Education Qualitative Coding Analysis

Education Qualitative Coding Analysis

Goal

Turn qualitative education data into a transparent coding process, evidence-backed themes, and a findings section.

Use After

Use after data sources and qualitative design are defined:

• education-qualitative-study-design
• education-mixed-methods-design
• education-action-research-design
• education-design-based-research
• education-policy-comparative-analysis
• education-program-evaluation

Do not expose the skill name to users. Present it as "质性材料编码分析" or "访谈/观察/文本分析".

Inputs

• Research questions
• Data type: interview, focus group, observation, document, policy, student work, AI feedback text, open-ended survey
• Corpus description and file list
• Chosen qualitative tradition if any: thematic analysis, grounded theory, case study, content analysis, discourse analysis
• Whether multiple coders are involved
• Language and anonymization requirements

Workflow

1. Prepare corpus:
   • transcribe audio/video
   • clean obvious transcription errors
   • anonymize names, schools, locations, accounts
   • assign document and participant IDs
2. Choose analysis approach:
   • thematic analysis for patterns of meaning
   • grounded theory for model/theory building
   • content analysis for categories/frequency
   • case analysis for bounded cases
   • discourse-oriented analysis for language/interaction patterns
3. Create initial coding unit:
   • sentence
   • paragraph
   • speaking turn
   • classroom episode
   • document section
   • artifact segment
4. First-cycle coding:
   • open coding
   • descriptive coding
   • in vivo coding
   • process coding
   • values/emotion coding where relevant
5. Build codebook:
   • code name
   • definition
   • inclusion/exclusion rules
   • example quote
6. Second-cycle coding:
   • axial coding
   • pattern coding
   • focused coding
   • category aggregation
7. Theme/model development:
   • theme names
   • theme definitions
   • relationships among themes
   • negative cases/contradictions
8. Write analytic memos throughout.
9. Build evidence table linking claims to excerpts.
10. Draft findings structure and discussion implications.

Tool Calls

Transcription

Whisper: https://github.com/openai/whisper
Otter: https://otter.ai/
讯飞听见: https://www.iflyrec.com/

Qualitative Coding Software

NVivo: https://lumivero.com/products/nvivo/
MAXQDA: https://www.maxqda.com/
ATLAS.ti: https://atlasti.com/
Dedoose: https://www.dedoose.com/
Taguette: https://www.taguette.org/
QualCoder: https://github.com/ccbogel/QualCoder

Lightweight Text Analysis

R:

install.packages(c("tidyverse", "tidytext", "quanteda", "readtext"))

Python:

pip install pandas nltk spacy scikit-learn

Use NLP only as support. Human interpretation remains required for qualitative analysis.

Output Format

1. Corpus Inventory

2. Coding Plan

3. Codebook

4. Category / Theme Table

5. Evidence Table

6. Findings Section Outline

Finding 1: [Theme name]
  - Main claim
  - Representative evidence
  - Variation across participants/cases
  - Negative or boundary case

Finding 2: [Theme name]
...

Approach-Specific Guidance

Thematic Analysis

Use when the goal is to identify patterns of experience, perception, practice, or meaning.

Steps:

1. Familiarize with data.
2. Generate initial codes.
3. Search for themes.
4. Review themes against data.
5. Define and name themes.
6. Write findings with evidence.

Grounded Theory

Use when the goal is to build a process model or explanatory theory from data.

Steps:

1. Open coding.
2. Constant comparison.
3. Axial coding.
4. Theoretical sampling if possible.
5. Selective coding.
6. Theoretical saturation judgment.
7. Model construction.

Output model:

Conditions -> Actions/Interactions -> Consequences

Content Analysis

Use when categories and frequencies matter, such as policy texts, textbooks, AI feedback comments, classroom discourse moves, or open-ended survey responses.

Steps:

1. Define coding categories.
2. Define coding unit.
3. Code sample and refine categories.
4. Code full corpus.
5. Count and interpret category patterns.
6. Combine frequency with qualitative examples.

Quality Rules

• Do not treat AI-generated codes as final findings.
• Every major theme needs multiple pieces of evidence unless it is explicitly a rare/negative case.
• Keep raw excerpt, code, category, and theme traceable.
• Distinguish participant claim from researcher interpretation.
• Include negative cases or contradictions when they matter.
• For classroom data, preserve context: lesson phase, activity, teacher/student role, task.
• For minor students, anonymize names and identifiable writing details.

User-Facing Closure

End with the next practical decision:

现在可以进入正式编码。你希望我先帮你做一版初始代码本，还是先根据 2-3 份材料做试编码，再修订代码本？