when you consider that these two teams have combined for 22 touchdowns in their combined 5 games. However, a bet offering a 200:1 payout when you have two very good defenses feels like a good longshot on which to put a little bit of action. To me, this is the equivalent of having a lucky number at the roulette table where you always put a penny at the table’s with a dollar minimum bet.
Working against me in this bet is the fact that the game is in the Superdome and, in case you haven’t seen him play, Colin Kaepernick is fast. Considering that next year’s Super Bowl is in East Rutherford, New Jersey, and the forecast for Sunday calls for highs of 37 degrees, 13 mph winds, and a few snow showers, this bet will look quite a bit better next year. However, there’s no way you’ll get a line of +20,000 next year in the Meadowlands. Five simoleons on nobody reaching the endzone.
SUMMARY
There you have it, I’m placed 100 simoleons on prop bets on the Super Bowl. Of my nine bets, only two are mutually exclusive (MVP race) so I could potentially go 8-9. If I were to do that with Flacco winning the MVP trophy, I’ll turn my 100 simoleons into 1,226.93 simoleons (1,224.43 if Kaepernick wins MVP).
Disregarding the fact that two of them are mutually exclusive, based solely on the lines, I have a 0.00014% chance of running the table. Of course, with my luck, Kaepernick will get hurt on the first series of the game and Alex Smith will come in and lead the 49ers to a thrilling victory, winning the MVP trophy by throwing four touchdowns (including the game winning two point conversion) after throwing a pick 6 to Ray Lewis to start his day. Jim Harbaugh will be doused in water (it’s better for you anyways) at the end of the game where the announcers forget to mention even once that Jim and John Harbaugh are related.
This, ladies and gentlemen, is why I don’t gamble.
As for who wins the game, the last I heard is that the 49ers are 4.5 point favorites. Part of me says that line is too generous and another part of me says it's not generous enough. My prediction is that it comes down to the kickers. David Akers (29-42 in the regular season) will miss his only attempt in the first quarter while rookie Justin Tucker (30-33 this year) while make three field goals, including one early in the fourth quarter, setting up Ray Lewis to make a huge tackle on fourth down inside of two minutes. Ravens 30-28.
Enjoy the Super Bowl everyone and I’ll be back next week to look at how well (or depressingly) these (fake) bets turned out.
Thursday, January 31, 2013
Sunday, January 20, 2013
Winning in the NFL
Another regular season is here and gone and seven more head coaches are searching for a new job. Well, technically six since it took Andy Reid about five minutes (rightfully so) to find a new job. Nevertheless, the common acronym Not For Long is more appropriate in the NFL than any other of the big professional leagues. While there are always examples of coaches let go by their teams after a curiously short stint (see Johnson, Avery of the Brooklyn Nets) in the other leagues, those instances are the exception, not the rule. In the NFL, the day after the final day of the regular season is known as Black Monday, the day when head coaches are given their pink slips. Sometimes it’s a veteran head coach that’s had more than a fair shot to win in a particular city but it feels like more often than not, the head coach is fired before he really has a chance to impart himself on the team.
Why is that important? When a new head coach is hired, he may get to hire his coaching staff but he doesn’t get to hire the general manager (unless he is the general manager) and for the most part, he inherits the players from the previous coach. Therefore, I could argue that in a coach’s first year, the expectations should be very low and gradually ramp up from there. Normally, I would say that a “fair shot” for a head coach would consist of four or five seasons but again, not all situations are normal. For instance, in his first four years as a head coach with their current team, Coach A was 22-42 and made one playoff appearance while Coach B went 39-25 and won the Super Bowl twice.
Coach B (Bill Belichick) inherited an 8-8 team that was the team’s worst record in the previous four years. Coach A (Jim Schwartz) inherited a team that from 2001-2007 went 31-81 before bottoming out in 2008, going 0-16.
At the end of the day, the question is very simple: how do you win in the NFL? There is an easy answer for those of you out there that are more sarcastically inclined and yes, the way to win in the NFL is to score more points than the other team. Now that it’s been said, we can move on with life… and this blog post.
