Book Reviews

Keeping Up With the Quants: Your Guide to Understanding + Using Analytics

Davenport, Thomas H., Jinho Kim. 2013. Keeping Up With the Quants: Your Guide to Understanding + Using Analytics. Harvard Business Review Press, 240 pp. $27.00.

As in his previous books Competing on Analytics and Analytics at Work, which he coauthored with Jeanne Harris, in Keeping Up With the Quants: Your Guide to Understanding + Using Analytics, Davenport seeks to provide managers and executives with a perspective on the capability of analytics. Kim, Davenport’s partner on this book, is a Wharton PhD and professor of statistics at the National Defense University in Korea. The authors open their book by clearly stating their goals: “We would like you to be an intelligent consumer of [advanced analytics], helping to frame the decision, asking questions about the data and the methodology, working to understand the results, and using them to improve outcomes for your organization” (p. 12). Although this sounds straightforward, I immediately recognized it as a deceptively ambitious set of objectives.

The authors organized the first half of the book around an analytics framework that consists of six steps: (1) problem recognition, (2) review of previous findings, (3) modeling, (4) data collection, (5) data analysis, and (6) results presentation and action. This structured approach to thinking about analytics is one of the most significant ideas that Davenport and Kim are trying to convey to their readers. Each chapter includes multiple stories told explicitly through this six-step lens. For some chapters, this works beautifully. For example, I enjoyed reading about a company called Transitions Optical that had successfully modeled and analyzed the impact of various marketing efforts to support better resource allocation. It was also fascinating to get the (admittedly condensed) description of how Bill Fair and Earl Isaac came up with the now ubiquitous FICO score. And the story of how Australian authorities solved an insider trading case using bank records and network theory was downright inspiring.

However, other stories were squeezed uncomfortably into this six-step framework; as such, I felt they were neither credible nor directly relevant to the book’s purpose. In several cases, the authors use this framework to describe landmark academic research efforts that spanned many years and undoubtedly evolved in a far less linear manner; examples include Murray and Gottman’s theory of marital conflict and Snowdon’s study of predictors of Alzheimer’s disease based on data gathered from a population of nuns. In another instance, the authors shoehorn the story of the Houston Rockets’ decision to trade Shane Battier into their six-step process; this felt like a gross oversimplification and ignored the critical role of the NBA salary cap on how and why that trade was made.

About halfway through the book, I had the realization that Keeping Up With the Quants was like a college radio station (“Welcome to KUWQ, broadcasting live from the center of the analytics universe…”). KUWQ is inveterately eclectic, featuring pop records (no analytics book is complete without an obligatory Gary Loveman reference), golden oldies (a wonderful story about a discovery by Archimedes), and some wonderfully weird tunes (uberGeek Garth, Sundem’s fictitious F_ido index for determining one’s suitability for pet ownership). KUWQ also makes up for its occasional lapses and omissions by offering songs, albeit ones that seem a little raw and unfinished, on several important topics that are just not getting the air time they deserve on more commercial channels; examples include the connection between creativity and analytics, suggestions on how managers can bolster their own analytic thinking and capabilities, and the value of relationships and trust in enabling data-driven decision making.

Ultimately, in addition to delivering a solid project framework, some wonderfully instructive stories, and a lot of other interesting related information, Davenport and Kim’s book also reminds us that much of what makes analytics work in practice is simply very hard to summarize and convey efficiently. The plain truth is that analytics, even from a managerial perspective, is a difficult sport to understand from the sidelines. Like a good coach, however, this book gamely tries to inspire us to get into the game and to do our best to keep learning once we get in there.

Vijay Mehrotra

Department of Business Analytics and Information Systems, University of San Francisco School of Management, San Francisco, California, [email protected]

Analyzing Evolutionary Algorithms

Jansen, Thomas. 2013. Analyzing Evolutionary Algorithms. Springer-Verlag, 255 pp. $109.00.

