Note: When clicking on a Digital Object Identifier (DOI) number, you will be taken to an external site maintained by the publisher.
Some full text articles may not yet be available without a charge during the embargo (administrative interval).
What is a DOI Number?
Some links on this page may take you to non-federal websites. Their policies may differ from this site.
-
Free, publicly-accessible full text available August 1, 2027
-
Free, publicly-accessible full text available June 1, 2027
-
Free, publicly-accessible full text available December 1, 2026
-
Free, publicly-accessible full text available December 11, 2026
-
Survey research faces mounting structural challenges: declining response rates, sample bias, block-wise missingness among at-risk respondents, and AI-assisted fraudulent completions in online panels. Large language models (LLMs) offer potential remedies, yet rigorous evaluations across the full survey workflow remain scarce, particularly in disaster contexts where data quality matters most. We present a five-stage framework for LLM integration covering questionnaire design, sample selection, pilot testing, missing-data imputation, and post-collection analysis, evaluated on the 2024 Hurricane Milton preparedness survey of Florida residents (n = 946). We introduce a Protection Motivation Theory (PMT)-constrained co-occurrence knowledge graph and benchmark seven LLM configurations against three classical imputation baselines. Four findings stand out. First, organizing retrieval around PMT causal structure and integrating all evidence in a single model call outperforms both unstructured retrieval and staged sequential inference on response prediction (mean absolute error 0.993 vs. 1.097 for standard retrieval-augmented generation). Second, our proposed Anchored Marginal Theory-Informed LLM (A-TLM) outperforms all classical baselines on root-mean-square error under disaster-relevant block-wise missing-not-at-random conditions (RMSE 1.439 vs. 1.496 for the next-best method) and achieves near-zero signed bias (-0.121) where the random-forest imputer produces the largest absolute bias (-0.631). Third, near-zero aggregate bias can mask opposing subgroup errors of substantial magnitude, with at-risk respondents systematically under-predicted; subgroup-stratified auditing is proposed as a minimum reporting standard. Fourth, hallucination risk in LLM-based survey analysis is architecturally manageable through grounded refusal. Together, these findings provide actionable guidance for disaster researchers seeking to integrate AI tools responsibly into survey workflows under urgent conditions.more » « lessFree, publicly-accessible full text available January 1, 2027
-
Free, publicly-accessible full text available December 1, 2026
An official website of the United States government
