Search results

Showing 1 – 4 of 4 results.

Self-published

AI-Supported Inquiry, AI Literacy, and Authentic Performance Among Preservice Teachers in Central China, 2024 (ICPSR 306435)

Released/updated on: 2026-06-03

Geographic coverage: Jiangxi, China

Time period: 2024-09-01--2024-12-31

This study examined the effects of QUEST+AI, an AI-supported inquiry model, on AI literacy and authentic performance among preservice teachers in Central China. The study used a nonequivalent-groups, quasi-experimental pretest–posttest design with two intact sections of an undergraduate Educational Research Methods course at a public university. Ninety-five preservice teachers participated, including 52 students in the experimental group and 43 students in the comparison group. Both groups received the same face-to-face course instruction over a 10-week period. The experimental group completed two QUEST+AI inquiry cycles with coached use of generative artificial intelligence, while the comparison group completed conventional homework assignments.

The data include participant demographic variables, pretest and posttest responses to a multidimensional AI literacy questionnaire, AI literacy total and subscale scores, final research proposal scores, and group assignment indicators. AI literacy measures cover use and application of AI, knowledge and understanding of AI, AI detection, AI ethics, AI creation, AI-supported problem solving, AI persuasion literacy, and AI emotion regulation. Authentic performance is represented by scores on a final educational research proposal, evaluated with a common rubric by two independent raters. The dataset also includes variables used in the study’s comparative analyses, including group condition, pretest scores, gender, grade level, age, and major.

Self-published

PromptTensor Prompt Bank (v1.0.1) (ICPSR 246323)

Released/updated on: 2026-02-25

Time period: 2025-08-11--2026-02-04

PromptTensor Prompt Bank (v1.0.1) is an English prompt dataset for LLM research and prompt-engineering experiments. It contains structured prompts labeled by domain/subdomain/intent, provided in JSONL format for easy loading and filtering.

Mirrors/DOIs: Zenodo (doi:10.5281/zenodo.18665254). Additional mirrors: Hugging Face, Kaggle, and GitHub.

Canonical page: https://prompttensor.com/datasets/prompttensor-promptbank-v1

Self-published

Análisis cualitativo asistido por LLMs: Una metodología híbrida para el estudio territorial de la participación ciudadana (ICPSR 239202)

Released/updated on: 2025-10-26

Geographic coverage: Colombia

Time period: 2024-01-01--2024-12-31

Prompts para el artículo: "Análisis cualitativo asistido por LLMs: Una metodología híbrida para el estudio territorial de la participación ciudadana"Esta investigación desarrolla una metodología que integra modelos de lenguaje en análisis cualitativo, argumentando que es posible superar limitaciones de escalabilidad sin sacrificar rigor interpretativo. Aplicada al estudio de participación ciudadana en Colombia, combinó transcripción automática, análisis asistido por IA y validación humana. Los resultados mostraron alta eficiencia (99 entrevistas analizadas en una semana) manteniendo profundidad analítica para identificar patrones territoriales, confirmando el potencial de este enfoque híbrido para la investigación en ciencias sociales.

Self-published

Enhancing Quantitative Analysis in Social Sciences with Large Language Models (LLMs): A Methodological Case Study in Educational Research (ICPSR 237744)

Released/updated on: 2025-09-29

The objective of this paper is to explore the potential of Large Language Models (LLMs) for assisting with quantitative data analysis in social science research. Specifically, it aims to introduce key concepts to help researchers effectively integrate LLMs into their workflows. For this purpose, we replicate a research paper in educational leadership on the relationship between school program coherence and student achievement. By leveraging LLMs to generate code for statistical tools like Mplus and R, researchers can streamline their data analysis, potentially saving time and effort. The quality of analytical code generated by LLMs can be influenced by the researcher’s understanding and application of concepts like context windows, LLM training data and training cut-off, model parameter settings like temperature, zero- and few-shot learning, and Retrieval-Augmented Generation. By describing and demonstrating the applications of these concepts, we aim to equip researchers with a basic toolset to leverage LLMs effectively to assist with coding for quantitative analysis.