Predicting the winner of the US 2024 elections using trust analytics
Katarzyna Budzynska, Ewelina Gajewska
TL;DR
The paper tackles predicting the outcome of the US 2024 presidential election by analyzing trust dynamics in social discourse rather than traditional sentiment. It introduces Trust Analytics (TrustAn) built on an ethos-mining pipeline (pprai), a RoBERTa-based classifier achieving macro-$F_1$ = 0.758 on polarised data, to label Reddit discussions as trust/distrust or neutral toward candidates. It computes a weekly trust slope $s_1$ and a trust/distrust ratio profile to track credibility shifts for Harris and Trump across key election events, concluding that Harris’ trust declines while Trump maintains a stable distrust/trust balance, leading to a predicted Trump victory. The work demonstrates the potential of automated ethos analysis of online political discourse as a complementary forecasting signal with practical implications for digital-era political analysis.
Abstract
A number of models and techniques has been proposed for predicting the outcomes of presidential elections. Some of them use information on the socio-economical status of a country, others focus on candidates' popularity measures in news media. We employ a computational social science approach, utilising public reactions in social media to real-life events that involve presidential candidates. Contrary to the popular approach, we do not analyse public emotions but ethotic references to the character of politicians which allows us to analyse how much they are (dis-)trusted by the general public, hence the name of the tool we developed: Trust Analytics (TrustAn). Similarly to major news media's polls, we observe a tight race between Harris and Trump with week to week changes in the level of trust and distrust towards the two candidates. Using the ratio between the level of trust and distrust towards them and changes of this metric in time, we predict Donald Trump as the winner of the US 2024 elections.