dagitty

dagitty constructs and analyzes directed acyclic graphs (DAGs) or causal Bayesian networks to support causal inference and the identification of covariate adjustment sets from observational data.

Key Features:

• Design and analysis of causal diagrams: Constructs and analyzes DAGs/causal Bayesian networks to identify strategies for mitigating confounding bias.
• Validation of assumptions: Tests whether assumptions encoded in a DAG are consistent with observed data.
• Error detection and model improvement: Evaluates statistical implications of DAG assumptions to detect model specification errors and inform model refinement.
• Consistency testing: Performs consistency checks between DAG-encoded assumptions and dataset implications.
• Enumeration of statistically equivalent DAGs: Enumerates statistically equivalent but causally different DAGs to reveal causal ambiguities and alternative specifications.
• Adjustment set identification: Identifies valid exposure–outcome adjustment sets that remain valid across statistically equivalent DAGs.
• Data-type testing and non-linearities: Tests DAGs against categorical, continuous, or combined data types, including non-linear relationships.
• R package integration: Provides an R package that integrates with the R platform for statistical computing and exposes functions for the above analyses.

Scientific Applications:

• Epidemiology: Determines covariate adjustment sets and assesses confounding structures for causal inference in observational epidemiologic studies.
• General causal inference: Represents and analyzes causal relationships across diverse scientific disciplines using DAG-based methods.

Methodology:

Consistency testing of DAG assumptions against data; enumeration of statistically equivalent DAGs; evaluation of statistical implications of DAG assumptions; identification of valid exposure–outcome adjustment sets; testing DAGs with categorical, continuous, or combined data including non-linearities.