Principal Components Analysis
VerifiedFactor Analysis
Run this test straight away on a free built-in teaching dataset — no data of your own needed — or bring your own. Either opens the guided workspace: variable setup, assumption diagnostics, results with effect sizes and confidence intervals, figures, and APA-ready reporting.
Loading teaching datasets…
Or use your own dataset
Loading your datasets…
Reduces many correlated variables into a few uncorrelated components that capture most of the variance.
PCA finds linear combinations (components) ordered by how much variance they explain — useful for dimension reduction, composite indices, or spotting structure. It's descriptive, not a latent-measurement model.
Worked example
Can 12 economic indicators be summarised by a few components?
PCA on 12 standardised indicators (n = 200), using a scree plot to choose the number of components.
Three components (eigenvalue > 1) explained 73% of the variance; the first component alone accounted for 29%.
A PCA reduced the 12 indicators to three components explaining 73% of the variance.
Try it yourself: Load this ready-made sample and follow the run above.
When to use it
- Reducing many correlated variablesYou want a few uncorrelated components that capture most of the variance. e.g. 12 economic indicators into 3 components.
- Building a composite indexThe first component is a common way to weight indicators into a single index.
- A quick look at structureScree and loadings hint at how many dimensions the data carry — as description, before any modelling.
When NOT to — use instead
- You want a latent-construct modelPCA components are weighted sums of the observed variables, not latent factors with error terms. → Exploratory Factor Analysis
- Confirming a hypothesised factor structureTo test a specified measurement model, fit it directly. → Confirmatory factor analysis
Assumptions (and what to do if they fail)
Check: Check sampling adequacy (KMO ≥ .6) and Bartlett's test before extracting; near-uncorrelated variables cannot be summarised.
If violated: PCA on essentially independent variables just relabels them — nothing is reduced.
Check: PCA on the covariance matrix lets large-variance variables dominate; use the correlation matrix (standardized) unless the scales are comparable on purpose.
If violated: One high-variance variable can hijack the first component.
Check: Do not interpret components as error-free latent traits.
If violated: Treating a component as a latent factor over-claims what PCA delivers.
Check: PCA captures linear structure; strong non-linearity is missed.
If violated: Curved structure is under-represented in the components.
Ready to run a Principal Components Analysis on your own data?
Guided setup, automatic assumption checks, effect sizes, figures and an APA write-up.
Run this test →