Steffen Wagner

In empirical quantitative research, purely associative models optimize empirical fit on observational data, yet they remain highly vulnerable to spurious correlations and systemic failures under structural breaks. This research project addresses these limitations by systematically identifying statistical signatures in passive data to reconstruct the underlying data-generating process as a Directed Acyclic Graph (DAG). Such causal graphs are essential because they define the structural network required to run unbiased interventional simulations, trace root causes, and guarantee predictive reliability under environmental shifts.
This talk reviews the mathematical foundations of causality and the core algorithmic classes of structural learning, including constraint-based, score-based, and functional frameworks. Special emphasis is placed on the integration of system-specific boundary conditions, analyzing how temporal lag constraints in longitudinal data are utilized to drastically reduce the combinatorial search space.
To demonstrate the utility and limits of these techniques, we outline the structural setups of two distinct research collaborations, transitioning from numerical time series data from environmental monitoring to categorical health-tech telemetry data. Since the project is currently in its initial phase, the session focuses primarily on the theoretical methodology, though preliminary findings will be presented for discussion depending progress.