SEM CFA reporting requirements aren't a style preference — they're a credibility test. Specialist reviewers form a verdict on rigour from the first paragraph of the analysis section.
Structural equation modelling and confirmatory factor analysis attract sharper peer review than almost any other quantitative technique. The reason is simple: the field has well-established reporting standards, and reviewers expect you to know them. Missing fit indices, unreported model modifications, or factor loadings buried in appendices are treated as carelessness — or worse.
This guide walks through what specialist reviewers actually want to see in CFA and SEM sections, the fit-index thresholds that hold up across journals, and the reporting moves that turn a vulnerable analysis into a defensible one. For broader methodology context, see our pillar on how to write a methodology section that reviewers respect.
What CFA and SEM each require
CFA and SEM share reporting infrastructure, but each demands its own minimum set. Treat the lists below as the floor. Specialist journals will expect more; general journals may accept slightly less, but never excuse missing fit indices.
- Model specification with item-to-factor assignments
- Estimator (ML, MLR, WLSMV) with rationale
- Standardised factor loadings (≥ 0.50 ideal, ≥ 0.70 preferred)
- Composite reliability (CR ≥ 0.70)
- Average variance extracted (AVE ≥ 0.50)
- Discriminant validity: AVE √ vs inter-factor correlations
- Full fit indices: χ²/df, CFI, TLI, RMSEA, SRMR
- Measurement model results (CFA) reported first
- Structural model with all path coefficients
- Standardised & unstandardised path estimates with CIs
- Indirect effects via bootstrapping (≥ 5,000 samples)
- R² for endogenous variables
- Common method bias diagnostics
- Fit indices for structural model, separately reported
Fit indices — the cutoffs that hold up
Fit-index cutoffs are debated in the literature, but most SSCI Q1–Q2 journals converge on the thresholds below. Report multiple indices (never just one), report exact values, and cite the source for your chosen cutoffs — typically Hu and Bentler (1999), Kline (2016), or Hair et al. (2019).
Acceptable fit (rather than excellent fit) is sometimes defended at CFI ≥ 0.90, RMSEA < 0.08, SRMR < 0.08 — with explicit citation to Hair et al. or Browne and Cudeck. Don't relax the cutoffs silently. State the threshold, cite the source, and report the achieved value.
If you used modification indices to improve model fit — correlating error terms, dropping items, freeing parameters — you must report every modification, justify each on theoretical grounds, and acknowledge that data-driven modifications reduce confirmatory status. Silent modifications are the single fastest way to lose specialist reviewer trust. Always retain the original model in supplementary materials.
Validity evidence reviewers expect
For CFA in particular, reporting validity evidence is non-negotiable. The expected set has three layers, each with its own threshold.
Convergent validity is supported by standardised factor loadings ≥ 0.50 (ideally ≥ 0.70), AVE ≥ 0.50 for each construct, and composite reliability ≥ 0.70. Report all three together, typically in a single measurement-model table.
Discriminant validity is most commonly demonstrated via the Fornell–Larcker criterion — the square root of each construct's AVE should exceed its correlation with any other construct. The HTMT ratio (Heterotrait–Monotrait) is increasingly preferred for stricter discriminant assessment, with cutoffs of 0.85 (strict) or 0.90 (lenient). Many Q1 journals now expect HTMT alongside the Fornell–Larcker.
Reliability sits alongside validity, but should never replace it. Cronbach's α tells reviewers about internal consistency, not construct validity. Report α, CR, and AVE together; never collapse them into a single claim.
If you're using PLS-SEM instead of CB-SEM
Partial least squares SEM has different reporting conventions. Fit indices in the traditional CB-SEM sense don't apply; instead, report SRMR, NFI, and the geodesic discrepancy as model-fit proxies, along with predictive measures (Q² values, PLSpredict). For measurement models, report outer loadings (≥ 0.70 preferred), reliability (rho_A and composite reliability), AVE, and HTMT for discriminant validity. Specialist reviewers in management and information-systems journals expect explicit acknowledgement that PLS-SEM is being used for prediction or composite modelling, not for theory testing in the covariance-based sense.
"Specialist reviewers don't reject SEM papers for imperfect fit. They reject them for opaque reporting of imperfect fit."
— Common refrain in editor-reviewer correspondenceThe four omissions specialist reviewers catch first
1. One fit index reported, others omitted. Reporting only CFI or only χ² makes reviewers assume the others were unfavourable. Always report at least one absolute (χ²/df, RMSEA, SRMR) and at least one incremental (CFI, TLI) index.
2. Modification indices used but not declared. Correlating error terms or freeing cross-loadings without disclosure is treated as analytical impropriety. Report every modification and justify each.
3. Bootstrap procedure missing for indirect effects. Reporting a Sobel test alone for mediation is increasingly considered insufficient. Specify the resampling procedure (5,000–10,000 bootstraps), the confidence interval type (bias-corrected accelerated), and the resulting CIs for all indirect effects.
4. Common method bias overlooked. For single-source self-report data, at minimum run Harman's single-factor test; ideally include a marker variable or unmeasured latent method factor approach. Reviewers ask about CMB on every cross-sectional self-report SEM submission they see.
Closing — write for the reviewer who'll know
The strongest SEM and CFA papers read as if the authors expected a specialist reviewer. Every cutoff is sourced. Every modification is disclosed. Every claim about validity is paired with its number. The section feels engineered rather than narrated — and that's exactly what specialist reviewers respond to.
Before submission, hand the analysis section to someone who has published SEM in a similar journal and ask: can you reproduce this model from this text? If the answer is yes, the reporting is done. If the answer is "almost, but I'd need a few more details," those are the same details reviewers will request — at the cost of an extra round.
Get your SEM section reviewed by someone who's published it.
Our editorial team includes PhD editors who have published SEM and CFA work in SSCI Q1/Q2 journals. They know what specialist reviewers ask for — and how to defend the modifications you've already made.