Significance Test in Studies of Nonrandomly Assigned Subjects: A Call for More Discussion

Abstract

There is a fundamental difficulty with a significance test in studies in which subjects are nonrandomly assigned, e.g., differences in strain, sex, age groups, and species, etc. The implausibility of the assumption of random sampling and the use of the significance test in such studies is pointed out. Inference without reliance on the significance test is needed not only in studies of subjects nonrandomly assigned but also in those in which assignment of subjects is random. It is emphasized that inference without the significance test remains a standard scientific approach, in spite of the widespread use of many statistical tests. A call is made for more discussion of the problem.