theoretical-sampling

# Theoretical Sampling Theoretical sampling is **data collection guided by the emerging theory**. You collect your next slice of data because analysis has revealed **gaps in conceptual development**—unknown properties, unclear conditions, unstable relationships—not because a sampling frame demands representativeness for its own sake. Use this skill whenever you must decide **whom to interview next**, **what settings to observe**, **which documents to request**, or **what follow-up probes** to add. --- ## What theoretical sampling is **Definition (classic GT)**: Sampling on the basis of **concepts** developed during analysis, aimed at **discovering variation** and **clarifying relationships** relevant to the emerging **core category**. **Primary question**: “What data do I need **next** to **develop** this category / test this hypothesis / clarify this boundary?” --- ## How it differs from other sampling logics | Sampling logic | Typical aim | GT contrast | |----------------|------------|-------------| | Probability sampling | Statistical generalization to population | Not GT’s primary goal | | Convenience sampling | Access/ease | Risky if never corrected by analysis | | Maximum variation (descriptive) | Showcase diversity | Helpful, but still can be **atheoretical** if not tied to categories | | Theoretical sampling | Develop categories & hypotheses | **Driven by analysis memos** | **Note**: early GT projects often begin with **purposive** access (who will talk). Theoretical sampling **takes over** as categories mature. --- ## Preconditions (before you sample “theoretically”) You should be able to articulate at least one of: - A **category** that needs **densification** (properties/dimensions unclear). - A **hypothesis** that needs **confronting** with new incidents. - A **negative case** gap (deviance underrepresented). - A **boundary condition** you cannot specify. If you cannot name the **analytic reason**, your “theoretical” sampling may be **convenience** in disguise—document that honestly. --- ## Sampling for variation vs confirmation ### Variation-seeking sampling Use when a category is **thin** or **monotonic** in your data. **Targets**: - Different roles, ranks, sites, histories, or stakes - Cases predicted to be **high** vs **low** on an emerging dimension - Settings where the phenomenon should **plausibly fail** **Analytic goal**: discover **dimensions** and **contingencies**. ### Confirmation-seeking sampling (disciplined) Use when a hypothesis is **plausible** but **fragile**—supported by only a few incidents. **Targets**: - Cases predicted to **repeat** the pattern under stated conditions - Cases predicted to **break** the pattern (falsification-friendly) **Analytic goal**: stabilize **conditional statements** without cherry-picking. **Caution**: “Confirmation” in GT is **not** seeking only supportive data; it means **testing** the emerging model comparatively. --- ## Writing sampling directives (memo template) ```text Category / hypothesis in focus: What is unknown (specific gap): Comparison needed (variation / boundary / mechanism): Sampling target (who/where/when): Eligibility criteria (inclusion/exclusion): Probes/questions to elicit relevant incidents: Ethical considerations / access constraints: What would count as “enough” for this gap (saturation note): ``` ### Example directive (illustrative) > “We have *softening risk framing* as a strategy, but unclear **when it backfires**. Next sample: participants who **failed** to secure manager support after soft framing; compare to **successful** cases. Ask for **moment-by-moment** account of manager response.” --- ## Interview / observation probes aligned with theoretical sampling Instead of only “tell me about X,” use **incident elicitation**: - “Walk me through the **last time** this became risky.” - “What happened **right before** and **right after**?” - “Who else was involved—what did they **do**?” - “Has it ever **gone differently**? What made it different?” These questions produce **compare-able** chunks. --- ## Relationship to saturation Theoretical sampling continues until relevant categories reach **theoretical saturation** (no new properties/relationships that matter). Sampling decisions should **update** as saturation signals appear. Signals you may be “done” with a category: - New data **repeat** known properties without refinement. - Negative cases **fit** refined boundary statements. - Additional interviews **do not** change memo outlines meaningfully. See `theoretical-saturation` for deeper assessment guidance. --- ## Documentation and ethics - Log **why** each participant/site was chosen **analytically** (audit trail). - Avoid **harmful targeting** justified as “theoretical”—ethics still governs inclusion. - Manage **power dynamics** when sampling deviant or vulnerable perspectives; prioritize **safety** and consent. --- ## Common mistakes - Confusing theoretical sampling with **snowballing** without analytic rationale. - Chasing **interesting stories** unrelated to core development. - Over-sampling **easy** voices because access is simple. - Stopping after a set *n* regardless of category development. - Writing proposals that pretend full theoretical sampling plan upfront—classic GT **cannot** finalize this before analysis. --- ## Key references - Glaser, B. G. (1978). *Theoretical sensitivity*. Sociology Press. - Glaser, B. G. (1992). *Basics of grounded theory analysis*. Sociology Press. - Glaser, B. G., & Strauss, A. L. (1967). *The discovery of grounded theory*. Aldine. --- ## Companion skills - `constant-comparison`, `memo-writing`, `open-coding`, `selective-coding` - `theoretical-saturation`, `glaserian-grounded-theory`