Latent Variable

A latent variable is a variable that cannot be observed. The presence of latent variables, however, can be detected by their effects on variables that are observable. Most constructs in research are latent variables. Consider the psychological construct of anxiety, for example. Any single observable measure of anxiety, whether it is a self-report measure or an observational scale, cannot [Page 697]provide a pure measure of anxiety. Observable variables are affected by measurement error. Measurement error refers to the fact that scores often will not be identical if the same measure is given on two occasions or if equivalent forms of the measure are given on a single occasion. In addition, most observable variables are affected by method variance, with the results obtained using a method such as self-report often differing from the results obtained using a different method such as an observational rating scale. Latent variable methodologies provide a means of extracting a relatively pure measure of a construct from observed variables, one that is uncontaminated by measurement error and method variance. The basic idea is to capture the common or shared variance among multiple observable variables or indicators of a construct. Because measurement error is by definition unique variance, it is not captured in the latent variable. Technically, this is true only when the observed indicators are (a) obtained in different measurement occasions, (b) have different content, and (c) have different raters if subjective scoring is involved. Otherwise, they will share a source of measurement error that can be captured by a latent variable. When the observed indicators represent multiple methods, the latent variables also can be measured relatively free of method variance. This entry discusses two types of methods for obtaining latent variables: exploratory and confirmatory. In addition, this entry explores the use of latent variables in future research.