Shaping Statistics for Success in the 21^st Century:
The Needs of Industry

Gary C. McDonald, Operations Research Department
General Motors Global Research and Development Operations

The title of this article indicates a scope clearly too broad to cover in its entirety. I will offer my view on the topic, based on my years of experience in the automobile industry, supplemented with input from my colleagues. While this background is primarily manufacturing based, it has elements in common with other industries. I will rely on the reader to make the appropriate associations and hopefully stimulate thought and discussion which will broaden and enrich the ideas I present. Looking into the future is exciting and provides and opportunity to take stock of many good things that have been accomplished in the past and to build enthusiasm for the new horizons. An interesting dialog on the future of statistics was held in 1985 to commemorate the 25^th anniversary of the founding of the Department of Statistics at the University of Wisconsin-Madison (Hill et al. (1988)). It’s interesting to check on how well our forethought was at that time, and if today’s assessment bears any similarity with that rendered earlier.

It is very difficult to forecast accurately the state of a technology, an industry, or a profession. I will not attempt to do so here. Alvin Toffler (1970), in a stimulating book about the future, described the accelerating rate of changes taking place in society, primarily due to technology. His description is more applicable today than it was almost three decades ago when the book was written. The personal computer is about two decades old and has had profound influence in all of our professional and personal lives. It would have been impossible to systematically forecast the expansive impact of that technology, say, three or four decades ago. In this discussion, I’ll look at some of the key elements in industry, categorized according to one taxonomy, as I know them today. Then I’ll comment on the trends which I sense evolving in industry and how they might shape the challenges and opportunities into the near future. The emphasis will be primarily, but not exclusively, on the statistical areas surrounding these trends. A few areas will be mentioned which have been and will be very important for industry and, to the extent of my knowledge, have not had much interaction with statistical sciences. These areas are important and might be ripe for some type of statistical interplay.

As shown in figure 1, the taxonomy for industrial activities around which I will organize my comments consists of six elements: Design; Engineering; Validation; Manufacturing; Sales & Service; and Consumer Research. I expand on each of these later in the paper by indicating what are current driving forces in these activities today, and what are significant trends which will affect these activities as we approach the 21^st century.

Figure 1. Taxonomy of Industrial Activities

A major goal for industry is to shorten lead time- the time from when a new product or service is conceived to when it is delivered to the customer. This is forcing companies to create faster development processes. This goal is driven by the competitive need to respond to market demands faster than the competition, to capture initial higher margins, and to reduce product development expense. There appears to be a high correlation between time in development and development expense. More demanding rates of change for technology in components of a product (e.g., electronic and computer based) are also driving the need for faster development times for products incorporating such components. Clark and Fujimoto (1991) describe the development of new products in the turbulent, demanding and highly competitive automotive world. Many of their findings are recommendations, however, also apply to other industries. Many of the forces driving the automotive industry to re-map their business processes are universal in their impact.

The major enabler for achieving this shortened lead time goal is information technology (IT) in its broadest sense. IT is a system with many hardware and software components. Hardware is the physical component of the system: computers, terminals, display systems, communication systems, physical interfaces to machines and people. Software is the broad array of programs that run on the computers. IT is the enabler which will lead to less reliance on computer simulation and math modeling for product development. It will permit real time measurement of every critical step in a manufacturing process. It will enable virtual builds of a product rather than physical builds- thus accelerating customer clinics for evaluation and guidance. These are just a few of the areas which will lead to a reorientation in the way that statistical sciences will be used.

The National Research Council (1995) has recently delineated an exciting research agenda around information technology for manufacturing and has identified many opportunities. In addition to the technology impact on industry, globalization of business opportunities and competition will continue to put pressure on product quality, cost, and affordability. Goldman et al. (1995) describes a new industrial order, deemed agile competition, evolving from the convergence of technology and competitive forces. Many leading companies today are developing business processes, agile competition processes, to respond effectively to uncertainty in consumer demand and business conditions. Statistical concepts are inherent in such processes.

In this section, the six taxonomy elements of Figure 1 will be discussed with respect to a few of the current driving forces and the trends which will affect these elements in the near future.