If we were to liken football history to that of baseball, we are certainly in the “Live Ball” era of football. For much of football’s history, three yards and a cloud of dust was the norm and the way to win was to get 3 ½ yards every play while holding your opponents to 2 ½ yards per play. There has been a gradual shift from the running game to the passing game over the past 40 years starting with players like Johnny Unitas and Joe Namath and continuing today with Tom Brady and Peyton Manning but nothing jumpstarted passing football like the rule changes following the 2004 season. The rule changes enacted were in favor of the quarterback and the receivers as fans seemed to tire of watching defensive backs mugging receivers every play.
Since then, quarterbacks have set just about every relevant record that matters with regards to playing the position. Yards and touchdowns in a season along with yards and touchdowns in a career all fell. In 2012, three rookies and three second year quarterbacks made up half the playoff field. One of the more senior quarterbacks, Aaron Rodgers, sat for three years waiting for Brett Favre to take his sideshow to New York and Minnesota. In 2012, two of those three rookies were the starting quarterbacks from the word “go” and Russell Wilson was the starter coming out of training camp.
All of this leads me to believe that success in the NFL revolves around the quarterback position. While this is completely logical (look at the win-loss records for Tom Brady and Peyton Manning over the past ten years), I wanted to find something more. Would there be a way to come up with a formula with which a football team could be built to win?
SAMPLE SIZE
One of the biggest problems with the NFL is the fact that there are only 16 games per season. Attempting to predict something like winning based on that sample is problematic at best. By comparison, there are 162 games in every baseball season so in many ways, one season (or even half a season) is enough to draw some significant conclusions. Essentially, what I’m trying to say is outliers will exist in any analysis that looks at patterns of success in the NFL. This just needs to be accepted so we can move on.
For this set, I used 2002 to 2011, a ten year period that, for the most part, occurred after that 2004 cutoff I spoke of earlier. The rules may not have changed yet at the beginning of that window, but some of the elite quarterbacks that are still dominating the league today were either in their prime or coming into their prime by the time the window started.
MEASUREABLES
I believe that the NFL is currently a passing league first, so one of the first things I wanted to include in any analysis is quarterback play. There are many ways to measure quarterback play but of all the data I have access to, I chose to use the passer rating metric. The new metric that has recently been introduced, QuarterBack Rating (QBR) is a better overall measure of how well a quarterback plays but data is only available for the past five seasons (2008 onward) and in an attempt to draw more significant conclusions, I want a larger sample size.
Turnovers are a very large part of football but especially so in the NFL. When it comes to the success of a quarterback, the pass rush of the opposing team can have a drastic impact so it would seem that sacks would also be an important piece of this analysis. Finally, while the rushing game has been devalued in the past ten years, it can still be very important. This year, the Vikings made the playoffs with a (putting it nicely) suspect quarterback and it was because they had a very reliable running game. So, rushing yards per game will also be included in this analysis.
At this point we need to remember one critical thing; there is another team on the field during every game. Therefore, I’m going to break this analysis into three parts:
OFFENSE – The goal of the offense is to score points. Therefore, how do passer rating, sacks allowed, turnovers committed, and rushing yards per game come together in terms of points on the scoreboard?
DEFENSE – How do those statistics (defensive instead of offensive) go into the number of points that a team’s opponent scores in a season?
EFFECTIVE WINNING PERCENTAGE (more on this in a moment) – How do the number of points scored and allowed go into a team’s winning percentage?
THE ONE-POSSESSION CONUNDRUM
When I first started looking at the mountain of data that I obtained, I noticed that no matter what I did, there were a number of stubborn outliers that I simply could not explain. After who knows how long of staring at numbers and manipulating them every way I could imagine it, it finally occurred to me; one possession games.
I don’t know if I was watching a football game or reading a piece about the 2012 incarnation of the Indianapolis Colts but it got me to thinking and I eventually asked whether I should even include one possession games in my analysis.
Off the top of my head, the average NFL game is comprised of anywhere from 100-150 plays, including special teams. Essentially, a one possession game (where the margin is 8 points or less) could have swung the other way if just one of those plays had happened differently. Given this logic, it follows that over time, a team’s record in one possession games would regress towards the mean, or a .500 winning percentage.
To calculate the “effective” winning percentage, I took the number of one possession games each team played, divided it in half and added that number to the wins and losses a team had from 9+ point games. So, if a team was 12-4 (for a .750 winning percentage) but went 6-2 in one possession games, I normalized this record by putting in a .500 record in one possession games so I removed the 6-2 record and put in a 4-4 record to go with the team’s 6-2 record in 9+ point games. The result is a record of 10-6 for an “effective” winning percentage of .625.
