Which Fidelity? Evidence on Improving Synthetic Participants: replication materials.

Replication materials for Hoefer, Zhang and Krastev (2026), “Which Fidelity? Evidence on Improving Synthetic Participants,” a cross-study analysis of methods that make large language models’ simulated survey and experimental responses match human data. The deposit documents how the evidence base was assembled and coded. It contains the raw OpenAlex and arXiv exports from twelve logged keyword queries (1,853 hits, 1,086 unique records), the decision and reason for every record at each screening stage (title, abstract and full text, yielding 116 method tests), the first author’s independent coding of a random sample of 100 screening decisions (95% agreement with Claude Opus 5.5, Cohen’s κ = 0.89), and the verdict for each of the 66 method-by-target cells drawn in the paper’s Figure 4. Searching and screening were done with the assistance of Claude Opus 5.5 (Anthropic) and audited as described in the README, which also gives the protocol, the queries and a codebook for every file. Licensed CC BY 4.0.