Factor Analysis

Exploratory Factor Analysis (EFA) in SPSS

Discover the latent constructs behind your survey items. We run EFA in SPSS with the right extraction method, rotation and factor retention criteria — then deliver APA-formatted loading tables, scree plots and a full interpretation ready for your Chapter 4.

PAF / ML

Extraction Methods

APA 7

Formatted Output

48hr

Typical Turnaround

Free

Initial Consultation

What Is Exploratory Factor Analysis?

Exploratory factor analysis (EFA) is a multivariate statistical technique that identifies the hidden structure underlying a set of observed variables. When a researcher administers a 30-item survey, respondents do not answer each item independently — their responses are driven by a smaller number of unobserved latent factors (constructs such as motivation, anxiety, or job satisfaction). EFA discovers these factors by analysing the pattern of correlations among items.

Unlike confirmatory factor analysis (CFA), which tests a pre-specified structure, EFA is data-driven: the researcher does not declare in advance how many factors exist or which items belong to which factor. The data determine the solution. This makes EFA the right choice when developing a new instrument, adapting an existing one to a new population, or exploring the dimensionality of a dataset for the first time.

EFA vs. PCA vs. CFA — When to Use Each

Technique Purpose When to Use SPSS Location
EFA Discover latent factors from observed variables New scale development, unknown factor structure, instrument adaptation Analyze → Dimension Reduction → Factor
PCA Reduce variables to fewer components Data reduction without theoretical interest in latent constructs Analyze → Dimension Reduction → Factor (extraction: Principal Components)
CFA Test a pre-specified measurement model Validating an existing scale, structural equation modelling SPSS Amos (not in base SPSS)

Key Concepts in EFA

Latent Factors

Unobserved constructs that cause the observed correlations among items. Each factor represents a dimension of the construct you are measuring (e.g. "emotional exhaustion" as a dimension of burnout).

Factor Loadings

Correlations between each observed variable and a factor. A loading of .70 means the factor explains 49% of the item's variance. Loadings above .40 are typically considered meaningful.

Communalities

The proportion of each variable's variance explained by the extracted factors. Low communalities (< .30) indicate the variable does not share much variance with others and may need removal.

Eigenvalues

Each factor's eigenvalue represents the total variance it explains. An eigenvalue of 4.5 means that factor accounts for as much variance as 4.5 individual items. Used in determining the number of factors to retain.

Assumptions and Prerequisites for EFA

Before running EFA, verify that your data meet these requirements. Violations lead to unstable or uninterpretable factor solutions.

1. Adequate Sample Size

The absolute minimum is 100 cases, but 200–400 is preferred. The subject-to-variable ratio should be at least 5:1 (Hair et al.) or ideally 10:1 (Nunnally). The best single indicator is the KMO statistic: if KMO is above .80, your sample is adequate regardless of the exact ratio.

KMO Value Interpretation (Kaiser, 1974)
.90+Marvellous
.80–.89Meritorious
.70–.79Middling
.60–.69Mediocre
.50–.59Miserable
Below .50Unacceptable — do not proceed

2. Sufficient Correlations Among Variables

Factor analysis works by modelling shared variance. If your variables do not correlate, there is nothing for factors to explain. Check the correlation matrix: most items should have at least one correlation above .30 with another item. Bartlett's test of sphericity must be significant (p < .05), confirming the correlation matrix is not an identity matrix.

3. Variable Measurement Level

Standard EFA in SPSS operates on Pearson correlation matrices, so variables should be continuous or approximate continuity (e.g. 5-point or 7-point Likert scales). For dichotomous or ordinal items with fewer than 5 categories, consider polychoric correlations. All variables in SPSS Variable View should be set to Scale measurement level.

4. No Extreme Multicollinearity or Singularity

The determinant of the correlation matrix should be above .00001. If it is zero or near zero, you have multicollinearity (two items are almost identical) or singularity. SPSS reports the determinant in the footnote of the correlation matrix. Remove one item from each highly correlated pair (r > .90) before proceeding.

5. Adequate Communalities

After extraction, examine the communalities table. Each variable's communality should be at least .30 (ideally above .40). Variables with very low communalities share little variance with the factors and should be considered for removal. With small samples (< 200), aim for communalities above .60 for stable results (MacCallum, Widaman, Zhang & Hong, 1999).

How to Run Exploratory Factor Analysis in SPSS — Step by Step

This workflow covers SPSS versions 25–29 and SPSS Subscription. Follow each step in order.

