What propaganda cannot do: the limits of persuasion

Series: How minds are moved · Living with it: self-defence for the mind · 1 of 4

In 2018 two political scientists, Joshua Kalla and David Broockman, pooled 49 field experiments on what campaign contact does to voters in American general elections: door knocks, phone calls, mail, ads. Their best estimate of the average effect on whom people vote for was zero. Not small. Zero. In the same years, much of the public conversation assumed that a few ads, a few bots or a clever data firm could steer an election. Both cannot be right in the same way, and the gap between them is the subject of this post.

This family is about living with manipulation, not just describing it. The first thing to know is what it cannot do.

Contents · 9 sections

TL;DR #

• Most measured persuasion is weak. Campaign contact in general elections averages roughly zero effect on vote choice; the Russian troll accounts on Twitter in 2016 show no measurable effect on attitudes or votes.
• Propaganda works best on people already inclined to agree. It mobilises, gives permission and coordinates more than it converts.
• When it does work, it can be deadly. Radio in Rwanda in 1994 and in Germany after 1933 had measurable effects, but mainly where listeners were already primed.
• Believing that manipulation is all-powerful is itself a trap: it excuses, it justifies censorship, and it makes people cynical enough to believe nothing.
• The practical stance: take manipulation seriously, and take your own judgement seriously too.

Minimal effects: an old finding that keeps coming back #

The idea is not new. In 1940, Paul Lazarsfeld and his team followed voters in Erie County, Ohio, through a presidential campaign. Their book, The People’s Choice (1944), found that the campaign mostly reinforced what people already thought; few switched. In 1960 Joseph Klapper summed up the research of his time: mass communication works mainly through existing attitudes and social groups, and it reinforces far more often than it converts.

Then came television, the internet and social media, and the fear returned with each one. The modern evidence is better, because it uses randomised experiments rather than surveys, and it points the same way.

Kalla and Broockman’s 2018 paper, published in the American Political Science Review, combined 40 earlier field experiments with 9 of their own. In general elections the average persuasive effect was zero. They found effects only in rare cases: when a candidate took an unusually unpopular position and a campaign invested heavily in finding the few voters who could be moved. Turnout can be moved; minds, much less.

The Russian Internet Research Agency’s 2016 Twitter campaign is the best-known modern fear. A 2023 study in Nature Communications by Gregory Eady and his co-authors linked a survey of Americans to their actual Twitter feeds. Exposure was extremely concentrated: 1% of users accounted for 70% of exposures, mostly people who already identified strongly as Republicans. Users saw 25 times more posts from national news media than from the Russian accounts. The study found no meaningful relationship between exposure and changes in attitudes, polarisation or voting. The operation is documented. Its effect on those votes is not.

Why people are harder to move than we think #

The cognitive scientist Hugo Mercier makes this argument at book length in Not Born Yesterday (2020). It is not in my library yet, so I am going by the published book and its reviews, not my own shelf. His case is that humans evolved “open vigilance”: we are open to information because it is useful, and vigilant because others may want to deceive us. We weigh a message by who sends it, whether it fits what we already know, and whether the argument holds. Mass messages, which cannot be tailored to each listener’s knowledge and trust, are weak by design.

My own library has an older version of the same lesson. Denise Winn’s The Manipulated Mind reviews the “brainwashing” scare after the Korean War and notes that the American prisoners who returned home “did not retain Communist ideals”. Many had collaborated to make life easier, not because they had been converted. Winn also notes that John Wesley, after his hellfire sermons, quickly divided new converts into groups that met at least once a week, while preachers who thought their work was done at conversion “soon lost most of those they had so dramatically won”. Conversion that is not supported by daily social life tends to wear off.

Where propaganda does work #

None of this means propaganda is harmless. The best studies also show where it bites.

In Rwanda in 1994, the radio station RTLM called for the killing of Tutsi. In a 2014 study in the Quarterly Journal of Economics, the economist David Yanagizawa-Drott used the hills that blocked radio reception in some villages and not others. He estimated that about 51,000 perpetrators, roughly 10% of the participation in the violence, can be attributed to the station. A large part of the effect came through spillovers into neighbouring villages: the radio coordinated people as much as it persuaded them.

In Germany, a 2015 study in the same journal by Maja Adena and her co-authors found that radio slowed Nazi growth before 1933, while the Weimar government controlled it. After the Nazis took control of it, radio increased denunciations of Jews and open antisemitism, but mainly where antisemitism had been historically high. Where it had been low, the same broadcasts had the opposite effect.

