jm-data-analysis
GitHub用于执行《Journal of Marketing》稿件的统计分析与报告。确保符合JM严格的透明度要求，包括报告实际p值、标准误和效应量。根据研究设计选择合适估计量，并提供识别、机制检验及稳健性分析，最终生成可复现的数据包。
Trigger Scenarios
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill jm-data-analysis -g -y
SKILL.md
Frontmatter
{
"name": "jm-data-analysis",
"description": "Use when running and reporting the statistical analysis for a Journal of Marketing (JM) manuscript — the right estimator for a big-tent design, JM's exact p-value \/ standard-error \/ effect-size reporting mandate, identification and robustness, and the JM Dataverse replication packet. Executes and reports the analysis; it does not design the study (jm-methods) or frame the contribution (jm-contribution-framing)."
}
Data Analysis & Reporting (jm-data-analysis)
When to trigger
- Data are collected and it is time to estimate and report
- You are unsure your estimator matches a big-tent design (experiment, panel, choice, qualitative)
- A reviewer will probe identification, robustness, or whether the effect is managerially meaningful
- You reached conditional acceptance and must assemble the JM Dataverse replication packet
JM's hard reporting mandate (non-negotiable)
JM's submission rules bake in statistical transparency. Empirical papers must report:
- Actual p-values — not thresholds such as "p < .05" or stars-only tables.
- Standard errors for estimates.
- Effect sizes — the magnitude of the effect, because JM judges substantive and managerial importance, not mere significance.
Report these throughout the main text and tables. A results section that shows significance without magnitude fails JM's substantive bar: an effect that is "significant" but trivially small rarely changes a managerial decision.
Choose the estimator that matches the design
| Design / claim | Estimator |
|---|---|
| Randomized experiment (lab/online) | ANOVA / regression with manipulation & attention checks |
| Field experiment with a firm/platform | Regression on randomized treatment; cluster-robust SE |
| Process / mechanism | Mediation (bootstrapped indirect effects); moderation-of-process |
| Preferences / trade-offs | Choice models (logit/mixed logit), conjoint, hierarchical Bayes |
| Observational panel / market data | FE / high-dimensional FE; DiD / event study; synthetic control; IV/2SLS |
| Customer-base dynamics | CLV / hazard / count models on longitudinal data |
| Limited / count dependent variable | Logit/probit, Poisson/negative binomial, Tobit as fits |
| Qualitative | Transparent coding: codes → themes → constructs, audit trail |
Cluster standard errors to the design's randomization/sampling structure.
Identification, mechanism, and robustness
- Identification (secondary data): state the source of exogenous variation; defend parallel trends (DiD), instrument relevance/exclusion (IV), or the donor pool (synthetic control). Address endogeneity head-on.
- Mediation: report indirect effects with bias-corrected bootstrap CIs (e.g., 5,000 resamples); report conditional indirect effects for moderated mediation.
- Moderation: report the interaction coefficient and plot simple slopes; report incremental variance.
- Robustness: alternative specifications, measures, and subsamples; rule out alternative explanations empirically; report exclusions and missing-data handling explicitly.
Effect size in managerial units
Translate effects into terms a decision maker reads: elasticities, lift in sales/conversion/CLV, willingness-to-pay in currency, welfare changes. This is how JM's dual mandate shows up in the results section.
JM Dataverse replication packet (at conditional acceptance)
JM's Research Transparency Policy applies to conditionally accepted revisions of manuscripts submitted on/after 2023-01-01. At conditional acceptance, deposit to JM's Dataverse: raw data files, analysis programs/scripts, and any details sufficient to replicate all reported analyses; for qualitative work, interview guides, coding procedures, and annotated examples. The packet is accessible to the processing Editor, not reviewers. A Data Availability Statement is required on the title page of the final manuscript. Preregistration is encouraged — supply anonymized links and attest you faithfully represented the preregistration. Under the AMA Journals policy, some conditionally accepted JM manuscripts may also go through a verification step in which a Coeditor assigns a Data Editor to review the Dataverse materials and submit a ScholarOne report.
Execution bridge (StatsPAI / Stata MCP)
Run the battery, don't just enumerate it. Full map:
execution-with-mcp. JM is empirical marketing — field experiments, panel/CRM data, and quasi-experiments; randomization inference for experiments, DiD / IV for observational claims.
- Many outcomes / specifications:
romano_wolf(step-down FWER) orbenjamini_hochberg— report the adjusted threshold. - OVB sensitivity:
oster_delta/sensemakr. - Inference:
wild_cluster_bootstrap(few clusters),twoway_cluster/conley; multilevel data → cluster at the right level. - Re-fit off one handle:
audit_result(result_id)lists the missing checks and the exactsuggest_functionfor each. - Exhibits:
etable/did_summary_to_latexfrom the handle — no retyped numbers.
Keep the decisive checks in the body and the exhaustive battery in the appendix. See the executed chain in the JF execution walkthrough.
Checklist
- Actual p-values, standard errors, and effect sizes reported throughout
- Estimator matches the design; SEs clustered appropriately
- Identification strategy stated and defended (secondary data)
- Mediation via bootstrap CIs; moderation via simple slopes + incremental variance
- Effect sizes translated into managerial units (lift, elasticity, WTP, welfare)
- Robustness, alternative explanations, exclusions, and missing data reported
- Replication packet prepared for JM Dataverse; Data Availability Statement drafted
- If selected for verification, Data Editor questions can be answered from the README/codebook/code
- Preregistration links + attestation ready (if applicable)
Anti-patterns
- Significance without magnitude — stars-only tables; "p < .05" with no effect size.
- Estimator/design mismatch — e.g., OLS ignoring panel structure or clustering.
- Unaddressed endogeneity in observational data.
- Baron-Kenny causal-steps only instead of bootstrapped indirect effects.
- Retrofitted replication — code that does not regenerate the reported tables.
Output format
【Estimator】experiment / field / choice / panel-DiD / qualitative; SE clustering ...
【Exact statistics】p-values + SEs + effect sizes reported? yes/no
【Identification】strategy + defense (DiD/IV/synthetic control) ...
【Mediation/Moderation】bootstrap CI / simple slopes reported? ...
【Managerial magnitude】lift / elasticity / WTP / welfare ...
【Robustness】[...]
【JM Dataverse packet】data + code (+ qualitative materials) ready? Data Availability Statement?
【Next step】jm-contribution-framing
Version History
- 1839142 Current 2026-07-05 13:49