  1. 1

    Prepare Your Data

    Open your .sav file. In Variable View, confirm all items are set to Scale measurement level. Check for missing data: EFA uses listwise deletion by default, so cases with any missing value on the included variables are dropped. If missingness exceeds 5%, consider multiple imputation before running EFA.

    Run Analyze → Descriptive Statistics → Descriptives to check for floor/ceiling effects, zero-variance items and distributional issues. Remove any item with zero or near-zero variance.

  2. 2

    Test Factorability — KMO and Bartlett's Test

    Analyze → Dimension Reduction → Factor

    Move all scale items into the Variables box. Click Descriptives: check KMO and Bartlett's test of sphericity and Coefficients (to view the correlation matrix). Click Continue.

    Decision rule: KMO must be above .60 (above .80 preferred). Bartlett's test must be significant (p < .05). If either condition fails, your data are not suitable for factor analysis — consider removing items with low anti-image correlations.

  3. 3

    Choose the Extraction Method

    In the Factor Analysis dialog, click Extraction. Choose:

    • Principal Axis Factoring (PAF) — the most common choice for EFA. Analyses shared variance only, excluding unique/error variance. Recommended when your goal is to identify latent constructs.
    • Maximum Likelihood (ML) — provides a chi-square goodness-of-fit test for the solution. Requires multivariate normality. Use when you want to test whether the extracted number of factors adequately reproduces the correlation matrix.
    • Principal Components — technically PCA, not EFA. Analyses total variance (shared + unique). Select only when your purpose is data reduction rather than construct discovery.

    Under Analyze, select Correlation matrix. Under Extract, initially select Based on Eigenvalue > 1 for a first pass. Check Scree plot. Click Continue.

  4. 4

    Determine the Number of Factors

    After the initial extraction, use multiple criteria to decide the number of factors:

    Criterion Rule Limitation
    Kaiser's criterion Retain factors with eigenvalue > 1.0 Tends to over-extract; unreliable with many variables
    Scree plot Retain factors above the "elbow" Subjective; multiple researchers may read it differently
    Parallel analysis Retain factors whose eigenvalue exceeds random data eigenvalue Gold standard but requires syntax or macro in SPSS
    Cumulative variance Aim for 50–60% total variance explained Arbitrary threshold; discipline-dependent
    Interpretability Each factor should have 3+ items and make conceptual sense Requires domain knowledge

    Once decided, return to Extraction, change Extract to Fixed number of factors, enter the number, and re-run.

  5. 5

    Select the Rotation Method

    Click Rotation in the Factor Analysis dialog. Choose:

    Rotation Type Assumes When to Use
    Varimax Orthogonal Factors are uncorrelated Most commonly reported; use when factors are independent
    Direct Oblimin Oblique Factors may correlate Recommended default for social/behavioural sciences
    Promax Oblique Factors may correlate Faster for large datasets; more aggressive oblique rotation
    Quartimax Orthogonal Factors are uncorrelated Minimises complexity of variables; tends to produce a general factor

    Check Rotated solution and Loading plot(s). Click Continue.

  6. 6

    Set Display Options

    Click Options. Check Sorted by size (groups items by their highest loading, making interpretation easier). Set Suppress small coefficients and enter .30 as the absolute value threshold. This hides trivial loadings from the output. Click Continue, then OK to run the analysis.

  7. 7

    Examine and Refine the Solution

    Review the output systematically:

    • Pattern matrix (oblique) or Rotated Component Matrix (orthogonal) — are items loading cleanly on one factor?
    • Communalities — any below .30? Consider removing those items.
    • Cross-loadings — any item loading above .32 on two factors? Remove and re-run.
    • Factor correlation matrix (oblique only) — correlations above .32 justify using oblique rotation; below .32 means orthogonal and oblique yield equivalent results.
    • Total Variance Explained — aim for 50–60% cumulative variance.

    Remove problematic items one at a time, re-running EFA after each removal. Continue until you have a clean, interpretable structure where every item loads above .40 on exactly one factor.

SPSS Menu Paths — Quick Reference

Procedure Menu Path Syntax Command
Factor Analysis (EFA) Analyze → Dimension Reduction → Factor FACTOR
KMO & Bartlett's Test Factor → Descriptives → KMO and Bartlett's test FACTOR /PRINT KMO
Correlation Matrix Factor → Descriptives → Coefficients FACTOR /PRINT CORRELATION
Scree Plot Factor → Extraction → Scree plot FACTOR /PLOT EIGEN
Reliability (after EFA) Analyze → Scale → Reliability Analysis RELIABILITY

Extraction Methods Explained

SPSS offers seven extraction methods. The two most appropriate for EFA are principal axis factoring and maximum likelihood.

