Home » A Statistician's Life, Statisticians in History

An Interview with Howard Wainer

1 September 2018 1,635 views No Comment

Sam Behseta

Howard Wainer gives a talk during his book signing (Uneducated Guesses) at the Princeton Library.

Howard Wainer gives a talk during his book signing (Uneducated Guesses) at the Princeton Library.

 

Howard Wainer, who writes the Visual Revelations column for CHANCE magazine, is distinguished research scientist (retired) at the National Board of Medical Examiners. He has won numerous awards and is a Fellow of the American Statistical Association and American Educational Research Association. His interests include the use of graphical methods for data analysis and communication, robust statistical methodology, and the development and application of generalizations of item response theory. He has published 20 books so far; his latest is Medical Illuminations: Using Evidence, Visualization & Statistical Thinking to Improve Healthcare.

Howard Wainer’s Visual Revelations column—appearing in CHANCE magazine, which is published by the ASA—is arguably the longest-running serial publication in the history of statistics: 25+ years of delving into the core events of the time, all from the point of view of a razor-sharp, uncompromising, and greatly responsible statistician. Now, that is a career unto itself, but then there are some 400 articles, volumes of books, and years of leadership at the Educational Testing Service and National Board of Medical Examiners—not to mention a long teaching career at the University of Pennsylvania. In recognition of his extraordinary contributions to CHANCE and the profession in general, former CHANCE Executive Editor Sam Behseta sat down with Howard for a friendly chat.

Sam Behseta: Howard, how did you become involved with CHANCE?

Howard Wainer: I got a call from Steve Fienberg after he and Bill Eddy started CHANCE. He knew of my interest in graphics and said there’s a column called Visual Revelations that Alan Paller had originated. Alan had written the first year or two’s worth, but soon he was to be leaving, so Steve asked if I would take it over for a while. I thought about it. I had ideas for about three columns, and thought that should carry me through most of the first year. And, surely, I would be able to think of something else for the fourth one. So, I figured I could certainly take it for a year and, with luck, I could get through two. So, I agreed.

Every year since, I’ve thought, “Well I’ve got another idea for one more column,” and so it just kept going. A bit like mathematical induction—if I had n-columns, I could always dream up one more. Anyway, I have enjoyed it so much that I just kept going. It has quickly been 25 years.

Sam Behseta: So, what’s the process for you? How does this work? I mean, obviously, you look around and observe and you have your professional stuff, but then how does it work? How do you decide a piece is suitable for CHANCE?

Howard Wainer: To some extent, it’s like relying on the kindness of strangers. I’ve relied on the kindness and wisdom of the editors, and the editors of CHANCE have been both wise and generous over the course of all these years. I figure about one idea out of every 50 is any good. Some people get one idea a week, and so they get one good one a year. John Tukey said he used to get 50 ideas before breakfast and, although even he—he claimed—had only one good one out of 50, it didn’t matter because he had plenty of them.

Wainer and his wife, Linda, on Hurricane Mountain in the Adirondacks

Wainer and his wife, Linda, on Hurricane Mountain in the Adirondacks

 

I don’t have 50 before breakfast, but I have a bunch of them. But there’s a filtering process that goes on. I pick one I think will be good, and then I draft something and send it to the editor and ask if it’s suitable? Obviously, I’m relying both on wisdom and generosity for the editor to say, “Yes, it is” or “No, it isn’t.”

Sam Behseta: Is there any cause célèbre you see as a source of future topics?

Howard Wainer: I’m becoming more convinced it’s going to be useful to take a more publicly Bayesian point of view now. I’ve been reading more about what is called the prosecutor’s fallacy. Basically, it’s calculating the wrong probability and not knowing how to do the Bayesian flip (or not realizing it should be done). Maybe that will be worth describing.

But the image of the initial audience for my column I have in mind is often that of the Stat 101 teacher who’s trying to present an idea to their class to illustrate a statistical principle. It has to be something they can do in one class without much math beyond algebra. My vision is that an instructor could go through the various columns and pick out a topic—a case study they can use to illustrate something the students can read and get some idea of how exciting it is to be in the field we are in.