Analyzing Evolutionary Algorithms presents a challenging task in theoretical studies in engineering and computer science. This book focuses on the theoretical analysis of evolutionary algorithms as one of the randomized algorithms in computer science. Although some recent advanced research explored and reported promising methods using runtime analysis, the literature has limited resources that provide an overview of the research developments, research methods, and rigorous proofs on bounds of expected time for a range of optimization problems. This book serves as a very useful source for researchers who are interested in exploring these challenging topics.

This book’s author is an active researcher in the field, focusing on theoretical analysis of evolutionary algorithms. He presents a number of useful methods used in runtime analysis for a range of well-known problems. Most researchers who are interested in the area of evolutionary algorithms could benefit from a deeper understanding of the topic.

The book starts with an overview of the background of the developments in evolutionary algorithms from both engineering and computer science aspects. After setting the goals of the book, the analysis of evolutionary algorithms as randomized algorithms, Chapter 2 provides basics including modules and parameters in evolutionary algorithms. This chapter, which is easy to follow even for researchers new to the area, also briefly overviews some other randomized heuristics that are referred to in later chapters.

Chapter 3 sets the context of the rest of the book, explaining three theoretical perspectives: Markov chains, scheme, and runtime analysis. Chapter 4 discusses black-box optimization, although it does not directly address evolutionary algorithms, and reveals limitations of general random heuristics.

Based on this background, Chapter 5 focuses on runtime analysis using a number of methods. These include fitness-based partitions, lower bounds, typical events, drift analysis, typical runs, delay sequences, and random family trees. In addition, this chapter considers various evolutionary algorithms to address a range of problem functions; examples include (1 + 1)EA, (1 + λ)EA, (μ + 1)∗EA, and (1 + 1)GA. The functions considered, such as ONEMAX, BINVAL, NEEDLE, LEADINGONES, LONGPATH_k, JUMP_k, RIDGE, and PLATEAU, all have specific features, and although rather simple, they facilitate theoretical proofs and can help the reader to gain a deeper understanding of evolutionary algorithms.

Chapter 6 explores four selected topics in evolutionary algorithms: crossover for steady-state GA, mutation for dynamic (1 + 1)EA, cooperative coevolution, and combinatory optimization problems. It also considers additional example problems, including the problems of minimum spanning trees and longest common subsequences.

Although the optimization problems examined are limited to specific functions and use the simplest variants of evolutionary algorithms (1 + 1)EA, the fundamental theory and proofs would be very useful to facilitate future developments in evolutionary algorithms. The book also includes insights for designing efficient evolutionary algorithms for more complicated optimization problems. In the Remarks section in each chapter, the author provides references and pointers to more advanced current developments in the area.

The book contains materials on the basics and background of theoretical proofs of runtime analysis for a range of optimization problems. A reader who has basic knowledge of evolutionary algorithms should find it relatively easy to understand. I highly recommend it for anyone who is looking to explore both the theoretical aspects of evolutionary algorithms and the practical aspects of designing more efficient algorithms.

R. Qu

School of Computer Science, Jubilee Campus, University of Nottingham, Nottingham, United Kingdom, [email protected]

Systems Analysis Tools for Better Health Care Delivery

Pardalos, Panos M., Pando G. Georgiev, Petraq Papajorgji, Britta Neugaard, eds. 2013. Systems Analysis Tools for Better Health Care Delivery. Springer, 178 pp. $129.00.

This slim and relatively expensive hardcover book is Volume 74 in a Springer series dedicated to optimization and its applications. On the back cover, the publisher claims it will benefit researchers and practitioners in healthcare, and lecturers and graduate students in engineering, applied mathematics, business administration, and healthcare; however, the book’s wide-ranging appeal is open to question.

The volume is a collection of eight separately authored chapters that focus on a very diverse selection of topics in which optimization approaches are applied to healthcare. In addition, a short four-page preface penned by the editors briefly introduces the field and summarizes the chapter contents.

