There were some good comments on my earlier negative ads zombie post, and Phil Arena has also posted a reply. Let me make a few more observations.
At least one commenter wonders how I can claim that there is any zombie at work here. After all, I note that the literature on negative advertising certainly doesn’t conclude that it never ever ever works. So why shouldn’t people go on believing that it works? The problem — the zombie — is that the belief that it works, as expressed in many different news articles and commentaries, pays little attention to context or nuance — that is, to why negative advertising might matter more in some races than others. I have no problem with context or nuance. Categorical statements are the zombie here.
Another commenter brings up the current GOP primary. Didn’t the ads matter there? What about Florida? I’m actually on record believing that the ads in Florida might have mattered. It’s consistent with what I wrote earlier: in general advertising will matter more when (a) one candidate can vastly out-spend opponents (as Romney did), and (b) when opinions about the candidates aren’t crystallized (as is true in the GOP primary). If we could get reporting about negative ads to start thinking about (a) and (b) and other such factors, the zombie would be dead.
But here’s another thing: say (a) and (b) hold. Then the question becomes: would negative ads be categorically more effective than positive ads? Do we conclusive evidence of that? Not by any stretch. So just because advertising might be more effective in certain circumstances doesn’t mean that negative ads will also be more powerful than positive ads in those circumstances. There is a lot of folk wisdom about whether and why and how to use negativity, little of which has been subjected to any kind of rigorous test.
Arena wonders about limits in the evidence about negative advertising. He seems to think these limits may be so great that no journalist who believes in the power of negative ads needs to revise his or her opinion in light of political science research. If I understand him correctly, the limit that arises is this: candidates who are leading have little incentive to “go negative” and candidates who are behind do have this incentive. So if negative advertising is being driven by how well candidates are doing in the race, how can we show that it actually affects how well they do in the race? What’s the cause and what’s the effect? How can we disentangle them?
Of course, there are also many experimental studies of negative ads that less vulnerable to this cause-and-effect problem. They don’t conclusively support the conventional wisdom that negative advertising “works,” although one can always ask whether their findings would generalize to the “real world.”
Perhaps more relevant to Arena’s point is this paper by Matt Blackwell, which I linked to in the zombie post. Blackwell writes:
bq. Candidates in these races change their tone over the course of the campaign, reacting to their current environment. A single-shot causal inference method compares campaigns that are similar on a host of pre-election variables in order to eliminate omitted variable bias. While this is often the best approach with single-shot data, such an approach ignores the fundamentally dynamic nature of campaigns: races that become close over the course of the campaign are more likely to go negative than those that are safe. Attempting to correct for this dynamic selection by controlling for polls leads to post-treatment bias since earlier campaign tone influences polling. The inappropriate application of single-shot causal inference therefore leaves scholars between a rock and hard place, steeped in bias with either approach.
Blackwell centers on the problem underlying interpretations of causes and effects: “races that become close over the course of the campaign are more likely to go negative than those that are safe.”
Blackwell uses some sophisticated modeling to try to deal with this problem, analyzing 176 gubernatorial and senatorial elections from 2000 through 2006. I’ll skip the statistics and get to the main finding, because it precisely illustrates what I was saying in my zombie post:
bq. Once I correct for the biases due to time, I find that negative advertising is an effective strategy for Democratic non-incumbents. This stands in contrast to the previous literature on negative advertising, which, according to Lau, Sigelman, and Rovner (2007), “does not bear out the idea that negative advertising is an effective means of winning votes.”
Now, as written, it would appear that this directly contradicts my argument in the zombie post, since I used the Lau, Sigelman, and Rovner meta-analysis to make my point. But if you look at his findings, you’ll actually see an illustration of my point about context and nuance. He finds that negative ads work for Democratic non-incumbents but only when aired close to Election Day. Among Democratic incumbents, negative ads also work only when aired close to Election Day — but they work in the opposite way intended, driving down vote share for those incumbents.
In other words, we have a finding that confirms the conventional wisdom but is limited to one set of candidates and to ads aired at a particular time. And then we have a finding for a different set of candidates that utterly contradicts the conventional wisdom. And this is in the article that probably does the best job to date of dealing with the problems that Arena raises.
Of course, it’s just one article. So we shouldn’t hang everything on it. Still I think that the academic literature on negative advertising isn’t so hamstrung by methodological problems that journalists and others can safely ignore it and go on believing that negative ads “work,” full stop. Context and nuance are necessary, and they will suffice to kill the zombie.