Principal Axis Factoring (PAF)

PAF estimates communalities iteratively, analysing only shared variance. It does not assume multivariate normality, making it robust to non-normal data. PAF is the recommended default for EFA in most social science, education and health research.

When to use: Developing a new questionnaire, adapting a scale to a new population, exploring the factor structure of Likert-type items, or whenever you want to identify latent constructs without strict distributional assumptions.

Maximum Likelihood (ML)

ML estimates factor loadings by maximising the likelihood that the observed correlation matrix was produced by the extracted factors. It provides a chi-square goodness-of-fit statistic, letting you test whether the specified number of factors adequately reproduces the data.

When to use: When your data are approximately multivariate normal and you want a formal test of model fit. ML is preferred in psychology journals and when you plan to compare the EFA solution to a subsequent CFA model.

Caution: ML requires larger samples (n > 200) and can produce Heywood cases (communalities > 1.0) with under-determined solutions. If you encounter convergence failures, switch to PAF.

Rotation Methods — Orthogonal vs. Oblique

Rotation does not change the total variance explained — it redistributes it across factors to make the solution easier to interpret.

Which rotation should I choose?

Start with Direct Oblimin (oblique). Examine the factor correlation matrix. If all factor correlations are below .32, the factors are essentially independent and orthogonal rotation (Varimax) will produce the same result — you can report either. If any correlation exceeds .32, oblique rotation is necessary because forcing orthogonality distorts the loadings.

Pattern Matrix vs. Structure Matrix

Oblique rotation produces two matrices. This distinction does not exist with orthogonal rotation.

Pattern Matrix

Contains the unique contribution of each factor to each variable, controlling for the other factors. This is what you report and interpret. Loadings represent regression weights.

Structure Matrix

Contains the total correlation between each variable and each factor, including shared variance through correlated factors. Used for supplementary information but not for primary interpretation.

How to Report EFA Results in APA Format

Follow this template for your dissertation results chapter or journal submission.

APA Factor Loading Table Template

Item Factor 1 Factor 2 Factor 3 h2
Item 7: I feel valued at work.82.71
Item 3: My supervisor respects me.78.65
Item 12: I trust leadership decisions.74.58
Item 1: My role uses my skills.65.49
Item 9: I feel burned out.85.74
Item 5: I dread going to work.79.68
Item 14: I am emotionally exhausted.72.55
Item 8: I can balance work and life.81.69
Item 2: I have flexible hours.76.61
Item 11: Workload is manageable.68.52
Eigenvalue 3.84 2.61 1.97
% Variance 25.6 17.4 13.1

Note. Factor loadings from pattern matrix using principal axis factoring with Direct Oblimin rotation. Loadings below .30 are suppressed. h2 = communality. Total variance explained = 56.1%.

APA Narrative Example

"An exploratory factor analysis was conducted on the 15-item Workplace Wellbeing Scale using principal axis factoring with Direct Oblimin rotation. The Kaiser-Meyer-Olkin measure verified sampling adequacy, KMO = .87, and Bartlett's test of sphericity was significant, χ2(105) = 1,842.63, p < .001, indicating the correlations were sufficiently large for factor analysis. An initial analysis was run to obtain eigenvalues for each factor. Three factors had eigenvalues above Kaiser's criterion of 1 and together explained 56.1% of the variance. The scree plot showed a clear inflection at the fourth factor, supporting a three-factor solution. Parallel analysis (1,000 random datasets) confirmed that only three factors had eigenvalues exceeding the 95th percentile of random data. The three-factor solution was retained. After rotation, Factor 1 (Organisational Trust) explained 25.6% of variance, Factor 2 (Emotional Exhaustion) 17.4% and Factor 3 (Work-Life Balance) 13.1%. All items loaded above .40 on their primary factor with no cross-loadings above .30. Factor correlations ranged from .24 to .41 (Table 2), supporting the use of oblique rotation. Communalities ranged from .49 to .74."

