So, did everyone do their reading for the week? Really? I see some nodding, but I see a lot of blank stares from the back of the room. You didn't even try, did you, Twist? You've been faking your way through this class from Day 1. And I don't want to hear about how your wife made you spend time with her. You've gotta make time!
Well, that's OK. We'll start off with a short discussion, giving you guys time to catch up before we have our pop quiz at the end of the class. Pop quiz!? Don't worry, if you've done the reading, it'll be a snap.
So what were we talking about again? Ah, yes, an economics-style analysis of AP voter behavior, using empircal tests to challenge conventional wisdom about college football (PDF). Confused? See last week's post for a synopsis. Now, we'll break it down (with this week's Top 25 interspersed throughout):
The AP poll is a Borda count, whereby each team receives points based on where they were ranked on each ballot (25 for 1st, 24 for 2nd, etc.), and then ranked according to total number of points. In the paper, for each piece of conventional wisdom being tested, the effect of the assumption under discussion is tested by measuring the change in each team's point total following a qualifying win or loss. These results are then averaged to see if, say, winning by a large margin has a statistically significant effect on a team's point total.
No movement in the top of our poll, although I did consider moving Penn State up after beating Ohio State on the road, and I thought Oklahoma State proved it belonged in the top 10 even in defeat in Austin. Some wins were better than others, but I didn't see anything this week that convinced me my opinions of these teams last week were wrong.
Does beating strong opponents help?
Sorta, but not really. Turns out, "while defeating a strong opponent does not help, losing to a strong opponent actually softens the blow of a loss. For example, losing to a team with an 8-3 record would actually decrease the negative point change from losing by more than 15% of the change for a loss." If you think about it, this makes sense. Since most of the Top 25 wins each week, teams often simply move up in order after winning, leapfrogging teams that lost. Often, it takes a loss (or at least a suspect win) by a highly ranked team to get voters to consider moving a lower-ranked team above them, whether they've been winning impressively or not.
Winners move up, losers move down. That's the basic idea here. Missouri gets a bump for absolutely decimating Colorado (58-0!), whereas Ohio State and LSU are treated relatively kindly for losing to tough opponents in Penn State and Georgia, respectively. Personally, I think we're being too kind to LSU, but I'm not sure who, exactly, I would rank above them. TCU drops a spot despite laying a 54-7 beating on Wyoming, who may be one of the 5 worst teams in Division I-A, entirely due to Missouri's leapfrogging of them. I'm going to blame this on our rotation of voters, as this week HydroTech fills in for the absent Yellow Fever.
Does the margin of victory matter?
Again, not really. In fact, it's the same effect as with the strength of the opponent. Blowout wins do not seem to matter at all. Rather, it's the margin of victory in a loss that makes a difference: "close losses actually help, and blowout losses hurt the most." Again, this seems to play to the mindless way in which many voters act: everyone moves up when they win, and only when they lose do the voters stop to consider the severity of the loss, and decide how many places to punish the loser.
I glibly call these voters 'mindless', but if you really sit down and think about it, at this point in the season, should the margin of victory or strength of opponent cause teams that both won to leapfrog each other? If I thought that, going into the weekend, Team A was better than Team B, and Team A wins a squeaker while Team B wins going away, should that cause me to change my mind about what I knew beforehand, possibly disregarding the previous couple months of play? Early in the season, wild poll swings are far more justified because voters should disregard their preseason expectations; after a month or so, however, a voter should need a good reason to forget what they thought they knew and change their mind.
Florida State is this week's ACC leader, which means that they're likely to eat it this weekend at Georgia Tech. Everyone else posts solid wins and moves up to occupy the spots vacated by last week's losers. Minnesota is still hanging around, mostly due to a soft-even-for-the-Big-10 schedule, including Northern Illinois, Bowling Green, Montana State and Florida Atlantic, but missing both Penn State and Michigan State. A thirteen-point loss at Ohio State has thus far, and may remain, their only encounter with real competition this year.
Is it better to lose early in the season or later?
According to this paper's findings, it's actually better to suffer a late (10th week or later) loss. "Rather than significantly hurting teams, losing late in the season actually helps them—it lessens the blow of a loss significantly." To wit:
Given the point estimate in Table 3 and the number of AP poll voters, losing late in the season implies that more than 3/4 of AP poll voters rank a team one place higher in their rankings after a late season loss than for an early season loss.
Interesting, no? Once again, you can come up with a common sense, almost obvious explanation for this behavior. Lose in week 1 and you're 0-1; in week 2, you're 1-1. Lose in week 8? You might only drop to 7-1, in which case there are a lot of teams that already have two or more losses, and there just isn't that far to fall. If BYU loses badly to TCU in week 2, they almost certainly drop out of the poll, but since they lost in week 8, they only dropped 9 spots.
The "when to lose" dilemma doesn't really apply this far down in the poll; all of these teams have multiple losses, and all have looked bad at least once this season. I really had to stretch myself to find 25 teams to rank this week, and these were the stiffs we came up with. I'll bet a couple of these teams go down this weekend.
3 open-ended questions
Okay, quiz time. Should be easy. Leave your answers in the comments; partial answers get credit too.
1) What do you think of these results? Do they make intuitive sense, or do you need to see mathematical proofs before you buy into the conclusions?
2) If voters become aware of the results this paper, do they become self-conscious of their own tendencies and begin to act counter to them?
3) Should voters act in the manner described, or do they unwittingly harm teams undeservedly by their collective actions, discouraging behavior (such as scheduling tough teams) that should be encouraged?
Extra credit (need to have done the reading on this one):
4) Did you spot any issues with the paper presented? Problems with the methodology, the mathematics, or the conclusions drawn? Does something seem "not quite right"?
Dropped Out: Pittsburgh (#15), South Florida (#16), Georgia Tech (#19), Kansas (#20), Boston College (#21), Northwestern (#24)
A bunch of losers here. Kansas was able to withstand road losses at South Florida and Oklahoma, in part because the Jayhawks were competitive in both games, but giving up 9 touchdowns at home to Texas Tech is a pretty good was to get ejected from the Top 25. Not that Pitt (54-34 losers to Rutgers) or Boston College (45-24 losers to North Carolina) looked any better, and losing to Indiana (Northwestern) or Louisville (South Florida), however close, is not considered a "good" loss.
OK, as to the last question, I don't claim to be an economist, and I have only a passing familiarity with some of the mathematics involved. However, I did have one quibble with the paper's conclusions: I don't believe it when it tells me that it's better to lose early rather than later.
Oh, the math is sound enough. I am certain that teams lose fewer points when they lose later in the season, due both to a) the teams below them having more losses, and b) the losing team having a larger track record of wins to counterbalance their loss. However, that does not mean that it would have been better to have lost late. The paper argues against me:
Even if one wished to argue that an early loss gives teams more time to make up ground in the rankings, the results here suggest that late losses leave teams with less ground to make up.
No, I'm still not buying it. Less ground to make up, sure, but less time to make up that ground. Comparing week 2 to week 3 is not at all the same as comparing week 8 to week 9. What we really want (and what this paper does not provide), is a comparison between week 2 to week 9. We should compare teams of similar initial standing who both end up with (say) a 7-1 record after 8 weeks. How do their point totals differ based on when their one loss was? The paper doesn't say, and indeed, the methodology given can't answer this question. It's not unknowable, just unknown within the framework of this analysis. Actually, it's too bad I don't have access to this data set put together by this paper, because I'd be really interested to find out the answer to this question.