When I did this, all of my regressions started to look much better than they originally had but while this solved one problem, it created another one. I appear to be getting closer to explaining how NFL teams can win but for the life of me, I have no idea how some teams do better in one possession games than others.
Out of all the regressions I conducted, the best r2 value I could get was about 0.19 and while the more statistically inclined of you might know that implication, for those of you who don’t, that’s not good. Essentially, in my search for an indicator, I was only able to find variables that could explain 19% of the variability of the dependent variable. For a perfectly linear relationship, the independent variable explains all of the variability in the dependent variable. When searching for a correlation, a regression isn’t worth much with an r2 value below 0.5.
I know that something dictates performance in close football games and if pressed, I’m inclined to say that it’s something subjective such as the quality of coaching. I don’t believe that it’s a coincidence that the only three teams that have won more than 60% of such games are New England Patriots (.708), Indianapolis Colts (.683), and the New York Giants (.600), while the only two teams to win less than 40% of such games are the Buffalo Bills (.375) and Detroit Lions (.338).
BY THE NUMBERS
For all of the regression analysis, I used the Ordinary Least Squares method and the advantage of this method is the r2 statistic I mentioned earlier. Using this method, I’ll be able to see how my variables mentioned above affect points scored and points allowed (and therefore winning percentage) and I’ll also be able to see how much of the picture is revealed by this analysis and how much is still murky.
Without further ado, the following equation determines how many point an offense will score in a season:
OFFENSE
Points = (RTN x 4.86222) – (Sck-All x 1.05405) + (Give x 1.5772) + (RYPG x 0.812841) – 144.606
RTN = passer rating of the team
Sck-All = sacks allowed
Give = turnovers committed
RYPG = rushing yards per game
DEFENSE
The following equation determines how many points a team will allow in a season:
Points = (OPP-RTN x 4.53191) – (Sck x 0.596579) + (Take x 1.42401) + (RYAPG x 0.89086) – 141.051
OPP-RTN = passer rating of the opponent
Sck = sacks
Take = turnovers a team creates
RYAPG = rushing yards allowed per game
WINNING
Finally, throw those together with effective winning percentage and you get the following equation:
EW% = (PF x 0.0214484) – (PA x 0.0208863) + 7.80691
EW% = effective winning percentage
PF = points scored
PA = points allowed
THAT’S A NICE BIT OF NUMERICAL NONSENSE, BUT WHAT DOES IT MEAN?
Fantastic question. First of all, the r2 values are all relatively high, which is desired. For points scored, the value was 0.750041 and for points allowed it was 0.743349 while for effective winning percentage, it was 0.928130. This tells me that there is more to scoring points than what I put into these equations (special teams contributions, third down efficiency, etc.) but for relatively straightforward analysis, I’d say this methodology was successful.
If we dig a little deeper into the equations, some interesting things emerge about the relative importance of the variables. For this following section, you have to keep in mind that when I’m talking about the effects of changing one variable, I’m holding all the others constant.
PASSER RATING
Given by the constant in the winning percentage equation, an “average” team in every part of this analysis would win 7.8 games. I realize that truly the average team wins 8 games but there isn’t one team that would have all of the league average statistics used in this analysis.
If a team’s passer rating improves by 10 points, from a league average 80.5 to 90.5 (a moderate but significant jump), that would imply that the offense would score 48.6 more points (4.86 x 10). Multiplying this by 0.02145 means that holding all else constant, an increase in 10 points of passer rating leads to 1.04 more wins, improving a record from 7.8-8.2 to 8.84-7.16.
SACKS ALLOWED
This season, the average number of sacks allowed was 36.5 while the fewest allowed was 20. Going from league average protection to the best pass protection in the league this season would increase a team’s point total by 17.39 points, or a bit more than a third of a win.
GIVEAWAYS
This is the one that doesn’t make intuitive sense. Given that the coefficient has a positive value, this implies that as a team turns the ball over more, they score more points. This is backed up by the same sign on the coefficient in the points allowed equation. The only possible explanation that I have is that a team that turns the ball over more is more aggressive and is more likely to score more points (call it the Brett Favre Effect).
Even though it makes no intuitive sense, the p-value for this variable was essentially zero. This means that the variable is statistically significant; in other words, it’s in the equation because it should be, it adds more descriptive value to the equation.
