Statistics Glossary → Model fit indices (χ², CFI, TLI, RMSEA, SRMR)
Model fit indices (χ², CFI, TLI, RMSEA, SRMR)concept
Measures that summarize, from different angles, how well the covariance structure implied by a confirmatory factor or structural equation model matches the observed data.
The model chi-square test assesses the null hypothesis that the model-implied covariance matrix equals the observed one exactly; because it rejects even trivial departures in large samples, it is not sufficient on its own. The comparative indices (CFI and TLI) show on a 0-1 scale how much the model improves on an independence model in which all variables are uncorrelated; TLI penalizes model complexity more. RMSEA measures misfit per degree of freedom, √(max(χ² - df, 0) / (df·(N - 1))), and is reported with a confidence interval. SRMR is the average size of the differences between observed and model-implied correlations.
Commonly cited cut-offs are about 0.95 or above for CFI and TLI, 0.06 or below for RMSEA (up to 0.08 acceptable) and 0.08 or below for SRMR; but these are rough rules derived from simulation studies under particular conditions, and they shift with model size, sample size and the size of the loadings. The indices complement one another and should be read together, along with the residual correlations. Good fit does not prove a model is correct: equivalent models with different theoretical structures can fit equally well.
An everyday example
A tailor makes a dress to a customer's measurements. The chi-square test asks whether the dress fits the body to the millimetre, and for a very fussy customer the answer is almost always no. CFI says how much better the dress fits than a shapeless off-the-peg sack; RMSEA how loose or tight it is per seam; SRMR the average gap between the measurements and the dress. A good tailor looks at all of them together.
In MerQur
In Confirmatory Factor Analysis, 'RMSEA denominator (N, N - 1)' and 'Extra output: Residual correlation matrix'; in Exploratory Factor Analysis, 'Extra output: Goodness-of-fit test and indices' (chi-square, TLI, CFI, RMSEA, BIC, RMSR); in Path Analysis, 'RMSEA denominator' and 'Residual correlations'. Structural Equation Modeling gives the fit indices and, with MLR, reports the scaled chi-square and scaled CFI/TLI/RMSEA. Item Response Theory reports M2/C2, RMSEA, SRMSR and CFI/TLI, and PLS-SEM reports the SRMR.
Analyses where this concept appears
Exploratory Factor Analysis (EFA)Confirmatory Factor Analysis (CFA)Structural Equation Modeling (SEM)Item Response Theory (IRT / Rasch)Path AnalysisPLS-SEM (Partial Least Squares Path Modeling)
Related concepts
Likelihood ratio test (LRT)Modification indicesMeasurement model and structural model