Research Design (asr-research-design)

ASR welcomes many methods but is exacting about each. The design must credibly link the argument (asr-theory-building) to evidence and rule out the leading alternative. Pick the section matching your method.

When to trigger

• Specifying identification, case selection, sampling, or site/informant logic
• A reviewer questioned causal claims, generalization, selection, or a confound
• Justifying why your design adjudicates the rival account from asr-literature-positioning

Quantitative / demographic

• Be honest about what the design identifies. Much of sociology is observational; distinguish description, association, and causation. If causal, state the assumptions (ignorability, parallel trends, exclusion) and defend them.
• Designs: panel/fixed-effects, DID/event study (modern staggered estimators, not naive TWFE), IV, RDD, matching/weighting with balance and sensitivity; decomposition for inequality; event history for timing; age-period-cohort for demographic change.
• Inference & sampling: respect complex survey design (weights, clustering, strata); cluster at the right level; report uncertainty.
• Sensitivity: how strong must an unobserved confounder be to overturn the result?

Comparative-historical

• Case selection by design logic (most/least-likely, paired comparison, deviant) — say what each case is a case of; justify the comparison and the counterfactual.
• Evidence: archives, secondary histories, administrative records; document provenance.
• Specify the causal form (necessary/sufficient conditions, conjuncture, path dependence; QCA where apt) and what evidence would have disconfirmed the argument.

Ethnographic / interview

• Site and informant selection justified theoretically, not by access alone; state positionality and access conditions.
• Plan how depth, saturation, and negative cases are handled; how claims map to observed evidence.
• Ethics/IRB, consent, and confidentiality planned from the start (see asr-data-and-transparency).

Network / computational

• Define boundary specification, missing-ties, and the generative process (e.g., ERGM/SAOM logic).
• For computational text/ML: validate against human-labeled samples; report stability.

The adjudication test (ASR-specific)

For the single strongest rival explanation: "If the rival were true rather than my argument, the evidence would look like ___; instead it looks like ___." If you cannot write it, the design does not yet identify the contribution.

What ASR referees demand of each design

ASR judges every tradition rigorously on its own logic — the unifying ASA standard is a design that connects argument to evidence and rules out the leading rival.

Worked micro-example (illustrative)

A comparative-historical study argues that labor-incorporation timing shaped later union strength across four countries.

Case logic: paired comparison (early vs. late incorporators), matched on industrialization level
Causal form: path dependence — early settlement locks in conditions C1–C2
Disconfirming evidence sought: a case with early timing but weak unions → would break the claim
Adjudication sentence: "If prior militancy drove this, the deviant case shows strong unions despite
  late timing; instead it shows weak unions, as the timing argument predicts."

The selection is design-driven, the rival is named, and the design specifies what evidence would have falsified the argument.

Referee pushback → ASR-specific fix

• "Association dressed as causation." → Restate what the design identifies and add a sensitivity bound or placebo; drop causal verbs you cannot defend.
• "Your cases look hand-picked." → Show the selection rule and what each case represents at the population level.

Calibration anchors

• Method-appropriate rigor, one disciplinary bar. ASR won't hold ethnography to a regression standard or vice versa — but every design must defeat its strongest rival.
• The adjudication sentence is the test. If you can't write "if the rival were true the evidence would look like ___," the design does not yet earn the contribution.
• Identification honesty travels. Stating plainly what observational data can and cannot establish reads as strength to a cross-method panel.

Execution bridge (StatsPAI / Stata MCP)

Estimate and audit the design, don't only describe it. Full map: execution-with-mcp. ASR is general sociology where observational designs dominate; foreground identification (DiD/IV/RDD), decomposition, and clustered inference.

• 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

• "Causal" language on a purely observational/associational design
• Naive TWFE on staggered timing; ignoring survey weights/clustering
• Convenience case or site selection dressed up as theory-driven
• Ethnographic claims with no account of negative cases or saturation
• A design that cannot distinguish your mechanism from the leading alternative

Output format

【Method】quant/demographic / comparative-historical / ethnographic / network-computational
【What it identifies】description / association / causation
【Key assumption(s) or selection logic】and how defended
【Rival ruled out】the adjudication sentence
【Robustness / negative cases / sensitivity】planned
【Next】asr-data-analysis

Supplementary resources

• ../../resources/external_tools.md — design/analysis packages and CAQDAS/QCA tooling
• ../../resources/official-source-map.md — ASR methodological breadth