Writers of Pro Football Prospectus 2008

Most Recent FO Features

HackenbergChr15.jpg

» SDA: From Death Valley to Happy Valley

While Saturday won't give us an answer to which is the best one-loss team, it is still a big week for conference races across the country.

10 Dec 2009

Walkthrough: December of Destiny

by Mike Tanier

Destiny ain't what it used to be.

David Garrard suggested this week that the 7-5 Jaguars are destined for greatness. "Yeah, it does feel like a year of destiny and that's great because 2007 felt like a year of destiny. To be at this point right now with seven wins; I bet if you talked to a lot of people before the season, they'd say, 'Yeah, right.'"

You remember the Super Bowl Champion 2007 Jacksonville Jaguars, right? They really were a team of destiny. They beat the undefeated Patriots in the playoffs, then won the Super Bowl when an Eli Manning pass bounced off David Tyree's helmet into Rashean Mathis' arms. This year's Jaguars are a lot like that team, which never existed.

Destiny's Jaguars are 7-5 on the strength of three-point wins against the Chiefs, the Rams (in overtime), and the Bills. They also beat the Jets by two. They have a 41-0 loss to the Seahawks, a 31-17 loss to the Cardinals, and a 30-13 loss to the Titans on their Resume of Fate. Just two weeks ago, the team that Garrard believes is blessed by Fortuna's loving caress lost 20-3 to the 49ers. Yes, Destiny's Jaguars are 1-3 against the woeful NFC West, their one win coming in overtime.

Clearly, Destiny has fallen upon hard times. Team of Destiny talk sounds silly coming from a wild card team that just won a playoff game. It sounds ridiculous coming from the quarterback of a second-place team a month removed from the playoffs. Garrard claims to be silencing doubters, but he's really silencing silence. No one doubts the Jaguars. No one believes in the Jaguars. No one even notices the Jaguars. Not even Destiny. Only 42,079 people attended the Jaguars-Texans game last week, a record low for a team that knows how to set low attendance records. Of course, if the Jaguars do win the Super Bowl, 400,000 people will claim to have been at the game where Chris Brown threw the interception that started the winning streak that ended with Josh Scobee's last-second field goal to beat the Saints in the Super Bowl.

By the way, Scobee is just 5-of-12 on field goals beyond 40 yards, so don't count on that game-winning kick, unless fate plays a hand.

I know I can be hard on the Jaguars. They are currently 21st in DVOA, down in the region where Redskins dwell. I am also too hard on those who engage in innocent magical thinking. What is Garrard supposed to think? That he's 7-5 only by grace of schedule? That there's no reason to practice or battle because the Colts will win the AFC? A little Chosen One mythology will keep the Jaguars striving for every win. And yes, thanks to the schedule and some luck, the playoffs are well within reach. If 2007 felt epic to Garrard, a first-round playoff loss to the Bengals will at least provide a little tingle.

In other news, Florida governor Charlie Crist thinks the Jaguars should draft Tim Tebow. Garrard's destiny may well lie elsewhere.

December Mourn

Last Monday at midnight, the odometer flipped. The warranty expired. Fairy Godmother hooked her wand to the charger and went to bed.

It was December 1. It was time for the Cowboys to stink, and for the Eagles to surge.

My skeptical mindset told me to doubt all of the December trends, even as the Eagles beat the Falcons and the Cowboys lost to the Giants, even as I wrote jokes about them for various articles. Saying the Cowboys are about to go into the can makes good copy, especially when writing to Eagles and Giants fans. Believing it is something else. The same is true about the Eagles. My heart wants to believe they have some holiday magic. My head tells me that the Eagles are always the Eagles, for better or worse, till death do us part.

Still, the numbers don't lie. Here are the Cowboys' win-loss records, by month, since 2005:

Month Record
September 11-5
October 12-8
November 17-4
Dec/Jan 7-12

The Cowboys are a .618 team overall, but a .368 team in December and January. By starting in 2005, we include their entire recent run of quality football. Go back to 2004, and you have a 6-10 team with players like Vinny Testaverde and Eddie George on the roster. No one cares what those two did in December.

Here are the Eagles, this time going all the way back to 2000, their first good year under Andy Reid:

Month Record
September 22-14
October 22-16
November 26-14-1
Dec/Jan 28-13

The Eagles are a .631 team but become a .683 team in December and January. Analyzing a decade of Eagles football means lumping pre- and post-Super Bowl teams, a team on the rise and a team in stasis, Duce Staley with Shady McCoy, Troy Vincent with Asante Samuel, Jeff Garcia with Kevin Kolb, and so on. The "December Eagles" we think of are the 2006-2009 team, 12-3 in December and January but 7-7-1 in November. The sample size is tiny, but the numbers are striking.

Is it unusual for a .618 team to have a run of games where they go 7-12, or a .631 team to have a winning percentage of .683 for a particular type of game? To phrase the question another way, is there a chance that the December record swings are just statistical noise? One way to tell whether statistical blips are really significant is binomial expansion, which tells us how likely a team is to win 28 games out of 41 -- or 30 or 40 for that matter -- given that their overall win expectance is .631. It's a relatively simple bit of algebra: create the binomial (.631W + .369L) to represent the win-loss probabilities, raise it to the 41st power, and find all the coefficients. A good calculator, like my TI-83 Math Teacher Service Revolver, helps.

Here are the probabilities that a team with a true .631 winning percentage will win 26 or more games, purely by chance, in a 41 game sample.

Wins Probability
26 .128
27 .122
28 .104
29 .079
30 .055
31 .033
32 .018
33 .008
34 .003
35 .001

A quick note here: the assumption that the Eagles are a .631 team is based on all 16 games, not on games from September through November. Take away December-January games, and their winning percentage drops to .614, making their December record more striking and the likelihood of sudden improvement occurring by chance smaller. Taking away successful games when measuring the odds of success creates a powerful selection bias, and I think it contributes to the perception that the Eagles do something drastically different in December.

There is a 10.4 percent chance that an Eagles-quality team will go 28-13, purely by luck of the draw, given a 41-game sample. Add the chances that the team would win 29, 30, or more games, and it creeps close to 30 percent. That's not overwhelming evidence, but it does suggest that random chance could be a major factor, if not the sole factor, in a December illusion.

If we start the study in 2006, it is harder to indict random chance as the sole culprit. The 2006-09 Eagles have a winning percentage of .593 but have gone 12-3 in December. The binomial is (.593W + .407L), raised to the 15th power. The chance of going 12-3 strictly by chance is 5.8 percent. It grows to just over eight percent when the chance of winning 13-15 games is factored in. It climbs to almost 20 percent if the chance of going 11-4 is included -- an 11-4 December record would still be cited as evidence of the Eagles late-season excellence -- but that is a bit of a dodge. The sample size is small, and events with a 5.8 percent chance of occurring happen daily, but there may be forces at work beyond random noise.

As for the Cowboys, here are the probabilities for winning seven games or less in December, given a .618 overall winning percentage:

Wins Probability
7 .0130
6 .0040
5 .0010
4 .0002

The chance that the Cowboys are winning seven games or less simply because of random factors adds up to about 2 percent. That could still just be the luck of the draw, but there's reason to look elsewhere.

Linked Fortunes

Binomial expansion of the win-loss records suggests that the Cowboys' December slip isn't just the result of random chance. The Eagles' upward trend, meanwhile, could be the result of random ups and downs, but there may be other forces at work. Win-loss record analysis is limited, however. Let's check in on the DVOA, which will take away schedule vagaries and fluky wins, telling us how well the two teams really play in December.

The 2008 Eagles enjoyed their highest DVOA of the season in a December victory over the Giants. Their other DVOA spikes occurred early in the year, and while they appear to have enjoyed a late-season upward trend, it's pretty slight. Remember that they lost to the Redskins, and that one of their best December performances came against the Browns, who had stopped scoring touchdowns.

The 2008 Cowboys are hard to analyze because their worst games occurred midseason, when Brad Johnson started at quarterback. Their December games were all in line with their early-season performance, except for their season-ending loss to the Eagles (-18%).

The 2007 Eagles, who finished 8-8, enjoyed three positive DVOA games in a row at season's end, but only one of those games was out of line with their season performance. More on that in a moment.

The 2007 Cowboys (13-3) posted DVOA ratings of -63% and -78% in December, plus a -11% DVOA against the Lions in a Week 14 win. They were clearly trending downward at season's end.

The 2006 Eagles (10-6) enjoyed their highest single-game DVOA of the year in Week 16; more on that game in a moment. The rest of their December percentages are in line with what they did in other months, save for back-to-back low-DVOA efforts against the Titans and Colts in November, when Donovan McNabb got hurt and Jeff Garcia was still rusty. Their trend line in Pro Football Prospectus 2007 is nearly flat.

The 2006 Cowboys posted negative DVOA in four of their five December games, with a season-low against the Saints (-56.2%) in Week 14. The team's overall DVOA for the year was 9.7%, so the late slump was significant, and the Cowboys' trend line slopes noticeably downward starting in mid-October.

Over the last three years, the Eagles' DVOA improvements in December have been modest, save for one-game spikes in 2007 and 2008. The Cowboys' DVOA declines were noticeable in 2006 and 2007, and they may have been masked in 2008 by a longer midseason slump.

Now here's the little detail I haven't mentioned: Remember the Eagles' out-of-line DVOA game in December of 2007? It came in a 10-6 win against the Cowboys, which earned a DVOA of 98%. Remember that season-high DVOA in December of 2006? It came in a 23-7 win against the Cowboys, earning a DVOA of 73%.

Remember last year's season finale, the game this website dubbed "Eagles Porn?" DVOA isn't impressed by long fumble and interception returns, so the Eagles' DVOA of 60% wasn't their best of the year. The Cowboys' -18% DVOA was already mentioned. It was the Eagles' third impressive December win against the Cowboys in three seasons.

The Eagles' December improvements have been slight, and their record is only slightly better than chance would dictate. The Cowboys' December declines have been more drastic, probably aren't the result of chance, and are borne out by DVOA for 2006 and 2007. The Eagles play the Cowboys every December, and the Eagles are 3-0 in those games.

Maybe the Eagles' perceived improvement is caused by the Cowboys decline.

