{"id":1003,"date":"2026-10-13T15:16:00","date_gmt":"2026-10-13T15:16:00","guid":{"rendered":"https:\/\/freestatisticaltools.com\/blog\/?p=1003"},"modified":"2026-10-13T15:16:00","modified_gmt":"2026-10-13T15:16:00","slug":"exploratory-factor-analysis-guide","status":"publish","type":"post","link":"https:\/\/freestatisticaltools.com\/blog\/exploratory-factor-analysis-guide\/","title":{"rendered":"Exploratory Factor Analysis (EFA): A Step-by-Step Guide for Survey Research"},"content":{"rendered":"\n

Exploratory factor analysis (EFA)<\/strong> finds the underlying structure in a set of items. You give it 20 questionnaire items; it tells you whether they group into, say, four coherent factors, and which items belong where. Researchers use EFA to develop new scales, check the structure of an adapted questionnaire, and reduce many items to a few meaningful dimensions.<\/p>\n\n\n\n

This guide walks through every decision in an EFA, with the cut-offs reviewers expect.<\/p>\n\n\n

<\/svg><\/span>Free tool \u00b7 runs in your browser<\/span>Exploratory Factor Analysis (EFA) Online<\/a>Free exploratory factor analysis online: KMO, Bartlett\u2019s test, principal axis factoring, varimax rotation, scree plot and sorted factor loadings.<\/span><\/span>Run an EFA \u2192<\/a><\/div>\n\n\n\n

EFA, PCA or CFA?<\/h2>\n\n\n\n
Method<\/th>Question it answers<\/th>Use when<\/th><\/tr><\/thead>
EFA<\/strong><\/td>Which latent factors explain the correlations among the items?<\/td>Developing or exploring a scale<\/td><\/tr>
PCA<\/strong><\/td>How can I summarize many variables in fewer components?<\/td>Data reduction, without a latent-variable model<\/td><\/tr>
CFA<\/strong><\/td>Does my hypothesized factor structure fit the data?<\/td>The structure is known from theory or prior research<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

PCA is often run under the name "factor analysis", but it is a different model: components contain all the variance, including error, while factors model only the shared variance. For scale development, EFA (for example with principal axis factoring) is usually preferred. For data reduction, use PCA<\/a>. To confirm a structure, use CFA<\/a> or CB-SEM Studio<\/a>.<\/p>\n\n\n\n

Step 1: Check that your data are suitable<\/h2>\n\n\n\n

Sample size.<\/strong> More is better. Common guidance: at least 5\u201310 respondents per item and 200+ overall, although well-defined factors with high communalities can be recovered with smaller samples.<\/p>\n\n\n\n

Item level.<\/strong> Items should be measured on at least 5-point scales and treated as roughly continuous. Reverse-score negatively worded items first.<\/p>\n\n\n\n

Correlations.<\/strong> Items should correlate moderately (roughly .30 to .80). Items that correlate with nothing won't load anywhere; items correlating above .90 are near-duplicates.<\/p>\n\n\n\n

KMO (Kaiser-Meyer-Olkin) measure of sampling adequacy:<\/strong><\/p>\n\n\n\n

KMO<\/th>Interpretation (Kaiser, 1974)<\/th><\/tr><\/thead>
\u2265 .90<\/td>Marvellous<\/td><\/tr>
.80\u2013.89<\/td>Meritorious<\/td><\/tr>
.70\u2013.79<\/td>Middling<\/td><\/tr>
.60\u2013.69<\/td>Mediocre<\/td><\/tr>
< .60<\/td>Unacceptable; don't factor<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

Bartlett's test of sphericity<\/strong> tests whether the correlation matrix is an identity matrix (no correlations at all). It should be significant<\/strong> (p < .05). With typical survey samples it nearly always is, so KMO is more informative.<\/p>\n\n\n\n

Step 2: Choose an extraction method<\/h2>\n\n\n\n