技能 数据科学 计量经济学数据分析与报告规范

计量经济学数据分析与报告规范

v20260724
ectj-data-analysis
本指南提供了严谨的计量经济学实证和模拟证据报告的完整标准。涵盖了蒙特卡罗模拟设计、证据图谱构建、通过详细记录确保可复现性,以及将理论假设与有限样本证据挂钩的方法。适用于高水平的统计研究。
获取技能
79 次下载
概览

EctJ Data Analysis

Use this when the method has to prove both statistical behavior and empirical usefulness.

Analysis checks

  • Keep Monte Carlo evidence focused. RES guidance asks that simulation results be summarized

compactly in the main text; use the supplement for details.

  • Include an empirical application that demonstrates applied value, even for theory-heavy

work.

  • Align simulations with the assumptions and failure modes from the theory section.
  • Compare against credible econometric alternatives, not only simplified baselines.
  • Report sample sizes, data-generating processes, tuning, seeds, software versions, runtime,

and convergence or failure diagnostics.

  • Show where the new procedure changes an applied conclusion, uncertainty interval, test

decision, or policy-relevant estimate.

Minimum evidence map

Before drafting results, create a one-page map with these rows:

  • Theory target: theorem, proposition, approximation, or diagnostic the simulation is meant to stress.
  • DGP grid: the smallest parameter grid that probes the boundary cases, not every imaginable design.
  • Competitors: incumbent estimator/test plus at least one strong practical alternative.
  • Failure diagnostics: convergence failures, non-positive matrices, weak identification, bandwidth/tuning

sensitivity, or coverage breakdowns.

  • Application payoff: the single empirical decision that changes because the method exists.

The main text should report only the rows that teach the reader why the method works and when it fails.
Full grids belong in the supplement or replication package.

Reproducibility ledger

Track every reported number in a ledger:

Manuscript itemScriptSeed/configOutput pathRuntime
Table/Figure X............

Use the ledger to decide what must be in the main replication path and what can remain optional.

Theory-to-simulation contract

EctJ referees read Monte Carlo sections as tests of the theory, and RES guidance caps the
main-text simulation summary near one page, so every theoretical claim needs exactly one matching
finite-sample exhibit:

Theoretical claimRequired Monte Carlo evidenceTypical display
Asymptotic normalityCoverage of nominal 95% intervals across nCoverage row per sample size
Size control of a testNull rejection rates near 5% at the relevant boundarySize table with nominal level in header
Local power gainPower against the incumbent under drifting alternativesOne power figure
Rate or bias reductionBias and RMSE relative to the strongest competitorCompact bias/RMSE panel
Tuning robustnessBehavior across bandwidth or penalty choices used in practiceSupplement grid, one-line main-text summary

A theorem with no matching row invites the classic EctJ objection that the asymptotics carry no
finite-sample evidence; a simulation with no matching theorem is decoration to cut.

Anchoring the DGP in the application

The other classic objection is a simulation design detached from the empirical illustration. Fix
it by calibration (illustrative numbers): if the application is a firm panel with N=180, T=12, and
residual serial correlation around 0.6, the core DGP should be N=200, T=12 with AR(1) errors at
rho in {0, 0.3, 0.6}, not an i.i.d. cross-section with n=10,000. State in the simulation preamble
which DGP parameters were estimated from the application data and which probe theoretical
boundaries. One calibrated design plus one boundary design beats six arbitrary grids at this
venue, and the pairing lets the empirical section reuse the simulation's vocabulary when it
explains why the new procedure changes the applied conclusion.

Computation reporting floor

  • State replication counts and justify them (illustrative floor: 1,000 draws for size claims,

more when coverage is pushed to the third digit).

  • Report wall-clock runtime for the main simulation and the application on stated hardware;

the EctJ replication policy makes these numbers checkable after conditional acceptance.

  • Log convergence failures per design cell and the handling rule (drop, restart, flag); silent

drops change rejection rates, and referees at this venue know it.

  • Name the software stack and versions in the simulation note, matching the replication README.

Execution bridge (StatsPAI / Stata MCP)

Run the battery, don't just enumerate it. Full map:
execution-with-mcp. The Econometrics Journal is a methods venue — estimator validity + simulation; pair estimates with diagnostics.

  • Many outcomes / specifications: romano_wolf (step-down FWER) or benjamini_hochberg.
  • OVB sensitivity: oster_delta / sensemakr.
  • Inference: wild_cluster_bootstrap (few clusters), twoway_cluster / conley.
  • Re-fit off one handle: audit_result(result_id) lists missing checks + the exact

suggest_function for each.

  • Exhibits: etable / did_summary_to_latex from the handle — no retyped numbers.

Decisive checks in the body, exhaustive battery in the appendix.
JF execution walkthrough.

Output format

[Evidence readiness] strong / adequate / weak
[Monte Carlo role] <theory validation or stress test>
[Empirical application role] <applied-value demonstration>
[Missing baseline or diagnostic] <item>
[Next analysis] <single run or table>
信息
分类 数据科学
名称 ectj-data-analysis
版本 v20260724
大小 5.7KB
更新时间 2026-07-28
语言