research-methods
Design and execute social/health research and apply the right methodology — conceptual & logic modeling, community-based participatory research (CBPR/YPAR), participatory and agent-based modeling, com
SKILL.mdskills/research-methods/SKILL.md
GitHubResearch methods — design, modeling, ethics, and domain context
Goal: bring the right methodology to a research task and ground it in what the researcher actually studies. This skill carries reusable method craft; pair it with a domain/collaborator profile of the researcher so recommendations land in context.
Researcher domain profile
Keep a profile of the researcher this install serves and read it before any research
task. It should hold their core domains (topics, constructs, recurring indices),
populations & settings (who, where, which agencies), and collaborators &
institutions (appointments, mentors and their specialties, community partners,
funders, HRPP/IRB contacts). This install's profile: LOCAL.md beside this skill, when
your copy keeps one — Researcher domain profile.
Conceptual / logic modeling
Build iteratively, in layers, then translate the diagram into prose backed by peer-reviewed (DOI-linked) literature:
- Start simple: one driver → outcome.
- Add pathways (e.g. direct exposure, secondary stressors, eco-anxiety) — multiple mechanisms, not one variable.
- Add moderators (social support, services, institutional trust, SES) — annotate how they're degraded by upstream forces.
- Put the structural root cause upstream (e.g. structural racism via a neighborhood track and a chronic-stress/allostatic-load track), not as a background moderator.
- Add feedback loops (outcome decline → increased vulnerability; an investment that severs a moderating layer).
- Treat intersectionality as a cross-cutting lens, not a box — boundary conditions that reshape the meaning of every node; name the specific axes at each tier (documentation status, disability/IEP loss, policing/stigma).
Community-based participatory research (CBPR / YPAR)
Engage participants as paid co-investigators, not subjects. Co-design research questions, instruments, sampling, consent, and data-governance protocols. Power-sharing safeguards seen to work: small-group-before-full-group, anonymous input channels, metacognitive check-ins, bilingual (EN/ES) materials, anchoring discussion in participant-produced artifacts to counter deference to adult/authority. Attend to immigration fear, research fatigue, and emotional safety / referral pathways. The community partner reviews all protocols before fielding. Frame participants as epistemic agents whose lived experience is primary data that administrative records and adult testimony can't replicate — and frame their exclusion as itself a mechanism of inequality the practice disrupts.
Participatory & agent-based modeling
- Participatory modeling (PM): causal loop diagramming (CLD) with participants (Zellner & Massey 2024), extended to agent-based models; participants rank strategies on feasibility × impact. CLD diagrams and ABM runs serve as analytic instruments.
- Agent-based simulation (NetLogo-style): patch-and-household model — each tick, mitigation projects complete and raise nearby housing quality/cost above a baseline; cost-burdened households evaluate moving via a utility function trading climate resilience vs. affordability; households that can't find affordable housing are "displaced," and at-risk-not-yet-moved households tracked via a displacement-pressure counter. Report sensitivity-analysis interpretation, not just runs.
Composite-index spatial analysis (climate gentrification)
Tract-level composite: Vulnerability = Environmental Exposures + Sensitivity − Adaptive Capacity. Indicators z-score standardized then rescaled 0–100 (preserves relationships, prevents negatives canceling positives), equally weighted, summed. Sub-indices: a Gentrification Index (tract typologies across observation windows, inflation-adjusted to constant dollars, mean imputation for missing data) and a Displacement Risk Index (social-vulnerability + housing-market sub-indexes, signs aligned, quartiled low/moderate/high). Adaptive capacity = count of resources within walking distance (≈400 m). Fold in community-nominated indicators (e.g. PM2.5, residential fires) from collaborative data-analysis workshops. Map at tract level with graduated shading.
Implementation-science & feasibility frameworks
- Bowen Feasibility Framework (acceptability, demand, implementation, practicality, adaptation, integration, expansion, limited-efficacy) — for piloting when no prior evaluation exists.