A win or two makes a huge difference in the probabilities listed above. Say the Cowboys went 2-1 in those December games, a reasonable result given that the Eagles were starting a backup quarterback in 2006 and had nothing to play for (while the Cowboys were fighting for home field advantage) in 2007. That would make the Cowboys 9-10 in December, the Eagles 10-5. The binomial expansions would change, because both overall winning percentages would change, but the Eagles win probability in particular would grow much closer to random chance: a 14.9 percent chance of winning 10 of 15 games with their new winning percentage, and well over 25 percent when the possibly of winning 11 or more games is factored in. The "other factors" besides random chance in the Eagles' December record could be Andy Reid's adjustments or Donovan McNabb's health. But it might be the Cowboys.

Eagles-Cowboys games also loom large in the perceptions of both regional and national fans. When you think of Cowboys meltdowns, the image of Brian Dawkins stripping Tony Romo pops into your mind immediately, right? Eagles fans forget Redskins losses after Cowboys wins. Three seasons of playoff implications and Terrell Owens story lines brought national attention to these games. If the Eagles didn't keep beating the Cowboys in December, there would be less talk of late-season voodoo in both cities.

I think the late-season voodoo is in one city. We don't usually talk about intangibles like leadership and character at Football Outsiders, but that doesn't mean they don't exist. The Cowboys get worse in December, whether due to Romo's lifestyle, the team's overall discipline, Owens' influence, a poor post-Parcells front office, the inability to handle cold weather, or some combination of all of them. The Cowboys are so bad that they make the Eagles -- who are always playing better than Philly fans think they are -- look better when they face off each December.

Of course, it's not just a December trend, it's a late-season trend. The samples are small. The Cowboys are working hard to address this very issue, even as they claim it doesn't exist. There's no reason to think the Cowboys will collapse this month, except for the fact that they play the Saints, Chargers, Redskins and Eagles: two superior opponents (maybe three), two tough divisional foes.

But there's more to the Cowboys' December Mourn than just talk.

Bad Santa

Part Two of a three-part series on football-related holiday gift ideas, most of which are awful. This week: juvenile literature, as in literature aimed at juveniles, not written by overgrown juveniles. Although ...

There's nothing better than exposing youngsters to the joys of reading about football. After all, today's elementary school fan is tomorrow's Walkthrough reader.

Ronde and Tiki Barber aren't just a Buccaneers cornerback and an annoying media gadfly. They are also twins! And authors of a series of children's books! By My Brother's Side tells what happens when Tiki breaks his leg and spends a summer watching Ronde hog all the sports glory. It's recommended for readers at about a third-grade level. "Young fans of the Barber brothers will enjoy this book. Others may wish for a plot with more action and substance," says a Booklist review by Todd Morning. Both Morning and a reviewer on Amazon praise illustrator Barry Root's watercolor-and-gouache images.

Game Day, another book by the twins, is aimed the first- and second-grade crowd. From the Simon and Schuster summary: "Tiki has had seven long touch-down runs this season, and Ronde is proud of his brother, but he can't help feeling a little down when Tiki gets all the glory." You get the impression that this conflict never quite got resolved back in the Pop Warner days. Barry Root is involved in this book as well, so there's probably more watercolor and gouache.

Not content to let football's second family hog all the easy-reader glory, the Mannings also got into the children's book field. Family Huddle, written by Peyton, Eli, and Archie Manning, is the story of a trip to grandma's, where there is much backyard football to be played. Family Huddle is most noteworthy for an appearance by Cooper Manning, the Chuckie Cunningham of the family. Cooper doesn't get a writing credit, but his existence as the oldest and wisest Manning brother is acknowledged.

In addition to instilling some old-fashioned family values, the book teaches kids some basic pass routes, like the buttonhook. At no point does it instruct youngsters to take the field an hour before each game and run through the passing tree. From Publisher's Weekly: "This writing team of two Super Bowl-winning quarterbacks and their former star quarterback father should score with young fans, despite some heavy-handed clich├ęs and syrupy dialogue." The art is digital, so no gouache, and the price tag ($16.95, marked down on Barnes & Noble to $12.23) makes this a gift best given to the Peyton or Eli-obsessed grade-schooler.

More discriminating (and older) readers will prefer Mike Lupica's Million Dollar Throw, about a 13-year-old with a golden arm and a chance to change his family's fortunes in a passing contest. Lupica holds several advantages over the Barbers and Mannings: He's an actual writer with a long list of adult and juvenile book credits, for starters. Lupica's books are great sports inspirationals aimed at preteen boys who may think that all books for their age group feature abstinence vampires.

Former Bears receiver Tim Green also writes challenging football books for early teens. Bookish boys are more likely to identify with Troy White, the protagonist of Football Genius and Football Champ than with a pint-sized Manning brother or a kid with a million-dollar arm. White isn't an athlete: He's a whiz kid who can predict winning plays, a skill that takes him from obscurity to a job in the Falcons front office. The life of this Football Outsider in Training isn't without difficulty: reporters run smear campaigns to belittle the teenage exec, and mom's boyfriend (an aging Falcons linebacker named Riley Coving ... er, Seth Halloway) is falsely accused in a steroid scandal. There's a lot for a 10- to 12-year-old to sink his teeth into, and the reviews suggest it makes a great starter thriller.

Next Week: Apparel.

Posted by: Mike Tanier on 10 Dec 2009

113 comments, Last at 15 Dec 2009, 4:25pm by Will Allen

Comments

1
by MJK :: Thu, 12/10/2009 - 12:36pm

Loved the article!

But then again, I have an advanced scientific degree and a passion for probability. I wonder if you're going to turn other readers off this week...

7
by billsfan :: Thu, 12/10/2009 - 1:11pm

I've always suspected that the number of readers of this site with advanced degrees was frighteningly high. Maybe this year's end-of-season survey should include a "highest level of education completed" question!

The only thing missing from this week's column was an obscure 80's pop-culture reference.

(I also like the Eagles)

25
by Kevin from Philly :: Thu, 12/10/2009 - 3:52pm

An article that confirms that the Cowboys suck in December? Who could get turned off by that?

2
by AnonymousA (not verified) :: Thu, 12/10/2009 - 12:42pm

Very disappointing. Stick to obscure witticisms; avoid sample size theater. Sad to think that on a supposedly stats-oriented website anyone would buy such a worthless "study".

43
by Sancho, Porto Alegre, RS, Brazil (not verified) :: Thu, 12/10/2009 - 7:59pm

This is the WALKTHROUGH! Geez, where is your sense of humor...

3
by Christopher (not verified) :: Thu, 12/10/2009 - 12:43pm

Wow, so the quarterback of a team that blew up its roster is excited to be in the playoff hunt, and you go and poop in his cheerios? While the comment is ridiculous, it reads like it was given in answer to the question "Does this feel like a year of destiny?"

Also, don't you sort of make his argument for him? The 7-5 Jaguars sure aren't 7-5 due to an abundance of competence.

4
by DrewTS (not verified) :: Thu, 12/10/2009 - 1:05pm

I think the mockery-worthy part of Garrard's quote was the bit about 2007 feeling like a year of destiny too. In what way is that encouraging to him, unless losing in the divisional round is the highest thing the Jaguars can ever hope to acheive? (OK, maybe he's onto something there.)

On a completely unrelated note, has anyone else started noticing word processor codes showing up in the captcha words? The other day, I saw an arrow, and today, I saw the indent symbol. How am I supposed to type that?

6
by DrewTS (not verified) :: Thu, 12/10/2009 - 1:08pm

By the way, the sponsor message on the 2007 Jaguars page on PFR is freaking priceless.

http://www.pro-football-reference.com/teams/jax/2007.htm

9
by billsfan :: Thu, 12/10/2009 - 1:19pm

I love the different rates for sponsoring pages:

2007 Patriots: $85
2007 Giants: $55
2008 Lions: $60
1997 Colts: $15

(I also like the Eagles)

24
by DelawareSteeler (not verified) :: Thu, 12/10/2009 - 3:52pm

the "go the the trade advisor for help" bit?

47
by mshray (not verified) :: Thu, 12/10/2009 - 8:19pm

That's not the most mockery-worthy thing I've heard Garrard say this season. A couple weeks ago on the Football Night in America pregame show Dan Patrick asked him about his come-from-behind 4th qtr drive & he said: "Well, I had the Lord on my side and He really came through for me."

So I guess God told Garrard that He had the Jags in His parlay that day. Or maybe God just told him that He has it in for the Jets. I really hate it when ahtletes (Trot Nixon comes to mind) say stuff like that. I was waiting for Garrard to say something appropriate after the loss to the mighty Forty-Niners, like: "Well God decided to humble me by making me lose two critical fumbles," but he didn't. So please, mock Garrard all you want.

And for what it's worth, I told my dad, a retired pastor, about this & he was every bit as ticked off as I was.

56
by Vincent Verhei :: Fri, 12/11/2009 - 12:36am

The only problem is that we've heard the "we won because God likes us better" so often, it's old now. Honestly, I'm sure you could find at least a dozen players after every game who would tell you this (and they would mean it).

58
by mshray (not verified) :: Fri, 12/11/2009 - 1:33am

No question whatsoever Vince, especially the part about it getting really old. I only mentioned it this time because A) Mike did a pretty good job of portraying Garrard as someone prone to other foolish utterances, and B) it sure looked like God was quite specifically not on his side in SF.

I suppose most people think of Kurt Warner right away on this topic, but from what I've observed, he is actually not as bad about this as many think. He is more likely to say (such as after the NFC championship last Jan) "I give all glory to God", i.e. implying that he doesn't need to take any for himself. And I've also heard him say that after a loss.

61
by Bobman :: Fri, 12/11/2009 - 5:18am

mshray, you mean Warner BLAMED God for a loss? Sweet. Well, I guess Kurt giveth....

I am under the impression that Warner said something like "Thanks You, Jesus!" after their SB win, which is pretty innocent, but one could infer that Jesus had particular interest in helping Kurt win that day.

79
by Bruce G. (not verified) :: Fri, 12/11/2009 - 2:45pm

Out of curiosity is it:

A) Only okay to thank God if I also blame him in equal proportion.

B) Only okay to thank God outside of the public eye.

