Method
This chapter details the specific analytical methods applied in Phase 1 of the study: (1) Content Analysis using Hwang et al.’s coding framework, (2) Qualitative Comparative Analysis (QCA), (3) Structural Topic Modeling (STM), and (4) Network Analysis. These four methods triangulate findings from multiple analytical angles.
Overview of Phase 1 Methods
Phase 1 Methods
├── Content Analysis (Comparative Mapping)
│ Hwang et al. 4-dimension coding → cluster derivation
│ Country-by-country, then cross-national comparison
│
├── QCA (fsQCA)
│ Conditions → Outcome: Explicit CT Integration
│ Sufficient condition paths; equifinality across countries
│
├── Topic Modeling (STM)
│ Inductive discovery of latent themes in course text corpus
│ Country × program level differences
│
└── Network Analysis
Co-occurrence network of coding attributes
Cross-national structural comparison (QAP)Method 1: Content Analysis and Comparative Mapping
Coding Framework (4 Dimensions)
Hwang et al.’s (2025) coding framework is applied identically across all three countries to ensure comparability.
| Dimension | Codes | Description |
|---|---|---|
| Curricular Themes | S, E, D, En, L | Strategy, Ethics, Data/Analytics, Entrepreneurship/Innovation, Literacy |
| Pedagogical Approaches | C, S, B, P, L | Case-based, Simulation, Brainstorming/Debate, Project-based, Lecture/Lab |
| Assessment Modes | A, R, F, Q | AI-Assisted Projects, Reflective Critiques, Faculty-graded, Quizzes |
| AI–CT Relationship | CT Level (H/M/L) + CT Linkage (E/I/A) | Bloom’s level × Explicit/Implicit/Absent integration |
Cluster Classification
Based on CT Level × CT Linkage matrix, schools are assigned to one of three clusters:
| Cluster | Definition |
|---|---|
| Explicit CT Leaders | CT Level = High AND CT Linkage = Explicit |
| Implicit CT Integrators | CT Level = Mid AND CT Linkage = Implicit |
| Nominal CT Adopters | CT Level = Low AND CT Linkage = Absent |
Inter-Coder Reliability
- Minimum 2 coders per language (English/Chinese/Korean)
- Pilot coding: 5 schools per country before full coding
- Target: Cohen’s Kappa ≥ 0.80
- Disagreements resolved through discussion and guideline refinement
Comparative Mapping Output
- Country-specific cluster distributions
- Cross-national cluster comparison (US vs China vs Korea)
- Alluvial / Sankey diagram visualization of national flows
Method 2: QCA (Qualitative Comparative Analysis)
Basic QCA Structure
[Case] × [Conditions] → [Outcome]
Case = each school (or school–program combination)
Conditions = variables expected to influence the outcome
Outcome = Explicit CT Integration (CT Level = High AND CT Linkage = Explicit)Outcome Variable
Preferred: fsQCA (fuzzy-set)
| Score | Definition | Operationalization |
|---|---|---|
| 1.0 | Fully in | Higher CT + Explicit Linkage |
| 0.67 | More in than out | Higher CT + Implicit, or Mid CT + Explicit |
| 0.33 | More out than in | Mid CT + Implicit, or Lower CT + Explicit |
| 0.0 | Fully out | Lower CT + Absent Linkage |
Condition Variables (6 Conditions)
| Condition | Logic | Calibration (fsQCA) |
|---|---|---|
| COUNTRY (decomposed) | National AI policy + AACSB maturity | Government AI Readiness Index; accreditation years |
| PROGRAM | Program level | 1 = MBA; 0 = Undergraduate |
| ACTIVE_PEDAGOGY | Active (Case/Simulation/Debate) vs Passive (Lecture/Project) | 1.0 = 2+ active types; 0.67 = 1 active + 1 passive; 0.33 = passive only (2+); 0.0 = lecture only |
| REFLECTIVE_ASSESSMENT | Reflective (AI Projects/Critiques) vs Traditional (Graded/Quiz) | 1.0 = A+R both; 0.67 = A or R + traditional; 0.33 = graded only; 0.0 = quizzes only |
| THEME_DIVERSITY | Number of curricular themes integrating AI | 1.0 = 3+ themes; 0.67 = 2 themes; 0.33 = 1 non-literacy theme; 0.0 = literacy only |
| STRATEGY_INNOVATION | Presence of Strategy (S) or Entrepreneurship/Innovation (En) theme | 1 = S or En present; 0 = neither |
Unit of Analysis
Recommended: School-level (Option A)
- Each school = 1 case (~US 40 + China 25 + Korea 20 = ~85 cases)
- PROGRAM included as condition variable to capture UG/MBA differences
- Country-specific supplementary analyses as scope conditions if needed
QCA Procedure
Step 1: Calibration - Transform each condition variable to 0–1 fuzzy set - Set calibration anchors: fully in (1.0), crossover (0.5), fully out (0.0) - Check distribution (histogram) — re-adjust if extreme skew
Step 2: Necessity Analysis - Test whether each condition is a necessary condition for the outcome - Consistency ≥ 0.90 → necessary condition
Step 3: Sufficiency Analysis (Truth Table) - Construct truth table → set frequency threshold (minimum 2 cases) - Consistency threshold ≥ 0.80 - Boolean minimization → derive parsimonious / intermediate / complex solutions
