What Is Factor Analysis in Simple Terms? A Plain-Language Guide with Examples

Factor analysis is a statistical method that groups many related survey questions or measurements into a smaller number of hidden "factors" — underlying themes that explain why those items tend to move together. If you have a 30-item questionnaire and suspect it's really measuring just three or four things, factor analysis is how you find out. Most people meet it while building or validating a scale for a thesis, and it feels intimidating because the output is a wall of decimals — but the core idea is genuinely simple.

Key Takeaways

• Factor analysis reduces many observed variables (like survey items) into a few underlying factors that explain the shared patterns among them.
• Use it to validate a questionnaire or discover the structure behind your items — for example, confirming that 12 items really measure "anxiety," "depression," and "stress."
• A factor loading is the correlation between an item and a factor; loadings above roughly .40 are usually considered meaningful.
• Exploratory factor analysis (EFA) discovers structure; confirmatory factor analysis (CFA) tests a structure you already expect.
• StatRyx runs factor analysis, checks the assumptions (KMO and Bartlett's test), and writes the results up in APA 7 format so you don't have to wrestle with SPSS syntax.

What Is Factor Analysis in Simple Terms?

Imagine you ask 200 people 20 questions about their wellbeing. Some questions are really about the same underlying feeling — "I feel nervous," "I worry a lot," and "I feel on edge" all tap into anxiety. Factor analysis looks at which questions rise and fall together across your respondents and bundles them into invisible groups called factors.

Think of it like sorting a messy playlist. You have 100 songs, but they naturally cluster into a handful of moods — upbeat, mellow, focus. Factor analysis does the sorting mathematically: it finds the smallest number of "moods" (factors) that still capture most of the pattern in your data.

The factors are called latent variables because you can't measure them directly — you can't ask someone "rate your anxiety factor from 1 to 7." You can only measure the observable items and let the statistics reveal the hidden theme behind them.

When Should I Use Factor Analysis?

Use factor analysis when you have many correlated variables and want to know whether they reflect a smaller set of underlying dimensions. The two most common reasons in real research are:

• Validating a questionnaire. You built or adapted a scale and want evidence that your items group the way you intended. This is the number-one use in psychology and education theses.
• Data reduction. You have 25 variables you'd like to condense into 4 or 5 composite scores before running a regression, so your model isn't cluttered and unstable.

You would not use factor analysis for a single-item measure, for purely categorical yes/no data (there are specialised versions for that), or when you have fewer participants than you have items. A common rule of thumb is at least 10 participants per item, with 200+ respondents being comfortable for a stable solution.

EFA vs CFA: What's the difference?

Exploratory factor analysis (EFA) is for discovery — you don't know how many factors exist and you let the data suggest a structure. Confirmatory factor analysis (CFA) is for testing — you already have a hypothesised structure (say, from a published scale) and you check whether your data fit it. Most students start with EFA; CFA usually comes later, often in a separate sample.

How Does Factor Analysis Actually Work?

Here are the moving parts in plain language:

• Correlation matrix: The analysis starts by measuring how strongly every item correlates with every other item. If nothing correlates, there are no shared factors to find.
• Factor extraction: The method (commonly principal axis factoring or maximum likelihood) pulls out factors that account for the shared variance.
• Eigenvalues: Each factor has an eigenvalue — how much total variance it explains. A traditional cut-off keeps factors with an eigenvalue greater than 1 (the Kaiser criterion), though a scree plot is often more reliable.
• Rotation: Factors are "rotated" (e.g. Varimax or Oblimin) to make them easier to interpret, so each item loads clearly on one factor. Rotation doesn't change the underlying maths — it just tidies the view.
• Factor loadings: The final table shows how strongly each item ties to each factor. A loading is essentially a correlation, ranging from −1 to +1.

A Worked Example with Real Numbers

Suppose you survey 250 university students with a 9-item wellbeing scale and run an EFA in StatRyx. First you check whether the data are suitable:

• Kaiser–Meyer–Olkin (KMO) = .84 — anything above .70 is good; this means the items share enough common variance to factor.
• Bartlett's test of sphericity: χ²(36) = 1,204.6, p < .001 — significant, so the correlation matrix isn't random. Both checks pass.

The analysis extracts three factors with eigenvalues above 1, together explaining 62% of the total variance. After Varimax rotation, the loadings look like this:

Reading the table: Items 1–3 load strongly on Factor 1 (all above .70) and weakly on the others — so Factor 1 is clearly anxiety. Items 4–6 cluster on Factor 2 (depression) and items 7–9 on Factor 3 (stress). The bolded loadings above .40 confirm your 9-item scale really measures three clean dimensions — exactly what you'd hope to report in a scale-validation study.

What Do the Numbers Mean?

• A loading of .78 means that item is strongly related to its factor — it's a good indicator of anxiety.
• A cross-loading of .21 on another factor is small and expected; you only worry when an item loads above about .40 on two factors, which makes it ambiguous.
• 62% variance explained means your three factors capture nearly two-thirds of everything the nine items measure — a solid result for the social sciences, where 50–60% is often acceptable.

If you're deciding between this and a related technique, see how it compares in our guide on principal component analysis vs factor analysis — they look similar but answer different questions (PCA summarises variance; factor analysis models a latent cause).

How to Report Factor Analysis in APA 7

Report the suitability checks, the extraction method, the number of factors, variance explained, and the rotated loadings. A clean APA-style write-up reads:

An exploratory factor analysis with maximum likelihood extraction and Varimax rotation was conducted on the nine wellbeing items (N = 250). The data were suitable for factoring, KMO = .84, Bartlett's test of sphericity χ²(36) = 1,204.6, p < .001. Three factors with eigen