In-Group Love, Out-Group Hate: A Framework to Measure Affective Polarization via Contentious Online Discussions

TL;DR

This work tackles the real-time quantification of affective polarization by introducing a discrete choice model that separates in-group love via $\alpha$ and out-group hate via $\beta$, with inertia $\delta$. It provides a logistic-regression–based estimation procedure for these parameters from social media data and validates the approach on COVID-19–related Twitter discussions about masking and lockdowns, showing it can reproduce observed polarization dynamics. Key findings include that $\alpha>\beta$ on both issues, partisans exhibit stronger affective polarization signals, and state-level geography modulates these parameters, yielding insight into the spatial heterogeneity of polarization. The framework offers practical implications for moderating online discourse and can be extended to multi-party settings and other contentious topics, contributing to computational social science by linking network dynamics, emotion, and opinion formation.

Abstract

Affective polarization, the emotional divide between ideological groups marked by in-group love and out-group hate, has intensified in the United States, driving contentious issues like masking and lockdowns during the COVID-19 pandemic. Despite its societal impact, existing models of opinion change fail to account for emotional dynamics nor offer methods to quantify affective polarization robustly and in real-time. In this paper, we introduce a discrete choice model that captures decision-making within affectively polarized social networks and propose a statistical inference method estimate key parameters -- in-group love and out-group hate -- from social media data. Through empirical validation from online discussions about the COVID-19 pandemic, we demonstrate that our approach accurately captures real-world polarization dynamics and explains the rapid emergence of a partisan gap in attitudes towards masking and lockdowns. This framework allows for tracking affective polarization across contentious issues has broad implications for fostering constructive online dialogues in digital spaces.

Figures (17)

• Figure 1: Example trajectories of the $\theta(t) = \left[ \theta^{\mathcal{B}}\left(t\right), \theta^{\mathcal{R}}\left(t\right)\right]$ on a fully connected graph based on Eq. \ref{['eq:FC_ODE']} (under the first definition of peer influence) for various parameter configurations. In each case, it is assumed that $\theta^{\mathcal{B}}\left(0\right) = \theta^{\mathcal{R}}\left(0\right)$ i.e., both groups initially have the same prevalence of the dynamic attribute.
• Figure 2: Example trajectories of the $\theta(t) = \left[ \theta^{\mathcal{B}}\left(t\right), \theta^{\mathcal{R}}\left(t\right)\right]$ on a fully connected graph based on Eq. \ref{['eq:FC_ODE']} (under the first definition of peer influence) for various parameter configurations. Unlike Fig. \ref{['fig:same_initialcondition_trajectories']}, the two groups initially have different prevalences of the dynamic attribute.
• Figure 3: Masking stances over time. Share of liberal and conservative users expressing pro-masking stance for (a) all users and (b) partisans.
• Figure 4: Lockdown stances over time. Share of liberal and conservative users expressing pro-lockdown stance for (a) all users and (b) partisans.
• Figure 5: Plotting trajectories of issue positions on masking for (a) all users and (b) political partisans.

Theorems & Definitions (2)

• definition 1: Net number of individuals with a stance normalized by degree
• definition 2: Net fraction of neighbors in each group with a stance