Completely randomized factorial analysis of variance using ranks

Several rank tests are available for the completely randomized, fixed effects multifactor design; however, none has been widely adopted nor has a systematic evaluation of the characteristics of these tests been undertaken. Rank tests due to Hettmansperger (1984), McSweeney (1967) and Puri & Sen (1985) are illustrated and evaluated on the basis of their extension to designs with more than two factors and small sample distributional properties (i.e. Type I error rate and power). The rank transform procedure of Conover & Iman (1981) is also discussed. Simulation results indicate that the F test is robust and shows good power across a variety of conditions. Among the non‐parametric procedures, the Hettmansperger test produced the largest power.