Buffalo Bills Going the Moneyball Route With Analytics

Nerval's Lobster writes "Can data-analytics software win a Super Bowl? That's what the Buffalo Bills are betting on: the NFL team will create an analytics department to crunch player data, building on a model already well established in professional baseball and basketball. 'We are going to create and establish a very robust football analytics operation that we layer into our entire operation moving forward,' Buffalo Bills president Russ Brandon recently told The Buffalo News. 'That's something that's very important to me and the future of the franchise.' The increased use of analytics in other sports, he added, led him to make the decision: 'We've seen it in the NBA. We've seen it more in baseball. It's starting to spruce its head a little bit in football, and I feel we're missing the target if we don't invest in that area of our operation, and we will.'"

Comments

Here's hoping the Browns go the same way

[afidel]

I for one hope the new Browns owner decides to go down a similar path, for a decade the Browns have squandered draft choices and money on flop after flop. Since they're in the market for a new GM and head coach now would be the perfect time to inject such a new system into the front office.

Re:Here's hoping the Browns go the same way

[pixelpusher220]

As a life long Bills fan, any person hoping their team 'follows the Bills' path is more certifiable than I am for being a life long fan ;-)

Too late?

[paiute]

Isn't it about 20 years too late to gain an edge on other pro franchises by following Moneyball? It's not like it is a secret weapon anymore.

Re:Too late?

[Trepidity]

It's catching on a bit later in sports other than baseball due to the difficulty (until recently) of collecting fine-grained statistics in many sports. Baseball is fairly discrete: it operates pitch by pitch, with a lot of down time in between. So a large number of relevant statistics can be tallied by hand, which is why we have piles of statistics dating back decades. For every pitch, you can mark down whether it was a strike or ball, whether the batter swung, where in the field it went to if hit, what the fielder did with it, what the runners did in response, etc.

For football (and even more so, soccer), a lot of the relevant information you'd get from watching replays is more "continuous" and harder to extract manually. Traditional statistics did measured things like passing completion percentage and yards gained by a running back, but they didn't collect data that could be used to quantify things like the quality of an offensive line, or of blockers, except indirectly through overall team performance. Now a lot more of that information is being automatically tallied using computer-vision algorithms churning through digitized camera footage.

Re:Too late?

[icebike]

This!

The fine grained statistics that you can pull out of baseball go back over a hundred years. Every game has these gathered more or less automatically these days, and even from a box score one can piece together stats on each player's abilities.

As a result there are plays that are never even attempted anymore because the stats are so precise at describing the likelihood of success/failure that they virtually dictate how the game is played. A player's value at any given point in time can be measured very precisely.

This level of detail is missing from Football, in part simply because too many bodies are in motion at once making it hard and tedious to map them, evaluate them, describe them, measure them, etc. Modern TV gear changes everything. Now all of this is possible after the fact by combining a few camera angles.

It will probably take 10years of game tape analysis just to define meaningful measures (statistics). Some will be useful, others will fall into disuse. But how we evaluate football players today will change drastically over time.

Re:Too late?

[ranton]

That's where statistics become useful! If a quarterback's throws are consistently dropped by a single receiver then it's the receiver's fault. This can be missed when looking at each play on an individual basis, but in aggregate over a season or two could reveal a significant weakness in the receiving corps or the quarterback.

But these are the kind of statistics they already have, and can already very easily analyze. The problem is that there just aren't enough data points. A good receiver may get 10 targets per game, and a quarterback is going to through on average 1 interception per game out of 40 attempts. That means an average receiver will probably have 4 passes intended for them intercepted each year. Lets say that half of interceptions are the receiver's fault, so that makes 2 per year. You simply are not going to get enough information when one player has 5 interceptions and another has 1. It just isn't enough data.

But a human (or AI) that knows the routes and can identify a poorly run route, a bobbled pass, and an overthrown pass could let you know who was actually at fault for each interception. It just takes a lot more data collection and analysis (a lot more).

Re:Too late?

[icebike]

Good example.

Too often the quarterback takes a bum rap for butter fingers down field, just like pitchers gain poor ERAs due to bad defense that allows easily defended ground balls to become runs scored, or pitchers that have bad win/loss records while playing with a horrible bunch of hitters.

Entire new measurements must be defined for football. How do you measure if the pass was catch-able by a competent receiver? Is it anything within arms length? Or is it more complex, taking into account direction of player motion? What about defensive coverage? Every one of these has to be assigned some form of measurement and then you have to start digitizing game tapes. It will take years to develop anything approximating what baseball has, but its probably long past due.

Re:Too late?

[Red Flayer]

, just like pitchers gain poor ERAs due to bad defense that allows easily defended ground balls to become runs scored

Just an FYI... ERA is Earned Run Average. It does not include runs arising from errors made by fielders (those are unearned runs).

Of course, there is still a very human factor in determining what defensive mishap is scored an error instead of a hit... not to mention great defenders who turn what would be a hit into an out because of their athletic range.

I say it is nearly impossible to do for football what Moneyball did for baseball. It's not just the scope of the stats that are important... it's also the team nature of football and the difference between systems.

In baseball, the average ball hit into play involves 5-6 people (catcher, pitcher, batter, fielder, baseman, sometimes an additional baseman). In football, the average play involves 22 people. The interdependencies are huge.

Plus you have the issue of players in different systems not playing to the same set of standards... how do you come up with equivalent metrics for players in positions whose role is different depending on the offensive scheme? Resort to video-game player stats for catching, awareness, blocking, etc? A wide receiver for the New York Jets is going to have a very different set of responsibilities than a WR for the Indianapolis Colts, and thus should be judged differently.

