Early Career Best Paper Award for Eduardo Ryo Tamaki

Eduardo is a doctoral researcher at GIGA Hamburg and member of the graduate centre "Effective and Innovative Policymaking in Contested Contexts" (EIPCC) at the Willy Brandt School of Public Policy. He and his co-author, Levente Littvay of the Hungarian Academy of Sciences, received an award at the first MethodsNET conference for their work on using AI for data collection.

Their paper introduces Chrono-sampling, a novel generative AI framework that functions as a social sciences time machine that enables researchers to recover historical public opinion data by simulating survey respondents from the past. Rather than simply mimicking surface-level distributions, Chrono-sampling allows Large Language Models (LLMs) to replicate context-dependent relationships—how attitudes shift across economic recessions, political transitions, and sociopolitical climates.The method combines two key mechanisms:

• Time-gating, which constrains the model's knowledge to a specific year in history
• Clio Personas, richly constructed biographical prompts based on real human profiles, which condition the model to behave like a historically grounded individual

Using data from the Reagan (1980s), Bush (2000s), and Obama (2010s) eras, the study shows that LLMs can accurately reproduce temporal shifts in key relationships—such as how retrospective and prospective economic evaluations correlate differently during recessions versus recoveries, or under different partisan regimes. These replicated patterns align with real human data and persist across robustness checks using multiple Large Language Models and prompt variants. Unlike earlier approaches, which preserved only strong or static associations, Chrono-sampling captures how patterns evolve with context.By demonstrating that LLMs can model not just distributions, but temporally-contingent causal-like structures, Chrono-sampling offers a powerful tool for extending time series, simulating counterfactuals, and studying historical public opinion where no data exist. This opens new frontiers for computational social science—particularly in periods where traditional data collection was impossible.