C) Never okay to thank God for anything.

I agree Garrard's choice of words was poor, but does that reflect poorly on God or on Garrard?

Is one allowed to thank God for winning if he also thanks Him when he loses, just for having the opportunity to be in the game? I'm of the opinion that some players (Warner included) if you got to ask them afterward, would likely feel this way...but more often than not, the 'losers' of the game are not interviewed in the post game for the national audience. And the only time we ever see losing side comments on a national level is when someone explodes and/or creates a controversy (thanks to our tabloid media mentality) i.e. the Mora Sr. 'Playoffs' gaffe or a failed 4th & 2 against an undefeated team.

82
by mshray (not verified) :: Fri, 12/11/2009 - 3:09pm

My dislike of Garrard's comment is that he specifically said God chose his side. This was fine in the Old Testament...heck, God didn't just help the Israelites win games, he specifically commanded them to carry out ethnic cleansing (c.f. the book of Joshua). However this is not what Jesus preached (c.f. the parable of the Good Samaritan).

My distinction with Warner is that he 'praises' God, and he's done that even when he lost a big game. I don't think he ever has or ever would blame God. I have no idea what Garrard thought after the game against SF. Maybe he blamed God, maybe he still thanked God for the opportunity to play, but I can be sure that he didn't say God was on his side and came through for him. Because how could God lose a fumble, much less two of them?

So yes, I think this reflects poorly on Garrard. At least if he's ever had a WWJD bracelet.

83
by Jesse W. (not verified) :: Fri, 12/11/2009 - 3:22pm

"Is one allowed to thank God for winning if he also thanks Him when he loses, just for having the opportunity to be in the game?"

I think so. I think most sensible God-fearing athletes are of this mindset: that God wants them to be thankful for the gifts and talents that allow them to play at such a high level, win or lose. When athletes credit God in interviews, I usually interpret what they're saying as fitting that perspective.

(And please, no jokes about "sensible" and "God-fearing" being an oxymoron.)

99
by Rick (not verified) :: Sat, 12/12/2009 - 9:56am

While Garrard's use of God's intervention is purely and simply absurd, I know what the players are TRYING to say, but often lack the education to communicate. Whether they win or lose, the benefit of God's presence allows them to play at the highest level.
God doesn't decide to strip the ball or allow it to bounce off someone's hands and into the hands of another player. But He does endow these athletes with a skill level which surpasses that of almost everyone who posts on these boards.

In essence, every time one of these guys thanks God for intervention, what they are really saying is that they are thanking God for giving them the skill level that they have to perform the task required of them. Unfortunately they simply aren't capable of saying it that way, because (like so many of us) either they lack the speaking skill to do so, or they confuse correlation with causation. God didn't CAUSE the fumble any more than He CAUSED the TD catch. What God did provide was a level of talent which allowed one athlete to make a catch, and allowed another to make a hit big enough to shake the ball loose.

It's unfortunate that people who love to get critical on the "God thing" lack the nuance to understand this and immediately attack the literal aspect of what is said by poor communicators. It's also unfortunate that some of these poor communicators may also lack a good understanding of statistics.

I have no issue with athletes crediting God for His role in their victories. Part of the reason you rarely hear anyone mention God when they lose is because you rarely hear losers interviewed. I have, in fact, heard several interviewees says "it's God's will" when they were asked how they felt after a loss. Since they didn't step up and say "God caused me to fumble", the anti-religious crowd like to point to the winners and point out how stupid they sound by comparing them to the rarely heard (and often soft spoken) losers.

Now, as far as the issue at hand goes - Jacksonville is no more the team of destiny than the 2008 Lions were. Of course, you could say the Lions WERE a team of destiny...and God must hate Detroit. In more ways than just football success is concerned. I'm not sure what happened there, but it must have something to do with some kind of human sacrifice for things to have gotten so bad in the Motor City.

100
by Gruntled (not verified) :: Sat, 12/12/2009 - 10:43am

Yes, some of them mean exactly that, but many of them mean something else, up to and including that God favors them or answered their prayers and caused their team to win. There is no way you can pile every religious person in this country into one box and assert that they all have exactly the same take on how God orders the universe.

Though I am now a long time member of the church of how the hell would I know, I was involved in some fundamentalist/evangelical/charismatic churches when I was much younger, and I can promise you that there are players who not only directly credit God for their individual or team's success, but also at least some who occasionally blame the devil for their failures, though up to now I have not seen an instance of anyone who was unwise enough to vocalize that to the press.

Oh, and coincidentally, members of some of the groups I am referencing also have a tendency to define other groups of people in narrow and simplistic terms.

64
by DeltaWhiskey :: Fri, 12/11/2009 - 7:16am

Was 2007 the year he threw a ridiculously low number of INTs?

78
by phillyfinfan (not verified) :: Fri, 12/11/2009 - 2:39pm

well, I think beating the Steelers in the wildcard round was their defacto superbowl. For a struggling franchise a playoff win is extraordinarily meaningful-something Eagles fans, who seem to take them as their due (until the championship game, of course), couldn't possibly get.

5
by Tim Wilson :: Thu, 12/10/2009 - 1:07pm

Don't your binomial expansion probabilities assume that all other factors are normalized between December and the rest of the Cowboys' year? This year, that is certainly not the case, as their schedule gets much tougher in December, and in 2007, the Cowboys benched their starters for the last 2 games of the season, half of their December schedule. In such a small sample size, variables like this can make a huge difference.

32
by mm (not verified) :: Thu, 12/10/2009 - 4:25pm

The DVOA analysis seems to confirm that something real is going on (since DVOA adjusts for schedule). However, schedule still seems the biggest answer here. After all, the Cowboys were undefeated last season when playing teams that didn't make their Conference title game.

51
by Tim Wilson :: Thu, 12/10/2009 - 9:21pm

Yep, agree with all your points. The analysis done here does not seem to factor in things like the Cowboys benching starters down the stretch in 2007 though (and DVOA wouldn't factor that in either).

I'm not saying that things like TO's late season injury or OL injuries should be takne into account, because every team has injuries and an inability to withstand them would indicate a lack of depth, but I do think that in this VERY small sample size, even a 2 game aberration in 2007 could make a significant difference.

8
by Monkey Business (not verified) :: Thu, 12/10/2009 - 1:12pm

Much ado is made about scheduling, but only two games a season are based on standings.

There's nothing redeeming about the Jaguars. They're headed full steam for a first round beating that will make 59-0 look like a hard-fought game between two playoff-tested opponents on the frozen-tundra of Lambeau-field.

27
by Brendan Scolari :: Thu, 12/10/2009 - 3:54pm

"Much ado is made about scheduling, but only two games a season are based on standings."

Yes, but at least 4 games will be different when comparing two teams. For example, let's compare the Saints and the Bucs. The Saints get to play the Rams and Lions while the Bucs get to play the Seahawks and Panthers. But the Saints also get to play the Bucs twice (two easy wins), while the Bucs have to play the Saints twice (two sure losses). Makes a pretty big difference, no?

93
by BigCheese :: Fri, 12/11/2009 - 7:20pm

Not to mention that the scheduling discussed here is between the team in december and the same team the rest of hte season.

- Alvaro

10
by Tim McKenna (not verified) :: Thu, 12/10/2009 - 1:52pm

The binomial calculations are correct, but they ignore the post-hoc nature of what you are doing. What we are seeing from the Cowboys isn't that unlikely if we instead think about what were the chances that 1 of the 12 or so above average NFL teams would suck this much over a four year period in December? The odds of that happening are much better and it is likely this focus on the Cowboys sucking is because they happened to be the above average team that has had a bad run in December.

Perhaps if someone could come up with a coherent explanation for why the Cowboys suck in December we could start thinking about whether the evidence backs that up. Instead the only explanations we have about why the Cowboys have been bad in December are the same old lame verbal diarrhea about character and heart. Perhaps the coaching staff is the problem but I haven't seen anything convincing about why Phillips (who admittedly is likely an idiot) is a worse coach in December than in September through November. Without a reasonable story to back this up the statistics we are seeing should probably be ascribed to chance.

17
by Pat (filler) (not verified) :: Thu, 12/10/2009 - 3:02pm

What we are seeing from the Cowboys isn't that unlikely if we instead think about what were the chances that 1 of the 12 or so above average NFL teams would suck this much over a four year period in December?

Um... I'm pretty sure the chances that 1 of the 12 would experience a 1-in-50 decline are about 1-in-4.

Perhaps if someone could come up with a coherent explanation for why the Cowboys suck in December

The most obvious reason for a team starting strong and finishing weakly is lack of depth. Is that the case for the Cowboys? Dunno, but if it wasn't fixed it'd persist, and it could be very difficult to see otherwise. (note that you don't need to lose players for lack of depth to hurt you, if you're inclined to play players more injury-questionable due to the quality of their backups.)

28
by tjmckenn :: Thu, 12/10/2009 - 3:56pm

1. A 25% chance of something happening is pretty high. Certainly wouldn't make the cutoff for rejecting chance as an explanation in the standard academic and legal framework (where admittedly arbitrary and somewhat silly cutoff is 5%).

2. This isn't really a reason for why the Cowboys suck in December. I don't recall their depth being a particular issue (but I don't really remember either way). Romo didn't get hurt in December I think, so that is one big position that hasn't been causing any depth issues.

33
by Pat (filler) (not verified) :: Thu, 12/10/2009 - 4:45pm

1. A 25% chance of something happening is pretty high

No, it's 1 in 4. That's what it is. It's not "high" or "low." It's 1 in 4. If we did 12 random draws like that 4 times in a row, we'd expect to get a result that off once.

It's a game. Of course you're never going to reject chance. You can only barely reject chance as an explanation for the results of the entire season. The fact that it's not the common result (i.e. it's less than 50%) should make you wonder, but of course it's not going to prove anything.

But the whole idea stems from the idea that wons/losses are random draws from a distribution centered around a "true" won/loss percentage, which is completely silly and absolutely false anyway, so you'd never expect it to be significant (to clarify, it's impossible for wons/losses to be random draws from a true won/loss percentage because one team's wins are another team's losses).