If we accept that it belongs, we can see that committing 10 more turnovers will lead to 15.77 more points scored. In 2012, this means that the difference between the best and worst performances for turnovers committed (23 turnovers) was worth 36.28 points, or 0.778 wins.
RUSHING YARDS PER GAME
When I started, I didn’t originally have any rushing stats in the analysis. Including this is basically a shout out to the incredible season that Adrian Peterson is having. Given that I’ve added a key component of my analysis because of one player, perhaps that shows just how good of a season he’s had.
Anyways, a league average offense rushed for 115.9 yards per game in 2012. The Redskins led the league (yes, even with Peterson, the Vikings didn’t lead the league in rushing) with 169.31 yards per game. That difference is worth approximately 43.4 points, or 0.93 wins.
SPELL IT OUT FOR ME…
Turnovers are important. Protecting the passer is important. Having a good running game is important. Nothing is nearly as important as having a good quarterback. In other words, if you were to replace a novelty quarterback with a passer rating of 72.9 (for sake of argument, let’s refer to him as Tim Tebow) with a first ballot Hall of Famer with a passer rating of 105.8 (let’s call him Peyton Manning), and everyone else plays just as well as they did over both years, the team will score 159.97 more points and win 3.43 more games.
OK, YOU’VE BEEN AVOIDING IT THIS WHOLE TIME… DID YOUR METHOD WORK?
Let’s put it this way… it shows what I wanted it to show and that’s that the team goes as the success of their quarterback goes. A quarterback’s play while throwing the ball is far more important than sacks allowed, turnovers, or rushing yards per game. On average, from 2002 to 2011, my equations were off by 1.4 wins on average and 73% of the time, it was within two wins of how a team actually did.
In 2004, the Tampa Bay Buccaneers went 5-11 when my method would have predicted a record of 8.9-7.1 due to quality play from the quarterback position. While it was predicted that the Bucs would have scored 362.8 points, they actually scored 301 (perhaps as a result of their 36 turnovers).
In 2009, the Colts went 14-2 but their numbers predicted a record of 9.3-6.7. That season, the Colts were outgained 126.5-80.9 in rushing yards per game, a deficit they most likely overcame with the help of a certain quarterback.
2012 RESULTS
This is how the 2012 season shaped up division by division with the records predicted by my methods next.
As can easily be seen, there were some hits and some misses but I would argue (as I mentioned before) that 16 games is a small enough sample size that it is easy to get outliers. So what happens when we start looking at teams over a longer time scale, say, 10 years?
Before I show you the results, I took the “different than expected” records in one-possession games and added that to the predicted number of wins. For instance, over the past ten years, the Patriots should have a record of 99.9-60.1 but their record in one-possession games was 46-19, or 13.5 games better than an even 32.5-32.5 record. When you add that in, the Patriots’ predicted record becomes 113.4-46.6. When you compare this to their actual 123-37 record, it compares favorably. While a difference of 9.6 wins might sound like a lot, you have to remember that over 10 years, that’s only one win per year. Here’s how the league looks:
The average difference between the team’s actual records and those predicted by my method? 3.31 wins per team over the ten year period so an average of a third of a win per year.
At the end of the day, if you take away only one thing from this post, you should take away that today’s NFL does indeed revolve around the quarterback position and there’s a reason why simply adding a player like Peyton Manning immediately makes the Denver Broncos Super Bowl contenders.
Essentially, this analysis confirms what we know simply watching games every Sunday during the fall: a good defense can take you far, a great running back can get you to the playoffs, but quarterbacks are ones that carry you to the Lombardi Trophy.
Why is that important? When a new head coach is hired, he may get to hire his coaching staff but he doesn’t get to hire the general manager (unless he is the general manager) and for the most part, he inherits the players from the previous coach. Therefore, I could argue that in a coach’s first year, the expectations should be very low and gradually ramp up from there. Normally, I would say that a “fair shot” for a head coach would consist of four or five seasons but again, not all situations are normal. For instance, in his first four years as a head coach with their current team, Coach A was 22-42 and made one playoff appearance while Coach B went 39-25 and won the Super Bowl twice.
