Mechanical Reliability Improvement—Probability and Statistics for Experimental Testing

The statistical aspects of this presentation are a bit tenuous (e.g., why not consider maximum likelihood estimation rather than moment-based estimation?), but I can see the need for the chapter given the intended audience. Chapter 6 provides some nice graphical tools for use in the detection of change. Chapter 7 discusses many different tests for trend and autocorrelation, as a complement to the graphical methods of Chapter 6. These tests are wellmotivated in the discussion. Chapters 8 and 9 provide statistical tests for detecting differences in moments and distributions. Many parametric and nonparametric tests are considered, and again, are well motivated. Chapter 10 considers methods for modeling change, as well as issues related to estimation (i.e., “calibration”) and veriŽ cation. I was disappointed that modern changepoint time series models were not discussed here, at least in general terms. In addition, the reliance on stepwise methods for variable selection in regression, rather than “best subsets” approaches, is also less than optimal. More discussion of estimation issues would have been helpful. Chapter 11 considers the simulation of hydrologic processes. Too much emphasis is placed on the mechanical generation of random numbers from various distributions. This seems out of place here, especially because the preceding discussion made no mention of many of these distributions (e.g., multinomial, Poisson, triangular). Chapter 12 discusses sensitivity analysis. I found this one of the book’s best chapters. A very clear connection is made between deterministic sensitivity analysis and sensitivity analysis in statistical models (e.g., regression). Chapter 13 is well written (by Glenn E. Moglen), but somewhat out of place. It considers frequency analysis in the context of geographic information systems (GIS). Although this is a very relevant topic, the chapter’s writing style is different from the earlier material. Overall, Modeling Hydrologic Change: Statistical Methods presents a useful introduction to issues concerning hydrologic change. There is, in the book’s early portion, a tendency to include terms/ideas before they have been introduced, but this is only mildly distracting. It would have been an improvement had the editor considered more advanced statistical modeling ideas. In particular, there is no mention of Bayesian methodology. Given the impact of Markov chain Monte Carlo in modeling systems with complicated dependence structures and uncertainty in data, process and parameters, this a nontrivial oversight. The Bayesian framework is perhaps the ideal paradigm for considering such uncertainty and changes in a hybrid deterministic/stochastic system such as is seen in hydrology.