The Best Learnings From One of The Most Influential Statisticians

[00:00:00] Alexander: Welcome to another episode of The Effective Statistician. Today, I have a guest that needs no introduction. I always wanted to say that. Stephen Senn. Hi, Stephen, how are you doing? 

[00:00:13] Stephen: I’m fine, thank you. 

[00:00:15] Alexander: Very good. So, as we are recording this summer [00:00:20] is finally starting. Here in Germany and also in Scotland where Stephen now lives.

[00:00:27] So thanks so much for agreeing to record this podcast. That has been on my list for quite some time and it’s great that we are now [00:00:40] really getting together. And so my first question is really about Your career, you have been in the industry for a very long time. You have worked at various companies, organizations.

[00:00:55] You also published quite a lot and to have, [00:01:00] you have a pretty big. Impact in the academic setting, both in terms of papers and but also in both in terms of books. So if you look back over your career, what are your biggest learnings in your career? 

[00:01:18] Stephen: Well, I would say the biggest [00:01:20] learning is the importance of luck.

[00:01:22] No, this is actually a serious point which I will maybe discuss later in the podcast anyway, but one shouldn’t underestimate the importance of luck. That’s not to say that, you know, everything is luck. That’s not true. But as we know from modeling in general, there’s a lot of what, which remains unexplained, and [00:01:40] we just have to call luck.

[00:01:41] In my case, I think I was, I was lucky on a number of things. First of all I came to, to Britain to, I wanted to study mathematics, but my background in Switzerland was not enough to prepare me for the British idea, which was basically that you should have studied maths and pretty much not much else from the age of [00:02:00] 16 to 18 if you intend to do mathematics at an English university.

[00:02:05] And so I looked for something else. And I found economics and statistics. And so I did that as an undergraduate degree. And I then found that I didn’t like economics. And so then I looked for something else to do because in the [00:02:20] meantime, at university in my third year, I’d met a first year student who is now my wife.

[00:02:24] And I wanted an excuse to stay on at the university. And the only course I could find, the only MSC course I could find for which I was remotely qualified was an MSC in computing and statistics. Okay. And so I did the MSc in computing and statistics and I discovered I don’t like computing. [00:02:40] So, so by process of elimination, I thought, Oh, maybe, maybe what I should be is a statistician, but I did not complete my MSc.

[00:02:49] I did not write up my project. And it was my wife later, my later wife, as we were married, who insisted that I go back and write up my project. And so some years after I should have graduated with an MSc, I [00:03:00] did graduate with an MSc. And that later enabled me to do a PhD and so forth. And if those things had not happened, I would not have had a career as a statistician.

[00:03:09] And there have been other things along the way where, you know, I made a lucky change of career and so forth. I started out working in the health service, did that for three years, then was a lecturer for [00:03:20] nine years in Scotland. In which I learned a lot about a lot of things, but not in a very deep way, because you were always having to lecture new courses and everything, and that was useful.

[00:03:30] And I enjoyed eight years, but I didn’t like the ninth year. I switched to the pharma industry, and I’d learnt my lesson, and after eight years, I switched again. [00:03:40] And I was a professor at University College London. And again, I learnt my lesson, and after eight years, I switched, and I went to Glasgow, where I was there for eight years.

[00:03:49] And then I switched, and I went to Luxembourg. And in Luxembourg, working in a research institute. I ran out of time after seven years. After seven years, I hit 65. I had to retire. I had no choice. I didn’t make my [00:04:00] eight year rule. But that’s basically what I’ve observed since then. And I feel I’ve been lucky because I’ve had experience in lots of different areas and managed to learn from a lot of very, very interesting and clever people about statistics as I’ve gone along.

[00:04:13] So that’s also partly been luck. 

[00:04:16] Alexander: Yeah. I luck is definitely [00:04:20] important factor. Yeah. I think that also calls for maybe a little bit more being humble about things. And also give a little bit grace to yourself. Yeah. If things don’t work out as fast or as good as you would like. To have in your career.

[00:04:39] Well, [00:04:40] sometimes, yeah, you don’t have the luck. Now. I also think opportunities come to those with a prepared mind. So yes, luck is one thing, but if you, if you change and you see this [00:05:00] opportunity, well, you also need to seize it. So what makes you kind of move and why did you switch? 

[00:05:11] Stephen: Well I think partly because I, I felt I set my first job in the health service.

