The general scientific approach, using complex systems models and methods, in psychology and the social sciences has been worked out in an open access book, published in 2024. The cascading transition modeling approach is worked out in detail.
The applied projects focused on the construction of cascading models for addiction, learning and opinions. For addiction, we have developed a formal model that integrates dual-process decision-making with social dynamics. The individual-level model is based on a well-known population dynamics framework from biology, where sudden outbreaks (analogous to high consumption in our case) can occur. The parameters of this model are influenced by consumption patterns within an individual's local social network, while changes in the network are driven by consumption behavior itself. This unique integration of individual and social dynamics enables us to explain empirical phenomena at both the individual and group levels. Examples include the segregation of user groups based on whether substances are illicit and the interaction between smoking and vaping. We hope to publish this work this year. In this sub-project we also analyze data on online gambling to gather evidence for the individual model in this work.
For opinion change, we already have a first model available that has been published. Attitudes or opinions are conceptualized as networks of beliefs, feelings and behaviors towards the attitude object. If these networks are balanced—primarily composed of positive connections—their behavior can be described by the cusp, a standard model of transitions. By constructing networks of cusps, we can model societal polarization. However, the assumption that attitude networks are inherently balanced is not self-evident. In this project, we developed LIAM (Learning Ising Attitude Model), which introduces learning into the network. Our findings show that even a basic form of learning, Hebbian learning, naturally leads to balanced networks. This not only supports our cascading transition modeling approach but also provides insight into various attitude-related phenomena, particularly how individuals resolve cognitive dissonance.
In this project we also tested some key predictions of the model. According to the model higher involvement leads to more polarization. We analyze involvement in five large-scale surveys on political orientation, EU attitudes, and COVID-19 vaccines across 79 countries over eight years. Using a new modality detection measure for ordinal data, we classify ideological divergence. Findings show involvement correlates with attitude extremity, supporting our model in COVID-19 vaccine data, but political and EU attitudes show no effect or contradict our predictions.
For learning, we developed a reciprocal model of practice and performance, where practice enhances performance, and both higher performance and improvements in performance accelerate the rate of practice. Between practice sessions, forgetting plays a role, decreasing performance. For the sub-processes of this model, we make use of standard models, such as the laws of practice and forgetting. It is shown that sigmoid learning curves lead to a bifurcation in skill development, resulting in either a positive feedback loop that fosters expertise or a negative one that leads to dropout. Based on this model we propose a number of intervention strategies. This model can be simplified to one differential equation allowing us is as model of one agent within an agent-based model. This will be a key objective in the coming years. Using data from online games and learning systems, we aim to test the assumptions and predictions of the model.
