[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7DKG07671969TYASQHB3VDJ":3,"skill-same:01M7DKG07671969TYASQHB3VDJ":60},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":16,"sourceHtmlUrl":17,"categories":18,"keywords":19,"publishedAt":20,"linkOnly":21,"markdownContent":22,"markdownBody":23,"frontMatter":24,"refName":25,"commitSha":26,"skillDirectory":27,"forksCount":14,"repositoryHtmlUrl":28,"repositoryHomepageUrl":6,"ownerHtmlUrl":29,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":30,"install":31,"resources":44},"01M7DKG07671969TYASQHB3VDJ","aer-identification",null,"Use when selecting, implementing, or stress-testing the causal identification strategy for an empirical economics manuscript — difference-in-differences (including staggered designs), instrumental variables (including weak-IV-robust inference), regression discontinuity, synthetic control, or shift-share \u002F Bartik. Apply before writing the introduction or results.","vc999999999\u002Fskillflux-catalog","vc999999999","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F104196942?v=4","01M7DKFQK2JRAY2C9EMD7FHBWG","jiangbing","01M7DKFQKQA5E0BWZ1WQY4G9WA",0,"MIT","https:\u002F\u002Fgithub.com\u002Fvc999999999\u002Fskillflux-catalog\u002Fblob\u002Fmain\u002Fintake\u002Fauto-empirical-research-skills-2026-10-09\u002Faer-skills\u002Fskills\u002Faer-identification\u002FLICENSE.txt","https:\u002F\u002Fgithub.com\u002Fvc999999999\u002Fskillflux-catalog\u002Fblob\u002Fmain\u002Fintake\u002Fauto-empirical-research-skills-2026-10-09\u002Faer-skills\u002Fskills\u002Faer-identification\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: aer-identification\ndescription: Use when selecting, implementing, or stress-testing the causal identification strategy for an empirical economics manuscript — difference-in-differences (including staggered designs), instrumental variables (including weak-IV-robust inference), regression discontinuity, synthetic control, or shift-share \u002F Bartik. Apply before writing the introduction or results.\n---\n\n# AER Identification\n\n## Overview\n\nIn AER-track empirical economics, **identification is the paper**. This skill routes among canonical designs, modern defaults, and referee-facing diagnostics.\n\nIf the design is fragile, return to `aer-topic-selection`; writing cannot save it.\n\n## When to Use\n\n- Designing the empirical strategy for a new project\n- The current strategy is TWFE \u002F first-stage F \u002F naive RDD and the referee will flag it\n- A prior submission was rejected on identification grounds and the design needs rebuilding\n- Choosing between two candidate identification strategies for the same question\n\n## Master Decision Tree\n\n```\nIs treatment assignment plausibly random conditional on observables?\n├── Yes, by design (RCT, lottery) → run the RCT analysis; register PAP via AEA RCT Registry\n└── No → identification must come from variation\n    ├── Sharp threshold in a running variable → RDD (sharp or fuzzy)\n    ├── Discrete policy change in some units, not others, over time → DiD\n    │     ├── Single treatment date → canonical 2×2 DiD\n    │     └── Staggered adoption → Callaway-Sant'Anna or Borusyak-Jaravel-Spiess\n    ├── Endogenous regressor + plausibly exogenous shifter → IV\n    │     ├── Shifter × pre-existing exposure shares → shift-share \u002F Bartik\n    │     └── Single instrument → weak-IV-robust inference if F \u003C 50\n    ├── One treated unit \u002F aggregate intervention → synthetic control\n    └── None of the above → reconsider the question\n```\n\n## Difference-in-Differences\n\n### Canonical 2×2 (single treatment date, two groups)\n\nUse TWFE if and only if:\n\n- Treatment timing is **simultaneous** for all treated units\n- The control group is **never treated**\n- Treatment-effect heterogeneity is implausible\n\nOtherwise, TWFE produces biased and often sign-flipped estimates.\n\n### Staggered Adoption (most modern applications)\n\n**Do not use TWFE.** Use one of:\n\n- **Callaway and Sant'Anna (2021)** — `csdid` (Stata), `did` (R). Identifies group-time average treatment effects (ATT(g,t)); estimands are doubly robust; supports event-study aggregation.