Can science-based interventions tamp down polarization?

An interdisciplinary coalition of behavioral and social scientists are working to bridge the ever-widening political divides. Their efforts offer some reason for optimism.

Long before the election cycles of 2024, 2020, or even 2016, anyone who followed US political news could sense it—a toxic, ever-escalating tide of alienation, suspicion, and loathing for the opposing side. For the “other.” For “them.”

Polarization is rampant, but there is cause for hope. An “exhausted majority” is tired of the partisan rancor. Research into interventions could suggest paths to finding or revealing common ground. Image credit: Dave Cutler (artist).

“It’s gotten worse and worse and worse, with no signs of getting better,” says Jay Van Bavel,* a psychologist at New York University (NYU) and co-author of a 2020 review on the subject (1). Call it what you will—tribalism, sectarianism, or an academic favorite, affective polarization—the mutual hostility and contempt started growing as early as the 1990s, he says, then accelerated rapidly in the 2010s. By the time COVID-19 hit in 2020, polarization had become a serious threat to national unity, democracy, and even health. “Thousands of people were dying needlessly because they didn’t trust medical advice, or their government, or mainstream media,” Van Bavel says. And now the harsh rhetoric of the 2024 campaign has widened the nation’s civic fault lines even further.

Yet, there are grounds for hope. In survey after survey, Stanford University sociologist Robb Willer says, “most Democrats and Republicans are united in a kind of ‘exhausted majority’ that is tired of partisan rancor and would prefer a different way forward” (2–6). And the vitriol of the 2024 election campaign doesn’t seem to have changed that fact, adds University of Rochester political scientist James Druckman, echoing a point made by many in the field. According to polls before and after the election (7, 8), Druckman notes, “people clearly sent a message that they care about basic issues like grocery prices, and not the rhetoric that extremists were pushing.”

Surveys also find that cross-partisan hostility is at least partly a matter of misperception: People tend to view members of the opposing party as being far more hostile and extreme in their views than most of them really are (9, 10). So, if people can somehow get past the media stereotypes and see each other as complex human beings, says Jan Voelkel, a sociology postdoc at Cornell University, “the image that emerges is so much better than the expectation that the conversations tend to go quite well.”

Researchers have had some success in their search for rigorous, reliable, and lasting ways to foster this kind of rapprochement. Indeed, the looming sectarian threat has brought a flood of new money, people, and ideas into polarization research, Van Bavel says, adding, “The quality of science has gotten so much better than when I was a grad student.” (See the Polarization and Trust Special Feature in PNAS Nexus, https://academic.oup.com/pnasnexus/issue/3/10#2052166-7821165.)

In particular, a new social science tool called a “megastudy” is offering greater clarity on which interventions work best for which issues (11, 12). A prime example is the Strengthening Democracy Challenge, a megastudy organized by Willer and a team of collaborators from across the country (13). Van Bavel calls it “the gold standard.”

The megastudy approach allowed the challenge team to compare the effects of 25 different interventions on about 32,000 online test subjects simultaneously and to collect some of the most reliable data so far on which interventions yield the biggest reductions in political tribalism. Among the notable findings: Cooling partisan animosity was surprisingly easy, with almost all of the 25 interventions producing at least some reduction. But lowering support for political violence and democracy suppression was a lot harder, with only a handful of the 25 showing promise. And a still-unsolved problem is how to sustain those animosity reductions after the interventions end—at which point participants find themselves back in the same divisive mediascape.

Going Big

A critical catalyst for the Strengthening Democracy Challenge came in 2019, when a nongovernmental organization (NGO) called the Civic Health Project gave Willer $250,000 to carry out a research project of his choice on depolarization and bridge-building across the partisan divide.

“Our perception was that there was a lot of scholarship describing the polarization problem, but not as much that was exploring and testing solutions,” says Kristin Hansen, who had walked away from her marketing-executive career and joined with management consultant Rob Romero to co-found the Civic Health Project in Menlo Park, California. “What do we do about polarization?” Hansen wondered. “What do we try? What do we test? How do we know what works?” Willer, she’d learned, had been looking to tackle those same questions ever since the divisive 2016 US presidential campaign.