Required Reporting Checklist

  • KMO value and Bartlett's test result (χ2, df, p)
  • Extraction method (PAF or ML) with justification
  • Rotation method (Varimax, Direct Oblimin, Promax) with justification
  • Criteria used to determine the number of factors (eigenvalues, scree, parallel analysis)
  • Number of factors retained and total variance explained
  • Pattern matrix loading table with communalities (suppress loadings < .30)
  • Factor correlation matrix (oblique rotation only)
  • Items removed and reason (cross-loading, low communality, low loading)
  • Eigenvalue and variance for each factor

Worked Example: Workplace Wellbeing Scale

A researcher develops a 15-item scale measuring workplace wellbeing. The sample is 312 full-time employees. Here is the full EFA workflow from start to finish.

Step 1: Data Screening

Descriptive statistics show all 15 items have means between 2.3 and 4.1 on a 5-point scale — no floor or ceiling effects. Missing data: 2.1% (MCAR by Little's test). Listwise deletion retains 305 complete cases.

Step 2: Factorability

KMO = .87 (meritorious). Bartlett's test: χ2(105) = 1,842.63, p < .001. The correlation matrix has adequate correlations (most r > .30) and determinant = .002 (above .00001 — no multicollinearity). Data are suitable for factor analysis.

Step 3: Initial Extraction

PAF extraction with eigenvalue > 1 produces three factors (eigenvalues: 4.82, 2.61, 1.97). Scree plot shows clear elbow at factor 4. Parallel analysis confirms three factors exceed random data. Cumulative variance = 56.1%. All three criteria converge on a three-factor solution.

Step 4: Rotation and Refinement

Direct Oblimin rotation applied. Item 6 cross-loads on Factors 1 and 2 (.38 and .35). Item 6 removed, EFA re-run. Item 10 has communality of .22. Item 10 removed, EFA re-run. Remaining 13 items load cleanly: no cross-loadings, all communalities above .40. Item 4 loads at .34 (below .40 threshold). Item 4 removed. Final solution: 12 items, 3 factors, 61.3% variance.

Step 5: Interpretation and Naming

Factor 1 — Organisational Trust (items 1, 3, 7, 12): items about feeling valued, supervisor respect and trust in leadership. Factor 2 — Emotional Exhaustion (items 5, 9, 13, 14): items about burnout, dread and fatigue. Factor 3 — Work-Life Balance (items 2, 8, 11, 15): items about flexibility, workload and personal time. Factor correlations: F1–F2 = −.41, F1–F3 = .38, F2–F3 = −.24.

Step 6: Reliability Check

Cronbach's alpha for each subscale: Organisational Trust α = .87, Emotional Exhaustion α = .84, Work-Life Balance α = .81. All exceed the .70 threshold. Item-total correlations range from .52 to .74 — no items need further removal.

12 Common Mistakes in Exploratory Factor Analysis

These are the errors dissertation committees and journal reviewers flag most often.

1. Using PCA when EFA is appropriate

PCA is data reduction; EFA identifies latent constructs. If your goal is scale development, use PAF or ML, not Principal Components.

2. Relying solely on eigenvalue > 1

Kaiser's criterion over-extracts. Always supplement with scree plot, parallel analysis and interpretability.

3. Using Varimax without checking factor correlations

If factors correlate above .32, orthogonal rotation distorts the structure. Run oblique first; use Varimax only if correlations are negligible.

4. Interpreting the structure matrix instead of pattern matrix

With oblique rotation, the pattern matrix shows unique contributions. The structure matrix inflates loadings through shared factor variance.

5. Keeping cross-loading items

Items loading .32+ on multiple factors indicate ambiguity. Remove them and re-run — a clean structure is more important than preserving items.

6. Removing multiple items at once

The factor solution changes with each removal. Remove one item, re-run, assess, repeat. Batch removal can produce an unstable solution.

7. Insufficient sample size

Running EFA with fewer than 100 cases or KMO below .60 produces unreliable results. Report KMO as your sampling adequacy measure.

8. Not reporting KMO and Bartlett's test

These are the minimum factorability checks. Omitting them is a guaranteed committee revision request.

9. Accepting factors with fewer than 3 items

A factor with only 1–2 items is under-identified and unreliable. Each factor should have at least 3 items (preferably 4+).

10. Not following up with reliability analysis

After EFA identifies factors, compute Cronbach's alpha for each subscale. EFA tells you the structure; reliability tells you if each factor measures consistently.

11. Confusing total variance with individual factor variance

Report both. After rotation, the variance is redistributed — the total stays the same but individual factor percentages change.

12. Not naming factors based on content

Factors should be named by what their items measure, not numbered. "Factor 1" is not a construct name — "Organisational Trust" is.

