Quick Reading Tips
-
Focus on the Default model row (your CFA) when collecting indices. -
Saturated model = perfect fit (reference). Independence model = zero correlation (poor fit reference). -
Report a balanced set: one absolute, one/two incremental, one parsimony, and one information criterion.
Fit Index Thresholds Summary
Detailed Breakdown
NPAR: Free parameters estimated.
CMIN (χ²): Model chi-square. Lower is better. Tests exact fit.
CMIN/DF: "Normed chi-square". Because χ² is sensitive to N, use this.
RMR: Root Mean Square Residual. Smaller is better.
GFI: Goodness-of-Fit Index (0-1). Absolute fit.
AGFI: Adjusted GFI (penalizes complexity).
Compares default model to independence model (worst case).
CFI (Comparative Fit Index) & TLI (Tucker-Lewis) are most common.
Rewards simpler models. Useful for comparing candidates.
PNFI & PCFI: Parsimonious NFI/CFI.
NCP: Noncentrality parameter. Building block for RMSEA.
FMIN: Minimum discrepancy per sample.
Rarely reported alone. Used for diagnostics.
RMSEA: Population misfit per df. Lower is better.
PCLOSE: Test of close fit (H₀: RMSEA ≤ .05). We want PCLOSE > .05.
Trade-off between fit and complexity.
AIC & BIC: Lower is better when comparing rival models.
Expected Cross-Validation Index.
Estimates how well model fits a similar sample.
Lower implies better expected replication.
Critical N. Smallest sample size where χ² becomes significant.
Reporting Guide & Example
Match journal style with this structure.
| AMOS Block | Statistic to Take | Why it matters | Typical Guideline |
|---|
Copy-ready Example Sentence
“The three-factor CFA showed adequate fit: χ²(132) = 227.61, p < .001, CMIN/DF = 1.724; RMSEA = .052, 90% CI [.040, .063], PCLOSE = .385. Incremental indices were high (CFI = .969, TLI = .964, IFI = .969, NFI = .930), and absolute fit was acceptable (GFI = .916). Parsimony indices were PNFI = .802 and PCFI = .836. Information criteria further supported adequacy (AIC = 305.61; BIC = 446.10).”