Coach B (Bill Belichick) inherited an 8-8 team that was the team’s worst record in the previous four years. Coach A (Jim Schwartz) inherited a team that from 2001-2007 went 31-81 before bottoming out in 2008, going 0-16.
At the end of the day, the question is very simple: how do you win in the NFL? There is an easy answer for those of you out there that are more sarcastically inclined and yes, the way to win in the NFL is to score more points than the other team. Now that it’s been said, we can move on with life… and this blog post.
If we were to liken football history to that of baseball, we are certainly in the “Live Ball” era of football. For much of football’s history, three yards and a cloud of dust was the norm and the way to win was to get 3 ½ yards every play while holding your opponents to 2 ½ yards per play. There has been a gradual shift from the running game to the passing game over the past 40 years starting with players like Johnny Unitas and Joe Namath and continuing today with Tom Brady and Peyton Manning but nothing jumpstarted passing football like the rule changes following the 2004 season. The rule changes enacted were in favor of the quarterback and the receivers as fans seemed to tire of watching defensive backs mugging receivers every play.
Since then, quarterbacks have set just about every relevant record that matters with regards to playing the position. Yards and touchdowns in a season along with yards and touchdowns in a career all fell. In 2012, three rookies and three second year quarterbacks made up half the playoff field. One of the more senior quarterbacks, Aaron Rodgers, sat for three years waiting for Brett Favre to take his sideshow to New York and Minnesota. In 2012, two of those three rookies were the starting quarterbacks from the word “go” and Russell Wilson was the starter coming out of training camp.
All of this leads me to believe that success in the NFL revolves around the quarterback position. While this is completely logical (look at the win-loss records for Tom Brady and Peyton Manning over the past ten years), I wanted to find something more. Would there be a way to come up with a formula with which a football team could be built to win?
SAMPLE SIZE
One of the biggest problems with the NFL is the fact that there are only 16 games per season. Attempting to predict something like winning based on that sample is problematic at best. By comparison, there are 162 games in every baseball season so in many ways, one season (or even half a season) is enough to draw some significant conclusions. Essentially, what I’m trying to say is outliers will exist in any analysis that looks at patterns of success in the NFL. This just needs to be accepted so we can move on.
For this set, I used 2002 to 2011, a ten year period that, for the most part, occurred after that 2004 cutoff I spoke of earlier. The rules may not have changed yet at the beginning of that window, but some of the elite quarterbacks that are still dominating the league today were either in their prime or coming into their prime by the time the window started.
MEASUREABLES
I believe that the NFL is currently a passing league first, so one of the first things I wanted to include in any analysis is quarterback play. There are many ways to measure quarterback play but of all the data I have access to, I chose to use the passer rating metric. The new metric that has recently been introduced, QuarterBack Rating (QBR) is a better overall measure of how well a quarterback plays but data is only available for the past five seasons (2008 onward) and in an attempt to draw more significant conclusions, I want a larger sample size.
Turnovers are a very large part of football but especially so in the NFL. When it comes to the success of a quarterback, the pass rush of the opposing team can have a drastic impact so it would seem that sacks would also be an important piece of this analysis. Finally, while the rushing game has been devalued in the past ten years, it can still be very important. This year, the Vikings made the playoffs with a (putting it nicely) suspect quarterback and it was because they had a very reliable running game. So, rushing yards per game will also be included in this analysis.
At this point we need to remember one critical thing; there is another team on the field during every game. Therefore, I’m going to break this analysis into three parts:
OFFENSE – The goal of the offense is to score points. Therefore, how do passer rating, sacks allowed, turnovers committed, and rushing yards per game come together in terms of points on the scoreboard?
DEFENSE – How do those statistics (defensive instead of offensive) go into the number of points that a team’s opponent scores in a season?
EFFECTIVE WINNING PERCENTAGE (more on this in a moment) – How do the number of points scored and allowed go into a team’s winning percentage?
THE ONE-POSSESSION CONUNDRUM
When I first started looking at the mountain of data that I obtained, I noticed that no matter what I did, there were a number of stubborn outliers that I simply could not explain. After who knows how long of staring at numbers and manipulating them every way I could imagine it, it finally occurred to me; one possession games.
I don’t know if I was watching a football game or reading a piece about the 2012 incarnation of the Indianapolis Colts but it got me to thinking and I eventually asked whether I should even include one possession games in my analysis.