But how to proceed? “I’d been trying to think of an experimental design that would tap the ideas of many, many researchers and practitioners in parallel,” Willer explains. He was especially eager to break out of the academic bubble and include practical insights from the “bridging movement”: a loose coalition of grassroots NGOs that had emerged by the dozens over the previous decade or so and taken names such as Civity, ListenFirst, Braver Angels, and More in Common. Some of these NGOs, including Hansen’s, were focused on organizing, fund-raising, and research. But most of them were conducting fieldwork, often setting up face-to-face conversations between small groups from opposing parties, then helping participants listen to and understand each other.

The solution to Willer’s design conundrum emerged in a brainstorming session with Hansen and Romero and combined two relatively new approaches in social science.

One, the megastudy, had been invented and named in 2016 by behavioral scientist Katherine Milkman and psychologist Angela Duckworth at the University of Pennsylvania’s Wharton School of Business. If you really want to tackle big societal problems, Milkman says, “you don’t want to just do science the old-fashioned way, which is lots of people coming up with their pet ideas in isolation, testing them in different populations—and then having no idea how to compare the results because it’s apples to oranges.” Instead, she and Duckworth had realized that you want to pit a bunch of promising ideas head-to-head, with test subjects drawn from the same population and results evaluated according to the same metrics. That way, Milkman says, you can compare magnitudes and cost-effectiveness directly and get a meaningful answer for which approaches work best under which circumstances.

Milkman, Duckworth, and their Wharton colleagues have successfully demonstrated the megastudy idea on their main research interest: finding better ways to nudge people into beneficial habits, such as saving for retirement, eating properly, getting their vaccines, or even working out regularly (14). Other labs have picked up on the megastudy approach to evaluate messaging in areas such as climate change (12, 15). But Willer’s lab in 2019 was the first to apply the idea to political polarization—albeit by combining it with a second approach, crowdsourcing (16, 17).

A megastudy compared the effects of 25 different polarization interventions on about 32,000 online test subjects simultaneously. While cooling partisan animosity was surprisingly easy, lowering support for political violence and democracy suppression was a lot harder. Image credit: Reprinted with permission from ref. 13 (AAAS).

Give the People What They Want

In many megastudies, the researchers draw up their own list of interventions to test, says Voelkel, who was then a PhD candidate in Willer’s lab and would become a lead researcher on the challenge. But the goal of this megastudy was to gather ideas that their lab had never thought of. In addition to the bridging-movement practitioners, Voelkel says, it was important to bring together “psychologists, political scientists, communication scholars, economists, sociologists, and people from this emerging field of computational social science.”

So Willer recruited a far-flung research team, including Druckman and Massachusetts Institute of Technology cognitive scientist David Rand as co-principal investigators, and put out a worldwide call for individuals or groups to submit candidate interventions. Because common bridging interventions, such as face-to-face encounters or in-person training sessions, would be logistically unmanageable, they restricted the submissions to digital formats that participants could access online and get through in 8 minutes or less. But other than those criteria, submitters had free rein. Video, audio, text, and interactive quizzes were fair game—the messages would be judged based only on how well they reduced three worrisome aspects of polarization: partisan animosity, support for partisan violence, and support for undemocratic practices such as voter suppression and refusal to accept election results.

Voelkel says the challenge team was astounded by the response: They received 252 submissions from 17 different countries—10 times the number for which they had funding. Willer had to double the size of the challenge’s advisory board in order to review all the candidates and settle on 25 submissions (see https://www.strengtheningdemocracychallenge.org/winning-interventions) that represented the full range of theories and approaches in the submission pool.

To carry out the experiment, the challenge team started with an online panel of just over 32,000 paid participants who’d been statistically matched to the US population in terms of age, gender, education, ethnicity, and party affiliation. Then, the team randomly assigned about 1,000 participants apiece to each of the final 25 candidate interventions and randomly designated about 5,000 more participants to serve as a control group.

Next, the participants were each given their assigned intervention, while the control group saw nothing. Then, everyone took a short survey with questions along the lines of, “If you had 50 cents, how much would you share with a stranger from the opposite party?” This survey was designed to gauge the participants’ attitudes on several variables that included the main three—partisan animosity, support for political violence, and support for undemocratic practices—then express the results as numbers between 0 and 100.

