Latentvariabler

Latentvariabler are variables that cannot be directly observed or measured, yet are inferred from observed data. They represent underlying constructs, processes, or states that give rise to the measurements. By modeling latent variables, researchers can capture structure in the data that is not immediately visible from the observed variables alone.

Latent variables are central to many statistical and machine learning approaches. In psychology, latent traits such as ability or personality are inferred from test scores. In statistics, latent factors are estimated in factor analysis to explain correlations among observed measures. In natural language processing, topics inferred by models such as latent Dirichlet allocation are latent variables representing themes in documents. In genomics and other fields, hidden sources of variation are often treated as latent variables to account for confounding or to discover meaningful structure.

A latent-variable model specifies probabilistic relationships between latent variables and observed variables. Latent variables can be discrete, as in hidden Markov models, or continuous, as in factor analysis or certain Bayesian models. Estimation typically relies on maximum likelihood or Bayesian inference, using algorithms such as the expectation-maximization (EM) algorithm, Markov chain Monte Carlo, or variational inference.

Identifiability and interpretation are important considerations. Latent variables are not always uniquely determined; their scale, orientation, or labeling may depend on model constraints. Consequently, interpretation should be guided by theory and validation.

Applications span psychology, education, marketing, biology, and NLP. While powerful, latent-variable models rely on assumptions about the data-generating process and can be sensitive to model misspecification and data quality.