Sam Behseta: What I noticed in your pieces is there is a sense of justice. You’re interested in social justice, largely speaking. Where does that come from? Do you feel a responsibility to comment on social issues? Or does it just come with it?

Howard Wainer: I’m not sure. If I had to say I had a religion, it’s the worship of evidence. You look at the evidence and follow where it leads. I guess that’s what you might call a weak prior in some important sense. In addition, my parents were children of the Depression, and so FDR was much closer to being a deity than a politician. So, the 12th commandment, certainly in my family, was thou shalt not vote Republican. So, I more or less absorbed democratic as well as Democratic ideals as part of mother’s milk.

Wainer's boat: Surprise.

Wainer’s boat: Surprise.

 
Sam Behseta: So, you were in Chicago in the ’60s? I was talking to Steve Fienberg and he was saying you were more or less there at the same time.

Howard Wainer: Yes. We came pretty much together. He was there a year ahead of me in ’69, and I came in ’70.

Sam Behseta: But you didn’t study statistics?

Howard Wainer: I was a faculty member, not a student. I was in something called the Committee on Methodology in the department of behavioral sciences.

I met Steve early in the fall of ’70; I had just arrived a few months before. They recruited two young assistant professors in statistics to count the votes in the faculty senate election. They use the Hare system, which is complicated because it is iterative. So, there were Steve and I with all these ballots. We had to arrange them in certain ways before we could start calculating the outcomes via the Hare system.

While we were doing this, I was complaining mightily about what a dopey waste of time this was. Steve turned to me very sternly and said, “Stop joking around. This is important—it’s faculty politics.” I knew at that point he and I were headed in different directions in our careers. Of course, the intervening 45 years have proved me completely correct.

Sam Behseta: Among all the pieces you’ve written in the last 25 years for CHANCE, do you have your own greatest hits?

Howard Wainer: I’m not sure. I’d have to look at all of them. How do you pick your favorite child? But I will say something else that is allied to this. I believe writing this column for CHANCE was the single best thing I’ve done in my career—for my career. Because what it’s done is forced me to write something intelligible every three months. As you know, I have collected 15 to 20 of these essays together every few years, reorganized them, rewritten them a little bit, written some interstitial material, and published a book. This has worked for four or five books so far. I don’t think I would have done this without CHANCE.

Sam Behseta: Meanwhile, you’ve been teaching constantly.

Howard Wainer: Well, I would say teaching intermittently. I’ve taught for the last 10 years—one course a year at Penn. I taught for a couple years at Princeton. When I was actually in the academic business, I was at Chicago and taught there full time.

But I view teaching the same way I view cooking. I enjoy cooking one meal a week. You can plan it, shop for it, work hard at it, put your heart into it, and then watch the enjoyment other people take in your efforts. Twenty-one meals a week is not for me. I think then you’re just throwing the stuff on the table.

I’ve taught one course a year for the last 10 years. I’ve worked hard at it, and I’ve enjoyed it immensely. The time came this year to cut back so I thanked Penn for putting up with me for all this time and stopped. They have a really wonderful department. I don’t know who ranks such things, but I would put them in the highest tier.

I’m honored that they let me be a small part of the department for this past decade. But now the time has come to stop. There’s a separation of three generations now between the students and me. They don’t get my jokes or many of my references. The students starting there this fall were born in 1995; most of my sport jackets are older than that. I have had fully adult women in my classes who were too young to date my children. That signals it’s time to stop.

Sam Behseta: Have you ever tried to use your pieces in CHANCE in your classroom? And what kind of feedback have you received?

Howard Wainer: It’s always been positive. In fact, there are several pieces that were coauthored with students. There was a piece on exploratory data analysis I did with a student named Danielle Vasilescu that was about overcrowding among nations. Then there was a piece I did with another student, Grace Lee, about the choice of independent variables required by dating services. It began when students asked why some dating services asked to include the length of your forefinger. No one could figure out why, but some of the girls were saying there was an old wives’ tale about the relationship between the length of the forefinger and the lengths of other body parts. Others suggested men often lie about their height, and finger length might provide a check on it.

