abstract
Four general criteria are proposed for the choice of a metrical solution: the solution should be compatible with the kind of causal effect in question, be easy to communicate, be unbiased by measurement errors, and be stable across subjects and situations. These criteria are illustrated for various randomized and non‐randomized designs. A distinction is drawn between judgement of an isolated effect and comparison of effects. There is no ‘one correct’ solution, and the metrical unit chosen for comparison will often be a pseudo‐common unit. Consequently, the problem of how to choose an appropriate metrical solution can only be resolved in a vague sense in many cases. It follows that several metrical alternatives, if available, should be applied when comparing effects. The problem of the pseudo‐common unit places several limitations on the possibility of attaining general causal laws.