Plenty of people have been doing advanced statistical analysis of football players for a very long time. It's a muddle at worst, and can inform scouts and GMs at best... but I believe the Bills will have very limited success if they plan on going "pure Moneyball" style.

I think what this is really all about is normalizing the market for players. Costs are not even close to being in line with the value of a player or position, and by sticking to their "Moneyball" rules, the Bills gain some negotiating leverage... and some of the costly mistakes they (and other teams) have made wrt free agent signings.

Sure, but....

[NeutronCowboy]

Ok, I can see it for baseball. There is close to no interplay between players (even on the pitching team, coordination is restricted to whether you can catch what someone throws at you), and strategy is restricted to positioning players where a batter tends to hit and to how aggressively you go after a pitcher or batter. You're also playing 162 games a year - you can get some pretty good numbers in that time. Basketball is a bit harder, but with only four other teammates on the floor and a fairly static match-up (guards don't face centers much, you have zone or man-defense, and strategy revolves around how much you go for inside battles versus outside shots), the possible factors that influence whether a shot is made or not is still pretty small. You're also playing 82 games and taking a significant number of shots in a game. Again, you have a decent data set to work with.

But football? There are 10 teammates on the field, quite a few of which get switched out every other snap. You have 52 people on the roster, with many of them active during every game (especially on defense). Strategic decisions can take specific players completely out of the game for long stretches (simplest example: you're behind in the game, and start throwing - does that mean your running backs now suck?). And finally: there's only 16 games in a season. Some people may see action only 2-3 times a game or see action in trivial circumstances (see: kicker, long snapper). So not only do you have a huge amount of variables influencing a single player's success, you will also have a hard time creating a metric for success (touchdowns and sacks are rare outcomes of a long string of events), and on top of that, you're frequently dealing with a data set that maybe consists of 100 data points for an entire year, and maybe of 10 points for some lower-rung players. And it's exactly in the lower rungs of the players where moneyball was so wildly successful. Everybody knows an Adrian Peterson and Derek Jeter when they see one, but what about the journey players who switch teams once a year? Moneyball pretty much addressed that problem in baseball, but I don't see it working in football.

The Bills might prove me wrong, but I see this instead turning into the problem Girardi had with the Yankees: making player decisions based on stats that are calculated with 5 data points leads to decisions that will come back to bite you in the long run. You might as well save the money and just flip a coin.

Didn't Work Out Too Well Before

[Snap E Tom]

Football is fundamentally different from baseball and basketball. It has a lot more strategy, deception, teamwork, and on-the-fly communication between players. Something that happens innocently on one side of the field often has tremendous consequences on the other side. All this is very hard to quantify in a statistical model. For example, if your star receiver is shut down for a game, that might be because he's drawing double or triple coverage. Sure, his stats are low, but your slot and split ends can now have a field day.

The San Francisco 49ers tried a sabermetrics in their crappy years this past decade. Pioneered by the head of player personnel Paraag Marathe, they fielded a bunch of .500 and sub .500 teams before they moved him more to the business end of things and went with more traditional executives at talent evaluation.

Re:Didn't Work Out Too Well Before

[Radres]

Maybe what football needs is sabermetrics on talent evaluators.

Not as relevant to football

[BadMrMojo]

The reason this works particularly well in baseball, basketball and hockey is the schedule. You have 162 games a year in MLB, for example. In the NBA and NHL, it's 82 games. That's a relatively substantial sample - each game only accounts for roughly 0.6 or 1.2% of the season record.

The NFL, on the other hand, has a 16 game season. A team having a particularly good or bad game carries 10 times the weight it does in baseball (just going off the percentage of the season's games). Also, unlike baseball, football's playoffs are single-elimination.

The reason analytics aren't as directly relevant to football is exactly the reason that I enjoy it immensely.

---

Legend for our friends abroad:
MLB = Major League Baseball
NHL = National Hockey League
NBA = National Basketball Association
NFL = National Football** League

** - yes, we're talking about American football, rather than the game known internationally as the game in which you kick a ball with your foot.

Depends on their effort

[ranton]

Their success will likely depend on how much effort they put into collecting data. If all they look at is the same statistics you can find at CBS Sports, Football Outsiders, etc. then it will probably not help at all. But if they really get serious about data collection, who knows how much insight they could gain.

There are about 130 plays per game, and 256 games per year. That is 33,280 plays to analyze each year. That would increase to about 135k if you include Division 1-A college games. If you had two guys spend 15 minutes analyzing each play (2 guys to reduce errors) then it would take 20 full time employees to do this each year. More if you want to get more immediate results after each week. There are plenty of ex-athletes that couldn't make the pros and are intelligent enough for the work. Probably somewhere around $2 million per year in salary ($500k if you only look at professional games).

Just think of all the information you could gain. The first team to get this right could probably greatly improve their overall defenses and their offensive lines (positions that are very hard to rate with stats). I wonder how many teams know how many seconds thier offensive tackles can block an average defensive lineman, adjusted for their quarterback's mobility on each play, and any number of other mitigating factors.

Analytics Won't Help you Buffalo

[jchawk]

Unfortunately it's poor ownership and overall lack of leadership that is forcing you to suffer season after season after season of terrible records.

This team is hopelessly lost. They have not made the playoffs since 1998 and haven't had a winning record since 2003.

Invest in proper coaches and support staff. Commit to building a franchise instead of quick picks that you think will instantly win you a super bowl. Teams don't win with one or two guys. It takes a good (not great) quarterback, a good running back (not great) and a couple of good receivers. Couple that with a consistent defense and you can win Championships.

Look at Pittsburgh or New England. Year after Year these teams are in the hunt and have won a truck load of trophies.

Re:Offense or Defense

[Airw0lf]

Baseball. Can you think of another sport where the defense is the team with the ball?

Cricket?