"Talking past each other": Issue ownership and microtargeting in Swiss online political ads

TL;DR

This study quantifies online political advertising in Switzerland's direct-democracy context, using 40k Facebook/Instagram ads (34,559 Swiss-centric) from 2021–2025. It combines Meta Ad Library data with referenda and federal-election annotations, revealing strong demographic microtargeting, pre-vote surges in referenda advertising, and a pattern of issue ownership where parties emphasize their own issues rather than debating shared ones. A machine-learning analysis shows that ad audience and topic features can effectively predict an ad's author, underscoring the distinct strategic profiles of major Swiss parties. The work highlights potential risks to deliberative democracy from microtargeting and content divergence, while advocating transparency and data-driven accountability in political advertising.

Abstract

Switzerland's unique system of direct democracy, characterized by frequent popular referenda, provides a critical context for studying the impact of online political advertising beyond standard electoral cycles. This paper presents a large-scale, data-driven analysis of 40k political ads published on Facebook and Instagram in Switzerland between 2021 and 2025. Despite a voting population of only 5.6 million, the ad campaigns were significant in scale, costing CHF 4.5 million and achieving 560 million impressions. This study shows that political ads are used not only for federal elections, but also to influence referenda, where greater exposure to ``pro-Yes'' advertising correlates significantly with approval outcomes. The analysis of microtargeting reveals distinct partisan strategies: centrist and right-wing parties predominantly target older men, whereas left-wing parties focus on young women. Furthermore, significant region-specific demographic variations are observed even within the same party, reflecting Switzerland's strong territorial divisions. Regarding content, a clear pattern of ``talking past each other'' is identified: in line with issue ownership theory, parties avoid direct debate on shared issues, preferring to promote exclusively owned topics. Finally, it is demonstrated that these strategies are so distinct that an ad's author can be predicted using a machine learning model trained exclusively on its audience and topic features. This study sheds light on how microtargeting and issue divergence on social platforms may fragment the public sphere and bypass traditional democratic deliberation.

Figures (9)

• Figure 1: Seven-day rolling average of ad impressions. The inset provides a daily view, disaggregated by language, for the 30-day period preceding the federal election.
• Figure 2: Comparison of the demographic reach by age and gender of each party's ads versus all other ads. Gender is measured using male-to-female odds ratios, and age using Wasserstein distances.
• Figure 3: Language-specific age distributions (bootstrapped mean and SD) for each party's ads compared to its complementary ad set. The Wasserstein Distance measures the divergence between the two.
• Figure 4: Male-to-female odds ratios for each party, relative to its complement, by language. Values are bootstrapped means, and 1 means perfect gender balance. An asterisk indicates that the 95% confidence interval does not include 1.0.
• Figure 5: Partial effect of days to referendum (t=0) on log-transformed Expenditure (blue) and Impressions (orange), shown by language and aggregate. Shaded areas represent 95% confidence intervals.