Introducing Multilevel Modeling

Kreft and DeLeeuw's Introducing Multilevel Modeling was written to mercifully place only limited mathematical demands on readers. It includes an interesting and readable account of intercepts and slopes as outcomes. However, in common with most other texts, it fails to accessibly develop parallels between multilevel modeling and more widely understood procedures. Instead, the authors' intent seems to be to highlight differences, perhaps to make the distinctiveness, virtues, and limitations of multilevel modeling clear to the reader. Whatever the value of their approach, I find it much easier to understand multilevel analysis as an extension of multiple regression, a connection which Kreft and DeLeeuw do not develop. Simple examples are presented throughout the book using MLn software. Models with more than two levels are not discussed. Growth models are mentioned in passing. The authors emphasize that the constraints imposed by multilevel analysis make it best suited to relatively simple models, but I think they over-emphasize this judgment. Models much more complex than those used in this text are usefully employed and widely reported in a variety of books and journal articles. The authors also treat close calls with regard to statistical significance as of dubious value. The reason for this is not given, and I much prefer setting an alpha level and then sticking with it, rather than second-guessing results. This is a brief text that addresses some issues of importance in the Frequently Asked Questions section at the end of the book. It is not a comprehensive reference, but that was not its intended purpose. It reads especially well the second time through, and gives a good deal of needed attention to centering, an important topic that is glossed over in most other multilevel texts.

As a foreword, I am a 2nd year psychology graduate student with ANOVA and multiple regression experience. That said, I've found this work to be clear, precise, and straight-forward in introducing the logic and concepts behind why one would wish to use a hiearchical linear model, as well as the foundation of said statistical design. The lack of emphasis on heavy math-based calculations will undoubtedly expedite training and use for those in the behavioral sciences. On that same note, formulae are interjected only when necessary and more often clarify than confuse a given point or concept. The only drawback of this text might be imparting exactly how one performs HLM. There are other books on the market that, though more technical and/or less clear, provide the heavy emphasis on analysis that this book does not quite have. Even so, I am firmly convinced that there is no better book for HLM in terms of laying out the theoretical foundation. An added bonus for me was the fact that ANOVA and regression are richly elaborated upon--a feature that many graduate students might appreciate as a refresher. In short, I highly recommend this book.

I liked the "comman-man-language" used by the authors to explain Multilevel Modeling. The use of MLn software, which is nearly 50% of the book, was a damper. With the commercial world dominated by SAS and SPSS, I would have liked the authors to give the examples for use with SAS

This book is the best intro to the subject that I've seen. The authors minimize the use of notation, mathematics and the like, and invoke the reader's intuition by developing some good, concrete examples. They present those examples as datasets (accessible on the web), as "run" regressions (i.e. with parameter estimates and standard errors), and graphically.They also demo how to "run" each of the examples on a PC, using the program MLn. If you don't use MLn (and I had never heard of it); this part of the book is less helpful. It would be great if, in an accompanying website perhaps, they were to demo the same analyses using other packages (e.g. SYSTAT, SAS and the like). But this book is really quite good, and a terrific addition to any applied social scientist's library.

When analysing data, the relationships between people that belong in the same classroom, live in the same street or suburb, are part of the same family or therapy group,etc., are often ignored. Multilevel or hierarchical linear modelling is a statistical technique for taking into account such dependencies, arranged in hierarchies (e.g., correlations between students within classrooms, correlations between classrooms within schools, correlations between schools within school districts). In other words, multilevel modeling techniques attempt to model the hierarchical relationships that are found in the real world. In the last 10 years or so there has been a growing number of books and software packages concerned with multilevel analyses. Introducing Multilevel Modeling is shorter and slightly less 'mathematical' than most and gives quite a good introduction to the subject. The book makes reference to the British MLn (MLWiN) computer program in its examples, whereas an introductory text arguably should have used the HLM program, for which a cutdown student version is available free. Taking group dependencies into account is extremely important, but unfortunately many researchers will be discouraged by the dry and heavy-going feel of these texts, which is so often the case with anything involving statistical theory. A highly approachable and readable book remains to be written, but Introducing Multilevel Modeling is probably the best of the current crop.