Exploratory factor analysis is a statistical method used to identify the latent (hidden) structure underlying a set of observed variables. It answers the question: "How many unobserved constructs explain the correlations among these variables?" Use EFA when developing a new scale, when you have no strong prior theory about factor structure, or when you want the data to reveal its own structure — before moving to confirmatory factor analysis (CFA) to test that structure on new data.

No single method is universally best. Use multiple criteria together: (1) Kaiser criterion — retain factors with eigenvalue > 1 (simple but can over-extract in large samples); (2) Scree plot — retain factors above the "elbow" in the eigenvalue plot; (3) Parallel analysis — compare your eigenvalues to those from random data of the same dimensions (generally most accurate); (4) Interpretability — the factor structure should make theoretical sense. If criteria disagree, examine solutions with ±1 factor and choose the most interpretable.

The Kaiser-Meyer-Olkin (KMO) measure assesses sampling adequacy — whether the pattern of correlations among variables is compact enough for factor analysis to produce distinct, reliable factors. KMO benchmarks: ≥ 0.90 = Marvelous, 0.80–0.89 = Meritorious, 0.70–0.79 = Middling, 0.60–0.69 = Mediocre, 0.50–0.59 = Miserable, < 0.50 = Unacceptable. A KMO ≥ 0.70 is typically required before proceeding with EFA.

EFA and PCA look similar but have fundamentally different goals. EFA assumes observed variables are caused by underlying latent factors — it separates common variance (shared among variables) from unique variance and error. It is a causal model for construct measurement. PCA creates linear composites that maximally explain total variance (common + unique + error) — it is purely a data-reduction technique with no causal assumptions. Use EFA when you believe latent constructs drive the correlations; use PCA when you only want to reduce dimensions efficiently.

Communality (h²) is the proportion of a variable's total variance explained by the retained factors. Values range from 0 (no shared variance) to 1 (all variance is shared). A communality of 0.65 means 65% of that variable's variance is captured by the factor solution. Low communality (h² < 0.30) suggests a variable does not fit well with the other variables and may need to be removed or that more factors should be retained. Very high communalities (> 0.95) may indicate multicollinearity issues.

Varimax is an orthogonal rotation that maximises the variance of squared loadings within each factor. It pushes loadings toward 0 or ±1, making simple structure clearer: each variable tends to load highly on one factor and near zero on others. Use Varimax when you have theoretical reason to believe the underlying constructs are independent (uncorrelated). If factors are likely correlated (e.g., anxiety and depression), use an oblique rotation like Promax or Oblimin instead — these are available in R's psych package or SPSS.

Rules of thumb vary: a common minimum is n ≥ 5 per variable, with n ≥ 100 as an absolute minimum for stable results and n ≥ 200 preferred. MacCallum et al. (1999) showed that adequate sample size depends on communalities and factor overdetermination: with high communalities (h² > 0.60) and well-determined factors (4+ salient loadings per factor), n = 100–150 may suffice. With low communalities or poorly defined factors, n ≥ 300 is recommended. Always report sample size and communalities alongside your EFA results.

A complete APA 7 EFA report should include: (1) extraction method (e.g., principal axis factoring), (2) rotation method (e.g., Varimax), (3) factor retention criteria and number of factors, (4) KMO value, (5) Bartlett's test statistics: χ², df, and p-value, (6) eigenvalues and cumulative variance for each retained factor, (7) loading threshold used for interpretation, and (8) a factor loading matrix table. Example: "An EFA using principal axis factoring with Varimax rotation revealed a 2-factor solution (KMO = .82, Bartlett's χ²(10) = 95.3, p < .001) explaining 64.2% of variance (Factor 1: 42.1%, Factor 2: 22.1%)." Use the auto-filled templates below for complete reporting examples.

Common causes of poor simple structure: (1) Wrong number of factors — try ±1 factor; (2) Cross-loading items — items loading ≥ threshold on two or more factors are ambiguous; consider removing them; (3) Wrong rotation — if factors are theoretically correlated, switch from Varimax to Promax/Oblimin; (4) Small sample — solutions with n < 100 are often unstable; (5) Inadequate measure diversity — you may need more items per expected factor (aim for 4+ items per factor). Always examine the scree plot alongside the Kaiser criterion, and consider parallel analysis for a more objective retention decision.

Yes — this tool performs standard EFA computations (correlation matrix, eigendecomposition, Varimax rotation, KMO, Bartlett's test) consistent with methods used in R (psych package), SPSS, and SAS. For peer-reviewed publications, we recommend verifying results with dedicated statistical software and reporting the verification. For coursework, theses, and preliminary research, this tool provides complete, citable results. Cite as: StatsUnlock (2025). Exploratory Factor Analysis Calculator. Retrieved from https://statsunlock.com/calculators/exploratory-factor-analysis-calculator/