[00:05:15] I learned some useful things, but I then came to the conclusion after three years that I wasn’t going [00:05:20] to progress or develop. And so I was looking for something completely different and I went to be a lecturer in it’s now a university, it’s now the University of Abertay Dundee, but at the time it was a, a, a college of technology.

[00:05:32] So I suppose, to be unkind, one would have said it’s a second class university, but let’s say a university with an emphasis on practical matters in [00:05:40] particular. And I learned a lot of statistics from that, also from a colleague, an older colleague of mine who’s since died, Gilbert Rutherford, who taught me a lot.

[00:05:49] I shared an office with him, so that was also lucky. So yeah, so one was prepared in a way to learn things. I sometimes found that I’d be told to give a course at short notice, and I [00:06:00] was then literally one week ahead of the students in learning stuff, you know, so it was, I had some quite, I had some classes where the students were older than I was.

[00:06:12] Yeah. So, you know, that also and of course in many ways they were much more knowledgeable about many things than I was. So you also have to be very, very [00:06:20] careful not to try and appear to be the all knowing all wise lecturer when, when that’s the case. 

[00:06:27] Alexander: I love that point. I think we should all embrace.

[00:06:31] more that attitude of constantly learning and knowing that we don’t know everything. [00:06:40] And that’s, that’s a, that’s another way to, to stay humble and to keep learning. So keep on learning and growing yourself. What is that? Is that one of the key reasons for you to kind of switch between different opportunities?[00:07:00] 

[00:07:00] Stephen: Yeah, I think so. That, that was one of the reasons. I think also yeah, learning opportunities. I think I was lucky at a particular time of my life when I went to work for Ciba Geigy. So it’s a forerunner company of Novartis in Baal, which is actually my hometown, Switzerland, although I had not lived there as a child, but officially I’m a Basler.

[00:07:19] [00:07:20] That was a wonderful opportunity and all sorts of practical problems presented themselves. But I was in a job where publishing methodological papers was not part of my job description. There were two departments in Sibagagi. One was, I used to say unkindly, Sibagagi had two departments. One was full of people with [00:07:40] lots of things to do and no time to think.

[00:07:43] And the other was full of people with lots of time to think and nothing to do, but that’s slightly unfair about the methodology department. But I was not in the methodology department. I was in the clinical trials department. But there were so many wonderful problems that came up all the time. And I thought, how am I going to do, what am I [00:08:00] going to do with this?

[00:08:00] And so I spent the evenings writing papers. So the papers were not written in company time, they were actually written in my time. But I had such wonderful examples that I was benefiting, and really I think it was lucky that I ended up in that situation, rather than in the methodology group, because having my nose [00:08:20] rubbed in practical matters all the time was really good for my development.

[00:08:23] I needed it. I needed to stop, you know, just imagining that. somehow fancy mathematical work or whatever was really what it was about. It’s part of it, but it’s not really what it’s about. 

[00:08:36] Alexander: Yeah. I love that kind of, if you [00:08:40] actually do the work the real challenges that prohibits you from working on problems that are not as important or not as relevant or you know, kind of ivory tower research and I think this is super important.

[00:08:58] Now, [00:09:00] you’ve wrote lots of lots of papers and also some books. So how did you do that? So, so what are your kind of recommendations for others who would also love to publish a lot? 

[00:09:14] Stephen: Well, I mean A lot of my early publications were short pieces, actually, funnily enough, commenting [00:09:20] on things that other people had written, which I disagreed with.

[00:09:22] In some cases, I think they were just technically wrong, although sometimes plausible. And I got into them simply because I was trying to repeat. I would sort of see something, I’d say, oh, that’s interesting, how does that work? And then I would sort of play around with it, try this, try that, try the other.

[00:09:39] [00:09:40] And then suddenly something would click, and I would say, well, maybe this isn’t right after all. And then I would try and find counterexamples or, you know, see where the problem was. And so that got me into publishing thought pieces where really there was no reason for the journal to refuse to publish them because [00:10:00] basically I was saying, look, this is not right.

[00:10:03] And, and then I gradually progressed to doing, to doing longer pieces like that. And then it was my wife again who suggested I write a book and I thought, well, what can I write a book about? You know, why? What am I? I’m an expert in anything. And she said, well, don’t you do a lot of crossover trials? And I said, Oh yes, I do.

