Chris Cillizza, the political blogger for The Washington Post, had an item the other day that asked an interesting question: Do debates really matter in shaping election outcomes? And he provided a mostly helpful answer, which is that lots of the time, they don’t. So kudos to Cillizza for stepping back to evaluate the actual significance of a political ritual that gets a lot of coverage because, as he notes, it’s useful to journalists—which is not the same thing as being important.

Still, it’s worth taking a moment for a closer look at Cillizza’s post, both to examine the debate question a little more closely, and also as an instance of how journalists sometimes embrace narratives even when they’re on guard against them.

Cillizza opens with the point that, while Democratic candidates and the Arkansas and Colorado Senate races are all furiously trying to claim victory in their recent head-to-head matchups, “it’s difficult to attach too much political significance to any one of” these debates. That’s largely because the audience for primary debates in statewide elections, which often have limited distribution, is tiny, even by the already-shrunken standards of the primary electorate.

Unfortunately, the research on the role of debates in primary and statewide elections is pretty scant, so we don’t have a strong empirical sense of whether this point holds up. But it seems right. Even if the debates are, in some sense, a better source of information about the candidates than advertisements or stump speeches, they won’t have much of an effect if people don’t see them.

We do have copious research on an area Cillizza doesn’t address in his post—the effect of debates in the general presidential election. And while there’s some disagreement in the findings, as expected, the strongest evidence is for a somewhat surprising conclusion: even though those debates do attract large audiences—for example, the first Obama-McCain debate drew over 52 million viewers, which amounts to more than 40 percent of the people who turned out on Election Day—they generally don’t have much of an effect on the outcome, either.

The issue is one of timing. In the image below, from the political scientist James Stimson’s book Tides of Consent, the black line shows the eventual winner’s share of the two-party vote in polls conducted during the presidential campaign; it’s a composite of five competitive races: 1976, 1980, 1988, 1992, and 2000. The diamonds show when debates were held:








(h/t Brendan Nyhan for the pointer to Stimson’s book)

As the graphic shows, there is a fair bit of voter decision-making during these campaigns—that’s the long upward trend in the spring and summer, peaking at around the time the conventions are usually held. But by the time of the debates, the line is essentially flat and the polls aren’t moving, because almost everyone who’s paying attention has been exposed to so much information that they’ve already made their choice. This isn’t to say that these debates are meaningless in terms of the result—if an election is close enough, as in 2000, everything is meaningful—but they’re not on the short list of things that matter, either.

So adding together what we think about primary debates and what we know about general election debates, we can make a guess about when debates might matter: the rare instances when a debate attracts a sizable audience of potential voters, but before those voters have made up their mind. Cillizza actually names a couple plausible contenders for this status in the second half of his post: the Scott Brown-Martha Coakley debate in Massachusetts, and the recent three-way debate in the UK, in which the Liberal Democrat Nick Clegg established himself as a serious challenger.

But those may be exceptions that prove the rule. The Massachusetts Senate race was a special election that had already begun to draw an unusual amount of attention. And the campaign was, by American standards, extremely abbreviated, so eight days before the election, when the debate was held, interest was high and some people were still choosing how, and whether, to vote. And in the UK case is also idiosyncratic: Clegg leads an established third party that is something of an electoral also-ran, and the debate—a novelty in British politics—gave him a platform alongside the two main contenders. That’s a unique constellation of circumstances.

So far, so good. But then, at the end of a post that began with a skeptical point of view, Cillizza gets a little credulous. Debates can matter when they “create windows of opportunity for underdog candidates to make up ground,” he writes, and singles out the October 2007 exchange in which Hillary Clinton stumbled over a question about whether illegal immigrants should get drivers licenses.

What seemed like an afterthought in a very contentious debate that centered on the Iraq war (among other weighty topics) wound up signaling the beginning of Clinton’s decline as the driver’s license equivocation came to symbolize everything voters disliked about her.

This might be true, but the evidence for it is pretty thin. The aggregate polling data for the 2008 Democratic primary shows that Clinton’s support was remarkably stable over a long period. This is really not surprising—she had been a national figure for a long time, and people had pretty settled opinions about her. It is the case that Obama’s support skyrocketed at about the time of that debate—but that’s probably a function of him consolidating support undecided voters who were already reluctant to back Clinton.

In other words, voters who didn’t back Clinton because of her “equivocation,” or for other factors, probably didn’t need any new symbols or signals. But journalists who had to update an old meme with some fresh news pegs did—and the debate stumble came in handy. As Stimson writes, after an election, we ask:

Why did the winner win? Why did the loser lose? It is tempting to use the debate to settle both questions, to find high moments that were good for the winner and lows that were bad for the loser.

But it’s at that point, he adds, that those much-maligned horse-race polls can be a corrective—because, they, “unlike the stories we tell after the fact, are not a social construction.”

Greg Marx is a CJR staff writer. Follow him on Twitter @gregamarx.