Agent Skills
› brycewang-stanford/Awesome-Journal-Skills
› jcp-data-analysis
jcp-data-analysis
GitHub针对JCP期刊实验数据的统计分析技能，涵盖ANOVA、回归及中介/调节分析。强调过程机制验证、严谨性报告标准（效应量、置信区间）及因果链设计，旨在提供符合后严谨改革时代规范的统计证据。
Trigger Scenarios
效应显著但过程证据不足
中介分析被质疑相关性或效力低
调节作用复杂或简单效应未探测
需按JCP严谨标准报告结果
心理过程测量信效度存疑
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill jcp-data-analysis -g -y
SKILL.md
Frontmatter
{
"name": "jcp-data-analysis",
"description": "Use when analyzing experimental data for a Journal of Consumer Psychology (JCP) manuscript — ANOVA\/regression on the effect, measured and experimental mediation, moderation and moderated mediation, measurement of the process, and the rigor-era reporting standards. Analyzes the studies; it does not design them (jcp-methods)."
}
Data Analysis (jcp-data-analysis)
When to trigger
- Your effect is significant but the process evidence does not yet hold up
- You ran mediation but a reviewer calls it correlational or under-powered
- A moderation is predicted but the interaction is messy or the simple effects are not probed
- You need to report results to JCP's post-rigor-reform standards (effect sizes, CIs, exclusions)
- The measure of your psychological process is noisy or its validity is in question
Analyze the process, not just the p-value
JCP's contribution is a mechanism, so the analysis must make the process visible and defensible. The headline test of the effect (typically ANOVA or regression with the manipulated IV) is necessary but not sufficient; the paper lives or dies on whether the mediation/moderation evidence supports the proposed psychological process and rules out rivals. Report estimates with effect sizes and confidence intervals, exact statistics, and full Ns before and after pre-specified exclusions. APA reporting style is the house norm.
The analysis toolkit by link in the chain
| Link | Standard analysis | What reviewers look for |
|---|---|---|
| Existence of effect | t-test / ANOVA / OLS with the manipulated IV | clean cells, effect size (d, η²), CI, no covariate fishing |
| Measured mediation | bootstrapped indirect effect (e.g., PROCESS / lavaan), bias-corrected CI | indirect effect with CI excluding 0; honesty that this is correlational evidence on the mediator |
| Experimental mediation | causal-chain design or manipulated-mediator analysis | the manipulation of M moves Y as the theory predicts |
| Moderation | regression interaction; ANOVA factorial | interaction term + probed simple effects (spotlight/floodlight), not just a significant interaction |
| Moderated mediation | conditional indirect effects (index of moderated mediation) | the index, with CI, and conditional indirect effects by moderator level |
Prefer experimental/causal-chain mediation and moderation-of-process over measured-mediator-only inference: JCP reviewers now treat a bootstrapped indirect effect on a self-reported mediator as suggestive, not dispositive, because it cannot establish the causal direction of M → Y.
Measuring the psychological process
- Validate the mediator measure: report reliability (α/ω) for multi-item scales; show the measure captures the intended construct and discriminates from confounds (mood, arousal, difficulty).
- Rule out alternative mediators statistically: include rival process measures and show the focal mediator carries the effect when they are modeled together.
- Avoid mediator-as-manipulation-check confusion: a manipulation check is not a mediator; the mediator is the downstream mental state.
Rigor-era reporting (post-2010s consumer-psych reforms)
- Report exact test statistics, p-values, effect sizes, and CIs — not just "p < .05."
- Disclose all conditions and measures collected; do not hide arms (the disclosure norm).
- Report sample size determination and adherence to (or deviation from) the pre-registration.
- State exclusions and their rule transparently, with Ns before/after.
- Avoid asterisk-only tables; report the numbers a reader needs to assess the process.
Execution bridge (StatsPAI / Stata MCP)
Run the battery, don't just enumerate it. Full map:
execution-with-mcp. JCP is experimental consumer psychology; randomization inference, mediation done right (mediate, not naive controlling-away), and family-wise corrections matter most.
- 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
- Effect reported with exact stats, effect size, and CI; cells and Ns clear
- Mediation uses bootstrapped/bias-corrected CIs; measured-only mediation is labeled correlational
- At least one stronger-than-Baron-Kenny process test where the claim is causal
- Moderation: interaction reported and simple effects probed (spotlight/floodlight)
- Moderated mediation: index of moderated mediation + conditional indirect effects
- Mediator measure reliability reported; rival mediators modeled and ruled out
- Exclusions pre-specified; all conditions/measures disclosed; preregistration deviations noted
Anti-patterns
- Indirect-effect worship: a significant bootstrapped indirect effect treated as proof of causal process
- Interaction without simple effects: a significant interaction with no spotlight/floodlight probing
- Covariate fishing: adding controls until the effect appears, undisclosed
- Hidden arms: dropping conditions or DVs that didn't work without reporting them
- p-only reporting: asterisks instead of effect sizes and CIs
- Mediator confound: a "mediator" that is just mood/difficulty the manipulation also moved
Output format
【Effect】test, stat, effect size, CI, cell Ns
【Mediation】measured / experimental; indirect effect + CI; correlational caveat if measured-only
【Moderation】interaction + probed simple effects (spotlight/floodlight)
【Moderated mediation】index + conditional indirect effects (if applicable)
【Process measure】reliability + rival mediators ruled out
【Rigor disclosures】exclusions, all conditions/measures, preregistration deviations
【Next skill】jcp-contribution-framing
Version History
- 1839142 Current 2026-07-05 13:29
