Project
Non-Linear Intensive Longitudinal Methods

Researchers are increasingly aware that studying psychological processes as they dynamically unfold over time and within people can yield valuable insights into the nature of these processes and give rise to effective treatments and interventions. This has led to a sharp increase in experience sampling methods and intensive longitudinal data (ILD). While ILD is increasingly conceptualized to come from complex non-linear systems (Gelfand & Engelhart, 2012; Nesselroade & Ram, 2004), researchers are currently ill-equipped to analyze non-linear processes meaningfully. These encompass a wide array of phenomena, including non-linear developmental trajectories (e.g. the development of intelligence; Savi et al., 2019), treatment effects (e.g., impacts on cancer patients), multistability and catastrophic regime switches (e.g., emotional fluctuations; Haslbeck & Ryan, 2022), and intricate measurement models (pertaining to the relationship between observed indicators and the latent variables they represent).
Diverse methods capable of handling non-linearity do exist, spanning from data-driven exploratory tools such as kernel smoothing, regression splines (Tsay & Chen, 2019), and Gaussian processes (Rasmussen & Williams, 2006), to more theoretical frameworks like state space modeling (Durbin & Koopman, 2012) and latent change scores (Cáncer et al., 2021). Despite their existence, research that extensively examines and compares these methods in the context of ILD and measurement models is scarce. Since many of these methods can capture several types of non-linearity, rely on different sets of assumptions, and allow researchers to draw distinct kinds of inference, it is oftentimes unclear which method is most suited in each situation. This complexity is further compounded by the potential for combining multiple methods to supplement theoretical models with data-driven techniques (Álvarez et al., 2009; Aue et al., 2015).
Our project aims to delve into non-linear intensive longitudinal methods, tailored to faithfully model the processes commonly encountered in the social sciences. To this end, we will review and evaluate current non-linear time series methods, adapt them to ILD, and combine non-linear models for structural relations and measurement in a new methodology. This endeavor is crucial to mitigate the inherent bias that is likely to exist in current ILD results stemming from the application of linear methods to non-linear data. Additionally, the insights these methods offer into the nature of underlying processes have the potential to impact theory development across various domains.

Supervisors
Dr. ir. J. Mulder
Dr. L.V.D.E. Vogelsmeier
Dr. J. Jongerling

Period
1 September 2023 – 1 September 2027