Design activities in the past have revolved primarily around physical models- models that could be seen, felt, tested, and modified. In the automotive industry such models are generally sculptured out of modeler clay. After the physical model is approved it is then digitized and represented with a mathematical formula. Computer-aided design (CAD) systems are beginning to play a significant role in this process. As we move to the 21^st century, CAD will play an even greater role eventually reversing the role of math model and clay model. Rather than developing a physical model to extract eventually a math model, the CAD system will be good enough to generate a computer model for approval and the subsequent cutting of a physical model will be for validation of design intent. Virtual reality computer systems will be playing an increasingly important role in developing quickly and inexpensively themes and concepts to be evaluated in consumer clinics and to be used in subsequent engineering analyses. Recently, The Econonmist (1997b) highlighted research initiatives leading to ever more ingenious ways of Making things inside computers before they are made in reality and thus cutting costs and speeding up development and production.

Engineering activities have been utilizing computer-aided engineering (CAE) methods, many of which are finite element type models, engaging in concurrent engineering (Prasad (1997)) primarily with the manufacturing activities, and implementing Design-for-Manufacturability (Anderson (1990)). This has moved engineering to a more systems oriented approach to the benefit of the entire enterprise. Regression modeling, response surfacing methods, and optimization methods have played a very significant role in these activities. Moving to the 21^st century, there will be a much greater emphasis on robust design of engineering specifications. DeVor et al. (1992), Chapter 21, provide a collection of case studies that will illustrate the value of the robust design approach especially when coupled with an analytic model for the performance measures of interest. The output will be product engineering specifications which are tolerant of the variability resident in the manufacturing processes. Likewise, there will be greater use of and need for mathematical modeling capability which incorporates variability of material parameters, geometrical specifications, and manufacturing conditions. Systems engineering will increase in importance as supplier leveraging and integration further affects redesigned business processes.

Validation, in the past, has been dominated by physical testing. This has involved both laboratory testing, primarily for components and sub-assemblies, and field durability testing, primarily for overall system validation. In this area, accelerated testing methods have been used extensively. Classical statistical estimation techniques, including reliability modeling, and design of experiments (DOE) methods are plating a very critical role. As we move forward, elements of physical testing will be replaced by computer simulation. Statistical and mathematical modeling will be increasingly challenged to effectively validate systems, rather than components and sub-assemblies, especially those that relate to expanded environmental and safety requirements which may extend for the lifetime of the product and beyond (e.g., re-cyclability). DOE will be used increasingly to guide computer analyses rather than physical testing and to develop strategies to validate underlying math models (see, for example, Sacks et al. (1989a and b)). The geometrical aspects of DOE for covering the design space fit well in this context; however, the notion of experimental error based on experimental units doesn’t fit so well.

Manufacturing has and is benefiting substantially from the use of statistical DOE, statistical quality control and sampling techniques, capability indices incorporating measures of variability, and the systematic use of quantitative and qualitative methods resulting in continuous improvement. Much effort has been expended over the last decade to learn and implement principles of lean manufacturing so well developed by Japanese manufacturers (Womack et al. (1990) and Jones (1992)). As we move to the 21^st century, there will be greater use of continuous measurement technologies leading to greater opportunities for real-time scheduling. As measurement, data collection, and sensor technology advances expanded applications for signal processing and control systems will emerge in the engineering, manufacturing, and service elements. Manufacturing throughput and response time will be key drivers increasing the need for and use of appropriate queing models (Goldratt and Cox (1992) and Hopp and Spearman (1996)). Agile and flexible manufacturing capability (Kidd (1994)) will become an effective enabler to meet future market uncertainty and drive economic order lot sizes lower and lower (to unity?).

Sales and Service has used statistical methods that have helped in the identification of root causes of problems based on observed symptoms. Thus signal detection techniques, diagnostics methods, fault tree analyses, and discriminant and classification analyses have all been used extensively. Problem solving approaches involving Pareto charts have also been used effectively. The 21^st century will see increased use of expert systems including neural networks (already being used in the area of credit approval). Bayesian analyses should have increasing applicability with the growth in data warehousing of historical information. Data mining, along with data warehousing, will be tested extensively in this area (Verity (1997)) as well as in the area of consumer research to be discussed in the next paragraph. There are many statistical rules-of-practice that will be challenged in this environment (Coy (1997)), and many opportunities for the statistical profession to guide and lead in this area of technology. For example, with the volume of products sold by General Motors there can be on the order of several million warranty records generated per month thus providing a fertile ground for data mining. Real time inventory control and logistics management will continue to grow in importance as industry adopts more systems thinking (versus local optimization within the business) and has the tools to monitor and track in real time all of the elements in the inventory and logistics chain.

