pageok
pageok
pageok
No meaningless games for NFL playoff teams.--

A few years ago, I noticed that one of the prevalent beliefs in the NFL is a myth: it is widely thought that it doesn't matter whether an NFL team bound for the playoffs wins its last regular-season game. Accordingly, many playoff teams rest some of their starters for their final regular season game, as the Chicago Bears did last year in their season-ending loss to Green Bay, a precursor of their loss in the 2007 Super Bowl. Like NFL coaches, many commentators also do not seem to understand that losing the last regular-season game is an excellent indicator of not making or not winning the Super Bowl.

Since the 1995-96 season, only one of the 12 Super Bowl winners (the 1999-2000 St. Louis Rams) lost their last regular-season game. Further, one would have to go back to the 1980s before finding a Super Bowl winner who lost its last regular season game by more than 7 points.

Why is this so? Of course, some strong teams need to win their last game to maximize home-field advantage in the playoffs, but other strong teams may simply be good enough to win even when they are not trying their hardest.

Whatever the reasons, winning your last regular-season game is an important indicator of your ability to win the ultimate prize: the Super Bowl.

Note that this analysis does not necessarily mean that resting your starters in the last game is the wrong strategy: perhaps you had almost no chance to win the Super Bowl anyway so resting them made no difference, or perhaps you are good enough to win the last game without your starters, in which case resting them might be the right move. But whatever the pre-game probabilities might be, after the game your chances of winning the Super Bowl are slim if you lost the final regular-season game (and especially if you lost by more than 7 points).

Daniel Chapman (mail):
Possibly because everyone rests starters, and the team that wins the superbowl is the one that has the depth to win their last regular season game with backups?

(Go Pack!)
12.30.2007 2:39pm
hattio1:
Uhhh,
To be statistically relevent, wouldn't you have to break those game ending numbers down by those that rested starters and those that didn't? And doesn't the limitation of looking at SuperBowl teams unnecessarily skew your data? Let's face it, if resting your starters would somehow guarantee a first round win, every team who was guaranteed to make the playoffs would rest...especially if they're San Diego.
12.30.2007 3:00pm
Siona Sthrunch (mail):
You say 1 out of 12 superbowl winners lost their last regular season game.

What fraction of superbowl winners lost, say, their antepenultimate regular season game?
12.30.2007 3:11pm
donaldk2 (mail):
I don't have the statistical resources to look it up, but how many Superbowl losers, other than the Bears, lost their final regular season game?

I will be glad to be informed otherwise, but my guess is that almost all of them won their finale.

To keep the statistically minded busy, why not compute Superbowl results as a function of winning their opening game, second game,...next-to-last game?

(Think reductio ad absurdum.)
12.30.2007 3:14pm
Ilya Somin:
I think this proves a correlation rather than causation. The better a team is the more likely it is to win the Super Bowl, and the more likely it also is to be able to win its last game even with the starters resting. Also, as you point out in the post, some very strong teams still need to win the last game in order to lock in a better playoff spot and therefore play their starters.
12.30.2007 3:25pm
Bruce:
What Ilya said.
12.30.2007 3:31pm
Hei Lun Chan (mail) (www):
Since the 1995-96 season, only one of the 12 Super Bowl winners (the 1999-2000 St. Louis Rams) lost their last regular-season game.

And only two of those 12 teams played in meaningless games (the 1998 Broncos being the other). The other teams were all playing for better playoff seeding.
12.30.2007 3:56pm
jrose:
Note that this analysis does not necessarily mean that resting your starters in the last game is the wrong strategy

That's not a strong enough statement. Something along the lines of "nothing in this analysis suggests a team should play its starters in a last game that has no bearing on playoff position" is more like it.
12.30.2007 3:57pm
Da Kid:
This just reinforces my belief that the Colts should play Manning, Addai, and Harrison the entire game tonight.

Not for the Colts' own benefit, of course, but to help ensure that the Browns make the playoffs.

