Skip to content

AI Readiness Scale (AIRS): Validated 12-item instrument. 7-phase psychometric validation (N=362): EFA→CFA→Invariance→SEM→Mediation→Moderation→Behavioral. Autonomy-centered UTAUT2 extension (R²=.819). Reproducible Jupyter analysis, intervention protocols, practitioner guidelines.

License

Notifications You must be signed in to change notification settings

fabioc-aloha/AIRS_Data_Analysis

Repository files navigation

AIRS Data Analysis Project

Artificial Intelligence Readiness Scale (AIRS)

Extending Model for Enterprise AI Adoption

ResearcherFabio Correa
InstitutionTouro University Worldwide
ProgramDoctor of Business Administration (DBA)
UpdatedDecember 30, 2025

📝 View Current Thesis Draft (PDF) | Defense Presentation | Chair approved - Defense scheduled


Project Status

Component Status Progress
Analysis Pipeline ✅ Complete 11/11 notebooks verified
Thesis Chapter 1 ✅ Complete Introduction (~2,800 words)
Thesis Chapter 2 ✅ Complete Literature Review (~5,200 words)
Thesis Chapter 3 ✅ Complete Methodology (~3,800 words)
Thesis Chapter 4 ✅ Complete Results (~3,200 words)
Thesis Chapter 5 ✅ Complete Analysis & Discussion (~4,100 words)
Thesis Chapter 6 ✅ Complete Conclusions (~2,500 words)
Bibliography ✅ Complete 93 references verified
Tables & Figures ✅ Complete 61 tables + 15 figures, all LaTeX formatted
IRB Approval ✅ Complete Letter embedded
Chair Review ✅ Approved Defense scheduled

DBA Thesis Positioning

This research bridges academic rigor with business impact:

Academic Contribution Business Contribution
Extends UTAUT2 with AI-specific constructs 16-item diagnostic tool for AI readiness
Rigorous EFA→CFA→SEM validation Intervention targeting for 4 user segments
Novel moderator discovery (Experience) Tailored training by experience level
Cross-population validation Leaders as AI champions (d=0.74-1.14)
Empirical gap closure Links to McKinsey/BCG adoption-value gap

Executive Summary

This research develops and validates the AI Readiness Scale (AIRS), a psychometrically sound instrument extending UTAUT2 for enterprise AI tool adoption. Through rigorous split-sample validation (N=523), we demonstrate:

Key Findings

Finding Evidence
Validated 8-factor model 16 items, CFI=.975, TLI=.960, RMSEA=.065
Price Value dominates β=.505 (p<.001), strongest predictor
Experience moderates adoption HM×Exp (p=.007) significant
Four user typologies identified Enthusiasts 16%, Cautious 30%, Moderate 37%, Anxious 17%
Leaders dominate tool usage d=0.74-1.14 across all AI tools
Model explains 85.2% variance R²=.852 in Behavioral Intention (8-factor diagnostic model)

Practical Impact

Organizations now have a validated 16-item diagnostic instrument for assessing AI adoption readiness with:

  • Theoretical grounding in UTAUT2 framework
  • Empirical validation across student and professional populations
  • Four distinct user segments for targeted intervention design

Sample Overview

Parameter Value
Full Sample N = 523
EFA Development n = 261 (50%)
CFA Holdout n = 262 (50%)
Random Seed 67

Role Distribution

Role n %
Student 216 41.3%
Professional 184 35.2%
Leader 123 23.5%

Validated Measurement Model

Model: 8 Factors, 16 Items

flowchart LR
    subgraph UTAUT2["UTAUT2 Core"]
        PE[Performance Expectancy]
        EE[Effort Expectancy]
        SI[Social Influence]
        FC[Facilitating Conditions]
        HM[Hedonic Motivation]
        PV[Price Value]
        HB[Habit]
    end

    subgraph Extensions["AI Extension"]
        TR[AI Trust]
    end

    subgraph Outcome["Outcome"]
        BI[Behavioral Intention]
    end

    PE -->|β=-.028| BI
    EE -->|β=-.008| BI
    SI -->|β=.136*| BI
    FC -->|β=.059| BI
    HM -->|β=.217*| BI
    PV -->|β=.505***| BI
    HB -->|β=.023| BI
    TR -->|β=.106†| BI
Loading