In other words, suppose you do this analysis and find that the Cowboys' decline was so strong, there was only a 0.01% chance of it ever showing up. It's still not significant, because the underlying model is impossible. Quibblers may say 'no, it is significant, because then if you go and look at those wins/losses, you'll find a reason for those losses, therefore it was significant'. The point is that the distribution of results here won't follow a binomial distribution until you get to large samples, because the results don't look like a random draw until the opponent looks average.

In fact, even using the binomial probability stuff as a first indicator isn't a great idea: an 0.01% 'chance' situation might have an obvious answer (one season, opponents, etc.) but a 50% 'chance' situation might be the one that when you apply a model that could conceivably be correct actually is the really unlikely one.

(n.b.: This is part of my criticism of people who do '4th down likelihood' silliness: treating something that cannot be a random draw as a random draw is just a recipe for 'fail')

(where admittedly arbitrary and somewhat silly cutoff is 5%).

You put a relatively-strict cutoff (although 5% is ridiculously weak for more data-driven sciences) because of the large number of analyses, and the fact that you typically can decrease the likelihood by increasing the data size or repeating analyses.

To defend Mike, though, all the criticisms are relatively pointless since he looks at DVOA then anyway.

40
by DaveRichters :: Thu, 12/10/2009 - 6:56pm

You say:

"1. A 25% chance of something happening is pretty high

No, it's 1 in 4. That's what it is. It's not "high" or "low." It's 1 in 4. If we did 12 random draws like that 4 times in a row, we'd expect to get a result that off once."

I say:

I think he was saying that he wouldn't be all that surprised if he flipped a fair coin and got heads twice in a row. Neither would I, so I agree that a 25% of something happening is pretty high. However, I also agree with you that it is 1 in 4, but I don't think you've brought much to the table with that observation.

You say:

"In other words, suppose you do this analysis and find that the Cowboys' decline was so strong, there was only a 0.01% chance of it ever showing up. It's still not significant, because the underlying model is impossible."

I say:

I'm not sure what you mean here. Are you saying that there is no difference in performance comparing December to not-December that you would believe? There's no problem with using the binomial distribution to see if the difference is what you could expect IF wins are a probability distribution. It doesn't matter at all if they are or aren't a probability distribution. If the p-value is sufficiently low then you've learned that it is unlikely to get such a difference without there being another factor. If the p-value indicates that it is likely to get such a result just by chance then there's nothing weird about the performance difference.

You say:

"You put a relatively-strict cutoff (although 5% is ridiculously weak for more data-driven sciences) because of the large number of analyses, and the fact that you typically can decrease the likelihood by increasing the data size or repeating analyses."

I say:

The 5% cutoff simply reflects the tradeoff between 2 types of possible errors. The community has decided that false alarms are way worse than misses.

44
by Pat (filler) (not verified) :: Thu, 12/10/2009 - 8:00pm

As an example: Teams A, B, C, D. Team A wins 0.75 of the time against B, C, but wins 0.25 of the time against D. Team B, C win 50% of the time against each other or D. If each team plays 1 game against the other, A, B, C will have a winning percentage of 0.583, B, C, will have a winning percentage of 0.417, and D will have a winning percentage of 0.583.

After 8 years, Team D is 6-2 vs Team A, and someone asks "does Team D just have team A's number?" and Mr. Statistics A comes out and says "well, no, because the chance of a team with a 0.583 winning percentage winning 6-2 or better is ~30%. Totally not surprising!"

Except Mr. Statistics B comes out and says "yes, and Team A has a winning percentage of 0.583 too: the chance of going 2-6 is 5% - small, but not too small. We've got 4 teams - we'd see something like this oh, 20% of the time for any given 8 year period."

But then Mrs. Statistics C (whose name is Jill Bames) comes out and says "look, you're crazy. If they play each other they can't both win 58% of the games. Excluding the games they play against each other, Team A wins 75%, Team D wins 50%. A 75% win team should win 75% of the time against a 50% opponent. The chance they go 2-6 is less than 0.4%. Team A can't beat team D. There's probably something there."

The p-value is an indicator of how likely an event would be to occur within the variation seen in the framework of your model. If the model's flawed (which it is, since it can't possibly represent the data set) then the p-value doesn't mean anything. Just note what Mike Tanier points out: the Cowboys swoon in December has boosted the Eagles record (and performance) in December as well. You can't say "you'd expect to see something like this in a league of 32 teams" because your model can't represent the league's distribution anyway.

53
by DaveRichters :: Thu, 12/10/2009 - 9:44pm

The issue is not whether NFL win/loss records are determined by a binomial probability distribution, they obviously aren't and the idea is completely ridiculous and perverse. The issue is whether or not the binomial distribution can describe the data, and I don't see why you think it can't. There is clearly a violation in the binomial assumption that each trial has the same probability of success, but my guess is that it is close enough even with that violation. But I certainly think that this article should have established that the test is an appropriate one given that violation.

57
by Pat (filler) (not verified) :: Fri, 12/11/2009 - 12:45am

The issue is whether or not the binomial distribution can describe the data, and I don't see why you think it can't.

I don't know why you think it can. For one team, over many seasons, it'll approximate one, via the central limit theorem. But teams don't play "average seasons," they play specific ones, and those seasons aren't the result of a binomial, independent process for each team. It's one thing to say that a team that's 0.400 over a 10 year span will win 8 games in X% of those seasons. It's another thing to say that they have an X% chance in a specific season.

But for the NFL as a whole? And keep in mind, that's what you're comparing against when you say "is the Cowboys' December swoon significant?" and conclude "well, 20% of the time you'll get a 2% outlier like this in the league..." (again, see example above) - that doesn't work. Not at all.

If you want a crazy example, if a team had a 90% chance against one team, and a 10% chance against another, their final winning percentage would average to 50%, but it wouldn't come close to looking like a binomial distribution.

There is clearly a violation in the binomial assumption that each trial has the same probability of success

Yeah, that's like... the entire definition of a process that results in a binomial distribution. Given that, I have no idea why you think it could look like one.

The point is that the NFL can be represented as a flip of 256 coins whose weights are determined by something like Bill James's log5 method between two teams. That's a reasonable model, and it actually works somewhat well. Determining the "true" winning percentages requires fitting a maximum likelihood, but hey, that's easy.

59
by DaveRichters :: Fri, 12/11/2009 - 1:47am

I doubt the log5 method would work in football. It is really only meaningful when win proportion is near .5. In that range, a binomial distribution work well, by the way. But the fact of the matter is that a binomial distribution can easily describe this, the only question is what the range in win likelihood is. Certainly with real win likelihoods of .9 and .1, say two of each over a four game stretch, there would be a big difference between the truth and the binomial prediction. But if they were in the .6 to .4 range, it is close enough.

I'd be amazed if the binomial could not describe the distribution of wins in the league in a given year, despite it obviously being theoretically dubious to do so.

80
by Pat (filler) (not verified) :: Fri, 12/11/2009 - 2:48pm

I doubt the log5 method would work in football.

Sure it would. Take a look at any of the game output functions used by maximum likelihood methods, and a log5 curve would approximate them to a good degree.

Log5 is derived under a basic assumption: "flip two weighted coins, A and B. If A=1,B=0, A wins. If B=1,A=0, B wins. If A=B, flip again." Work out the math, and you get (A-BA)/(A+B-2AB). It has nothing to do with a win proportion near 50% - it's pretty much the absolute simplest way you can convert a "team is a weighted coin" model into a workable model for a league.

I'd be amazed if the binomial could not describe the distribution of wins in the league in a given year, despite it obviously being theoretically dubious to do so.

Be amazed then, as it can't. The NFL averages to 0.5 winning percentage, obviously, and a distribution of 32 draws of 16 with a win percentage of 0.5 doesn't fit the distribution of wins in any league year.

95
by DaveRichters :: Fri, 12/11/2009 - 8:44pm

It does have to do with a win proportion near .5. The winning percentages are basically "if team A played average teams, what proportion would they win?" There's no reason to think that log5 works well to predict what happens when a very good team plays a very poor team.

But you are probably correct that the binomial distribution can't model the win distribution very well. With p=.5 and n=16, the variance is 4 wins, which is almost certainly too narrow for reality. I thought I would be way more amazed!

97
by Pat (filler) (not verified) :: Fri, 12/11/2009 - 10:48pm

The winning percentages are basically "if team A played average teams, what proportion would they win?"

log5 is simply a method for determining a "winner" of two weighted coin flips. How the weights (the winning percentages) are determined is a completely different question. You could estimate it from the season winning percentage (pretty accurately in a game like baseball, less so in a game like football) or by maximizing the total likelihood from the season that occurred.

Said another way: log5 is just an example of a function such that given two numbers tA, tB that represent team A, B's "true strength," log5(tA,tB) is the fraction of times that A beats B (Surprisingly, the form of the function doesn't really matter all that much).

There's no reason to think that log5 works well to predict what happens when a very good team plays a very poor team.

There's no reason to think log5 would work well ever - you can trivially define the rules of a game such that it fails, even if the winning percentages are near 50%. You use it because it's a first-order approximation. It's not going to fail by much unless the game isn't well modeled as a first-order contest (e.g. the example above) - that is, if you've got A, B, C, and you know the number of times B beats C on average, and the number of times A beats B on average, you can determine the number of times A beats C on average (And the NFL specifically can be explained entirely as first-order contests).

102
by DaveRichters :: Sat, 12/12/2009 - 1:41pm

There are some inherent assumptions that are being made that are not established, and most likely not true. I agree that log5 would be better than using .5 for a test like the one done here; I'm just pointing out fairly obvious flaws. One is that the method was only tested for teams with winning proportions near .5, so it is in no way established that you'll have good estimates in football where winning proportions have far more variance. It says a 14-2 team would beat a 2-14 team 98% of the time, and that seems too high to me, partly because Drew Brees has some chance of getting hurt.

Moreover, the log5 method assumes that there are no interactions between teams, so that the two numbers tA, tB that were derived from win results against a distribution of teams can be meaningfully used. This assumption is trivially false, and I believe it is universally accepted. In fact, your earlier example with teams A, B, C, and D included interactions.

In general, though, I agree with your point that the log5 is superior, but there are many distributions that are better than the binomial for this one. The binomial will underestimate the probability of a poor December, so at least the test establishes a lower bound (given that we know the Cowboys didn't lose against a weaker than average schedule).

