Recovering Bistable Systems from Psychological Time Series

Abstract

Conceptualizing mental disorders as complex dynamical systems has become a popular framework to study mental disorders. Especially bistable dynamical systems have received much attention, because their properties map well onto many characteristics of mental disorders. While these models were so far mostly used as stylized toy models, the recent surge in psychological time series data promises the ability to recover such models from data. In this paper we investigate how well popular (e.g., the Vector Autoregressive model) and more advanced (e.g., differential equation estimation) data analytic tools are suited to recover bistable dynamical systems from time series. Using a simulated high-frequency time series (measurement every six seconds) as an ideal case we show that while it is possible to recover global dynamics (e.g., position of fixed points, transition probabilities) it is difficult to recover the microdynamics (i.e., moment to moment interactions) of a bistable system. Repeating all analyses with a sampling frequency typical for Experience Sampling Method studies (measurement every 90 minutes) showed that the recovery of the global dynamics was still successful, but no microdynamics could be recovered. These results raise two fundamental issues involved in studying mental disorders from a complex systems perspective: first, it is generally unclear what to conclude from a statistical model about an underlying complex systems model; and second, if the sampling frequency is too low, it is impossible to recover microdynamics. In response to these results we propose a new modeling strategy based on substantively plausible dynamical systems models.

Bridging the gap between complexity science and clinical practice by formalizing idiographic theories: a computational model of functional analysis

“…For practical reasons, variables are often measured within the same time-scale (usually once a day or every few hours), potentially leading to biased estimates in dynamical models. A recent simulation study suggests that using the most commonly applied ESM time-intervals results in data models that are largely unable to recover the micro dynamics of a system [49]. A stronger focus on theory and the utilization of clinical knowledge could therefore be helpful in informing relationships in the estimated model that cannot reasonably be captured by commonly used ESM data.…”

Modeling Psychopathology: From Data Models to Formal Theories

“…Data is simulated from the Panic Model, the full specification of which is given by (D. J. Robinaugh, Haslbeck, et al, 2019), using the statistical programming language R. We use the Panic Model to generate time-series data of 1000 individuals, on a single minute time scale, for 12 weeks, using Euler's method with a step size of .001. This yields a total of n t = 12, 0960 repeated measurements per person.…”

Complexity in psychological self-ratings: implications for research and practice

“…Two variables that have a stable causal connection in a complex system may exhibit non-stationary correlation with each other, even to the point that the correlation coefficient switch signs [100]. In a simulation study on psychological dynamics, Haslbeck and Ryan [101] found that a VAR model cannot retrieve the complex systems model that generated the data. Sugihara et al [100] propose convergent cross mapping as a possible alternative for Granger causality.…”