\n- **Borusyak, Jaravel, and Spiess (2024)** — imputation estimator.\n- **de Chaisemartin and D'Haultfœuille (2020)** — `did_multiplegt`.\n- **Sun and Abraham (2021)** — interaction-weighted estimator for event studies.\n\n**Required diagnostics:**\n\n1. Goodman-Bacon decomposition to show the share of weight from \"forbidden\" comparisons under TWFE\n2. Event-study plot with the imputation or Callaway-Sant'Anna estimator\n3. Pre-trends test reported as the joint test, not just the visual\n4. Heterogeneity by treatment cohort\n\n### Pre-Trends\n\nA flat pre-trend is necessary but not sufficient. Report:\n\n- Visual event-study plot with 95% confidence intervals\n- Formal joint test of pre-period coefficients (p-value)\n- Honest DiD (Rambachan-Roth 2023) sensitivity bounds for the post-period\n\n## Instrumental Variables\n\n### Weak Instruments\n\nThe first-stage F > 10 rule is **obsolete**. Modern conventions:\n\n- Just-identified models: report **Anderson-Rubin (AR) confidence sets** as primary inference; AR keeps size under weak instruments.\n- For F \u003C 50: 2SLS confidence intervals are unreliable; AR is required, not optional.\n- Stock-Yogo TSLS-bias critical values assume homoskedasticity and rarely fit clustered settings.\n\nUse `weakivtest` (Stata), `ivDiag` (R), or the Olea-Pflueger effective F statistic.\n\n### Exclusion Restriction\n\nThe IV's credibility depends on a story, not a test. State the exclusion restriction in **one sentence** in the introduction and defend it with:\n\n- Institutional narrative (one paragraph)\n- A placebo regression where the instrument predicts an outcome it should not affect\n- Sensitivity analysis: how much exclusion-restriction violation would overturn the result (Conley et al. 2012)\n\n### Shift-Share \u002F Bartik\n\nTwo valid sources of identification, with very different implications:\n\n1. **Exogenous shares (Goldsmith-Pinkham, Sorkin, Swift 2020)** — argue that pre-existing exposure shares are conditionally exogenous; report the Rotemberg weights and inspect the top-5 industries driving identification.\n2. **Exogenous shocks (Borusyak, Hull, Jaravel 2022; Adão, Kolesár, Morales 2019)** — argue that aggregate shocks are as-good-as-random; report shock-level inference.\n\nPick one explicitly. Do not hand-wave between the two.\n\n## Regression Discontinuity\n\n### Modern Defaults\n\n- **Local linear regression with a triangular kernel.** Polynomials of order > 1 are discouraged (Gelman-Imbens 2019).\n- **MSE-optimal bandwidth (Calonico-Cattaneo-Titiunik 2014)** with the robust bias-corrected confidence interval. Use `rdrobust`.\n- **Donut RDD** if bunching near the cutoff is a concern.\n- **Covariate adjustment** for efficiency; main result must hold without it.\n\n### Required Diagnostics\n\n1. McCrary (2008) \u002F Cattaneo-Jansson-Ma (2020) density test for manipulation of the running variable\n2. Balance tests on predetermined covariates at the cutoff\n3. Placebo cutoffs away from the true threshold\n4. Bandwidth sensitivity — show the estimate across at least three bandwidths\n5. Visual RD plot using `rdplot` with the binning method explicitly stated\n\n## Synthetic Control\n\n### When Appropriate\n\n- One (or few) treated units\n- Long pre-treatment outcome series (≥ 10 periods)\n- A large donor pool of plausibly comparable untreated units\n- Aggregate intervention (policy at the country, state, city level)\n\n### Modern Extensions\n\n- **Generalized synthetic control (Xu 2017)** for multiple treated units\n- **Augmented synthetic control (Ben-Michael, Feller, Rothstein 2021)** for bias correction\n- **Synthetic DiD (Arkhangelsky et al. 2021)** combining SCM and DiD weighting\n\n### Required Diagnostics\n\n1. Placebo (in-time): apply SCM to pre-treatment fake intervention dates\n2. Placebo (in-space): apply SCM to every donor as if it were treated; report the distribution of placebo effects\n3. Permutation inference \u002F Fisher exact p-value\n4. Weight vector reported in the appendix; donors with > 10% weight discussed\n\n## Field Experiments and RCTs\n\nIf the paper uses a field experiment:\n\n- **Register with AEA RCT Registry** before the intervention begins. AEA journals require this prior to submission.