Finally, a comparison between the control group’s numbers and those from the treatment groups showed the impact of each of the 25 candidate messages.

Reaching Across the Aisle

One of the most striking findings of the study was that almost all the messages produced statistically significant reductions in partisan animosity, with some doing so by more than 10 points out of 100. Hence, the Stanford researchers found that reducing polarization “is not something that is impossible to do, but is actually quite easy to do,” says Palma Strand, research director for the bridging organization Civity.

For example, one simple, but very effective, strategy used by several of the challenge’s final 25 contestants was to remind participants that Americans of both parties embrace core values such as democracy, free speech, and respect for election results. Some of these messages also included quizzes or videos using survey data to correct misperceptions about the other side. Contrary to a common belief among Democrats, for example, most Republicans say they would never support laws making it harder for Democrats to get elected. “It’s about our shared identities and the fact that the divides between us are not as big as we think,” says Van Bavel, who co-authored an NYU submission that used this approach and lowered partisan animosity scores by 9 points relative to the controls.

But perhaps the most commonly used strategy among the 25 contenders was equally simple and effective: show that individuals from the other side are not stereotypes, but actually complex, sympathetic human beings.

Civity’s entry was a good example (see https://sshs.qualtrics.com/jfe/form/SV_6VjWZrKyGaCI5IW?Condition=Civity_Storytelling). The heart of it was a set of five 45-second videos, each showing a different person looking into the camera and, in their own words, giving viewers their answers to three questions: what story people may miss about them, how they surprise people who may stereotype them, and what they bring to their community. This was about the simplest presentation the Civity team could think of, says organization CEO Malka Kopell: “It was like watching five people having one side of a two-way conversation.” But even these brief glimpses into strangers’ inner lives seemed to leave viewers in an empathetic frame of mind. Civity’s submission was another of the challenge’s top performers, reducing partisan animosity scores by more than 9 points.

This outcome makes perfect sense from a social science standpoint, Van Bavel says. Knowingly or not, many of the most effective interventions were based on evidence-based theories that have been around since the 1950s (18, 19). Among the most important of these theories is the contact hypothesis (18), which says that face-to-face interactions can indeed decrease tensions and stereotypes—especially when the two sides meet as equals and are working together on a common goal.

“Submitters may have thought that by targeting partisan animosity, they would reduce all three metrics. But an important lesson from this study is you cannot assume that.”

—Jan Voelkel

A vivid example of this dynamic (see https://sshs.qualtrics.com/jfe/form/SV_6VjWZrKyGaCI5IW?Condition=Contact_Project) was submitted by a team from Bowdoin College in Maine. The submitting researchers simply asked viewers to watch “Worlds Apart,” a Heineken beer ad from 2017 that had first aired in England and then gone viral on social media. The 4.5-minute video first introduces viewers to two strangers who are unaware of their profound disagreements on feminism, two with equally strong disagreements on climate change, and yet another two who differ on transgender identity. The video follows each pair as they assemble a flat-pack bar and engage in some nonpolitical getting-to-know-you conversation. Finally, each individual is told their teammate’s views on the issue about which they feel most deeply and is given a choice: either leave or stay and keep talking with their teammate over a (Heineken) beer. They all stay.

Although this Heineken ad has been sharply criticized—not least for suggesting that intractable differences could be solved over a beer—the vast majority of people who’ve watched it since 2017 have reacted quite positively (20). Certainly, the Stanford challenge’s participants did: Viewing the Bowdoin College submission reduced their partisan animosity scores by almost 11 points compared to the control group, more than any other entry in the contest.

Research has suggested that the difference between metrics for “in-party love” and “out-party hate” has greatly increased since 1980, implying a disturbing trend that has people hating the other party much more than they love their own. Image credit: Reprinted with permission from ref. 1 (AAAS).

Pride and Prejudice

Unfortunately, Voelkel says, the good news about partisan-animosity reduction has to be tempered by another striking finding from the study. None of the 25 entries were especially good at reducing scores on the other two measures of sectarian sentiment: support for undemocratic practices and support for partisan violence.

