This repository contains the statistical analysis workflows, visualization code, and supplementary materials for Paper 3 of a three-paper research series examining artificial intelligence adoption among K-12 educators in Jordan.

Research Question: What factors predict educators' willingness to recommend AI tools to colleagues, and do perceptual beliefs mediate the relationship between AI use and endorsement intentions?

Despite 84.7% of educators adopting AI tools, endorsement intentions varied substantially:

Key Insight: AI use showed no independent effect once perceptions were controlled—endorsement operates entirely through belief formation.

77.7% of the AI use → endorsement relationship was mediated through perceptions:

                    ┌─────────────────────────────┐
                    │   Perceived Student Effects │
         a₁=0.65*** │         (55.1%)             │ b₁=0.74***
              ┌─────┴─────────────────────────────┴─────┐
              │                                         │
              ▼                                         ▼
┌──────────┐      ┌─────────────────────────┐      ┌────────────┐
│ AI Tool  │      │   Teacher Work Effects  │      │Endorsement │
│   Use    │─────►│         (14.4%)         │─────►│ Intentions │
└──────────┘      └─────────────────────────┘      └────────────┘
              │   a₂=0.31**            b₂=0.41***  │
              │                                    │
              │   ┌─────────────────────────┐      │
              └──►│  Efficiency Comparisons │──────┘
         a₃=0.28* │         (8.3%)          │ b₃=0.26**
                  └─────────────────────────┘
                           
                  Direct Effect: c' = 0.19 (ns)
                  Total Effect: c = 0.87***

Implementation barriers showed counterintuitive positive effects on endorsement:

Interpretation: Educators who overcome barriers become stronger advocates—likely due to cognitive dissonance (effort justification) and enhanced self-efficacy.

• Sample: N = 189 K-12 educators from 32 schools in Jordan
• Design: Cross-sectional survey with ordinal outcome
• Data Collection: April–August 2024 via face-to-face paper survey

Ordinal Regression:

logit[P(Y ≤ j)] = αⱼ - (β₁·AI_Use + β₂·Student_Effect + β₃·Work_Effect + 
                        β₄·Efficiency + β₅·Barrier_Count + covariates)

Mediation Model:

• X (Predictor): AI Use Status (binary)
• M (Mediators): Student Effects, Work Effects, Efficiency Comparisons
• Y (Outcome): Endorsement Intentions (ordinal, treated as continuous for mediation)
• Covariates: Gender, subject area, sector, √experience

ordinal-mediation-advocacy/
├── README.md                          # This file
├── LICENSE                            # MIT License
├── .gitignore                         # Git ignore patterns
├── requirements.txt                   # Python dependencies
├── GITHUB_SETTINGS.md                 # Repository configuration guide
│
├── data/
│   ├── data_dictionary.md             # Variable definitions and coding
│   └── survey_instrument.md           # Original survey (Arabic & English)
│
├── analysis/
│   ├── jasp/
│   │   ├── ordinal_regression_workflow.md    # Step-by-step JASP guide
│   │   └── mediation_analysis_workflow.md    # Mediation in Python
│   └── results/
│       └── summary_tables.md          # All statistical tables
│
├── visualizations/
│   └── scripts/
│       ├── mediation_path_diagram.py         # Figure 1
│       ├── effect_decomposition_bar.py       # Figure 2
│       ├── predicted_probabilities.py        # Figure 3
│       └── barrier_effects_comparison.py     # Figure 4
│
└── paper/
    └── [Published PDF - see DOI]

# Python environment
pip install -r requirements.txt

# JASP (download from https://jasp-stats.org/)
# Version 0.19.3 or later recommended

1. Prepare Data

   • Review data/data_dictionary.md for variable coding
   • Ensure variables match specifications

2. Run Bivariate Analyses (JASP)

   • Follow analysis/jasp/ordinal_regression_workflow.md
   • Generate chi-square tests and effect sizes

3. Fit Ordinal Regression (JASP)

   • Build nested models (Demographics → Constraints → Perceptions → Full)
   • Test proportional odds assumption (Brant test)

4. Conduct Mediation Analysis (Python)

   • Follow analysis/jasp/mediation_analysis_workflow.md
   • Run bootstrap resampling (5,000 iterations)

5. Generate Visualizations (Python)

   cd visualizations/scripts
   python mediation_path_diagram.py
   python effect_decomposition_bar.py
   python predicted_probabilities.py
   python barrier_effects_comparison.py

1. Lead with student outcomes — Evidence of student learning gains (55% of advocacy effect) is far more persuasive than productivity benefits
2. Create peer testimonial programs — Leverage educators who overcame barriers as powerful advocates
3. Move beyond technical training — Pedagogical conviction, not tool proficiency, drives institutional diffusion

1. Invest in demonstrable evidence — Fund pilots that generate measurable student outcome data
2. Address access equitably — Availability barriers paradoxically strengthen advocacy among successful adopters
3. Prioritize multilingual tools — Language-barrier educators show unique advocacy pathways

@article{rababah2025advocacy,
  title={Analyzing the Gap Between AI Adoption and Advocacy: An Ordinal 
         Regression and Mediation Analysis Among K-12 Educators},
  author={Rababah, Anfal},
  journal={Zenodo},
  year={2025},
  doi={10.5281/zenodo.17519209}
}

Anfal Rababah
Independent Researcher, Jordan
📧 Anfal0Rababah@email.com
🔗 ORCID: 0009-0003-7450-8907

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

• Statistical analyses conducted using JASP (open-source)
• Visualizations created with Python's matplotlib and seaborn
• Mediation analysis implemented following Hayes' PROCESS methodology
• AI language models (Grammarly, Claude Sonnet) assisted with manuscript preparation