\n- **Pre-analysis plan (PAP)** posted before unblinding. Per Olken and others, keep the PAP moderate in scope — pre-specify primary outcomes and the analysis specification, leave exploratory work clearly labeled as such.\n- **Power calculations** in the manuscript or appendix.\n- **Multiple-hypothesis correction** if more than one primary outcome.\n- **Attrition** documented and tested for differential attrition by treatment arm.\n\n## Mechanism vs. Identification\n\nA common confusion: **identification answers whether X causes Y; mechanism answers why.** Mechanism evidence should not weaken the identification of the main effect. Run:\n\n- Subgroup heterogeneity (does the effect concentrate where theory predicts?)\n- Mediation analysis only if the mediator is itself plausibly exogenous (rare)\n- Auxiliary outcomes consistent with the proposed channel\n\n## Red Flags for Referees\n\n- TWFE on staggered data with no Goodman-Bacon decomposition\n- First-stage F = 12 cited as evidence of instrument strength\n- RDD with a polynomial of order 4\n- Synthetic control with no placebo inference\n- DiD with a \"control group\" of eventually-treated units\n- IV exclusion restriction defended only by \"we control for X\"\n- Quoting an Angrist-Pischke citation as a substitute for showing the diagnostic\n\n## StatsPAI Tool Bindings\n\n\u003C!-- tool-bindings -->\nWhen a StatsPAI MCP server is connected, **select the validated tool, then let it\nrun the estimator** — do not hand-roll the design. The methodological *choice*\nstill comes from the decision tree above; this table is only the execution\nsurface. The full registry and chained workflow live in\n`skills\u002Faer-statspai\u002FSKILL.md`.\n\n| Design \u002F diagnostic | Call (StatsPAI) | Do not hand-roll |\n|---|---|---|\n| Staggered DiD ATT(g,t) | `callaway_santanna` then `aggte` | a pooled two-way fixed-effects regression |\n| Imputation \u002F dCDH variants | `did_imputation`, `sun_abraham`, `did_multiplegt` | event-study leads and lags assembled by hand on staggered data |\n| Forbidden-comparison weight | `bacon_decomposition` | eyeballing whether TWFE is \"probably fine\" |\n| Pre-trends (joint, not visual) | `event_study`, `pretrends_test` | declaring parallel trends from a plot alone |\n| IV under weak instruments | `ivreg` with `anderson_rubin_ci`, `effective_f_test` | a first-stage-F-only 2SLS table |\n| Shift-share \u002F Bartik | `bartik` | a Bartik IV with no Rotemberg-weight report |\n| RDD (local-linear, RBC) | `rdrobust`, `rdbwselect`, `rdplot`, `rddensity` | a high-order global-polynomial RDD |\n| Synthetic control | `synth`, `augsynth`, `gsynth`, `sdid`, `synth_time_placebo`, `synth_loo` | SCM with no placebo or leave-one-out inference |\n| Design \u002F estimator selection | `detect_design`, `preflight`, `recommend` | guessing the estimator before profiling the data |\n\u003C!-- \u002Ftool-bindings -->\n\n## Repository Resources\n\nBundled with the installed skill, no repository checkout needed --- read it\nbefore the repo resources below:\n\n- `references\u002Festimator-playbook.md` --- per-design estimands, modern defaults, diagnostics, and referee-objection response scripts\n\nWhen working from the repo or plugin bundle, load only the relevant resource:\n\n- Estimator defaults, package calls, diagnostics, and citations: `docs\u002Fmethods-reference.md`\n- Staggered DiD implementation: `templates\u002Fstata\u002F03_main_did.do`, `templates\u002Fr\u002F03_main_did.R`, or `templates\u002Fpython\u002Fmain_did.py`\n- Worked empirical examples: `examples\u002Faer-exemplars.md` and `examples\u002Fmodern-aer-exemplars.md`\n\nUse the methods reference before prose: it fixes the estimand,\ndiagnostic, inference method, and citation that the manuscript must report.