What Industry Needs
Jon R. Kettenring
Bell Communications Research
Industry needs holistic statisticians who are nimble problem solvers. They need to be able to work smoothly on teams and to communicate effectively about their work. Many of industry's statistics problems are interdisciplinary and involve exciting challenges in unplowed areas. Industry needs people who thrive on such opportunities.
The industrial world continues to change at an amazing pace. In telecommunications, for example, it seems that the basic parameters of the business are in constant flux. The information age is here, but exactly what does this mean?
Along with the uncertainties are immensely challenging and exciting problems that need serious work, and much of this work is statistical in nature. Examples close to home include network reliability, software quality, and dealing with massive data sets. Generally speaking, there is a need to transform data into information (Nair and Pregibon, 1993) and intelligence that cuts across much of modern industrial life.
The major problems are usually of such complexity that progress is best made when they are tackled by cross-organizational interdisciplinary teams. Since many of these problems cry out for statistical thinking, it is natural for statisticians to be full partners on the teams. However, to be successful, statisticians need not only to bring their statistical expertise to the table but also to integrate themselves effectively with the rest of the team. Ultimately, the only thing that really matters from the industrial perspective is not the statistics per se but the impact of the total effort on the problem at hand. (See Duffy (1993) for a similar perspective.)
With this background in mind, it should be clear why industrial employers are particularly interested in hiring statisticians with a holistic view of the world as well as relevant and highly honed technical skills. Statisticians who can serve as bridge scientists (Lowrence, 1985, p. 7) are especially cherished.
The title of this symposium and the text of this paper reflect the presumption that a statistics education should be interdisciplinary. This concept is discussed in the next section, which is followed by an airing of the question of how a modern interdisciplinary statistics education can serve the needs of industry and by concluding comments.
Interdisciplinary Statistics Research and Education
Literature on interdisciplinary research is helpful for clarifying what is involved in interdisciplinary activities. Porter and Rossini (1986) distinguish this form of research, which involves substantial integration across disciplines, from multidisciplinary research, which is less integrated, and cross-disciplinary research, which implies only that two or more disciplines are involved.
The concept of interdisciplinary research is itself controversial. Caudill and Roberts (1951) point out numerous pitfalls. Seyle (1975, pp. 176-177) discusses how difficult it is to achieve interdepartmental scientific cooperation in the university. Bauer (1990) argues that interdisciplinary work involves surmounting cultural differences and is basically intractable.
However, there is no reason to be so pessimistic concerning statistics and its involvement with other disciplines. The starting point for the IMS Panel's report on cross-disciplinary research in the statistical sciences (IMS Panel, 1988) is the argument that ''the driving force behind the development of modern statistics has been the need to solve practical problems." No doubt much of this work has been cross-disciplinary rather than interdisciplinary in the full integrated sense. Still, the emergence of the fields of biostatistics, econometrics, and psychometrics exemplifies involvement that has gone "all the way." Gnanadesikan (1990), while raising concerns about how vigorously cross-disciplinary opportunities in statistics are being pursued, points out opportunities in several areas, for example, molecular biology. The mission of the recently formed National Institute of Statistical Sciences is to provide a structure for cross-disciplinary research involving universities, industry, and government. In short, historical momentum, current opportunities, and some supportive organizational structures could all lead to acceleration of collaborative activities between statistics and other areas.
Within the university, there appears to be increased interest in building bridges between disciplines. Gardiner (1987) observes that leading universities are changing their emphasis from information gathering to information processing, which would certainly include a major role for statistics. He argues that interdisciplinary research groups, distinct from the usual discipline-based departments, are a vehicle for fostering such work.
Against this background, it is clear that there are, or can be, creative environments at the university that will support both cross-disciplinary or interdisciplinary research and educational activities in the field of statistics. Impediments, such as tenure concerns, are also recognized; recommendations for dealing with some of them are laid out in the IMS Committee's report on cross-disciplinary activities (IMS Committee, 1990).
The degree of integration across disciplines that is achieved in the educational process is probably less important than the commitment to teach statistics in a way that emphasizes teamwork, communication, and solving real problems. Students who have these values and associated skills ingrained in them are the most likely to be successful in modern industry.
What Needs To Be Done?
The potential for defining a successful holistic statistics education program—one that is right for the needs of industry, in particular—is within the grasp of most statistics departments.
The pieces are already in place in many cases, others are being experimented with, and still others have already been proposed. Departments that are unsure about how to proceed can obtain useful advice by surveying graduates who are working in industry. What may be most difficult is finding the commitment and resources to "put it all together." Since there is so much agreement on the need for change (for example, see Hogg, 1991), and those programs that lag behind are likely to find it increasingly difficult to attract and place good students, the chances of success would seem to be high.
Basic building blocks of any program should continue to be well-rounded statistical knowledge including subjects such as data analysis, statistical computing, sampling, linear models, experimental design, time series, multivariate analysis, and so forth. To make these courses really click, one suggestion is to make sure that they are interspersed with as much real data experience as possible (Singer and Willett, 1990; Cobb, 1991). A related suggestion is to assign instructors to these key courses who themselves have had substantial real experience with data (Gnanadesikan and Kettenring, 1988). Would a medical student want to learn surgery from professors who have never done it? These suggestions take on special urgency for courses on data analysis. For these courses, data-savvy instructors are essential!
