Pigache R M, Graham B R, Freedman L
Biol Psychol. 1976 Sep;4(3):213-35. doi: 10.1016/0301-0511(76)90006-5.
Treatment effects can often depend on baseline (pre-treatment) variables. This is ignored in many studies, or else the relationship is removed statistically by analysis of covariance. The latter method, however, assumes that slopes are equal and also loses information on the base-line contribution. Instead, the Johnson-Neyman technique does not assume that slopes are equal and furthermore permits examination of the baseline contribution. The method was devised for education research and is here extended to biological studies. However, the method assumes that variances are equal for the regressions, which might not be always so. The mmodification proposed accommodates inequalities of variance, whether intrinsic to the regressions or resulting from differences in group size. The Johnson--Neyman technique is discussed in relation to alternative analyses and, in appropriate situations, is considered to yield more information. Furthermore, with the refinement described, it involves even fewer assumptions and becomes more powerful.
治疗效果往往取决于基线（治疗前）变量。许多研究忽略了这一点，或者通过协方差分析在统计上消除这种关系。然而，后一种方法假定斜率相等，并且还会丢失关于基线贡献的信息。相反，约翰逊 - 内曼技术不假定斜率相等，而且还允许检查基线贡献。该方法是为教育研究设计的，在此扩展到生物学研究。然而，该方法假定回归的方差相等，而实际情况可能并非总是如此。所提出的修正方法考虑了方差的不平等，无论是回归本身固有的，还是由组大小差异导致的。本文讨论了约翰逊 - 内曼技术与其他分析方法的关系，并认为在适当情况下，它能产生更多信息。此外，通过所描述的改进，它所涉及的假设更少，且更具效力。
