Effects of personalised content on affective polarisation in referendum campaigns - what donated data from Instagram can tell us

Social media platforms have become central pillars of today’s communication landscape, deeply shaping how political information is presented, disseminated and consumed. The use of social media has increased in recent years, both among campaign actors, as an additional channel for campaigns, and among citizens, as a source for information. This development is paralleled by a trend towards rising levels of affective polarisation across many democracies, sparking widespread speculations about a potential association between these two phenomena. Yet, studies examining the question whether social media use is linked with higher levels of affective polarization have so far yielded mixed findings. Most previous research focuses on the US-context and is based on survey data that relies on self-reported social media use. Moreover, it remains unknown which kind of content is responsible for triggering affective polarisation. The present study addresses these gaps in two ways. First, we focus on the hitherto understudied area of direct democratic campaigns, which have been linked to the emergence and intensification of new and existing group identities in previous research. Building on this strand, and integrating it with the literature on political personalisation, we argue that the exposure to personalised communication styles, frequently employed by political actors in direct democratic online campaigns, influences the level of affective polarisation among citizens. Second, we adopt an innovative methodological approach by combining digital trace data (DTD) from Instagram with survey data collected through a national post-vote survey in Switzerland (DDS-21 project). The survey consists of a stratified random sample of around 2500 Swiss citizens from the three big linguistic regions. Out of these respondents 50 respondents, who use Instagram, donated their DTD. Compared to self-indicated media use measured in surveys, the use of DTD allows for a nuanced analysis of use frequency and analysis of the content users are exposed to, due to the richness of each data donation with its several thousand data points. Moreover, DTD collected by data donation – compared to other methods such as tracking – is unaffected by the data collection as the data is saved automatically and only collected afterwards. The study focuses on two popular initiatives that were put to vote in November 2025. The two proposals fit the purpose of this article very well. The first proposal, which’s main demand was the introduction of an inheritance tax on high inheritances, polarised on a very traditional left right axis with leftist supporting and conservatives opposing the proposal. The second proposal, which’s main demand was the introduction of a general civil service, does not fit well into existing party cleavages. Only the smallest of the six big Swiss parties supported the ballot and opposition came both from left and right, making the proposal interesting as citizens from all sides could find reasons for and against the proposal. The findings of this study contribute to our understanding of the role of social media content in the emergence of in- and out-group affects in issue specific groups and opinion formation processes.