Agent Skills
› brycewang-stanford/Awesome-Journal-Skills
› jcp-methods
jcp-methods
GitHub针对JCP期刊的实验设计与方法技能，涵盖操纵检验、混淆控制、先验功效分析及多研究因果链构建。用于隔离心理机制、强化过程证据及满足高严谨性要求，不涉数据分析。
Trigger Scenarios
不确定实验能否隔离机制
审稿人质疑操纵有效性
中介效应缺乏强过程证据
样本量未进行先验功效分析
决定预注册或注册报告
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill jcp-methods -g -y
SKILL.md
Frontmatter
{
"name": "jcp-methods",
"description": "Use when designing or stress-testing the experiments for a Journal of Consumer Psychology (JCP) manuscript — manipulations, controls and confounds, manipulation checks, power, the multi-study chain, and pre-registration. Designs the studies that test the process; it does not analyze them (jcp-data-analysis)."
}
Experimental Design & Methods (jcp-methods)
When to trigger
- You have a mechanism but are unsure which experiments can actually isolate it
- A reviewer questions whether your manipulation moves the construct it claims to
- Mediation rests on a single measured mediator and you need stronger process evidence
- Sample sizes were chosen by convention, not by an a priori power analysis
- You are deciding whether to pre-register, or whether to take the Registered Report route (待核实)
JCP's design bar: causal isolation of a psychological process
JCP is an experimental journal: random assignment, manipulated independent variables, and measured (or manipulated) mediators are the default. The design must do more than show an effect — it must rule out that the effect is anything other than the proposed process. Since the rigor reforms of the 2010s, JCP reviewers expect adequately powered cells, clean manipulations with checks, pre-registered confirmatory studies where feasible, and a multi-study package whose studies each test a different link in the causal chain rather than re-running the same demonstration.
The multi-study chain
A persuasive JCP package walks the mechanism:
- Existence — establish the effect cleanly with a strong manipulation and a behavioral or consequential DV where possible.
- Mediation — show the process carries the effect. A measured mediator is the floor; manipulating the mediator (or moderation-of-process) is far stronger and increasingly expected.
- Moderation — a theory-predicted moderator that switches the process on/off; this is the most convincing process evidence.
- Boundary / robustness — generalize across stimuli, populations, and operationalizations so the effect is not stimulus-bound.
Vary the operationalization across studies (different manipulations of the same construct, different DVs, different samples) so a reviewer cannot attribute the result to one idiosyncratic stimulus.
Manipulations, controls, and confounds
- Manipulation checks: include a check that the IV moved the intended construct and nothing else. A manipulation that also shifts mood, difficulty, or attention is confounded.
- Pretest stimuli: pretest scenarios/images/copy so conditions differ only on the focal dimension.
- Confound audit: ask of every manipulation, "what else changed?" — fluency, plausibility, arousal, social desirability, demand. Design controls (yoked conditions, neutral comparisons) before collecting data.
- Demand and attention: attention/comprehension checks, funnel debrief for hypothesis guessing, and a design that does not telegraph the prediction.
- Process-of-mediation: prefer causal-chain (experimental-mediation) or moderation-of-process designs over Baron-Kenny on a measured mediator, which reviewers now treat as weak causal evidence.
Power and samples
- Run an a priori power analysis (effect-size assumption justified by pilot or prior literature, not back-solved) and report target N before data collection.
- Pre-specify exclusion rules (attention checks, completion, duplicates) in advance; report Ns before and after exclusions.
- For online panels (Prolific/MTurk/CloudResearch), document the platform, screening, and data-quality safeguards.
Pre-registration and transparency as design choices
- Pre-register confirmatory studies (OSF/AsPredicted): hypotheses, design, sample size, analysis plan, exclusions. JCP encourages preregistration with an analysis plan (检索于 2026-06；以官网为准); report deviations transparently.
- Plan the open data + materials deposit and the code from the start (see jcp-submission), not at acceptance.
- Consider a Registered Report for a strong confirmatory theory test where the result should publish regardless of direction (待核实 whether JCP currently offers this format).
Execution bridge (StatsPAI / Stata MCP)
For the empirical / causal lane, estimate and audit rather than only specify. 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.
detect_design→recommend→ fit withas_handle=true→audit_resultto enumerate the checks the design owes.- Panel / staggered DiD:
callaway_santanna/sun_abraham+bacon_decompositionhonest_did_from_result. IV:effective_f_test+anderson_rubin_ci. RDD:rdrobust+mccrary_test.
- Experiments: randomization-based inference and
romano_wolffor the many-outcome family-wise correction reviewers expect.
Match the toolchain to the reviewer pool, and report the effect size the venue wants. A run end-to-end (synthetic data, real returns) is in the JF execution walkthrough.
Checklist
- Each study tests a distinct link: existence → mediation → moderation → boundary
- At least one strong process design (manipulated mediator or moderation-of-process), not only measured mediation
- Manipulation checks confirm the IV moved the construct and not a confound
- Stimuli pretested; conditions differ only on the focal dimension
- A priori power analysis; target N and exclusion rules pre-specified
- Confirmatory studies pre-registered; deviations to be reported
- Materials, data, and code deposit planned now
Anti-patterns
- One study, big claim: a single experiment asked to carry a process contribution
- Measured-mediation-only: Baron-Kenny on a self-report mediator presented as causal process
- Confounded manipulation: the IV also shifts mood/difficulty/arousal and there is no check
- Convenience N: 50/cell because "that's what we usually run," with no power plan
- Stimulus-bound effect: one scenario, one product, generalized to "consumers"
- Post-hoc exclusions: dropping participants until p < .05 (a rigor-era red flag)
Output format
【Study chain】existence → mediation → moderation → boundary (map studies)
【Process design】measured mediator / manipulated mediator / moderation-of-process
【Manipulation + check】IV, the construct it moves, the check, the confounds controlled
【Power & samples】a priori N per cell, exclusion rules, platform
【Pre-registration】which studies, registry, what is locked
【Transparency plan】data + materials + code deposit
【Next skill】jcp-data-analysis
Version History
- 1839142 Current 2026-07-05 13:29
