Why information campaigns may backfire: Voter heterogeneity and network effects

Political information campaigns that perform well in focus groups and pilot studies may backfire when applied at scale. This column uses a field experiment in the 2023 presidential election in Argentina to demonstrate how a campaign can affect voters through both direct and indirect channels. A small but significant share of voters disagreed with an information leaflet about an outsider candidate. These outraged voters were more active in communicating with their social networks than voters who agreed with the campaign message, creating a net result that was the opposite of the campaign’s intended impact.

In recent years, the world has seen a historically unprecedented rise of populism, with populist parties and politicians not just receiving more votes but also winning elections and getting into power (Schularick et al. 2021). Many economists and political scientists have explained the success of populist narratives with the rise of social media, which provides a platform for anti-incumbent, misleading, and polarising messages (Campante et al. 2023). Can mainstream politicians stand up to disinformation and misinformation through political information campaigns providing fact-checking and correcting misleading statements with expert opinions?

Unfortunately, designing effective political information campaigns is notoriously difficult. Even experienced campaign practitioners are often wrong when predicting which messages would be persuasive (Broockman et al. 2024). Furthermore, many political campaigns are not only ineffective but can even backfire, producing the effects opposite to those intended by their designers (e.g. Hewitt et al. 2024).

In our recent paper (Egorov et al. 2026), we show that the difficulty of predicting campaign results can be explained by a combination of two factors: voter heterogeneity and network effects. Each has been separately documented in the literature, but not in combination. Multiple studies have shown that the effects of exposure to political information vary sharply by prior beliefs or demographics (e.g. Enikolopov et al. 2011, Adena et al. 2015, Kendall et al. 2015, Kalla and Broockman 2018).

The literature has also shown the importance of spillovers and network effects of political communication (e.g. Ames et al. 2016, Satyanath et al. 2017, Arias et al. 2019, Cruz et al. 2020, Guriev et al. 2021, Blattman et al. 2024, Duarte et al. forthcoming). Political information campaigns usually reach only a part of the electorate. The exposed voters then communicate with their neighbours and friends. Therefore, the overall effect of the campaign depends not only on the exposed voters’ reaction to the campaign message but also on the differential intensity and persuasiveness of communications by different groups of exposed voters to their non-exposed peers.

We refer to the campaign’s effects on the exposed voters as its direct effects and the spillover impacts on the non-exposed voters as its indirect effects. We develop a methodology to measure both direct and indirect effects and apply this methodology in a randomised field experiment. We find that both direct and indirect effects are substantial and can have opposite signs. In our experiment, the direct effect goes in the direction intended by the campaign but is dominated by the countervailing indirect effect, thus leading to the campaign’s overall backfiring.

Pinning down direct and indirect effects of a political information campaign

We studied the direct and indirect effects of a political information campaign during the 2023 presidential election in Argentina. This election was unexpectedly won by an anti-elite outsider with little political experience, Javier Milei. Partnering with a local NGO, we randomised the distribution of 15,000 leaflets that presented expert assessments of the likely economic consequences of two of Milei’s policy proposals: dollarising the Argentinian economy and replacing public education with vouchers. The information campaign was conducted in Salta province, where Milei got the largest share of voters in the primary (almost 50%), excluding the capital city of Salta.

We separately identified the campaign’s effects on directly exposed voters and on non-exposed voters in the same precinct, at a very granular level, thanks to three unique features of Argentina’s electoral system. First, official results are reported at the sub-precinct mesa (Spanish for “table”) level. Voters are assigned to mesas alphabetically. Second, parties must publish membership lists, which allowed our partner NGO to target specific voters within specific mesas. Third, Argentina holds a mandatory primary before the first round of the presidential election, which allows us to match mesas with similar baseline characteristics into treatment and control groups.

We randomised precincts into treatment groups, in which some mesas were treated, and control groups that had no treated mesas, and randomly varied campaign intensity within the treated precincts. This design enabled us to separately measure the campaign’s direct effects on the electoral outcomes among treated voters in treated mesas and the indirect effects on voters in untreated mesas in the same treated precincts.

The contrast between the direct and indirect effects of the campaign was striking. Both effects were large but had opposite signs. In the case of the information campaign on the implications of Milei’s education policy proposal, for every 100 leaflets, Milei lost about 20 votes among those directly treated but gained roughly 30 votes among their untreated neighbours, yielding a net increase of 10 votes in the election’s first round. These direct and indirect effects persisted in the runoff round as well. The same pattern reappeared in a preregistered replication experiment before the runoff.

Figure 1 Direct and indirect effects of information campaign on Javier Milei’s vote in the first round of the 2023 Argentinian presidential election

Why the backlash?

Our follow-up survey experiment sheds light on the mechanism. Participants were asked to imagine either believing or disbelieving a campaign leaflet that criticised an outsider politician’s policy proposals. Those who imagined disbelieving the critical leaflet were significantly more likely to report that they would discuss it with friends and neighbours and that their goal would be to convince others to support the outsider. In contrast, those who accepted the message rarely felt motivated to talk about it.

The fact that the outraged minority speaks louder than the persuaded majority echoes insights from social psychology that anger and moral outrage are powerful motivators for communication (Redlawsk 2006, Webster and Albertson 2022). This explains how a locally effective message can produce an opposite result overall.

Our survey results also suggest that this mechanism is particularly likely to operate when a campaign criticises an outsider. Respondents randomised into imagining that they did not believe the leaflet were more likely to report that they would have reacted differently had the leaflet criticised a mainstream candidate, indicating that voters perceive campaigns against outsiders as distinct from those against incumbents or, more generally, established candidates. A possible interpretation is that criticism of mainstream candidates is viewed by voters as common and expected, whereas criticism of outsider candidates may be perceived as an attempt by corrupt elites to protect their position.

Rethinking the design of political information campaigns

Most campaign evaluations capture only direct persuasion, not the social diffusion that follows. Our findings highlight that indirect effects can be large, opposite in sign, and even dominant. This helps explain why campaign messages that test well in focus groups or randomised trials, which measure effects only on directly exposed voters, may underperform or even backfire in real electoral contexts, where voters talk to each other.

Our experiment relies on the unique features of the Argentinian electoral system. But the takeaways from our analysis are relevant for explaining a wider global puzzle: why anti-elite narratives are so resilient to corrective information. Recognising that communication does not stop at individual exposure but continues through conversations over social networks, both offline and online, helps in designing interventions that avoid backlash.

Campaign designers should explicitly measure and take into account voter heterogeneity and differential social diffusion. Evaluating messages in focus groups or small online samples is insufficient when spillover effects can flip the overall impact. This lesson potentially extends beyond electoral politics and has implications for all information campaigns within organisations or across societies.

Source : VOXeu