108
by Pat (filler) (not verified) :: Mon, 12/14/2009 - 9:32pm

There are some inherent assumptions that are being made that are not established, and most likely not true.

There aren't assumptions. It's just a method for deciding the victor of two weighted coin flips. The only assumptions are that the weights are independent (or can be cast into an independent form) which is exactly the same as saying that it's a first-order contest.

It says a 14-2 team would beat a 2-14 team 98% of the time, and that seems too high to me, partly because Drew Brees has some chance of getting hurt.

All that says is that the team wasn't an 0.875 winning percentage team. That's a criticism of the method for determining the weights, not the method for the victory.

Moreover, the log5 method assumes that there are no interactions between teams, so that the two numbers tA, tB that were derived from win results against a distribution of teams can be meaningfully used.

That's what I said in the second paragraph. log5 is going to be a reasonable assumption so long as the contest is first-order.

This assumption is trivially false

That's actually wrong. The NFL is, to an extremely good approximation, well-described as a series of first-order contests. There may be actual second-order effects, but they are not required to describe the league.

In fact, your earlier example with teams A, B, C, and D included interactions.

Yes, that's the point. That's why you use something like log5 - which assumes no interactions - to demonstrate that it's not consistent with that. A binomial comparison makes no sense, which you can easily verify by trying to construct a positive/negative control.

but there are many distributions that are better than the binomial for this one.

It's not about the choice of distribution. It's the choice of model. Any probability distribution for wins in a season will fail to be a good model for something like a 'probability of this happening' test. Any one.

Even if you used the actual distribution of wins itself! Why? Because the results of a team's season aren't independent from the other 31 teams - you can't have more than 4 16-0 teams in the entire league, for instance, due to the schedule format. That's why I keep pointing to Mike's conclusion regarding the Eagles "December surge." Wins simply aren't an independent variable.

That's why you have to set up the problem such that you are measuring something independent - namely, "wins minus expected wins," which could be made to be independent.

110
by DaveRichters :: Mon, 12/14/2009 - 10:17pm

There aren't assumptions. It's just a method for deciding the victor of two weighted coin flips. The only assumptions are that the weights are independent.

Right. So we agree. I think.

All that says is that the team wasn't an 0.875 winning percentage [sic] team. That's a criticism of the method for determining the weights, not the method for the victory.

The log5 method is defined by the way it determines weights. I think transforming the win proportions first would be a good idea.

There may be actual second-order effects, but they are not required to describe the league.

I think you'd be better off, when assigning win probability for a game, to be aware of, say, the starting QB missing the game with injury or whether the game is home or away, or lots of other things.

Any probability distribution for wins in a season will fail to be a good model for something like a 'probability of this happening' test. Any one.

You can just have a game as the level of analysis, even for the binomial, so that each trial results in a win for one team and a loss for another. But the log5 is a probability distribution specifically for determining the probability of a win for any game. So I'm not sure what you mean to say.

111
by Pat (filler) (not verified) :: Tue, 12/15/2009 - 3:40pm

The log5 method is defined by the way it determines weights. I think transforming the win proportions first would be a good idea.

I really don't know why we keep disagreeing here. Maybe we're using "weight" differently or something. log5 says nothing about how you determine tA, tB. It only defines f(tA, tB). If you have a team that you determine has tA = 0.875 and another that you determine has tB = 0.125 and team B actually wins 10% of those games (so f(tA, tB) = 0.9) then your determination for tA, tB was off, and you go back and adjust those accordingly.

Bradley-Terry models and Elo-type rankings (which are just stepwise versions of that) use a different sigmoidal function (typically f(tA, tB) = tA/(tA+tB)) but that's just for simplicity in fitting since it makes the log likelihoods all pretty and simple.

I think you'd be better off, when assigning win probability for a game, to be aware of, say, the starting QB missing the game with injury or whether the game is home or away, or lots of other things.

Better off than what? A full-blown model taking into account injuries, field location, etc. is of course going to be better. There's more information. It's also a heckuva lot more work.

But the step from "assume wins are independent and determined by a random variable with weight equal to a winning percentage" to "assume wins are determined by a log5 comparison using winning percentages from non-common games" is huge - after all, you'll now actually be able to represent a given season, and your statistics won't be complete bull. Going from there to a maximum-likelihood model is a similarly big improvement (you'd now be able to represent multiple seasons well).

And I think going from that to a full-blown crazy model would actually only result in marginal improvement for a situation like this.

But the log5 is a probability distribution specifically for determining the probability of a win for any game.

log5 determines win probabilities for a single contest. Not for a season. That's the difference: you can only get distributions which match the NFL's actual season distribution by simulating the actual contest distribution, since they have structure.

Just as an example, imagine trying to figure out how often, given a probability distribution of team strengths P(tX), a team would go undefeated. How would you do it? Throw a random variable from P (tA), generate 16 random variables from P (t_i), multiply f(tA,t_i) for all 16 t_i, repeat 32 times and sum it? Yeah, that won't work, because the probability of five 16-0 teams is strictly zero, not "infinitesimally small", so your summation was incorrect as the probabilities were joint (the number of times even 2 16-0 teams shows up is very different than what you'd expect).

Now, you might say "well, yes, but in this case, we're only concerned about one team." Yes, that's true - but when you want to calculate how likely it is that a team like that shows up in the league - you've just added the league structure, and all other 31 teams.

71
by Kevin from Philly :: Fri, 12/11/2009 - 1:25pm

If some scientist said their was a 25% chance a dinosaur-killing sized asteroid would hit Earth next year, I'd say that was REALLY high.

81
by Pat (filler) (not verified) :: Fri, 12/11/2009 - 2:50pm

And if a Wall Street investor said there'd be a 25% chance that if you invested $10,000, there'd be a 25% chance you'd quadruple your investment - otherwise you'd lose it - you'd probably say it was too low.

92
by Kevin from Philly :: Fri, 12/11/2009 - 7:18pm

Actually, that's a pretty good representation of my stock picking record. At least the "lose it all" part.

11
by nat :: Thu, 12/10/2009 - 1:52pm

One small nit to pick about the binomial expansion stuff.

You didn't choose the Eagles and the Cowboys at random. You chose them because you had already noticed they were outliers as far as December records go.

So, you should be asking "how likely is it that the league's outlier has a December record like this?", not "how likely is a typical team of this strength to have a record like this?". The math gets tougher, and you need to make assumptions about the distribution of teams' strengths. But you get thirty-two attempts to get an outlier, not just one, and you will find that outliers like these are pretty common.

12
by jebmak :: Thu, 12/10/2009 - 2:09pm

Good point, I hadn't thought of that. Thanks for bringing it up.

16
by Neoplatonist Bolthead (not verified) :: Thu, 12/10/2009 - 2:52pm

It's hard to talk about W/L records in a given month with any real confidence. You can talk about W/L records in general, and infer, say, that the Colts have been a better team than the Raiders over the last few years. But for one month, you have such a funky sample size that any observations are hard to trust. Sure, there's some reason to suspect that the Cowboys have something going on that causes them to play well at first but run out of steam late in the season, the same way that one can make the opposite supposition about the Chargers or Eagles. But n<30 for any period of time in which n December games could be considered to have been played by the same team (as opposed to a bunch of mostly-different guys in the same uniform).

Me, I think it is real. In the '07 preseason, when the Cowboys just manhandled the Chargers, I looked at my buddy and said "those Cowboys are unstoppable. Look how big their line is." His response was, "yeah, those linemen are really big and athletic. How many of them will still be playing in December? They're going to run out of gas." Is that the reason? Maybe, maybe not. Certainly, I think the team leaders and coaches play a role in teams that gain or lose momentum at the same time every year, but I have no idea how I'd try to prove that in a team sport with a short season and rapid player turnover.

Some teams do seem to get more motivated for one thing or another, and it doesn't seem to happen at the same level with the same teams all the time: Indy is known for lightening up once they have what they want, Oakland and KC play better in each other's houses, New England is generally very good at avoiding back-to-back losses, San Diego really turns it on when they wear their alternate unis. This stuff is all strictly psychological; that doesn't mean it's not real.

23
by tjmckenn :: Thu, 12/10/2009 - 3:51pm

Do you really believe things like SD's uniforms and NE not losing back to back are real? It seems like the SD uniform thing is exactly the kind of conclusion you should not be drawing. Lots of teams play in alt unis. One of them (SD) does well. Concluding that SD performs better in its alternative uniform is supported by the data, but is that really a good explanation for what you are seeing?

NE plays well most weeks, not just the week after a loss. If NE really did have a special skill at avoiding back to back losses shouldn't they have managed to pull out the game last week?

Now Indy lightening up after they have secured (homefield, playoffs, etc) is reasonable as one would expect most teams to not try as hard once winning doesn't really matter.

19
by Temo :: Thu, 12/10/2009 - 3:23pm

Was going to respond with more, but what this guy said is sufficient.

Not that I'm criticizing, however; it was a pretty good analysis in total as far as sports goes. It wouldn't pass muster for publication or anything, but whatever.

There probably is something going on.

18
by Pat (filler) (not verified) :: Thu, 12/10/2009 - 3:12pm

You chose them because you had already noticed they were outliers as far as December records go.

Not sure I'd agree. If the 49ers had a December swoon of similar degree, it's doubtful it would've been noticed. The Eagles and Cowboys were chosen because they were high profile (i.e. playoff) teams with outlying December records.

Which means it's more likely that the number of teams of interest is actually 12, not 32. But you'd be better off making a distribution of December changes for the entire league - if the Cowboys are the far outlier, it's not that interesting. If there already are 1 or 2 other teams that far off... well, it becomes somewhat interesting.

The math gets tougher

Not really. You can make a rough estimate by just taking (1-p)^32 (or 12 if you buy the logic above), where p is the chance probability of your outlier. Or you can be even more sleazy and just compare the number to 1/32 (or 1/12). If the number of teams of interest is really 32, you're likely to get a team like this more often than not (it's above 50%). If it's 12... well, it still could easily be random, but 1 in 4 or 5 starts to make me curious (n.b.: my curiosity threshold is much weaker for games than it is for science).

