THE EXPANSION of research in chronic diseases in recent years has been accompanied by more frequent application of the longitudinal approach, a methodology long familiar to investigators in the field of human growth and development [l]. With experience the special problems of this complex research strategy have become clearer. High on the list of these problems are the restrictions on the size and nature of the study population often imposed by cost and management considerations, cumulative losses of subjects to observation through non-cooperation and migration, severe limitations on the opportunity to test consistency of findings from one study to another, and the long wait for final results. Because of the difficult problems of design, content and execution, the reasons for undertaking a longitudinal study and the unique qualities of this approach periodically come under critical scrutiny. Briefly stated, justification for undertaking this type of inquiry rests on the decision that it can best lead to an understanding of the interaction between time and changes in a characteristic. For this purpose, the distinguishing feature of the longitudinal approach in its data collection phase consists of repetitive observations over a long period of time on the same individuals or on a subsample of these individuals. In its analytical phase, it attempts to measure change over time in a study population whose composition is defined as of the beginning of the time interval. Under certain circumstances all of the necessary data may be available in records established before the investigation starts. However, this paper is concerned primarily with the more usual situation in which access to patients is required. Before undertaking a longitudinal study, more than a casual glance is given by many investigators to the comparatively more tractable cross-sectional methodology. As defined here, the data collection process of this approach consists of observations made at a particular point in time, and the analysis is concerned with variability in a