On first impressions, this book is a mixed collection of contributions that vary widely in scope, technical and research content, accessibility, style, and quality. Chapter 1 presents a clinical application that shows how semi-infinite linear programming methods can be used to optimize dosage in large-scale radiotherapy treatments. This is both a technical and interesting study. Chapter 2 compares portable asset management in hospitals. This chapter, which has clear objectives, uses simulation modeling to compare management strategies and review the use of various technologies. Chapter 3, which discusses the use of stochastic integer programming in healthcare, is a review rather than a research report. It cites a range of examples, including geographical resource utilization, the surgical theatre room, and staff scheduling, and it highlights potential and specific challenges and limitations (e.g., the challenge of encompassing dynamic aspects of healthcare in the modeling approach). Chapter 4 is also more of a discussion than a specific research project; it provides a general overview of optimization in e-health by drawing on a range of sources and listing areas of application. I found this chapter to have few specific examples and lack detail and depth. Instead, it concentrates on generic aspects, issues, and problems and makes overgeneralized statements. For example, it devotes only four rather bland sentences to solutions and recommendations for the issues identified. Chapter 5 looks at nurse scheduling using integer programming. The topic is poorly presented; it has little or no practical context, such as information on how or whether the solution was applied. In this chapter, 16 pages are appendixes devoted to sparse data tables (12 of these pages have data content of simply ones and spaces). The chapter seems to be crying out for its own optimization algorithm to address the efficient use of paper. Chapter 6 gives an interesting treatment of an increasingly important area—clinical data mining. It focuses specifically on optimizing treatment patterns and uses adverse events in colon cancer as a case study. Again, however, the discussion section is scant and includes little information on dealing with how, where, or when the analysis was applied or if it had any impact. Chapter 7, also more of a discussion than primary research, provides an overview of the pharmaceutical supply chain. It compares structures, technologies, and delivery mechanisms for drug administration between distributers and patients. I found Chapter 8 to be the most satisfying section of the book, because it looks at automated nurse scheduling in a Swedish hospital and begins to address the issue of context with good evidence of a collaborative research approach, engagement with health services staff, and implementation issues.

In general, this volume’s clear weakness is its fragmented and patchy nature. It lacks an extensive overview of the field, which would give an overall context to its varied contributions. Such a discussion should clearly outline the specific challenges of applying optimization in healthcare, highlight distinctive features of the domain, list the key areas of implementation, and outline the alternative methods available. Given that the range of applications the book covers is far from exhaustive, this discussion is particularly important.

In common with much research literature in this field, little in this book confronts the central challenges of implementing systems analysis in healthcare or addresses in detail the collaborative context essential to a successful and effective application. For example, I could find little in the text that touched on an evaluation of the work described.

The book will be of some interest to researchers who apply these methods in healthcare, especially in the specific areas covered in the chapters. Each chapter contains generally good lists of references so that the reader can do additional research on particular topics. Beyond this, however, I feel that the volume fails markedly to live up to the promise of its title.

Martin Pitt

University of Exeter Medical School, Exeter, United Kingdom, [email protected]

Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie, or Die

Siegel, Eric. 2013. Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie, or Die. John Wiley & Sons, 320 pp. $28.00.

According to its author, “This book is about the most valuable achievements of computerized predictions, and the two things that make it possible: the people behind it, and the fascinated science that powers it” (p. 3). Further, he defines “predictive analytics (PA) as the technology that learns from experience (data) to predict the future behavior of individuals in order to drive better decisions” (p. 11). He describes machine learning as “computers automatically developing new knowledge and capabilities by furiously feeding on modern society’s greatest and most potent unnatural resource: data” (p. 3).

Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie, or Die is an interesting and useful book that illustrates the value of analyzing the huge volume of available data and discovering statistically significant patterns that can lead to better decisions. No doubt exists about the value of such discoveries and their considerable contribution to many diverse fields of endeavor, which the author summarizes in tables in Chapter 3. My problem is that the book is overoptimistic about the contributions of PA as a new field of endeavor and its serious limitations. Let me explain.