29
by nat :: Thu, 12/10/2009 - 4:04pm

It's human nature to treat 1 in 5 events as signifying something. But in the real world, it's very easy to find them.

If you look at the points for and against for the teams in the AFC South, and read the numbers backwards (e.g. read 201 as 102), the points for AND the points against BOTH match the actual ranking of the teams. The odds against that are HUGE. (1/4!) * (1/4!) = (1/576). Even allowing for 8 divisions, that's still a 1 in 72 chance. Wow!

Of course, I just cast around a bit looking for patterns. It didn't take long. I assure you I didn't need to do 576 checks of this depth to find one. Which is why I don't get excited about 1 in 5 odds on arbitrary stuff like December record.

My personal opinion is that the Cowboys are either unlucky or a bad December team, but I can't tell which from this analysis. I hate 'em, so either way is okay by me.

35
by Pat (filler) (not verified) :: Thu, 12/10/2009 - 5:23pm

But in the real world, it's very easy to find them.

They occur 20% of the time. They're easy to find if your search space is large - that is, if you have something you can search a lot.

I assure you I didn't need to do 576 checks of this depth to find one.

Actually, you probably did - well, not 576 - as you noted you only need ~72 anyway, and you the most common number of trials needed to find a 1/72 chance isn't 72. This is the one thing that even statisticians and scientists don't understand all that well: your brain is a fantastic pattern recognition engine. Even if your conscious brain didn't think it was doing a ton of tests, it probably was (Hence the reason why it's nearly impossible to correct an a posteriori tuned search for number of trials).

But still, the reason it was easy to find is that your search space was essentially unbounded. Dismissing any given 20% result because you can generate an "apparent" 20% result in a huge search space is a bit of a strawman.

It's also a bit disingenuous because you didn't offer a hypothesis - you just offered a pattern. If you stated it as a hypothesis (I dunno - "the NFL is fixed to generate a pattern in the AFC South"?), it's easily disprovable, since it'll almost certainly go away next week. It has a low apparent a posteriori probability, but that's because of the high trials factor that wasn't accounted for.

In this case, the hypothesis was "there exists a top-level team in the NFL that declines in December" - and the result was "eh, maybe it might be the Cowboys." That result can then be retested with no trials factor in the future.

Which is why I don't get excited about 1 in 5 odds on arbitrary stuff like December record.

Looking for declines of a given team at the end of a season isn't nearly as large a search space. Since there is a hypothesis here, a 20% a posteriori chance probability isn't exactly a big deal, but since it's probably the largest one, it's worth testing this year, in which case you don't have the trials factor to compete with.

It's not worth "getting excited over," but it is worth considering.

41
by DaveRichters :: Thu, 12/10/2009 - 7:11pm

You say:

This is the one thing that even statisticians and scientists don't understand all that well: your brain is a fantastic pattern recognition engine.

I say:

You are half correct. Scientists understand this quite well, but I agree that many statisticians don't. Considering the insufferably large literature in the scientific community studying just how amazing the human brain is specifically at pattern recognition, I would say you are making a strange claim. In truth, the "one thing" that scientists don't understand all that well is why reviewer #2 is such a jerk.

45
by Pat (filler) (not verified) :: Thu, 12/10/2009 - 8:06pm

You are half correct. Scientists understand this quite well

Oops, I wasn't specific enough. Neuroscientists understand this quite well. That's not what I was trying to imply - sorry about that. Look around in the literature in non-neuroscience fields at people doing "blind" searches and trying to correct for the trials factor, and you'll be amazed.

I guess I mainly meant "scientists trying to do statistics."

54
by DaveRichters :: Thu, 12/10/2009 - 10:01pm

Yeah, yeah, neuroscience gets all the credit. But all the best findings re: how-awesome-the-brain-is come from experimental psychology. In general, in my experience, psychologists are also worlds better than neuroscientists at statistics.

42
by Will Allen (not verified) :: Thu, 12/10/2009 - 7:41pm

Yeah, I hate'em, too, so I just enjoy the spectacle, without regard to underlying cause. I sometimes feel bad about it, though, because I like a lot of their players, and the Cowboy fams at this site seem like decent folk, so it just seems wrong to enjoy something that causes such discomfort to those I have no issues with. My loathing of Jones is rather large, unfortunately. Watching his face, in the closing moments of a first round playoff loss, is really pleasurable. Doing so in the closing moments of a season ending regular season loss is better than 25 year old scotch.

66
by Dales :: Fri, 12/11/2009 - 9:04am

What did Felix ever do to you?

Oh, wait...

72
by Kevin from Philly :: Fri, 12/11/2009 - 1:29pm

Ugh. Paint thinner is better than 25 year old scotch. Try Irish whiskey - much better tasting.

76
by DaveRichters :: Fri, 12/11/2009 - 1:59pm

As far as Irish brands go, Redbreast is excellent. But please, you can't possibly suggest that paint thinner is better than Scotch and expect to be taken seriously. The Scots make some VERY drinkable whiskey.

89
by Kevin from Philly :: Fri, 12/11/2009 - 7:08pm

You're probably right, but I've tasted scotch a few times, and (as the card in the liquor store might read) there's always a hint of iodine and a distinct "NJ Turnpike, Exit 15" note.

94
by billsfan :: Fri, 12/11/2009 - 7:33pm

At least it wasn't Exit 12!

(I also like the Eagles)

113
by Will Allen (not verified) :: Tue, 12/15/2009 - 4:25pm

This is late, but now, now, let us at least attempt to be ecumenical in our regard for The Lord's nectars! If I make a remark of admiration for the lovely spirits on one side of the Irish Sea, that is in now way to be interpreted as a negative remark for those on the other side! Or for their fellow sainted liquids from Kentucky, for that matter! Sheesh!

13
by Agronomist (not verified) :: Thu, 12/10/2009 - 2:24pm

What exactly is a "scientific" degree? I suppose he was just being facetious.

While the effects of "heart" and "desire" have largely been demythologized in baseball, it's not as easy with football. There are cold hard facts about talent in football, sure; and although you can't will yourself to hit a Rivera cutter, you certainly can will or wile your way to that extra yard. The Cowboys look like a team that can line up and steamroll over anyone, but apparently that's not challenging enough for a Princeton grad with a "historical" degree. Not to mention the head coach has to go to work every day with the genius heir apparent breathing down his neck.

I don't know. I'm not saying I know why the Eagles do so much better in December than the Cowboys, but I wouldn't discount coaching or character or team culture any more than I would random probability. That being said, the Cowboys will now proceed to go 7-0.

30
by tgt2 (not verified) :: Thu, 12/10/2009 - 4:13pm

Wile and will are very different things. You're setting up a false equivalence in your comparison between baseball and football.

As for an actual comparison, I see no reason to say heart and desire impact football more than baseball. Your silly platitudes about trying harder for that extra yard are just as silly as baseball platitudes about trying harder to hit a cutter.

63
by pouringlizards (not verified) :: Fri, 12/11/2009 - 7:10am

I think the issue is that you can't will yourself to be more co-ordinated, but can will yourself to keep pushing even when it hurts to do so. I would say that's where the 'heart and desire' elements come in. Whilst I don't like to prove a general point with a single example, that Stafford TD after the dislocated shoulder is a good example of a time where 'trying harder' to get that extra play (not yard, but same difference I think) does make the difference.

Of course, there's still physical limits, but I think calling it a silly platitude is overlooking the different nature of the two sports.

67
by DaveRichters :: Fri, 12/11/2009 - 9:39am

With some tasks a mere increase in effort is helpful and with other tasks it can be detrimental. A RB that increases effort might gain an extra yard, but maybe a QB will make worse reads. A pitcher might throw a fastball a little faster and a hitter might swing at bad pitches more often.

I have no faith in platitudes, but it is clear from experimental data that an increase in effort is sometimes helpful, sometimes harmful, and is task-dependent.

101
by pouringlizards (not verified) :: Sat, 12/12/2009 - 1:28pm

Oh, of course it's task-dependent, and I take your point about QB reads. I could add, say, WRs snatching at the ball rather than playing with 'soft hands,'leading to a critical drop at a late stage of a game.

Look at the tasks involved in football- the finesse players are actually a very small part of what's going on. I'm talking about a defender trying to strip the ball and a RB hanging on, a Defensive Lineman pushed to the ground but scrambling and crawling through a block to get to the QB, a guard struggling to hold Albert Haynesworth up at the point of attack. You can't say that all that stuff's not effort-dependent.

103
by DaveRichters :: Sat, 12/12/2009 - 1:44pm

Yes, I completely agree with you. I think there is a far bigger role for effort in football than in baseball.

14
by Todd S. :: Thu, 12/10/2009 - 2:24pm

Mike, good stuff on the book recommendations. I still remember fondly my time spent in the 7th grade library, checking out sports books like "Crazy Legs McBain" and "Dugout Tycoon."

15
by Stevis (not verified) :: Thu, 12/10/2009 - 2:47pm

I'll let others pick at the binomial expansions.

My question is this: at what age level do you think it's appropriate to introduce readers to Monday Night Jihad?

21
by Temo :: Thu, 12/10/2009 - 3:43pm

The life of this Football Outsider in Training isn't without difficulty

Wouldn't actually being in an NFL front office make him an insider and not an outsider?

Edit: And how has Theo Epstein not written a similar book by now?

20
by Bobman :: Thu, 12/10/2009 - 3:37pm

My three sons love the three Barber brothers books they have--gifts from an aunt who is probably a Giants fan. We read and re-read them all the way home on a flight from NJ a few sumemrs ago.

The Garrard destiny stuff was priceless. But I think you missed the point--He means they are destined to have 7 wins. Mission accomplished!

68
by Anonymousse (not verified) :: Fri, 12/11/2009 - 10:25am

Jaguars were "destined" for 12 wins in 2007. If Jaguars have equal "destiny", they're going to win at least one playoff game.

Silly comment by Garrad, and obviously the 2009 Jaguars are the 2008 Dolphins, and will get crushed by whichever decent team doesn't get the first round.

