Anonymised matched-panel dataset and full statistical audit trail for "Scaffolding and fading generative AI support shapes dependency and anxiety among undergraduate law students"

Published: 20 November 2026| Version 1 | DOI: 10.17632/gggvjcpnrz.1
Contributor:
RMM Pradeep

Description

Human-AI Interaction, Dependency, and Academic Anxiety in Legal Education Overview This dataset supports a five-month, single-cycle design-based research study investigating the impact of structured generative artificial intelligence (AI) integration on student self-efficacy, dependency, and academic anxiety. Collected within a first-year Bachelor of Laws (LLB) cohort at a Sri Lankan state university, the data capture student responses across three progressively structured human-AI interaction phases: M1: Unstructured use M2: AI as Reviewer M3: AI as Minimal Reviewer Dataset Structure & Contents The repository includes quantitative survey metrics and qualitative reflections: Cross-Sectional Samples: Likert-scale responses gathered at three chronological milestones ($n_1 = 126$, $n_2 = 107$, $n_3 = 122$). Longitudinal Panel: A matched dataset ($n = 51$) tracking individual trajectories across all phases. Qualitative Data: A collection of 355 open-ended student reflections providing contextual insight into behavioral and emotional shifts during AI scaffolding and support withdrawal. Variables & Methodology Core Metrics: Evaluates self-efficacy, perceived dependency, and academic anxiety using validated psychometric scales and custom instruments. Statistical Analysis Files: Includes scripts and processed outputs for Friedman, Wilcoxon signed-rank tests (longitudinal panel), and Kruskal-Wallis tests (full cross-sectional samples), alongside correlation analyses examining baseline dependency and final anxiety outcomes ($\rho = 0.38, p = 0.007$). Reuse Potential This dataset offers valuable empirical insights for researchers, educational technologists, and curriculum designers studying GenAI pedagogy, equity in digital learning, and the psychological impacts of scaffolding withdrawal in higher education.

Files

Steps to reproduce

1. Dataset Acquisition: Access the raw survey data (cross-sectional Likert-scale responses and matched longitudinal panel data) and qualitative open-ended reflection transcripts from the repository. 2. Data Preparation and Cleaning: Filter responses across the three distinct implementation phases: unstructured use (M1), AI as Reviewer (M2), and AI as Minimal Reviewer (M3). Handle missing entries for the longitudinal panel subset (n = 51). 3. Quantitative Analysis: Execute non-parametric statistical tests in software such as R or SPSS. Run Friedman and Wilcoxon signed-rank tests on the longitudinal panel data, and Kruskal-Wallis tests on the full cross-sectional samples (n_1 = 126, n_2 = 107, n_3 = 122) to evaluate changes in self-efficacy, academic anxiety, and perceived dependency. 4. Correlation Analysis: Perform Spearman rank-order correlation to evaluate the predictive relationship between baseline AI dependency and final anxiety outcomes (rho = 0.38, p = 0.007). 5. Qualitative Analysis: Code the 355 open-ended student reflections using thematic analysis to cross-validate quantitative trends regarding scaffolding withdrawal and emotional impacts.

Institutions

Categories

Artificial Intelligence, Education, Higher Education, Legal Studies

Licence