Many statistics courses often start with the birthday problem. The instructor asks for students willing to bet, for example, $50 that at least two persons in the classroom have the same birthday. Several students usually take the bet because they miscalculate the probability of such an event and believe that it is unlikely. When the class size exceeds 50 students, the instructor usually wins. For those who do not understand probabilities, the instructor’s predictive power seems inexplicable. But such power is not mysterious; given the number of students in the class, a simple statistical formula can provide the exact answer. The same is true about PA. PA models are based on empirically estimated equations determined from historical data. Such models cannot exhibit either machine learning or meta learning; they simply reflect historical information and are effective for as long as such information remains valid.

The idea that a simple statistical equation, model, or decision rule, based on empirical data, can outperform human judgment goes back to the middle 1950s when Meehl (1954) was published. A large number of studies have since confirmed his finding, thus opening the door to the widespread use of empirically based regression equations, models, or decision rules as a more accurate alternative to unaided, inconsistent human judgment. Using such decision rules, however, has four problems:

• They are based on the average, and although, as the author points out, there are more white swans than black swans, a single black one can be catastrophic and wipe out all the benefits of the innumerable white ones (Taleb 2010).

• They assume independence, which can create serious problems in the case of a meltdown, such as the 2008–2009 major recession when everything when wrong.

• They are incapable of modifying their decisions in the face of environmental or other changes, because they are incapable of realizing that such changes have taken place. Only humans can recognize the need to reestimate the model’s parameters and (or) modify its variables.

• Because large numbers of firms and institutions use them, they are fast becoming competitive requirements instead of competitive advantages. Total quality was a strong competitive advantage two decades ago; it is now practiced by all firms and provides no competitive rewards.

A major theme of Predictive Analytics is that data are growing at an incomprehensible exponential speed, actually doubling every three years; therefore, the most valuable action to take is to learn how to use these data to make predictions, as many organizations have already done. Netflix and dating services are great examples. The book describes how to build PA models and talks about their advantages and limitations, concluding that unless firms learn to exploit such advantages and overcome their limitations, they will not be able to survive. PA starts by conducting a broad, exploratory analysis, testing many predictors. In so doing, it uncovers surprising findings; for example, vegetarians are less likely to miss their flights.

The idea of PA is not new. It started with step-wise regressions; it was then called data mining and later number crunching (Ayres 2007). The interesting question is always how to identify and profit from consistent patterns in available data while avoiding spurious correlations; however, there is a big difference between identifying such patterns and using them for predictive purposes. The book mentions that terrorists do not buy life insurance; however, such a finding is valueless because it would be trivial to buy life insurance to fool investigators. It may sometimes be more advantageous to go against the recommendations of PA, particularly if the great majority of industry players based their decisions on such recommendations. Strategically, for example, it would make more sense to concentrate on a customer segment that is excluded when mortgage decisions are made by a PA model that most banks use.

My view is that Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie, or Die is an interesting book worth reading because it provides many fascinating examples and practical advice on developing PA models. Their predictive value is restricted, however, to many small decisions based on the average behavior of the crowds (Surowiecki 2005). Silver (2012) is an excellent book to read to understand what can and cannot be predicted and the advantages and limitations of modeling. No matter what PA can contribute, users must always be skeptical because predictions can always go wrong (Makridakis et al. 2009).

Spyros Makridakis

INSEAD, Fontainebleau, France, [email protected]

Unrelenting Innovation: How to Build a Culture for Market Dominance

Tellis, Gerald J. 2013. Unrelenting Innovation: How to Build a Culture for Market Dominance. John Wiley & Sons, 310 pp. $34.95.

In Unrelenting Innovation: How to Build a Culture for Market Dominance, Gerald Tellis assembles past innovation research mingled with in-depth business cases to support his thesis: business culture drives innovation. His thesis is admittedly complex to prove. As he explains in Chapter 1, when examining the failure of great incumbent firms, culture is an internal, subtle, and soft factor. Past innovation research has often focused on more testable, tangible factors, such as patents or financial resources; however, Tellis makes an excellent practical argument for culture by exploring incumbent failures among formerly great technology firms. He then transitions to cases of both start-ups that grew to dominate their markets and incumbents that continued to innovate successfully.

