Axial coding agent

Description

Using an LLM agent to propose relationships between open codes, such as conditions, actions and consequences, around central categories, for researchers to evaluate and refine.

Axial coding, in Strauss and Corbin's grounded theory, reassembles open codes by relating them to a category through the paradigm model: causal conditions, context, intervening conditions, actions or interactions, and consequences. An axial coding agent receives the open codes with their evidence and proposes candidate categories and relationships, each with supporting quotes and counter-evidence. Because axial coding is interpretive, the agent is a hypothesis generator: researchers test proposed relationships against the data and constant comparison. Rigour comes from requiring evidence for every proposed relation, asking explicitly for disconfirming cases, comparing the agent's proposals with independent researcher memos, and documenting accepted and rejected relations.

Attributes

Loss type
Latent
Effort
Weeks

Steps

  1. Prepare the open-code table with definitions and linked quotes from prior passes.
  2. Prompt the agent with the paradigm model and a schema for relations (category, code, relation type, evidence quotes, counter-evidence).
  3. Run the agent per candidate category and collect proposed relations.
  4. Have researchers independently write axial memos for the same categories before reviewing the agent's output.
  5. Compare, test each proposed relation against the data, and accept, revise or reject with reasons.
  6. Iterate with constant comparison until categories are saturated, and record the final paradigm model.

Sources

  • Whisper.cpp
  • spaCy
  • Claude API (multi-turn agent)
  • asyncio
  • pandas
  • Prefect/Airflow for pipeline orchestration
  • Strauss, A. & Corbin, J. (1990) Basics of Qualitative Research
  • Corbin, J. & Strauss, A. (2015) Basics of Qualitative Research, 4th ed.
  • Saldaña, J. (2021) The Coding Manual for Qualitative Researchers, 4th ed.
  • The Inquiry Series, C13 — The Full Agentic Research Pipeline — Whisper → spaCy → AI Coder → Synthesis

Also known as

  • LLM-assisted axial coding
  • AI-supported category relating

Categories

  • User Research
  • Qualitative Analysis & Synthesis