Leveraging Large Language Models for Survey Research

Project Directors Dr. Alexander Wenz MZES-funded 2024 – 2027

Research question/goal:

The recent advances and availability of large language models (LLMs) have created new opportunities for research in the social and behavioural sciences. This project investigates to what extent LLMs can be used to facilitate two aspects of survey data collection: (1) questionnaire development and evaluation and (2) survey interviewing. First, we examine to what extent LLMs can identify potential problems in survey questions compared to existing evaluation methods. Second, we assess to what extent an LLM-based chatbot can collect high quality survey data compared to a standard web survey. Third, we investigate whether the LLMs’ performance can be improved with prompt-engineering and fine-tuning.

Current stage:

Pilot work for the project was presented at international conferences and in a colloquium at the Institute for Employment Research. A first article, examining biases in large language models, has appeared in Social Science Computer Review, and a second is currently under review.


Publications

Journal Articles

  • Von der Heyde, Leah, Anna-Carolina Haensch, Alexander Wenz (2025): Vox Populi, Vox AI? Using Large Language Models to Estimate German Vote Choice. Social Science Computer Review, tba, tba, 1-23. More

Presentations

  • Haensch, Anna-Carolina, Leah Von der Heyde, Alexander Wenz, Bolei Ma (2025): United in Diversity? Contextual Biases in LLM-Based Predictions of the 2024 European Parliament Elections. [11th Conference of the European Survey Research Association (ESRA), Utrecht, 14/07/2025 - 18/07/2025]. More
  • Haensch, Anna-Carolina, Leah Von der Heyde, Alexander Wenz, Bolei Ma (2025): United in Diversity? Contextual Biases in LLM-Based Predictions of the 2024 European Parliament Elections. [AAPOR 80th Annual Conference, St. Louis, 14/05/2025 - 16/05/2025]. More
  • Haensch, Anna-Carolina, Leah Von der Heyde, Alexander Wenz (2024): Can large language models predict how people vote? Evidence from Germany. [10th International Conference on Computational Social Science (IC2S2), Philadelphia, PA, 17/07/2024 - 20/07/2024]. More
  • Von der Heyde, Leah, Anna-Carolina Haensch, Alexander Wenz (2024): Can large language models estimate how people vote? Evidence from Germany. [29. Wissenschaftlicher Kongress der Deutschen Vereinigung für Politikwissenschaft, Göttingen, 24/09/2024 - 27/09/2024]. More