Multilevel Analysis: Techniques and Applications [Paperback]

Hox provides a good, conventional tratement of multilevel modeling, one that is much better than De Leeuw's on-line review would have the reader suspect. When struggling with this material for the first time, moreover, Hox's one-page treatment of models with more than two levels is worth the price of the book. His cautionary comments alert the reader to the sometimes intractable complexity that may be occasioned by even three-level models, much less four or more.

Kelvyn Jones takes issue online with this admonition, and, no doubt, there are informative three level models. But Hox's observation is still eminently applicable. In my experience, the amount of work required to make the transition to three-level models is underestimated in most textbook accounts.

Part of the problem inheres in making more and more difficult specification decisions in the absence of readily interpretable guidance from theoretical and substantive literature. Beyond that, models with three or more levels quickly become statistically very complex. The number of random component variances and covariances increases dramatically with he addition of predictors with random slopes. Parallels between two-level and three-level models are a good deal less obvious when it comes to actually specifying three-level models. Model building facility takes practice.

In spite of all this, three-level models can be useful, providing insights that otherwise would not be available. However, off-handed assumptions that three-level regression models are just straightforward extensions of two-level models may lead us to expect too much. Three-level models are uniquely complex, and their effective application demands more theoretical and substantive knowledge than is typically available.

OK, Hox's one-page warning did not contain all this material, certainly not enough information to actually buy the book for just one cautionary page. Nevertheless, until I stumbled on that page, I struggled more with, and gave much more attention to models with more than two levels than they usually deserve.

Another real virtue of Hox's book is that, in contrast to most other texts dealing with multilevel models, it gives adequate attention to the really interesting topic of constructing intervals for random intercepts and slopes, providing estimates of how much they vary group to group. In some instances, the degree of variability is startlingly large, making clear that fixed components, as usually reported, can be very misleading.

For most readers, Hox's book is not easy, but it's clear that the author understands that the complexity of the material will make it difficult for most of us to quickly grasp. It is obvious from the patient, largely non-mathematical nature of his presentation that he wants folks who have paid for his book to benefit from an investment of time and effort in understanding multilevel modeling. He does this, moreover, while covering a broader range of topics than most texts of this kind.

All tolled, Hox's book certainly deserves the four stars I've given it. Another edition is scheduled to be published in 2010, and it deserves a look.

Overall this is an excellent text for understanding and conducting mulitlevel analysis. Hox makes the promise that the data sets used in the book would be made available online. At the time of this review Aug 12 2011 the data is not available.

Very useful text and has helped me move forward in my research. It situates the multilevel analysis procedure in the continuously evolving context of multivariate analysis and outlines some of the major parameters for conceptualizing and building your own models. The text is very accessible for someone who is considering multilevel modeling and has some basic background in regression analysis and other multivariate techniques. It is written in a very non-technical and non-algebraic language with numerous examples.

› Go to Amazon.com to see all 3 reviews  4.3 out of 5 stars