If you’ve ever created a survey using Likert scales (“Strongly disagree” → “Strongly agree”), you may have ended up with a pile of questions that all seem related—but you’re not sure how they fit together.
That’s where factor analysis comes in. It’s a statistical method that helps you uncover the hidden structure behind your survey items—essentially grouping questions into meaningful categories (called factors).
This guide will walk you through how to run factor analysis step by step.
Step 1: Understand the Basics
- What is factor analysis?
A statistical method that reduces many survey items into a smaller number of factors. - Why use it?
To see if your questions are measuring the same underlying concept. For example, 10 Likert items about “job satisfaction” might actually group into 2 factors: work environment and career growth.
Step 2: Prepare Your Data
Before you jump into analysis:
- Check your sample size. A common rule of thumb: at least 5–10 responses per item, with a minimum of ~100 participants.
- Ensure your variables are correlated. Factor analysis works best if items are moderately correlated.
- Clean your data. Handle missing values, reverse-code negatively worded items, and make sure all Likert responses use the same scale (e.g., 1–5 or 1–7).
Step 3: Choose the Right Method
There are two main types:
- Exploratory Factor Analysis (EFA): Use when you don’t know the factor structure beforehand.
- Confirmatory Factor Analysis (CFA): Use when you already have a theory about how items should group.
💡 For beginners, start with EFA.
Step 4: Run the Analysis
Here’s how you might do it in practice (example in SPSS, but R, Stata, or Python can do the same):
- Go to Analyze → Dimension Reduction → Factor.
- Select your Likert-scale items.
- Choose Extraction method (e.g., Principal Axis Factoring).
- Look at KMO test (Kaiser-Meyer-Olkin > 0.6 is acceptable).
- Run the Bartlett’s Test of Sphericity (should be significant).
- Decide how many factors to keep (check eigenvalues > 1, scree plot, or parallel analysis).
- Apply rotation (Varimax for independent factors, Oblimin for correlated ones).
Step 5: Interpret the Results
- Factor loadings: Show how strongly each question relates to a factor.
- Generally, loadings above 0.40 are considered meaningful.
- Cross-loadings: If an item loads on more than one factor, decide whether to drop or revise it.
- Naming the factors: Look at which items cluster together and give that group a descriptive label (e.g., “Work-Life Balance,” “Leadership Support”).
Step 6: Validate the Factors
- Check Cronbach’s alpha for internal consistency (α > 0.7 is acceptable).
- If you plan to publish, consider running a Confirmatory Factor Analysis (CFA) on a separate dataset to confirm your structure.
Example Scenario
Imagine you surveyed employees with 12 Likert-scale questions about job satisfaction. After running EFA, you find:
- Factor 1: Questions about pay, promotions, and career growth → “Career Satisfaction”
- Factor 2: Questions about teamwork, leadership, and recognition → “Workplace Support”
Now you can analyze each factor separately instead of juggling 12 individual items.
In short
Running factor analysis on Likert-scale data might sound intimidating, but it’s really about finding patterns in your questions.
- Use EFA to explore.
- Use rotation to make results easier to interpret.
- Always check reliability before drawing conclusions.
Once you’ve done it a couple of times, you’ll find factor analysis is one of the most powerful tools in your research toolkit.
