hrm-data-analysis

Data Analysis (hrm-data-analysis)

When to trigger

• You have nested data (employees in units/firms) and need the right multilevel model
• A mediation/moderation hypothesis needs a defensible test (not just a significant indirect effect)
• A measurement model (CFA) must establish discriminant validity before structural tests
• A reviewer challenges the aggregation, the estimator, or asks for robustness
• Qualitative data need a transparent, auditable coding and trustworthiness account

Match the estimator to the data structure

Data / claim	Estimator	What referees will check
Individuals nested in units; cross-level effects	HLM / mixed models (random intercepts/slopes)	Variance decomposition; ICC justifying multilevel; correct level for each predictor
Latent constructs + structural paths	SEM (with measurement model first)	CFA fit (CFI/TLI ≥ ~.95, RMSEA ≤ ~.06, SRMR ≤ ~.08); discriminant validity (AVE > shared variance)
Mediation (the HR black box)	Bootstrap indirect effect CIs; multilevel mediation if cross-level	Theorized mechanism, not inference from significance alone; 1-1-1 vs. 2-1-1 structure stated
Moderation / interaction	Product terms; simple slopes; interaction plot	Centering; region of significance; power; theory for the slope change
HR system → firm performance (panel)	Fixed-effects / DiD / IV	Endogeneity strategy; clustered SEs; pre-trends if DiD
Meta-analysis	Random-effects (e.g., Hunter–Schmidt / HVZ)	Coding reliability; heterogeneity (I², Q); publication-bias checks; moderator analysis

Multilevel and SEM discipline (HRM's bread and butter)

• Justify going multilevel. Report ICC(1)/ICC(2); if essentially zero between-unit variance, a multilevel model is not warranted — say so.
• Group-mean center lower-level predictors when testing within-unit effects; grand-mean center for cross-level; state which and why (the choice changes the meaning of the coefficient).
• Measurement before structure. Run the CFA and establish discriminant validity before interpreting structural paths; a saturated SEM with a poor measurement model is not evidence.
• Mediation is a theory claim. Report the indirect effect with bias-corrected bootstrap CIs, but the mechanism must have been theorized a priori; do not back-fill the mechanism from a significant indirect path.
• Aggregation evidence travels with the analysis. r_wg, ICC(1), ICC(2) belong in the results, tied to the composition model from hrm-methods.

Robustness and transparency HRM expects

• Report alternative specifications (controls in/out, alternative operationalizations of the HR system) and show the focal effect is stable.
• Address endogeneity for adoption/performance claims (FE, DiD, IV) and report clustered standard errors at the assignment level.
• Provide effect sizes in practitioner-meaningful terms (e.g., a 1-SD increase in HPWS is associated with X% higher productivity) — HRM rewards results an HR leader can act on.
• For qualitative work, give a transparent audit trail: data structure (first-order codes → second-order themes → aggregate dimensions), coding reliability or consensus process, and trustworthiness (member checks, triangulation).
• Follow Wiley's data-availability policy: include a data-availability statement and prepare materials for sharing where permitted (检索于 2026-06；以官网为准).

Checklist

• Estimator matches the data structure (nesting modeled; latent constructs in SEM)
• ICC reported and the multilevel choice justified
• CFA fit + discriminant validity established before structural interpretation
• Centering choice stated and matched to the effect being tested
• Indirect effects via bootstrap CIs; mechanism theorized a priori
• Endogeneity addressed; SEs clustered at the right level
• Effect sizes translated into practitioner-meaningful magnitudes
• Qualitative: transparent data structure + trustworthiness account
• Data-availability statement prepared per Wiley policy

Anti-patterns

• OLS on nested data: ignoring clustering and inflating significance
• Structure before measurement: interpreting SEM paths with a failing CFA
• Mediation by significance: claiming a mechanism from a significant indirect effect never theorized
• Centering silence: not stating group- vs. grand-mean centering in multilevel models
• p-value-only results: no effect sizes, no practitioner translation
• Aggregation without evidence: a unit-level construct with no r_wg/ICC
• Opaque qualitative coding: themes with no audit trail or reliability account

Output format

【Journal】Human Resource Management (Wiley "HRM")
【Skill】hrm-data-analysis
【Data structure】nested / latent-SEM / panel / meta / qualitative
【Estimator】HLM / SEM / FE-DiD-IV / bootstrap mediation / RE meta
【Measurement】CFA fit + discriminant validity status
【Multilevel】ICC reported; centering choice
【Mediation/moderation】indirect-effect CIs; interaction plot; a-priori mechanism?
【Robustness】alt specs / endogeneity / clustered SEs
【Practitioner magnitude】effect translated to an actionable number
【Data policy】availability statement prepared? 检索于 2026-06；以官网为准
【Next skill】hrm-contribution-framing