The Lion and Gun

In response to: Our Dangerous Narratives.

The point Ross makes is an important one. Narratives, particularly as applied to daily minutiae, often tell us more about the biases and interests of those imposing them than they do about reality. Breathless explanation of daily poll movements -- usually just noise -- and the ups and downs of the Dow throughout a day or a week has kept an awful lot of talking heads in business. (The most ridiculous example might have been Matt Drudge's recent attempt to construct a 'comeback' narrative because McCain pulled within two points of Obama in one of three voter turnout models in one of seven daily tracking polls for one day last week.) Indeed there is a danger in ignoring uncertainty and the unpredictable, just as there is a danger in assuming that a purported cause will produce the same effect in vastly different circumstances. After years of study I can assure you: history does not, as a rule, repeat itself.

Nevertheless, the argument can be taken too far.

Actions which are almost completely unpredictable individually often become quite regular in aggregate; likewise, changes which are essentially random from minute-to-minute can form real and really explicable patterns over a broader time horizon. Thus very complex sociological or economic or psychological processes can manifest themselves in relatively simple and discrete relationships. The result is that there are some explanations of cause-and-effect that are logical, often quantifiable, sometimes repeatable, relatively predictable, and occasionally persistent through time. Demographic voting patterns, for example, are very predictable; they change from one election to another, yes, but typically by minute (and typically foreseeable) amounts.

Let's return, for the moment, to polling. If Obama rises in an average of the daily trackers by two or three points over the course of the next couple of days I'm sure there will be a lot of speculation (or, worse, declaration) that it was due to the Powell endorsement. We can't know, however, if such an increase in Obama's polling would represent an actual change in his level of support or just coincident noise; and if it turns out to be the former it still remains problematical to ascribe it to any single cause (such as Powell's endorsement).

On the other hand, there are relationships of cause-and-effect in polling that are predictable and repeatable. Perhaps the most obvious and persistent is the convention bounce: the degree to which polling shifts in the wake of a given party's national convention is predictable to a fair degree of precision; thus it is entirely reasonable to ascribe Obama's brief bounce at the end of August, and McCain's longer bounce at the beginning of September, to such. The same could be said for the choice of running-mates, as well as the tendency of polling to tighten as the election draws near.

There are also relationships which are by their nature less easily or precisely predictable but no less explicable. Typically if a crisis occurs during an election the cause of which is (rightly or wrongly) ascribed to the incumbent party there will be a statistically significant shift to the challenger; the same applies when there is a general feeling of malaise (as measured by right-track/wrong-track polls). An exogenous crisis or a feeling of optimism, on the other hand, will favor the incumbent party. Economic problems boost support for Democrats; foreign policy problems boost support for Republicans.

These relationships, of course, change gradually over time, and the degree to which they manifest themselves is, unlike convention bounces, not predictable (nor are, in many cases, the predicating circumstances). None of that argues against their existence, which has been established multiple times over multiple elections. Thus it is reasonable to say that a terror attack against the United States would, all else being equal, likely boost John McCain's numbers to an unpredictable extent. And while it would stretch credulity to ascribe a short-lived three-point Obama bump to the Powell endorsement, it is reasonable to conclude that much of his sustained rise from mid-September through mid-October was the result of the financial crisis.

Finally, there are factors the veracity of which is more difficult to establish but which are still useful to consider in explaining relationships and predicting results. Negative advertising, successful messaging, and the character of a candidate all probably tend to affect election results, although it is mostly impossible to quantify or predict their effects (and, I suspect, their supposed effects are generally overblown).

For example, Obama's polling slowly but steadily declined after his overseas trip this summer. There are four likely explanations for this erosion in his numbers: 1) the tendency, ceteris paribus, for Presidential races to tighten over time; 2) the downside of a short-lived bounce from Obama's trip (and potentially a larger and longer-lived bounce from securing the nomination a month earlier); 3) the effectiveness of McCain's negative advertising, which began about that time; 4) random noise or unknown quantities. The cause could have been any combination of those factors, or any one factor alone. In real life there are always too many variables to understand, and, yes, too many unknowns. Nevertheless, considering the existence of bounces in other forms, the predictable tendency of races to tighten, and the roughly quantifiable effectiveness of attack ads in general (as opposed to any one given attack ad), any one or all of those first three 'narratives' provides a reasonable, if uncertain, explanation.

The problem, then, isn't narratives but their over-use, over-reliance on them, and the tendency to ignore valid narratives that cut against eventual results. The last, especially, is the problem with explanations of John Kerry's loss in 2004.

It's true that the final result in Ohio (and nationwide) was roughly within the margin of error. That is not, however, to say that the result was in any sense 'random', or that explanations imposed upon it are meaningless, for two reasons:

First, polls leading up to election day for weeks predicted almost precisely the outcome that was seen. Any one of those polls showed a result well within the margin of error. Taken in aggregate, however, it was extremely unlikely that so many polls would so persistently show Kerry down by, say, two points simply by coincidence. He really was down by two points. The decisions of those few thousand Ohio voters that made the difference between victory and defeat were predictable.

Second, and more important, the question that's being answered isn't, 'Why did Kerry lose a close race?' but, instead, 'Why didn't Kerry win by a substantial margin?' When talking about the things that Kerry's campaign did wrong, and that Bush's campaign did right, pundits and political scientists aren't imposing a narrative on a result within the margin of error because there's no reason why the result had to be within the margin of error. If Kerry had run a better campaign it's conceivable the election would have been a blowout, and it is reasonable to ask why that didn't happen. The danger is to forget the other side of the story: why didn't Bush win by ten points? The fact that the race was so close tells us that both campaigns had things going for them and things going against them. Thus, as I've said, it's true that pundits tend to see the Kerry loss in hindsight as having been the inevitable result of what he did wrong without considering all of the things he did right, or that Bush did wrong. In this sense the narrative that has been applied is false, but it is false by sin of omission: incomplete, not incorrect.

The lesson, then, is that day-to-day tea leaf reading is often fruitless and misleading; reliance on narratives can be dangerous because it discounts the random and unknowable and because it implies that results are repeatable when often they are not; and narratives often fail to tell the whole story, or emphasize one part of the story to the expense of another. Nevertheless, they are important not simply as simplifications but because they often provide cogent and quantifiable explanations that really do reflect reality, and in some situations they are repeatable and predictable. Narratives often help us to understand the world as it is and thus make better decisions. We just have to remember that they don't tell the whole story. And that we're not half as clever as we think.