aa r X i v : . [ s t a t . O T ] A p r Statistical Science (cid:13)
Institute of Mathematical Statistics, 2015
A Conversation with Richard A. Olshen
John A. Rice
Abstract.
Richard Olshen was born in Portland, Oregon, on May 17, 1942. Richard spenthis early years in Chevy Chase, Maryland, but has lived most of his life in California.He received an A.B. in Statistics at the University of California, Berkeley, in 1963, anda Ph.D. in Statistics from Yale University in 1966, writing his dissertation under thedirection of Jimmie Savage and Frank Anscombe. He served as Research Staff Statisticianand Lecturer at Yale in 1966–1967.Richard accepted a faculty appointment at Stanford University in 1967, and has heldtenured faculty positions at the University of Michigan (1972–1975), the University ofCalifornia, San Diego (1975–1989), and Stanford University (since 1989). At Stanford,he is Professor of Health Research and Policy (Biostatistics), Chief of the Division ofBiostatistics (since 1998) and Professor (by courtesy) of Electrical Engineering and ofStatistics. At various times, he has had visiting faculty positions at Columbia, Harvard,MIT, Stanford and the Hebrew University.Richard’s research interests are in statistics and mathematics and their applications tomedicine and biology. Much of his work has concerned binary tree-structured algorithmsfor classification, regression, survival analysis and clustering. Those for classification andsurvival analysis have been used with success in computer-aided diagnosis and prognosis,especially in cardiology, oncology and toxicology. He coauthored the 1984 book
Classi-fication and Regression Trees (with Leo Brieman, Jerome Friedman and Charles Stone)which gives motivation, algorithms, various examples and mathematical theory for whathave come to be known as CART algorithms. The approaches to tree-structured cluster-ing have been applied to problems in digital radiography (with Stanford EE ProfessorRobert Gray) and to HIV genetics, the latter work including studies on single nucleotidepolymorphisms, which has helped to shed light on the presence of hypertension in certainsubpopulations of women.Richard also has a long-standing interest in the analyses of longitudinal data. Thisincludes a detailed study of the pharmacokinetics of intracavitary chemotherapy withsystemic rescue (with Stephen Howell and John Rice). Related efforts have focused on“mature walking,” concomitants of high cholesterol, and aspects of glomerular filtrationin patients with nephrotic disorders (with Bryan Myers). With the late David Sutherland,Edmund Biden and Marilynn Wyatt, he coauthored the monograph
The Developmentof Mature Walking . Richard’s other stochastic-statistical interests include exchangeabil-ity, conditional significance levels of particular test statistics, CART-like estimators inregression and successive standardization of rectangular arrays of numbers.Richard is an elected Fellow of the IMS, the AAAS, the ASA and the IEEE. He is aformer Guggenheim Fellow and has been a Research Scholar in Cancer of the AmericanCancer Society.
John A. Rice is Emeritus Professor, Department ofStatistics, 367 Evans Hall, University of California,Berkeley, California 94720-3860, USA e-mail:[email protected].
This is an electronic reprint of the original articlepublished by the Institute of Mathematical Statistics in
Statistical Science , 2015, Vol. 30, No. 1, 118–132. Thisreprint differs from the original in pagination andtypographic detail. J. A. RICE
Fig. 1.
Richard Olshen, 2010.
Rice:
It’s a great pleasure to interview you,Richard. We go back a long way to UC San Diego inthe 1970’s. I’d like to begin with your distant past,if your memory goes back that far. What childhoodinfluences drew you into mathematics and statisticsand medical science?
Olshen:
My father was a Ph.D. student of HenryLewis Rietz of the Rietz Lectures and the IMS. Hehad a Ph.D. in mathematics from the State Uni-versity of Iowa. He was a very troubled person, butevery once in a while he had a clear view of things.He knew a fair bit of mathematics and some statis-tics, such as it was when he was younger. Interestingideas were always in the air. I remember when I wasa child wondering if there were more real numbersthan integers.
Rice:
At what age was that? Do you know,roughly?
Olshen:
I think I was nine. I remember learningsomething about transfinite arithmetic and Cantor;but that was many years ago, and I don’t remembermany of the details. As far as statistics goes, I was ajoint Statistics and Mathematics major at Berkeley,but the person who advised my letter of the alphabetin the Department of Mathematics wanted me totake a course that I didn’t want to take; so I droppedthe Mathematics part.
Rice:
Oh, is that how you ended up a Statisticsmajor? I know you went to Berkeley, presumablybecause you had an interest in Mathematics.
Olshen:
I was recruited to go to the University ofChicago for no good reason I could ever discern. Myfather wouldn’t hear of me going to the Universityof Chicago. A woman of whom I was very fond wasgoing to Berkeley, and I thought it was a prettygood school. It wasn’t Stanford. Stanford was not awelcome word in my house.
Rice:
Oh, really?
Olshen:
When I was a junior in high school, mymother and I went to college admissions night atBurlingame High School. We lived in Burlingame,California, then, which is near the San Francisco air-port. We were sitting in a room, and the woman whowas somehow in charge of outreach from Stanfordgot up; and the first words out of her mouth were,“Life doesn’t end if you don’t get into Stanford.”My mother grabbed my arm and pulled me out ofthe room and said, “You’re not applying there.”
Rice:
Good for your mother!
Olshen:
That didn’t appeal to her aesthetic.
Rice:
What was the course in statistics that drewyou into the subject at Berkeley?
Olshen:
It wasn’t so much a course as it was aperson. I had taken probability when I was a sopho-more. In those days, you could actually do Volume1 of Feller. Now it’s somewhat forbidden becausethe problems are too hard. I was pretty good at it.
Fig. 2.
Richard’s high school graduation picture. TakenSpring 1958, age 16.
CONVERSATION WITH RICHARD A. OLSHEN Fig. 3.
Richard as a graduate student at Yale. Taken Spring1965.
Then, when I was a junior in college I met the lateDavid Freedman. I believe I was in the first coursehe taught at Berkeley.
Rice:
That must have been very early in his career.
Olshen:
That was in 1961. He was somebodyI wanted to please. He was a stern guy and ob-viously very sharp. In those days, it was clear,both at Berkeley and at Yale, that the very youngfaculty—and David was certainly young then—thatthe young faculty looked on the able students ascompetitors for their jobs, and so that tension wasalways there. He was pretty secure and didn’t reallyfeel this way, but there were others who seemed tohave that attitude.
Rice:
You thought that at both Berkeley andYale?
