This book is about policy change in the U.S. national government. Drawing on their exhaustive collection of policy data for the last three decades, Bryan Jones and Frank Baumgartner make a very persuasive argument for their “punctuated equilibrium” theory of public policy. The authors detail a catalog of reasons why policy changes are typically small, including (among other things) individual-level “bounded rationality” and institutional properties like “friction.” These many factors combine to produce policy that appears to change incrementally most of the time, but there are infrequent moments in which major change is observed. The amount of nonincremental change is linked to the nature of political institutions and to the stages of the policy process. The final stage of the process—legislative change—is costly and slow, producing infrequent change, while changes in the earlier stages are less costly and therefore more frequent.

According to the statistical argument that frames the entire book, the theory of incrementalism predicts that the distribution of policy change ought to appear “Normal.” Supposing that a sum of separate random variables is a description of the policy process, as the incrementalists seem to imply; then very general statistical results indicate that the frequency distribution ought to be bell shaped. (This is one of the central limit theorems.) With that null hypothesis clearly pinned down, the authors then offer a wealth of data to indicate that the distribution of observed change is not Normal. Their whole argument is nicely summarized in Figure 4.14 (p. 111), which shows that the distribution of U.S. national budget changes is a sharply spiked distribution with long tails. As they demonstrate, the severity of the deviation from normality is predictable according to their theories of institutions and policy process.

The authors have wrestled with a set of very difficult research problems that have taken them off the beaten path in political science. They are apologetic in their presentation, asking for the reader's patience in considering the shape of statistical distributions. Possibly fearing the alienation of readers who lack an interest in mathematical details, they make their statistical argument in an entirely verbal format. Their effort to keep the presentation clear and readable is certainly successful, but some of the more mathematically inclined readers may feel that a more rigorous presentation would be suitable. Even key statistical concepts, such as the Normal, Pareto, or double-exponential distributions, are not given explicit mathematical definitions. The terms “kurtosis” and “leptokurtotic” are introduced on page 111, but formal definitions are not found until a footnote on page 181. One might wish that there were a technical chapter or appendix with formal definitions of the distributions, a mathematical derivation that justifies the use of these particular distributions, and a formal comparison with similar models from other fields (especially Zipf's Law). In some sections of the book, the authors seem to hint that policy change might reflect a mixed stochastic model, one which combines draws from a Normal distribution during “incremental times” and some other distribution during the infrequent periods of change. A formalization of such a characterization would be most welcome.

Without a doubt, this book will be very useful in graduate seminars. It touches on many of the biggest ideas in social science, ranging from theories of individual behavior and rationality, to “middle level” phenomena like political organizations, to the highest level, institutions and policy change. It is also written beautifully. The simple elements of writing—paragraph construction and word selection—are accomplished with so much grace that the presentation is more reminiscent of a trade paperpack than an academic monograph. The central contentions are clearly spelled out, without mathematics. The early research on incrementalism is described in such a rich and insightful way that the presentation should serve as an example for dissertations for some time to come.

Some comments in the early portion of this book might distract the reader from its central themes, or even give the reader the wrong impression. There are derisive comments (without argument or citation) about the Bush administration's policy toward Iraq (p. 8) and former Attorney General John Ashcroft (p. 15) that seem out of place. The book is framed around the idea that humans are uncertain about the truth and have a difficult time managing decisions, and yet the authors seem certain of their own political judgment. Similarly, on page 17, one finds the assertion “This means that learning is not Bayesian” without an effort to explain what the Bayesian approach might be or how exactly it is inconsistent with the authors' description of learning. The frequency of these distracting comments is much lower after the first chapter.

The most resounding theme of the book, which is established persuasively, is that the distribution of observed change (in a variety of indicators) is not Normal. With that contention firmly established, we are left with plenty of research to do. While we can safely reject the elementary version of the incremental theory, we are not confident about what factors cause spiked distributions. Political scientists, along with economists, sociologists, ecologists, physicists, geologists, and geographers, are confronted with the apparent ubiquity of power law distributions in aggregate data. In a very real sense, the existence of spiked distributions is an overdetermined outcome, one that is consistent with many different models of the behavior of individuals, groups, and institutions. Like all truly valuable research projects, The Politics of Attention thus answers some questions definitively and also clearly frames questions for future study.