Improving Cross-Cultural Survey Simulation with Calibrated Value Personas

2026-05-15

AI Summary

This study addresses the challenge that large language models struggle to accurately replicate the true opinion distributions of populations in underrepresented countries during cross-cultural survey simulations. The authors propose a novel personality construction approach grounded in cultural values: textual descriptors are extracted from survey data reflecting core cultural dimensions, and personas are sampled according to the target population’s value distributions. A calibration mechanism is further introduced to enhance response diversity and opinion fidelity. Innovatively integrating cultural values directly into prompt engineering—replacing conventional proxies based on sociodemographic or personality traits—the method significantly reduces prediction errors across multiple countries. Notably, it improves simulation accuracy for low-resource nations, narrows the performance gap with dominant cultural contexts, and yields response distributions closely aligned with those of human respondents.

Abstract

Large language models (LLMs) are increasingly used to simulate human opinions and survey responses, but their ability to reproduce population responses across cultures remains limited. Existing persona-based prompting methods typically rely on sociodemographic or personality traits, which are only indirect proxies for the values that shape human responses. We propose a value-based persona construction method that derives textual descriptors from survey responses capturing core cultural dimensions. By sampling value profiles from target populations and aggregating LLM responses across personas, we obtain population-level predictions grounded in observed value distributions. We further introduce a calibration procedure that improves response diversity while preserving estimated opinions. We show that our approach reduces prediction error across countries, with the largest improvements observed in underrepresented populations. This substantially narrows the performance gap between countries aligned with dominant LLM priors and those that are less represented in training data, while also yielding response distributions that closely match human diversity.

Problem

• cross-cultural survey simulation
• large language models
• value-based personas
• population-level prediction
• cultural bias

Innovation

• value-based personas
• cross-cultural survey simulation
• LLM calibration
• population-level prediction
• response diversity