How Large Must an Associational Mean Difference Be to Support a Causal Effect?

With their recent publication, our statistics expert Michael Höfler and his collaborators Ekaterina Pronizius and Erin Buchanan gave a Christmas present to the scientific community.

They discuss bias in observational studies due to unconsidered common causes of factor and outcome, what quantities this bias depends on, and how it can be accounted for. They suggest higher confirmation thresholds than those used to test for associations. They also show how to visualise the quantities and how they affect whether a causal effect is supported or not. An R package and a Shiny app called ViSe is provided.

For more information you can find the full article here