ABSTRACT
The treatment of outliers has a long history. Many of the most influential statisticians have made contributions to the detection and treatment of outliers in data, their work spanning over a century, often devoting substantial portions of their career to this study (details can be found in excellent reviews of this field; Barnett & Lewis, 1994; Hawkins, 1980). More recently, multilevel modelling methods have been used to diagnose outliers with theoretical work on diagnostics in multilevel models with particular application to the detection of outliers (Hilden-Milton, 1995; Hodges, 1998), and a number of practical procedures for dealing with outliers, applied to cross-sectional educational data (Langford & Lewis, 1998). However, although there have been many recent examples of the application of multilevel and related growth models to longitudinal data, both on physical attributes and psychological processes, with both single and multivariate dependent variables (Goldstein, 1989, 1995; Hoeksma & van der Boek, 1993; Longford, 1993; Rogosa & Saner, 1995; Sayer & Willett, 1998; Ware, 1985), and although longitudinal missing data has received some attention (Diggle, 1994; Diggle & Kenward, 1998; Goldstein, 1995), to our knowledge no application of multilevel modeling methods has been made to the detection of outliers in longitudinal data.
