In traditional statistics, we often look for the "average" effect. We ask: Does a new teaching method improve test scores? Does a specific treatment reduce symptoms? While these questions are useful, they assume that the population is a monolithic block. They ignore the possibility that a treatment might work wonders for one group of people while doing nothing—or even causing harm—to another.
Researchers across medicine, psychology, and sociology are increasingly moving away from this "variable-centered" approach toward a person-centered framework. By using Latent Class Analysis (LCA) and Latent Profile Analysis (LPA), they are uncovering hidden subgroups within populations that share unique patterns of behavior, symptoms, or beliefs.
The Fallacy of the Average
Imagine evaluating the quality of life for seniors in nursing homes. A standard analysis might give you a single average score. However, as Chunqin Liu and colleagues demonstrate in Quality of life profiles and its association with predictors amongst Chinese older adults in nursing homes: a latent profile analysis, that average hides the truth. By applying LPA, they found distinct "profiles" of residents: some who thrive due to high psychological resilience and others who struggle despite having similar physical health.
The core intuition of LCA/LPA is that patterns matter more than individual variables. Instead of correlating "Variable A" with "Variable B" across the whole sample, these models cluster individuals who exhibit similar configurations of multiple variables. This allows researchers like Dennis Fok, Richard Paap, and Philip Hans Franses to identify unobserved heterogeneity—statistically grouping people who respond to external "shocks" (like price cuts or medical interventions) in the same way, as seen in Modeling dynamic effects of promotion on interpurchase times.
Identifying Hidden Phenotypes
In clinical settings, this approach is revolutionary for "precision medicine." Rather than treating a disease as a single entity, LCA helps identify "phenotypes"—subgroups with different biological or clinical trajectories.
For instance, Mingqiang Wang and Yue Jin used LCA in Identifying subgroup of severe community-acquired pneumonia based on clinical metagenomics to discover two distinct classes of pneumonia patients. They found that corticosteroids significantly reduced mortality in one group but were ineffective in the other. Without LCA, the benefit for the first group would have been "washed out" by the lack of response in the second, leading to a misleading "average" result that suggests the treatment doesn't work.
Similarly, in Dissecting the phenotypic heterogeneity in sensory features in autism spectrum disorder, Julian Tillmann and Mirko Uljarević used Factor Mixture Modelling—a hybrid of LCA and factor analysis—to show that sensory challenges in autism aren't just a matter of "more or less" severity, but fall into distinct qualitative categories that correlate with different social behaviors.
Mapping Mindsets and Archetypes
Beyond medicine, these techniques are used to map the "internal landscape" of human beliefs and behaviors.
- Education: Ehsan Latif and Xiaoming Zhai use LCA to categorize how students interact with AI in Conversational AI as a Catalyst for Informal Learning, identifying "learner archetypes" that range from deep explorers to superficial fact-checkers.
- Sociology: Yang Hu applied LCA in Attitudes toward transnational intermarriage in China to reveal that "cosmopolitan" attitudes aren't a single spectrum; instead, people cluster into multidimensional typologies based on their unique mix of economic and cultural values.
- Public Health: During the pandemic, Zhimin Niu and Songli Mei used LPA in Latent profile and network analysis of risk perception to track how different subgroups of students processed risk over time, finding that some groups developed "psychological immunization" while others remained highly anxious.
Why It Works: The Power of Probabilistic Membership
Unlike traditional "hard" clustering (like K-means), LCA is a probabilistic model. It doesn't just say "You are in Group A." It calculates the probability that an individual belongs to a specific latent class.
This nuance is critical for researchers like Dominik Hangartner and Luke Keele. In Profiling Compliers in Instrumental Variables Designs, they use these principles to describe the traits of "compliers"—a latent group of people who only take a treatment because they were encouraged to do so. By "profiling" this hidden group, researchers can determine if their findings are actually applicable to the general public.
By shifting the focus from variables to people, Latent Class and Profile Analysis provide a more honest map of human diversity. They allow us to stop designing for the "average person" and start designing for the real, distinct subgroups that actually exist.