Parametric Social Identity Injection and Diversification in Public Opinion Simulation

TL;DR

This paper introduces PSII, a framework that injects demographic attributes into LLMs to improve diversity and realism in public opinion simulations, addressing the limitations of current homogeneous responses.

Contribution

The paper proposes Parametric Social Identity Injection (PSII), a novel method for controlling social identities in LLMs at the representation level, enhancing diversity in opinion simulations.

Findings

PSII significantly reduces divergence from real survey data.

PSII increases diversity and inter-group differences in LLM responses.

The method improves the fidelity of public opinion models.

Abstract

Large language models (LLMs) have recently been adopted as synthetic agents for public opinion simulation, offering a promising alternative to costly and slow human surveys. Despite their scalability, current LLM-based simulation methods fail to capture social diversity, producing flattened inter-group differences and overly homogeneous responses within demographic groups. We identify this limitation as a Diversity Collapse phenomenon in LLM hidden representations, where distinct social identities become increasingly indistinguishable across layers. Motivated by this observation, we propose Parametric Social Identity Injection (PSII), a general framework that injects explicit, parametric representations of demographic attributes and value orientations directly into intermediate hidden states of LLMs. Unlike prompt-based persona conditioning, PSII enables fine-grained and controllable identity modulation at the representation level. Extensive experiments on the World Values Survey using multiple open-source LLMs show that PSII significantly improves distributional fidelity and diversity, reducing KL divergence to real-world survey data while enhancing overall diversity. This work provides new insights into representation-level control of LLM agents and advances scalable, diversity-aware public opinion simulation. Code and data are available at https://github.com/halsayxi/PSII.