That is the pattern across the strongest evidence. Propaganda is weak at turning a neighbour into an enemy. It is strong at telling people who already resent that neighbour that they are allowed to act, that everyone else will act too, and when.

Two other effects are real and quieter. Repetition makes statements feel more true, the “illusory truth” effect; it is small for any single exposure but cumulative. And the media are better at setting what we think about than what we think. The choice of which stories are big is itself a form of influence, which is why the next post is about counting voices.

The trap: believing in all-powerful manipulation #

If people are not easily moved, then the belief that they are can do harm of its own.

It excuses. “They were brainwashed” explains away people who disagree with you without listening to them. It is often said about the other side’s voters, rarely about one’s own.

It is sold. The UK Information Commissioner closed its investigation into Cambridge Analytica in 2020. According to the reporting of its letter to Parliament, it found that the firm’s models and methods were exaggerated, that the company had oversold the effectiveness of its “psychographic” targeting, and that there was no evidence the harvested Facebook data was used in the Brexit referendum campaign. The firm’s own sales pitch and its critics’ fears described the same magic, and the regulator found less of it. The Commissioner also warned that the tools themselves are not going away.

It justifies censorship. If any false message can turn the public, then the public cannot be trusted with false messages, and someone must filter them. Every side has used this argument when it held the filter. It is the bridge from fear of propaganda to the censorship family of this series.

It breeds cynicism. If everything is manipulation, nothing can be trusted, and a person who trusts nothing is not free; they are simply available to whoever shouts last. A 2016 RAND report on Russian state media called its model a “firehose of falsehood”: high volume, many channels, rapid repetition and no commitment to consistency. My inference, not RAND’s wording: a public that has given up on telling true from false is exactly what such a firehose needs.

What it feels like from the inside #

Two feelings, and both are worth noticing. The first is helplessness: the sense that “they” can make people believe anything, so your own judgement does not matter. The second, its twin, is the special feeling of being one of the few who see through it, while the “sheep” are fooled. Both feelings flatter someone. The first flatters the manipulator; the second flatters you. Neither matches the evidence, which says that most people, most of the time, are stubborn, sceptical and attached to what they already think.

How to notice it #

• A claim that a campaign, ad or troll farm “swung” an outcome, without a measured effect. Ask: how many people saw it, who were they, and what changed?
• Exposure numbers (“reached 100 million”) presented as effect numbers. Reach is not persuasion.
• “They were brainwashed” used to explain people you disagree with, but never people you agree with.
• Calls to restrict speech based on how powerful the speech is supposed to be.
• Your own drift into “nothing is true” after a long period of reading about manipulation.
• Messages that tell you that you are already right and are now allowed to act. That is where propaganda is strongest.

Frequently asked questions #

Does political advertising change how people vote? #

In US general elections the pooled experimental evidence says the average effect is close to zero. Kalla and Broockman (2018) found persuasive effects only in rare cases, such as unusually unpopular candidates and heavily targeted campaigns. Effects on turnout and in primaries can be larger.

Did Russian trolls change the 2016 US election? #

The operation is documented. A 2023 study in Nature Communications found that exposure on Twitter was concentrated among 1% of users, mostly strong partisans, and found no measurable relationship with changes in attitudes or votes. It did not study other platforms or channels such as hacked emails.

Can propaganda cause violence? #

Yes, in some conditions. A 2014 study estimated that RTLM radio in Rwanda accounts for about 10% of the participation in the 1994 genocide, around 51,000 perpetrators. Its effect worked largely by coordinating people already exposed to hostility.

What is “open vigilance”? #

Hugo Mercier’s term for how people evaluate communication: open to useful information, but checking the source, plausibility and arguments of what they hear. On his account, mass persuasion fails more often than it succeeds.

If manipulation is weak, why worry about it? #

Because “weak on average” is not “harmless”. Small effects matter in close contests; propaganda can mobilise people already inclined to act; and repetition, agenda-setting and censorship shape what people get to see at all.

The lesson #

The honest picture is in between. Manipulation is real, documented and sometimes deadly, and it mostly works by finding people who are already halfway there. It is not a remote control for minds. That matters because the opposite belief makes us easier to handle: it tells us to distrust our neighbours, to accept filters, or to give up on truth. The best defence is not fear of propaganda. It is confidence in your own vigilance, used every day.