Step 4: Interpretation - Identify sufficient condition paths (what combinations lead to Explicit CT) - Confirm equifinality (do different paths exist by country?) - R packages: QCA, SetMethods
Expected Scenarios
| Scenario | Description |
|---|---|
| A: Universal path | “Active Pedagogy × Reflective Assessment × Strategy Theme” → Explicit CT regardless of country |
| B: Equifinality | US: Active Pedagogy × Strategy; China: AI Policy × Theme Diversity; Korea: Reflective Assessment × MBA |
| C: Necessary condition | Active Pedagogy necessary but not sufficient alone — requires combination with other conditions |
QCA Coding Worksheet Template
| School_ID | Country | Program | Active_Ped | Reflect_Assess | Theme_Div | Strat_Innov | AI_Policy | Outcome_CT |
|-----------|---------|---------|------------|----------------|-----------|-------------|-----------|------------|
| US_001 | US | MBA | 1.0 | 0.67 | 1.0 | 1 | 1.0 | 1.0 |
| CN_001 | China | MBA | 0.33 | 0.67 | 0.67 | 1 | 0.67 | 0.67 |
| KR_001 | Korea | UG | 0.67 | 0.33 | 0.33 | 0 | 0.5 | 0.33 |Method 3: Topic Modeling (Structural Topic Model — STM)
Purpose
Inductively discover latent themes beyond the 5 pre-defined curricular themes → triangulate with deductive coding.
Design
- Corpus: Course description and syllabus text collected from all sampled schools
- Recommended model: STM (
stmpackage in R)prevalence ~ country + program_leveldirectly models country-level topic differences
- Alternatives: LDA (easier to interpret), BERTopic (semantic similarity)
Procedure
Step 1: Corpus preparation - Collect course/syllabus text by country - Preprocessing: tokenization, stopword removal, stemming (language-specific) - Construct English unified corpus (post-translation) or country-specific corpora
Step 2: STM execution - Determine number of topics: searchK() function, k = 5–25 range - Model fitting: prevalence ~ country + program_level - Output: Top keywords per topic, representative documents, topic proportions
Step 3: Analysis - Statistical test of topic distribution by country (estimateEffect) - Match STM topics against 5 pre-defined themes - Interpret newly discovered themes
Output location: docs/05_분석/Quan/
Method 4: Network Analysis
Purpose
Visualize and compare co-occurrence patterns among coding attributes across countries.
Network Types
| Network Type | Description |
|---|---|
| Bipartite (school × coding attributes) | Community detection → compare with existing 3 clusters |
| Co-occurrence (attribute × attribute) | Weighted network by co-occurrence frequency. Centrality analysis to identify core categories |
| Country-comparison | QAP (Quadratic Assignment Procedure) statistical test of structural differences between country sub-networks |
Procedure
Step 1: Network construction - Convert coding data to adjacency matrix - Co-occurrence network: create edges between coding categories co-occurring in the same school - Weight = co-occurrence frequency
Step 2: Network analysis - R packages: igraph, tidygraph, ggraph, sna - Whole network: degree centrality, betweenness centrality, community detection (Louvain algorithm) - Country sub-networks: same analysis repeated per country - Cross-country structural comparison: QAP
Step 3: Visualization - Force-directed layout for whole and country-specific networks - Node size = centrality; edge thickness = co-occurrence frequency; color = coding dimension
Output location: docs/05_분석/Quan/
Triangulation Logic
| Method Pair | What is triangulated |
|---|---|
| Content Analysis ↔︎ Network Analysis | Do community-detected clusters match qualitative mapping clusters? |
| Pre-defined themes ↔︎ STM topics | Do inductively discovered topics align with deductive coding themes? |
| QCA paths ↔︎ Mapping clusters | Do QCA sufficient condition paths correspond to sub-patterns within clusters? |
When methods diverge, the divergence itself becomes a finding — reporting methodological sensitivity differences as a contribution.
Software and Tools
| Purpose | Tool |
|---|---|
| Coding data management | Excel / Google Sheets |
| Topic Modeling | R (stm, tidytext, quanteda) |
| Network Analysis | R (igraph, tidygraph, ggraph, sna) |
| QCA | R (QCA, SetMethods) |
| Visualization | R (ggplot2, ggraph) + Quarto |
| Reference management | Zotero / BibTeX |
| Collaboration | Google Drive + GitHub Pages (public/) |
References
- Hwang, T., Park, B., Kim, H., & Kim, S. (2025). AI and critical thinking in business education: A first mapping of AACSB schools in the USA. Journal of Business and Management Cases.
- Ragin, C. C. (2008). Redesigning social inquiry. University of Chicago Press.
- Roberts, M. E., Stewart, B. M., & Tingley, D. (2019). stm: An R package for structural topic models. Journal of Statistical Software, 91(2), 1–40.
- Csardi, G., & Nepusz, T. (2006). The igraph software package for complex network research. InterJournal, Complex Systems, 1695.