Off the top of my head, the average NFL game is comprised of anywhere from 100-150 plays, including special teams. Essentially, a one possession game (where the margin is 8 points or less) could have swung the other way if just one of those plays had happened differently. Given this logic, it follows that over time, a team’s record in one possession games would regress towards the mean, or a .500 winning percentage.
To calculate the “effective” winning percentage, I took the number of one possession games each team played, divided it in half and added that number to the wins and losses a team had from 9+ point games. So, if a team was 12-4 (for a .750 winning percentage) but went 6-2 in one possession games, I normalized this record by putting in a .500 record in one possession games so I removed the 6-2 record and put in a 4-4 record to go with the team’s 6-2 record in 9+ point games. The result is a record of 10-6 for an “effective” winning percentage of .625.
When I did this, all of my regressions started to look much better than they originally had but while this solved one problem, it created another one. I appear to be getting closer to explaining how NFL teams can win but for the life of me, I have no idea how some teams do better in one possession games than others.
Out of all the regressions I conducted, the best r2 value I could get was about 0.19 and while the more statistically inclined of you might know that implication, for those of you who don’t, that’s not good. Essentially, in my search for an indicator, I was only able to find variables that could explain 19% of the variability of the dependent variable. For a perfectly linear relationship, the independent variable explains all of the variability in the dependent variable. When searching for a correlation, a regression isn’t worth much with an r2 value below 0.5.
I know that something dictates performance in close football games and if pressed, I’m inclined to say that it’s something subjective such as the quality of coaching. I don’t believe that it’s a coincidence that the only three teams that have won more than 60% of such games are New England Patriots (.708), Indianapolis Colts (.683), and the New York Giants (.600), while the only two teams to win less than 40% of such games are the Buffalo Bills (.375) and Detroit Lions (.338).
BY THE NUMBERS
For all of the regression analysis, I used the Ordinary Least Squares method and the advantage of this method is the r2 statistic I mentioned earlier. Using this method, I’ll be able to see how my variables mentioned above affect points scored and points allowed (and therefore winning percentage) and I’ll also be able to see how much of the picture is revealed by this analysis and how much is still murky.
Without further ado, the following equation determines how many point an offense will score in a season:
OFFENSE
Points = (RTN x 4.86222) – (Sck-All x 1.05405) + (Give x 1.5772) + (RYPG x 0.812841) – 144.606
RTN = passer rating of the team
Sck-All = sacks allowed
Give = turnovers committed
RYPG = rushing yards per game
DEFENSE
The following equation determines how many points a team will allow in a season:
Points = (OPP-RTN x 4.53191) – (Sck x 0.596579) + (Take x 1.42401) + (RYAPG x 0.89086) – 141.051
OPP-RTN = passer rating of the opponent
Sck = sacks
Take = turnovers a team creates
RYAPG = rushing yards allowed per game
WINNING
Finally, throw those together with effective winning percentage and you get the following equation:
EW% = (PF x 0.0214484) – (PA x 0.0208863) + 7.80691
EW% = effective winning percentage
PF = points scored
PA = points allowed
THAT’S A NICE BIT OF NUMERICAL NONSENSE, BUT WHAT DOES IT MEAN?
Fantastic question. First of all, the r2 values are all relatively high, which is desired. For points scored, the value was 0.750041 and for points allowed it was 0.743349 while for effective winning percentage, it was 0.928130. This tells me that there is more to scoring points than what I put into these equations (special teams contributions, third down efficiency, etc.) but for relatively straightforward analysis, I’d say this methodology was successful.
If we dig a little deeper into the equations, some interesting things emerge about the relative importance of the variables. For this following section, you have to keep in mind that when I’m talking about the effects of changing one variable, I’m holding all the others constant.
PASSER RATING
Given by the constant in the winning percentage equation, an “average” team in every part of this analysis would win 7.8 games. I realize that truly the average team wins 8 games but there isn’t one team that would have all of the league average statistics used in this analysis.
If a team’s passer rating improves by 10 points, from a league average 80.5 to 90.5 (a moderate but significant jump), that would imply that the offense would score 48.6 more points (4.86 x 10). Multiplying this by 0.02145 means that holding all else constant, an increase in 10 points of passer rating leads to 1.04 more wins, improving a record from 7.8-8.2 to 8.84-7.16.