[00:10:18] And she said, well, why don’t you write a [00:10:20] book about that? So, so I did. So that was yeah, so I don’t, I don’t know. It’s partly depends. Well, I have to say I’m very, very lucky because although I’m Swiss, my mother was English. So yeah. And English is my first language. And I think that I sometimes get somewhat annoyed with my British colleagues because they [00:10:40] underestimate the sacrifices that people who are not native speakers have to make in order to make the same progress in publishing.

[00:10:48] So there’s a terrific advantage. In being a native English speaker, in that sense, you know, I’ve always spoken English some of my life, I was born in Germany, in fact, in Saarbrücken, [00:11:00] so some of my life, I also spoke German as my other language, but I switched, and then some of my life, I spoke French as my other language, so the net result was that I ended up thinking in English, but that has been an advantage for publishing, no question. 

[00:11:13] Alexander: So when you actually write a paper, how do you start?

[00:11:18] Stephen: Well, a lot of it, a lot of [00:11:20] it would involve some mathematics and obviously to a certain extent one has to have solved the mathematical problems one wants to present. And so it will involve thinking about a particular problem. I’m working on a paper at the moment, actually with, with two other statisticians.

[00:11:38] And there, I [00:11:40] think that that’s been a back and forth between us, but a lot of my papers have been single author papers. And then, then yes, you’ve got to, you’ve got to have decided what particular problem you have solved. You’re stating in this paper. particular, you know, this paper, even if it’s not a mathematical problem, you’ve got to decide how to solve it.

[00:11:53] And the way in which I usually solve these things is I spend a lot of time thinking about them before I write down [00:12:00] anything. So you know, really to understand what’s involved. And then if I want to say something about analysis of covariance and what exactly are the key features of this and so forth, think about it and think about it.

[00:12:13] And then, then start writing. There has to be a stage at which you should write. I mean, you should write [00:12:20] the mathematical bit that you want to do, that you actually try and sketch out the ideas that you’re trying to present and get those sorted out. By the way, I also have a piece of software that I regularly use to help me solve mathematical problems.

[00:12:33] I’ve been using that since 1993, I think, or 94. [00:12:40] And that is Mathcad, which is a mathematical software package used mainly by engineers. There are very few statisticians who use it. But so I think maybe more statisticians might use Mathematica or also MATLAB to a certain extent, which is a slightly different function.

[00:12:58] And so that enables me to [00:13:00] do quite a lot of theoretical development. So I often, I often will create quite an elaborate Mathcad solution to a particular problem so I can see exactly how it goes. It’s very easy to read on the whole because the languages, although it’s a programming language, it looks just like mathematics.

[00:13:18] So if you want to do an integral, [00:13:20] you actually write down the integral. Okay. Look exactly as if it was in a paper. So that already helps in, in, in a certain way in in developing a lot of stuff. And also as regards problems, I always say that anybody who wants to be a serious statistician has to understand all solutions, two ways [00:13:40] at least.

[00:13:41] One is the formal mathematical way, this works because, and then you can work through a particular proof. But also there’s the intuitive way, if it doesn’t, doesn’t make any sense intuitively, what’s going wrong. 

[00:13:55] Alexander: Yep, yep, yep. 

[00:13:57] Stephen: And some, sometimes it’s taken me very late in my career [00:14:00] to, to, to realize this.

[00:14:01] For example, I, I realized some years ago that a common thing that’s said about sequential analysis in the frequentist mode is not correct. Sequential analysis doesn’t in fact bias in the way that some people think it does if you don’t adjust, provided that you wait [00:14:20] according to the amount of information produced.

[00:14:22] So if you were doing a meta analysis and you discovered that all the trials that are going into the meta analysis were in fact sequential analysis trials, some of them had finished early, some of them had gone to the end and everything like that, then in that case provided you weight them according to the amount of information, so the trials that [00:14:40] stopped early would.

[00:14:41] get less weight than the trials that came to the end. And in that case, you can show, and I have a paper in Pharmaceutical Statistics, that the biasing factors cancel out. They cancel out by weighting, because anything which is optimistically good is actually likely to have had less information, and is then down weighted by the trials that compete to the end.[00:15:00] 

[00:15:00] And what people have forgotten was, if a trial goes to the end, Then it can’t have been particularly good in the early phase, because otherwise it would have stopped early. And so once you mix the two types of trials together, equally, then appropriately, as in the fixed effect measurement analysis, then it cancels out.

