ABSTRACT
Many analyses in statistics are concerned with what would happen if the units in a population changed their values of a categorical variable (their status), while keeping all other circumstances (their background) unaltered. If we could control the process by which the values of this categorical variable are assigned, the problem would be resolved by experimental design-by assigning the values of the variable to the studied units completely at random. Numerous problems in social sciences cannot be studied by this approach because they relate to variables that cannot be controlled externally. Often they are recorded a long time after their values are set and would in normal circumstances remain unchanged. In any case, the units to be observed (individuals, households, schools, businesses, or the like) would reject such control for any purpose. Their right to do so is protected by legal code and ethical standards. Neither could we contact a unit twice; first to arrange the assignment, and then, some time later (e.g., in several years’ time), to observe the outcome.