22
by Bobman :: Thu, 12/10/2009 - 3:44pm

Oh, and what effect of locking in a playoff slot is taken into account with the PHI/DAl Dec records analysis? If you look just at Dec, the Colts bite (well. compared to their Sept-Nov record). 2006 excepted, they usually start the JV at least one game, and in 2005 Sorgi saw about 11 quarters of action, most of it in the last 2-3 weeks.

Now I know the NFCE is generally a season-long bloodbath with nary a clear winner until the bitter end, but with such small sample sizes, even one or two games lost because of the start 'em/rest 'em debate would skew the results a bunch, no? Did either team rest starters during this sample timeframe?

26
by Temo :: Thu, 12/10/2009 - 3:53pm

I know the Cowboys did in 2007, pulling their starters at half time of a messy, rainy game vs. the Redskins in week 17.

They did not lock up a spot in 2006 and didn't make the playoffs in 2008.

In 2005, I think they were eliminated before they ever took the field for the late game in week 17, making the game pointless. No starters were rested, but lets just say it's not like they went down swinging.

31
by dmb :: Thu, 12/10/2009 - 4:20pm

In that Week 17 game against the Redskins in 2007, the Cowboys weren't exactly gangbusters before they pulled the starters ... but it wouldn't be shocking if the Cowboys treated it as a practice, since (I believe) they already had a bye wrapped up.

34
by Temo :: Thu, 12/10/2009 - 4:58pm

They didn't look good and they did have the bye wrapped up. And it was a cold, rainy day in Washington with nothing on the line except for Romo wrapping some Cowboys team records (which I thought was silly).

Again, that's really the only "excuse" game they have.

36
by Spielman :: Thu, 12/10/2009 - 5:32pm

I'd love to have heard Cooper Manning's reaction to the Family Huddle book. I'm going for something along the lines of:

"Wait. You guys have more money than God, you decided to do a children's book about our childhood, and you used my likeness but cut me out of a writing credit and the profit? Seriously? You sons of bitches! Er... sorry, Mom."

37
by Temo :: Thu, 12/10/2009 - 5:39pm

Sorry Grandma too, I guess.

62
by Bobman :: Fri, 12/11/2009 - 5:26am

Awesome. I think Cooper, as a CFA or similar, handles a lot of family investments. With that in mind...

"Hey, Peyt, Eli, congrats on that awesome book. Good tims, I'll tell ya. Here, I took your advance checks and bought you some really great stocks at bottom of the barrel prices. Here, Eli, are yor Worldcom shares, and for you, Peyton, Enron, because they're based in Houston and that's part of the AFC South. Sweet, hunh?"

Oldest brothers have a way about them..... Don't you think if at some family gathering Archie tells the boys to get in the car for a little outing, Cooper automatically gets shotgun? Or if Olivia comes along and the boys are stuck in back seat, Eli is crammed into the middle, on the receiving end of double-barrelled wet willies? Oh yeah.

38
by Key19 :: Thu, 12/10/2009 - 5:41pm

Well, here's my best shots of reasoning:

1. Cowboys have a big offensive line that to me isn't in the best of shape. They tend to regress the longer the season goes on and they have almost no depth at O-line in case someone actually does go down. Anyone who saw Cory Procter play last December can attest to the lack of depth part.

2. Strength of schedule. Call me crazy, but I don't think most teams would have a winning record in December given their last three seasons:

2007: DET, PHI, CAR, WAS
2008: PIT, NYG, BAL, PHI
2009: NYG, SD, NO, WAS, PHI

I see two, MAYBE four games out of thirteen that you would really go into thinking "I fully expect the Cowboys to win." DET and CAR are two, and depending on your impression of Washington, maybe two more. All of the others have been absolutely brutal games. Last year, all four of those teams made the Divisional Round of the Playoffs. This year, the only team that is out of it right now is Washington. I'm not really sure why people expect that the Cowboys should just beat teams that are in the top 25% of the league. It is going to be very tough inherently. Give them a schedule of TB, CLE, WAS, and CAR and let's see how they do. If they lose 3 of those 4, then yes, we have a legitimate problem. But there's no shame in just being a pretty good team that isn't as good as great teams. There's a pretty strong concentration of hard games there in December. If that stretch came in September every year, everyone would say "WHY CAN'T DALLAS WIN IN SEPTEMBER OMG THEY JUST NEVER CAN START FAST" instead of "OMG THEY GET YELLOW THIGHS EVERY DECEMBER BECAUSE THEY CAN'T WIN WHEN IT COUNTS."

That's just how I feel.

39
by dmb :: Thu, 12/10/2009 - 5:57pm

I think schedule might be part of it, but keep in mind that their DVOA dropped noticeably, too -- and that takes the scheduling into account.

48
by Xeynon (not verified) :: Thu, 12/10/2009 - 8:45pm

Right. While that DVOA dropoff has to have been affected by playing their backups for two games in 2007, the Eagles have also played their backups in December games in the relevant timeframe without suffering the dropoff.

I don't know that "heart", "character", "intensity" or the like can really be discussed meaningfully given that they can't be quantified, but my own pet theory is that there are psychological factors involved. Over the past several seasons, the Cowboys have been constantly asked about their tendency to collapse in December, which after awhile has to become an annoying distraction if nothing else. Then there's the fact that T.O. in all his locker room dividing glory was around to throw temper tantrums or make a spectacle of himself or whatever the last few years. Finally, there's coaching - I can attest that Reid is generally a coach who adjusts very well between games. If you beat his team the first time you play them one way, he will generally come out with a different approach and probably beat you the second time you play unless you do something unexpected, as Arizona did last year. Phillips seems much more inclined to stay the course and stick with the strategies that won him the first matchup. Just an impression, but it does match the data.

50
by Pat (filler) (not verified) :: Thu, 12/10/2009 - 8:54pm

Depth would still explain that too, though. A team with weak depth will collapse a lot more by resting starters than one with strong depth.

Then there's the fact that T.O. in all his locker room dividing glory was around to throw temper tantrums or make a spectacle of himself or whatever the last few years.

Owens from 2006-2008 was an unusally large part of the Cowboys offense, too - if he declined over the season (due to, y'know, being old) that'd basically produce the same effect as weak depth (well, I guess it is weak depth, if you choose not to use your backups when the starter isn't as effective).

52
by Tim Wilson :: Thu, 12/10/2009 - 9:22pm

Owens was NOT a big part of the Cowboys offense during December of 2007, when he was hurt for 1.5 games and when he was benched (along with the rest of the starters) for 2 more.

55
by tuluse :: Thu, 12/10/2009 - 10:15pm

Which again would explain a worse play.

60
by t.d. :: Fri, 12/11/2009 - 3:15am

If dvoa accurately described how good a team was at that point, wouldn't it remain flat? I mean, it should be boosted against tough teams, and weakened against poor teams, so it should be schedule independent, and yet, when the Giants had their 5-0 start against shit teams, they were rated as an elite team. Dvoa was oblivious to the fact that they were a deeply flawed team that hadn't yet faced an opponent capable of exploiting their weaknesses yet. So if it's taking that into account, it isn't doing an adequate job of taking it into account.

65
by pouringlizards (not verified) :: Fri, 12/11/2009 - 7:40am

Well, we don't know as much about those crappy teams at that point, which is one thing. The other point is that we hadn't SEEN those flaws in the Giants. DVOA does seem to have a decent predictive ability once we get a decent sample size, but it isn't all-seeing, all-knowing from the outset. The discrepancy between DVOA and team records only really gets interesting when you get to later in the season, when there's more data for it to work with.

87
by Spielman :: Fri, 12/11/2009 - 5:53pm

If teams played at the same level every week, sure. I don't think you'll find very many people making that claim, though.

77
by sswoods (not verified) :: Fri, 12/11/2009 - 2:38pm

#2 is the key here, fellas.

Going back to 2005, since the article does, here's their Dec. record against teams that finished:

12-4: 1-1
11-5: 1-2
10-6: 1-3
9-6-1: 0-1
9-7: 0-1
8-8: 1-1
7-9: 3-0
6-10: 0-1 (05)
3-13: 0-1 (06)

Two bad losses, in terms of the quality of the opponent. Otherwise, they're playing quality teams, and can't be Philly (or Washington) in December. Expect a stumble this season because their schedule includes 12-0, 7-5, 9-3, 8-4, and 3-9.

46
by jgrenci@zoomint... :: Thu, 12/10/2009 - 8:11pm

"In other words, suppose you do this analysis and find that the Cowboys' decline was so strong, there was only a 0.01% chance of it ever showing up. It's still not significant, because the underlying model is impossible. Quibblers may say 'no, it is significant, because then if you go and look at those wins/losses, you'll find a reason for those losses, therefore it was significant'. The point is that the distribution of results here won't follow a binomial distribution until you get to large samples, because the results don't look like a random draw until the opponent looks average.

In fact, even using the binomial probability stuff as a first indicator isn't a great idea: an 0.01% 'chance' situation might have an obvious answer (one season, opponents, etc.) but a 50% 'chance' situation might be the one that when you apply a model that could conceivably be correct actually is the really unlikely one."

not exactly what Pat (filler) means here. The underlying model is impossible?? does that mean it has biases? if so, yeah, well many many models have biases, but we still work with them because it might be the best we have. I am not suggesting we get careless in studies...

.01% by the way still means .01% NO MATTER THE SAMPLE SIZE. For example, the probability of flipping a coin ten times in a row to be a head is 1/1024 = .1%. Small sample size, but nonetheless a small percentage.

Of course, there are several faults with the Cowboy situation which has been brought up by several people. the biggest is that we are looking at things after the fact. A true hypothesis starts with a hypothesis and THEN collects data.

the other problem is that we have not adjusted for strength of schedule, resting players, etc.

49
by Pat (filler) (not verified) :: Thu, 12/10/2009 - 8:47pm

not exactly what Pat (filler) means here. The underlying model is impossible??

Yes, it is. You're modeling an NFL team as 16 flips of a weighted coin. This obviously can't represent the NFL as a whole.

is that we have not adjusted for strength of schedule

...which is essentially what I'm saying.