-Da Kid, President, Jim Sorgi Fan Club.
12.30.2007 4:19pm
Steve2:
So what this means is... if Washington wins their way into the playoffs, we're looking at Green Bay over the AFC in the Super Bowl? I like it!
12.30.2007 5:45pm
Steven Lubet (mail):
Every Super Bowl team -- both winners and losers -- has necessarily won at least two, and sometimes three, intermediate playoff games. It is hard to imagine that the final regular season game would have a delayed impact three games later. Thus, a more interesting question is whether teams that have rested starters, and therefore lost their last regular season games, tend to lose their first playoff games.
12.30.2007 6:58pm
DNL (mail):
Prof. L:

Can't the same be said for World Series winning teams? I'm pretty sure that very few of them -- in fact, none of them! -- lose week 17 of the NFL season. Same for NBA teams and NHL teams, and even snooker champions.
12.30.2007 7:12pm
alias:
What Ilya said ... obviously. wtf?
12.30.2007 7:32pm
DNL (mail):
On a more serious note, the main reason why your (incredibly small) sample size gives you those results is simple:

A team that won in week 17 is more likely to have a favorable stance in the playoffs than a team that lost in week 17, with a few exceptions:
* When a dominating team in a weak conference locked up home-field advantage already
* When a weak team wins an even weaker conference
* When a strong team runs away with a wildcard spot

(There are outliers, as evidenced by 2007, but that's the bulk of the data.)

Basically, in most other situations, a playoff-bound team has incentive to win beyond the junk science espoused in this post. Either it's that they need to win to get into the playoffs, or they are playing for a bye, or a higher seed and/or home-field advantage. In each of these cases, a win correlates to a better playoff posture, which arguably causes more Super Bowl wins.

The situations I delineated, of course, are when a team will be most able to sit their starters, and in most cases, those teams are weaker than the competition. The 2006 Bears were the best team in a terrible conference. This year, the 2007 Chargers basically won the AFC West by default. Same goes for the '07 Bucs and the NFC South. Finally, the Jacksonville Jaguars and New York Giants are able to (although the latter did not) rest their starters, and it's pretty easy to show that neither are in the top 2 in their respective conferences.

While I think one can make the case that a #1 or even #2 seed resting their starters going into a bye leads to more upsets than are warranted in the second round of the playoffs, I haven't yet seen it tested; and it really has nothing to do with the thesis posited by this post.
12.30.2007 7:37pm
byomtov (mail):
This is junk. Lots of good criticisms have been made. Here's more.

Obviously a Super Bowl winner is likely to win its last game ( as well as its third game, or eighth game, or whatever) because it's one of the strongest teams. With only twelve data points it's hard to say that game 16 is more predictive than others. Even if it is, there are a couple of other possible reasons:

An opponent out of contention at the end of the season may not be giving its best, so the chance of being upset is lower than earlier in the season.

A playoff team may have built a good record early and faltered late due to injuries or for some other reason. Such a team is more likely than a healthy one to lose both its last game and a playoff game.

jrose wrote:

"nothing in this analysis suggests a team should play its starters in a last game that has no bearing on playoff position" is more like it.

True.
12.30.2007 8:05pm
byomtov (mail):
One more thing. Suppose that Super Bowl winners turn out to have a better record in game 16 than in any other regular season game. This really wouldn't tell us anything anyway. Among games 1-16 there has to be some game in which the SB winners had the best record. Why would the fact that the magic number is 16 tell us anything?
12.30.2007 8:50pm
yahonza (mail):
And why only measure since the 1995-96 season? That seems like an arbitrary choice. Nothing in the rules of football changed that year, so why ignore the previous years? That seems like choosing data that supports the theory, while ignoring data that doesn't.

I have to agree with the comment above: the impact of loosing game 16 can't project itself three games later to the superbowl. Kind of a silly idea, if you think about it.
12.30.2007 9:53pm
James Lindgren (mail):
I am often surprised by the comments here.

1. There seems to be considerable confusion here about correlation and causation. I couldn't have been clearer that winning the last game is an INDICATOR (not a CAUSE).

I wrote:

"losing the last regular-season game is an excellent indicator of not making or not winning the Super Bowl"

and

"Whatever the reasons, winning your last regular-season game is an important indicator of your ability to win the ultimate prize: the Super Bowl."