Model Fit (CFA Holdout, n=262)

Index Value Threshold Interpretation
CFI .975 ≥.95 Excellent
TLI .960 ≥.95 Excellent
RMSEA .065 ≤.08 Good
SRMR .048 ≤.08 Excellent

Reliability

Factor α CR AVE
Performance Expectancy .803 .804 .673
Effort Expectancy .859 .861 .756
Social Influence .752 .763 .621
Facilitating Conditions .743 .750 .601
Hedonic Motivation .864 .865 .763
Price Value .883 .883 .790
Habit .909 .909 .833
AI Trust .891 .891 .804

Hypothesis Testing Results

ID Hypothesis Result Evidence
H1a PE → BI (+) ❌ Not Supported β=-.028, p=.791
H1b EE → BI (+) ❌ Not Supported β=-.008, p=.875
H1c SI → BI (+) ✅ Supported β=.136, p=.024
H1d FC → BI (+) ❌ Not Supported β=.059, p=.338
H1e HM → BI (+) ✅ Supported β=.217, p=.014
H1f PV → BI (+) ✅ Supported (Strongest) β=.505, p<.001
H1g HB → BI (+) ❌ Not Supported β=.023, p=.631
H2 TR → BI (+) ❌ Marginal β=.106, p=.064
H3 Experience moderates ⚠️ Partial HM×Exp p=.007
H4 BI predicts usage ✅ Supported ρ=.69, p<.001
H5 Role differs in usage ✅ Supported Leaders > Others

Analysis Pipeline

All notebooks in airs_experiment/ are complete and verified:

Phase Notebook Purpose Status
0 00_Create_Split_Samples Sample preparation
1 01_EFA_Experiment Factor exploration
2 02_CFA_Experiment Model validation
3 03_Measurement_Invariance Group equivalence
4 04_Structural_Model Hypothesis testing
5 05_Mediation_Analysis Indirect effects
6 06_Moderation_Analysis Interaction effects
7 07_Tool_Usage_Patterns Behavioral validation
8 08_Qualitative_Feedback Thematic analysis
9 09_Comprehensive_Review Gap analysis
10 10_Final_Synthesis Integration

Repository Structure

AIRS_Data_Analysis/
├── airs_experiment/          # Analysis notebooks and outputs
│   ├── data/                 # Processed datasets
│   ├── plots/                # Generated visualizations
│   ├── results/              # JSON output files
│   └── tables/               # CSV summary tables
├── thesis/                   # DBA thesis documentation
│   ├── chapters/             # Chapter files (1-6)
│   ├── tables/               # Publication-ready tables (9)
│   ├── figures/              # Exported figures
│   ├── references/           # Bibliography (93 refs)
│   ├── EXECUTIVE_SUMMARY.md  # Standalone executive summary
│   ├── PRACTITIONER_BRIEF.md # Standalone practitioner brief
│   ├── THESIS_SUMMARY.md     # 5-page thesis summary
│   ├── DEFENSE_PRESENTATION.md # Committee defense (28 slides + backup)
│   ├── PRESENTATION_SLIDES.md # Committee presentation draft
│   └── DRAFT 07.pdf          # Current compiled thesis
├── research/                 # Literature and industry sources
│   ├── AIRS literatyre review.md        # Theoretical foundation
│   ├── Bridging the AI Adoption Gap...  # Industry benchmarks (585 lines)
│   ├── Meta-Analysis of UTAUT...        # Blut et al. 2022 (84 pages)
│   └── Venkatesh_AOR_2021.md            # AI-UTAUT research agenda
├── plan/                     # Analysis planning documents
├── data/                     # Raw data files
├── docs/                     # Additional documentation
└── scripts/                  # Utility scripts

