Can there be a trade off between internal and external validity

Proponents of randomised controlled trials (RCTs) see in them the potential to revolutionise empirical methods in the social sciences. RCTs are experiments that randomly allocate participants into two groups, one that receives a tr eatment (the treatment group) and one that receives no treatment or a placebo (the control group). This design extracts the causal effect of the treatment. RCTs can be compared to other empiricals methods along the two dimensions of internal validity, i.e . how good they are at finding the true causal relationship in the studied population, and external validity, i.e. how good the discovered causal relationship can be generalised to other populations. While there is broad consensus that RCTs are better in i nternal validity than orthodox empirical methods such as regressions, there is much controversy whether this advantage comes at a cost to external validity. Even more, there are competing views about the relationship between internal and external validity that revolve around the question whether there can be a trade off between them. In this paper, I use a formal approach to define internal and external validity and show that a trade off is conceptually possible and how it might arise in practice.