In-Context Learning for the Imputation of Public Opinion Data with Large Language Models

2026-06-08

AI Summary

This study addresses the challenge of missing values in public opinion surveys caused by nonresponse by introducing in-context learning (ICL) with large language models (LLMs) to survey data imputation—a novel application in this domain. Leveraging 150 opinion variables from 15 waves of the American Trends Panel, the authors systematically evaluate multiple ICL strategies under missing completely at random (MCAR), missing at random (MAR), and missing not at random (MNAR) mechanisms. Results demonstrate that the proposed approach significantly reduces absolute imputation error across all missingness mechanisms, with the greatest improvement observed under MNAR. The optimal ICL configuration achieves nearly 95% overall confidence interval coverage, with interval widths only one-half to one-fifth those of the conventional MICE-PMM method, substantially outperforming existing imputation techniques.

Abstract

Large language models have been widely evaluated as simulators of individual survey responses. In practice, however, fully unobserved responses are rare; the dominant problem is partial non-response. Imputation aims to restore the overall structure of a survey dataset by filling in these missing values. It has its own well-defined evaluation criteria and differs fundamentally from prediction. We propose to impute missing survey data through in-context learning (ICL). We systematically evaluate ICL design choices across different missingness mechanisms (MCAR, MAR, MNAR) on 150 opinion variables spanning 15 waves of the American Trends Panel. Compared to well-established statistical methods for data imputation like MICE PMM, our ICL approach consistently reduces absolute error across all missingness mechanisms, with the largest gains under non-random missingness (MNAR). Notably, the best-performing specification (gpt-oss-120b with 100 in-context examples) achieves near-nominal aggregate coverage (approaching the 95% level) with confidence intervals two to five times narrower than MICE PMM. We publish a Python package with an sklearn-like API to enable easy deployment of our method using local and proprietary LLMs.

Problem

• imputation
• partial non-response
• public opinion data
• missing data
• survey data

Innovation

• In-Context Learning
• Data Imputation
• Large Language Models
• Survey Non-response
• Missing Data Mechanisms