Few of the entries achieved reductions on these variables that were statistically different from zero—not even the Heineken ad. And those that did better could be quite inconsistent. Take a video submitted by a Stanford group (see https://sshs.qualtrics.com/jfe/form/SV_6VjWZrKyGaCI5IW?Condition=Democratic_Fear). It showed just how bad democratic collapse has been in countries such as Venezuela, Russia, and Turkey, which have had riots in the streets and police repression. The scenes of violence were supposed to remind viewers that they shouldn’t take democracy for granted.

And the intervention accomplished this—sort of. It reduced support for undemocratic practices by almost 5 points, making it one of the best on that metric. But it actually boosted support for partisan violence by more than 2 points, possibly because it included shots of the January 6, 2021, assault on the US Capitol. As Voelkel points out, the most conservative participants may have felt that the attack was justified.

Sorting all this out will be a job for future research, Voelkel says. “Submitters may have thought that by targeting partisan animosity, they would reduce all three metrics,” he says. “But an important lesson from this study is you cannot assume that.” Support for political violence and for undemocratic practices appear to be distinct attitudes that need to be understood on their own terms.

Fighting a Toxic Environment

Meanwhile, Van Bavel points to an even more daunting task for future research: achieving staying power amidst a complex, multifaceted media environment. No matter how effective these messages are in the lab, he says, “the effect is going to be offset when people go back to watching news, or tuning into social media, or hanging out with friends who are hyper-partisan.” This is why no one was surprised by the challenge’s follow-up interviews, which showed that the effects of even the best interventions had suffered up to a 60% decline after two weeks.

On the other hand, Willer says, “we never thought the challenge was going to give us ‘one weird trick’ where you just show somebody a 5-minute video and fundamentally change how they think about rival partisans for all time.” The idea, he says, was always to identify effective bridging strategies that could then be extracted from these online messages and reworked into a form suitable for repeated messaging in the real world.

That could take a while, says Matthew Levendusky, a political scientist at the University of Pennsylvania. “The Strengthening Democracy Challenge was a great way of testing a lot of ideas,” he says. “But scaling these strategies is not easy, and I have yet to hear a great mechanism for doing that from anyone.”

Still, lots of people are trying, starting with the challenge team itself. Thanks to an additional grant from Stanford, the team has begun field tests of four interventions from the challenge (see https://www.strengtheningdemocracychallenge.org/grants). In each case, Willer says, “researchers are pairing up with bridging organizations to evaluate how well the findings from our experiment apply out in the world.” So, for example, the Minnesota Council of Churches is showing Civity’s personal-storytelling video in small-group settings to test how it affects participants’ feelings of partisan animosity and mutual empathy. In another example, a consortium of several nonprofits is showing a version of Stanford’s democracy-collapse video on Instagram and in focus groups of young adults, then testing how it affects participants’ feelings about the importance of democracy. Reports from these field tests should start appearing in about a year, Willer says.

Other researchers, looking for better ways to counter online toxicity and misinformation, are experimenting with tactics such as injecting challenge-style bridging messages into social media streams and down-ranking posts that are filled with name-calling, insults, or links to low-credibility news sources (21, 22).

Of course, not even the most effective of these tactics can accomplish much of anything until established platforms such as Facebook, YouTube, or X start building these strategies into their own feed algorithms, which is hardly a given. But the real world isn’t waiting, Hansen says.

Starting well before the 2024 election, she says, the bridging movement has tried to satisfy the growing appetite for conflict-management training in legislatures, schools, workplaces, local governments, veterans’ organizations, and more. Indeed, Hansen says, “I have come to describe the bridge-building movement first and foremost as a skill-building movement.” And perhaps the single most important skill is what might be called the golden rule of bridge-building: Listen to others as you would have them listen to you—not by judging or lecturing them, but by trying to understand who they are and where they are coming from.

This kind of listening won’t make deep disagreements go away, Civity’s Kopell says. But it can help committees and communities find productive solutions to even the most toxic, divisive issues, instead of endlessly gridlocking themselves into angry disputes that go nowhere. “People do not want to sit down and talk about real stuff,” Kopell says, “unless they trust and respect each other.”