To the Editor.—
From a statistical perspective, data in which the same subjects are repeatedly measured during time (ie, longitudinal data) present special problems. Repeated observations are, in general, correlated with each other, so that the display of overall time-point mean values represents a significant loss of within-subject information. This problem is further complicated when two longitudinally measured variables are jointly examined in the same subjects.For heuristic purposes, we take as an example the study by Greden et al,1 who graphically presented the "association" between clinical improvement (as measured by the Hamilton Rating Scale for Depression) and a steady decrease of postdexamethasone plasma cortisol concentrations in a sample of depressed patients during the course of treatment (Figure, top). The original intent of this Figure was to demonstrate that a strong positive association exists between changes in these two variables during time; however, this need not necessarily be the