Rectification Difficulty and Optimal Sample Allocation in LLM-Augmented Surveys

Zikun Ye, Hema Yoganarasimhan

Large Language Models can generate synthetic survey responses at low cost, but their accuracy varies unpredictably across questions. We study the design problem of allocating a fix…