AI evidence extraction

A Multidimensional Psychometric Scale for Measuring AI Dependency in Academic Research

Mugaahed Abdu Kaid Saleh, Abdellatif Sellami · International Journal of Human-Computer Interaction · 2026

AI-generated evidence extraction, verified across multiple analytical personas. Not a substitute for the peer-reviewed original.

9/10

Relevance

1/4

Quality (LMQS)

Evidence

Citations

0.00

FWCI

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.1080/10447318.2026.2629522

Methodology & findings

Study design

Mixed-methods scale development study across five sequential phases: Phase 1 (qualitative): focus group discussions (8-10 participants per session) and literature review with thematic analysis (Cohen's Kappa=0.82); Phase 2 (N=248): exploratory factor analysis (PCA with varimax rotation); Phase 3 (N=219): confirmatory factor analysis using SmartPLS 4.0 with hierarchical component modeling; Phase 4 (N=200): structural equation modeling to assess nomological validity; Phase 5 (N=588): cross-sample validation using CFA.

Sample

N = 1255, 5 groups

Primary method

Exploratory Factor Analysis (PCA with varimax rotation, Kaiser normalization using SPSS v.29); Kaiser-Meyer-Olkin (KMO) test and Bartlett's test of sphericity; Confirmatory Factor Analysis (CFA) using SmartPLS 4.0 with Type 1 hierarchical component modeling; Harman's single-factor test and marker variable analysis for common method variance assessment; Confirmatory Tetrad Analysis (CTA-PLS); Heterotrait-Monotrait (HTMT) ratio for discriminant validity; Structural Equation Modeling (SEM) for nomological validity testing; Cronbach's alpha and composite reliability for internal consistency; Average Variance Extracted (AVE) for convergent validity; Content Validity Index (CVI) computation; Modified Kappa (κ*) for expert agreement adjustment; Thematic analysis for qualitative focus group data; Cohen's Kappa for inter-coder reliability.

Main result

The study developed and validated a 15-item multidimensional AI Dependency Scale across five phases. "The three dimensions of cognitive reliance, emotional reliance, and behavioral reliance collectively account for AI dependency among academic researchers. Through exploratory factor analysis, we identified three distinct factors representing cognitive, emotional, and habitual behaviors associated with AI use. Confirmatory factor analysis further validated the scale's robustness and reliability." The scale demonstrated strong psychometric properties, with overall reliability of 0.86 and factor-specific reliabilities ranging from 0.73 to 0.90. Cross-sample validation in Phase 5 (N=588) confirmed generalizability with excellent model fit (SRMR=0.044). Notably, "The analysis revealed a statistically significant relationship between the two constructs (b ¼ 0.488, p < 0.001)" demonstrating that AI dependency is positively associated with attitudes toward plagiarism.

Reports effect sizes.

Research paradigm

Mixed-methods (qualitative + quantitative positivism)

Author conclusions

"This study introduces a validated instrument to measure AI dependency in academic research environments, conceptualized through cognitive, emotional, and behavioral dimensions. By integrating theoretical frameworks from psychological and educational research, the AI Dependency Scale advances current understanding of how scholars engage with generative AI tools. The findings not only demonstrate the scale's psychometric robustness but also highlight its relevance in predicting ethical vulnerabilities, particularly permissive attitudes toward plagiarism. In doing so, this research provides a critical resource for future inquiry into technology-mediated academic behavior and offers a foundation for institutional policies that promote ethical and balanced AI usage."

Risk of bias

Selection bias: Stratified purposive sampling may not represent all academic researchers; Self-report bias: Paper-based questionnaires relying on self-reported AI dependency; Common method variance: Addressed through Harman's single-factor test and marker variable analysis; Attrition: Overall response rate of 80.96% across phases, ranging from 76.3% to 86.9%; Temporal bias: Cross-sectional design prevents causal inference; Selection bias: purposive sampling with stratification may not represent all doctoral researchers; Self-report bias: participants self-reported AI dependency and plagiarism attitudes; Common method variance (CMV) addressed through Harman's single-factor test and marker variable analysis, but residual CMV risk remains; Geographic limitation: Indian higher education context may not generalize globally; Attrition: response rates ranged from 76.3% to 86.9% across phases, suggesting potential non-response bias; Selection bias: Stratified purposive sampling may not be fully representative of all researcher populations; Self-report bias: All data collected via self-administered questionnaires, no behavioral measures; Attrition: Response rates varied across phases (76.3%-86.9%), though overall completion rate was 80.96%; Phase-specific samples: Each phase involved independent samples; potential for sampling variation; Recruitment method: On-site recruitment during data collection may introduce selection bias favoring accessible participants; Common method variance: While Harman's single-factor test and marker variable analysis were conducted to address this, reliance on single measurement method remains a limitation

Limitations

The authors state several limitations: "First, the scale was validated with academic researchers, which was appropriate given the study's focus on scholarly engagement with generative AI
However, the generalizability of findings to other academic populations, such as undergraduate and postgraduate students, remains limited." Additionally, "the scale captures psychological and behavioral dimensions of dependency, it does not incorporate dispositional traits such as impulsivity or neuroticism, which are emphasized in the I-PACE model." Furthermore, "the study was conducted within the Indian higher-education context, which may limit the immediate generalizability of the findings."

Open questions raised

Generalizability to undergraduate and postgraduate students beyond doctoral researchers
Integration of dispositional personality traits (impulsivity, neuroticism) from the I-PACE model
Expansion beyond plagiarism to other forms of academic misconduct (unauthorized collaboration, ghost authorship, content fabrication)
Cross-cultural and multi-institutional validation across diverse international samples
Generalizability to undergraduate and postgraduate students beyond PhD researchers
Incorporation of dispositional traits such as impulsivity or neuroticism emphasized in the I-PACE model

Extracted from: pdfAgreement 60%

Explore related topics

Responsible AI Use Research Integrity Mixed Methods