Capturing fluctuations in multivariate intensive longitudinal data

Abstract

While the popularity of intensive longitudinal designs continues to grow and the technology to collect intensive data advances, the methods to analyze this type of data remain lagging behind. Traditional analyses for “non intensive” longitudinal data may not properly capture complex variable interrelationships, are not equipped to handle the high dimensionality of these relationships, and are unable to capture the rates at which these relationships may change. This paper introduces a novel method for analyzing intensive longitudinal data that incorporates dimension reduction techniques and time series analyses. The method is a three-phase, “bottom up” approach where the data is first analyzed person to person. In the first phase, distance parameters are determined for each individual. In the second phase, optimal distances between the variables are computed for each participant across all time points. Lastly, a one-dimensional solution is computed across all time points for each participant. A first order autoregressive model was fit to each individual's solution vector to examine intra individual dynamics and allow for comparisons of interindividual trajectories. Thus, the method constructs a one-dimensional parameterized representation of the longitudinal data at each time point while preserving the structure of the relationships between variables over time. While this approach is novel in its application to intensive longitudinal data, the proposed method is a promising alternative for analyzing intra- and inter-individual differences over time.