Makuch et al. (1990) argue that an "optimal curriculum" for preparing statistics students to work in industry should consist of "topics from statistics, mathematics, computer science, and their applications." This is excellent generic advice. More attention needs to be given to correcting the imbalance between mathematics and computer science stemming from historical emphasis on the former. Makuch et al. go even further, suggesting that a graduate-level dual major in statistics and computer science would best prepare students for industry. The main point is clear: computing and related skills are vital for contemporary industrial — and probably all — statisticians.
Many statistics programs now have courses on and opportunities for consulting. Such activities are surely worthwhile, but they may not go nearly far enough in terms of giving students genuinely collaborative problem-solving experiences. In this sense, opportunities for students to work with interdisciplinary centers on the campus may be more fruitful and even lead to path-breaking cross-disciplinary thesis problems.
For some programs, closer ties with local industry may be a way of building an environment for experiential learning, a concept that has been emphasized by Snee (1993) and others. Opportunities for increased university-industry interactions seem to be wide open and can take many directions. Faculty and students both may be able to join problem-solving teams in industry. Industrial statisticians may be able to identify hot research topics for university statisticians to tackle. Experiences drawn from work with local industry should help faculty to enliven their courses. Whatever routes are traveled, increased contact between university and industrial statisticians should yield useful payoffs for both groups.
It is sometimes said that statistics students with subpar communications skills should take jobs in industry because they would not do well in the classroom. Unfortunately, they are unlikely to do well in industry either. In fact, there has been repeated discussion and concern about how serious a problem this is for industrial statisticians. McDonald (1988) emphasizes this in the context of communicating with management. Hoadley and Kettenring (1990) focus on the communications gap between statisticians and engineers or physical scientists, and explore many of its dimensions. Caulcutt (1987) raises similar issues from a British perspective. One
is left with the impression that there has been very little attempt by statistics programs to respond to this very serious and legitimate concern. Statistics students must somehow learn to give an effective talk, to write a report that managers can understand, and to generally cultivate their interpersonal skills for statistical consulting and cross-disciplinary collaboration.
Industry needs — perhaps more than it ever has — nimble problem solvers with first-rate statistical skills. Solid statistical thinking is required for integrated solutions to many of industry's toughest problems. Statistical training should be truly holistic because that is the way statisticians are expected to operate in the real world.
To reach this state in university statistics education, closer working ties with industry would be worthwhile. This need not be painful. In fact, the results could well be renewed vigor between sectors that are sometimes unnaturally far apart.
Experiences along these lines should be shared so that progress can be made as quickly as possible. For example, Rutgers University has started a dialogue with statisticians in local industry about ways of working more closely together. What are they learning that would be of benefit to others?
The spirit of interdisciplinary work should shine through the education process. In both industry and the university, "as in farming, the most fertile soil may be that under the fences rather than at the center of long-established fields" (McHenry, 1977, p. ix).
Thanks to Ram Gnanadesikan, Innis Sande, and Vijay Nair for helpful comments.
Bauer, H. H. 1990. Barriers against interdisciplinarity: implications for studies of science, technology, and society (STS) . Science, Technology, and Human Values 15:105-119.
Caudill, W., and B. H. Roberts. 1951. Pitfalls in the organization of interdisciplinary research. Human Organization 10:12-15.
Caulcutt, R. 1987. Statistics in industry: a failure of communication. Statistician 36:555-560.
Cobb, G. 1991. Teaching statistics: more data, less lecturing. Amstat News 182:1,4.
Duffy, D. E. 1993. Comment on "Is industrial statistics out of control?". Stat. Sci. 8(4):380-384.
Gardiner, J. J. 1987. Building bridges between the disciplines: collaboration in an information processing university. Eur. J. Eng. Educ. 12:121-127.
Gnanadesikan, R. 1990. Looking ahead: cross-disciplinary opportunities for statistics. Am. Stat. 44:121-125.
Gnanadesikan, R., and J. R. Kettenring. 1988. Statistics teachers need experience with data. Coll. Math. J. 19:12-14.
Hoadley, A. B., and J. R. Kettenring. 1990. Communications between statisticians and engineers/physical scientists (with discussion). Technometrics 32:243-274.
Hogg, R. V. 1991. Statistical education: improvements are badly needed. Am. Stat. 45:342-343.
Institute of Mathematical Statistics (IMS) Committee. 1990. Report on Cross-Disciplinary Activities. IMS Bull. 19(6):716-719.
Institute of Mathematical Statistics (IMS) Panel. 1988. Cross-Disciplinary Research in the Statistical Sciences. (Reprinted in Stat. Sci. 5(February):121-146, 1990.)
Lowrence, W. W. 1985. Modern Science and Human Values. New York: Oxford University Press.
Makuch, W. M., G. J. Hahn, and W. T. Tucker. 1990. A statistical computing curriculum to meet industrial needs. Am. Stat. 44:42-49.
McDonald, G. C. 1988. Communicating with management. Chance 1:42-44.
McHenry, D. E. 1977. Academic Departments: Problems, Variations, and Alternatives. San Francisco: Jossey-Bass.
Nair, V. N., and D. Pregibon. 1993. Comment on "Is industrial statistics out of control?". Stat. Sci. 8:391-394.
Porter, A. L., and F. A. Rossini. 1986. Multiskill research. Knowledge: Creation, Diffusion, Utilization 7:219-246.
Seyle, H. 1975. From Dream to Discovery. New York: Arno Press.
Singer, J. D., and J. B. Willett. 1990. Improving the teaching of applied statistics: putting the data back into data analysis. Am. Stat. 44223-230.
Snee, R. D. 1993. What's missing in statistical education? Am. Stat. 47:149-154.