ESRA 2027 sessions by theme

Back to Overview of Sessions

Session Details

Widespread agreement on the utility of silicon samples and related methods (digital twins, synthetic respondents, etc.) for survey research remains elusive, despite exponential growth in research on the topic.

This session brings together perspectives on the topic from both academia and industry, with the aim of identifying a current consensus view, while highlighting areas of continued disagreement.

Preferred topics for this session would use LLM generated synthetic data to:

- Improve estimation or prediction efficiency—e.g., using Prediction Powered Inference, etc. (c.f., Hullman, Broska, et al., 2025)
- Reduce bias in survey estimates to obtain more accurate point estimates for general populations or sub-populations

Both failures and successes in using silicon samples in one or both of these contexts are welcome contributions, as are theoretical perspectives grounded in empirical evidence and simulations.

Within the session, the organizers will provide a topic framing given the specific presentations, and will synthesize those presentations into a proposed point-in-time consensus for further group discussion.