Practical applicability of missingness DAGs in complex longitudinal studies

Ascertain the feasibility of collecting study-specific knowledge on missing data mechanisms, integrating this information into realistic missingness directed acyclic graphs (m-DAGs), and establishing identification and recoverability results in complex longitudinal settings; evaluate whether m-DAGs can be applied effectively beyond theoretical contexts.

Background

While causal missingness graphs have advanced theoretical understanding, the authors emphasize the need to demonstrate their real-world applicability in complex longitudinal biomedical studies. This includes eliciting reasons for missingness from domain experts, encoding them in m-DAGs, and then deriving identification and recoverability results.

The paper presents a longitudinal case study to probe this practicality but explicitly notes uncertainty about how well such knowledge can be integrated and how difficult the derivation of identification and recoverability results is in realistic settings.

References

It is unclear how well knowledge on missing data can actually be collected, then integrated in a realistic causal graph, and how difficult the mathematical exercise of establishing identification and recoverability results in such a complex, yet realistic setting is. Can m-DAGs make their way from blackboards to actual applications?

— Recoverability of Causal Effects under Presence of Missing Data: a Longitudinal Case Study  (2402.14562 - Holovchak et al., 2024) in Section 1 (Introduction)