constant-comparison

# Constant Comparative Method The constant comparative method is the **engine** of classic grounded theory. It converts raw experience into **concepts** and concepts into **theory** by systematic comparison. Without comparison, coding collapses into **summary**; with comparison, coding becomes **analytic**. Use this skill whenever you are generating codes, merging categories, defining properties/dimensions, testing a core category, or evaluating saturation claims. --- ## What constant comparison accomplishes - **Refines codes** (definitions, inclusion/exclusion boundaries). - **Surfaces properties and dimensions** of categories. - **Suggests hypotheses** about conditions, strategies, and consequences. - **Prevents forcing** by continually confronting ideas with **new** and **deviant** data. --- ## Four stages (Glaser & Strauss, 1967) and how they operate today Classic texts describe comparing as progressing through **comparing incidents applicable to each category**, **integrating categories and their properties**, **delimiting the theory**, and **writing theory**. In practice, these **overlap**—but the sequence is still pedagogically useful. ### Stage 1 — Compare incidents to incidents (within emerging codes) **Goal**: Ensure a code is **conceptually consistent** and **data-grounded**. **Micro-procedure**: 1. Pick a code (e.g., *delaying disclosure*). 2. Collect **several** incidents tagged with that code. 3. Ask: **What is similar?** What differs? 4. Update the code definition; split into two codes if differences are **systematic**. **Output**: tighter definitions; early **properties** (“kinds of delaying,” “degrees of delaying”). ### Stage 2 — Compare incidents to concepts (refinement) **Goal**: Use new data to **test** the conceptual label. **Micro-procedure**: 1. Read a new incident. 2. Ask: “Is it still **this** category, or a **variant**, or a **different** category?” 3. If it stretches the definition uncomfortably, **rename/split** rather than “make it fit.” **Output**: dimensional thinking; boundary conditions; negative cases queued. ### Stage 3 — Compare concepts to concepts (integration) **Goal**: Build **relationship statements** among categories. **Micro-procedure**: 1. Take two mature categories (e.g., *psychological safety* and *disclosure timing*). 2. Ask: **Do they co-occur?** **Does one enable/constrain the other?** **Under what conditions?** 3. Write a **memo hypothesis** and mark it as provisional. **Output**: contingent hypotheses; later fodder for **theoretical coding**. ### Stage 4 — Delimiting comparisons (selective phase) **Goal**: Stop expanding sideways; deepen **core-related** comparisons. **Micro-procedure**: 1. Hold the **core category** constant. 2. Compare new incidents for **what they add** to core-linked hypotheses. 3. Drop comparisons that **do not change** your integrated outline. **Output**: theoretical density without endless sprawl. --- ## Comparison at each level (cheat sheet) | Level | Question | Typical yield | |------|----------|----------------| | Incident ↔ incident | “Same meaning?” | Code splits/merges | | Incident ↔ concept | “Still fits?” | Property/dimension refinement | | Concept ↔ concept | “How connected?” | Hypotheses + theoretical codes | | Model ↔ negative case | “Where breaks?” | Boundary conditions | --- ## Practical techniques you can use in a coding session ### 1) Flip-flop comparison Take two incidents that **look opposite**. Force yourself to name the **conceptual axis** that distinguishes them (a **dimension**). ### 2) Worst-case / best-case contrast Select extremes within your sample (if available). Ask what conditions produce the difference. ### 3) Silent-voice comparison Ask: “Who is **not** represented here?” Negative cases may be **data absences**—note as a **sampling** issue, not only analytic failure. ### 4) Source triangulation comparison Compare **interview** claims to **observation** or **documents** where possible. Treat discrepancies as **analytic gold**. ### 5) Time-slice comparison Compare **early vs late** interview moments within a longitudinal account (if applicable). --- ## Comparison matrices (lightweight templates) ### Matrix A — Property discovery ```text Category: ________ Incident ID | Evidence snippet | Candidate property | Notes | ``` ### Matrix B — Hypothesis testing ```text Hypothesis: If [condition], then [strategy], leading to [outcome]. Supporting incidents (IDs): Contradicting incidents (IDs): Revised hypothesis: ``` ### Matrix C — Merge/split decisions ```text Code A vs Code B: Similarities: Systematic differences: Decision: merge / split / keep separate Rationale (comparative evidence): ``` --- ## Output format (session deliverables) End each analytic session with: ```text Comparisons executed (brief list): Code changes (rename/split/merge): New/updated hypotheses (bullets): Negative cases discovered: Next comparison targets (specific incidents/sources): ``` This becomes part of your **audit trail**. --- ## Common failure modes - **Coding without comparing** → pretty labels, weak theory. - **Comparing only supportive incidents** → confirmation bias. - **Over-merging** too early → lost variation; **under-merging** → synonym sprawl. - **Comparing quotes** instead of **concepts** → stays descriptive. - **Stopping comparison** when the story “feels done” instead of checking **saturation** criteria. --- ## Relationship to memoing and sampling - **Memos** store the *results* of comparisons as hypotheses. - **Theoretical sampling** targets the *next* comparisons you need but cannot yet make. --- ## Key references - Glaser, B. G., & Strauss, A. L. (1967). *The discovery of grounded theory*. Aldine. - Glaser, B. G. (1992). *Basics of grounded theory analysis*. Sociology Press. - Glaser, B. G. (1978). *Theoretical sensitivity*. Sociology Press. --- ## Companion skills - `open-coding`, `selective-coding`, `memo-writing` - `theoretical-sampling`, `theoretical-saturation` - `glaserian-grounded-theory`