Fig. 3.

Example of one-factorial design of three conditions that share a common activation pattern. (A) The data consist of 5 measures of the control condition, which provides a measure of the common activation pattern (first row in matrix U), followed by 5 measurements for the 3 conditions each. (B) Due to the common activation, the sample correlations (light gray line) between mean patterns are much higher than the true correlations (line). Prior subtraction of the mean pattern of the control condition (dark gray dashed line) lowers the estimates, but still overestimates the correlation and underestimates the differences. Using the pattern component model (black dashed line), valid estimates can be obtained. (C) Multidimensional scaling based on the estimated correlation coefficients shows distortions of similarity structure for sample correlations between mean patterns.