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

Abstract

Background Psychopathology research is changing focus from group-based “disease models” to a personalized approach inspired by complex systems theories. This approach, which has already produced novel and valuable insights into the complex nature of psychopathology, often relies on repeated self-ratings of individual patients. So far, it has been unknown whether such self-ratings, the presumed observables of the individual patient as a complex system, actually display complex dynamics. We examine this basic assumption of a complex systems approach to psychopathology by testing repeated self-ratings for three markers of complexity: memory, the presence of (time-varying) short- and long-range temporal correlations; regime shifts, transitions between different dynamic regimes; and sensitive dependence on initial conditions, also known as the “butterfly effect,” the divergence of initially similar trajectories. Methods We analyzed repeated self-ratings (1476 time points) from a single patient for the three markers of complexity using Bartels rank test, (partial) autocorrelation functions, time-varying autoregression, a non-stationarity test, change point analysis, and the Sugihara-May algorithm. Results Self-ratings concerning psychological states (e.g., the item “I feel down”) exhibited all complexity markers: time-varying short- and long-term memory, multiple regime shifts, and sensitive dependence on initial conditions. Unexpectedly, self-ratings concerning physical sensations (e.g., the item “I am hungry”) exhibited less complex dynamics and their behavior was more similar to random variables. Conclusions Psychological self-ratings display complex dynamics. The presence of complexity in repeated self-ratings means that we have to acknowledge that (1) repeated self-ratings yield a complex pattern of data and not a set of (nearly) independent data points, (2) humans are “moving targets” whose self-ratings display non-stationary change processes including regime shifts, and (3) long-term prediction of individual trajectories may be fundamentally impossible. These findings point to a limitation of popular statistical time series models whose assumptions are violated by the presence of these complexity markers. We conclude that a complex systems approach to mental health should appreciate complexity as a fundamental aspect of psychopathology research by adopting the models and methods of complexity science. Promising first steps in this direction, such as research on real-time process monitoring, short-term prediction, and just-in-time interventions, are discussed.

Studying Behaviour Change Mechanisms under Complexity

“…By examining intensive longitudinal data, we have shown similar results to a plethora of studies, which find complex dynamics in ecological momentary assessment data [ 72 , 115 , 122 , 138 ]: Non-linear, non-ergodic, non-independent dynamics, which defy traditional assumptions. Moreover, the empirical case example of a single person shows the importance of observing change over long periods of time: There are several recurring “ways of being”, patterns of experience which change dynamically within the individual.…”

“…This, too, implies that interaction-dominant causation is a more plausible framework for the behaviour change domain, wherein effects emerge (and are conditional upon) the system’s holistic multivariate dynamics, with everything potentially taking place simultaneously in a circularly causal manner. Interaction-dominant dynamics are also characterised by heavy-tailed distributions [ 49 , 66 , 76 ] such as the log-normal distribution [ 77 ], which are common in psychological data [ 78 , 79 ], as well as the presence of long-range temporal correlations and power-law scaling [ 72 , 80 , 81 ]. Importantly, interplay happens not just between variables, but also between their temporal dynamics: Processes taking place on fast timescales (e.g., lack of physical activity) modulate slow-timescale processes (e.g., development of obesity, lower energy levels), which then feed back, affecting fast-timescale processes [ 41 ].…”

“…Indeed, exact prediction beyond some short-term horizon is convincingly shown to be impossible in complex and not-so-complex systems across sciences, as indicated since Poincare’s discovery of the famous three-body problem [ 110 ]. For a discussion on evaluating the aforementioned prediction time horizon in psychological self-ratings, see [ 72 ].…”

“…But even time-varying autoregressive models operate under the linear regression framework, with its accompanying assumptions, such as normally distributed errors. Furthermore, in Figure 2 , we have limited ourselves to investigating the lag-1 relationships, whereas long-range dependencies are common in ecological momentary assessment data [ 72 , 92 , 115 , 122 ].…”

Too many zeros and/or highly skewed? A tutorial on modelling health behaviour as count data with Poisson and negative binomial regression

“…However, while incorporating count distributions into these methods represents some improvement, some recent work suggests that these do not adequately model EMA/ESM data (Hasselman & Bosman, 2020 ). Repeated observations over time within an individual are clearly not independent, but traditional linear methods tend to model influence only from the previous observation (a lag of 1 in autoregression), whereas human memory can provide influences from weeks or months ago (Hasselman & Bosman, 2020 ; Olthof, Hasselman, & Lichtwarck-Aschoff, 2020 ). Similarly, for behaviour change, we would expect to see potentially marked abrupt change, change in variability, and other phenomena not modelled within linear methods.…”