In most social science experiments, there is no direct measure of concepts. Instead, we use observable measures to quantify the value of the hidden trait.
Latent variables are the hidden or unobserved elements we’re measuring in this experiment.
Before being used, latent variables must also be tested and proven to be valid and reliable indicators. Otherwise, you will be inferring the value of an unobservable concept using assumptions.
There are several techniques for measuring latent variables, and each one is effective in different use cases.
In this article, we’ll go over what latent variables are, how to measure them, their significance, and examples.
What Are Latent Variables
Variables in research are nonconstant, measurable, and controllable entities. But here’s the thing: some variables are easier to quantify than others.
Latent variables are those variables that are measured indirectly using observable variables. So, rather than measuring things that can’t be quantified, we infer the value using variables we can quantify.
For example, there’s no unit for measuring employee dedication or user experience. But these variables are important to improving workplace culture and product value.
Instead of hoping for the best, researchers measure observable variables related to these unquantifiable variables to deduce their values.
For example, a user experience researcher creates a survey to gauge user satisfaction with a web design. The survey questions can’t measure user experience directly; rather, they infer user satisfaction using a couple of factors.
What Is a Latent Variable Model?
Latent variable models postulate how the properties of observable variables (indicators) relate to latent variables. There are several categories of latent variable models.
Types of Latent Variable Models
Factor Analysis
This method allows you to compress large amounts of variables into smaller measurable variables (factors).
Factor analysis employs a combination of statistical methods to identify latent variables from collected data (manifest variables). This technique is best suited for market research, psychological, and sociological studies.
For example, you want to assess a newly launched e-commerce website using a survey. The survey would most likely include questions about:
- Ease of Navigation
- website loading speed
- website aesthetics
- their likelihood of returning to the website
- Probability of recommendation to family and friends
These variables are quite a handful, but they are not the primary focus of your survey. You’re conducting this research to see how smooth the user experience on the new website is.
So, you’re measuring the underlying variable by combining the results of these surveys. While there are various techniques for conducting factor analysis, they all have one thing in common: you’re shrinking massive amounts of data to get to the underlying concepts; latent variables.
Latent Trait Analysis
Latent trait analysis employs statistical models to represent the connection between latent traits and their observed variables. This model establish a link between the characteristics of the items being measured.
This method is mostly used in educational and psychological evaluation.
For example, trying to determine the motive for a suspect who may have committed a crime. With this technique, you’re measuring the opportunity to commit the crime, the attribute the suspect possesses to commit the crime, and relating them to the underlying concept- motive.
Latent Class Analysis
With this technique, you’re identifying different variables within a data set that exhibit the same observable attributes, then categorize them as a subgroup. These subgroups help determine the underlying variable you want to measure by assessing them into groups.
The most common use cases for latent case analysis are in health, psychology, and social sciences. For example, you want to group people based on their shopping habits (observable habits) into different types of customers(latent classes) for your fashion store.
With this technique you’d probably find out some of your customers are impulsive shoppers, some are window shoppers, some are discount hunters and some are Driven, Shoppers.
You could also find out why your customers are in these groups.
Examples of Latent Variables
Health and Fitness
Medical professionals can’t exactly measure your overall well-being. Instead, what medical professionals do is carry out different physical tests to infer how good or bad your health is.
These tests determine the quality of your health because medical professionals have been trained to use the observed value from these tests to differentiate between healthy and unhealthy patients.
Also, there’s no actual measurement to determine your fitness. But with observable attributes like your weight, the time it takes for you to get tired, physical strength, endurance, and more.
Even in the qualities to measure your fitness, some of them are still latent e.g physical strength and endurance. But you can only infer one’s physical strength by the activities they can participate in and how long they participate in that activity without getting tired and stopping.
Personality
Personality is a very complex variable to measure, but we all measure it in some sort of way or the other. The most common method to measure personality is factor analysis.
For example, if you’re meeting someone for the first time, you’re judging the person’s character by the characters they exhibit. For example, the person is seemingly interested in conversations out their field of expertise, this is curious, it shows that the person’s openness to experience the big personality traits is high.
Of course, this isn’t exactly an accurate way to determine one’s personality, you’d have conducted these tests over time to be sure. They are also studied specially designed to determine personalities avoiding bias.
Intelligence
Most of us have taken some kind of intelligence test, either an IQ or an EQ test. IQ tests analyze people’s problem-solving, critical and logical reasoning, age, and experience to determine their level of intelligence.