Olshen:
Well, those were two good places to getjobs. Yes, I was surprised, but that was the sensethat I had, especially at Yale.
Rice:
How many statistics majors were in yourclass? It must have been a very small number.
Olshen:
I don’t know. There were more than 10and not more than 30.
Rice:
It’s grown substantially in the last few years.There are more than 300 now at Berkeley.
Olshen:
There’s a change in enrollment at Stan-ford, too, although there is no undergraduate majorin statistics at Stanford. There’s a Math/Comp Scimajor on which Bradley Efron has worked hard. It’sreally good, and it’s one of the best undergraduatemajors at Stanford. I don’t know how many students it has, but quite a few. The master’s program in theDepartment of Statistics was almost nonexistent 10years ago. Now it has 90+ students, and people areclamoring to get into the master’s program. Peoplefrom all over the world who would have been Ph.D.students 20 years ago, bright kids.I think statistics serves these young people well. Itteaches them something about computing. It teachesthem something about statistical inference. I thinkthese are all good things to know, no matter whatthey choose to do. It’s hard to learn much subjectmatter very well if you’re less than 20, but in yourearly twenties it’s a good idea to learn what you can.That means juniors and seniors in college and firstand second year graduate students.
Rice:
That’s what you did in going from Berkeleyto Yale. Was that transition a big change?
Olshen:
Well, yes and no. It was not a changein difficulty. They were equally difficult. Berkeleyand Yale were both, for me, really, really hard. Ifthey had been one percent harder I couldn’t havedone either one. But Berkeley’s statistics was moredecision theoretic. I would say more of Wald’s de-scendents than anything else. Yale, because FrankAnscombe was the founder of its Department ofStatistics, was much more British, much more Fish-erian, much more likelihood oriented. I was the firstPh.D. student there.What was deemed important in statistics was dif-ferent in the two places. But as a student, one isfaced with challenges of various sorts, and thosechallenges were formidable for me in both places.
Rice:
What led you to go to Yale for your graduatestudies?
Olshen:
Well, I thought, perhaps, that I wouldgo to Princeton; and David Freedman, who in-fluenced me at Berkeley, was a friend of FrankAnscombe, who was then at Princeton. Anyway,I visited Princeton the summer of 1962, and ul-timately was not admitted to the Department ofMathematics at Princeton, anyway. But I didn’tknow that then, when in January of 1963, I got apersonal letter from Frank, which included an appli-cation to Yale. Frank said, “I’m moving from Prince-ton to Yale. Do you want to come to Yale? If you do,here’s an application.” I asked David, “Is there any-body good at Yale?” because I didn’t know muchabout it. For personal reasons, I wanted to get faraway from the San Francisco Bay area. I thoughtNew Haven was far enough.
J. A. RICE
David said, “Oh, yeah. There are lots of great peo-ple at Yale.” He mentioned some of them. He wasright. I said, “Fine. I’ll go there.” That’s how I endedup there.
Rice:
Which Yale faculty had a strong influenceon you?
Olshen:
During my four years in New Haven, Yalein general and Hillhouse Avenue there in particularwere exciting places to be. Frank Anscombe was aremarkable statistician who, in retrospect, kept hisconsiderable mathematical skills too hidden. Sev-eral of us, especially Frank and Phyllis Anscombe,recruited Jimmie and Jean Savage to Yale in thespring in 1964. Jimmie’s last name, not his originalfamily name, was nonetheless well chosen. I believethat his fame in statistical history is deserved. Es-pecially my last two years in New Haven he was ex-traordinarily generous with his time, usually spend-ing an hour with me every day, most time spentworking on mathematical problems. Jimmie’s abili-ties were remarkable. For example, unaware of thework of Kolmogorov and Arnold before him, Jimmiesolved a variation of Hilbert’s thirteenth problemby himself. Unfortunately, few solutions of the vari-ous problems we discussed ever led to publications.The late Shizuo Kakutani taught measure theoryand many aspects of probability. Paul L´evy was theoriginator of much probability Kakutani taught; so,too, were the studies of Markov processes and er-godic theory by him and Yoshida. Some of the prob-lems in measure theory he asked us owed to (sepa-rate) books by Kuratowski, Sierpinski, and Haus-dorff, though students were left to discover themourselves. Alan James taught multivariate analysisfrom his unique perspective that combined as seri-ous computation as was possible then with the studyof matrix groups. There was a year long course inutility theory and game theory by Johnny Aumann,and much else, too.
Rice:
Let me look ahead in time here. Your careerhas had a remarkable trajectory. I think one of yourfirst publications was on asymptotic properties ofthe periodogram.
Olshen:
That was my thesis.
Rice:
Oh, I didn’t know that. One of the most re-cent was on some cytokine bead assays. I don’t knoweven what they are. But I wonder, before we go intosome of these areas in more detail, as an overview,are there landmark topics that you’ve visited duringyour career that sketch out the contours of this tra-jectory? We’ll return to some of these more in depth later, I’m just trying to get a sense of scope and flownow.
Olshen:
I think of statistics as a triangle. There’sa computational part, at which you’re very expert,and I’m not; a mathematical part, which I think isone of my strengths; and subject matter stuff. I can’tdo all three corners of the triangle, or at least I’m notvery good at one of them; and so I try to do the othertwo. I grew up in the Sputnik era. Mathematics wasone of those things that was in the air, and so that’swhat we did. When I was a freshman in college, Iremember being in a class where we did Hardy’s
ACourse in Pure Mathematics . We tried to do theproblems, which were pretty tough for this 17-yearold.
Rice:
Yes, that was an era of intense interest andenthusiasm for mathematics and mathematics edu-cation. It was a heady time to be a math major.
Olshen:
As far as subject matter goes, my feelingis that it’s hard for me, maybe because I’m slow, tobe much of a dabbler. I’ve encountered a few top-ics that have really interested me, and I’ve tried tostay with them long enough so that I could learnenough to be of use. I think that if you’re going todo statistics, then you have to meet subject matterpeople on their turf. In order to do that, you haveto eat humble pie, a lot of it sometimes, and be will-ing to take your lumps, and just try your best tolearn whatever subject it happens to be. There havebeen four or five subjects in my life that I’ve triedto learn. Probably I’ve not learned any of them verywell, but it’s not for lack of trying.That’s always been my attitude. There was themathematics on the one hand, and there was tryingto learn subject matter areas on the other. Together,they’ve been pretty much a full time job.
Rice:
After you got your Ph.D., you moved arounda bit, spending time at Columbia, Michigan andStanford. Then you landed in San Diego in 1975.What led you to come to San Diego? I was veryhappy you did, of course.
