While educational researchers frequently collect data from a sample of individuals on a sample of variables, they sometimes collect data involving samples of: (1) subjects; (2) variables; and (3) occasions of measurement. A multistage procedure for analyzing such three-mode data is presented, focusing on effect sizes and graphic confidence intervals. The procedure was illustrated with a data set involving the perceptions of educational equity issues by medical school admissions officials. Admissions officials at 144 medical schools were asked to rate the extent of their agreement with 30 statements using a 5-point Likert scale; 91 subjects replied, yielding a response rate of 63.2%. A 91 x 30 x 4 matrix defined the three-mode data set. In the first phase, the referent mode was collapsed into the subject mode to yield a matrix that was subjected to principal components analysis. The first six components were extracted, and rotated to the varimax criterion. The second step involved retrieving the referent mode involving the four ethnic groups (Black students, Hispanic students, other minority students, and non-minority students). This was done by computing the 95% confidence interval about the mean factor score on each of the six components consisting of the ratings by the admissions officials of a single referent group. One figure illustrates the relative simplicity and graphic features of this approach; one table presents the item markers of the six components; and one appendix contains the matrix. (SLD)