Jmf research design

Research Design (jmf-research-design)

JMF accepts quantitative, qualitative, and multi-method work but is demanding about each. The design must connect the framework (jmf-theory-and-conceptual-framework) to evidence while respecting the unit of analysis (individual, dyad, family, household, cohort) and the non-independence of people who share a relationship. This skill is mode-aware: pick the section that fits your work.

When to trigger

• Specifying the design, sample, measures, and identification strategy
• A reviewer questioned selection, causal claims, generalizability, or the handling of dyads
• Designing a longitudinal, dyadic, family-level, or comparative study
• Preparing a pre-analysis plan or planning a replication

Quantitative — longitudinal / family demography

• Life-course / panel: growth curves, cross-lagged panel, fixed effects within persons or couples, change-score models; be explicit about timing, sequencing, and time-varying covariates.
• Event history / survival: discrete- or continuous-time hazards for marriage, cohabitation, divorce, fertility; competing risks where multiple exits are possible.
• Family demography: rates, life tables, decomposition, standardization; complex-survey weights, clusters, and strata applied correctly.
• Selection is the default rival. State how you address it — fixed effects, sibling/twin designs, propensity methods with sensitivity, natural experiments/IV, or honest scoping to association.

Quantitative — dyadic & family-level

• Couples/dyads: Actor–Partner Interdependence Model (APIM), dyadic SEM, multilevel models with members nested in couples; distinguish distinguishable vs. indistinguishable dyads.
• Families with multiple members/children: multilevel/mixed models; account for clustering within families; model coparenting and sibling structure where relevant.
• Measurement: validate constructs; report reliability; test invariance across partners, groups, or time before comparing.

Experiments (lab / survey / field)

• Preregister design and primary analyses; report power/MDE for the relevant unit (often the dyad or family, not the person); pre-specify subgroups. Address attrition, manipulation checks, and ethics/ IRB and consent for couples, children, and families.

Qualitative / multi-method

• Sampling and case logic stated by design (theoretical, maximum-variation, paired), not convenience; say what the case is a case of.
• Multi-perspective family data (both partners, parents and children): plan how interdependent accounts are analyzed and reconciled.
• Integration in mixed methods: explicit joint displays; say what each strand adds.

The selection-and-interdependence test (JMF-specific)

For the strongest rival (usually selection), write: "If selection rather than my mechanism drove this, the data would look like ___; instead they look like ___." Then confirm the model respects non-independence of partners/family members. If either fails, the design does not yet identify the contribution.

Design-defensibility table JMF referees apply

Design feature	Vulnerable	Likely reviewer verdict
Unit of analysis	Person-rows for a couple-level question	"Unit mismatch — re-specify"
Selection rival	Asserted away	"Selection into family transitions"
Temporal structure	Cross-section for a dynamic process	"Cross-sectional claim about a dynamic process"
Non-independence	Independence assumed	"Dyadic dependence ignored"

For the flagship journal of family science, the design section is judged on whether the unit of analysis and temporal leverage fit the family process being theorized. JMF welcomes quantitative, qualitative, and multi-method work, each on its own rigor bar.

Worked micro-example (illustrative)

A study claims cohabitation "causes" lower relationship quality from one cross-sectional wave (illustrative).

• Vulnerable: a between-person snapshot where married respondents score 0.4 SD higher — read as selection (who marries) plus a cross-sectional claim about a process that unfolds over time.
• Strengthened: a panel tracking couples from cohabitation forward, using within-couple change as partners transition to marriage, the couple as the unit, members modeled via APIM. The estimand becomes within-couple change in dyadic quality (illustrative −0.08 SD). Adjudication: "If sorting drove this, gaps would predate cohabitation."

Referee-pushback patterns and the design fix

• "Selection into family transitions." Add within-unit leverage (couple/sibling FE), a sensitivity analysis, or a credible natural experiment; else scope the claim to association.
• "Dyadic dependence ignored." Specify APIM, dyadic SEM, or multilevel nesting; test whether dyads are distinguishable rather than assuming.
• "Cross-sectional claim about a dynamic process." Re-anchor on a longitudinal estimand matching the life-course logic of the framework.

Calibration (hedged): preregistration and unit-appropriate power are encouraged but evolving — confirm against the journal's current author guidance.

Execution bridge (StatsPAI / Stata MCP)

Estimate and audit the design, don't only describe it. Full map: execution-with-mcp. JMF is quantitative family demography/sociology; emphasize identification, selection, and decomposition methods.

• detect_design → recommend → fit with as_handle=true → audit_result.
• Observational causal claims: staggered DiD (callaway_santanna / sun_abraham + bacon_decomposition + honest_did_from_result); IV (effective_f_test + anderson_rubin_ci); RDD (rdrobust + mccrary_test).
• Experiments: randomization-based inference, romano_wolf for many-outcome family-wise control, and mediate for mediation (not naive controlling-away).
• Sensitivity: oster_delta / sensemakr for observational claims.

Report the effect size in interpretable units; route the full battery to the appendix/supplement. A run end-to-end (synthetic data, real returns) is in the JF execution walkthrough.

Anti-patterns

• Treating couple or family data as independent observations
• "Causal" language on an observational design with unaddressed selection
• Naive cross-sectional snapshots for inherently longitudinal family processes
• Ignoring panel attrition or differential dropout in family studies
• Convenience case selection dressed up as theory-driven

Output format

【Mode】longitudinal-demographic / dyadic-family / experiment / qualitative-mixed
【Unit of analysis】individual / dyad / family / household / cohort
【Estimand or claim】what is being identified/shown
【Selection handled】the adjudication sentence
【Non-independence】how clustering/dyad structure is modeled
【Robustness/sensitivity】planned checks
【Next】jmf-data-analysis

Supplementary resources

• ../../resources/external_tools.md — dyadic/multilevel/survival packages and family datasets
• ../../resources/official-source-map.md — JMF methods scope and replication guidance