How To Do Factor Analysis In Spss

How to Do Factor Analysis in SPSS

Factor analysis is a powerful statistical technique used to uncover the underlying structure of a set of observed variables. SPSS (Statistical Package for the Social Sciences) provides a user‑friendly interface for conducting both exploratory and confirmatory factor analyses. Even so, researchers in psychology, sociology, marketing, and many other fields rely on it to reduce data dimensionality, validate scales, and identify latent constructs. This guide walks you through the entire process—from preparing your data to interpreting the output—so you can confidently apply factor analysis in your own projects Simple, but easy to overlook. Worth knowing..

Detailed Explanation

At its core, factor analysis seeks to explain correlations among a larger set of variables by a smaller number of unobserved (latent) factors. Imagine you have a questionnaire with 20 items measuring various aspects of job satisfaction. And rather than treating each item separately, factor analysis can reveal that these items actually cluster into, say, three broader dimensions: intrinsic satisfaction, extrinsic satisfaction, and work‑environment satisfaction. Those dimensions are the factors; each observed item loads (i.But e. , correlates) with one or more factors to varying degrees Which is the point..

There are two main types of factor analysis you can run in SPSS:

1. Exploratory Factor Analysis (EFA) – used when you do not have a pre‑specified hypothesis about the number or nature of factors. SPSS’s Factor procedure (under Analyze → Dimension Reduction → Factor) performs EFA by extracting factors, rotating them for interpretability, and providing statistics such as Kaiser‑Meyer‑Olkin (KMO) measure of sampling adequacy and Bartlett’s test of sphericity.

2. Confirmatory Factor Analysis (CFA) – used when you have a theoretical model you want to test. While SPSS does not have a built‑in CFA module, the AMOS add‑on (which integrates with SPSS) allows you to specify and evaluate measurement models using structural equation modeling techniques Less friction, more output..

For most beginners, the focus is on EFA because it helps you discover the factor structure before moving to more rigorous hypothesis testing. The steps below assume you are conducting an exploratory factor analysis.

Step‑by‑Step or Concept Breakdown

Below is a detailed, numbered workflow you can follow in SPSS version 27 or later. Each step includes the menu path, key decisions, and what to look for in the output.

1. Prepare Your Data

   • Ensure all variables are scale (interval or ratio) or at least treated as continuous.
   • Check for missing values; SPSS will exclude cases listwise by default unless you specify pairwise or imputation.
   • Consider reversing negatively worded items so that higher scores consistently reflect the same direction of the construct.

2. Assess Suitability for Factor Analysis

   • Go to Analyze → Descriptive Statistics → Correlate → Bivariate to inspect the correlation matrix. Look for many correlations above .30; if most are near zero, factor analysis may not be appropriate.
   • Run the Factor procedure once to obtain the KMO and Bartlett’s test: Analyze → Dimension Reduction → Factor.
   • In the Descriptives button, check KMO and Bartlett’s test of sphericity. A KMO value > .60 (ideally > .80) and a significant Bartlett’s test (p < .001) indicate adequate sampling.

3. Extract Factors

   • Still in the Factor dialog, click Extraction.
   • Choose a method: Principal Axis Factoring (PAF) is common for EFA when you want to uncover latent constructs; Principal Components Analysis (PCA) is used when the goal is purely data reduction.
   • Decide how many factors to retain. Options include:
     • Kaiser criterion (eigenvalue > 1) – automatic but can overextract.
     • Scree plot – look for the “elbow” where eigenvalues level off.
     • Parallel analysis – compare eigenvalues to those from random data (requires a macro or external tool).
   • Set the number of factors accordingly (e.g., select “Fixed number of factors” and enter 3).

4. Rotate the Solution

   • Click Rotation and choose an orthogonal (e.g., Varimax) or oblique (e.g., Direct Oblimin) method.
   • Use orthogonal rotation if you assume factors are uncorrelated; oblique if you expect them to correlate (common in psychological scales).
   • Keep the default Display option to show the rotated factor matrix.

5. Request Additional Output

   • In the Options button, you can request Coefficient Display Format (e.g., sort by size) and Missing Values handling (e.g., replace with mean).
   • Click Continue, then OK to run the analysis.

6. Interpret the Output

   • Communalities: Values indicate the proportion of each variable’s variance explained by the extracted factors. Low communalities (< .30) suggest the item may not fit well.
   • Total Variance Explained: Look at the cumulative percentage after rotation; a solution explaining 50‑70 % of variance is often considered acceptable in social sciences.
   • Rotated Factor Matrix: Identify which items load strongly (typically ≥ .40 or .50) on each factor. Items that load on multiple factors (cross‑loadings) may need revision or removal.
   • Factor Correlation Matrix (if oblique rotation): Examine the correlations among factors; high correlations (> .70) may suggest the factors are not distinct.

7. Refine and Re‑run

   • Based on the initial run, you may decide to drop poorly performing items, adjust the number of factors, or try a different extraction method.
   • Repeat steps 2‑6 until you achieve a clean, interpretable solution.