\n\n## Identification Gate\n\nDo not advance to robustness or writing until, for the chosen design, **all** are true:\n\n- [ ] A modern estimator is used — no TWFE on staggered data, no first-stage-F-only IV, no high-order-polynomial RDD\n- [ ] Every required diagnostic for the design (see the per-design lists above) is run and reported\n- [ ] Inference matches the design — cluster-robust \u002F AR \u002F wild bootstrap \u002F permutation, not default OLS SEs by reflex\n- [ ] The identifying assumption is stated in one sentence, ready to drop into the introduction\n- [ ] No item in \"Red Flags for Referees\" is present\n\n### Gate Record Mini-Example\n\nWrite the gate decision before routing onward:\n\n```text\nSTRATEGY: IV\nFIRST STAGE: effective F = 7.8; 2SLS CI is not primary\nROBUST INFERENCE: AR 95% CI = [-0.14, 0.52]\nPLACEBO: beta = 0.003 (p = 0.71)\nDECISION: advance with directional headline only\n```\n\n## Handoff\n\n```text\nSTRATEGY: \u003CDiD | IV | RDD | SCM | shift-share | RCT>\nMODERN ESTIMATOR USED: \u003Cyes \u002F no \u002F which>\nREQUIRED DIAGNOSTICS REPORTED: \u003Clist>\nINFERENCE METHOD: \u003Crobust \u002F cluster-robust \u002F AR \u002F wild bootstrap \u002F permutation>\nWEAK-IV \u002F TWFE \u002F POLY-ORDER RED FLAGS: \u003Clist or \"none\">\nNEXT SKILL: aer-robustness\n```\n\n## Anti-Patterns\n\n- Defending an old design (\"the prior literature used TWFE\") when modern estimators exist\n- Reporting OLS-with-controls as the main specification and IV\u002FRD as \"robustness\"\n- Using more than one identification strategy as if they were independent confirmations when they share identifying variation\n- Footnoting the identifying assumption instead of stating it in the introduction\n","# AER Identification\n\n## Overview\n\nIn AER-track empirical economics, **identification is the paper**. This skill routes among canonical designs, modern defaults, and referee-facing diagnostics.\n\nIf the design is fragile, return to `aer-topic-selection`; writing cannot save it.\n\n## When to Use\n\n- Designing the empirical strategy for a new project\n- The current strategy is TWFE \u002F first-stage F \u002F naive RDD and the referee will flag it\n- A prior submission was rejected on identification grounds and the design needs rebuilding\n- Choosing between two candidate identification strategies for the same question\n\n## Master Decision Tree\n\n```\nIs treatment assignment plausibly random conditional on observables?\n├── Yes, by design (RCT, lottery) → run the RCT analysis; register PAP via AEA RCT Registry\n└── No → identification must come from variation\n    ├── Sharp threshold in a running variable → RDD (sharp or fuzzy)\n    ├── Discrete policy change in some units, not others, over time → DiD\n    │     ├── Single treatment date → canonical 2×2 DiD\n    │     └── Staggered adoption → Callaway-Sant'Anna or Borusyak-Jaravel-Spiess\n    ├── Endogenous regressor + plausibly exogenous shifter → IV\n    │     ├── Shifter × pre-existing exposure shares → shift-share \u002F Bartik\n    │     └── Single instrument → weak-IV-robust inference if F \u003C 50\n    ├── One treated unit \u002F aggregate intervention → synthetic control\n    └── None of the above → reconsider the question\n```\n\n## Difference-in-Differences\n\n### Canonical 2×2 (single treatment date, two groups)\n\nUse TWFE if and only if:\n\n- Treatment timing is **simultaneous** for all treated units\n- The control group is **never treated**\n- Treatment-effect heterogeneity is implausible\n\nOtherwise, TWFE produces biased and often sign-flipped estimates.\n\n### Staggered Adoption (most modern applications)\n\n**Do not use TWFE.** Use one of:\n\n- **Callaway and Sant'Anna (2021)** — `csdid` (Stata), `did` (R). Identifies group-time average treatment effects (ATT(g,t)); estimands are doubly robust; supports event-study aggregation.\n- **Borusyak, Jaravel, and Spiess (2024)** — imputation estimator.\n- **de Chaisemartin and D'Haultfœuille (2020)** — `did_multiplegt`.\n- **Sun and Abraham (2021)** — interaction-weighted estimator for event studies.