Intraindividual Variability and Short-Term Change

Modern developmental science posits individuals as part of a large, multifaceted dynamic system. Yet, traditional statistical methods are not in accord with this conceptualization of individual behavior. We are interested in developing methods apt for developmental science that consider the dynamic nature of human behavior. A key opportunity for understanding processes involves the increased interest and availability of intense longitudinal data (i.e., time-series data) of psychological phenomena. These data allow us to investigate the dynamic structure of processes and the dynamic interplay among variables. Recently, we examined the benefits of modeling intra-individual variability with individual-level models for predicting distal outcomes (Castro-Schilo & Ferrer, 2013). Current work aims to extend latent variable approaches for analyzing intra-individual variability and allow researchers to answer questions related to multivariate dynamics.


Multivariate Dynamics and Long-Term Change

Often times, longitudinal investigations are carried out by collecting data across largely spaced time intervals (i.e., panel data). Under certain assumptions, these data are also informative for understanding within-person dynamics. In this domain, we have introduced modeling techniques for characterizing and predicting linear and non-linear change (e.g., Grimm, Castro-Schilo, & Davoudzadeh, 2013). These techniques capitalize on the explicit specification of latent change scores. Indeed, when researchers are interested in studying change, explicit representation of change in statistical models has clear advantages. One key benefit is the possible inclusion of change as a predictor, rather than only an outcome. In several projects, we adapted latent change score models to take advantage of the latter feature (e.g., Henk & Castro-Schilo, in press). For example, in collaboration with colleagues, we looked at changes in attachment to school of Mexican-origin children as predictors of a series of important developmental outcomes (Castro-Schilo, Ferrer, Hernández, & Conger, conditionally accepted). We found that changes in school attachment that occur during the transition to middle school are central to future peer competence, attachment to teachers, schooling aspirations, and intentions to engage in substance use. Similarly, we applied this modeling technique to investigate effects of changes in positive affectivity in later cognitive decline among elder Mexican-Americans (Castro-Schilo, Fredrickson, & Mungas, in preparation). Results from this study suggest a positive dynamic relation between positive affect and cognition at old age.

Current work is focused on facilitating exploration of multivariate dynamics in the latent change score framework (Henk & Castro-Schilo, in preparation). The main innovation of this research consists on providing applied researchers with a modeling approach that can directly test hypotheses in which change (free of measurement error) is both a predictor and an outcome, and where multiple processes are considered simultaneously.


Psychometric Issues: Multitrait-Multimethod Models
To accurately capture short-term or long-term change, we must ensure appropriate measurement of our constructs. When multiple traits -such as personality characteristics, behaviors, or abilities- are measured with multiple methods -such as different measurement procedures or different informants,- Multitrait-Multimethod (MTMM) models can be used to examine construct validity. We use these models to gain insight into the quality of our data and to account for unwanted biases, particularly when data are self-reported. In recent work, we showed the benefits of collecting MTMM data and the biases that arise from existing MTMM models when examining associations of traits with external variables (Castro-Schilo, Widaman, & Grimm, 2013; 2014). Currently, we are testing alternative MTMM models that deal with the limitations of their predecessors. We are also working on the synergy between MTMM models and longitudinal investigations.