SPSS · EFA

Exploratory factor analysis in SPSS: run it, read it and clean up the rotated matrix

Items jump to the wrong factor, one item loads on two columns, KMO is under 0.5? A step-by-step EFA guide with the item-removal process that gets you a clean rotated matrix.

⏱ ~6 min readSources: Hair et al. (2019) · Kaiser (1974)Updated:

Cronbach’s alpha passed, so you confidently run the EFA and get back a mess: items from factor A in factor B’s column, one item loading on two columns, a different number of factors from your model. This guide covers exploratory factor analysis in SPSS step by step: the conditions, the clicks, how to read the four tables that matter, and how to clean up the rotated matrix so the result stands up in front of your committee.

Rotated matrix with three factors, with item PV2 flagged for cross-loading at 0.468 and 0.552
Item PV2 loads on two factors, only 0.084 apart.

1What exploratory factor analysis is and when you need it

EFA groups observed items that correlate strongly into factors. That lets you check convergent validity (items of one factor load strongly on it) and discriminant validity (they do not load strongly on other factors).

Cronbach’s alpha→EFA→Correlation & regression→or CFA / SEM

2Conditions before you run EFA

  • Include only items that passed Cronbach’s alpha.
  • Sample size of at least 5 responses per item, and preferably 10 (Hair et al., 2019).
  • Use quantitative or Likert items only; leave demographics out.
✦ Together or separately?

For regression models it is common to run EFA separately for the independent variables and for the dependent variable. When preparing for CFA or SEM, many authors run all scales together.

3How to run exploratory factor analysis in SPSS

🖱️ Clicks
  1. Analyze→Dimension Reduction→Factor
  2. Move the items to the Variables box.
  3. Descriptives: tick KMO and Bartlett’s test of sphericity.
  4. Extraction: Method Principal components (regression) or Principal axis factoring (preparing for CFA), Based on Eigenvalue > 1.
  5. Rotation: Varimax (with Principal components) or Promax (with Principal axis factoring).
  6. Options: Sorted by size and Suppress small coefficients at 0.3.
✦ Why suppress at 0.3 rather than 0.5?

Hide everything under 0.5 and you cannot see the cross-loadings between 0.3 and 0.5, which are exactly what you are looking for. See them first, then decide.

4Reading the 4 EFA tables that matter

Four EFA conditions: KMO, Bartlett, eigenvalue and variance explained, factor loadings
The four conditions of an acceptable EFA result.

① KMO and Bartlett’s Test

  • KMO ≥ 0.8: Good · 0.5 to 0.8: Acceptable (Kaiser, 1974)
  • KMO < 0.5: The data is not suitable for factor analysis
  • Bartlett Sig. < 0.05: Items are correlated enough for EFA

② Communalities

How much of each item’s variance the factors explain. An item with a very low communality (under about 0.5) usually also loads weakly, so it is one to look at.

③ Total Variance Explained

  • Eigenvalue > 1: The rule for keeping a factor
  • Cumulative % ≥ 50%: The factors explain enough of the variance

④ Rotated Component Matrix

The most important table. Each item should load strongly on exactly one factor:

  • Loading ≥ 0.5: The common threshold
  • By sample size (Hair et al., 2019): about 0.30 at n = 350, 0.50 at n ≈ 120, 0.75 at n = 50
Example: a rotated matrix (Varimax)
Item123
SQ20.842
SQ10.815
SQ30.776
PV10.803
PV30.769
PV20.4680.552
TR10.821
TR20.794
TR30.731

KMO = 0.812; Bartlett Sig. = .000; 3 factors with eigenvalues above 1, explaining 68.4% of the variance. But PV2 loads on both factor 1 (0.468) and factor 2 (0.552), a gap of only 0.084. That is a cross-loading to deal with.

5A messy rotated matrix: 4 situations and fixes

Cross-loading item

Loads on two factors with a small gap (under 0.3 is a commonly used cut-off). Consider removing it, especially if the wording could be read two ways.

Low loading

The item reaches 0.5 on no factor. Usually a vague or off-topic question, and a candidate for removal.

Item jumps to another factor

An item from factor A sits squarely in factor B. Reread it: if that makes theoretical sense, you may rename the factor; if not, remove it.

Wrong number of factors

Two factors merge or one splits. Check eigenvalues, the scree plot and theory; you can fix the number of factors (Fixed number of factors) to test it.

🔁 The removal process
  1. Remove the single worst item (heaviest cross-loading or lowest loading).
  2. Re-run the EFA, because the matrix changes every time.
  3. Repeat until every item loads ≥ 0.5 on one factor and KMO and variance explained pass.
  4. If a factor lost items, re-run Cronbach’s alpha for it.
⚠️ Do not remove at all costs

Removing too many items can empty a concept of its meaning. Every removal needs a reason, statistical and content-based, and a record. This is a section committees love to question.

6How to report EFA in your thesis

✍️ A paragraph you can adapt

“EFA on the 9 observed items of the independent variables, using Principal Components with Varimax rotation, gave KMO = 0.812 (> 0.5) and a significant Bartlett’s test (Sig. < .001), so the data was suitable for factor analysis. Three factors were extracted at an eigenvalue of 1.236, explaining 68.4% of the variance (> 50%). Item PV2 cross-loaded on two factors and was removed; all remaining items then loaded above 0.5 on their intended factors.”

Practice on the rotated matrix of your own model

The Light Data’s model canvas produces the SPSS tables (Cronbach’s alpha, KMO, the rotated matrix, correlation, regression) on simulated data shaped by the model you draw. It is a convenient way to get familiar with each table before you run real data. Simulated data is never for reporting.

🎯 EFA checklist
  • KMO ≥ 0.5 (ideally ≥ 0.8) · Bartlett Sig. < 0.05.
  • Eigenvalue > 1 · variance explained ≥ 50%.
  • Each item loads ≥ 0.5 on exactly one factor, with no cross-loading.
  • Remove one item at a time, and re-run alpha if you removed any.

Frequently asked questions

What is a good KMO value?

A KMO of 0.5 or above is the minimum for factor analysis and 0.8 or above is good (Kaiser, 1974). Bartlett’s test should have Sig. below 0.05.

What factor loading is acceptable?

0.5 or above is the common rule. Hair et al. (2019) relate it to sample size: about 0.30 for 350 responses, 0.50 for around 120 and 0.75 for 50.

How do I handle cross-loading items?

An item that loads on two factors with little difference between them (a gap under 0.3 is a commonly used cut-off) is a candidate for removal. Remove one item at a time, worst first, and re-run the EFA.

Should I run EFA on independent and dependent variables together?

For regression models it is common to run EFA separately for the independent variables and the dependent variable. When preparing for CFA or SEM, many authors run all the scales together.

Principal Components with Varimax, or Principal Axis Factoring with Promax?

PCA with Varimax is common in theses using regression. When preparing for CFA or SEM, Principal Axis Factoring with Promax is often recommended, because factors are usually correlated in practice.