Reflection on modern methods: Understanding bias and data analytical strategies through DAG-based data simulations

Abstract

Directed acyclic graphs (DAGs) are increasingly used in epidemiology to identify and address different types of bias. The present work aims to demonstrate how DAG-based data simulation can be used to understand bias and compare data analytical strategies in an educational context. Examples based on classical confounding situations and an M-DAG are examined and used to introduce basic concepts and demonstrate some important features of regression analysis, as well as the harmful effect of adjusting for a collider variable. Other potential uses of DAG-based data simulation include systematic comparisons of data analytical strategies or the evaluation of the role of uncertainties in a hypothesized DAG structure, including other types of bias such as information bias. DAG-based data simulations, like those presented here, should facilitate the exploration of several key epidemiological concepts, DAG theory and data analysis. Some suggestions are also made on how to further expand the ideas from this study.