ABSTRACT
Dynamic factor analysis (DFA) is a statistical technique used to identify lagged structure in covariance matrices. DFA was developed to overcome one of the main limitations of Cattell’s P-technique factor analysis (Cattell, 1963; Cattell, Cattell, & Rhymer, 1947), which did not include the specification of any lagged relations. P-technique factor analysis, formulated by Cattell to examine within-person fluctuations in personality traits, consists of the factor analysis of multivariate time series data comprised of scores on multiple manifest variables measured on a single individual across multiple time points. Correlations or covariances are then computed among the manifest variables, so the correlations represent the degree to which manifest variables covary across time. When subjected to factor analysis, the resulting factor structure provides information about the latent dimensions that underlie intraindividual variability on the measured variables, but does not reflect any time-related dependencies among manifest or latent variables. The different techniques that are subsumed under the label of DFA were developed precisely to account for time-related dependencies among manifest and latent variables, as these time-related dependencies can be revealing of mechanisms associated with dynamic processes that underlie causal relations among variables across time.
