Seminar in applied mathematics and statistics

SPEAKER:  Anne Helby Petersen (Biostatistics, University of Copenhagen)

TITLE: Bridging the gap: How to make empirical researchers discover causal discovery?

ABSTRACT: Statistical methods for causal discovery has been available for a long time, but their practical use within the health sciences is almost non-existent. This is not due to lack of relevance: Epidemiologists have long been devoted to estimating causal effects from observational data, and to this end, causal discovery methods could in principle be very useful. Nonetheless, these methods have not been adopted by empirical researchers. We intend to bridge this gap between theory and practice by proposing modifications to existing methods for causal discovery that may help make them be part of an epidemiologist's statistical toolbox. We propose a temporal extension of the PC algorithm for causal discovery which, together with an accompanying R package, makes it easy for empirical researchers to make use of their prior knowledge about temporal relationships among variables in order to infer causal models for their observational data. Moreover, we propose a shift towards a more qualitative interpretation of the causal models resulting from the algorithms, which we believe will make them more useful to empirical research. We showcase this empirical use of causal discovery in an application concerning life course development of depression in a cohort of Danish men.

This is joint work with Merete Osler and Claus Ekstrøm.


Upcoming events (after September 9):