ANALYSIS OF COHORT EFFECTS IN MIXED LONGITUDINAL DATA SETS

Mixed longitudinal designs are among the most efficient for the study of growth and developmental processes. In this approach, one studies s everal (birth) cohorts, each for a relatively short length of time, an d then links the growth curves for the individual cohorts together to obtain the growth curve for the entire length of time spanned by the a ges of the subjects in all cohorts. Thus, e.g., in the Nijmegen Growth Study, three cohorts were each studied for 5 years, the intent being to join the three curves together to form a single curve covering the entire period from 4 to 14 years of age. In order for this approach to be valid, there either should be no cohort effects (secular trends) o r the fitted curve must be adjusted in some way to correct for such ef fects if they exist. The question thus arises as to how one should tes t for the presence of cohort effects and what one should do about them if found. The problems which may arise using height and weight data f rom the Nijmegen Growth Study are illustrated. In particular for girls , height and weight both show cohort effects (at 9.25 years of age) wh en the raw data are used. If, however, the observed data are used to e stimate the values at the target age, and these values are used in the comparison, differences between the cohorts are no longer significant . The problems are further illustrated using data from a mixed longitu dinal data set of cleft lip and palate patients and data from the Nati onal Dutch Growth Study 1980.