Factor analysis is a technique that helps researchers study a concept that cannot easily be measured
Very often, research involves examining concepts that are not easy to measure. How do you measure quality of life, mental health, or social support if you do research in psychology, for example? Or if you’re an economist, how would you measure entrepreneurialism, risk aversion or corporate social responsibility?
To conduct research on something you can’t measure directly, you might need a statistical technique called factor analysis. This technique is used when you have a latent variable—a concept/idea that’s not easy to measure directly—and you’re trying to understand it by measuring observed variables. Let’s look at an example.
Suppose that you are on a research team looking to understand stress among students at a college campus. To carry out the research, you need to come up with some way of measuring what you’re studying. What questions would you ask or what data would you gather to measure the stress experienced by college students? The list of variables could be very long.
You could ask students for information about the demands of their academic program, such as course load and difficulty level. You could get information from students about their volunteering, their commute and their job commitments. You could also ask them about their physical exercise, their diet, their sleep patterns and the number of days they’re off sick. You could ask questions about their emotional states and psychological outlooks, such as questions about worry, optimism amd social connectedness.
Factor analysis is an important tool to help sort out which of these variables are more important for measuring a latent variable. It’s a very complicated technique that requires a good foundation of statistics to understand. But in essence, the technique is about examining the data you’ve got to see whether scores of certain measured variables tend to move in the same pattern.
In the above example, a factor analysis might show that answers related to commute, paid work and extracurricular activities tend to share a pattern—in other words, they are correlated to one another. That observation leads you to the recognition that those questions all get at an underlying concept of time pressure or time availability. Likewise, from the results of your factor analysis you might learn that the answers you get for questions about diet, sleep, exercise and sick days also share a similar pattern. In effect, these measures of diet, sleep and exercise are observable variables that together help you measure the latent concept of physical health.
It’s possible you also learn from the factor analysis that some questions you’ve asked do not share a pattern in their responses to any other variable—that is, they are not correlated to any of the other questions. This might help you understand how you could shorten your survey.
As this example demonstrates, sometimes you’re working with more than one latent variable at a time. The main focus of your study—stress experienced by college students—is itself a latent variable. But it can be understood as a composite of other latent variables—academic demands, time pressure, physical health and mental health.
There are two types of factor analyses. One is called exploratory factor analysis, where the analysis helps you discover the underlying structure of your data—that is, which variables group together. The other type of factor analysis is called confirmatory factor analysis. That’s what you would use when you already have a good theory or hypothesis about how observed variables relate to the latent factors, and all you need is to run a test to see if the data support your theory.
In sum, factor analysis has many uses. It helps you identify factors you can measure that underlie the variables you are interested in. It helps you see groupings of similar variables so you can choose one variable to represent many. With a reduced number of variables, you can more easily create measurement tools such as questionnaires.
Source: At Work, Issue 89, Summer 2017: Institute for Work & Health, Toronto