Learn about DAGs and DAGitty
To understand what DAGitty does, it is fundamental to first understand what DAGs do (and don't do). On this page, I provide some starting points for you to learn more about DAGs. The functions of DAGitty itself are described in the manual.
However, the manual provides only very little introduction to DAGs. Therefore you may be interested in these additional resources.
Interactive Tutorials and Examples
This is meant to become a list of DAGs aspects of that are illustrated interactively using DAGitty itself. Currently this list is still very short, but I hope it will grow more in the future.
- Parents, children, ancestors, descendants ... If you are confused by all the graph terminology, here's our little game to learn about it!
- What is a confounder? A mediator? A proxy confounder? These are all concepts that can only be defined by invoking causal language. Learn in our tutorial on covariate roles how to spot such variables in s DAG.
- "Table II Fallacy" is a nice idiom introduced by Westreich and Greenland. It refers to the problems with interpreting the coefficients in a multiple regression model causally, including the widespread belief that coefficients in such models are "mutually adjusted". Learn here how to properly interpret coefficients in multiple regression models!
- The article d-Separation Without Tears, taken and adapted from Judea Pearl's textbook "Causality", explains the key concept of d-separation in simpler words.
- The Simpson Machine illustrates which causal structures can lead to Simpson's paradox, and that valid covariate adjustment sets cannot be found without causal considerations.
- The Single-Door Criterion can be used to identify structural parameters (direct effects) in structural equation models, even when the model as a whole is not identifiable. This short tutorial explains what the single-door criterion can, and cannot, do.
- "Graphical Causal Models" by Felix Elwert is the most accessibly written general introduction to DAGs that I know of.
- "Reducing bias through Directed Acyclic Graphs" by Ian Shrier and Robert W. Platt is a nicely written piece on the specific issue of covariate adjustment.
- Epidemiologists may also like "Causal Diagrams for Epidemiologic Research" by Greenland, Pearl, and Robins.
- And then there's Judea Pearl's textbook "Causality", which is certainly comprehensive and thought-provoking, but not very accessible.