Olshen:
Well, I was happy, too. There were acouple reasons. First of all, they would have me,which was not a trivial matter. Second of all, theygave me tenure ab initio. Since I had had tenureat the University of Michigan anyway, and offers oftenure at other places, that was important to me.When I came to San Diego my billet, or whatever itwas called, was joint between Mathematics and theSchool of Medicine. I was interviewed by the Dean
CONVERSATION WITH RICHARD A. OLSHEN of the Medical School, who asked, “Are you reallyinterested in medicine?”I was interested enough to say, “If you hire me, I’llbe faithful to the medical school’s welfare.” I meantit, and I tried to be. The idea of doing mathematicsand medicine always appealed to me. Rice:
Having a foot in each of these places on cam-pus didn’t create a cognitive dissonance?
Olshen:
I don’t know. In that respect, nothing haschanged very much. My job titles have changed, butnothing about me in that respect has changed. Inever stopped to ask. I think that people are drivento do what they’re going to do. It’s not fruitful to askwhy. One does what one does. If it’s robbing banksor hurting people, that’s not an admissible strategy;but you can look after your career and pursue whatinterests you or do what you think you can do. I’venever stopped to ask.
Rice:
One thing you did at UCSD during thetime you were there was to create a real presencefor statistics, particularly in the medical school, inwhich it hadn’t had much presence before. I waswondering: how did you go about doing that? It canbe socially and culturally difficult.
Olshen:
UCSD was started, as you probably know,as a university campus in the 1960s, as opposed tobeing merely the Scripps Institute of Oceanography,which had existed since the early 20th century. Itwas founded by Roger Revelle, an amazing man. Hefought hard against prejudices that were ruled illegalby the 1964 civil rights law. He brought scientificactivity to a part of the world where it hadn’t beenso much before.But if Roger Revelle had a blind spot, it was thathe just didn’t like statistics. There was never a De-partment of Statistics at UCSD like there were atvarious other UC campuses, as you well know.A lot of problems in medicine really involve sta-tistical issues, and not just in medicine, but in a lotof scientific areas. Think of the validation of the dis-covery of the Higgs boson, for example. It seemed tome that there was a vacuum, that there was a needfor people interested in interpreting data. I don’tknow that I filled it very well.
Rice:
There must have been a few key people inmedicine who helped you fill that vacuum.
Olshen:
There were. One of the things that helpedpromote that was that in the late 1970s there wasan attempt to get National Cancer Institute desig-nation for a Cancer Center at UCSD. The leader ofthe effort was John Mendelsohn. There was a group of people including not only Mendelsohn, but alsoSteve Howell, Mark Green and Ivor Royston. Theywere eclectic, but real dynamos, all of them in theirown ways.They included me. I think that was certainly onepath. Another path that I think was really helpfulto me at UCSD was that UCSD had this traditionof cardiovascular medicine. Gene Braunwald of Har-vard had been at UCSD briefly. He brought JohnRoss and Jim Covell and other people there. Therewas this huge presence in cardiology. John Ross wasthe leader of it when I was there. Many of these peo-ple were really smart. They operated on dogs andwhat have you, so it was a little grisly what theydid. I felt I’d learned from them. It was a pleasureto be involved in their projects. There was some-thing called the Specialized Center for Research inIschemic Heart Disease, and they included me.A third avenue was the Gait Lab in Children’sHospital and Health Center. Again, that was inter-disciplinary. It involved a surgeon, an engineer anda nurse; I was the fourth of them. We didn’t publishmany things, but I think what we did was prettygood.
Rice:
Yes, your work on gait was an importantearly stimulus to the development of functional dataanalysis.
Olshen:
Those were three areas that I think wereenabling to me. There were many other good thingsat UCSD that came later. Psychiatry is a big dealat UCSD, and eventually I got involved in the Cen-ter for Neurobehavioral AIDS. Anyway, those weresome of the avenues. The thing they all had in com-mon is that I had much to learn.
Rice:
In the Department of Mathematics, whereyou had your other foot, what people did you learnfrom especially?
Olshen:
Well, of course, coming to UCSD, I wasgrateful because Ingram Olkin at Stanford had spo-ken with Murray Rosenblatt. Ingram didn’t give meany reason to be optimistic, but Rosenblatt wasthe senior person in the statistical community atUSCD, and the whole reason I got interested in pe-riodograms in the first place owed to the famousbook by Grenander and Rosenblatt.
Rice:
I remember that you knew that book quitewell.
Olshen:
Well, I had read it from the first letter tothe last. I can’t say that I memorized it, but prettyclose. Murray was there. He was certainly an influ-ence. I knew that Adriano Garsia was at UCSD. He
J. A. RICE had given basically a two line proof of the maximalergodic theorem; it led to a quick proof of the er-godic theorem, which is something that had begunat Yale in some sense with Josiah Willard Gibbs. Ihad a Josiah Willard Gibbs Fellowship at Yale whenI came there, so I felt some connection with thatwork. Michael Sharpe was somebody I had knownsince graduate school.
Rice:
Oh, that’s right. He was a graduate studentat Yale, too, wasn’t he?
Olshen:
He was the first person I met in NewHaven. I remember talking to Michael, who was fromTasmania, which seemed like it was pretty far away.He had been an honor student. I guess in their sys-tem, you did three years of college, and then if youwere really good, you did a year of honors; he haddone honors with the celebrated E. J. G. Pitman,father of your celebrated colleague Jim Pitman. Iremember coming home after spending about a halfhour in the Yale Co-op chatting with Michael, andI remember telling Vivian, my wife at the time, “Ifeverybody around here is as good as this guy, I’min big trouble.”Michael was very well educated, and he was quitesmart, and that was evident, I would say, after about45 seconds. After 30 minutes, I was thoroughly in-timidated. I remember that Michael detested thecold in New Haven; he came to San Diego in part be-cause he read through books on temperatures in thecontinental United States, and he wanted a high av-erage temperature and as small a difference as pos-sible between the max over the month and the min.
Rice:
San Diego is pretty much an optimum inthat metric in the US.
Olshen:
He said, “I’m going there,” and he did.Anyway, and of course, you were there, and youwere interested in time series and all that stuff. Ididn’t feel like Stanford was the right place for meto be pursuing that. There were a lot of reasons whyUCSD seemed like a good place. There were a lot ofvery bright, very able people.However, I think there was a downside in thatSan Diego got to be a really good place becauseit rapidly hired a bunch of people who were verygood, but who were unhappy where they were. Theyweren’t unhappy where they were because of wherethey were; they were unhappy with the place be-cause of who they were.The medical school actually was different fromsome of the rest of the campus, because as medical
Fig. 4.