Consumer research has utilized surveys extensively to help characterize the marketplace, identify customer needs, and assess the level of satisfaction with the product and service (e.g., J.D. Power and Associates, Agoura Hills, CA). Consumer clinics and focus groups have been a major component of assessing future consumer response to products and services in the design phase of development. Econometric methods and dynamic models have been used to model the market response to a portfolio of products at specific prices and market conditions. Extensive scenario evaluations are then based on these models to assess robustness of strategies. Future work in this area will utilize more societal and technology trend information. Foe example, building on such trends Coates et al. (1996) provide a fascinating collection of fifteen scenarios of the world in 2025. There is an immense challenge in this area for statistical sciences. There are very large data bases with such information along with global benchmarking assessments. The challenge is to synthesize and focus such information to lead to effective decision making. Companies are striving to understand market segmentation and provide valued niche products. Statistical modeling and computer algorithms that help with such "slicing and dicing" will be in high demand. Emphasis will be placed on qualifying and interpreting relevant demographic data, life style characteristics, and consumers’ future needs with a goal of developing robust business plans. Environmental concerns such as global warming, emissions reduction, and re-cyclability will continue to challenge statistical modeling and forecasting capabilities (The Economist (1997a)). The development and application of methods to detect change points will continue to be very important as new manufacturing processes, new materials, and new technologies globally enter the environment.

I would like to conclude this section with a few comments on information technology (IT). The development of IT is the single most impactful technology to affect industry business practices (National Research Council (1995)). As examples, data warehouses, data mining, intelligent software agents, the internet and its follow-on offer tremendous benefits to industry. Each of these elements do have statistical sounding ingredients which need attention and, in a sense, legitimization. There is real opportunity for large scale interdisciplinary research and applications in this context. The National Institute of Statistical Sciences will, I hope, play a leading role for our statistical profession in developing and obtaining broad based support for such studies.

The preceding assessment has some important implications for the statistical education of those professionals who will enter industry. The educational topic has been extensively addressed by a committee of ASA (1980) and more recently by Snee (1993), as well as many others. I don’t view the needed changes as revolutionary changes. Rather I see it in terms of trends which increase emphasis in some areas and decrease emphasis in others. In this manner, I’ll try to state some observations in the form of how I would change my education (if I could do it over) to match the requirements for the positions I see today. Beyond the specific suggestions to follow, there is a need for the industrial statistician to move from a posture of passive consultant to one of active leadership. I suggest:

• Less emphasis on statistical hypothesis testing and significance levels;
• Less emphasis on measure theory and probability theory;
• Stress deeper subject matter knowledge (e.g., engineering, biology, physics, etc.) in a specialized area of interest;
• Broader topical knowledge rather than deeper knowledge of statistical theory;
• Greater emphasis on visual analyses and interpretation coupled with data analysis;
• Greater emphasis on communication skills;
• Greater emphasis on queuing theory along with stochastic processes and models;
• Greater emphasis on signal processing, pattern recognition, and control systems;
• Greater emphasis on interdisciplinary studies and problem formation;
• Greater attention to organizational systems and processes rather than methodology for specific problem solving;
• Greater emphasis computing skills linked to common business systems (e.g., spreadsheets) and on database issues (e.g., organization and management);
• Greater emphasis on building effective business cases; and
• Further focusing decision theory on real decision making in a business context.

From an industry perspective, I think the value of statistical training is greatly enhanced when it is coupled with expertise in a relevant subject matter. Perhaps we should consider future statistical training to be inherently interdisciplinary and design curricula which lead to joint majors emphasizing subject matter expertise with statistics being a strong value-added component. Bio-statistics is a good example of how such interdisciplinary couplings can be done beneficially. I think this would help in expanding career growth opportunities for the statistician in industry and place he or she in a stronger position to assume higher levels of operational responsibility. The career trajectories of statisticians in industry should be given strong consideration in designing the curricula and training.

The Committee on Science, Engineering, and Public Policy (1995) looked into the educational experiences required for scientists and engineers so that they may contribute effectively to national, scientific, and technological objectives. In their study, they addressed question such as:

• Given present career paths, what are the most appropriate structures and functions for graduate education?
• What should be the nation’s goals for graduate science and engineering education?
• Are we producing the right number of PhDs?

A large portion of the committee’s recommendations deal with enhancing the versatility of education and highlighting the long range disadvantage of overspecialization. They argue for more intimate knowledge of business and commerce and ability to develop and market ideas. They stress need for interdisciplinary PhDs in such areas as environmental science, engineering, and biomedicine. These are areas where the statistical sciences have already embarked on substantive interdisciplinary programs.

This article was prepared from an industry perspective and in the spirit of the leadership secrets described so effectively by Slater (1994) in writing about GE’s Jack Welch. Such "secrets" include: look reality in the eye and don’t flinch; be ready and eager to rewrite your agenda; and don’t get stuck in the past.

