jams-data-analysis

Data Analysis & Reporting (jams-data-analysis)

When to trigger

• Data are collected and it is time to estimate and report
• You are unsure whether the estimator matches the design or the data structure
• A reviewer says "the analysis does not support the inference" or "report effect sizes"
• Significance is reported but the managerial magnitude is missing

Choose the estimator that matches the design

| Design / claim | Estimator |

|---|---|

| Latent constructs + structural paths (survey) | Covariance-based SEM (Mplus / lavaan / AMOS); PLS-SEM when prediction or formative constructs dominate |

| Nested data (consumers in stores, firms in industries) | HLM / multilevel models; random intercepts/slopes; report ICC |

| Mediation (process) | Bootstrapped indirect effects (PROCESS / lavaan), bias-corrected CIs; report the indirect effect, not just Baron–Kenny steps |

| Moderation / moderated mediation | Interaction term + simple slopes; conditional indirect effects (index of moderated mediation) |

| Experiment (factorial) | ANOVA / regression; estimated marginal means; planned contrasts; effect sizes per cell |

| Panel / observational causal | FE / DiD (modern staggered estimators); cluster-robust SE |

| Endogenous marketing regressor | IV/2SLS or Gaussian-copula control function; report first stage / instrument strength |

| Discrete choice / demand | Logit/probit; random-coefficient (mixed) logit |

| Meta-analysis | Random-effects effect-size synthesis; moderator meta-regression; publication-bias diagnostics |

Match SE clustering to the sampling/assignment structure (participant, store, market, firm).

JAMS reporting conventions

• APA results style. Report exact statistics (coefficients, SEs or t-values, CIs, exact p where shown). Avoid asterisk-only tables where the journal asks for precision; let the magnitude, not the star count, carry the result.
• Effect sizes and uncertainty, always. Standardized coefficients, R²/f², η²/Cohen's d, or odds ratios as the model requires — significance without magnitude is not a JAMS result.
• SEM reporting: measurement model first (loadings, AVE, CR, discriminant validity), then the structural model (standardized paths, R² for endogenous constructs, overall fit: CFI, TLI, RMSEA, SRMR).
• PLS reporting: loadings/weights, CR, AVE, HTMT, R², Q² (predictive relevance), and f²; bootstrap the path significances.

Translate every result into a managerial magnitude

This is the JAMS-distinguishing step. For each headline result, write a ledger row before drafting the results paragraph:

| Result | Theory point it supports | Required statistic | Managerial magnitude |

|---|---|---|---|

| Main path / treatment effect | which hypothesis / mechanism is confirmed | std. coef. + CI / d | sales lift, share, CLV, margin, retention, brand-equity points |

| Mediation (process) | which mechanism carries the effect | indirect effect + bias-corrected CI | why the process matters for the decision |

| Moderation (contingency) | when the effect strengthens/reverses | interaction + simple slopes | the managerial guardrail / segmentation rule |

| Robustness / alternative model | which threat (CMV, endogeneity) is reduced | same discipline as the main result | whether the conclusion's direction/size holds |

If the managerial-magnitude column is empty, the result is not yet ready for a JAMS results section.

Execution bridge (StatsPAI / Stata MCP)

Run the battery, don't just enumerate it. Full map:

[execution-with-mcp](../../../shared-resources/empirical-methods/execution-with-mcp.md). JAMS is empirical marketing with much survey-based SEM; the chain below serves causal / quasi-experimental designs and many-outcome corrections.

• 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](../../../Journal-of-Finance-Skills/resources/worked-examples/02-execution-walkthrough.md).

Checklist

• [ ] Estimator matches design and data structure; SE clustering correct
• [ ] SEM: measurement model reported before structural; full fit indices given
• [ ] PLS: HTMT, R², Q², f² reported; paths bootstrapped
• [ ] Mediation via bootstrapped indirect effects with bias-corrected CIs
• [ ] Moderation: simple slopes + index of moderated mediation where relevant
• [ ] Effect sizes and uncertainty reported throughout (APA style)
• [ ] Every headline result has a managerial-magnitude translation
• [ ] Robustness addresses the design's specific threat (CMV / endogeneity / pre-trends)

Robustness that targets the design's real threat

Generic robustness ("we also ran model B") rarely persuades JAMS reviewers; the robustness must answer the specific threat to the genre's inference:

• Survey/SEM: rule out CMV with a marker-variable / CFA-marker model and report whether paths survive; test an alternative measurement specification; show results hold on a holdout or second sample.
• Secondary data: placebo tests, alternative instruments, pre-trend/parallel-trends evidence, sensitivity to the identifying assumption, and alternative fixed-effect structures.
• Experiment: replication across stimuli/samples, a confound-ruling-out study, and a test of the alternative-mechanism account.
• Meta-analysis: sensitivity to coding decisions, trim-and-fill / PET-PEESE for publication bias, and influence diagnostics for outlier studies.

State, for each robustness check, which threat it neutralizes — a list of checks with no mapped threat reads as box-ticking.

Anti-patterns

• Baron–Kenny causal-steps mediation instead of bootstrapped indirect effects
• Reporting fit indices but no standardized paths or R²
• Significance with no effect size and no managerial magnitude
• Ignoring nesting (consumers within stores) and clustering
• A weak/untested instrument, or endogeneity waved away
• Asterisk tables that hide the size of the effect
• Robustness checks listed with no statement of which threat each addresses

Output format

【Design】survey-SEM / PLS / HLM / experiment / panel-causal / choice / meta
【Estimator】matches design? SE clustering: [...]
【Measurement (if SEM/PLS)】AVE/CR/discriminant + fit/HTMT: pass/fix
【Effect sizes + uncertainty】reported (APA)? pass/fix
【Mediation/moderation】bootstrapped indirect / simple slopes: done?
【Managerial-magnitude ledger】every headline result translated? yes/fix
【Robustness】design-specific threat addressed: [...]
【Next skill】jams-contribution-framing