SACKS ALLOWED
This season, the average number of sacks allowed was 36.5 while the fewest allowed was 20. Going from league average protection to the best pass protection in the league this season would increase a team’s point total by 17.39 points, or a bit more than a third of a win.
GIVEAWAYS
This is the one that doesn’t make intuitive sense. Given that the coefficient has a positive value, this implies that as a team turns the ball over more, they score more points. This is backed up by the same sign on the coefficient in the points allowed equation. The only possible explanation that I have is that a team that turns the ball over more is more aggressive and is more likely to score more points (call it the Brett Favre Effect).
Even though it makes no intuitive sense, the p-value for this variable was essentially zero. This means that the variable is statistically significant; in other words, it’s in the equation because it should be, it adds more descriptive value to the equation.
If we accept that it belongs, we can see that committing 10 more turnovers will lead to 15.77 more points scored. In 2012, this means that the difference between the best and worst performances for turnovers committed (23 turnovers) was worth 36.28 points, or 0.778 wins.
RUSHING YARDS PER GAME
When I started, I didn’t originally have any rushing stats in the analysis. Including this is basically a shout out to the incredible season that Adrian Peterson is having. Given that I’ve added a key component of my analysis because of one player, perhaps that shows just how good of a season he’s had.
Anyways, a league average offense rushed for 115.9 yards per game in 2012. The Redskins led the league (yes, even with Peterson, the Vikings didn’t lead the league in rushing) with 169.31 yards per game. That difference is worth approximately 43.4 points, or 0.93 wins.
SPELL IT OUT FOR ME…
Turnovers are important. Protecting the passer is important. Having a good running game is important. Nothing is nearly as important as having a good quarterback. In other words, if you were to replace a novelty quarterback with a passer rating of 72.9 (for sake of argument, let’s refer to him as Tim Tebow) with a first ballot Hall of Famer with a passer rating of 105.8 (let’s call him Peyton Manning), and everyone else plays just as well as they did over both years, the team will score 159.97 more points and win 3.43 more games.
OK, YOU’VE BEEN AVOIDING IT THIS WHOLE TIME… DID YOUR METHOD WORK?
Let’s put it this way… it shows what I wanted it to show and that’s that the team goes as the success of their quarterback goes. A quarterback’s play while throwing the ball is far more important than sacks allowed, turnovers, or rushing yards per game. On average, from 2002 to 2011, my equations were off by 1.4 wins on average and 73% of the time, it was within two wins of how a team actually did.
In 2004, the Tampa Bay Buccaneers went 5-11 when my method would have predicted a record of 8.9-7.1 due to quality play from the quarterback position. While it was predicted that the Bucs would have scored 362.8 points, they actually scored 301 (perhaps as a result of their 36 turnovers).
In 2009, the Colts went 14-2 but their numbers predicted a record of 9.3-6.7. That season, the Colts were outgained 126.5-80.9 in rushing yards per game, a deficit they most likely overcame with the help of a certain quarterback.
2012 RESULTS
This is how the 2012 season shaped up division by division with the records predicted by my methods next.
As can easily be seen, there were some hits and some misses but I would argue (as I mentioned before) that 16 games is a small enough sample size that it is easy to get outliers. So what happens when we start looking at teams over a longer time scale, say, 10 years?
Before I show you the results, I took the “different than expected” records in one-possession games and added that to the predicted number of wins. For instance, over the past ten years, the Patriots should have a record of 99.9-60.1 but their record in one-possession games was 46-19, or 13.5 games better than an even 32.5-32.5 record. When you add that in, the Patriots’ predicted record becomes 113.4-46.6. When you compare this to their actual 123-37 record, it compares favorably. While a difference of 9.6 wins might sound like a lot, you have to remember that over 10 years, that’s only one win per year. Here’s how the league looks:
The average difference between the team’s actual records and those predicted by my method? 3.31 wins per team over the ten year period so an average of a third of a win per year.
At the end of the day, if you take away only one thing from this post, you should take away that today’s NFL does indeed revolve around the quarterback position and there’s a reason why simply adding a player like Peyton Manning immediately makes the Denver Broncos Super Bowl contenders.
Essentially, this analysis confirms what we know simply watching games every Sunday during the fall: a good defense can take you far, a great running back can get you to the playoffs, but quarterbacks are ones that carry you to the Lombardi Trophy.
Subscribe to:
Posts (Atom)
