Jmgmt Data Analysis

Data Analysis (jmgmt-data-analysis)

When to trigger

• The model is fit but a reviewer questions the measurement model or fit indices
• Mediation/moderation is tested in a way a JOM methods reviewer would challenge
• Nested data are being analyzed without modeling the nesting
• A meta-analysis needs the right estimator, corrections, and heterogeneity diagnostics
• Results are reported with significance asterisks and no effect sizes

The JOM analysis bar

JOM houses some of the field's leading research-methods scholars and runs methods reviews, so analysis is read by an unusually demanding audience. The expectation is a transparent measurement model before the structural model, effect sizes and confidence intervals alongside tests (not p-stars alone), and analysis choices that match the level and design set in jmgmt-methods. Report enough that the analysis is reconstructable from the paper and the (anonymized) data transparency table.

Branch paths

Branch A: SEM / CFA (latent-variable micro models)

• Report the measurement model first: standardized loadings, reliability (composite reliability/ω, not only α), AVE, and a discriminant-validity check (AVE vs. squared inter-construct correlations, or HTMT).
• Report multiple fit indices (χ²/df, CFI/TLI, RMSEA with CI, SRMR) — never a single index.
• For mediation, report the indirect effect with a bootstrap CI (e.g., 5,000 resamples, bias-corrected); do not infer mediation from a Sobel test alone or from two significant paths.
• Compare the hypothesized model against a theoretically motivated alternative.

Branch B: Multilevel / HLM (nested data)

• Report null-model ICC(1)/ICC(2) to justify multilevel modeling; group-mean-center level-1 predictors where the theory concerns within-group effects, and say which centering and why.
• Model random slopes when testing cross-level moderation; report the variance components.
• Test cross-level interactions with adequate level-2 N; do not over-interpret a cross-level slope from few groups.

Branch C: Regression / interactions (archival or single-level)

• Report standardized and unstandardized coefficients, robust/clustered SEs as appropriate, and effect sizes (ΔR², f²).
• Probe interactions: simple slopes at ±1 SD, an interaction plot, and a region-of-significance (Johnson–Neyman) where useful.
• For curvilinear claims, test the quadratic term and report the turning point with a CI.
• Carry the endogeneity strategy from jmgmt-methods through to the estimates (first-stage strength, exclusion logic).

Branch D: Meta-analysis

• Use the chosen model (Hunter–Schmidt psychometric vs. Hedges–Olkin random-effects); report k, total N, corrected mean effect, 95% CI, and the 80% credibility interval.
• Report heterogeneity (Q, I², τ²); run moderator/meta-regression tied to competing theories.
• Run publication-bias diagnostics (funnel/trim-and-fill, Egger's, PET-PEESE or selection models) and discuss robustness.

Robustness the JOM audience expects

Because JOM reviewers are methods-literate, anticipate the standard robustness asks rather than waiting for them: report the focal result under alternative specifications (with/without controls, alternative operationalizations), show it is not driven by influential cases, and — for archival work — report the endogeneity-corrected estimate alongside the naive one so the reader sees how much the correction moves the coefficient. Park the full robustness battery in the online supplement and summarize it in a sentence in the main text; the 50-page limit makes the supplement essential, not optional.

Worked vignette (illustrative)

A multilevel paper claims a cross-level moderation: team climate strengthens the individual-level link between role clarity and performance. The weak version reports a significant level-2 × level-1 product term and stops. The JOM-grade version reports the ICC(1) = .18 that justifies HLM, grand-mean-centers the level-2 moderator and group-mean-centers the level-1 predictor (stating why), fits a random-slope model, reports the cross-level interaction with its CI, and plots the simple slopes of role clarity at high vs. low team climate. The plot, not the p-value, is what convinces a reviewer the moderation is real and correctly modeled.