A Larger Perspective

John C. Bailar III

McGill University

As academic statisticians, we are missing the boat. We are barking up the wrong tree. We do not see what is plainly before us. We are kidding ourselves when we think that "our" kind of statistics is vital to the welfare of the nation and the world.

More and more, despite occasional appearances otherwise, we as academic statisticians are talking to ourselves. Even at this symposium we talk about how to do the old things better and more broadly, not about what we could offer to society, and what most needs to be done. Think about the whole range of the really big problems of the day: violence, crime and criminal justice, education and industrial productivity in the broadest senses, unemployment, the balance of trade, federal deficits, the health and welfare of millions of disadvantaged persons, urban rot, racial and ethnic tensions, and homelessness.

The kinds of statistics that we teach in undergraduate and especially in graduate programs have almost nothing to contribute to anything that matters on the scale of these problems. Instead, we teach about new abstractions in statistical theory, or we teach about new applications of theory to what are, in this context, tiny problems with tiny generalizations and tiny implications. I would certainly agree that every graduate student in statistics needs very substantial training in statistical theory, partly because of important applications, but partly also to recognize the many situations in which some theoretical development is not appropriate. In my experience, the latter has been more common than the former.

My own field, biostatistics, may illustrate a trend. Biostatistics began as an effort to break away from the sterile patterns of academic statistics, to develop and support practitioners who understand the problems and who contribute substantially to their solution. However, for many years biostatistics has been drifting back into this same kind of introspection, the navel gazing, the almost exclusive focus on theory, often theory that has no conceivable, important, real use. We fiddle while Rome burns.

Then we wonder why the world passes us by. Why is it that the economists, for example, are often on the front pages of our newspapers, or testifying to Congress, or making major decisions about public policy, but not the statisticians? Is it that we do nothing on that scale of importance, nothing that merits that kind of attention? I fear that the answer is "yes."

We teach what we enjoy teaching and what we know how to teach, not what the world needs. Think about that litany of problem areas I just recited. The solutions to those problems could profit enormously from sound statistical data, soundly analyzed. But the difficulties that block our understanding on these problems have little to do with probability models or random variation or all those other good things that make up statistics in academia today. They have to do, rather, with the vast range of other kinds of uncertainties, that is, what we broadly call bias. Bias dominates randomness almost everywhere. Think about your own past training and the training that many of you now deliver to new generations of students. What fraction of that training is or was devoted to bias? What fraction deals in any direct way with the big problems of this year?



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 18
Modern Interdisciplinary University Statistics Education: Proceedings of a Symposium A Larger Perspective John C. Bailar III McGill University As academic statisticians, we are missing the boat. We are barking up the wrong tree. We do not see what is plainly before us. We are kidding ourselves when we think that "our" kind of statistics is vital to the welfare of the nation and the world. More and more, despite occasional appearances otherwise, we as academic statisticians are talking to ourselves. Even at this symposium we talk about how to do the old things better and more broadly, not about what we could offer to society, and what most needs to be done. Think about the whole range of the really big problems of the day: violence, crime and criminal justice, education and industrial productivity in the broadest senses, unemployment, the balance of trade, federal deficits, the health and welfare of millions of disadvantaged persons, urban rot, racial and ethnic tensions, and homelessness. The kinds of statistics that we teach in undergraduate and especially in graduate programs have almost nothing to contribute to anything that matters on the scale of these problems. Instead, we teach about new abstractions in statistical theory, or we teach about new applications of theory to what are, in this context, tiny problems with tiny generalizations and tiny implications. I would certainly agree that every graduate student in statistics needs very substantial training in statistical theory, partly because of important applications, but partly also to recognize the many situations in which some theoretical development is not appropriate. In my experience, the latter has been more common than the former. My own field, biostatistics, may illustrate a trend. Biostatistics began as an effort to break away from the sterile patterns of academic statistics, to develop and support practitioners who understand the problems and who contribute substantially to their solution. However, for many years biostatistics has been drifting back into this same kind of introspection, the navel gazing, the almost exclusive focus on theory, often theory that has no conceivable, important, real use. We fiddle while Rome burns. Then we wonder why the world passes us by. Why is it that the economists, for example, are often on the front pages of our newspapers, or testifying to Congress, or making major decisions about public policy, but not the statisticians? Is it that we do nothing on that scale of importance, nothing that merits that kind of attention? I fear that the answer is "yes." We teach what we enjoy teaching and what we know how to teach, not what the world needs. Think about that litany of problem areas I just recited. The solutions to those problems could profit enormously from sound statistical data, soundly analyzed. But the difficulties that block our understanding on these problems have little to do with probability models or random variation or all those other good things that make up statistics in academia today. They have to do, rather, with the vast range of other kinds of uncertainties, that is, what we broadly call bias. Bias dominates randomness almost everywhere. Think about your own past training and the training that many of you now deliver to new generations of students. What fraction of that training is or was devoted to bias? What fraction deals in any direct way with the big problems of this year?