69
by Mansteel (not verified) :: Fri, 12/11/2009 - 11:26am

One issue regarding the probabilities calculated for December win pct: Though it seems as if you want to calculate the probability of e.g. the Cowboys Dec record 2006-2009 being 7-12, what we actually find surprising is that they have any such stretch of games in which they perform so poorly. E.g, if their September (rather than December) record was 7-12 we'd be asking the same type of questions (but not at this time of year, of course). Thus the probability we need to calculate is the Cowboys going 7-12 or thereabouts IN ANY GIVEN MONTH rather than in December specifically.

To (hopefully) clarify things by analogy: In a hand of Texas Hold'em, the flop comes up JJJ. Someone says, "Wow, what are the chances of that?" To answer, you'd want to find the prob of getting any three of a kind because that is what they are surprised about--they're not surprised at getting Jacks specifically.

By the way, this is why you can get "duelling probability experts" in court cases: there is often confusion/disagreement on what probability should actually be calculated.

84
by mm (not verified) :: Fri, 12/11/2009 - 5:01pm

Well, we can quickly do a really simple version of this question.

If you approximate the NFL as 4 months long, which each month having 4 games (yeah, sometimes December has 5 games), then December has a 1/4 chance of being the worst month.

The chance of December being the worst month 4 years in a row is (1/4)^4=1/256, so in a 32 team league that wouldn't come up every year by random chance, but it wouldn't be a shock to see it happen.

The chance of the any one month (whether it be October, November, etc.) being the worst month 4 years in a row) is (1/4)^3=1/64, which is pretty frequent for a 32 team league.

You can argue that December hasn't just been run-of-the-mill bad for the Cowboys, but unusually bad, which would make it even more rare than this model suggests.

85
by sswoods (not verified) :: Fri, 12/11/2009 - 5:23pm

mm, your argument has merit only if the schedule is equal in all instances; of course, that is not the case. As has been mentioned already, the 'Boys schedule has been quite difficult in December the last few years, especially the last two. Their results would likely been far different if they played even average teams as opposed to a four game stretch of 12, 12, 11, and 9 win teams (in the case of '08). A point also made earlier in the comments.
Has Dallas played poorly in December compared to other months recently? I think they have, but not by a great degree - their problem seems to be an inability to beat Philly and Washington in Dec. more than anything else.

96
by mm (not verified) :: Fri, 12/11/2009 - 10:16pm

If your team plays December the same as the other months (ie fatigue from the season and players freaking out over the playoff picture aren't affecting your team more than its competitors), then there is a 1/4 chance it will be your worst month. If one year your 'worst' month happens because of scheduling, there's a 1/4 chance the worst month for scheduling will be December. If your one year your 'worst' month happens because of a particular run of short-term injuries, there's a 1/4 chance that the primary month those happen in is the last one.

The model works for any cause that can happen each month. It's only a quick calculation of how often random chance would produce 4 years in a row where December is the worst. If it isn't a particularly low number, then its more likely the cause of the Cowboys plight isn't 'choking', and is something random, like scheduling (with this model, it could be something different every year).

Of course, for some teams, their 'worst' month is 3-1...for the Patriots 2 years ago, their worst month was 4-0. The actual odds of having 4 seasons like the Cowboys would have to be lower than this model

98
by tuluse :: Sat, 12/12/2009 - 3:27am

Except a month isn't 1 event it's 4, and the problem isn't that it's the worst month, it's the degree to which it's bad. So your model doesn't work at all.

106
by mm (not verified) :: Sat, 12/12/2009 - 9:21pm

Except a month isn't 1 event it's 4

Actually, its 4 games, or dozens of drives, or hundreds of plays. We're comparing a team's results over one month compared to other months.

he problem isn't that it's the worst month, it's the degree to which it's bad.

Looking at DVOA, perhaps, but we'd have to look at how DVOA is distributed for most teams to say that. Looking at the record only (which is what the national media does), I don't think so. If you added the worst months of an above average team for 4 straight years and came up with 7-11, would that be strange? I don't think so. The odd thing to me is that the worst month has been December 4 years in a row.

107
by mm (not verified) :: Sat, 12/12/2009 - 9:48pm

OK, as a quick comparison, I decided to look at the NY Giants, another above average team over the last 4 years. Checking wiki for their records the last 4 seasons, I found their record for each month the last 4 years.

2005- SEP- (2-1) OCT- (3-1) NOV- (2-2)* DEC-(4-1) TOTAL-(11-5)
2006- SEP- (1-2) OCT- (4-0) NOV- (1-3)* DEC-(2-3) TOTAL-(8-8)
2007- SEP- (2-2) OCT- (4-0) NOV- (1-2)* DEC-(3-2) TOTAL-(10-6)
2008- SEP- (3-0) OCT- (3-1) NOV- (5-0) DEC-(1-3)* TOTAL-(12-4)

* worst month of the year

So their overall record for 4 years is 41-23, barely better than the Cowboys 40-24. While the Cowboys went 7-11 (38.9%)over December, the Giants went 5-10 (33.3%) over their "worst months". The only thing that sticks out is that the Giants didn't have their "worst month" as the same calender month each year (For a moment I thought it might happen, until I looked over the 2008 record).

104
by sswoods (not verified) :: Sat, 12/12/2009 - 3:27pm

Let's look at it this way. A given team plays a balanced schedule each year, something like this:

4 elite teams
4 above average teams
4 below average teams
4 poor teams

Each month, this given team plays one team from each group. If said team has worse results consistently in the last month, then the model has merit. But, if the said team plays, in the last month, 3 elite teams and 1 above average team, the model doesn't hold true, because it doesn't take into account the difficulty of schedule. Any given team, for the most part, will have better results against weaker opponents than stronger ones; if most of the weaker opponents are played early in the schedule but the stronger ones are played later, one would expect a decline in production at the end of the schedule. This is exactly what we find with the Cowboys, especially the last two seasons, and see happening again this season. The 'Boys aren't the only ones - the Giants looked like worldbeaters during the easy portion of their schedule, but have struggled once the competition improved. An above average team playing Dallas' schedule would be expected to have better results through the first three months but then fade in December - we see this all the time in College Football, shouldn't we also recognize this aspect in the pros? The point is random chance isn't the only factor here, that schedule plays an incredible part.

105
by mm (not verified) :: Sat, 12/12/2009 - 9:18pm

But, if the said team plays, in the last month, 3 elite teams and 1 above average team, the model doesn't hold true, because it doesn't take into account the difficulty of schedule

I'd like you to re-read what I typed. Tell me precisely where I said scheduling was not the 'reason'.

Scheduling is a random factor here. The whole point is too quickly look at how likely or unlikely something random like scheduling could be the cause.

Your NFL season is essentially September, October, November, December.

What are the odds December will be the most difficult in terms of scheduling? 1/4. So what are the odds December will be the most difficult in terms of scheduling 4 years in a row? (1/4)^4. So the model works fine for that 'cause'.

However, the model can work if any other 'random' cause produces the 'worst' month. What if one year your worst month happens because of a handful of injuries + a bad officiating call? What are the odds that that month will be December? 1/4.

What if one year your injuries are spread out through the year, and the schedule is pretty much spread out? You'll still have one 'worst' month, even if the record is 2-2 or even 3-1. If you're team isn't freaking out over December, there's still a 1/4 chance December will be that worst month, just by chance.

The model isn't about which cause, just as long as it has an equal chance of occurring each month.

Given that the Cowboys are a team that went 40-24, I didn't think it odd that their worst month from each year added together produced a 7-11 record (ignoring this year). After all, more losses will occur in your worst month than in any other month. The odd thing was that the worst month was December every year. So I examined the odds of that happening to a team from chance alone.

109
by Pat (filler) (not verified) :: Mon, 12/14/2009 - 9:43pm

Scheduling is a random factor here.

Yeah, I really don't think that's true. I think the NFL quite regularly puts certain games on the schedule at certain points (Christmas Philly/Dallas, for instance), and since teams aren't entirely randomly good/bad, there's a decent chance there'll be a bias there.

112
by Pat (filler) (not verified) :: Tue, 12/15/2009 - 3:58pm

Also worth noting is that the math isn't correct, since you've got ties - that is, you're asking "is December less than or equal to (November and October and September)?"

Which means that the chance of a month being the "worst" month is higher than 1/4, and dependent upon winning percentage. Take two months, A and B, each with 4 games, all results (4-0, 3-1, 2-2, 1-3, 0-4) likely (so 25 total possibilities). The number of times that month A has fewer or equal wins to month B is 15/25, not the 50% as you would expect. This gets higher as the team gets better or worse (since the distributions get tighter) - that's easy to see since it has to approach 1 as the winning percentage goes to 100% or 0%.

70
by 4tuna (not verified) :: Fri, 12/11/2009 - 1:14pm

Family Huddle is most noteworthy for an appearance by Cooper Manning, the Chuckie Cunningham of the family.

A little help please. Who the hell is Chuckie Cunningham?

73
by Kevin from Philly :: Fri, 12/11/2009 - 1:34pm

Chuckie Cunningham was the older brother of Richie Cunninham on the show "Happy Days". He disappeared after the first season, and was never seen again until the final season, when he was written back in to replace Ron Howard (who had just left the show).

88
by Spielman :: Fri, 12/11/2009 - 7:02pm

No. Ron Howard was out of the show for several seasons, not one, and Chuck never came back. Ted McGinley was kind of the replacement for Howard, but his character was named Roger Phillips.

91
by Kevin from Philly :: Fri, 12/11/2009 - 7:15pm

My bad. I guess I stopped paying attention the last few years.

74
by nat :: Fri, 12/11/2009 - 1:35pm

A character from the TV series Happy Days who was written out of the show in its second season, never to be heard from again. He was the main character's older brother.

Who they hell is Cooper Manning?

90
by Kevin from Philly :: Fri, 12/11/2009 - 7:12pm

That's what Archie would like to know.

75
by OMAR :: Fri, 12/11/2009 - 1:45pm

+1 for "abstinence vampires"

Nothing was ever so disappointing as a gift book aimed at one's age group.

86
by Andrew B :: Fri, 12/11/2009 - 5:31pm

Cowboys:

2005: 2-3
2006: 2-4 (including a 3 point win at the end over the Giants)
2007: 2-3 (and one win was by one point over the Lions!)
2008: 1-3
2009: 0-1

That would be 7-14. Even worse than the record you give them.

The Original Andrew