Covariance-based SEM in AMOS stands or falls on model fit — the evidence that your theorised model reproduces the observed data. Reviewers scan the fit table first. The good news: only a handful of indices matter, and each answers a specific question.
The indices that matter
Chi-square and CMIN/df
The chi-square tests exact fit but is oversensitive with large samples — nearly everything 'fails'. Convention therefore uses the normed chi-square (CMIN/df): values below 3 are generally acceptable, below 2 good.
CFI and TLI (incremental fit)
These compare your model to a baseline model of no relationships. CFI and TLI ≥ 0.90 indicate acceptable fit; ≥ 0.95 is the stricter standard many journals prefer.
RMSEA (absolute fit, with parsimony penalty)
Root Mean Square Error of Approximation punishes unnecessary complexity. ≤ 0.08 is acceptable, ≤ 0.06 good; report the 90% confidence interval, ideally with its upper bound under 0.08.
SRMR (residual-based fit)
The standardised average difference between observed and implied correlations. ≤ 0.08 is the usual bar.
Best practice is one index per family — e.g. CMIN/df, CFI (or TLI), RMSEA with CI, and SRMR. Cherry-picking the single index that flatters your model is a reviewer red flag.
When fit is poor: the legitimate moves
- 1Check the measurement model first — run CFA before the structural model; low loadings and cross-loading items are the usual culprits.
- 2Inspect standardised residuals to find where misfit concentrates.
- 3Use modification indices with theory, not greed — freeing error covariances within the same construct can be defensible; adding paths purely to chase fit is not.
- 4Reconsider the model — sometimes the honest conclusion is that the theory needs revision, which is a finding, not a failure.
What examiners actually probe
Not whether every index clears the strictest bar, but whether you (a) reported a fair set of indices, (b) justified any respecification with theory, and (c) validated the measurement model before testing structure. The two-step logic — CFA, then structural model — is the defence that works. Our AMOS mentoring rehearses exactly this.
Choosing between CB-SEM and PLS-SEM in the first place? Read What is SmartPLS and SPSS vs AMOS vs SmartPLS.
Frequently asked
What if one fit index passes and another fails?+
Common, especially CFI passing while RMSEA protests (or vice versa) — they penalise different things. Report all honestly, diagnose the source of misfit, and justify your judgement; selective reporting is worse than imperfect fit.
Is chi-square significance a problem in large samples?+
A significant chi-square with n in the hundreds is expected and not disqualifying — the test is oversensitive at scale. That's why CMIN/df and the approximate-fit indices carry the interpretive weight.
phdguide's mentors are senior academics, former supervisors, statisticians and publication specialists with 25+ years of combined experience guiding MBA, MPhil and PhD scholars from topic to viva.
Ethical, compliant guidance: We provide academic support, mentoring, analysis, editing and structuring — not authorship. Your work stays compliant with university policies.