\n\n**Required diagnostics:**\n\n1. Goodman-Bacon decomposition to show the share of weight from \"forbidden\" comparisons under TWFE\n2. Event-study plot with the imputation or Callaway-Sant'Anna estimator\n3. Pre-trends test reported as the joint test, not just the visual\n4. Heterogeneity by treatment cohort\n\n### Pre-Trends\n\nA flat pre-trend is necessary but not sufficient. Report:\n\n- Visual event-study plot with 95% confidence intervals\n- Formal joint test of pre-period coefficients (p-value)\n- Honest DiD (Rambachan-Roth 2023) sensitivity bounds for the post-period\n\n## Instrumental Variables\n\n### Weak Instruments\n\nThe first-stage F > 10 rule is **obsolete**. Modern conventions:\n\n- Just-identified models: report **Anderson-Rubin (AR) confidence sets** as primary inference; AR keeps size under weak instruments.\n- For F \u003C 50: 2SLS confidence intervals are unreliable; AR is required, not optional.\n- Stock-Yogo TSLS-bias critical values assume homoskedasticity and rarely fit clustered settings.\n\nUse `weakivtest` (Stata), `ivDiag` (R), or the Olea-Pflueger effective F statistic.\n\n### Exclusion Restriction\n\nThe IV's credibility depends on a story, not a test. State the exclusion restriction in **one sentence** in the introduction and defend it with:\n\n- Institutional narrative (one paragraph)\n- A placebo regression where the instrument predicts an outcome it should not affect\n- Sensitivity analysis: how much exclusion-restriction violation would overturn the result (Conley et al. 2012)\n\n### Shift-Share \u002F Bartik\n\nTwo valid sources of identification, with very different implications:\n\n1. **Exogenous shares (Goldsmith-Pinkham, Sorkin, Swift 2020)** — argue that pre-existing exposure shares are conditionally exogenous; report the Rotemberg weights and inspect the top-5 industries driving identification.\n2. **Exogenous shocks (Borusyak, Hull, Jaravel 2022; Adão, Kolesár, Morales 2019)** — argue that aggregate shocks are as-good-as-random; report shock-level inference.\n\nPick one explicitly. Do not hand-wave between the two.\n\n## Regression Discontinuity\n\n### Modern Defaults\n\n- **Local linear regression with a triangular kernel.** Polynomials of order > 1 are discouraged (Gelman-Imbens 2019).\n- **MSE-optimal bandwidth (Calonico-Cattaneo-Titiunik 2014)** with the robust bias-corrected confidence interval. Use `rdrobust`.\n- **Donut RDD** if bunching near the cutoff is a concern.\n- **Covariate adjustment** for efficiency; main result must hold without it.\n\n### Required Diagnostics\n\n1. McCrary (2008) \u002F Cattaneo-Jansson-Ma (2020) density test for manipulation of the running variable\n2. Balance tests on predetermined covariates at the cutoff\n3. Placebo cutoffs away from the true threshold\n4. Bandwidth sensitivity — show the estimate across at least three bandwidths\n5. Visual RD plot using `rdplot` with the binning method explicitly stated\n\n## Synthetic Control\n\n### When Appropriate\n\n- One (or few) treated units\n- Long pre-treatment outcome series (≥ 10 periods)\n- A large donor pool of plausibly comparable untreated units\n- Aggregate intervention (policy at the country, state, city level)\n\n### Modern Extensions\n\n- **Generalized synthetic control (Xu 2017)** for multiple treated units\n- **Augmented synthetic control (Ben-Michael, Feller, Rothstein 2021)** for bias correction\n- **Synthetic DiD (Arkhangelsky et al. 2021)** combining SCM and DiD weighting\n\n### Required Diagnostics\n\n1. Placebo (in-time): apply SCM to pre-treatment fake intervention dates\n2. Placebo (in-space): apply SCM to every donor as if it were treated; report the distribution of placebo effects\n3. Permutation inference \u002F Fisher exact p-value\n4. Weight vector reported in the appendix; donors with > 10% weight discussed\n\n## Field Experiments and RCTs\n\nIf the paper uses a field experiment:\n\n- **Register with AEA RCT Registry** before the intervention begins. AEA journals require this prior to submission.\n- **Pre-analysis plan (PAP)** posted before unblinding. Per Olken and others, keep the PAP moderate in scope — pre-specify primary outcomes and the analysis specification, leave exploratory work clearly labeled as such.\n- **Power calculations** in the manuscript or appendix.\n- **Multiple-hypothesis correction** if more than one primary outcome.