As Vijay Govindarajan states in the foreword, “Success breeds complacency, lethargy, or arrogance” (p. xiii). Chapters 1 and 2 discuss innovation failures among incumbents. One of the strengths of this section is its focus on strong incumbents that had excellent innovation histories. Their eventual innovation failures cannot be tied to factors such as lack of resources, intellectual property, or management talent. The so-called incumbent’s curse (Rajesh and Tellis 2000) often comes into play, where incumbents fail to focus on the future, are averse to risk, or do not cannibalize successful products. Chapter 2 goes into detail on how alternatives to cannibalizing a firm’s successful product fail. For example, when incumbents acquire firms with successful technology (as opposed to making their own), their stock prices suffer. Tellis also employs industry histories to prove that the evolution of a new technology is not limited to the classic technology S-curve. The older technology can counter with higher performance on its technology curve after the newer technology has initially surpassed it, or a newer technology can improve drastically after it seems to have hit technology limits.

Chapters 3 and 4 expand on embracing risk and focusing on the future, respectively. Tellis points out that missed innovation opportunities (Type 2 errors) do not initially seem costly, thus creating a bias toward saving money by aggressively screening innovations. His examples of Amazon, Toyota (with the Prius), Facebook, and Federal Express demonstrate how risky these now commonplace ventures seemed in their beginning stages. Chapter 4 points out psychological biases that hinder future planning, such as the hot-hand, paradigmatic, and commitment biases. Particularly interesting was how, in this information age, the availability bias (Tversky and Kahneman 1974) is a problem for innovation managers. The more informed managers are, via today’s news on current innovation, the less likely they are to prepare for the future. Chapter 4 also explains how to use information to track emergent customers for new trends.

In Chapters 5–7, Tellis explains traits of successful innovation cultures: providing incentives for enterprise, encouraging internal markets, and empowering innovation champions at all levels of the firm. Chapter 5 contains many anecdotes about failure and explains how firms can vary the type of incentive (moral, social, and monetary) to elicit more innovative behavior from employees. One lesson to take away from the chapter is the need to regularly review incentive programs. Both GM and IBM are cited for starting with well-intentioned incentive programs that devolved or corroded over time. Chapter 6 describes how internal markets should lead to idea generation, bigger talent pools, and more efficient resource allocation. One section of this chapter cautions about potential negative outcomes of markets and gives examples of internal markets, such as divestitures and research contests. Chapter 7 outlines the characteristics of innovation champions; they have a vision for the future of mass market, are mavericks, have the conviction to persist even when success seems unlikely, and are willing to take risks. Tellis points out how champions may also emerge from the bottom, citing the Google’s associate product manager program as an example. Some additional examples of how younger and (or) less well-placed employees have championed innovation would have been helpful here.

The final chapter will be of particular interest to academics. Tellis compares his thesis of culture driving innovativeness with several other competing factors. For example, in examining disruptive technology as a factor, his research shows that incumbents are more likely than entrants to introduce disruptive technology. The book concludes with a study of worldwide firms, which shows that culture has a greater effect on innovativeness of firms than the other competing macro and micro factors.

In terms of potential follow-up research on the book’s themes, a further explanation on the role of conflict in enabling successful innovation culture seems intriguing. Innovation failures at Kodak and Sony seem partly the result of an overemphasis on consensus, while successful innovators such as Toyota seem more willing to encourage debate. In addition, the book’s emphasis on larger incumbent firms leaves some room to explore whether innovation culture has as great an effect on small firms transitioning out of their nascent phase. Finally, studying how to manage information to properly update and alter firm culture seems promising.

This book is well positioned to benefit both practitioners and academics. Its examples and clear language make it highly readable. I highly recommend the final chapter, which offers an intriguing direct comparison of culture with other well-respected innovation factors espoused by Schumpeter, Christensen, Diamond, and others, for further study. The assembled cases also offer thoughtful content for potential classroom discussions. I enthusiastically recommend this book.

John N. Angelis

Department of Decision Sciences, Saunders College of Business, Rochester Institute of Technology, Rochester, New York, [email protected]