Perhaps the dating service was going to use the length of the forefinger (Who’s going to lie about that?) as a proxy for height. So, Grace went and gathered forefinger lengths and heights of a number of students, looked at the correlation, and calculated the standard error. It turned out that finger length isn’t much of a predictor, because the standard error of the estimate of your height from your forefinger yields bounds of about plus or minus six inches, and that wasn’t helpful.

Sam Behseta: One thing I tell my students is that statisticians should be able to write well because that’s what they do. They analyze data and then they write about it. And it often comes as a surprise to them because their thought process is, “I’m in a quantitative field; I’m not in literature or anything like that.” What’s your thought on that?

Books and More Books
Since this interview, Howard Wainer has retired, and we wanted to know what he has been up to. Of course, it involves books.

Name one or two favorite blogs or books you have read.

The only blog I ever read is Andrew Gelman’s, but even that only episodically. The top four books currently on my bed-stand are:

1. Ari Shavit’s marvelous My Promised Land
2. Volume 18 in Patrick O’Brian‘s spectacular Aubrey/Maturin 20-volume series of continuing sea-faring novels
3. Cass Sunstein’s Impeachment
4. Walter Isaacson’s Leonardo Da Vinci

What have you been doing to fill your time since you retired?
Nothing has changed except I no longer commute to Philadelphia (I don’t miss it) and I no longer receive a pay check (I miss that).

I have been working hard at completing the graphics history book, and I spend a fair amount of time as an expert witness (mostly in trials of teachers accused of helping their students cheat). Since there is no place in heaven for those who prosecute teachers, I always defend. School districts always seem to make the same statistical errors in conducting their investigations. I have been thinking of putting together my testimonies into some sort of guide for both teachers’ unions and school districts.

I also spend a fair amount of time in the gym and, when the weather is fine, sailing on the Surprise (my Alerion 28).

My new book [with Michael Friendly] was tentatively titled On the Origin of …, but that title is almost certainly to be changed (too Darwinian). I don’t know what it will be. It is planned to be published in April 2019 (if Friendly and I have the wit and energy to get it done by then). It is sort of a history of graphics—told mostly as a series of stories about the origins (and originators) of graphic forms. We begin with cave drawings of mammoths from Paleolithic times and how they influenced our understanding of what mammoths looked like and move forward through the 17th century (van Langren’s one-dimensional plot of longitude) to Playfair, Minard, Snow, and many others. We also include W.E.B. DuBois and his collaboration with Minard in graphing the Great Migration. There is also a very nice chapter, principally Michael’s, on the history of graphing motion that shows how we finally understood how horses run and cats land on their feet.

Howard Wainer: It’s come as a surprise to me, too. When I applied to college, I eliminated all colleges that required an essay. So, I ended up going to Rensselaer Polytechnic Institute. I hated writing because I wanted to do math. So, the joy I take in writing now astonishes me. But of course, I think one’s success as an academic is related to many variables. Horsepower, of course, focus, grit, energy, good taste in problems, and the need to tell stories. You have to tell stories.

You have to construct some kind of narrative around what it is you’re doing. No one’s going to pay attention to your stories unless you’re interesting and grammatical. And of course, the more you write, the more practice you get, and the better you are at this. When I was at Chicago, one of the advantages of being a faculty member was that you could sit in on any course you wanted. So, when I learned there was a course on writing Saul Bellow gave, I jumped at the opportunity to audit it.

One of the things I’ve observed about mathematicians is most of us are open to criticism of our mathematics, but not our writing. The response is usually, “Who are you to criticize my writing?” But taking this course with Bellow, the “Who are you?” is easy to answer. He has a Nobel Prize, a Pulitzer, National Book Awards, etc. So, I would write things, and he would give them back to me with comments. I would pay attention.

