A new publication became available, again an ‘educational’. However, this time the topic is new. It is about the application of directed acyclic graphs, a technique widely used in different areas of science. Ranging from computer science, mathematics, psychology, economics and epidemiology, this specific type of graphs has shown to be useful to describe the underlying causal structure of mechanisms of interest. This comes in very handy, since it can help to determine the sources of confounding for a specific epidemiological research question.
But, isn’t that what epidemiologist do all the time? What is new about these graphs, except for the fancy concepts as colliders, edges, and backdoor paths? Well, the idea behind DAGs are not new, there have been diagrams in epidemiology since years, but each epidemiologist has his own specific ways to draw the different relationship between various variables factors. Did you ever got stuck in a discussion about if something is a confounder or not? If you don’t get it resolved by talking, you might want to draw out the your point of view in a diagram, only to see that your colleagues is used to a different way of drawing epidemiological diagrams. DAGs resolve this. There is a clear set on rules that each DAG should comply with and if they do, they provides a clear overview of the sources of confounding and identify the minimal set of variables to account for all confounding present.
So that’s it… DAGs are a nifty method to talk the same idiom while discussing the causal questions you want to resolve. The only thing that you and your colleague now can fight over is the validity of the assumptions made by the DAG you just drew. And that is called good science!
The paper, with first author MMS, appeared in the methodology series of the journal Nephrology Dialysis and Transplantation, can be found here in pdf, and also on my mendeley account.
bob siegerink 