Richard, taken in the backyard of his home in DelMar, CA, in 1977. schools go, the medical school wasn’t very cranky.Or at least I didn’t perceive it as being so.
Rice:
One of the best things that happened to youat San Diego was that you met and married Susanand expanded your family.
Olshen:
Yes, well, I was in a pretty sorry shape.I was a single parent.
Rice:
How did you meet?
Olshen:
Oh, I met Susan because I was a sin-gle parent living in Del Mar Heights. There weretwo women in the neighborhood, Sandy Petersonand Gail Goldberg. They used to help me, becauseI didn’t know about the Hebrew school, I didn’t
Fig. 5.
From left to right: David Perlman, Michael Perlman,Elyse Olshen, Adam Olshen, Richard Olshen. Picture taken in1978 in La Jolla, CA.
CONVERSATION WITH RICHARD A. OLSHEN Fig. 6.
Richard and Susan Olshen on the banks of theCharles River, Spring 1980. know about piano lessons; I didn’t know about soc-cer teams. If something came up, I would ask oneof them, “Should my child go to this school or thatschool or this team or this teacher or whatever?”One day Gail said to me, “Richard, my husband’spartner’s wife has a friend, Sue Heller, in La Jolla;and she’s separated from her husband; and if youdon’t call and ask her out for dinner, I’ll never speakto you again.” So I called her.
Rice:
That’s a forceful matchmaker!
Olshen:
I said, “Sue Heller is the name of the wifeof my pediatrician.” I said to Susan, “If you’re thewife or former wife of my pediatrician, then I’m notgoing near you with a ten-foot pole because one ofthe few things that’s going well in my life is thepediatrician. I really like this guy. He takes goodcare of my children, and I like him. So if you’re thatSue Heller I don’t want to get anywhere near you.”She said enough to preclude her being the wife of thepediatrician. I said, “Well, OK. Do you want to go toLescargot for dinner?” She was taken aback becauseit was a nice restaurant. But the thing about it wasthis: it was really awkward for me to go there bymyself. I will say Susan was totally flabbergastedwhen I took her there but I said, “It has nothing todo with you, I like this place and I can’t come hereby myself.” Gallant I was.So I met Susan in 1977. We got married in 1979.What’s this, 2013? It’s been awhile.
Rice:
It certainly has. In 1977, changing the topica bit, you were beginning to get involved withCART, which at that time. . .
Olshen:
Oh it was before then.
Rice:
At that time, it seemed to me quite noveland esoteric. Now it’s a very standard tool that ev-erybody learns; it’s widely used.
Olshen:
I started in with CART in 1974, at theStanford Linear Accelerator Center. I was in the
Fig. 7.
From left to right: step-son Stephen Heller, step-daughter Rachel Miller, son Adam Olshen, and daughter ElyseOlshen Kharbanda, taken at Adam’s wedding to Manisha Desai in 2001.
J. A. RICE
Computation Research Group. Jerry Friedman wasmy boss there, and he was very interested in binarytree structured rules. They started out as rules forquick searches because you can imagine if you wantto find a nearest neighbor and you build a tree downto where there’s one observation per terminal node,you’re going to be able to find nearest neighborspretty easily. Jerry was interested in using this forclassification. I got interested in the application side,which had to do with a lead and plastic sandwich ofparticles originating in a bubble chamber. Also, LouGordon and I worked on the mathematical side.By 1977, I was into CART. There was no bookthen. The book didn’t come until six or seven yearslater, depending on how you count.
Rice:
In 1977, weren’t Leo Breiman and ChuckStone also involved?
Olshen:
Leo and Chuck were definitely involved.There were basically three groups of two, Jerry andLarry Rafsky, Chuck and Leo and Lou Gordon andI. Larry Rafsky was busy with other things anddidn’t really pursue this very extensively. Lou some-how never became part of the milieu, but I becamefriendly with Chuck because I had known him in myprobability life. David Siegmund and I had workedon a problem that Chuck ended up doing. Then in1975 there was a meeting at UCLA where nearestneighbors and trees and what now are called supportvector machines, but in those days were called vari-able width kernels, were very much in the air. Therewas in CART history a famous technical report thatcame in 1979 from a place where Leo did consultingin Santa Monica and where he dragged Chuck. Itwas called Technology Services Corporation.
Rice:
I remember seeing that report. It was quitesomething, very forward looking, for its time.
Olshen:
Chuck was pretty well versed in trees sev-eral years before then. I remember he had written apaper that he submitted to
The Annals of Statistics .Richard Savage was the editor. He had showed it toJohn Hartigan who didn’t speak well of it, I guess.I called up Savage and gave him a piece of my mind,not that I had any to spare, and not what he wantedto hear.
Rice:
Weren’t you a discussant of that paper?
Olshen:
Yes. It went from being rejected to beinga discussion paper.There were trees involved in that, and there weresome personal rivalries that were buried in the dis-cussion, Lou Gordon’s and my first paper on CARTfor classification was published a year later. Jerry had published something in one of the IEEE jour-nals in 1976.Then, I think it was 1981, Chuck manipulatedthings in the following sense. Chuck’s older boy,Danny, had a Bar Mitzvah. There was assigned seat-ing at the reception, and Chuck went out of his wayto make sure that I was seated next to Leo.Leo and I talked for several hours about treestuff. Somehow that led to a manuscript, and thatmanuscript existed for quite a long time. Some of itwas medical stuff that I wrote and some of it wasmathematics. Regarding the latter, I wrote the ini-tial draft; and Chuck completely rewrote it. Sevenof the first eight chapters were from Leo. I remem-ber vividly Chuck saying that, “with Leo the first90 percent is easy and the last 10 percent is reallyhard. With me, if you can understand the notation,it’s all there.” What happened is that Leo took whatChuck wrote, read it and he really didn’t like it.
Rice:
But both pieces survived in the final book,right?
