This book describes a framework for projecting mortality rates by age groups (17), sex (2), causes of death (24), and countries (191). The framework is composed of statistical models that generalize linear regression for time-series analysis. New methods are introduced for incorporating prior information, and mortality data were mostly derived from the World Health Organization database. The authors also include various explanatory variables such as gross domestic product, tobacco consumption, fat consumption, etc. Given the enormous task at hand, the authors acknowledge the lack of data and the need to pool data from similar cells. Instead of the traditional practice of pooling based on similar coefficients, the authors pool based on similarities across the expected values of the dependent variables.

In addition to describing the framework proposed by the authors, this book illustrates some model results of age-sex-country-cause-specific mortality projections, including illustrations of choices available to the user. Many existing methods currently used in projecting mortality are described in the book and a comparison of model results between the new framework and some existing models is given. This book was intended to be accompanied by a software package that allows the user to use this mortality framework, but to include many of their own choices on a variety of adjustable settings (degree of smoothness, explanatory variables, and model specifications). This software would be very helpful in understanding the framework and potential user choices and evaluating results. This reviewer cannot comment on the software package, since it was not included with the review copy provided.

I found the book informative and interesting and a positive contribution to projection methodology. It is useful to develop and examine alternative ways of projecting mortality and this book presents some novel approaches. It also allows the user to choose a number of model specifications, thereby adding flexibility to the model. However, it is difficult to prove that one model can be said to outperform others without explaining what criteria would be used to judge the winning model, and whether the criteria would include subsequent outcomes, fewer input assumptions, historical fits, etc. It would also have been helpful to have expanded on the data descriptions used for the modeling.