A Quarter Century of Advances in the Statistical Analysis of Longitudinal Neuropsychological Data

Objective: Over the last 25 years, there has been an unprecedented increase in federal funding for large-scale longitudinal studies, many of which collect neuropsychological or neuroimaging outcome measures. These studies have collected data from thousands of study participants across multiple waves of data collection over many years. With the increased availability of longitudinal data, data sharing policies have become more liberal, thereby offering significant opportunities for interested researchers to carry out their own longitudinal research with these data. At the same time, these opportunities have stimulated new conceptualizations of longitudinal change and have led to the development of novel approaches toward analysis of longitudinal data. My aim is to review these new conceptualizations and novel data analytic approaches. Method: In this article, I describe the state of the field a quarter century ago with respect to available longitudinal studies, and I trace the growth of federally funded longitudinal studies over the last 25 years by describing 18 of these projects, many of which are still collecting data. In the second part of this article, I describe changes in the methods used to analyze longitudinal data, transitioning from the paired t test and repeated measures ANOVA to latent change scores, linear mixed effects modeling, and latent growth curve models. Changes in the approach to management of missing data are also discussed. Conclusions: Future studies should abandon traditional longitudinal analytic methods in favor of contemporary approaches given their increased power, greater accuracy, and widespread availability. General Scientific Summary This article describes a number of federally funded longitudinal research studies in cognitive aging and Alzheimer's disease that have made significant contributions to our understanding of longitudinal changes in normal and pathological aging, and it underscores the rapid rise of these funded studies over the last 25 years. The article also considers significant shortcomings and limitations associated with traditional approaches to longitudinal data analysis, such as the paired t test and repeated measures analysis of variance. Finally, contemporary methods are presented for management of missing data and for longitudinal data analysis, including linear mixed effects modeling, latent growth curve analysis, latent change scores, and generalized estimating equations, and the article illustrates the advantages of these contemporary approaches over traditional methods of analysis.