What it means
The method starts from correlations. If several variables consistently rise and fall together, the reasonable explanation is that something common sits behind them, and factor analysis estimates those common somethings, which are called factors, along with how strongly each variable relates to each one.
The strength of a variable's relationship to a factor is called its loading, and loadings are the output people actually read. A question about delivery speed loading at 0.85 on a factor while a question about price loads at 0.05 tells you the factor is about service, not cost.
In finance and business the technique appears in three main places. It reduces survey and market research data to a few meaningful dimensions, it builds risk models that explain why many assets move together, and it constructs credit or behavioural scorecards from dozens of correlated inputs.
Two variants get confused. Exploratory factor analysis lets the data suggest how many factors exist and what they contain, while confirmatory factor analysis tests whether a structure you already believe in actually fits, and the two answer quite different questions.
The honest limitation is that factors are mathematical constructs, not facts. The software will name nothing; a human decides that Factor 1 means "perceived service quality", and that interpretation can be wrong, self-serving or simply an artefact of which questions were asked in the first place.
In practice
Real-world examples.
Example
A retail bank runs factor analysis on a 30-question customer satisfaction survey and finds three factors: ease of everyday transactions, confidence in staff advice and pricing fairness. The board stops reporting 30 separate scores and starts managing three, which changes what branch managers are held accountable for.
Example
An asset manager applies factor analysis to the daily returns of 400 shares and discovers that most co-movement is explained by four factors, one of which behaves like sensitivity to interest rates. The firm then hedges that exposure directly rather than trading individual positions.
Example
A subscription business feeds 45 usage metrics into a factor analysis to build a churn model. The metrics collapse into two factors, depth of feature adoption and frequency of logins, which become the two numbers displayed on the customer success dashboard.
Think of it
“Factor analysis finds the hidden drivers-common underlying forces explaining patterns in returns.
Formula
Calculation
Each variable is modelled as: Variable = (Loading 1 x Factor 1) + (Loading 2 x Factor 2) + Unique component
Two derived numbers do most of the interpretive work. A variable's communality is the sum of its squared loadings, showing how much of that variable the factors explain. A factor's eigenvalue is the sum of squared loadings down the factor's column, showing how much total variance that factor accounts for.
Suppose five survey questions load on a single retained factor with loadings of 0.9, 0.8, 0.7, 0.6 and 0.5.
Squared loadings: 0.81, 0.64, 0.49, 0.36 and 0.25
Eigenvalue = 0.81 + 0.64 + 0.49 + 0.36 + 0.25 = 2.55
Variance explained = 2.55 / 5 variables = 0.51, or 51%
The first question has a communality of 0.81, meaning the factor explains 81% of its variation, while the last has a communality of 0.25, so three quarters of that question is about something the factor does not capture.Case study
Seen in the real world.
This fictional example is provided for illustrative purposes. Ashfield Grocers collected 26 store performance metrics every month and its regional directors argued endlessly about which ones mattered. An analyst ran a factor analysis and found that 22 of the 26 metrics loaded onto just three factors, which the team labelled stock availability, staffing consistency and store condition.
The interesting result was the four metrics that loaded on nothing. Each of those measured something genuinely independent, including local competitor openings, and the company had been treating them as interchangeable with the other 22 in its ranking system.
In this illustrative case the monthly pack was rebuilt around the three factor scores plus the four standalone measures. Meetings shortened, and because the factors had plain-language names, store managers could finally act on the reporting rather than debating which of 26 numbers to chase.
Watch out
Common mistakes.
- Treating factors as real, physical things rather than statistical summaries that a human has chosen to name and interpret.
- Running the analysis on far too few observations, where a rough working minimum is several times as many cases as variables, which produces unstable and unrepeatable loadings.
- Retaining too many factors because the software offers them, then straining to invent a story for a factor that explains very little of the variance.
Questions
People also ask.
How many factors should I keep?
Keep the ones that explain a meaningful share of variance and that you can describe in plain language, using an eigenvalue above 1 and a scree plot as starting guides rather than rules.
Is factor analysis the same as principal component analysis?
No, they are closely related but factor analysis models shared variance and assumes underlying causes, while principal component analysis simply repackages total variance into new axes.
Do I need a statistician to use it?
For a one-off exploratory look the software is accessible, but for anything driving capital allocation or pricing you want someone who can judge the assumptions and test stability.
From the founder's library
Take it further with the book.
Build your financial confidence beyond this definition. Shihan's full-length guide, Accounting Fundamentals, takes the same plain-English approach and turns it into a complete, practical playbook for non-finance managers, business owners and students - with chapter-end quiz answers and presentation slides included.
25% off with code MMHQ25, applied at checkout. Priced in USD - checkout may show the equivalent in your local currency.
View the book and save 25%Related