Repeated-measures analysis of variance

Repeated-measures analysis of variance is concerned with study designs in which the same patient is observed repeatedly over time. Such data are also referred to as longitudinal data, or panel data. In analyzing these data, we must take into consideration the fact that the error components of repeated observations on the same patient are usually correlated. This is a critical difference between repeated-measures and fixed-effects designs. In a repeated-measures experiment, the fundamental unit of observation is the patient. We seek to make inferences about members of a target population who are treated in the same way as our study subjects. Using a fixed-effects method of analysis on repeated-measures data can lead to wildly exaggerated levels of statistical significance since we have many more observations than patients. For this reason, it is essential that studies with repeated-measures designs be always analyzed with methods that account for the repeated measurements on study subjects. Example: effect of race and dose of isoproterenol on blood flow Lang et al. (1995) studied the effect of isoproterenol, a β-adrenergic agonist, on forearm blood flow in a group of 22 normotensive men. Nine of the study subjects were black and 13 were white. Each subject's blood flow was measured at baseline and then at escalating doses of isoproterenol. Figure 11.1 shows the mean blood flow at each dose subdivided by race. The standard deviations of these flows are also shown. At first glance, the data in Figure 11.1 look very much like that from a fixed-effects two-way analysis of variance in that each blood flow measurement is simultaneously affected by the patient's race and isoproterenol dose. The fixed-effects model, however, provides a poor fit because each patient is observed at each dose.