Measure Individual Ideology From Social Media Short Texts With a Dynamic Lexicon

🔎 The Problem

Political language shapes civic discourse, and social media lets ordinary users adopt elite rhetoric. Standard ideological text-scaling tools break down on the short, informal posts typical of platforms such as Twitter, preventing meaningful individual-level inference.

🔧 What the Method Does

Introduces the first viable approach to estimate individual-level ideological positions from social media content. The dynamic lexicon approach places social media users—elites, parties, or citizens—on a shared ideological dimension despite the brevity and informality of posts.

📚 How Validity Was Assessed

• Demonstrates correlation with existing measures of ideology across multiple political contexts.
• Validates performance in multiple languages.
• Tests predictive power by forecasting out-of-sample, individual-level voting intentions.

📈 Key Findings

• Short, informal social media text can yield meaningful ideological estimates when modeled with a dynamic lexicon tailored to platform language.
• Estimated positions align with established ideology measures across contexts and languages.
• Ideological estimates contain derivative signal strong enough to predict individual voting intentions out-of-sample.
• Properly modeled social media data can capture derivative signal more effectively than discrete scales used in many traditional surveys.

❗Why It Matters

This method opens a new avenue for measuring political ideology at scale and at the individual level using digital trace data. It enables comparative, multilingual analysis of ideology across elites and citizens and improves the capacity to link language use to political behavior.