In 2020, at least $6 billion will be spent on political campaigns in the U.S.—over $2 billion on the presidential election alone. These astronomical numbers fuel fears that public opinion will be manipulated, either through traditional techniques like TV ads, cold calling and mailings, or with more sophisticated technology that uses big data for targeted advertising.
But are these fears justified? For nearly two decades, political scientists have systematically tested the effectiveness of political campaigns. To find out whether a campaign strategy works, the best solution is to use randomized control trials. For example, you can randomly select certain neighborhoods to receive a campaign mailing supporting a particular candidate. If those areas vote for the candidate in greater numbers, it should be because the mailing influenced their choice.
This is what Alan Gerber of Yale University, a pioneer in the field, did in one of the first such studies, published in the journal American Behavioral Sciences in 2004. Nearly 100,000 households received mailings in favor of a congressional candidate. Were these people more likely to vote for the candidate? Not one bit. (More precisely, they were 0.2% more likely to do so, which is statistically insignificant.) The study was convincing because it was big: As a rule, the larger the sample, the more solid the conclusion.
In some of the other studies that Prof. Gerber conducted, he found that campaign mailings did have an effect on votes. But these experiments had smaller sample sizes, and their results were contradictory—in one of them, mailings seemed to make people less likely to vote for the candidate. The best way to make sense of such studies is to pool them in a metaanalysis, a statistical test that aggregates the results of many experiments to see if robust patterns emerge.
Last year, in a paper published in the American Political Science Review, political scientists Joshua Kalla of Yale and David Broockman of Stanford looked at all the studies that used randomized trials to test the effectiveness of political campaigns, adding nine of their own studies for good measure. The whole spectrum of campaign tools was covered— mainly canvassing, phone calls and mailings, with a few studies focusing on TV and online ads. The researchers’ conclusion was unambiguous: “The best estimate of the effects of campaign contact and advertising on Americans’ candidates choices in general elections is zero.”
Is all the money spent on political campaigns wasted, then? Not necessarily. Even if each individual campaign strategy has a tiny effect, using them together, or repeatedly,
might multiply their strength. And in primary elections, it turns out that campaigns can have larger (though still modest) effects.
In one of Prof. Gerber’s original experiments, a mailing in a primary election increased the candidate’s vote margin by 2.7%, a significant change. The main reason why campaigning is more effective in a primary is that voters can’t rely on the typical shortcut of party identification, saying simply “I’ll vote for the Democrat” or “I’ll vote for the Republican.” Voters also have weaker opinions and less information about the candidates than in general elections. That makes them more amenable to persuasion, as they learn new information about the candidates. These experiments might seem dated. Who uses snail mail or answers their landlines anymore? The internet is the future, and online campaigns can buy data by the cartload to better aim their targeted messages. Yet there is no evidence that online political ads are any more powerful than old-fashioned TV spots. Indeed, there are good reasons to think that all online ads, not just the political ones, have little impact. Randall Lewis, a researcher at Google, a company that makes 90% of its money selling online advertising, found (in an unpublished paper co-authored with Justin Rao of Microsoft) that the effects of online ads are so small and variable that it is essentially impossible to measure their return.
What about targeted advertising, which was famously used by the political consulting firm Cambridge Analytica in the 2016 presidential election? Ads that take into account an individual’s tastes and preferences can have a small effect on product sales—translating into a few dozen extra purchases after millions of people have been exposed to the ads. But no one has any idea how to obtain even those small effects for political ads—not even Cambridge Analytica, whose work was “pop psychology B.S.,” as Luke Thompson, vice president for politics and advocacy at the Republican analytics firm Applecart, told the L.A. Times in 2018. Why are people so hard to persuade? And does this mean that we are so hopelessly pigheaded that all efforts at changing our minds are wasted? When we encounter a message that challenges our views— like being asked to vote for a candidate we don’t already favor— our first reaction is usually to reject it. We change our minds only if we are provided with good arguments, ideally in the context of a discussion and from a source we perceive as competent and trustworthy. Gaining voters’ trust or engaging them in proper discussion is very hard to do en masse, which is why large-scale persuasion nearly invariably fails to convince us.
When provided with the right reasons by the right people, however, we do change our minds. Profs. Broockman and Kalla found
FROM TOP: MICHAEL BUCHER/THE WALL STREET JOURNAL; NICOLE MILLMAN/REDUX PICTURES
this sweet spot in a 2016 study published in the journal Science, in which canvassers engaged people in a 10-minute conversation on transgender rights. They offered information and arguments and asked the voters to remember a time when they had been “judged negatively for being different,” so that they could better understand the plight of transgender people. This intervention reduced prejudice, and the changes seem to have been genuine, as they were still present three months later.
The huge sums poured into political campaigns rarely achieve such lasting effects, however. So if you’re worried that next year’s election will be gamed by big data analytics or targeted advertising, you can relax—or, better yet, focus on what to do about more effective techniques for manipulating elections, such as gerrymandering and voter suppression.
Prof. Mercier is a cognitive scientist at the Jean Nicod Institute in Paris. His new book, “Not Born Yesterday: The Science of Who We Trust and What We Believe,” will be published by Princeton University Press on Jan. 28.
Evidence shows that ads almost never affect us the way a personal conversation can.
When we encounter a message that challenges our views, our first reaction is usually to reject it.
A volunteer distributes flyers for Sen. Bernie Sanders’s presidential campaign in Manchester, N.H., last month.