Factor analysis
Factor analysis models shared patterns among items through unobserved factors; exploratory and confirmatory variants differ in how much structure you specify first.
Fit
When to use it, and when not to
Use it when
- You want to examine the internal structure of a set of related items.
- You have a proposed construct and data suited to estimating shared item variation.
Not when
- You only need agreement between raters. Use an agreement statistic. Read: Bland-Altman method comparison
- You need evidence that a questionnaire covers the intended experience. Assess content validity with its target users. Read: Measurement properties and the designs that test them
Before you start
What you need
- Items, response scales and the intended constructExplain why the items should share a factor and whether responses are ordered categories.
- An exploratory or prespecified structureFor confirmation, state which items load on each factor before fitting.
- A sample suited to model complexityPlan for loading strength, item categories and missing responses; no fixed persons-per-item rule guarantees adequacy.
- The response model fits the item scales.How to check: Use methods suited to ordered responses rather than treating few categories as continuous without justification.
- The model explains the main shared item variation.How to check: Inspect remaining associations and cross-loadings. Do not add correlated errors solely to chase fit.
- Confirmation is separated from exploratory choices.How to check: Use a new sample or a defensible split. Report any changes made after looking at fit.
Plain language
How it works
A factor is an unobserved variable used to explain why items vary together. A loading links an item to that factor. Residual variation remains outside the shared factor.
Exploratory factor analysis estimates a plausible structure with fewer fixed loading restrictions. Confirmatory factor analysis tests a structure you specify first. Correlated factors may be appropriate.
Choose factor count using theory and empirical checks. Parallel analysis compares variation explained by candidate factors with that expected from random data. Fit statistics do not settle the construct. The cited scale studies show exploratory and confirmatory work in measurement development.
Worked example
A worked example
- Setting
- Researchers examine a new symptom questionnaire.
- Question
- How much of one standardized item is explained by a single factor?
- Approach
- Assume one factor with unit variance, an item with unit variance, loading 0.8 and an uncorrelated residual.
- Result
- The factor explains 64 in 100 variance units: 64 of 100. The residual accounts for 36 in 100 variance units.
- What it means
- Squared loading gives explained variance under this simple model. It does not mean the item is accurate for 64 in 100 patients.
In practice
Recent studies PubMed search · not vetted
Watch out
Common mistakes
The one to remember
A factor is a model of shared item variation, not proof that a construct exists or that a score is valid. Exploration and confirmation on the same data are not independent evidence.
- Treating an exploratory solution as independent confirmation.Instead: Test the proposed structure with fresh data.
- Calling a good fit proof of validity.Instead: Combine structural evidence with content and other evidence for the intended use.
Inside Mallard
How Mallard uses this
Plans do not link to this method yet. Mallard does not run this analysis.
analysis.factor_analysis, the id a plan cue records.Related Learn pages
Related designs and methods
Evidence
Sources & further reading
- Factor analysis and evaluation of one-year test-retest reliability of the 33-item Childhood Trauma Questionnaire in Chinese adolescents (opens in a new tab) · Front Psychol · doi 10.3389/fpsyg.2024.1384807
- Development and Psychometric Validation of the TF-Col: A Patient- and Caregiver-Reported Measure of Financial Toxicity in Oncology (opens in a new tab) · Value Health Reg Issues · doi 10.1016/j.vhri.2026.101695
- A RATIONALE AND TEST FOR THE NUMBER OF FACTORS IN FACTOR ANALYSIS (opens in a new tab) · Psychometrika · doi 10.1007/BF02289447