Nonrandom Assignment in ANCOVA

A specific form of nonrandom assignment to treatment groups, the “alternate ranks” design, was investigated. This design eliminates the possibility of a correlation between the covariate and treatment, and rules out experimenter bias in assignment of subjects to groups. With covariate means equated by this design, the parameter estimates and F values were precise and unbiased following ANCOVA. The frequencies of rejection under the null hypothesis conformed closely to the expected number of rejections at four different alpha levels. The alternate ranks design is especially recommended for experimenters who anticipate difficulties meeting ANCOVA assumptions. Interpretation of results is also facilitated by matching groups in this manner.