When to use it
- •You have an established scale, or an EFA result from a separate sample, and want to confirm it.
- •You need to establish measurement validity before running a structural model.
- •A reviewer has asked for evidence of convergent and discriminant validity.
Assumptions to check first
- •The factor structure is specified before looking at the fit — respecifying until the model fits is exploratory work wearing confirmatory clothes.
- •Continuous indicators and multivariate normality for maximum likelihood estimation (robust estimators exist when this fails).
- •Sufficient sample size — commonly cited minimums start around 200, with more needed for complex models.
What to report
- •χ² with degrees of freedom and p, even though it is nearly always significant in realistic sample sizes.
- •Approximate fit indices: CFI, TLI, RMSEA (with its confidence interval) and SRMR.
- •Standardised factor loadings for every indicator.
- •Composite reliability and average variance extracted (AVE) if you are claiming convergent validity.
APA 7 example
The three-factor measurement model showed acceptable fit, χ²(51) = 98.7, p < .001, CFI = .96, TLI = .95, RMSEA = .052 [90% CI .038, .066], SRMR = .043. All standardised loadings exceeded .60 and were significant at p < .001.
Numbers are illustrative — the Advisor generates this sentence with your own results.
The mistake reviewers catch
The familiar cut-offs (CFI ≥ .95, RMSEA ≤ .06, SRMR ≤ .08) come from Hu and Bentler (1999) simulations under specific conditions, and were never intended as universal pass marks. Report them, but do not treat a CFI of .94 as a failed study.
Run it on your own data
Paste or upload your dataset — nothing leaves your browser. The Analysis Advisor checks the assumptions above, runs cfa, and drafts the APA Methods and Results text with your numbers in it.
Open the Analysis Advisor →Related analyses
Frequently asked questions
When should I use cfa?
Does my hypothesised measurement model actually fit the data? You have an established scale, or an EFA result from a separate sample, and want to confirm it.
What do I need to report for cfa in APA 7?
χ² with degrees of freedom and p, even though it is nearly always significant in realistic sample sizes. Approximate fit indices: CFI, TLI, RMSEA (with its confidence interval) and SRMR. Standardised factor loadings for every indicator. Composite reliability and average variance extracted (AVE) if you are claiming convergent validity.
What is the most common mistake with cfa?
The familiar cut-offs (CFI ≥ .95, RMSEA ≤ .06, SRMR ≤ .08) come from Hu and Bentler (1999) simulations under specific conditions, and were never intended as universal pass marks. Report them, but do not treat a CFI of .94 as a failed study.