Under pressure Studying complex and causal systems in psychopathology

In this dissertation, I focused on two questions, (1) can we, and if so how, assess to what extent a complex dynamical system is in such a state that it can transition between two stable states, and (2) how well are we able to estimate a causal graph when we combine observational and experimental data. Chapter 2 provided an overview on various techniques that can be used to estimate network structures. Two models, the Cramer model and the Empirical Mean Field Approximation, were described and illustrated using empirical data. In chapter 3, we theoretically showed that it is possible to reduce a multidimensional dynamical system to a single equation, which in turn may be used to estimate a system’s dynamical properties. In other words, the mean field model that is introduced here can be used to infer whether or not a system is in a space where two stable states exist, or in a space where only one stable state exists. chapter 4 expanded on this work and combined the mean field model with maximum likelihood estimation to estimate the parameter of interest in the mean field model. With this parameter, we could then assess the expectancy of an individual to transition between two stable states.
The second part of this dissertation focused on causality. Chapter 5 studied different algorithms to estimate causal graphs. Here, we argued that using observational data alone will not give the entire causal picture. By combining observational data with experimental data, it is possible to detect meaningful causal relations that would otherwise stay undetected. Not only did we show the advantage of the combination of observational and experimental data in a simulation study, chapter 6 showed that this approach also results in interesting and meaningful causal relations when empirical data are used.