OCR for page 18
Modern Interdisciplinary University Statistics Education: Proceedings of a Symposium When a "statistician" does take on the big problems and big programs, that person is very likely to be someone who has little or no formal statistical training. Ask yourself, for example, about the leadership of the big federal statistical agencies — somebody mentioned a dozen; I might count as many as 15. These include, for example, the Bureau of the Census, the National Center for Health Statistics, and the Bureau of Labor Statistics. How many of these 15 agencies have as directors someone who claims statistics as his or her primary discipline? I think of just two. How many directors would benefit from a deep understanding of statistics at the highest levels? Is it less than 15? For many years, the person who held the title of Chief Statistician of the United States was not a statistician, but an economist. (I refer here to a prior incumbent; the person who now holds this office, also trained as an economist, is in fact a very accomplished statistician.) I said that we talk about how we could do the old things better rather than how we might be doing the new things that need to be done. I discussed this matter briefly with Wayne Fuller of Iowa State University, who reminded me that if we go even further back in our history, our intellectual grandparents and great-grandparents in statistics did indeed devote themselves to these big problems, and good analysts found ways to use flawed and incomplete data to derive sound conclusions and practical recommendations. But that tradition has been allowed to slip into the hands of other disciplines. For example, it has been epidemiologists rather than statisticians who have spent much effort in recent years on two areas critical to statistical analysis. One is understanding the nature of confounding and the effects of efforts to reduce its influence. The other is developing a taxonomy of bias. This taxonomy has some very important, big, practical implications. Work in both of these areas seems to be almost unknown to academic statisticians. Again, it is other disciplines that have tackled the thorny issues of teaching persons who may be highly expert in some area of application but have the equivalent of grade school training in statistics about the use, sometimes even the correct use, of powerful computational and inferential tools. We have not wanted to get our hands dirty. Jon Kettenring has mentioned here a set of five points that he considers important; I agree on all of them. One point was the need for data-savvy instructors. Three of the other four also rotate around this need for an understanding of what the world is truly all about. What do we do at present in academic settings to attract instructors who are data-savvy? What do we do to nurture them? What do we do to retain them when it comes time to consider tenure? I have not said that current areas of great interest in statistical theory are unimportant or that our training is useless. Those areas are important and our training in those areas can be vital, but they cover only a small fraction of what needs to be done. For a moment now, let us think in top-down fashion. How many academic statisticians of the 1993 variety does this country really need for the two essential purposes of replicating themselves and producing a continuing flow of new theoretical results? Then ask, How many academic statisticians do we have? It is my assessment that the present supply is already a lot greater than the demand; that the total professorate substantially exceeds what is needed for the really essential tasks of what is now academic statistics. Conversely, that professorate is far short of what we would need to take on the job I have outlined.

OCR for page 18
Modern Interdisciplinary University Statistics Education: Proceedings of a Symposium Are you with me so far? Do you think I am talking about your neighbor? I am talking about myself, and I am talking about you. I have heard several questions asking speakers how they would solve problems that they have noted, and it would be quite fair to ask me the same questions. One thing we should do in the academic setting is to focus far more than at present on inference in the face of bias, sometimes serious bias. It was precisely this kind of concern that got me into the quantitative study of health risks due to hazards such as asbestos or dioxin (my current area of statistical interest, and it is statistics). In such risk assessment, uncertainties commonly range over three or more orders of magnitude. That is real uncertainty, and it is virtually all from bias. It is not going to matter much whether an animal experiment produces 20 versus 22 versus 25 cancers among 50 animals when you deal with that sort of uncertainty. Another recommendation is to use big, real, important examples to teach theory. Any instructor who makes up a set of data to illustrate something has failed the students. If you cannot find a real example, there is something wrong with what you are trying to teach. We are going to have to make cuts in the present curriculum. I am sorry; everything we now teach is surely important, but some things are more important than others, and we need to consider priorities. I do not think anybody else here has specifically brought out the need for dropping large parts of the things that we have traditionally taught, but there is not room for everything. Perhaps we all have a wish to teach our students everything we have learned over a period of, say, 30 years; it might be better to understand that they can and will continue to learn from their professional experiences, just as we did and still do. We really do not have to teach them everything we know. We need to provide students with a substantially increased scope for electives. The curriculum that I recently inherited as chair of a department was chock-full, so the students had virtually no scope for electives, and that is one of the first things I am changing. Maybe you will look at your own curricula to see whether there is something that could be done about this, even to the point of allowing and encouraging graduate students in statistics to take more than occasional courses in substantive areas. How about giving our students encouragement and support to spend a summer, or a term, or even a whole year in supervised work in some applied operation in government or industry? Some programs do this now, but not nearly enough. My last point is that we must actively seek out and hire (and then give tenure to) statisticians who are especially good at doing and teaching applications. Will you give your next academic ticket to someone who has published half a dozen trivial papers in some annals, and will lead students who will do more of the same, or to someone who has used the tools of statistics to transform a national debate and who will lead students to do more of that? Will these suggestions produce a different kind of graduate? Of course they will. That is the point of it all.