Factor analysis for time series data
G. David Garson · 2022
Factor analysis for time series data is covered in this chapter on functional PCA (FPCA), which aids in understanding a time series, such as wages over time. The worked example focuses on wages as the variable of interest. In functional PCA, the time series for the focal variable is decomposed into multiple functions (curves) which explain variation over time. The computed set of functional components are the “modes of variation”. For a single time series such as wages, there will be multiple modes of variation. The first FPCA component is the mode which explains the most variation, with later components explaining successively less variation. Longitudinal PCA is implemented using the FPCA() function of the “fdapace” package. Also called “principal component analysis through conditional expectation” (PACE), this is a nonparametric procedure. PACE is a flexible alternative to random effects modeling of longitudinal data. FPCA scores also may be used for outlier detection.