\n- **Attrition** documented and tested for differential attrition by treatment arm.\n\n## Mechanism vs. Identification\n\nA common confusion: **identification answers whether X causes Y; mechanism answers why.** Mechanism evidence should not weaken the identification of the main effect. Run:\n\n- Subgroup heterogeneity (does the effect concentrate where theory predicts?)\n- Mediation analysis only if the mediator is itself plausibly exogenous (rare)\n- Auxiliary outcomes consistent with the proposed channel\n\n## Red Flags for Referees\n\n- TWFE on staggered data with no Goodman-Bacon decomposition\n- First-stage F = 12 cited as evidence of instrument strength\n- RDD with a polynomial of order 4\n- Synthetic control with no placebo inference\n- DiD with a \"control group\" of eventually-treated units\n- IV exclusion restriction defended only by \"we control for X\"\n- Quoting an Angrist-Pischke citation as a substitute for showing the diagnostic\n\n## StatsPAI Tool Bindings\n\n\u003C!-- tool-bindings -->\nWhen a StatsPAI MCP server is connected, **select the validated tool, then let it\nrun the estimator** — do not hand-roll the design. The methodological *choice*\nstill comes from the decision tree above; this table is only the execution\nsurface. The full registry and chained workflow live in\n`skills\u002Faer-statspai\u002FSKILL.md`.\n\n| Design \u002F diagnostic | Call (StatsPAI) | Do not hand-roll |\n|---|---|---|\n| Staggered DiD ATT(g,t) | `callaway_santanna` then `aggte` | a pooled two-way fixed-effects regression |\n| Imputation \u002F dCDH variants | `did_imputation`, `sun_abraham`, `did_multiplegt` | event-study leads and lags assembled by hand on staggered data |\n| Forbidden-comparison weight | `bacon_decomposition` | eyeballing whether TWFE is \"probably fine\" |\n| Pre-trends (joint, not visual) | `event_study`, `pretrends_test` | declaring parallel trends from a plot alone |\n| IV under weak instruments | `ivreg` with `anderson_rubin_ci`, `effective_f_test` | a first-stage-F-only 2SLS table |\n| Shift-share \u002F Bartik | `bartik` | a Bartik IV with no Rotemberg-weight report |\n| RDD (local-linear, RBC) | `rdrobust`, `rdbwselect`, `rdplot`, `rddensity` | a high-order global-polynomial RDD |\n| Synthetic control | `synth`, `augsynth`, `gsynth`, `sdid`, `synth_time_placebo`, `synth_loo` | SCM with no placebo or leave-one-out inference |\n| Design \u002F estimator selection | `detect_design`, `preflight`, `recommend` | guessing the estimator before profiling the data |\n\u003C!-- \u002Ftool-bindings -->\n\n## Repository Resources\n\nBundled with the installed skill, no repository checkout needed --- read it\nbefore the repo resources below:\n\n- `references\u002Festimator-playbook.md` --- per-design estimands, modern defaults, diagnostics, and referee-objection response scripts\n\nWhen working from the repo or plugin bundle, load only the relevant resource:\n\n- Estimator defaults, package calls, diagnostics, and citations: `docs\u002Fmethods-reference.md`\n- Staggered DiD implementation: `templates\u002Fstata\u002F03_main_did.do`, `templates\u002Fr\u002F03_main_did.R`, or `templates\u002Fpython\u002Fmain_did.py`\n- Worked empirical examples: `examples\u002Faer-exemplars.md` and `examples\u002Fmodern-aer-exemplars.md`\n\nUse the methods reference before prose: it fixes the estimand,\ndiagnostic, inference method, and citation that the manuscript must report.\n\n## Identification Gate\n\nDo not advance to robustness or writing until, for the chosen design, **all** are true:\n\n- [ ] A modern estimator is used — no TWFE on staggered data, no first-stage-F-only IV, no high-order-polynomial RDD\n- [ ] Every required diagnostic for the design (see the per-design lists above) is run and reported\n- [ ] Inference matches the design — cluster-robust \u002F AR \u002F wild bootstrap \u002F permutation, not default OLS SEs by reflex\n- [ ] The identifying assumption is stated in one sentence, ready to drop into the introduction\n- [ ] No item in \"Red Flags for Referees\" is present\n\n### Gate Record Mini-Example\n\nWrite the gate decision before routing onward:\n\n```text\nSTRATEGY: IV\nFIRST STAGE: effective F = 7.8; 2SLS CI is not primary\nROBUST INFERENCE: AR 95% CI = [-0.14, 0.52]\nPLACEBO: beta = 0.003 (p = 0.71)\nDECISION: advance with directional headline only\n```\n\n## Handoff\n\n```text\nSTRATEGY: \u003CDiD | IV | RDD | SCM | shift-share | RCT>\nMODERN ESTIMATOR USED: \u003Cyes \u002F no \u002F which>\nREQUIRED DIAGNOSTICS REPORTED: \u003Clist>\nINFERENCE METHOD: \u003Crobust \u002F cluster-robust \u002F AR \u002F wild bootstrap \u002F permutation>\nWEAK-IV \u002F TWFE \u002F POLY-ORDER RED FLAGS: \u003Clist or \"none\">\nNEXT SKILL: aer-robustness\n```\n\n## Anti-Patterns\n\n- Defending an old design (\"the prior literature used TWFE\") when modern estimators exist\n- Reporting OLS-with-controls as the main specification and IV\u002FRD as \"robustness\"\n- Using more than one identification strategy as if they were independent confirmations when they share identifying variation\n- Footnoting the identifying assumption instead of stating it in the introduction\n",{"name":5,"description":7},"main","18f52829480a4e8df7dd5d9c13ce3b5284311891","intake\u002Fauto-empirical-research-skills-2026-10-09\u002Faer-skills\u002Fskills\u002Faer-identification","https:\u002F\u002Fgithub.com\u002Fvc999999999\u002Fskillflux-catalog","https:\u002F\u002Fgithub.com\u002Fvc999999999","2026-10-11 20:16:46",{"source":8,"skillName":5,"commands":32,"prompt":42,"zipAvailable":43},[33,36,39],{"tool":34,"command":35},"npx","npx skills add vc999999999\u002Fskillflux-catalog --skill aer-identification",{"tool":37,"command":38},"bunx","bunx skills add vc999999999\u002Fskillflux-catalog --skill aer-identification",{"tool":40,"command":41},"pnpm","pnpm dlx skills add vc999999999\u002Fskillflux-catalog --skill aer-identification","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fvc999999999\u002Fskillflux-catalog\u002Fblob\u002Fmain\u002Fintake\u002Fauto-empirical-research-skills-2026-10-09\u002Faer-skills\u002Fskills\u002Faer-identification\u002FSKILL.md",true,[45,51,56],{"path":46,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":49,"sourceHtmlUrl":50},"agents\u002Fopenai.yaml","REFERENCE","参考文档",212,"https:\u002F\u002Fgithub.com\u002Fvc999999999\u002Fskillflux-catalog\u002Fblob\u002Fmain\u002Fintake\u002Fauto-empirical-research-skills-2026-10-09\u002Faer-skills\u002Fskills\u002Faer-identification\u002Fagents\u002Fopenai.yaml",{"path":52,"resourceType":53,"resourceTypeDesc":54,"sizeBytes":55,"sourceHtmlUrl":16},"LICENSE.txt","LICENSE","许可证",1067,{"path":57,"resourceType":47,"resourceTypeDesc":48,"sizeBytes":58,"sourceHtmlUrl":59},"references\u002Festimator-playbook.md",17337,"https:\u002F\u002Fgithub.com\u002Fvc999999999\u002Fskillflux-catalog\u002Fblob\u002Fmain\u002Fintake\u002Fauto-empirical-research-skills-2026-10-09\u002Faer-skills\u002Fskills\u002Faer-identification\u002Freferences\u002Festimator-playbook.md",[61,69,77,85,93,101],{"skillNo":62,"name":63,"nameZh":6,"description":64,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":65,"sourceHtmlUrl":66,"categories":67,"keywords":68,"publishedAt":20},"01M7DKTKB8RM9NFNA9VD5QNM59","scientific-schematics","Generates scientific diagram drafts using Nano Banana 2 AI with smart iterative refinement. Uses Gemini 3.7 Flash for quality review. Refines when the review requests improvement, with at most two generations. Specialized in neural network architectures, system diagrams, flowcharts, biological pathways, and complex scientific visualizations.","https:\u002F\u002Fgithub.com\u002Fvc999999999\u002Fskillflux-catalog\u002Fblob\u002Fmain\u002Fskills\u002Fresearch\u002Fscientific-schematics\u002F1.0.0\u002FLICENSE.md","https:\u002F\u002Fgithub.com\u002Fvc999999999\u002Fskillflux-catalog\u002Fblob\u002Fmain\u002Fskills\u002Fresearch\u002Fscientific-schematics\u002F1.0.0\u002FSKILL.md",[],[],{"skillNo":70,"name":71,"nameZh":6,"description":72,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":73,"sourceHtmlUrl":74,"categories":75,"keywords":76,"publishedAt":20},"01M7DKTE1BMWDNRQ6JHSB8YZVM","scientific-critical-thinking","Evaluates scientific claims and evidence quality. Applies to experimental design