[00:15:18] But this took me, I don’t know why it took me [00:15:20] so long to realize this. I suddenly felt good. I’ve been thinking about sequential analysis. Incorrectly for all this time. And that’s an example where the back of my mind I was worried about something. Because it didn’t make intuitive sense. 

[00:15:38] Alexander: Yep. Yep. Yep.

[00:15:39] Stephen: And then in [00:15:40] the end I could look at it. I give you one more example. In my book on crossover trials, even the second edition I write that trials in infertility are not suitable for crossover trials. And the reason I said that was that if a couple conceives in the first treatment Then in that case, obviously, I’m not going to try the second one.

[00:15:57] So essentially, to the extent that a [00:16:00] crossover trial infertility is, in infertility, is successful, it will actually not provide the period for, the information for both treatments. But then suddenly it occurred to me, well, this is very odd, because what am I doing? It’s like, as if I had a parallel group trial.

[00:16:19] And then I [00:16:20] said, right, okay, that’s the end of the parallel group trial. Now, those of you that it didn’t work out for, you can try the other treatment. Now, everybody says a parallel group trial is okay. So why should collecting extra information, once you have completed a parallel group trial, why should this be completely unacceptable?

[00:16:39] That’s illogical. [00:16:40] It must be possible, allowable to have extra information. The challenge is how would you use the extra information? and then I had a PhD student boy Cano ti that I set to work on the, on the problem. And we have a paper together explaining exactly how you can do [00:17:00] this. And now again, I tend to think, how could I have been so stupid to think that this was true?

[00:17:05] Alexander: Yeah, it’s very nice to hear from someone like you that first, it takes quite some time to overcome these challenges and said we can correct ourselves over time. Yeah. I think this [00:17:20] is really, really important. We have a development over our career. And we definitely make mistakes early on.

[00:17:27] We have wrong assumptions or misperceptions about things. But that shouldn’t stop us from correcting us later on. So. If you, [00:17:40] you mentioned kind of publish to solve a specific problem, now when I think about kind of these different problems, there’s very often kind of side problems or you can make some bigger or, you know, how, how do you [00:18:00] limit a paper so that it’s actually not going to too broad.

[00:18:06] Stephen: Oh yeah, I suppose that’s a, that’s a problem. And most papers tend to tend to end up being longer than I intended. It’s true. Yeah, it’s sort of annoying [00:18:20] sometimes because also sometimes you find that you you deliberately didn’t cover a sort of development from a paper because you sort of said, well, yes, that would happen. And it’s sort of obvious. And you know, anybody who wants to can join the, join the dots and you don’t do it.

[00:18:36] And then sometime later you find that somebody said, yeah, but what they [00:18:40] overlooked was, you know, such and such, you didn’t overlook it. You actually just decided not to put it in the paper. So, so, but nevertheless, it’s there is a limit, I think. Of course, most most journals have some sort of a formula or informally as to what they would accept.

[00:18:54] And sometimes you have to propose to have a part one, a part two paper or whatever, [00:19:00] and publish, publish something later. I think it’s Yes, it’s similar to although, you know, I can’t always say that I’ve kept my paper to a reasonable length I do pride myself and I don’t go over the time a lot of thought to me when I’m giving a lecture yeah, and That’s also a similar discipline [00:19:20] and I get really annoyed at conferences where the Chairman says something, oh, there’s just time for one quick question when in actual fact there isn’t time because the speaker has used up all the time that was allotted for the question.

[00:19:34] Alexander: Yeah. Yeah. I find this also super annoying, 

[00:19:38] Stephen: really, really [00:19:40] annoying. And it’s just not, it’s not, I mean, you know, the speakers have to be, have to be disciplined in that way. And the same is true, I think, to a certain extent with papers, there’s a, there’s a, and certainly when speaking, there is a difference between, which is something you have to understand is what you want to say and what your audience can hear.

[00:19:58] Can 

[00:19:59] Alexander: you elaborate a [00:20:00] little bit on that one? 

[00:20:01] Stephen: Well, I think that you’ll find that some speakers choose to, to, to stuff their head. Okay. their lectures with details, which really won’t come across in a, in a lecture and submission time. You’d have to have a super, super intelligent person, someone who was really up on the field, almost to the level that you [00:20:20] were to understand if you’re trying to try and get that across in 20 minutes or whatever you’ve been given or 45 if you’re, you know, some sort of invited speaker.