He always referred to writing as his craft, never as his art. He explained the reason he has had success is that he worked hard at it. Of course, he also has some game. So, I work hard at it. I write and rewrite. It’s fortunate I enjoy reading my own work so much that I can reread the same thing several times and constantly pick away at it to try to find le mot juste. So, I emphasize the importance of clear writing to all those with whom I work.

Reading broadly is sometimes daunting. For when you read something written by a real pro, the tendency is to throw up your hands in surrender and think, “I could never do that.” That may be true, so you have to focus on clarity instead. Style may come later, but clarity must come first. There are lots of wonderful statisticians who wrote terribly. How many papers/books begin with such catchy phrases as “Let I be an index set.” A long way from “Call me Ishmael.” John Tukey was perhaps the most influential statistician of his time, but no one would confuse his writing with Hemingway’s.

Sam Behseta: Did you know him personally? How did you interact with him?

Howard Wainer: I was a student and he was a faculty member. And then I subsequently worked with him for about 30 years. When I was at ETS, I would see him frequently. He was a remarkably generous man. And so, when I had an especially thorny problem, I would go talk with him about it. He’d always have something valuable to offer.

One important lesson he taught to so many of us was about the evil of hubris. He said hubris is a bad thing because—he didn’t say this, but it was certainly implicit in what he was saying—because people are—the phrase he used was—“people are different.” And you don’t learn anything if you don’t both listen to other people and respect what they have to say. He was probably the best hubris destroyer you could imagine. You could go talk with him about anything and quickly realize you’re not as smart as you thought you were. John frequently understood your problem better than you did before you arrived, but if he didn’t, he certainly would before you left.

As nearly as I can tell, John knew pretty much everything. There’s a wonderful story about him that I believe comes from Colin Mallows. John would sometimes get annoyed because people would ask him ridiculous questions just to see if he knew the answer. So, Colin suggested, if you want to find out how to milk an elephant, don’t ask John directly. Just come in and start talking about elephants in general and eventually, when the conversation gets around to how to milk one, he’ll tell you.

Sam Behseta: Were you involved in the whole discussion between Bayesianism and frequentism, and did you take sides early on, or did it not matter to you?

Howard Wainer: No. I have gradually become what is pejoratively referred to as an opportunistic Bayesian. That’s if it works, I will use it, and if it doesn’t, I won’t. But certainly, my training was never from a Bayesian perspective. In fact, only recently did I learn why John never seemed to be a fan of Bayesian methods. This attitude was a mystery to me. I couldn’t understand why Tukey wasn’t more positive toward Bayesian methods as they became more practical.

Recently, the book The Theory That Wouldn’t Die, about Bayes’ Theorem, explained that Tukey was using Bayesian methods during the Cold War period on matters of national security, but he chose not to make it public because it conferred a tactical advantage to the United States. I guess that carried over into the predictions of election outcomes for NBC as well.

But I think he probably, and this is just a surmise on my part—I don’t pretend to understand him—but I have a feeling he was against dogmatism because it stood in the way of the flexibility required for a first-rate scientist. And many of the Bayesians of the period were certainly dogmatic. So, he might have been in favor of Bayesian methods, but not Bayesians. If you remember statistics in the ’50s, it really was almost warfare. My friend Sam Savage tells how his 9-year-old brother was accosted at some Stanford faculty party by somebody who told him, “Your father is deeply deluded.” He was talking to a 9-year-old and yet he felt it was important to point out the moral weaknesses of a Bayesian attitude.

Sam Behseta: I guess you had to identify yourself with a camp back then. And it didn’t make much sense because I think, at some juncture, it was an ideological warfare between the two camps.

Howard Wainer: Well, it’s almost over now. Because there are problems you can’t solve any other way.

Sam Behseta: At this juncture, with the whole big data and the immediate interest in solving problems, which are at the core of science, it is possibly more efficient to be a pragmatist.

