Despite increasing data collection facilities, selection bias is a widespread issue in most epidemiological research. This arises when either a proper sampling design cannot be implemented or, when possible, units fail to comply. When participation or compliance is linked to the phenomena under investigation, conclusions based on the available information are necessarily distorted.

Two different situations will be addressed. The first when no information is available on units not included in the sample. The second when some information is available, either at aggregate or individual level (this second related to missing data).

Some recent contributions are presented that combine results from stratified case-control design and secondary outcome analysis. Although the issue arises not only in causal inference, we will mainly focus on causal effects and make use of causal graphs where a binary indicator of selection is included as a node.  

Elena Stanghellini is Professor of Statistics at the University of Perugia. After completing her Ph.D. in 1995 at the University of Florence, she was Research Fellow at the Open University, to then join the University of Perugia in 1997. In 1999 she was Jemolo Fellow at the University of Oxford. Currently she is Affiliated Guest Professor at Umeå University. Her research focuses on Graphical Markov Models, both from the theoretical and applied view point. She has been working on identification of Graphical Markov Models with unobserved nodes, capture-recapture methods, causal inference and distortion induced by informative selection or drop-out. In 2020, she was Member of the Advisory Board of National Bureau of Statistics (ISTAT) for the design and implementation of sample surveys to measure and monitor epidemiological parameters of COVID-19 pandemic.

Contact Christiana Kartsonaki for RSS Oxford Local Group and RSS Medical Section