[00:20:28] And you have to ask yourself, well, what can they understand? And the other thing that you have to. have to think about is whether you really want to spend the time in covering the [00:20:40] algebraic derivation of a particular result, or whether instead you want to show them a plot of some function which incorporates what the what the result is, and to try and give them some intuitive understanding for, you know, why the result is the way it is.

[00:20:54] So that’s That’s attention there. And I think also in writing papers, you can do a similar [00:21:00] thing by deciding to put difficult material into an appendix, which I often think is a, is a wise thing to do, so that you don’t interrupt the main flow of the argument with intricate Derivation, which you can then leave to the appendix, but, but, you know, taste will differ, I suppose, regarding [00:21:20] that.

[00:21:20] Alexander: I completely agree that whenever you communicate, you need to take care of the audience. And you need to adjust to the to the different situations. Yeah. If you have a presentation, then communication is synchronous. Yeah. So [00:21:40] people kind of need to follow you at your speed that you’re giving the presentation.

[00:21:46] And if you’re too detailed, or if you’re too fast. It becomes nearly impossible for people to follow. And so that is an actually disservice to the audience. And [00:22:00] for me, that’s basically. bad presentations. Yeah. So providing an idea, providing kind of an overall understanding an intuitive kind of understanding of how the solution looks like is something that you can bring across.

[00:22:17] In a presentation [00:22:20] for asynchronous communication like a paper. Yeah. People can spend much more time. Yeah. And then of course, the other nice thing about the paper is that it’s not necessarily sequential, you know, like a presentation. [00:22:40] Well, it starts, it has a middle and it has an end. Well, a paper, you can, you know, flip around, you can read it and, you know, go back, you can go to the appendix, read another reference whatsoever, you know, discuss about it with some colleagues.

[00:22:57] So all of that you [00:23:00] can do with a paper you can’t do for a presentation. So always. Make sure that you adjust your communication to these kind of different things. I think it’s a great key point. Yeah. You have given lots of presentations over your career. What do you think [00:23:20] about the presentation skills of statisticians in general?

[00:23:25] Stephen: I’ve I’ve listened to a lot of good presentations as well. So I’m, I think on the whole statisticians are not too bad. I think there are some simple tricks. that one can learn. I was fortunate that when I was at [00:23:40] the Dundee College of Technology, I went on a sort of course about the technology of presentation.

[00:23:47] But remember that this was the era before computer projection. When we were preparing stuff on acetates, you know, if we were going to a, to a conference, it would be a pile of acetates [00:24:00] that you then, you then use and you presented them. But nevertheless, there was some useful things that were taught to me, some small tricks that people overlook.

[00:24:08] One strange thing is the use of typeface. A lot of people imagine that using block capitals makes things easier to read. Oh no. This is actually the opposite. And [00:24:20] they’re confusing two particular things. One is if you write out something in handwriting, then usually you can read someone having written in block capitals quicker than you can read their own handwriting because their handwriting is idiosyncratic and you have a much bigger interpretation task.

[00:24:36] in the, in the handwriting. But actually when it comes to print, it reverses the [00:24:40] case. And the word shape is much, much clearer if you’re not using uppercase. And so you find people we’ve got a slide for for the audience and everything on the slide is in uppercase, and I think, why, why are you doing that?

[00:24:54] This is really stupid, you know? You should use lowercase and, and it’s quicker to read. And there’s been [00:25:00] quite a lot of research done, done on that. And so this is something that was hammered home to me when I was when I was a young lecturer in this institution, which took, took teaching quite seriously.

[00:25:10] And that’s one point. The other point is the importance of sticking to time, as I said. I think the point another point which I think statisticians are [00:25:20] quite good at is you should use you should try and graph things as much as possible. So, you know, for example, a typical error that people make, I think, in presenting simulation material.

[00:25:32] I also think there’s far too much simulation used, by the way. Not enough theory. There should be more theory, less simulation. But anyway. In presenting [00:25:40] simulation material, they give you, they give you a table which is full of figures. And they say, and I appreciate that this is hard to read, but, you know, well, if you appreciate it’s hard to read, why did you put this up in the first place?