Howard Wainer: Yes. The Shibboleth is, “Does it work?” Fifteen years ago or so, I teamed up with a very young Eric Bradlow (who has since become an eminent professor at Penn). He was a dyed-in-the-wool Bayesian, and he taught me a great deal. We did a book together that was a fully Bayesian approach to doing test theory, and it forced me to learn (and eventually love) the stuff. At the time, I wasn’t fully conversant with the estimation methods that made Bayesian procedures practical. MCMC [Markov chain Monte Carlo] became my favorite initials.

Sam Behseta: When I talk to PhD students, sometimes they think every PhD in statistics, when they finish their education, they ought to go into academia. But my story is, listen, actually that’s a very small minority. A good chunk of PhD graduates will end up in other sectors and can be extraordinarily effective in shaping policies. How rewarding has the experience of working outside academia been to you?

Howard Wainer: Well, I’ll tell you. I certainly came out of graduate school with the same point of view that you expressed—I was training to become a professor. For no good reason, other [than] who is doing the training—professors. It also seemed like a nice life. So that’s the direction I wanted to go.

I found, as I pursued this path, there were aspects of being an academic that are very attractive—certainly the teaching part. The part I didn’t like, because it was too difficult—at least for me—was discerning what were good problems to work on. When I first got out of graduate school, I went to Temple University and I was unhappy. I thought, well, that’s because Temple isn’t Princeton. I left quickly and got a job at The University of Chicago. Chicago is as good a university as you can find. Chicago treats its faculty as well as you can expect to be treated. It’s a wonderful place, and the colleagues I had were terrific. And still I wasn’t happy.

I found that there was an enormous amount of intellectual horsepower being used to solve trivial problems. That wasn’t everybody, obviously. There is certainly wonderful and important work that comes out of academics, but since I had yet to develop good taste in the choice of problems, the university life was ill suited to me—at least at that time. So, I left Chicago and went to Washington. I worked there during the Carter administration. Toward the end of my time there, I met with Don Rubin, who was—at that point—preparing to go to work for the EPA [Environmental Protection Agency] and was assembling an all-star cast to go with him. He had already recruited Paul Rosenbaum and Rod Little, and it was flattering to be considered part of such a group. And Don had some interesting ideas. But the more we spoke about it, the clearer it became that what I was interested in doing was essentially substituting (never replacing) for Don at the Educational Testing Service (ETS), which he was leaving to go to EPA. And so, I went in that direction. I even rented Don’s Princeton house from him while he lived in Washington.

At ETS, we had real problems—real in the sense that if you solved one, it could have important and immediate consequences. For example, there was one problem having to do with improving the efficiency of the shipping of tests to the hundreds of testing centers. You can’t send them too early or send too many extras because of security issues. You can’t send them too late because, if they arrive late, that doesn’t help. How do you know how many to send? At the time, they had ad hoc rules in which, on the last day that would allow normal shipping, they would send the number of tests equal to 10 percent more than the number of people who had so far registered for that particular site. If more were needed, they would overnight them at the last minute. With hundreds of testing centers, this was expensive.

Paul Holland was looking at this problem and said if X is the number who registered by shipping time last year and Y is the number that showed up eventually, we need a rule to predict Y from X. Statisticians know how to do this. It turned out that regression equation saved ETS some number of hundreds of thousands of dollars every year. That’s a nice reward for doing a regression.

ETS, in their generosity, then gave the statistics group an annual grant of a fairly large number of thousands of dollars a year to support whatever research we felt was worthwhile within our group. You can see how such work could be rewarding—and we actually did other things of broader interest. There were all sorts of important and interesting problems that appeared because ETS was a data-rich environment with a lot of resources. So, I found it to be an advantage, at least for me.

Sam Behseta earned his PhD from Carnegie Mellon University and is a professor at California State University, Fullerton. His main research area is statistics in neuroscience. Behseta has both trained and mentored undergraduate and graduate students.

Editor’s Note: This interview originally appeared in CHANCE, Volume 28, Issue 1, Winter 2015.

1 Star2 Stars3 Stars4 Stars5 Stars (No Ratings Yet)
Loading...

Comments are closed.