SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsmw-identification

mw-identification

Design & creation

View repository

Updated 10-15 20:09

About this skill

Causal identification strategy (mw-identification)

Trigger timing

  • The empirical body has only OLS + control variables, worried about rejection
  • DID uses TWFE but does not respond to recent criticism of heterogeneous treatment (Goodman-Bacon, de Chaisemartin, Sun-Abraham, Callaway-Sant'Anna)
  • IV first stage F is weak / doubts about instrument endogeneity
  • Planning to use double machine learning but unsure how to report it

Design priority

Preference ranking of Management World editorial board members (strong → weak):

  1. Policy shock + DID (including staggered / continuous treatment)
  2. Regression discontinuity (clear policy threshold)
  3. Instrumental variables (strong instrument + exclusion argument)
  4. Propensity score matching + DID
  5. Synthetic control method
  6. Double machine learning / causal forest
  7. OLS + endogeneity discussion (easy to reject)

Branch paths

Branch A: DID

  • Is it staggered? → Must use Goodman-Bacon decomposition + Callaway-Sant'Anna or Sun-Abraham
  • Parallel trends test: event study plot must be drawn
  • Placebo: randomly assign treatment group 500–1000 times
  • Should the "bad comparison" weights of the Bacon decomposition be reported?

Branch B: IV

  • First-stage F must be ≥ 10(weak instrument → use Anderson-Rubin or weak-IV-robust CI)
  • The exclusion argument requires at least 3 paragraphs: theory / institutions / placebo
  • Has the reduced form been reported?

Branch C: RDD

  • Has the McCrary test been done?
  • Bandwidth: Is the optimal bandwidth used (Calonico-Cattaneo-Titiunik) + robustness for at least 3 bandwidths?
  • Covariate smoothness test

Branch D: DML

  • Report the number of sample splits + cross-fitting
  • Report the choice of nuisance functions (lasso / random forest / xgboost)
  • Provide robustness for at least 5 different ML learners

Execution bridge (StatsPAI / Stata MCP)

Actually run the designand audit it, not just describe it. For the complete mapping, see execution-with-mcp. Management World emphasizes empirical work in the Chinese context + policy operability; identification + economic magnitude, while qualitative/case studies follow their own standards.

  • detect_design → recommend → use as_handle=true to fit → audit_result to list the checks still owed.
  • **Observational causality:** staggered DID (callaway_santanna / sun_abraham + bacon_decomposition + honest_did_from_result）；IV（effective_f_test + anderson_rubin_ci）；RDD（rdrobust + mccrary_test）。
  • **Experiment:** randomization inference + romano_wolf to control the error rate for multiple outcome families.
  • Sensitivity:oster_delta / sensemakr。

Report in the main texteconomic magnitude, with the full battery in the appendix; every number must be reproducible. For a real end-to-end example, see JF execution walkthrough. If StatsPAI/Stata is not connected, switch to resources/code/ and mark unverified numbers.

Must-check list

  • Parallel trends / smoothness / weak instrument tests have all been done
  • Placebo tests have been done (random treatment timing / random treatment units)
  • The clustering level for standard errors in the main regression is reasonable (individual / individual+time / treatment level)
  • Is there "two-stage" identification (first confirm exogeneity, then estimate the treatment effect)
  • Has the "treated units' anticipation" problem been addressed

Anti-patterns

  • TWFE + staggered but no discussion of heterogeneous treatment bias
  • IV using "exogenous event × previous-period endogenous variable"—reviewers will ask "why does the previous period not affect the current period"
  • "We believe the policy is exogenous to firm decisions" but no evidence is provided
  • RDD uses a truncated bandwidth but does not report bandwidth sensitivity

Identification Strategy Decision Table for This Journal

Management World is sponsored by the Development Research Center of the State Council and is a top-tier authoritative comprehensive CSSCI journal that gives equal weight to management and economics. Its review expectations for identification strategies differ markedly from those of purely methodological journals: editorial board members place greater emphasis on "whether the policy shock genuinely serves major national strategies and whether the identification logic can be told as a Chinese institutional story," rather than on the novelty of the instrument. The table below anchors review decisions in the context of this journal (for uncertain formatting details, the editorial office's latest manuscript guidelines shall prevail):

