Key points are not available for this paper at this time.
A test is made of the hypothesis that little error is introduced into multivariate analysis by the assignment of a linear metric to ordinal data. It is shown that previous attempts to test this hypothesis neglected a number of relevant variables. A more thorough test is made which attempts to include all of the relevant variables more or less simultaneously. It is concluded that the error that may result from the arbitrary assignment of equal distance scoring to ordinal variables is greater than has been previously reported. The main reason why the present test of the little error hypothesis reaches a conclusion at variance with previous tests is that they considered only extreme correlations with an outside variable-either perfect or very low.
Frank S. Henry (Thu,) studied this question.