技能 数据科学 治理研究设计方法论

治理研究设计方法论

v20260724
govern-research-design
这是一份用于治理、制度和公共政策领域的实证研究设计指南。它提供了从因果推断(如DiD、IV、RDD)到定性案例分析、混合方法整合的完整方法论框架。本技能的核心目标是指导用户构建严谨的研究设计,确保其分析能够有力地排除所有可能的替代性解释,从而确立研究的独特贡献。
获取技能
378 次下载
概览

Research Design (govern-research-design)

Governance welcomes any rigorous approach but is demanding about each. The design must credibly connect the argument (govern-theory-building) to comparative/institutional evidence and rule out the strongest rival institutional explanation. This skill is mode-aware: pick the section that matches your work. (For the conceptual contribution, this is the empirical-design variant.)

When to trigger

  • Specifying identification, case selection, or comparative design
  • A reviewer questioned causal claims, case choice, country selection, or an institutional confound
  • Choosing governance/institutions measures (V-Dem, QoG, WGI, etc.) and defending them
  • Justifying why the design adjudicates the rival account from govern-literature-positioning

(a) Comparative / causal designs (governance & institutions)

  • Identification first. State the estimand and the assumptions that license a causal reading (parallel trends, exclusion, continuity, ignorability). Defend them; don't assert them.
  • Reform DiD / event study. When a reform rolls out across units/countries over time, use modern staggered-adoption estimators (Callaway–Sant'Anna, Sun–Abraham, de Chaisemartin–D'Haultfœuille, Borusyak et al.) — not naive TWFE, which is biased under heterogeneous/dynamic effects. Show pre-trends and event-study leads/lags.
  • Cross-national panels. Justify fixed effects (country, year), the level of clustering, and what is identified off within-country vs. between-country variation.
  • IV / RDD where applicable. IV: first-stage strength, exclusion, weak-IV-robust inference. RDD (e.g., electoral or threshold-based reform rules): density/manipulation tests, bandwidth robustness.

(b) Qualitative / comparative-historical

  • Case selection by design logic (most/least likely, typical, deviant, paired comparison) — not convenience. Say what each case is a case of, and how the selection adjudicates the argument.
  • Process tracing with explicit tests (hoop, smoking-gun, straw-in-the-wind, doubly-decisive); state what evidence would have disconfirmed the argument in each case.
  • QCA where used. Justify calibration of set membership, the truth table, and consistency/coverage thresholds; report and interpret limited diversity, not just the solution formula.
  • Source transparency. Archives, interviews, fieldnotes — plan how they will be documented and cited (see govern-transparency-and-data).

(c) Mixed methods

  • State the integration logic up front: does the qualitative work generate, test, or explain the quantitative result (or vice versa)? Sequencing and the role of each strand must be deliberate.
  • Show where the strands converge and own where they diverge — divergence is informative, not a flaw to bury.

(d) Measuring governance & institutions (caveats)

  • V-Dem, QoG, WGI, Bertelsmann, ICRG, etc. are estimates, not facts. Report the version, the construct each index actually captures, and the measurement model's uncertainty (e.g., V-Dem posterior credible intervals). Do not treat composite indices as ground truth.
  • WGI in particular aggregates perceptions and is endogenous to outcomes — flag this when it sits near the dependent or treatment variable.
  • Show results are not an artifact of one index: triangulate across measures where the concept allows.

The rival-institutional adjudication move (Governance-specific)

For the strongest rival institutional explanation, write one sentence: "If the rival were true rather than my argument, the cases/data would look like ___; instead they look like ___." A design that cannot distinguish your account of governing from the leading institutional alternative has not yet identified the contribution.

Execution bridge (StatsPAI / Stata MCP)

Estimate and audit the design, don't only describe it. Full map: execution-with-mcp. Governance is public administration and institutions research — comparative and causal designs on governance reforms; the chain serves its quantitative-causal lane, while comparative-historical / qualitative work uses its own standards.

  • detect_designrecommend → fit with as_handle=trueaudit_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

  • Naive TWFE on a staggered reform; clustering below the level of treatment assignment
  • "Causal" language on a cross-national correlation the design only supports as association
  • Convenience country selection dressed up as theory-driven case logic
  • Treating V-Dem/WGI/QoG scores as exact, ignoring index uncertainty and construct mismatch
  • A design that cannot rule out the leading rival institutional account

Output format

【Mode】comparative-causal / qualitative / comparative-historical / mixed
【Estimand or claim】what is being identified/shown
【Key assumption(s)】and how each is defended
【Governance measures】index + version + uncertainty/construct caveat
【Rival ruled out】the rival-institutional adjudication sentence
【Robustness/sensitivity】planned checks
【Next】govern-data-analysis

Supplementary resources

信息
Category 数据科学
Name govern-research-design
版本 v20260724
大小 6.61KB
更新时间 2026-07-28
语言