SPSS Syntax for Exploratory Factor Analysis

Reproducible syntax for every step. Copy these into your SPSS Syntax Editor and modify the variable list for your dataset.

Complete EFA with PAF and Direct Oblimin

FACTOR
  /VARIABLES item1 item2 item3 item4 item5 item6 item7
    item8 item9 item10 item11 item12 item13 item14 item15
  /MISSING LISTWISE
  /PRINT INITIAL KMO EXTRACTION ROTATION FSCORE
  /FORMAT SORT BLANK(.30)
  /PLOT EIGEN
  /CRITERIA FACTORS(3) ITERATE(25)
  /EXTRACTION PAF
  /CRITERIA ITERATE(25)
  /ROTATION OBLIMIN
  /METHOD=CORRELATION.

Maximum Likelihood with Varimax

FACTOR
  /VARIABLES item1 TO item15
  /MISSING LISTWISE
  /PRINT INITIAL KMO EXTRACTION ROTATION
  /FORMAT SORT BLANK(.30)
  /PLOT EIGEN
  /CRITERIA FACTORS(3) ITERATE(25)
  /EXTRACTION ML
  /ROTATION VARIMAX
  /METHOD=CORRELATION.

Parallel Analysis (Horn's Method)

* Parallel analysis macro for SPSS.
* Download rawpar.sps from Brian O'Connor's website and run:

INCLUDE 'C:\path\to\rawpar.sps'.
RAWPAR
  NVARS = 15
  NCASES = 312
  NDATSETS = 1000
  PERCENT = 95.

Reliability Analysis per Factor

* Factor 1: Organisational Trust.
RELIABILITY
  /VARIABLES=item1 item3 item7 item12
  /SCALE('Organisational Trust') ALL
  /MODEL=ALPHA
  /STATISTICS=DESCRIPTIVE SCALE CORR
  /SUMMARY=TOTAL.

* Factor 2: Emotional Exhaustion.
RELIABILITY
  /VARIABLES=item5 item9 item13 item14
  /SCALE('Emotional Exhaustion') ALL
  /MODEL=ALPHA
  /STATISTICS=DESCRIPTIVE SCALE CORR
  /SUMMARY=TOTAL.

* Factor 3: Work-Life Balance.
RELIABILITY
  /VARIABLES=item2 item8 item11 item15
  /SCALE('Work-Life Balance') ALL
  /MODEL=ALPHA
  /STATISTICS=DESCRIPTIVE SCALE CORR
  /SUMMARY=TOTAL.

What You Receive With Every EFA Analysis

1

Complete SPSS Output (.spv)

KMO test, Bartlett's test, eigenvalues, scree plot, unrotated solution, rotated pattern/structure matrices, communalities and factor correlation matrix — all in the original SPSS output file.

2

SPSS Syntax File (.sps)

Every command documented with comments. Includes the initial extraction, each iterative refinement run and the final solution. Fully reproducible.

3

APA Factor Loading Table

Publication-ready table with items sorted by factor, loadings, communalities, eigenvalues and variance explained. Formatted to APA 7th edition standards.

4

Written Interpretation

Plain-language narrative explaining the EFA results, factor naming rationale, items removed and why, and reliability coefficients. Ready for your Chapter 4 or Methods section.

5

Scree Plot & Factor Plot

High-resolution scree plot and factor loading plot exported as .png files at 300 DPI, formatted for publication.

6

Free Revisions

Committee requests different extraction or rotation? Want to test 3 vs. 4 factors? We revise at no extra cost until your supervisor signs off.

Frequently Asked Questions

What is exploratory factor analysis?

Exploratory factor analysis (EFA) is a multivariate statistical technique used to uncover the underlying latent structure of a set of observed variables. It identifies clusters of variables that covary strongly, grouping them into factors that represent unobserved constructs. For example, 30 survey items measuring workplace attitudes might reduce to four factors: job satisfaction, autonomy, teamwork and management trust. EFA is exploratory because the researcher does not specify the number of factors or which variables load onto which factors in advance — the data determine the structure.

When should I use EFA in my research?

Use EFA when you are developing or adapting a questionnaire, survey or measurement instrument and you do not have a pre-specified factor structure to test. Common scenarios include: developing a new scale (e.g. a burnout inventory), validating a translated instrument in a new population, reducing a large item pool to a manageable number of subscales, or when you want to understand the dimensionality of your data before running inferential analyses. If you already have a hypothesised structure, use confirmatory factor analysis (CFA) instead.