- CFIR (5 domains) preferred over RE-AIM when multilevel + equity attention is needed; map each method onto a framework domain so the framework is visible in the data-collection plan.
Measures / data collection
Mixed methods, often with instruments co-designed via YPAR rather than pre-selected validated scales. Methods seen: surveys & intercept surveys, semi-structured interviews, focus groups, field observation, gallery walks, intergenerational workshops. Go beyond deficit constructs — name solastalgia, climate/racial grief, allostatic load, displacement anxiety. Always pair each data source with an analysis/interpretation framework (thematic analysis, grounded theory, member checking, ABM sensitivity interpretation) — a source list is not an analysis plan.
Human-subjects protections (IRB / HRPP)
From AAHRPP-accreditation consulting work (the engagement this came from: LOCAL.md
beside this skill, when your copy keeps one — Human-subjects protections: where this
came from):
- AAHRPP three-domain framework structures everything: I. Organization, II. IRB/Ethics Committee, III. Researcher & Research Staff. Review documents element-by-element against these domains.
- Plan vs. Investigator Manual: a Plan (e.g. UW–Madison HRP-101, the standard Huron Toolkit "HRPP Plan") carries structure, definitions, roles/responsibilities, scope, document control. The plain-language researcher voice lives in the companion Investigator Manual (HRP-103). Match the model document to the document type — don't conflate them.
- Institutional-document hallmarks: definitions up front; a roles-and- responsibilities spine; explicit "overseen vs. not overseen" scope lists; the principle that protections do not vary by funding; agency-by-agency regulatory mapping; formal document control; continuity-of-operations provisions.
- FWA & registered IRBs: an HRPP holds a Federalwide Assurance and may operate
multiple registered IRBs (this install's worked instance:
LOCAL.mdbeside this skill, when your copy keeps one — FWA & registered IRBs: the worked instance). - A standalone Human Subjects Protection section in a proposal covers IRB, data security, confidentiality, risk, and mitigation — dedicated, not scattered.
Standing caveats:
- AAHRPP element numbers (III.1.E, I.1.D, …) must be verified against the current Evaluation Instrument before use in a submission — AAHRPP periodically renumbers.
- Distinguish planning artifacts (reviews, standards docs, restructure blueprints) from final products (the rewritten manual/Plan still has to be drafted from them).
- Treat already-approved stakeholder source material as a reference input, not something to regenerate.
- Tooling reality: some "PDF" files are ZIP archives of page images with OCR, or Markdown; auth/robots-restricted Box/Drive links can't be fetched — ask for the file.
Literature synthesis
When asked to "find articles that support this model," return peer-reviewed sources with links/DOIs, mapped to the specific construct/node they support. Recurring anchors: Crenshaw; Bailey et al.; Marcuse; Dantzler (2021); Pulido (2016); Rucks-Ahidiana (2022); Massey & Denton; Bates (2013); Desmond (2012); Ozer et al. (2020); Zellner & Massey (2024); Bowen et al. (2009); Damschroder et al. (CFIR); Binet et al. (2019).
Before you call this done
- Conceptual model has the structural root cause upstream, intersectionality as a cross-cutting lens (not a box), and feedback loops — not a single driver→outcome arrow.
- Every data source is paired with an analysis/interpretation framework — a source list is not an analysis plan.
- CBPR/YPAR work frames participants as paid co-investigators / epistemic agents, and the community partner reviews all protocols before fielding.
- Human-subjects material sits in one dedicated section (IRB, data security, confidentiality, risk, mitigation), and any AAHRPP element numbers are verified against the current Evaluation Instrument.
- Literature claims are backed by peer-reviewed, DOI-linked sources mapped to the specific construct/node they support.
Related skills
- information-architecture: the structure of a long paper or institutional document: audience, thesis, section order.
Capturing learnings
Record new methods, framework choices, instrument designs, IRB/AAHRPP specifics, and
domain/collaborator updates in LEARNINGS.md beside this skill, when your copy keeps one.
Read before a research task; update after.