validity, biases and confounders, statistical interpretation, evidence grading frameworks (GRADE, Cochrane Risk of Bias), and teaching critical analysis. Supports evidence appraisal and identifying flaws; formal peer review writing belongs to peer-review.","https:\u002F\u002Fgithub.com\u002Fvc999999999\u002Fskillflux-catalog\u002Fblob\u002Fmain\u002Fskills\u002Fresearch\u002Fscientific-critical-thinking\u002F1.0.0\u002FLICENSE.md","https:\u002F\u002Fgithub.com\u002Fvc999999999\u002Fskillflux-catalog\u002Fblob\u002Fmain\u002Fskills\u002Fresearch\u002Fscientific-critical-thinking\u002F1.0.0\u002FSKILL.md",[],[],{"skillNo":78,"name":79,"nameZh":6,"description":80,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":81,"sourceHtmlUrl":82,"categories":83,"keywords":84,"publishedAt":20},"01M7DKT74MD9SFF7K4ZGQJ5PAY","scientific-brainstorming","Facilitates evidence-aware scientific ideation with independent generation, structured discussion, explicit assumptions, transparent evaluation, adversarial review, and decision logs. Use for early-stage research brainstorming or prioritizing candidate directions; hand off empirical validation, study design, ethics or regulatory review, and clinical questions to appropriate experts or skills.","https:\u002F\u002Fgithub.com\u002Fvc999999999\u002Fskillflux-catalog\u002Fblob\u002Fmain\u002Fskills\u002Fresearch\u002Fscientific-brainstorming\u002F1.0.0\u002FLICENSE.md","https:\u002F\u002Fgithub.com\u002Fvc999999999\u002Fskillflux-catalog\u002Fblob\u002Fmain\u002Fskills\u002Fresearch\u002Fscientific-brainstorming\u002F1.0.0\u002FSKILL.md",[],[],{"skillNo":86,"name":87,"nameZh":6,"description":88,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":89,"sourceHtmlUrl":90,"categories":91,"keywords":92,"publishedAt":20},"01M7DKT1837E2PSV5Y6EQV4A4B","aer-tables-figures","Use when constructing or revising regression tables, descriptive statistics tables, or figures after results are estimated and before submission for an AER, AER:Insights, or AEJ manuscript. Implements AER booktabs house style, regression-table layout, and figure-note conventions.","https:\u002F\u002Fgithub.com\u002Fvc999999999\u002Fskillflux-catalog\u002Fblob\u002Fmain\u002Fskills\u002Fresearch\u002Faer-tables-figures\u002F1.0.0\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fvc999999999\u002Fskillflux-catalog\u002Fblob\u002Fmain\u002Fskills\u002Fresearch\u002Faer-tables-figures\u002F1.0.0\u002FSKILL.md",[],[],{"skillNo":94,"name":95,"nameZh":6,"description":96,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":97,"sourceHtmlUrl":98,"categories":99,"keywords":100,"publishedAt":20},"01M7DKSW7MJQ8A88TJC1H293PB","aer-statspai","Use when aer-identification has fixed the design, after methodology choice and before aer-robustness or aer-tables-figures, to run an AER-track analysis with StatsPAI — the agent-native Python engine and MCP server for causal inference, robustness, sensitivity, and publication-ready table export.","https:\u002F\u002Fgithub.com\u002Fvc999999999\u002Fskillflux-catalog\u002Fblob\u002Fmain\u002Fskills\u002Fresearch\u002Faer-statspai\u002F1.0.0\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fvc999999999\u002Fskillflux-catalog\u002Fblob\u002Fmain\u002Fskills\u002Fresearch\u002Faer-statspai\u002F1.0.0\u002FSKILL.md",[],[],{"skillNo":102,"name":103,"nameZh":6,"description":104,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":105,"sourceHtmlUrl":106,"categories":107,"keywords":108,"publishedAt":20},"01M7DKSQKV1429VSPVVHAM1QQM","aer-robustness","Use when the main empirical results exist but the manuscript lacks the robustness, heterogeneity, mechanism, and placebo checks that AER referees will demand. Apply after aer-identification and before aer-introduction so that the value-added paragraph can reference these tests.","https:\u002F\u002Fgithub.com\u002Fvc999999999\u002Fskillflux-catalog\u002Fblob\u002Fmain\u002Fskills\u002Fresearch\u002Faer-robustness\u002F1.0.0\u002FLICENSE","https:\u002F\u002Fgithub.com\u002Fvc999999999\u002Fskillflux-catalog\u002Fblob\u002Fmain\u002Fskills\u002Fresearch\u002Faer-robustness\u002F1.0.0\u002FSKILL.md",[],[]]