Models for Intensive Longitudinal Data
Classical longitudinal analysis is mainly focused on examples to some ten occasions. New technologies lead to longitudinal databases with a considerably higher intensity and a substantially larger volume of data, for which the new term ‘intensive longitudinal data’ (ILD) is used. The number of occasions in ILD may be hundreds or thousands. However, the main difference between ILD and other models pertain to the scientific motivations for collecting ILD, the nature of hypotheses about them and the complex features of the data. The main themes in ILD modeling are: (i) the complexity and variety of individual trajectories, (ii) the role of time as a covariate, (iii) effects found in the covariance structure, (iv) relationships that change over time, (v) interest in autodependence and regulatory mechanisms.
The book is a collection of eleven papers (arranged as chapters) written by different authors. The introductory chapters focus on multilevel models and on marginal modelling through generalized estimating equations. Later chapters describe methodological tools from item response theory, functional data analysis, time series, state-space modeling, stochastic differential equations, engineering control systems, and models of point processes. Theory is illustrated on real data drawn from psychology, studies of smoking and alcohol use, brain imaging and traffic engineering. Some authors have supplied programs and source code examples. They are available at a website accompanying the book. By the way, the formula on page 118, line 6, should read sm=c22m-1+c22m. The remark on page 130 that the order p of an autoregressive process is often determined heuristically should be complemented by another remark that the order p is also often determined using AIC, BIC and similar criteria. This collection contains many interesting models and practical examples. The volume can be attractive reading for statisticians working in biostatistics and behavioural and social sciences.