Home » A Statistician's Life, Additional Features, Influential Mentors, Mentoring, Statisticians in History

Interview with Terry Speed

1 September 2016 7,773 views One Comment
Jean Yang, professor at the University of Sydney School of Mathematics and Statistics interviews her mentor, Terry Speed.
Terry Speed

Terry Speed

Tell us a little about your research interests and university life before your PhD.

At school and as an undergraduate, I always had broad interests, but I wouldn’t call them research interests, just interests. The usual stuff: how, why, when, and where. Science, art, literature, some sociopolitical things (e.g., concerning the death penalty [against]). I was unfocused.

How did you become interested in statistics? Did you always want to pursue a study in mathematics or statistics?

I started university wanting to do medicine. My interest in medicine came from being a sickly kid who spent a great deal of time at home and in hospitals, and hence I had a lot of first-hand experience with the medical world. In university, I aimed for medical research, not medical practice, because I lacked the “bedside manner.” For me, in the late 1950s, bedside manner included lying to the patient. I saw it first-hand with a school friend who died of leukemia. Everyone kept encouraging her to think she would get better, while letting us—her schoolmates—know she was dying. That seemed unfair to me then, and still does.

I lasted one term in my combined science-medicine course before giving up the medicine goal and switching to science only. That involved adding another math course, not dropping anything. I call that failure #1, as it was a change in direction; I’d gotten my goals wrong.

In the second year, I dropped chemistry and physics, and in my third year, genetics, after which I found myself doing just mathematics and statistics. For the next 10 years or so, these two played a roughly equal part in my life.

There was a period during my (part-time) PhD when I thought I might become a high-school mathematics teacher, and so I got trained for that, but ended up not pursuing it, (failure #2). In that 10 years, I tried research in pure maths (failure #3), probability theory (failure #4), and theoretical statistics (failure #5), after which I had a 4.5-year spell in science management (failure #6).

By the time I arrived at Berkeley, I was a statistician with an interest in all applications, including biology, and most particularly, genetics. As I have said elsewhere, fail early, fail often.

Who was your PhD thesis adviser, and who are the academics who influence you the most in your statistical thinking?

My PhD adviser was Peter Derrick Finch. He was a very good researcher in probability, statistics, and mathematics. He did not have a PhD, and was rather proud of that. He had broad interests and got bees in his bonnet—he followed his passions, rather than sticking to one area. I share a few, but not all, of his qualities, though whether the sharing is association or causation we’ll leave to another time.

Other academics who influenced me were people who got me interested in particular topics: MJD White and PA Parsons for genetics; Evan J. Williams for Fisherian statistics; Gordon B. Preston for algebra; Emanuel Strzelecki for functional analysis; K. R. Parthasarathy for things like Lie groups; EJ (Ted) Hannan for stochastic processes; and Dev Basu for critical thinking about classical statistical inference. Dev became a Bayesian, but I resisted, though conceding the validity of many of his arguments.

Behind these live people were four remote geniuses who influenced me at a distance in time and space: John von Neumann (1903–1957); R. A. Fisher (1989–1962) and Norbert Weiner (1894–1964); and Andrey Kolmogorov (1903–1987).

At Berkeley, you were involved as an expert in O.J. Simpson’s trial. Would you describe how you became involved, and had you been interested in forensics before?

I got involved in the Simpson case because I was asked and I said yes. I did so because I believe in innocence until proven guilty and that the prosecution has to prove its case beyond a reasonable doubt. I was asked because I was known to be critical of the way in which statistics was used in DNA forensics, and I still am. This is not the place to explain why in detail, but let me just mention that I find it hard to see meaning or justice in the so-called “match probabilities” of 1 in 1,020 still being presented to courts, when there are between 109 and 1010 people on the planet and the calculation omits any consideration of lab errors such as sample contamination, mislabeling, or the like—errors which are regularly found to occur.

I have had a long-standing interest in the legal treatment of evidence, and in criminology more generally. Fifty years ago, I was an expert witness in a murder trial in Melbourne—that of Ronald Ryan. There, the issue was geometry, the trajectory of a bullet fired during a prison escape. I was asked then because other, more senior mathematicians in the city had said no. Belief in the rights of defendants wasn’t widespread then and still isn’t. Ryan was the last person executed in Australia, and I look forward to the day that can be said about someone in the USA.

Research and Career

Who are/were your mentor(s) at different stages of your career, and what is the most significant lesson you took away from your mentor(s).

