Agent Skills
› brycewang-stanford/Awesome-Journal-Skills
› jms-data-analysis
jms-data-analysis
GitHub针对JMS论文，执行并辩护定量（回归/SEM/稳健性）或定性（编码/溯因/可信度）分析。解决内生性、稳健性及数据到构念的追溯性问题，确保分析结果能有力支持理论主张。
Trigger Scenarios
审稿人质疑内生性或稳健性
定性研究缺乏可审计的数据路径
中介/调节效应未按要求分析
审稿人认为分析不支持论点
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill jms-data-analysis -g -y
SKILL.md
Frontmatter
{
"name": "jms-data-analysis",
"description": "Use when the execution and credibility of the analysis is the bottleneck for a Journal of Management Studies (JMS) manuscript — regression\/SEM and robustness for quantitative work, OR coding, abduction, and trustworthiness for qualitative work. Runs and defends the analysis; it does not design the study (jms-methods) or build exhibits (jms-tables-figures)."
}
Data Analysis (jms-data-analysis)
When to trigger
- Estimates are in but reviewers question endogeneity, robustness, or the indirect-effect claim
- A qualitative analysis reaches findings but the path from data to constructs is not auditable
- Effects hinge on a single specification with no robustness
- A mediation/moderation result is reported without the analysis JMS expects
- A reviewer says "the analysis does not support the claim" or "I can't see how you got here"
The JMS analysis bar — two idioms, one standard
JMS judges analysis by whether it credibly supports the theoretical claim, in whichever idiom the study uses. Quantitative work is held to identification and robustness standards; qualitative work is held to trustworthiness and transparency standards. Use the path that matches your design; do not import quant criteria (p-values, effect sizes) to judge a qualitative paper, or qualitative looseness into a quantitative one.
Quantitative path
- Specification & estimator: match the estimator to the data structure (OLS/GLM, fixed effects for panels, SEM for latent constructs and full mediation models, multilevel models for nested data). State why.
- Mediation done right: test indirect effects with bootstrapped confidence intervals (not Baron–Kenny steps alone); but remember an indirect effect is evidence for a theorised mechanism, not a substitute for theorising it.
- Moderation: plot the interaction; report simple slopes and the region of significance; do not over-read a marginal interaction.
- Endogeneity & robustness: run the identification strategy planned in
jms-methods(FE, IV/2SLS, DiD, matching) and a robustness battery — alternative measures, alternative samples, controls in/out — each tied to a named threat, not a fishing expedition. - Measurement evidence: report reliability (alpha/CR), convergent/discriminant validity (AVE), and CFA fit; address CMB with a designed test, not only Harman.
Qualitative path
- Coding transparency: show the move from first-order codes → second-order themes → aggregate dimensions; a reader should be able to trace a quote to a construct.
- Abductive logic: make the iteration between data and theory explicit — surprising observations, the candidate explanations considered, why the retained one fits best. JMS rewards visible abduction, not a tidy after-the-fact story.
- Evidentiary support: a representative-quotes table tying each theme to data; report disconfirming/negative cases and how they refined the model.
- Trustworthiness: state the procedures used (audit trail, member checking, inter-coder reliability where appropriate, prolonged engagement) so credibility is demonstrable.
- From narrative to mechanism: for process work, show what drives the transitions across phases, not just the sequence.
Execution bridge (StatsPAI / Stata MCP)
Run the battery, don't just enumerate it. Full map:
execution-with-mcp. JMS mixes qualitative and quantitative management research; the chain below is for the quantitative-empirical lane.
- 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
- Path chosen (quantitative / qualitative) and matched to the design
- Quant: estimator fits the data; mediation via bootstrapped CIs; interactions plotted with simple slopes
- Quant: each robustness check tied to a named threat; CMB addressed by design; CFA/validity reported
- Qual: first-order → second-order → aggregate-dimension chain is auditable
- Qual: abductive reasoning visible; representative quotes table; negative cases reported
- Qual: trustworthiness procedures stated
- The claim never exceeds what the analysis supports
Anti-patterns
- Mechanism by mediation: claiming a process exists only because the indirect effect is significant
- Robustness theatre: a wall of checks that never names the threat each one rules out
- p-hacking / specification mining: the one significant model among many, presented as the model
- Quote-mining: cherry-picked quotes with no systematic coding behind them
- Tidy abduction: a too-clean narrative that hides the messy data-theory iteration reviewers want to see
- Idiom confusion: judging a qualitative paper by sample size and significance, or a quant paper by "richness"
Output format
【Path】quantitative / qualitative
【Quant】estimator + why; mediation (bootstrap CI); moderation (simple slopes); robustness→threats; CMB/CFA
【Qual】coding chain (1st→2nd→dimensions); abduction made visible; quotes table; negative cases; trustworthiness
【Claim support】does the analysis carry the theoretical claim? gaps …
【Next step】jms-tables-figures
Version History
- 1839142 Current 2026-07-05 13:47
