Do the 24 items of your questionnaire really measure the four dimensions you named in the methodology chapter? If you adapted a scale, translated it or built it yourself, a committee may ask for evidence of its structure. Exploratory factor analysis is the usual first step: it shows which items group together, which items belong nowhere, and how many dimensions the data support.
This article is part of our guide How to Write a Master’s Thesis by the Mastermind PhD team and relates to our Statistical Analysis service. Sources last checked: November 2026.
SPSS makes the procedure look like one click, but it hides seven decisions behind default settings, and several of those defaults are not what methodologists recommend. This guide walks through each decision, the output, and how to report it.
What is exploratory factor analysis? It is a multivariate technique that identifies a smaller number of latent factors explaining the correlations among a set of observed variables, typically questionnaire items. It is called exploratory because the number of factors and the item-factor assignments are estimated from the data rather than fixed in advance, as they are in confirmatory factor analysis.
When do you need exploratory factor analysis?
Use it when you develop a new scale, when you adapt or translate an existing scale for a different culture or population, or when previous studies disagree about a scale’s structure. If the structure is well established and you want to test it, confirmatory factor analysis (CFA) in AMOS or similar software is the stronger choice. Some theses run exploratory factor analysis on one half of the sample and a CFA on the other half.
Decision 1: Is the sample large enough?
There is no single rule. Common guidance includes at least 5 to 10 participants per item and an absolute minimum around 100 to 200 cases. Research reviewed by Costello and Osborne (2005) shows that larger samples produce more stable solutions, and that strong loadings and several items per factor reduce the sample needed. A 30-item scale with 150 respondents is a weak basis for a confident exploratory factor analysis.
📱 Ask a statistician about your data — Send a short description of your instrument and sample size on WhatsApp and get a suitable analysis plan. Every analysis comes with the raw SPSS file and a results report.
Decision 2: Are the data suitable? KMO and Bartlett’s test
| Check | Criterion | Meaning |
|---|---|---|
| Kaiser-Meyer-Olkin (KMO) | At least .60; Kaiser (1974) called .80 meritorious and .90 marvelous | Correlations are compact enough to form factors |
| Bartlett’s test of sphericity | p < .05 | The correlation matrix is not an identity matrix |
| Individual MSA values | Diagonal of the anti-image correlation matrix above .50 | Each item fits the set; remove items below .50 one at a time |
| Correlation matrix | Many correlations above .30, none above about .90 | Enough shared variance, no near-duplicate items |
Decision 3: Which extraction method?
SPSS defaults to principal components, which is strictly a data-reduction technique rather than factor analysis. For identifying latent constructs, Costello and Osborne (2005) recommend a true factor method: principal axis factoring when normality is doubtful, or maximum likelihood when the items are approximately normal, which also gives a goodness-of-fit test. Many published studies still use principal components; whichever you choose, name it in the exploratory factor analysis section of your methodology.
Decision 4: How many factors should you keep?
- Kaiser’s criterion (eigenvalues above 1) is the SPSS default and tends to keep too many factors.
- The scree plot: keep the factors above the point where the curve flattens.
- Parallel analysis compares your eigenvalues with those from random data and is widely regarded as the most accurate method. It needs a syntax file or an online calculator, since SPSS menus do not offer it in older versions.
- Theory: a solution must be interpretable. A factor with two items or no clear meaning is not a real dimension.
Run the analysis with the number suggested by parallel analysis and the scree plot, then compare it with one factor more and one less.
Decision 5: Which rotation?
Rotation makes the solution easier to interpret. Orthogonal rotation (Varimax) assumes the factors are uncorrelated; oblique rotation (Direct Oblimin or Promax) allows them to correlate. Dimensions of a psychological or social construct almost always correlate, so oblique rotation is the more realistic default. If the factor correlation matrix shows correlations below about .32, an orthogonal rotation gives similar results (Tabachnick & Fidell, 2019).
With oblique rotation, interpret the pattern matrix, which shows each item’s unique relationship with each factor.