To illustrate, your score on an analogies test or on a reading comprehension test is an INDICATOR of how you might perform in school, not a CAUSE of doing better in school. Getting a higher grade on an analogies test does not necessarily cause you to do better in school, though I wouldn't rule out the possibility of some causal influence.

Similarly, I did not argue that winning the last game CAUSES teams to win the Super Bowl, but raher that it is a good indicator of whether you have a slim chance of winning the Super Bowl.

It is doubly unfortunate that several commentators actually endorsed the "correlation is not causation" criticism—and not one of you pointed out that I explicitly presented it as an indicator that works for whatever reasons.

2. Others of you treated my analysis as if I were addressing which of 2 teams who make the Super Bowl would win the game. Again, I quite explicitly posted that "losing the last regular-season game is an excellent indicator of NOT MAKING or not winning the Super Bowl." Thus, the main effect might be in the first or second playoff games, not the Super Bowl itself (I haven't collected or broken down the data that way because my thesis is not specific to one game). The idea that the indicator would take 3 playoff games to manifest itself is a simple misreading of my argument and the evidence I presented.

3. There are 12 playoff teams each year, 4 with first round byes and 8 who play the first weekend. As I count, in the last 12 years, of the 42 playoff teams who lost their last game, only one won the Super Bowl. Of the 102 playoff teams who won their last game, 11 won the Super Bowl. Thus, the odds of winning the Super Bowl are 395% higher for teams who won their last game.

Similarly, among the 48 bye teams, none of the 15 teams who lost their last game won the Super Bowl, while 6 of the 33 bye teams who won their last game (18%) won the Super Bowl.

4. BTW, thanks go out to Hei Lun Chan, who actually adds new relevant information (assuming the facts he presents are true).

Jim Lindgren
12.30.2007 10:44pm
byomtov (mail):
I couldn't have been clearer that winning the last game is an INDICATOR (not a CAUSE).

Jim,

It's not even an indicator. Read my second comment, for one reason. There are potentially others.
12.30.2007 11:23pm
Bruce:
Jim, you sure did use the word "indicator," but I think a natural reading of your post (whether you meant it or not) is as advice to teams *before* playing their 16th game, rather than as advice to commentators on how to interpret the results of a 16th game. Since the team might actually have some control over those results, then if winning the 16th game is an indicator and the team knows this *in advance* things start to get a little weird -- just as they did for Calvinist Predestination theorists. If those who are predestined to go to heaven act a certain way, and you force yourself to act in that way, can you feel comfortable you are one of the Elect or not?
12.30.2007 11:40pm
Ilya Somin:
There seems to be considerable confusion here about correlation and causation. I couldn't have been clearer that winning the last game is an INDICATOR (not a CAUSE).

I apologize to Jim for misinterpreting his post. However, I think that this language in the post was susceptible to that kind of minsinterpretation:


it is widely thought that it doesn't matter whether an NFL team bound for the playoffs wins its last regular-season game. Accordingly, many playoff teams rest some of their starters for their final regular season game, as the Chicago Bears did last year in their season-ending loss to Green Bay, a precursor of their loss in the 2007 Super Bowl. Like NFL coaches, many commentators also do not seem to understand that losing the last regular-season game is an excellent indicator of not making or not winning the Super Bowl.


To me, this seems to be a criticism of at least some NFL coaches for resting their starters in the last regular season game and for underestimating the importance of winning it. Certainly, it seemed to me that you were criticizing the Bears for resting their players and suggesting that that had something to do with their loss in the Super Bowl. By contrast, if it really is just an indicator rather than a cause, then there is no grounds for criticizing teams such as the Bears who rest their players in the last game when they have nothing to play for.
12.30.2007 11:48pm
James Lindgren (mail):
Ilya,

Thanks.

I also specifically wrote that "this analysis does not necessarily mean that resting your starters in the last game is the wrong strategy."

Further, when I asked why this indicator works, I wrote that some strong teams need to win (and are strong enough to do so) and other teams are strong enough to win even when they don't need to. Both explanations have to do with underlying team strength.

