Whereas in an experiment, the researcher can do random allocation to ensure that the relationships do not come from prior group differences.