Article Text

Download PDFPDF

Multilevel modelling of health statistics
  1. J Merlo
  1. Department of Community Medicine, Malmö University Hospital, Malmö, Sweden; juan.merlo{at}

    Statistics from

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

    Edited by A H Leyland, H Goldstein. New York: Wiley, 2001.

    This is a collaborative work that reunites many of the best authors in the area of multilevel analysis. The book fulfils the need of explaining multilevel modelling to researchers in the health sciences. It has a practical orientation, focusing on a series of applications of relevance by means of thoroughly explained, pertinent examples, rather than simply presenting algebraic formulations.

    The book opens with a good introductory chapter that provides a didactically sound general survey of multilevel concepts and models, and is well complemented by “Context and composition” in Section 12.4. If studied in conjunction with some instructional aids (see and, it is probable that this chapter will increase the number of multilevel analyses performed in the field of community health. After this highly readable beginning, the book moves on to address statistical techniques in sections on “Modelling repeated measurements”, “Binomial regression” (with an interesting example concerning equity in health care access), “Poisson regression”, “Multivariate models”, “Outlier, robustness, and detection of discrepant data”, “Modelling non-hierarchical structures”, “Multinomial regression”, and “Spatial analysis”. There is also a chapter about the time honoured “Institutional performance” (although the reader should not fail to supplement it by reading its longer version1). A well presented chapter on “Sampling” by Snijders provides a good understanding of sample size calculations, and the similarities and differences between multistage samples and multilevel study designs. Finally, there is a useful overview of the software currently available (this reviewer personally uses MLwiN).

    The pedagogical challenge of explaining multilevel analysis to the health epidemiologist is an ambitious one, and this volume is an important step in the right direction. Leyland and Goldstein have provided a tool that will not only benefit health epidemiologists who already possess statistical skills, but should also considerably improve the dialogue between epidemiologists and statisticians.

    Multilevel analysis is nothing new for many mathematicians, statisticians, economists, sociologists, geographers—and even certain dentists and pharmacologists. For the health epidemiologist, however, it is represents a major scientific advance. In all likelihood, multilevel analysis would have satisfied Thomas S Kuhn, when he wrote: “My main goal is to urgently demand a change in the perception and in the evaluation of the known data”.2 But whether “changes in the perception and in the evaluation of known data” are always welcome may be another question.