John R. Skalski, Kristen E. Ryding, and Joshua J. Millspaugh. 2005. Elsevier-Academic, San Diego, California. xiii + 636 pp., 90 text figures. ISBN 0-12-0088773-8. Cloth, $69.95.—Collection of data on the age and sex composition of birds harvested by state and federal agencies has a long history, whereas less effort has been focused on estimation via capture-recapture, distance sampling, and other procedures. Yet in reading the literature on wildlife statistics, one would obtain the opposite impression, with most of the advances since the 1960s having been in areas of capture-recapture, tag recovery, and distance sampling (e.g., Otis et al. 1978, Burnham et al. 1980, Brownie et al. 1985, Pollock et al. 1990, Buckland et al. 1993, Williams et al. 2002). By contrast, there has been relatively little progress in the analysis of count- based data since the development of these methods between the 1940s and 1960s. Until recently, little formal statistical theory existed for many of these methods, so that variance estimates, confidence intervals, and assumption tests were generally unavailable. This book attempts to remedy the situation.

Chapter 2 provides an excellent review of population dynamics, especially of harvest theory, which is important because many of the data sources later considered derive from hunter and angler harvests. Subsequent chapters cover count-based approaches (direct counts, harvest surveys, age, and sex ratios) but also include methods such as capture-recapture and distance sampling for comparison and assumption testing. Chapters cover estimation of sex ratios (chapter 3), productivity and survival (chapters 4 and 5), harvest and harvest morality (chapter 6), population change (chapter 7), population indices (chapter 8), and abundance (chapter 9). Chapter 10 provides examples using multiple approaches to estimate parameters. Despite the chapter title, most of these are not “integrated” analyses. A notable exception is a study of Ring- necked Pheasants (Phasianus colchicus) in which change-in-ratio, catch-effort, and capture- recapture data were incorporated into a single likelihood, providing more precision (and fewer assumptions) than each method separately.

The material in chapter 8 on finite sampling is very well written, but because these concepts apply generally, coverage earlier in the book would have been better. A useful appendix on “Statistical Concepts and Theory” covers maximum-likelihood estimation, interval estimation, hypothesis testing, and other topics. However, I could find no mention in the book of bias, accuracy, or precision—which is surprising, given the fundamental importance of these concepts to estimator assumptions.

The authors have done a thorough job of gathering many disparate methods together and providing a comprehensive description of data structures, statistical models (including likelihood formulas where possible), and assumptions. In several cases (e.g., chapter 3), they have also incorporated detection probabilities into the statistical models, so that, given appropriate data, parameters of biological interest can be estimated without making critical and untestable assumptions. Each chapter closes with a schematic decision tree, which can be used to guide selection of an appropriate sampling-estimation approach.

Unfortunately, few of the methods described here can provide, by themselves, reliable inference on populations. In contrast to methods such as capture-recapture, distance sampling, and detection-adjusted visual counts, most do not provide data that can be used to avoid untenable assumptions, or test those that cannot be avoided. In chapter 3 (pp. 65–66), the authors note that “in the absence of auxiliary information about detection rates [sex ratio] is not estimable…[so that]…in populations with different detection probabilities for males and females, an unbiased estimate…is not possible.” However, sex-biased detection is common, and I question the value of a methodology that is not robust to an assumption that cannot be tested. Likewise, many of the methods described for harvest mortality (chapter 6, p. 287) “require the detection process to be stationary before, during and after the periods of harvest.…[but] the data usually collected by these techniques are insufficient alone to assess the validity of [these assumptions].” More serious is the use of vertical life-table (VLT) analyses to estimate age-specific survival and other parameters (chapter 5). Here, restrictive assumptions are required, including stationarity (λ = 1) and stable age distribution (SAD), which are seldom true in practice, especially in harvested populations. Unfortunately, these assumptions cannot be tested with the most common data structures (e.g., single, time-specific age distributions, as obtained via harvest surveys). Finally, it is not true (p. 163) that the assumption of age stability can be relaxed if the population is at SAD for a portion of the year. The SAD assumption is related to the fact that time-specific age distributions are, by definition, a mixture of ages from several cohorts, and this mixture only reflects age-specific survival when the age distribution is both stable and stationary between years.

A number of methods described are so-called “indirect methods” and are not statistical- estimation procedures per se, but model-based projections. In some cases (e.g., sex-ratio projections), these require independent estimates of survival to project sex ratios, and typically they also require assuming SAD and λ = 1. Despite a disclaimer (p. 76) that these are projections, they are later referred to as “estimates,” and the “estimation” [sic] formulas are indicated by hats. Unfortunately, this blurs the distinction between estimates (which are functions of data) and model projections (which may or may not be) and will be confusing to managers, who may see these views of “estimation” as interchangeable. Likewise, methods for “estimating” λ (chapter 7) based on Lotka and Leslie models provide projection of λ at SAD, and so cannot be used with estimates that assume that SAD and λ = 1 have been attained.

Chapter 8 covers population indices, and the authors frankly admit (p. 360) that “without [auxiliary information on detection probabilities] it is usually impossible to assess [the validity of the index assumptions]…. Thus population indices are, in part, a matter of faith….” I concur with this, and with the statement that “science is not based on faith but rather on the interpretation of empirically derived facts.” However, they assert that “despite the inferential weaknesses of indices, they remain a cornerstone of wildlife science.” I cannot deny that these methods have appeal in being “cost efficient” (well, at least low-cost—the benefits are debatable), but the contradiction here is jarring. The authors later refer to “calibration” of indices, but the examples principally involve conversion between index and estimate units, without verification of the assumptions of either.

Overall, this book is a useful contribution to the literature on wildlife demography. My concern is that the book will be cited by some in defense of index and count methods in lieu of other, more robust approaches (e.g., Williams et al. 2002), or more recent work on count-based occupancy estimation (MacKenzie et al. 2006). Wildlife ecology is better served by sampling to obtain data that can be used in statistical models without requiring assumptions that are likely false—and that, when false, lead to unreliable inferences.