jar-methods
GitHub专为JAR论文研究设计与识别策略优化，解决因果推断瓶颈。涵盖选择自然实验、断点回归或工具变量等识别变异源，应对内生性威胁，规范样本构建与复现流程，确保结论可信。
触发场景
安装
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill jar-methods -g -y
SKILL.md
Frontmatter
{
"name": "jar-methods",
"description": "Use when the research design and identification strategy are the bottleneck for a Journal of Accounting Research (JAR) manuscript — choosing the setting, source of identifying variation, and sample to support a causal accounting claim. Designs the study; it does not run the estimation, clustering, or robustness (jar-data-analysis)."
}
Research Design & Identification (jar-methods)
When to trigger
- The design is a panel regression with no source of identifying variation
- The claim is causal but the variation is observational/endogenous
- A referee says "this is correlation, not causation" or "the channel is unidentified"
- You must choose between an archival, experimental, analytical, or field design
JAR's dominant design: empirical-archival capital markets
JAR's defining methodology is large-sample empirical-archival capital-markets research (financial-statement and market-data econometrics in the Ball-Brown lineage). The journal also publishes experimental, analytical/modeling, and field-study work, and the Registered Reports track is well suited to higher-outcome-risk designs that require new data collection. The bar across all of them is credible identification: a referee must believe the estimate reflects the economic effect you claim, not an omitted variable, reverse causality, or selection.
Find identifying variation (the core archival problem)
| Theoretical claim | Identification that earns it |
|---|---|
| Effect of a rule/standard | Staggered or sharp adoption as a natural experiment (modern DiD) |
| Effect at a threshold | Regression discontinuity (e.g., index inclusion, size cutoffs) |
| Effect of an endogenous firm choice | IV/2SLS with a defensible instrument, or a shock to the choice |
| Information content of a disclosure | Short-window event study around the release |
| Causal mechanism under control | Experiment (lab/online/field), often paired with archival evidence |
- Anticipate the threats by name: omitted correlated variables, reverse causality, selection into treatment, and measurement error. State which one is the binding concern and how the design neutralizes it.
- Parallel trends / pre-trends for DiD; bandwidth and manipulation tests for RD; instrument relevance and exclusion for IV — plan the diagnostics now, not in the rebuttal.
- Construct measurement: define accounting constructs (earnings quality, disclosure, audit quality, comparability) with measures used in prior JAR work, and plan validation.
Sample and reproducibility from the start
Specify the sample frame, screens, and data vintages (Compustat/CRSP/I/B/E/S/Audit Analytics/EDGAR). Because JAR requires posted data and code, design the pipeline to be top-to-bottom reproducible from raw extracts, recording exclusion rules and access dates.
Execution bridge (StatsPAI / Stata MCP)
For the empirical / causal lane, estimate and audit rather than only specify. Full
map: execution-with-mcp. JAR is archival/empirical accounting; foreground identification around disclosure and regulation shocks, with modern DiD where adoption is staggered.
detect_design→recommend→ fit withas_handle=true→audit_resultto enumerate the checks the design owes.- Panel / staggered DiD:
callaway_santanna/sun_abraham+bacon_decompositionhonest_did_from_result. IV:effective_f_test+anderson_rubin_ci. RDD:rdrobust+mccrary_test.
- Experiments: randomization-based inference and
romano_wolffor the many-outcome family-wise correction reviewers expect.
Match the toolchain to the reviewer pool, and report the effect size the venue wants. A run end-to-end (synthetic data, real returns) is in the JF execution walkthrough.
Checklist
- A specific source of identifying variation is named (shock/threshold/instrument/manipulation)
- The binding endogeneity threat is identified and the design addresses it
- Diagnostics planned (pre-trends / bandwidth / first-stage / manipulation checks)
- Accounting constructs measured with validated, prior-literature measures
- Sample frame, screens, and data vintages specified; pipeline reproducible
- If high-risk/new-data: Registered Reports (Stage 1 protocol) considered
Anti-patterns
- Kitchen-sink OLS with controls standing in for identification.
- Endogenous regressor treated as exogenous with no strategy.
- Weak/implausible instruments failing relevance or exclusion.
- Staggered DiD with two-way FE ignoring heterogeneous-treatment-timing bias (use modern estimators).
- Event windows fished for significance; ad hoc bandwidths in RD.
Output format
【Design】archival-NE / RD / IV / event-study / experiment / analytical / field
【Identifying variation】shock / threshold / instrument / manipulation
【Binding threat】OVB / reverse causality / selection / measurement — addressed by ...
【Diagnostics planned】pre-trends / bandwidth / first-stage / balance ...
【Constructs & measures】definitions + validation
【Sample & reproducibility】frame, screens, vintages; data/code pipeline
【Next step】jar-data-analysis
Resources
../../resources/official-source-map.md— official JAR/Chicago Booth/Wiley URLs (accessed 2026-06-01)../../resources/external_tools.md— archival data sources and identification tooling (reghdfe / csdid / rdrobust / ivreghdfe)
版本历史
- 1839142 当前 2026-07-05 13:24