Although I am now a big believer in mentoring, I’m embarrassed to say I wasn’t conscious of it playing much of a role in my research career. Maybe I’m down playing it, and if so, apologies to all my mentors out there—if you are still alive! Of course I learned an enormous amount from my collaborators and students, but it seems odd to call them mentors. Mostly, I felt I was trying hard to catch up or keep up with people at the same time as [I was] trying to keep being myself. Perhaps it is just that 50+ years ago, mentoring wasn’t the big deal it is now. Either way, I can’t recount a significant lesson … but then I’ve always felt I am a slow learner. Maybe a lesson is on the way.

Do you believe in strong involvement or ties with the industry? How should an academic go about achieving this?

I have always had a great interest in and respect for industry. As an undergrad, I had a summer job at ICI in Melbourne and learned a lot from the work of George Box, written when he was at ICI in the UK just over a decade earlier. As an academic, I’ve always liked to forge links with government and industry statisticians, in part because they employ our graduates, in part because they have problems we can sometimes help solve, and in part because I’m curious about how our discipline gets used in what I like to call the “real world,” distinct from the “imaginary world” of academia.

While in Perth, I used to take my students out to field experiments laid out by statisticians from the then department of agriculture. It was fun walking around a Latin square or a split plot experiment of fruit trees, and later we would analyze the previous year’s data.

On one occasion, I went up in a very small plane with someone from the local bureau of census who was checking dwellings in remote rural areas for his master list. The idea was that the pilot would swoop down, making a low pass above each possible housing unit, to determine whether it was occupied. My brown paper bag saw a lot of action on that trip, and I concluded I didn’t have the stomach for census work.

How do you achieve it? Be interested and respectful. Show your interest and respect.

Interdisciplinary Research

What motivated you to change from statistics (analysis of variance) to genetics and bioinformatics? Was there an event, a memorial project, or a collaboration that sparked or marked the beginning of your transition?

I fell in love with genetics (including DNA) in my first biology course in 1961 and have never deserted her. In 1963, I wrote an undergraduate honors thesis on R. A. Fisher’s work on the survival of a single mutant gene. There has been no event, and no transition. If, like me, you were a statistician interested in genetics and DNA in the 1970s and 1980s, then with no further effort on your part, you became labeled a bioinformatician. I like to think I didn’t find bioinformatics; it found me.

You must find interdisciplinary research and applied statistics rewarding; however, can you comment on the challenges and dangers for transitioning into interdisciplinary research?

I have written three or four IMS Bulletin columns about this topic, so I refer readers to them. Keyword: interdisciplinary! Of course interdisciplinary work is the ideal refuge for someone who likes to avoid being pigeon-holed, and I am certainly like that. You can always be too mathematical for the biologists and too biological for the mathematicians. More seriously, I do like to try to bridge these disciplines and be comprehensible to both if I can.

In addition to bridging the disciplines between biology and mathematics, what is your opinion about encouraging more collaboration between mathematics and statistics?

This is a tricky issue. Not surprisingly, many—perhaps most—mathematicians see the mathematics in statistics as pretty low level and uninteresting—as mathematics—and I think they are right. We can go a long way with ancient notions such as Taylor expansions. But, of course, some more recent developments in mathematics can have a big impact on statistics, or the analysis of data more generally. I think wavelets are a great example, and more recently, compressed sensing. I suppose my view is that formal collaboration between the two disciplines doesn’t seem necessary unless a well-defined statistical problem definitely reduces to a well-defined mathematical problem. More generally, I see the goals of mathematicians and statisticians as different.

You mentioned to me once you will always give anyone on campus an hour of your time. That stuck in my mind. Do you ever find such things taking too much of your time?

People sometimes say to me, “Thanks for talking/listening to me—you must be so busy. …” To that, I occasionally reply, “If I wasn’t talking/listening to people like you, I wouldn’t be busy.” So no, it never takes up too much time; it’s what I should be doing with my time. The rest is noise (a phrase I like).

I have witnessed you working with a large number of collaborators. Can you describe how it is working with so many people? Do you ever find you lose track of your project?

I have broad interests. I’m curious and restless and will try anything (once). As a result, I usually (always?) say yes when someone asks for help. Do I ever lose track? Frequently. I usually have to begin resumed discussions with collaborators by going back to square two.

Have you had any difficult collaborators, and what is your strategy with them?

If a collaborator is a pain, I might find myself very busy with other collaborators and have trouble responding to them quickly. But as a pain myself, I have a pretty high tolerance for pain, so it happens less than you might think.

It is often a fine line between collaborative research and consulting. Do you have any suggestions for how to strike a balance between collaborative research and consulting?

Balance is a concept I’ve enjoyed studying in experimental design, but it’s never been a strong point in my personal or professional life. Plus, I don’t like giving advice, so I’ll let readers find their own (im)balance here. No suggestions, apart from do lots of both!

There is a surge in demand for “data scientists.” Do you see this as the same demand for statisticians?