Identification Scenario Leaning of This Journal's Editorial Board High-Risk Rejection Signal
National-level pilot policies (free trade zones, registration-based IPO system, low-carbon pilots) + staggered DID Strongly recognized; fits the "serving national strategy" positioning No diagnosis of heterogeneous treatment bias
Administrative approval threshold + RDD Recognized (institutional threshold is clear and credible) Threshold manipulation without McCrary
Bartik / share-shift instrumental variable Cautiously accepted; requires a three-part syllogism for exclusion Only a single sentence: "we believe it is exogenous"
Pure cross-sectional OLS + control variables High probability of rejection Writing the endogeneity discussion as a disclaimer
Cross-country panel + methodological innovation Mismatched with this journal's positioning; recommend submitting elsewhere Absence of the Chinese context

Common Rejection Patterns (Identification Dimension)

  • "Policy exogeneity" is only asserted without institutional evidence—this journal's editorial board requires the pilot selection mechanism to be written into the institutional background section
  • Staggered DID uses TWFE throughout, with not a word about the literature on heterogeneous treatment bias since 2021
  • The instrumental variable "looks exogenous," but shares common macroeconomic drivers with firms' total factor productivity, and the exclusion argument is absent
  • Using overseas data to argue about Chinese issues; no matter how clean the identification, it will be hard to pass the "relevance to national strategy" hurdle

Mini Walkthrough Example: Digital Transformation and Firms' Total Factor Productivity

A fictional manuscript is run through according to this skill's decision rules (all numbers below are illustrative and only demonstrate the judgment process):

  • Setup: Using a 2018 provincial "enterprise cloud adoption" subsidy pilot as the shock, examine the impact of digital transformation on firms' total factor productivity (TFP_LP). The sample is a panel of A-share manufacturing listed companies from 2012–2022, N≈18,600 (illustrative).
  • Walkthrough Branch A (DID): The pilot is rolled out in batches → judged as staggered → triggers the rule "must use Goodman-Bacon decomposition + Callaway-Sant'Anna." Illustrative results: the Bacon decomposition shows a "bad comparison" weight of 11% (illustrative); after switching to the CS estimator, the dynamic effect coefficient is 0.032 (illustrative), close to the TWFE estimate of 0.041 but more stable.
  • Parallel Trends: In the event-study plot, the coefficients for periods -5 to -1 are all insignificant (illustrative), and become significantly positive starting in period 2 after treatment.
  • Placebo: Randomly draw treatment timing and repeat 1,000 times; the true estimate falls outside the right tail 1% of the distribution (illustrative).
  • Journal Context Check: The subsidy pilot corresponds to the "Digital China" national strategy → passes the strategic relevance gate; the TFP improvement corresponds to "high-quality development" → the policy implication is clearly anchored.
  • Verdict: The identification strategy can proceed to the next stage (mw-mechanism-heterogeneity), but the institutional background chapter needs to supplement evidence on the exogeneity of pilot selection.

Reviewer Probing Patterns + Journal Context Revisions

  • Probe: "Were the pilot regions not already better in digital foundations?" → Journal revision: In the institutional background, cite the pilot selection documents to prove that the selection criteria were unrelated to firm TFP, and add a firm-level pre-trend balance table.
  • Probe: "Could the subsidy have simultaneously changed financing constraints, making the mechanism impure?" → Revision: Explicitly incorporate financing constraints as a competing mechanism into mw-mechanism-heterogeneity, rather than avoiding it.
  • Probe: "Why not use the more cutting-edge DML?" → In this journal's context, the editorial board prefers "identification logic that is clear and reproducible"; DML can be used as a robustness check rather than the main result, avoiding method for method's sake.

Calibration Anchors

The identification forms of empirical papers already published in this journal are mostly "national-level policy shock + staggered DID + heterogeneity-robust bias diagnostics + placebo + multiple robustness checks," and the identification chapter is strongly coupled with the institutional background chapter. New methods (DML, synthetic DID) usually appear in the robustness section rather than the main results. The specific annual format and method preferences are subject to the editorial office's latest manuscript guidelines.

Output Format

【识别策略】DID / IV / RDD / DML / 其他
【已完成检验】[平行趋势, 安慰剂, 弱工具, ...]
【缺失检验】[...]
【聚类层次】...
【国家战略相关性】强 / 中 / 弱
【下一步】mw-mechanism-heterogeneity