You are here:
Using Directed Acyclic Graphs to Support Causal Interpretations from Data Analysis for Environmental Assessments
Citation:
Carriger, John F. Using Directed Acyclic Graphs to Support Causal Interpretations from Data Analysis for Environmental Assessments. SETAC North America 42nd Annual Meeting: SETAC SciCon4, N/A, N/A, November 14 - 18, 2021.
Impact/Purpose:
This presentation will provide an introductory exploration of directed acyclic graphs (DAGs) and their utility for causal interpretations with observational data. The DAGs are the foundation of Bayesian networks but can be used beyond Bayesian network analysis to examine the possibilities for causal interpretations and to detect biases in causal interpretations from sampling and analysis. The presentation will discuss key data analytical concepts like confounding, overcontrol, and selection bias and why DAGs are suited for their identification in the context of environmental assessment.
Description:
Directed acyclic graphs (DAGs) can represent the causal structure of a problem by relating variables through arrows in a cause-effect fashion. When using DAGs in a Bayesian network, conditional probabilities are established for the relationships between the variables. However, these probabilities are often less stable for different contexts than the qualitative relationships in the DAG structure itself. The stability of an appropriately constructed DAG makes it very useful for understanding causal relationships behind the data used for prediction and diagnosis. Recent research has explored how DAGs can be used to better support causal understanding across analytical tasks. When developing statistical models, the strength of correlations are often not directly interpretable in a causal sense. Still, they are regularly interpreted in such a manner. Known biases in causal interpretations can occur from not understanding the causal structure behind the data including unidentified causal associations with the predictors and between the predictors and unmeasured variables. The DAGs provide a platform for making any causal inferences from data analysis transparent. Understanding the data generating process with DAGs can also help to gain insights and prevent erroneous causal interpretations. This presentation will explore how data analysis for environmental assessments can be assisted with the use of DAGs and the limitations. Stylized examples of relevance for environmental assessment will be provided along with classical cases of how DAGs have been used to recognize the causal relationships behind associations. The application of DAGs to analysis of data can enhance causal interpretations and prevent the conflation of influential factors when the structure of the data generating process is identifiable.