Many people's first explanation for polarization is the media: cable news, talk radio and, above all, social media. The research here is lively and often surprising. It supports some popular beliefs, weakens others, and shows that the link between screens and division is less direct than it looks.

From three networks to many channels

In the 1960s most Americans got television news from three broadcast networks aiming at the broad middle. Cable, talk radio and later the internet broke that pattern. The political scientist Markus Prior argued that the main effect of more choice was that people uninterested in politics could avoid news entirely, while the most interested could consume much more of it. The result was a wider gap between the engaged, who tend to be more partisan, and everyone else.

Partisan outlets clearly exist and find large audiences on both left and right. Studies of cable news suggest it can shift viewers' votes a little. But most Americans consume relatively little political news of any kind, and audiences for the most partisan outlets, while influential, are a minority of the public.

The echo chamber thesis

The popular picture of the internet is the echo chamber or filter bubble: algorithms show people only what they already agree with, and views harden. Evidence for this is weaker than many assume. Studies of web browsing have found that most people's news diets overlap more than expected, and that heavy partisans often do see the other side's arguments, frequently in order to mock them. Exposure alone does not produce agreement.

Puzzles in the evidence

Three findings complicate the social media story.

First, timing and age. In 2017 the economists Levi Boxell, Matthew Gentzkow and Jesse Shapiro found that in the United States polarization grew most among older Americans, the group least likely to use the internet and social media. That does not rule out an online effect, but it means social media cannot be the whole story.

Second, exposure to the other side. In a 2018 experiment led by the sociologist Christopher Bail, Twitter users were paid to follow a bot that retweeted messages from the opposing side for a month. Republicans who did so became more conservative, and Democrats became slightly more liberal, though that change was not statistically clear. Simply seeing the other side's loudest voices did not build understanding; if anything it confirmed people's worst impressions.

Third, large platform experiments. During the 2020 United States election, researchers working with Meta ran studies with tens of thousands of consenting Facebook and Instagram users, published in Science and Nature in 2023. They changed feeds to chronological order, reduced content from like-minded sources, or removed reshared posts. Over about three months, these changes did not measurably change users' polarization, though some changed what users saw and how much they engaged. Critics note that three months is short, that the platforms had already shaped people for years, and that effects on the whole society may differ from effects on individuals.

Where media effects may be real

Research points to some more specific channels. Social media rewards content that provokes outrage, and posts attacking the other side tend to get more engagement than posts praising one's own. That may not change the average user's views much, but it changes what kind of political voice gets amplified, giving the most hostile voices outsized visibility. That in turn feeds false polarization: people see the most extreme members of the other side and mistake them for typical members.

The decline of local newspapers may also matter. Studies have found that when a local paper closes, voting becomes somewhat more nationalized and straight-ticket, as people lose information about local candidates and fall back on national party labels.

A fair verdict

Few researchers think media are irrelevant, and few think they are the root cause. A reasonable reading is that media changes magnify divisions that have deeper sources in sorting, identity and institutions, and they especially distort our picture of the other side. Societies with similar internet use differ greatly in polarization trends, as Level 6 shows, which also suggests technology is not destiny.

Worked example: one viral post

Suppose a member of Party X posts an angry, insulting comment about Party Y. It gets a million views, mostly from Party Y members sharing it in outrage. Meanwhile ten thousand ordinary Party X members post nothing at all about politics that day.

What has happened? Party Y members now have a vivid example of what "Party X people" are like. Nothing in their views of policy has changed, and the post may not have persuaded anyone. But their mental image of the other side has been drawn from its most hostile member, not its typical one. Multiply this daily and you get a public with exaggerated beliefs about how extreme and hateful opponents are. The research suggests the most useful question to ask about any viral political post is: how typical is this person?