Using directed acyclic graphs to determine if multiple imputation can be valid
Providing practical tools for assessing whether the ‘missing at random’ assumption is plausible for multiple imputation in a full sample or subsample.
Missing data are a common challenge in epidemiology and health research. The missing at random (MAR) assumption is required for unbiased estimation of all analysis model parameters when using multiple imputation (MI) to account for missing data. Establishing whether this assumption is plausible can be difficult in practice, especially when several variables are incomplete. In this talk, Dr Paul Madley-Dowd will present a framework that uses directed acyclic graphs (DAGs) to help researchers assess whether the MAR assumption is plausible in the presence of multiple incomplete variables.
When the MAR assumption is not plausible, it may still be possible to unbiasedly estimate exposure-outcome effects using an alternative approach called subsample-MI. Paul will explain the rationale behind subsample-MI and present an extension of our DAG-based framework that helps researchers assess the appropriateness of subsample-MI.
Drawing on examples from methodological and applied epidemiology research, Paul will demonstrate how the approach can be used to inform analysis decisions and improve transparency around missing data assumptions. The session is aimed at epidemiologists and health researchers who routinely analyse incomplete data.
Speaker
- Dr Paul Madley-Dowd - Research Fellow, University of Bristol
Event notices
- Please note that you can join this event in person or you can join the session remotely
- Please note that the recording link will be listed on this page when available
Admission
Contact