Olshen:
Well, they did; but they survived in funnyway. I’ve told this story before in the pages of
Statis-tical Science , and I’ll try to be brief. Basically whathappened was at one point Susan and I came up toBerkeley and were visiting Chuck for some reasonthat I don’t remember. We went to what used tobe a very good open air sandwich shop on Hearst,just below Euclid on the north side of the street. Thethree of us ran into Leo and Jerry. At that point, Leoand Chuck hadn’t spoken to each other for a longtime, and the manuscript lay dormant. Leo was al-ways the gallant one and he said “Why don’t we gettogether after lunch at my office, and we’ll hammerthis out?”Susan said fine, and she had a book in her purse.She said, “I’ll go to the library and read my book”and I said “No you won’t.” I knew that Leo hadthis gallant aspect to him, that Leo would never beharsh in front of a woman. “You’re coming to ourmeeting.”We came to Leo’s office, the four of us; Jerry andI were always willing to compromise on almost anyreasonable thing. Leo and Chuck didn’t get along allthat well even though they were colleagues. I got achair and made sure that Susan sat on it betweenLeo and Chuck: Leo and Jerry on one side; Chuckand me on the other.I knew that if we were ever going to agree on any-thing that that was the right environment, and wedid. We came to some ground rules about who was
CONVERSATION WITH RICHARD A. OLSHEN allowed to criticize whom, and that I would be thearbitrator. I would try to write things so that it readlike a book, and make the glossary and the table ofcontents and what have you. The book was finishedsometime in 1983 and was published in late 1983. Rice:
Yes, I still go back to it and read it for in-sights. When I think I understand something, andthen I realize that I don’t, I go back and read itagain.
Olshen:
We tried pretty hard. Of course now, it’ssomewhat pass´e. That was, of course, before boost-ing, though we certainly realized that if you havea base rule for classification, observations clearlymarked for one class or the other aren’t the hardparts. The hard parts are observations near theboundary. The idea of boosting made sense, butmaking science out of that is not a trivial matter.I have the impression now there are lots of whatI consider pretty good classifiers out there. Thereare neural nets done properly and support vectormachines, because Vapnik had this bully pulpit andwrote a book. There’s boosted CART. Then laterLeo got into random forests. Those are just somethat come to mind.I don’t really think that the hard part of mostclassification problems is whether you choose a sup-port vector machine or boosted CART. I think thehard part is knowing what features to include. Knowing what features to include gets you themain digit in error rates and risk. Whether it’s sup-port vector machine or boosted CART or somethingelse matters less; it’s easy to fool any of them. But,if you’re any good at what you’re doing you’ll know,“Gee, I don’t think I want to use a random forest forthis because there are a lot of features and most ofthem are noise; and I could fool it.” Or, “I know thedecision boundary is really smooth and a straightline so vanilla CART doesn’t make sense becausethe boundary doesn’t have the saw tooth.” Well,you should know your subject matter well enoughto know that, and if you do you can usually be apretty good guesser as to what to use. But it mat-ters whether you include this or its square or theproduct of these two things or whatever.
Rice:
Another hard thing about classificationproblems is not, as you say, it’s not whether youuse support vector machines or random forests, buthow you actually construct the training set, whereit comes from, and what it’s relation to the test setis. That’s often really quite nontrivial.
Olshen:
Of course.
Rice:
I think it’s frequently glossed over.
Olshen:
Well, the assumption of internal cross val-idation is that the joint probability structure of thepredictors and the outcome are the same; and what
Fig. 8.
This photo was taken in front of the old Sequoia Hall at Stanford in 1975. The occasion was a gathering to discusswhat role statistics research might have in environmental problems. The cast of characters is: Back row (left to right): BradEfron, John Tukey, Paul Switzer, Herb Robbins, Tom Sager, not identified, Ray Faith, not identified, Richard Olshen. Middlerow (left to right): Don McNeill, Yash Mittal, Elizabeth Scott, Don Thomsen, Gary Simon. Front row (left to right): GeoffWatson, Peter Bloomfield, Persi Diaconis, Jerzy Neyman, Ingram Olkin. J. A. RICE you’re testing is not. Does that make sense? In alot of applications, it doesn’t. You see that all thetime in medicine. Just for an illustration: Supposeyou have a truck and you go to the county fair andyou do mammograms. You could have some classifierand it will be trained because you’ll go to some med-ical center and pull out 500 records of people whohave breast cancer and 500 people who didn’t. Butin the county fair the prevalence/priors are maybeone out of 500 or something like that. It’s very differ-ent. You’re basically talking about different regionsof the feature space, different base rules, and thething that worked for 500 versus 500 may not workvery well for one versus 250.
Rice:
And the joint dependence structure of thecovariates can be different.
Olshen:
Yes. In that part of the feature space,it might. There are all kinds of things that can gowrong, and it’s amazing that in 2013 that one stillneeds to say such things out loud because these aremistakes that are common today. It’s not like, “Oh,in olden days people did things this fallacious way.”Olden days may be 20 minutes ago.
Rice:
Your interests changed. You went to Stan-ford; I think it was in 1989. At some point in theSchool of Medicine, there you became interested ingenomics. That changed a lot of what you did. Howdid that transition take place?
Olshen:
Well, I’d always been interested in genet-ics. My first wife and girlfriend who drew me toBerkeley in the first place wrote a thesis about thegenetics of mating latency in fruit flies. Basically,the idea is that you had sites and people knew thatthe outcomes were discrete. One of the big things onthe table then was, “How many genes were involved?How many sites?” These days you’d say, how manySNPs were involved in producing a particular phe-notype?That’s a deconvolution problem because the phe-notype you see is the product of some vector of geno-types plus some noise, you must deconvolve the sumof things that matter and the noise.I had a long standing interest in those problems.Then in the 1990s I was very fortunate at Stan-ford, just as I had been in the Laboratory for Math-ematics and Statistics at UCSD, that I had veryable assistants. My assistant at Stanford, BonnieChung, said that I got a phone call from Victor Dzauwho was then the Chief of the Division of Cardiol-ogy at Stanford and the Chair of the Departmentof Medicine. Later he left Stanford. He and others were starting a project that ultimately was calledSAPPHIRe, the Stanford Asia and Pacific Programon Hypertension and Insulin Resistance. It involvedpeople who I didn’t know but should have. DavidBotstein was one, and there were various others. NeilRisch was somebody upon whom we could lean tohelp with calculations.So for reasons that I don’t know, Victor somehowgot my name and I knew who he was even thoughI doubt he knew much about me, and said, “We’regoing to be writing this grant Saturday morning.”Well. . .I didn’t take to being anyplace at eight o’clockSaturday morning. But I finally got there at nineo’clock, having dragged myself out of bed, becauseI realized that this was the big leagues; and eventhough finding genes that predispose to hyperten-sion is really tough, it seemed like something Ishould get involved in. That was in the 1990s. Sincethen things grew. The technology grew—one of mystudents worked for a company, Affymetrix, that dida lot of SNP genotyping and invented some of thetechnologies.That technology was developed by a man in engi-neering and his daughter. Part of my life has beenin Electrical Engineering at Stanford. The man isFabian Pease. It involves embedding something inplastic and shining laser light on what binds to it,the complimentarity of nucleic acids, and the bend-ing of laser light.The bending of light leads to an inverse physi-cal problem of making an inference. I’m not goingto go into details because there are other places toread about it. But the point is that virtually allthose technologies, SNP technology, expression tech-nology, and now protein chips, in some sense theyare all the same. Those are nifty problems.They get harder the bigger the molecules youare embedding in the plastic are. That’s why theproteins are really tough. They tend to be hugemolecules, and they don’t have very many bindingsites.I never got very much involved in gene expression,but I’ve certainly been involved in the proteins andthe actual SNPs themselves. Once again, there’s atriangle. There are SNPs; there is then gene expres-sion; and then the actual proteins that your bodysees.One thing has led to another, and a lot of problemshave come up related to that, one of them beingimmunology, very broadly defined. That’s how I got
CONVERSATION WITH RICHARD A. OLSHEN into this SAxCyB and protein arrays. The statisticsof it is not very foreign. Rice:
Another activity, of course, that has con-sumed your time at Stanford and your interests is allyour work on image compression with Bob Gray andhis colleagues. It’s easy to see a path from CART tothat in broad brush. How did that begin?
