When and why do (survey) experimental treatment effects generalise?

Abstract not yet added