Thesis Documentation

Completed

Document Location Words
Chapter 1: Introduction thesis/chapters/01_introduction.md ~2,800
Chapter 2: Literature Review thesis/chapters/02_literature_review.md ~5,200
Chapter 3: Methodology thesis/chapters/03_methodology.md ~3,800
Chapter 4: Results thesis/chapters/04_results.md ~3,200
Chapter 5: Analysis & Discussion thesis/chapters/05_analysis_discussion.md ~4,100
Chapter 6: Conclusions thesis/chapters/06_conclusions.md ~2,500
Appendices A-H thesis/appendices.md Complete
Tables 4.1-4.9 thesis/tables/ 9 tables
Bibliography thesis/references/bibliography.bib 93 refs

Build & PDF Generation

See PDF_GENERATION.md for complete documentation on:

  • Full thesis build (thesis/build-thesis.ps1)
  • Standalone document conversion (thesis/convert-to-pdf.ps1)
  • Prerequisites and troubleshooting

Key Methodological Decisions Documented

Decision Rationale Location
4 constructs dropped (VO, EX, ER, AX-orig) Poor reliability (α = .301–.582) Ch 3 §3.4.4, Ch 4 §4.2.1
AIRS vs UTAUT2 comparison ΔR² = .016 (modest improvement) Ch 4 §4.3.5
H5 mediation untestable EX/ER excluded from model Ch 5 §5.6.2, §5.7.1

Quick Start

Requirements

pip install -r requirements.txt

Key dependencies: pandas, numpy, scipy, factor_analyzer, semopy, pingouin, scikit-learn, matplotlib, seaborn

Run Analysis

# Navigate to experiment folder
cd airs_experiment

# Run notebooks sequentially
# 00 → 01 → 02 → ... → 10

Use the AIRS-16 Scale

The validated 16-item scale measures 8 predictor constructs plus Behavioral Intention:

# Calculate construct scores (mean of 2 items each)
# 8 Predictor Constructs (16 items)
PE = mean(PE1, PE2)  # Performance Expectancy
EE = mean(EE1, EE2)  # Effort Expectancy
SI = mean(SI1, SI2)  # Social Influence
FC = mean(FC1, FC2)  # Facilitating Conditions
HM = mean(HM1, HM2)  # Hedonic Motivation
PV = mean(PV1, PV2)  # Price Value (STRONGEST predictor β=.505)
HB = mean(HB1, HB2)  # Habit
TR = mean(TR1, TR2)  # AI Trust

# Outcome Variable (4 items)
BI = mean(BI1, BI2, BI3, BI4)  # Behavioral Intention

Key References

Technology Acceptance

  • Venkatesh, V., et al. (2003). User acceptance of information technology. MIS Quarterly, 27(3), 425-478.
  • Venkatesh, V., et al. (2012). Consumer acceptance and use of information technology: Extending UTAUT. MIS Quarterly, 36(1), 157-178.

Scale Development

  • DeVellis, R. F., & Thorpe, C. T. (2021). Scale development: Theory and applications (5th ed.). Sage.
  • Hair, J. F., et al. (2019). Multivariate data analysis (8th ed.). Cengage.

AI Adoption

  • Shin, D. (2021). The effects of explainability and causability on perception, trust, and acceptance. IJHCS, 146, 102551.

Full bibliography: thesis/references/bibliography.bib


Citation

@phdthesis{correa2025airs,
  author = {Correa, Fabio},
  title = {Artificial Intelligence Readiness Scale: Extending Model for Enterprise AI Adoption},
  school = {Touro University Worldwide},
  year = {2025},
  type = {Doctoral dissertation}
}

License

Code: MIT License Documentation: CC BY 4.0


Last Updated: December 30, 2025 Version: 7.1 (Defense presentation prepared)

About

AI Readiness Scale (AIRS): Validated 12-item instrument. 7-phase psychometric validation (N=362): EFA→CFA→Invariance→SEM→Mediation→Moderation→Behavioral. Autonomy-centered UTAUT2 extension (R²=.819). Reproducible Jupyter analysis, intervention protocols, practitioner guidelines.

Topics

Resources

License

Code of conduct

Contributing

Stars

Watchers

Forks

Packages

No packages published

Languages