心理学的測定とノンパラメトリツク法

抄録

Psychological scales are either nominal, ordinal, interval or ratio. As no direct measurement of the psychological (intervening) variables is possible, overt behavioral variables must first be measured. The latter are based on physical scales, which may be called phenotypic or peripheral scales as contrasted with indirect, genotypic scales of psychological variables.
Thus, relationships between the two types of scales should be made clear in any psychological measurement. Many people believe that transformation of variables for homogeneity of variance, for instance, is done purely from statistical expediency. This, however, does not seem to be correct on the basis of the fact that we are always dealing with psychological or genotypic scales and not numbers per se. Transformation of scores is only legitimate if it improves the meaning of the psychological variables.
Traditional statistical tests may be used for psychological scales provided that they are additive or they have equal units. Non-parametric tests, on the other hand, are most appropriate to ordinal scales. Attention is called, however, to the fact that some nonparametric methods assume additivity of the scale and that, therefore, they should not be used for the ordinal scale.
Three uses of nonparametric statistics may be mentioned. First, the methods may be applied to ordinal and nominal scales. Second, they can be used for interval or ratio scales in place of traditional statistical methods because nonparametric tests do not have such assumptions as normal distribution, homogeneity of variance and the like. Third, nonparametric methods are superior to traditional methods in the simplicity of computation for small samples.