Olshen:
Well that started out because I was atStanford on sabbatical in 1987 and 1988. There wasa graduate student in electrical engineering namedPhil Chou, who’s now at Microsoft Research, a bril-liant person. Jerry Friedman, my CART colleague,was supposed to be on his orals committee, andJerry wasn’t able to go to the exam. He asked,“Would you go?” I was just a visitor, but it seemedof interest and I went. Phil was clearly terrific. Histhesis adviser was Bob Gray, who was the master ofcompression. Bob’s student Eve Riskin saw that thepruning algorithm that’s Chapter 10 of the CARTbook, that came from the Technologies Services Cor-poration tech report, really applied to image com-pression. Think of a binary tree and you could thinkof bits telling you to go left or right, and you canthink of the average number of bits you need, andthat’s just the average depth of the tree.If you are building large trees and pruning themback, you’d be faced with what amounts to the sameproblem in both cases. Anyway, when I came backto Stanford in 1989, there was a phone call fromBob who was looking for somebody with whom tocollaborate, and he had problems in image compres-sion of various sorts. I was asked to help, and I did.I’m not sorry I did; it’s been an interesting chapterof my life.We studied malignant masses in the mediastinumand in lungs by CT. We studied flow through majorblood vessels in the chest by MR. We studied digitalmammography, which turns out to be a really hardsubject, and also satellite images.
Rice:
There’s another thing you’ve been involvedwith at Stanford that I know much less about, theData Coordinating Center. You haven’t told memuch about it in the past.
Olshen:
Well, what I thought it was to be and theway it’s turned out aren’t the same. My motivationwas very simple. It used to be that when anybodyhad his or her favorite algorithm for doing classifica-tion of whatever, it was always, and I mean always,tried out on the UC Irvine database. I don’t everwant to hear again about the UC Irvine database.I thought, there’s so much going on at Stanford. Why don’t we just organize something at Stanfordand get Stanford data and use them for standardsin somebody’s support vector machine or whatever?I decided to organize something: the Data Coordi-nating Center. My hope sort of panned out, andsort of did not. It still exists, but it’s turned intoa boutique operation that does very fancy databasethings, mostly for Stanford’s Cancer Institute. Fur-thermore, HIPAA laws have intervened. It’s not atrivial matter to get data from somebody’s experi-ment on human beings to a statistician, or an engi-neer, or somebody who may have something to sayabout, “Yes, this person will get a malignant dis-ease,” or, “Yes, this person has hypertension,” orwhatever.But I got that started before I knew the weight ofHIPAA laws upon us. My efforts were a reaction tomy being sick after the 107th time that I saw some-thing from the Irvine database. Some of the thingsI was involved in at Stanford had to do with nephrol-ogy, that is to say, with kidneys. I got involved with agroup in Phoenix; one of the NCI branches. NIDDKis there, and I worked with a friend in his lab atStanford. I knew that in the database at UC Irvineis the Pima database. I knew that there are PimaIndians in Arizona, because there’s a reservationthere. They have hardscrabble biological cousins innorthern Mexico who are skinny and not hyperten-sive. The people in Arizona are insulin resistant, andthey’re fat; and you can wonder why.This seemed interesting because this suggestedthat there was some gene by environment inter-action going on, so that played into CART, intomy interest in that. It played into my history withnephrology, and I realized, and maybe this is pre-sumptuous of me, that probably many of the peopleusing the Pima Indian database in the UC Irvinecollection for testing their algorithms didn’t knowanything about hypertension, or Pima Indians. Thatoffends my aesthetic. Maybe it’s because I’m so poorcomputationally, but I’ve seen myself as a partici-pant in people’s activities, but not more than that.
Rice:
Let me probe a bit further into your role ininterdisciplinary studies. You’ve talked about sev-eral of them, and you said one of the things youbring to them is humility; but actually, as a statis-tician, you bring more. You’re working with smartengineers, or you’re working with smart MDs, butyou’re bringing something as a statistician.
Olshen:
I hope so. J. A. RICE
Rice:
You’re bringing something to the table. Iwonder if you could articulate what you think thatis.
