# Research Study Design

> Turn a business brief into a complete research proposal linking qualitative exploration and quantitative validation, with sampling, analysis and delivery plans. Use for overall study design or proposal review; use the questionnaire skill for item-level specifications.

- Skill: `frankglendon/research-study-design` (Agent Skill, multi-file: 4 files)
- Install (CLI): `npx skillmds@latest add frankglendon/research-study-design`
- Raw SKILL.md: https://api.skillmd.com/api/skills/frankglendon/research-study-design/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Research & Search
- Author: frankglendon (https://skillmd.com/u/frankglendon)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/frankglendon/research-study-design

---


# Design an end-to-end research study

The host authors a reviewable proposal. This skill supplies planning guidance, not fieldwork, statistical execution or an independent agent runtime.

1. Read the brief and authorized references. Separate client requirements, historical proposals, hypotheses and findings. Record the decision, markets, population, period, budget and intended use. Follow existing user authorization; progress with explicit reversible assumptions where appropriate. Do not invent supplier access, prices, experience or credentials.
2. Map each business decision to research questions, method, sample/analysis base, analysis, deliverable and action condition. Remove modules without a decision use. Preserve regional scope and verify existing operations before calling a study first entry.
3. Choose the necessary stages. For mixed-method work, explain how qualitative findings challenge hypotheses and become quantitative constructs/stimuli, and how quantitative results could change recommendations. Use [design and review guidance](references/design-review.md); two unrelated method lists are insufficient.
4. Specify qualitative method, recruitment states, sample/city rationale, interview length, module probes, stimuli, coding and negative cases. Do not multiply every balancing attribute into tiny cells. Shop-alongs/diaries drawn from interviewees are subsets, not additional unique participants. Themes are not population percentages and assumed personas are not validated segments.
5. Specify the quantitative population, frame/source, screening, quotas, base/boost samples, questionnaire order, measures, pretest, QC, weighting and analysis bases. Size critical comparisons rather than just the total. An opt-in sample does not support a probability-sampling margin of error or guaranteed representativeness. NPS uncertainty is not a binary proportion formula. Randomized stimulus responses do not establish real-world revenue lift.
6. Name the data, method, assumptions and fallback for each analysis. Check segment stability/assignability, distinguish perceived prices from elasticity, and narrow claims or add data when bases are insufficient. A described method does not mean its estimator is implemented.
7. Use the [proposal outline](references/outline.md) to cover deliverables, dependencies, schedule, staffing roles, costing quantities, risks and unresolved decisions. Adapt depth to the brief rather than copying historical sample sizes or fixing a page count.
8. Review coverage, sample arithmetic, time budgets, local versions and method integration. Distinguish a draft ready for review from authorization/readiness to conduct fieldwork. Hand the frozen survey design to `research-questionnaire`, its coded specification to `research-codebook`, and stage status to `research-workflow`.

Label the artifact as a proposal until research has been conducted. Do not prefill segment sizes, preferences or growth outcomes. If exporting Office, use the host's document/presentation tools and complete Microsoft Open XML SDK and visual checks. Keep client originals local; public examples must be generic or synthetic.