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) or
benjamini_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
exact suggest_function for each.
- Exhibits:
etable / did_summary_to_latex from 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
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
Source: brycewang-stanford/Awesome-Journal-Skills → Journal-of-Management-Studies-Skills/skills/jms-data-analysis/SKILL.md
1---2name: jms-data-analysis3description: 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).4---5
6
7# Data Analysis (jms-data-analysis)
8
9## When to trigger
10
11- Estimates are in but reviewers question endogeneity, robustness, or the indirect-effect claim
12- A qualitative analysis reaches findings but the path from data to constructs is not auditable
13- Effects hinge on a single specification with no robustness
14- A mediation/moderation result is reported without the analysis JMS expects
15- A reviewer says "the analysis does not support the claim" or "I can't see how you got here"
16
17## The JMS analysis bar — two idioms, one standard
18
19JMS 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.
20
21## Quantitative path
22
23- **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.
24- **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.
25- **Moderation**: plot the interaction; report simple slopes and the region of significance; do not over-read a marginal interaction.
26- **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.
27- **Measurement evidence**: report reliability (alpha/CR), convergent/discriminant validity (AVE), and CFA fit; address CMB with a designed test, not only Harman.
28
29## Qualitative path
30
31- **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.
32- **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.
33- **Evidentiary support**: a representative-quotes table tying each theme to data; report disconfirming/negative cases and how they refined the model.
34- **Trustworthiness**: state the procedures used (audit trail, member checking, inter-coder reliability where appropriate, prolonged engagement) so credibility is demonstrable.
35- **From narrative to mechanism**: for process work, show *what drives the transitions* across phases, not just the sequence.
36
37## Execution bridge (StatsPAI / Stata MCP)
38
39Run the battery, don't just enumerate it. Full map:
40[`execution-with-mcp`](../../../shared-resources/empirical-methods/execution-with-mcp.md). JMS mixes qualitative and quantitative management research; the chain below is for the quantitative-empirical lane.
41
42- **Many outcomes / specifications:** `romano_wolf` (step-down FWER) or
43 `benjamini_hochberg` — report the adjusted threshold.
44- **OVB sensitivity:** `oster_delta` / `sensemakr`.
45- **Inference:** `wild_cluster_bootstrap` (few clusters), `twoway_cluster` / `conley`;
46 multilevel data → cluster at the right level.
47- **Re-fit off one handle:** `audit_result(result_id)` lists the missing checks and the
48 exact `suggest_function` for each.
49- **Exhibits:** `etable` / `did_summary_to_latex` from the handle — no retyped numbers.
50
51Keep the decisive checks in the body and the exhaustive battery in the appendix. See the
52executed chain in the [JF execution walkthrough](../../../Journal-of-Finance-Skills/resources/worked-examples/02-execution-walkthrough.md).
53## Checklist
54
55- [ ] Path chosen (quantitative / qualitative) and matched to the design
56- [ ] Quant: estimator fits the data; mediation via bootstrapped CIs; interactions plotted with simple slopes
57- [ ] Quant: each robustness check tied to a named threat; CMB addressed by design; CFA/validity reported
58- [ ] Qual: first-order → second-order → aggregate-dimension chain is auditable
59- [ ] Qual: abductive reasoning visible; representative quotes table; negative cases reported
60- [ ] Qual: trustworthiness procedures stated
61- [ ] The claim never exceeds what the analysis supports
62
63## Anti-patterns
64
65- **Mechanism by mediation**: claiming a process exists only because the indirect effect is significant
66- **Robustness theatre**: a wall of checks that never names the threat each one rules out
67- **p-hacking / specification mining**: the one significant model among many, presented as the model
68- **Quote-mining**: cherry-picked quotes with no systematic coding behind them
69- **Tidy abduction**: a too-clean narrative that hides the messy data-theory iteration reviewers want to see
70- **Idiom confusion**: judging a qualitative paper by sample size and significance, or a quant paper by "richness"
71
72## Output format
73
74```text
75【Path】quantitative / qualitative
76【Quant】estimator + why; mediation (bootstrap CI); moderation (simple slopes); robustness→threats; CMB/CFA
77【Qual】coding chain (1st→2nd→dimensions); abduction made visible; quotes table; negative cases; trustworthiness
78【Claim support】does the analysis carry the theoretical claim? gaps …
79【Next step】jms-tables-figures
80```
81
82---
83
84**Source:** [`brycewang-stanford/Awesome-Journal-Skills`](https://github.com/brycewang-stanford/Awesome-Journal-Skills) → `Journal-of-Management-Studies-Skills/skills/jms-data-analysis/SKILL.md`
thedixitjain