Olshen:
One answer might be an example. Some-thing just came up in the Workshop in Biostatistics,that I ran at Stanford for many years, and for whichI am now ably assisted by Chiara Sabatti, who doesmost of the heavy lifting.Imputation is a big deal in genetics these days.People make inferences about the single nucleotidepolymorphisms at sites for which they have no data.They may actually sequence a half a million sites ifthey do a lot, maybe many fewer if they are morespecialized.To impute they use something called haplotypes.My understanding of what a haplotype is, is thatthere are long strings of DNA, and if I’m at a givenpoint and there are five points nearby and I knowwhat those are, then I must be part of such andsuch a cluster and, therefore, I can read out fairlyfar. OK? What the genome is, then, is a bunch ofhaplotypes strung together. I’m going to even forgetabout the randomness of the fact that the partitionof humanity is very coarse. One can ask, “What’s theprobability mechanism that generated these thingsin the first place?”After querying people in a large audience that in-cluded some people who know genetics far betterthan I do, it seemed that because this imputationis done with so called hidden Markov models, thereneeds to be something that’s at least approximatelyMarkovian there. What is it?I was able to get out of the discussion that what’sMarkovian are these so-called haplotypes that getlaid down. Well that means that the marginal distri-bution of the individual sites is certainly not Marko-vian. But what is it?Well, people compute now the covariance functionof sites. You can do that, but then you have to askyourself, is the covariance function you compute con-sistent with that of a mixture of Markov processes?You should be able to answer questions like that, be-cause you should know the probability mechanismthat generated the data in the first place.That’s our job—to try to make those inferences.I don’t see those kinds of questions being asked. Youask what I bring to the table, maybe it’s a sensitiv-ity to things like what I’ve cited. That’s an exampleof something that’s sort of statistical, sort of proba-bilistic. One could think, “What kind of tests wouldyou do if you got data on genotypes to figure out if something was a mixture of Markov processes ornot, and necessarily consistent with how haplotypesare said to be generated?” That’s a question that itseems to me is worth asking. So far as I can tell, ithasn’t been asked.
Rice:
I’m thinking about what you’ve just beensaying about this example and about numerous in-teractions with young people, both statisticians, andnonstatisticians. I’m thinking particularly aboutpeople who attend your biostat seminars, aboutgraduate students and post docs. What advice doyou give them if they say, “I’d like to be doing thiskind of thing, this interdisciplinary thing in the fu-ture.” Do you tell them, “Go out and learn aboutMarkov processes?” What do you say?
Olshen:
No. Well, first of all, hardly anybody everasks. But of those few who do, my only advice wouldbe that anything you learn is to the good. In par-ticular, anything one can learn in mathematics isto the good because it may come up in the future,and it certainly sharpens the mind. Anything youcan learn about the subject matter is fine. But themost important thing you have to learn is you haveto learn how to learn, because, at least in my life,the things that I do every day didn’t exist as prob-lems when I was a student. The world has changed.I don’t know if it has changed for the better, butit’s changed. One is constantly having to learn newthings.To summarize, the main things to learn are pa-tience, learning how to learn, learning how to bea student for the rest of your life. Because if yougo into some academic work, you are going to bea student for the rest of your life, and not onlythat—I was speaking with Iain Johnstone aboutthis the other day because the question came up inconversation—I think you have to enjoy the chase.The chase might mean working on problem three inChapter Seven.It might mean the fact of trying to understandSNPs that are combined with some environmentalfactors to predispose to insulin resistance or hy-pertension. It might mean any one of a number ofthings. But if you don’t enjoy and get some chargeout of just whatever the chase is, then you are notgoing to be very happy; and you’re probably not go-ing to be able to do much either, and there’s a lotto do.
Rice:
You said you have to be a student. I thinkas you get older it’s hard to find the time to be astudent.
CONVERSATION WITH RICHARD A. OLSHEN Fig. 9.
Richard with Peter Bickel and Erich Lehmann atBerkeley in 2005. Peter is a long time friend and collaborator.The late Erich was Richard’s adviser his sophomore year atUC Berkeley in 1960–1961.
Olshen:
One has no choice.
Rice:
You have to really want it, or else it’s notgoing to happen.
Olshen:
That’s the only choice there is. One isa student. I don’t know what it would be like tobe a super genius. But I can say what’s like to beme. If you just have maybe better than average butnot such spectacular gifts, then you just have to bewilling to plug away and to be patient and crossyour fingers, and hope for the best. But one of thethings, also, that I think, because this has comeup in conversations far removed from this discus-sion lately is this: it’s really nice when people comealong afterwards and they come up with a simpleproof of something. You think, “That’s great.” Butthe first person that got there didn’t know, didn’tknow what the answer was. Maybe yes, maybe no,maybe this, maybe that. To me, that’s the hard partand the fun part of every subject. In that respect,there is no disconnect between medicine and math-ematics. They are just hard things to do. They’rethings one doesn’t understand and one crosses onesfingers and hopes that one will learn to explain somephenomenon. I can say in my case that I’ve certainlybeen disappointed many times. That maybe it’s justbecause I’ve made unfortunate choices.But I think the people who are most successfulhave been successful at least in part because they’vebeen wise about how to spend their time. Every-body’s only got so much time. There are a few super geniuses, but there are not enough to populate allthe universities.But some people are clearly better than others atpicking things to work on. Afterward it’s easy to say,“If I had thought of that. . . ” Well, the point is thatyou didn’t.Well it’s just like you and Bernard’s finding theeigenfunctions and my gait stuff. After the fact, Isee that’s a kind of obvious thing to do. Not thatI know how to form confidence intervals for thosepredictions very well. A lot of things are easier inhindsight than they were in foresight.
Rice:
Yes. Foresight’s limited. I was thinkingabout yours. I was trying to put myself in your po-sition when you were working on the fluctuationsof periodograms. Then in light of things we’ve justbeen talking about, if you try to look ahead fromyour point of view then, what things would mostsurprise you about statistics? There’ve been lots ofchanges. Are there particular things which you justwouldn’t have envisioned, which have surprised youespecially?
Olshen:
I think that modern computing haschanged the world. It will never be the same, andit shouldn’t be. I think that it’s not settled yet. Be-cause there is one view of this world that says, “Well,I don’t know if this is a good model for x , y or z ; soI’ll simulate.” You’ll simulate five cases and they’llall come out heads! I can toss a coin five times andit’ll come out all heads, and I’ll think that’s a twoheaded coin; but maybe that just happens that itcame out heads five times. Then on the other hand ifyou’re just curmudgeonly and say, “Well I won’t be-lieve a word unless I can prove some theorem aboutit,” you almost never can. What is the right wayto be? I have no idea. Maybe 100 years from now,if the world doesn’t blow itself up or poison itself,then maybe people will figure that out better.I’m 70 years old. Actuarial chances are that I’mnot going to live that much longer, and my healthisn’t so terrific. We just get just this little slice oftime. I’m not a hyper-religious person, but I do tryto read the Torah portion in the Old Testament ev-ery week.The Hebrew is really beautiful. I can’t translate alot of it, but what I can translate is really good. Notonly that, but in books that one reads, one realizesthat hundreds and thousands of years ago there weresome really smart people who wrote great stuff. AsBradley Efron reminds me, if Mozart hadn’t lived, J. A. RICE it isn’t that somebody else would have written “DonGiovanni.” We wouldn’t have “Don Giovanni.”But whatever we’ve done, and nobody in sciencedoes that much because you know it’ll all be redis-covered somehow, in some fashion anyway. What weknow from the past is just a distillation of what hap-pened. Who knows if what’s distilled and thought tobe so nice now was in its day thought to be so nice!I’ve had occasion recently to be interested in thedistribution of the sum of independent uniform ran-dom variables. It just came up as a matter of so-called meta analysis and all this. It’s not a trivialmatter, because you can think of picking a point ona hypercube and a plane sliding through the hyper-cube. But hypercubes have corners, and they screwup distributions. Well, so I’ve learned that in 1920stwo very smart people, one named J. O. Irwin andthe other Philip Hall, who went on to become a fa-mous mathematician, figured out how to do that.They published in
Biometrika . Rice:
Figured out how to do what?