The 21^st century holds tremendous opportunities for the statistical profession. The profession’s impact on industry will depend on its ability to build an the rapidly growing world of information technology. The greatest difference between the assessment given in this article and that of Hill et al/ (1988), just one decade ago, is the recognition that information technology is the key to success. The vitality and growth of the profession is strongly linked to it agility, i.e., its ability to contribute leadership to industry and society by building a relevant research base and innovative applications, in a timely fashion, to critical problems.

I would like to acknowledge the helpful discussions and suggestions from many of my colleagues at the General Motors Global Research and Development Operations.

American Statistical Association (1980). Preparing statisticians for careers in industry, The American Statistician, Vol. 34, pp. 65-80.

Anderson, D.M. (1990). Design for Manufacturability: Optimizing Cost, Quality, and Time-to-Market. CIM Press, Lafayette, CA.

Box, G. and Luceno, A. (1997). Statistical Control by Monitoring and Feedback Adjustment. John Wiley & Sons, Inc., New York, NY.

Clark, K.B. and Fujimoto, T. (1991). Product Development Performance: Strategy, Organization, and Management in the World Auto Industry. Harvard Business School Press, Boston, MA.

Coates, J.F., Mahaffie, J.B., and Hines, A. (1996). 2025: Scenarios of U.S. and Global Society Reshaped by Science and Technology. Oakhill Press. New York, NY.

Committee on Science, Engineering and Public Policy (1995). Reshaping the Graduate Education of Scientists and Engineers. National Academy Press, Washington, DC.

Coy, P. (1997). He who mines data may strike fool’s gold, Business Week, June 16, 1997, p.40.

DeVor, R.E., Chang, T-h, and Sutherland, J.W. (1992). Statistical Quality Design and Control: Contempory Concepts and Methods. Macmillan Publishing Company, New York, NY.

Goldman, S.L., Nagel, R.N., and Preiss, K. (1995). Agile Competitors and Virtual Organizations: Strategies for Enriching the Customer. Van Nostrand Reinhold, New York, NY.

Goldratt, E. M. and Cox, J. (1992). The Goal, 2^nd revised edition. North River Press. Inc., Great Barrington, MA.

Hill, W.J., Hunter, W.G., Pfeifer, C.G., Marquardt, D. W., Snee, R.D., McDonald, G. C., Duncan, J.W., and Joiner, B.L. (1988). Statistics – a road to the future, Chance, Vol. 1, No. 1, pp. 38-44.

Hopp, W. J. and Spearman. M.L. (1996). Factory Physics: Foundations of Manufacturing Management. Richard D. Irwin, Inc. Chicago. IL.

Jones, D.T. (1992). Beyond Toyota production system: the era of lean production. In Manufacturing Strategy: Process and Content (Voss, C.A., ed.), pp.189-210, Chapman & Hall, London.

Kidd, P.T. (1994). Agile Manufacturing: Forging New Frontiers. Addison-Wesley Publishing Company, New York, NY.

National Research Council (1995). Information Technology for Manufacturing: A Research Agenda. National Academy Press, Washington, D.C.

The Economist (1997a), A warming world, June 28, 1997, pp.41-42.

The Economist (1997b), The immaterial world, June 28, 1997, pp.86-87

Sacks, J., Schiller, S.B., and Welch, W.J. (1989a). Design for Computer Experiments, Technometrics, Vol. 31, No. 1, pp. 41-48.

Slater, R. (1994). Get Better or Get Beaten! 31 Leadership Secrets from GE’s Jack Welch. Richard D. Irwin, Inc., Burr Ridge, IL.

Prasad, B. (1997). Concurrent Engineering Fundamentals: Integrated Product Development. Prentice-Hall, Inc., Upper Saddle River, NJ.

Sacks, J., Welch, W.J., Mitchell, T.J., and Wynn, H.P. (1989b), Design and Analysis of Computer Experiments, Statistical Science, Vol. 4, No. 4, pp. 409-435.

Snee, R.D. (1993). What’s missing in statistical education? The American Statistician, Vol. 47, No. 2, pp. 149-154.

Toffler, A. (1970). Future Shock. Random House, Inc., New York, NY.

Verity, J.W. (1997). Coaxing meaning out of raw data, Business Week, February 3, 1997, pp. 134-138

Womack, J.P., Jones, D.T., and Roos, D. (1990). The Machine that Changed the World. Rawson Associates, New York, NY.