Real Examples

Example 1: Job Satisfaction Survey
A researcher administers a 15‑item job satisfaction scale to 500 employees. After checking missing data (<2 %) and reversing two negatively worded items, they run an EFA with Principal Axis Factoring and Direct Oblimin rotation. The KMO is .88 and Bartlett’s test is significant (p < .001). The scree plot shows a clear elbow after three factors, explaining 62 % of the variance. The rotated matrix reveals:

• Factor 1 (Intrinsic Satisfaction): items 1, 2, 3, 4 load .68‑.81.
• Factor 2 (Extrinsic Satisfaction): items 5, 6, 7, 8 load .55‑.74.
• Factor 3 (Work‑Environment): items 9‑15 load .48‑.76.

All communalities exceed .Worth adding: 45), supporting a three‑dimensional model. 30‑.On the flip side, 40, and factor correlations are modest (. The researcher proceeds to compute scale scores by averaging the items loading on each factor and uses them in subsequent regression analyses Simple as that..

**Example 2: Customer Perception of a New

Example 2: Customer Perception of a New Mobile Banking App
A fintech company launches a redesigned mobile banking platform and wants to understand how users experience its various functionalities. A panel of 320 customers completes a 22‑item questionnaire covering aspects such as security confidence, transaction ease, visual design, customer support, multi‑factor authentication, and personalized alerts. Two items show missing responses (≈1 % of the total), which are handled by pairwise deletion in SPSS And it works..

The analyst follows the workflow outlined earlier:

1. Data Preparation – The dataset is entered, missing values are flagged, and two negatively worded items (e.g., “I feel uneasy using the app’s biometric login”) are reverse‑scored to align with the overall scoring direction.

2. EFA Request – Using Analyze → Dimension Reduction → Factor, the extraction method is set to Principal Axis Factoring (PAF) because the goal is to uncover latent constructs rather than merely describe variance.

3. Extraction Settings – The default Scree plot is retained, and the number of factors is initially left to the program’s automatic extraction (which will be refined after inspecting the scree) Simple, but easy to overlook..

4. Rotation Choice – Direct Oblimin rotation is selected to allow the emerging factors to correlate, reflecting the realistic expectation that, for example, perceived security and ease of use often overlap in users’ minds.

5. Additional Output – In the Options dialog, the analyst checks Coefficient Display Format (sorted by absolute size) and enables Missing Values handling via pairwise deletion. The rotated factor matrix is displayed for review Not complicated — just consistent..

6. Interpretation of the First Run

   • KMO & Bartlett’s Test – KMO = .91 (meritorious) and Bartlett’s χ² is highly significant (p < .001), confirming that the correlation matrix is suitable for factor analysis And that's really what it comes down to..

   • Scree Plot – A clear elbow appears after the fourth component, suggesting a four‑factor solution That's the part that actually makes a difference..

   • Total Variance Explained – The initial (unrotated) solution accounts for 73 % of variance; after rotation, the cumulative variance rises to 78 %—well within the 50‑70 % range commonly accepted for social‑science research Took long enough..

   • Rotated Factor Matrix – The matrix highlights four coherent dimensions:

     Factor 1 – Security Trust: items 1, 4, 9, 14 load .71‑.84.
     Factor 2 – Transaction Efficiency: items 2, 5, 10, 15 load .68‑.79.
     Factor 3 – Interface Aesthetics: items 3, 7, 12, 18 load .62‑.73.
     Factor 4 – Support & Personalization: items 6, 11, 16, 20 load .59‑.71 Small thing, real impact..

   • Communalities – All items exceed .45, indicating acceptable representation within the factor structure.

   • Factor Correlation Matrix – Correlations among the four factors range from .32 to .48, suggesting moderate interrelatedness without redundancy.

7. Refinement – The initial solution is deemed satisfactory, but two items (e.g., “I feel confident using the app’s chat support”) show modest loadings (.38) and a cross‑loading on Factor 4 and Factor 2. After discussion with subject‑matter experts, the team decides to drop the item, re‑run the analysis, and verify that the remaining items still meet the loading threshold (≥ .40). The revised run yields an identical factor pattern with improved overall model fit (cumulative variance = 80 %).

Next Steps – The researcher now creates subscale

scores by averaging the items within each factor, computes reliability coefficients (Cronbach’s α > .In real terms, 80 for all subscales), and exports the factor scores for use in subsequent regression or segmentation analyses. A summary table documenting item‑total statistics, factor loadings, and variance explained is prepared for inclusion in the final report. This systematic workflow ensures that the exploratory factor analysis is not only statistically sound but also interpretable and actionable for stakeholders.

Short version: it depends. Long version — keep reading It's one of those things that adds up..

So, to summarize, conducting EFA in jamovi—when guided by clear methodological steps and thoughtful interpretation—offers researchers a transparent and efficient path from raw survey data to validated multi‑dimensional constructs. By following this structured approach, analysts can confidently uncover the latent structure underlying user‑experience instruments while maintaining rigor and reproducibility throughout the analytical process.