Decision 6: Which items should you keep?
- Loadings: a common threshold is .40; Tabachnick and Fidell (2019) suggest .32 as a minimum. Set “Suppress small coefficients” to the threshold under Options so the matrix is readable.
- Cross-loadings: an item that loads above the threshold on two factors with a difference below about .20 does not belong clearly to either.
- Communalities: items with very low communality (below about .30 to .40) share little variance with the others.
- Factors: aim for at least three items per factor.
Remove items one at a time and rerun the exploratory factor analysis, because each removal changes all the other values. Record every removal and the reason; examiners will ask.
Decision 7: How do you run it in SPSS?
- Analyze, then Dimension Reduction, then Factor.
- Move the items into Variables.
- Descriptives: tick KMO and Bartlett’s test of sphericity, and Anti-image.
- Extraction: choose Principal axis factoring or Maximum likelihood, tick Scree plot, and set the number of factors if you have decided it.
- Rotation: choose Direct Oblimin or Promax.
- Options: Exclude cases listwise, Sorted by size, Suppress small coefficients at .40 (or your chosen value).
- Click OK, and save the syntax with Paste so the analysis can be reproduced.
How do you report exploratory factor analysis?
Report the sample size, the extraction method, the rotation, the KMO value, Bartlett’s test, the criteria for the number of factors, the variance explained, the items removed with reasons, and a table of the pattern matrix.
The following is an illustrative example of the wording, not real results: “An exploratory factor analysis using principal axis factoring with Promax rotation was conducted on the 24 items (N = 412). The KMO measure was .89 and Bartlett’s test was significant, χ²(276) = 5,127.40, p < .001. Parallel analysis and the scree plot supported four factors, which together explained 58.3% of the variance. Two items were removed because of cross-loadings.”
Add the loading table, sorted by factor, with the factor correlations in a note. Then report Cronbach’s alpha or McDonald’s omega for each retained factor.
📱 Ask on WhatsApp now — 5 years of experience across all disciplines and 14 academic services. Two-stage payment (80/20); revisions are free until supervisor approval, within 30% of the chapter.
Frequently asked questions
What is the difference between exploratory factor analysis and PCA?
Principal component analysis summarizes total variance into components; factor analysis models only the shared variance and assumes latent factors cause the item responses. For scale validation, a true factor method is usually recommended, although results are often similar with many items and high communalities.
What is the difference between exploratory and confirmatory factor analysis?
EFA estimates the structure from the data. CFA tests a structure you specify in advance and gives fit indices such as CFI and RMSEA. They answer different questions and should not be run on the same data to confirm each other.
Can I use exploratory factor analysis on an established scale?
You can, especially after translation or adaptation to a new population. If the original structure is well supported, a CFA is the more direct test.
What if an item loads on the wrong factor?
Check its wording, translation and reverse coding first. If it still loads elsewhere, either keep it on the factor where it loads with a theoretical justification or remove it. Do not move it silently.
What KMO value is acceptable?
Values of .60 or above are usually considered acceptable, and .80 or above good. Kaiser’s labels range from unacceptable below .50 to marvelous at .90 and above.
Sources
- Costello, A. B., & Osborne, J. W. (2005). Best practices in exploratory factor analysis: Four recommendations for getting the most from your analysis. Practical Assessment, Research, and Evaluation, 10, Article 7. https://doi.org/10.7275/jyj1-4868
- Kaiser, H. F. (1974). An index of factorial simplicity. Psychometrika, 39(1), 31–36. https://doi.org/10.1007/BF02291575
- Tabachnick, B. G., & Fidell, L. S. (2019). Using Multivariate Statistics (7th ed.). Pearson.
- Field, A. (2018). Discovering Statistics Using IBM SPSS Statistics (5th ed.). SAGE.
- Pallant, J. (2020). SPSS Survival Manual (7th ed.). Open University Press / Routledge.
Sources last checked: November 2026. This article is general educational guidance and does not replace your university regulations or your supervisor’s instructions. How we prepare our content.