About my research

My research develops and applies new statistical and causal inference methods for social science in the era of artificial intelligence (AI). Alongside advancing methodological foundations, I use these tools to study substantive questions about human–AI interactions and the broader implications of AI and algorithms for society. My current work centers on the following interconnected research programs.

Causal and Statistical Inference for Texts, Images, and Videos Using GenAI

Unstructured data, such as texts, images, and videos, are central to modern communication. However, their statistical analysis poses substantial challenges because they contain many intertwined attributes whose effects are difficult to isolate. In this research program, I develop frameworks that use the internal representations of generative AI models to learn latent structures and conduct valid causal and statistical inference for many different settings.

1. A formal framework for using GenAI in causal and statistical inference: Imai and Nakamura (2026a, 2026b)
2. Dynamic causal inference for unstructured data: Nakamura and Imai (2026+); Nakamura, Breuer, Crespin, Dietrich, and Imai (2026+)
3. Software (gpi-pack): Nakamura and Imai (2025, maintainer)

Valid Measurement and Statistical Inference with Text, Image, and Video Annotations Under Challenging Circumstances

Large language models (LLMs) and other AI systems are increasingly used as annotators, but their predictions inevitably contain measurement error, disagreement, and systematic biases. In this research program, I develop statistical frameworks that enable valid measurement and downstream inference in challenging settings, including those in which annotation quality is low, only indirect validation is feasible, or no ground-truth labels are observed. The goal is to make AI-assisted measurement and data collection both statistically principled and practically useful.

1. General strategies and recommendations for AI-assisted annotation: Nakamura, Tan, and Yean (2026+)
2. Efficient statistical inference with low-quality and noisy annotations: Nakamura (2025+)
3. Downstream analysis when only indirect validation is available: Nakamura and Egami (2026+)

AI for Social Science

Human–AI interaction is becoming an increasingly important subject of social science research. In this research program, I develop methodological frameworks for studying how people engage with AI systems and how these interactions shape social and political outcomes. I also examine how AI-mediated interactions can be used to improve the practice of social science research.

1. Identification and estimation strategies for human–AI interactions: Nakamura (2026+)
2. Using AI interviewers to probe respondents’ understanding of survey questions: Hinck, King, Nakamura, and Stewart (2026+)