On the cover

Abstract

tion of scale patterns assuming perfect scalability. But such success may be based on any one or more of the three bases mentioned above. Since CR is not a measure solely of scalability, it cannot be an accurate measure of scalability. That a high CR can result from the chance factor of extreme marginals has been recognized in sociological and psychological literature, and at least one political scientist has called attention to the problem. But the solution adopted has been to warn against extreme marginals and to promote an awareness of the relation between the marginals and the Guttman coefficient. For judicial cases this has meant excluding, for purposes of computing CR, all cases in which the vote of the justices has split 9-0, 8-0, 8-1, and 7-0. But this can cover up the extremeness of individuals. Moreover, any cutting point for the exclusion of cases must be arbitrarily or subjeotively chosen and thus the CR will still reflect the extremeness of whatever marginals are included. So, even if the most extreme cases are left out, the coefficient continues to reflect the magnitude of the marginals and, therefore, measures something other than scalability. An automatic correction for extremeness which avoids the subjective treatment of the problem has been devised by MenzelZ and is known as the coefficient of scalability (CS ). CS=l-E/ME, where E is total inconsistencies or errors in the scale and ME is the maximum errors possible. Maximum errors for items, ME(items) = I x N ~ ~ , ~ , . where I is the number of cases, N is the number of justices, and f,,. is the frequency of response for the modal category of a case. Maximum errors for individuals, ME(ind . )=IxN-z s where s is the subscore for the modal category of a justice. Since the coefficient of scalability measures scalability only, the ratio of CS/CR delineates that portion of the CR resulting from the factor of marginal extremeness. The level of significance for CS has been tentatively set at .60 to .65 by Menzel but, of course, this is no more objective than Guttman’s choice of -90 for a significant CR. Since in Table I we have used cases in which the marginals are as extreme as those eliminated under the standard described above, we may ask the extent to which the CR of .963 is a measure of scalability. The ME to be inserted in the formula CS = 1 E/ME is the smaller of the h4E for items and the ME for individuals, since the smaller of two maxima is the effective one. This turns out to be ME for cases, whioh in the case in question is 75. Thus CS = 1 8/75 or 394. Since this figure is considerably above the .60-.65 level and only very slightly below the .90 level we may conclude that an unusually high degree of scalability is present in our scale. We therefore reject H, for H, and conclude that the justices of the Supreme Court responded to civil liberty cases in the 1959 term in ternis of one dominant variableattitude toward deprivation of civil liberty.