Description
Tests whether statistically derived clusters correspond to genuinely different people, goals and contexts by investigating exemplar members through interviews, replays or diaries.
Statistical validity does not make a segment meaningful. This method samples high-membership and boundary users from each cluster and investigates them qualitatively, through session replays, interviews or short diaries, to check whether the cluster reflects a coherent motivation or context, or an artefact of features and scaling. A blind sorting step, where researchers group exemplar profiles without knowing labels, tests whether clusters are recognisable. Findings feed back into feature engineering and naming. Rigour comes from purposive sampling (core and boundary cases), blinding where possible, explicit criteria for confirming, splitting or merging clusters, and documenting disconfirming evidence.
Attributes
- Discipline
- Qualitative Analysis & Synthesis
- Loss type
- Latent
- Goal
- Segment users, Understand why
- Stage
- Definition
- Data
- Mixed
- Effort
- Weeks
Steps
- Select 4-6 high-membership users and 2-3 boundary users per cluster for investigation.
- Review session replays or event timelines for each, then conduct 30-45 minute interviews or 1-week diaries with a subset.
- Ask researchers blind to cluster labels to sort anonymised profiles into groups and compare with algorithmic clusters (e.g. agreement above 70%).
- Code interviews for goals, triggers and contexts, and compare themes within versus across clusters.
- Decide for each cluster whether to confirm, rename, split, merge or discard it, and record the evidence.
- Feed findings into revised features or segmentation and repeat if major changes occur.
Sources
- Cluster Analysis for Researchers (Charles Romesburg)
- Pattern Recognition and Machine Learning (Christopher Bishop)
- Methods Atlas (Vivek Gopinath)
- Designing and Conducting Mixed Methods Research (John Creswell & Vicki Plano Clark)
- Cluster Analysis for Researchers
- Pattern Recognition & ML (Bishop)
- Methods Atlas (Gopinath)
- The Inquiry Series, Ch 51 — Behavioral Clustering & Probabilistic Segmentation
Also known as
- Qualitative cluster validation
- Mixed-methods segment validation
Categories
- User Research
- Qualitative Analysis & Synthesis