Olshen:
How to compute the distribution of thesum of IID uniforms. That sounds like a simple exer-cise, but just try to do it. It’s not so simple. You caninvert a Fourier transform if you’re good at invertingFourier transforms, but that involves complex inte-grals. It turns out that the essential computationfor that, and this is a footnote that Karl Pearsonput in
Biometrika , the essential computation thatenables you to compute the distribution of the sumof IID uniforms was done by Euler, who apparentlydidn’t know anything about applications and couldnot have cared less. Good for him, but was thatworth anything in those days?How did I get interested in that? I got interestedin it because it had to do with combining indepen-dent tests into one test of significance. If you thinkthat the null hypotheses are true, then you’ve got auniform draw on the unit interval. You’ve got, collec-tively, a point on the unit cube. Fisher’s minus twicesummation thing results in a hyperbolic neighbor-hood of zero. What if you wanted a linear neighbor-hood of zero? This is something my son Adam gotme into.Well, the question is easy to state, but the answersaren’t always easy to come by. I think that what iseven the right thing to do in given applications isfar from obvious.
Rice:
You have been quite a valuable mentor toyoung people. Is there anything you can say aboutthat process?
Olshen:
There are few guidelines. It says in Torahthat there are two classes of people in the world ofwhom you must never be jealous, your children andyour students. That’s one set of guidelines.Another thing: some cultures have a severe, if im-plicit, concern about respect for elders; whereas Jew-ish culture has in it a healthy skepticism of the wis-dom of elders. Now that I am old, I wouldn’t minda little more respect; but I think that it can be over-done because the future is for young people.My attitude is that no future was built on thebacks of 70-year olds. The future is in young people.If you think that the young people are what willbecome us (and we won’t be here to see what theydo) then you would like for them to look back onyou perhaps favorably to the extent that what youinstilled in them was something worthwhile.
Rice:
You’ve been in academic institutions rep-resenting statistics in one way or another, depend-ing on the institution. Academic structures and ed-ucation are changing, the roles of statistics can bedifferent in different universities, depending on theenvironment, and those environments are changing.
Olshen:
I think statistics is in a really difficultplace, because it has to justify itself as having some-thing of its own, on the one hand, and being a ser-vant of other fields on the other. You and I havetalked about that. I think that’s a scenario that ishard for university administrators to understand.
Rice:
Well, it’s a strength and simultaneously aweakness.
Olshen:
That’s true. It’s a perpetual problem, andI don’t think it’s going to go away. However, thereare other people trying to eat our lunch. Computerscience is, for example. To me it is about data struc-tures and related subjects. These are fields aboutwhich statisticians could do well to know more. How-ever, to too great an extent, computer science is re-discovering the wheel. I think that in classification,for example, or machine learning, there is much toomuch encroachment by computer scientists.
Rice:
What are your plans for the future? Whatare you looking forward to doing?
Olshen:
Don’t know. I think about that, but I haveno idea. I mean, I realize that one useful purpose Ican serve is to be a babysitter for grandchildren.That’s important. That’s clearly a task that I amdeemed able to do.
Rice:
Congratulations.
Olshen:
Beyond that? I don’t know, more of thesame. I’m trying to get some papers done now. I
CONVERSATION WITH RICHARD A. OLSHEN can’t run as fast as I used to. I used to be sharperthan I am now. All I’ve ever had is just the ability toreact to situations that weren’t always of my choos-ing and weren’t always enviable either. My health ispretty poor.I’m trying to write a monograph on the successivenormalization of rectangular arrays of numbers, andI see there’s lots to do, and I don’t know if I’ll getto that. But I hope to. Rice:
Well maybe it gets back to Yogi Berra, right?It’s hard to predict what’s going to interest you inthe future. Would you have predicted five years agothat you’d be interested in normalizing rectangulararrays? Probably not.
Olshen:
No. That came up as a challenging math-ematical problem. But I see that it has practicalconsequences. It’s like making inferences about vec-torial data, whether you look at covariances or cor-relations, you learn different things from each one;and that’s inescapable. I’m also trying to rewritesomething for some referees now that has to do withdefining insulin resistance rigorously and finding ifthere are SNPs and candidate genes that predisposeto it. I just finished something with my son, Adam,on ribosomal profiling.There’s another project that has to do with HIV.HIV used to be an acute disease and you’d get it andyou were dead quickly. Drugs now really prolong life,but they are pretty potent stuff. They’re pretty bad,and you have to worry. If somebody is going to bealive for 10, or 15, or 20 or 30 years, you’d betterworry about whether the potion you are giving isgoing to cause heart disease, or kidney disease orsomething else. There are ways of trying to makethose inferences.
Rice:
We’re very fortunate to be in a professionwith so many opportunities, aren’t we?
Olshen:
Yes, it’s a pretty good deal. I rememberin San Diego at the Rosenblatt’s house many yearsago, the late Errett Bishop asked, “What would youdo if you could do anything? Would you work inalgebraic geometry, do this or do that. . . ?”
Fig. 10.
Richard Olshen and John Rice in the Fall of 2013,Berkeley, CA.
Rice:
Or, constructive mathematics. Of course!
Olshen:
I said, “Errett, I would do exactly whatI’m doing. I would just be better at it because I’dbe smarter.”
Rice:
He must have been very disappointed bythat answer.
Olshen:
Disappointed? He didn’t believe me! Butthat’s what I think. I told him, I said, “I’d do exactlywhat I’m doing. I’d just be better at it.” He was veryupset; he didn’t like that at all. But I thought thatwas an honest reply. I think that a lot of peoplewho have jobs as statisticians of some form or otherdeep down believe that. That’s how they conducttheir lives. Unfortunately, it’s going to be an ongoingnecessity to justify ones existence as a statistician;but it is an honorable way to conduct your life.