What is the difference between EFA and PCA?

EFA and principal component analysis (PCA) are often confused but serve different purposes. EFA models shared variance among variables to identify latent factors — it assumes observed variables are caused by underlying constructs plus unique error. PCA is a data reduction technique that creates linear composites (components) explaining maximum total variance, including error variance. Use EFA when your goal is to discover latent constructs underlying your items. Use PCA when you want to reduce dimensionality without theoretical interest in latent variables. In SPSS, both are accessed through the same Factor Analysis dialog but differ in the extraction method selected.

How many participants do I need for EFA?

Common guidelines include: a minimum of 300 cases (Tabachnick & Fidell), a subject-to-variable ratio of at least 10:1 (Nunnally), or at least 5 cases per variable with a minimum of 100 total (Hair et al.). However, the most defensible approach is to report KMO: values above .80 indicate your sample is adequate regardless of the exact N. For dissertation committees, a sample of 200–400 with KMO above .80 is rarely questioned. If your KMO is below .60, your data are not suitable for factor analysis regardless of sample size.

What is factor rotation and which method should I use?

Factor rotation repositions the factor axes to achieve a simpler, more interpretable structure — it does not change the mathematical solution, only how variance is distributed across factors. Orthogonal rotation (Varimax is most common) assumes factors are uncorrelated; oblique rotation (Direct Oblimin or Promax) allows factors to correlate. In social and behavioural sciences, constructs almost always correlate, so oblique rotation is usually the better default. If the factor correlation matrix shows all correlations below .32, orthogonal and oblique solutions are essentially identical and you can report the simpler orthogonal solution.

How do I determine the number of factors to extract?

Use multiple criteria rather than relying on a single rule: (1) Kaiser's criterion — retain factors with eigenvalues above 1.0, though this tends to over-extract; (2) scree plot — look for the 'elbow' where eigenvalues level off; (3) parallel analysis — compare observed eigenvalues to those from random data of the same dimensions (the gold standard; use the syntax or macro in SPSS); (4) cumulative variance explained — aim for 50–60% in social sciences; (5) theoretical interpretability — the final number should make conceptual sense. When criteria disagree, extract different numbers, compare the solutions and choose the most interpretable one.

What is an acceptable factor loading?

A loading represents the correlation between a variable and a factor. Common thresholds: .30 = minimal acceptable, .40 = more important, .50 = practically significant (Hair et al., 2019). For clean structure, each item should load above .40 on one factor and below .30 on all others. Items that load above .32 on two or more factors are cross-loading and are typically removed. Items that do not reach .40 on any factor are weak and may also be removed. Report the pattern matrix (not the structure matrix) when using oblique rotation.

How do I report EFA results in APA format?

APA reporting for EFA should include: (1) KMO value and Bartlett's test result; (2) extraction method (principal axis factoring or maximum likelihood); (3) rotation method and justification; (4) number of factors retained and the criteria used to decide; (5) total variance explained by the retained factors; (6) a factor loading table showing all items and their loadings (suppress loadings below .30 for readability); (7) communalities; (8) factor correlation matrix if oblique rotation was used; (9) any items that were removed and why. Present the loading table with factors as columns and items as rows, sorted by loading magnitude within each factor.

What do I do if my items cross-load?

Cross-loading occurs when an item loads at .32 or above on two or more factors. Options: (1) remove the item and re-run the analysis — this is the most common approach; (2) assign the item to the factor where it loads highest, if the difference between loadings is at least .20; (3) consider whether the item conceptually belongs to both factors, which may indicate the factors are not cleanly separated. After removing cross-loading items, re-run the EFA from scratch because the solution changes with every item removed. Document every removal in your methods section.

Can I run EFA on Likert scale data?

Yes, with caveats. Individual Likert items (e.g. a single 5-point item) are technically ordinal, but EFA in SPSS operates on Pearson correlation matrices, which assume continuous data. This is acceptable when you have at least 5 response options per item and reasonably normal distributions. For items with 4 or fewer categories, or severely skewed distributions, consider using polychoric correlations (available via SPSS syntax or the R psych package). Most dissertation committees accept Pearson-based EFA on 5-point or 7-point Likert items as standard practice.

Need Help With Your Exploratory Factor Analysis?

Send your dataset and research questions. We'll run the full EFA, determine your factor structure and deliver APA-formatted results within 48 hours.