The simple scalability of documents

s before asking them to make any pair comparison judgments. Figure 1 nicely displays the order effects discussed by Eisenberg and Barry (1988), but intended in their case. In this case the bias results from presentation to subjects as the first pair, the set members judged most and least useful by them. The unfortunate coincidence of this pair as the first pair presented causes a strong degree of intransitivity to appear in Figure 1 for the first two columns and also results in a marginally acceptable score for stress. The effect is removed simply by dropping the first column of choice ratios from the matrix to create the values appearing in Figure 2. Figure 3 represents the earlier data from the work of Rees and Schultz (1967) obtained by the method of constant sums. The reader will note that of all four data sets so represented, this matrix displays the greatest degree of orderliness and regularity. This is remarkable, since Rees and Schultz (1967) used only 16 subjects to obtain this data, as opposed to the 51 subjects providing the data for Figures 1 and 2. This suggests that the method of constant sums may be both more efficient and more reliable than the methodology employed in this experiment. In this method, described in Comrey (1950), subjects divide 100 points between each member of a pair to express their choice of the member evoking the strongest response. In the present experiment, subjects made a binary choice between pair members. An additional experiment reported by Rorvig (1987) also confirms the superiority of this method of pair comparison judgment elicitation. The results of the transitivity and independence tests appear in Tables 1 and 2.