This repository contains the analysis materials for a published research study examining AI implementation barriers among K-12 educators in Jordan. Using person-centered analytical approaches, this study identifies distinct educator constraint profiles that challenge traditional assumptions about barrier accumulation in educational technology implementation.

Key Finding: Implementation barriers don't accumulate additively—educators facing cost constraints have 71% lower odds of simultaneously reporting skill deficits (φ=-.246, p<.001), revealing distinct subpopulations requiring differentiated support strategies.

This is Paper 1 of a three-paper research series examining AI adoption among K-12 educators in Jordan (N=189).

┌─────────────────────────────────────────────────────────────────────────────┐
│                         SAME SAMPLE: N=189 K-12 Educators                   │
└─────────────────────────────────────────────────────────────────────────────┘
                                      │
          ┌───────────────────────────┼───────────────────────────┐
          │                           │                           │
          ▼                           ▼                           ▼
   ┌─────────────┐           ┌─────────────┐           ┌─────────────┐
   │   PAPER 1   │           │   PAPER 2   │           │   PAPER 3   │
   │  (This Repo)│           │             │           │             │
   ├─────────────┤           ├─────────────┤           ├─────────────┤
   │ WHO faces   │           │ WHAT drives │           │ WHY do some │
   │ which       │──────────►│ adoption?   │──────────►│ advocate?   │
   │ barriers?   │           │             │           │             │
   ├─────────────┤           ├─────────────┤           ├─────────────┤
   │ • 8 cluster │           │ • Perception│           │ • 77.7%     │
   │   profiles  │           │   dominance │           │   mediated  │
   │ • Cost-Skill│           │ • 92.3%     │           │ • Student   │
   │   inverse   │           │   accuracy  │           │   effects   │
   │   (φ=-.246) │           │             │           │   primary   │
   └─────────────┘           └─────────────┘           └─────────────┘

Title: Hierarchical Cluster Analysis of AI Implementation Constraints: Uncovering Heterogeneous Barrier Patterns in Education

Author: Anfal Rababah

DOI: 10.5281/zenodo.17519099

Citation:

@article{rababah2025cluster,
  title={Hierarchical Cluster Analysis of AI Implementation Constraints: 
         Uncovering Heterogeneous Barrier Patterns in Education},
  author={Rababah, Anfal},
  journal={Zenodo},
  year={2025},
  doi={10.5281/zenodo.17519099}
}

1. What are the prevalence rates and demographic distributions of AI integration among K-12 educators?
2. How do implementation constraints (cost, availability, skill, language) distribute across demographic subgroups?
3. What pairwise co-occurrence patterns exist among constraints?
4. What distinct educator subgroups emerge from hierarchical cluster analysis of constraint profiles?

• N = 189 K-12 educators (3 excluded from initial 192 for logical inconsistencies)
• 32 schools randomly selected from 72 total schools in northern Jordan
• Stratified sampling by: geographic location, school level, gender composition, sector

• Face-to-face paper surveys (April–August 2024)
• Survey administered in Arabic, translated to English for analysis
• Open-ended responses coded into binary barrier indicators

Demographics (Independent):

• Gender (male/female)
• Teaching subject (scientific/non-scientific)
• Educational sector (public/private)
• Teaching experience (4 categories: ≤5, 6-10, 11-15, ≥16 years)

Outcomes:

• AI adoption (binary)
• AI usage frequency (5-level ordinal)
• Barrier types (4 binary indicators)

• 84.7% overall adoption rate
• Teaching subject was the only significant predictor (scientific: 92.2% vs non-scientific: 80.8%, p=.040)

Significant inverse relationship: Cost × Skill (φ=-.246, p<.001, OR=0.29)

• Educators with cost barriers had 71% lower odds of reporting skill deficits
• Suggests distinct constraint profiles rather than additive barrier accumulation

Eight distinct profiles identified (R²=0.778, silhouette=0.650):

ai-barriers-cluster-analysis/
│
├── README.md                    # This file
├── LICENSE                      # MIT License
│
├── data/
│   ├── data_dictionary.md       # Variable definitions and coding
│   └── survey_instrument.md     # Original survey (Arabic + English)
│
├── analysis/
│   ├── jasp/
│   │   └── analysis_workflow.md # JASP analysis procedures
│   └── results/
│       └── summary_tables.md    # Key statistical outputs
│
├── visualizations/
│   ├── scripts/
│   │   ├── cluster_profiles.py  # Cluster heatmap visualization
│   │   ├── cooccurrence_network.py  # Network diagram
│   │   └── frequency_charts.py  # Bar charts and distributions
│   └── figures/
│       └── [generated figures]
│
├── paper/
│   └── Rababah-2025-ClusterAnalysis-AI-Barriers.pdf
│
└── requirements.txt             # Python dependencies

All statistical analyses were conducted in JASP:

1. Descriptive Statistics: Frequencies module for prevalence rates
2. Chi-Square Tests: Frequencies → Contingency Tables
3. Non-parametric Tests:
   • Mann-Whitney U for binary group comparisons
   • Kruskal-Wallis for multi-group comparisons
4. Cluster Analysis:
   • Machine Learning → Clustering → Hierarchical
   • McQuitty linkage method
   • Euclidean distance metric
   • BIC optimization for cluster selection

Python was used exclusively for creating publication-quality figures:

# Example: Cluster profile heatmap
import matplotlib.pyplot as plt
import seaborn as sns
import pandas as pd

# Cluster z-scores from JASP output
cluster_data = pd.DataFrame({
    'Cluster': [1, 2, 3, 4, 5, 6, 7, 8],
    'Cost': [-0.61, 1.62, -0.61, 1.62, 1.62, 1.62, -0.61, -0.61],
    'Availability': [-0.54, -0.54, -0.54, 1.84, -0.54, 1.84, 1.84, 1.84],
    'Skill': [0.27, -0.48, 0.43, -0.24, -0.24, 0.28, -0.29, 0.28],
    'Language': [-0.29, -0.29, 3.40, -0.29, 3.40, 3.40, -0.29, 3.40]
})

This project showcases:

• Survey Research Design: Stratified sampling, instrument development, cross-cultural translation
• Statistical Analysis: Chi-square, non-parametric tests, effect sizes (φ, Cramér's V, odds ratios)
• Machine Learning: Unsupervised clustering, model validation (silhouette, BIC)
• Data Visualization: Network diagrams, heatmaps, publication-quality figures
• Research Communication: Academic writing, statistical reporting (APA style)

The findings support differentiated intervention strategies:

• Cross-sectional design (single timepoint)
• Self-reported measures
• Geographic concentration (single Jordanian city)
• Small sample sizes in some clusters (n=2-4)

Anfal Rababah
Independent Researcher, Jordan
📧 [email protected]
🆔 ORCID: 0009-0003-7450-8907

This project is licensed under the MIT License - see the LICENSE file for details.

• K-12 educators who participated in the survey
• JASP development team for open-source statistical software