I'm sure that neiher of us would rule out the possibility of a causal effect, but since I wouldn't want to confuse correlation with causation, I quite properly limited my stated conclusions to the existence of the indicator.

BTW, this year may be the toughest test yet for this last-game indicator: this weekend 7 of the 12 playoff teams lost their last game, more than any year I've looked at so far.

Jim Lindgren
12.31.2007 12:12am
Ilya Somin:
I also specifically wrote that "this analysis does not necessarily mean that resting your starters in the last game is the wrong strategy."

Yes, I saw that. But I interpreted it (apparently wrongly) as suggesting that in some cases it is indeed the wrong strategy, even if that isn't "necessarily" true in all cases.

BTW, this year may be the toughest test yet for this last-game indicator: this weekend 7 of the 12 playoff teams lost their last game, more than any year I've looked at so far.

Yes, I agree. However, I hope and expect that the Patriots will win, thereby providing additional support for your theory.
12.31.2007 1:17am
alias:
sorry for the rudeness. I hit post and then thought better of it, but wasn't quick enough with the mouse to close the window. Lawprofs doing the armchair sports analyst thing apparently strike a nerve with me.
12.31.2007 1:33am
seconding Da Kid's sentiments:
I wish someone had sent this to Tony Dungy. Call it "resting for the playoffs" if you want, but the Colts basically threw the game today. Poor Browns.
12.31.2007 2:32am
Paul McMahon (mail):
I've long doubted that particular fantasy of the talking sports heads. One reason is that with the large scale use of substitutions, platooning, there is really a smaller difference between the "starters" and their relief.

The exceptions would be in particular positions, I suppose, like the quarterback. But there we have numerous rules to protect those players, so the risk is minimal.

Now, if the "rest" hypothesis applied to the coaching staff ....
12.31.2007 7:49am
jrose:
I quite properly limited my stated conclusions to the existence of the indicator.

Your critcism of the Bears coaching staff is not consistent with this claim.
12.31.2007 8:56am
byomtov (mail):
The way to show that this is in fact an indicator, and not just a random occurrence, is to look at the seasons before 1996 and see if the pattern holds. By "pattern" here I mean something like "SB winners have a significantly better record in their last regular season game than in any other regular season game."

Until then, you got nuthin'.
12.31.2007 10:26am
Andrew Okun:
Even though your last season game result isn't used to determine post-season results, there are lots ways choice of play in that last game could affect post-season play.

1. Information. Some of you have said that the last game result is a good indicator. Best to give no indication at all then, to the degree possible. Play weird to make the result misleading. Best of all, forfeit the game. Won't affect your post-season standings and your team is shrouded.

2. Player condition. Obvious, and clearly the factor governing most teams play.

3. Team psychology. Play to win if your team needs to come into the post-season pumped and active, and other teams need to keep fearing you. You risk exhaustion and loss, but may gain mental high ground. Alternatively, play badly and lull the others into overconfidence.

4. Other teams' conditions. Play hard and injure some of the other teams' players and confidence. I don't know the details... do playoff teams play each other in the last game? If so, and they're tired and you're well rested, perhaps attrition is an advantageous move.

Lots of fun to be had with supposedly unimportant game.
12.31.2007 1:20pm
byomtov (mail):
Lots of fun to be had with supposedly unimportant game.

True. They could also be used to work on unusual plays. Shouldn't teams take the opportunity to practice onside kicks, for example?
12.31.2007 2:37pm
Da Kid:
I hate you, Jim Sorgi.
12.31.2007 4:25pm
DeezRightWingNutz:
Jim,

If you hope to avoid confusion of correlation and causation in the future, I'd suggest using the term "correlation," "correlates, or "correlated" rather than indicates. I understand what your wrote, but when I first read it, my first thought was, "correlation, not necessarily causation." I think using some form of the word "correlate" would've led me to interpret your remarks as intended (probably because I learned to distinguish correlation from causation, not indication from causation).

Also, I hate USC.
1.1.2008 5:48pm