New Statistical Methods for Analysing Social Structures: an introduction to multilevel models

Abstract An introductory account is given of developments in multilevel modelling of educational and other social data. The technique is introduced with some simple examples and its importance is explained. Examples of applications in a number of areas are given, including repeated measures designs, school effectiveness studies, area‐based studies and political opinion sample surveys. Almost all data collected in the social sciences have some form of inherent hierarchical structure, and this structure should be reflected in the statistical models that are used to analyse them. It is suggested that multilevel techniques and associated software packages have reached the stage when they can and should be applied routinely in the analysis of social data, and that failure to do so can result in potentially serious misinterpretations.