For emotional intelligence, gather data on how people react to situations to determine their level of compassion and empathy. Both types of intelligence tests have one thing in common: they don’t have a measurable unit.
Individual ability to perform well on these tests, on the other hand, indicates a high level of emotional or intellectual intelligence.
Effects & Implications of Latent Variables
Establish Relationship Between Unrelated Concepts
The most important implication of using latent variables is that it allows you to infer the value of concepts we can’t directly observe by using concepts we can see. Sometimes the observable data to help us determine latent variables has no relationship between them, but when combined, they help us find patterns or factors that help us measure latent variables.
For example, there is no direct measure to quantify the unemployment rate in a specific region, but we can infer the rate from two separate or unrelated data sets: the number of employed people in the region and the number of unemployed people in that region.
In this case, the relationship between these two separate data is the latent variable. Latent variables don’t just help us measure unobservable traits, they also help us establish connections between concepts to find meaningful concepts.
Drawing Meaningful Conclusion with Data
In most cases, the main reason we need to figure out how to measure a latent construct is that it has a significant impact on something but we can’t physically quantify how much or how much it impacts it.
For example, to determine why one city’s crime rate is higher than others, an experiment would need to determine the motivation behind these crimes.
Following that, in a survey to find out why City A crime rate is way lower than City B which is next to it, respondents answered that they have low motivation to commit a crime.
According to this study, an environment can influence the behavior of its occupants. The next step is to determine what motivates people to commit crimes in City B but not in City A; there are environmental factors that inhibit motivation to commit a crime in City A but may be present in City B.
The factors that would most likely motivate people to commit crime differ from city to city and would most likely include their living conditions, general values, and others.
From the crime rate to the motivation to commit a crime, they are all latent constructs. But by collecting and analyzing information from residents, it will be easier to determine and improve the major factor influencing people to commit crimes in one city or the other.
How Do You Find & Measure the Latent Variables
We can’t physically measure latent variables, but we can infer their value by measuring observable variables. To determine the value of a latent variable, we quantify the observable variable and establish a relationship between it and the latent variable.
Here are the steps to find latent variables in research:
Defining the Reflective Items
A latent construct typically has multiple indicators that infer or reflect its value. These indicators, also known as items, are what is collected as data during a latent construct study.
Also, these indicators must possess some characteristics that qualify them as an effective method of quantifying the latent construct.
Developing and Generating Latent Variables
Developing indicators (latent variables) essentially means creating a foundation for what we want to measure. That is, questions these indicators should answer about the construct we want to quantify.
We can now determine if these items are an effective means of measuring the latent construct we’re studying based on the questions and statements we want to answer and prove.
After the development of the indicators, the next step is to determine the form in that the latent variables would be measured; this is known as item generation.
For example, there is a study being conducted to determine the level of commitment that people over the age of 25 have to their jobs. The latent variable measured in this study is the commitment to work; the items would be questions that reflect the level of commitment employees above 25 years have toward their jobs.
In this case, item generation would entail defining the format of the question; would it be rating scale questions or yes or no questions, the number of questions, etc.
Pretesting and Validity Test
Before proceeding to measure a latent construct, first, verify the effectiveness of the latent variables you’ll be using to measure this concept on a small scale.
This test will determine whether the previous steps in item development and generation are an efficient means of determining the latent construct you want to measure.
After testing and proving that the indicators are effective, you must retest them to ensure that they are a reliable and valid means of measurement.
If the latent variables produce consistent values when measured multiple times, it indicates that it is reliable. If the results of the preliminary testing correlate with the expected results of previous experiments, it is valid.
Sample Selection and Data Collection
Once the latent variables have been determined to be valid methods of measuring the concept, the next step is to choose a sample that fits into the population of interest that the latent variable is to measure.
Following that, conduct your research, record responses or observations, and then compile the data into a platform where it can be analyzed.
Data Analysis
The final step is to analyze the collected data to determine the value of the latent variables. There are several techniques to this, including factor analysis, latent class analysis, and latent trait analysis.
Conclusion
Latent variables are important but unobservable variables that cannot be directly measured. As a result, we infer their values using observable variables by establishing a relationship between these observable variables and the latent variables.
However, there are several steps to validate the effectiveness of using the observable variable before we can use it to infer the value of the latent variables.