This question and the next are hard to answer briefly. I think the answer to the first has to be no, for if statisticians were delivering precisely what those demanding data scientists want, the demand would indeed be for statisticians. But it is definitely not. To that extent, statisticians have missed some of the boats. Having said that, it is clear to me that much of what is being sought in data scientists can be supplied by statisticians with the right training and attitude. There are misunderstandings on both sides about what statistics and statisticians can and should do, but there are certainly many things we don’t do, at least not to the extent needed in many problems. Of course, many of these relate to computing, but some are about roles.

What is your opinion about how the restructuring and emergence of data science courses affect statistics as a discipline?

I see it as a necessary correction for our community. There is no doubt in my mind that there are real and important issues here, mixed in with the usual human characteristics of exaggeration (hype), oversimplifying, competition for resources (turf battles), and so on. We just have to hang in there, argue our case, concede, and correct our failings, knowing that our discipline is a very important part of what is needed, even if it is not everything.

Recently, you have campaigned for women in academia, and I heard WEHI is starting a child care center. Can you tell us why this is important for you?

I think it goes back to my spending a large amount of my childhood at home with my mum. Plus, I have three sisters. So I had plenty of opportunity when young to see the world from the viewpoint of girls and women. I noticed quite early on that “It’s harder for girls,” [which is also] the title of a book and a short story by the Australian writer Gavin Casey that I read as a kid. Putting this together with the fact that working for social justice was a feature of our household and that I married a feminist, you get a male feminist.

At school, I was occasionally top of the class in one thing (e.g., arithmetic or mathematics), but a girl would sweep the field in everything else. So I never got any sense that girls were academically less capable than boys—more the contrary. When I saw how all this worked out later on in girls’ careers, I couldn’t help but be struck by the injustice of it all. My main regret is being slow to embrace this cause.

Mentoring

I saw you have an IMS Bulletin column about mentoring. How important is mentoring in statistics or interdisciplinary research? Do you believe the role of mentoring has increased in its importance over the last 40–50 years?

I do think mentoring is important, not just in statistics or interdisciplinary research, but in most occupations. The world is now a tougher place than it was when I was younger. I came of age in an era when we didn’t need to think about jobs; we just graduated and chose from among several reasonable options. So I definitely agree that the role of mentoring has increased in importance over the last 40–50 years (i.e., over my career). I think it has come with the increasing corporatization of universities and research institutes and the increasing professionalization of disciplines like ours. Failing, drifting, being a jack-of-all trades, things like that are not encouraged these days. Having career plans, getting guidance, avoiding mistakes, this is how it is. To avoid sounding too much like an old fool, I’ll stop here, but let me repeat: Mentoring is important, especially for women and other groups under-represented in senior roles in academia and elsewhere. White males have been mentoring white males for millennia in Western society; it’s so common it is unconscious. But as soon as someone different from the norm comes along—a woman rocket scientist or brain surgeon—the need is greater and the opportunities are fewer.

I understand you don’t like to give advice, but is there any discussion or something you would encourage postdocs to consider regarding their careers?

Talk to lots of people, but don’t do what they recommend. In particular, read my article, “Never Ask for or Give Advice, Make Mistakes, Accept Mediocrity, Enthuse.”

What do you believe is the most pressing question in bioinformatics or statistics today? If you were a graduate student or ECR today, what research area would be the most fascinating for you?

I’ve always hated being asked, “What are the big questions in …,” so I tend to reply, “I work on little questions, topics of no consequence.” See above: contrary. See also my column “How to Do Statistical Research.” I do think I tend to think about unfashionable things, perhaps unconsciously avoiding competition. As for what will be the most fascinating research area for me today, as you know, I’m currently obsessed with removing unwanted variation using negative controls. It’s not going to be solved completely any time soon. But, of course, other things catch my fancy: heterogeneity, evolution, single cells, many cells, tissues, organisms … Who knows when one of these will become an obsession.

I have enjoyed reading your columns in the IMS Bulletin. What motivates the various topics in that column?

My motivation mainly comes from listening and talking to people, though some comes from my personal obsessions. I’m interested in small and occasionally not-so-small issues that slip through the cracks: history, trends, the unspoken, the unspeakable. Oscar Wilde called fox hunting “the unspeakable in pursuit of the uneatable.” I wish I’d said that.

Jean Yang

Jean Yang

Professor Jean Yang is an applied statistician with expertise in translational bioinformatics. She was awarded the 2015 Moran Medal in Statistics from the Australian Academy in recognition of her work on developing methods for molecular data arising in cutting-edge biological and medical research.

1 Star2 Stars3 Stars4 Stars5 Stars (2 votes, average: 5.00 out of 5)
Loading...

Comments are closed.