Applied Analysis Note 5
A Note on Good Model Fit
Model fit is based on how close the model-implied covariance matrix is to the observed covariance matrix
- Chi-Square should be low, P-value high
- CFI > 0.95 (max 1) [Bentler. Psychol Bull, 1990; 107:238-46]
- TLI > 0.95 (max 1) [Hu & Bentler, 1999; SEM 6:1-55]
- RMSEA < .05 (min 0) [Hu & Bentler. Psychol Meth, 1998; 4:424-53]
Good web resource: http://davidakenny.net/cm/fit.htm
Some details:
The confirmatory fit index (CFI) is calculated as:
where subscript 0 implies the null model, and subscript 1 implies the proposed model.
The Tucker Lewis index (TLI), also known as the non-normed fit index, is calculated as:
The Root mean squared error of approximation (RMSEA) is calculated as: