Abstract

Self-selection, informative drop-out and variable-dependent sampling schemes are common issues in many epidemiological studies. They all give rise to selection bias, which impedes identification of causal estimands, too. Some authors formalize this problem by adopting a graphical model where the selection mechanism is treated as a node on which conditioning happens. Within this framework, we review some recent contributions and explore the extension to the longitudinal setting, for which some solutions were also proposed.

Based on joint work with: Elena Stanghellini, Chris Caroni, Konstantina Gourgoura and Taiki Tezuka.

References

Doretti M., Geneletti S., and Stanghellini E. (2016) Tackling non-ignorable dropout in the presence of time-varying confounding. Journal of the Royal Statistical Society – Series C.

Doretti M., Genbäck M., and Stanghellini E. (2024) Mediation analysis with case-control sampling: Identification and estimation in the presence of a binary mediator. Biometrical Journal.

Mathur M. B. and Shpitser I. (2024) Simple graphical rules for assessing selection bias in general-population and selected-sample treatment effects. American Journal of Epidemiology.

Stanghellini E., Doretti M., and Tezuka T. (2024). O note on “Simple graphical rules for assessing selection bias in general-population and selected-sample treatment effects” by Mathur and Shpitser. American Journal of Epidemiology.