Diego, Tom, and Thomas, South Queensferry, September 2016
DECEMBER 2016 Issue of ISBA Bulletin
Me, Arnold Zellner, Gordon Kaufmann, Wes Johnson, Carl Morris, and Shanti Gupta, First ISBA CONFERENCE, Hotel Nikko, San Francisco, 1993.
FROM THE EDITOR
Did you miss the email about the ISBA election results? We have them in this issue. Wonder how ISBA@NIPS went? Also in this issue! A final highlight is the interview with one of the ISBA founders, Thomas Leonard. He was also the first editor of this publication, then called the ISBA Newsletter. This follows on from last issue’s interview with another of our former editors, Manuel Mendoza.... I’ll start making notes for when I am interviewed in 25 years time! But in all seriousness, these interviews are an important way of preserving the history of our society. This was sadly driven home, as this final issue for 2016 was being put together, by the passing of two ISBA fellows, Hajime Wago and Stephen Fienberg. We have included brief obituaries, and hope to have some longer reflections on the impact of these important Bayesians in the March issue. In the meantime, while we did not have an ISBA interview with Stephen Fienberg, there is a fascinating interview with him from Statistical Science in 2013, which you can find here: https://arxiv.org/pdf/1310.2442.pdf. Both he, and Thomas Leonard in this issue, have thoughts about how our expertise and professionalism as statisticians can make the world a better place–a fitting topic to contemplate as we enter a new year. With my best wishes for the year to come, your Bulletin editor, Beatrix Jones
Beatrix Jones
IN MEMORIUM, STEVE FIENBERG
Stephen E. Fienberg, a fellow of ISBA and friend to many members, passed away December 14, 2016, in Pittsburg, Pennsylvania. He was born in Toronto, Ontario, in 1942. After completing a Bachelor’s degree at the University of Toronto he went on to obtain a Masters and PhD from Harvard under the supervision of Frederick Mosteller. He worked at the University of Chicago and University of Minnesota before taking a position at Carnegie Mellon University in 1980, where he spent the remainder of his career. By that time he was well know for his work on categorical data, but also for interdisciplinary work in the Social Sciences.
Steve Fienberg was my academic friend, I first met him in 1977 during a conference on Categorical Data Analysis in Newcastle, which was organised by Robin Plackett, He will be surely missed,
EXCERPTS FROM MY ISBA INTERVIEW (from page 11)
. I tend to receive lots of ridicule whenever I cite my Technometrics1975 paper, often out of the blue. In fact, Steve Fienberg made a meal of it during his after dinner speech at Valencia 6, just before I played Rev Thomas Bayes returning from Heaven in the comedy skit. As a special case, my probability model reduces to a time series model which represents a sort of stochastic volatility, though at least one worthy Bayesian has laughed his head off at this suggestion, and another firmly denied it when I suggested a touch of d´eja vu.
EUGENICS AT UCL
That takes us to the polymath Sir Francis Galton. He was a very great man. He may have been the first person to do the conjugate analysis for the normal distribution with specified variance. He also invented a very strange sort of machine; which tried to calculated posterior probabilities from prior probabilities via a simulation process. That reminds me of modern day MCMC. However, because of recent protests by students, minority students in UCL, I’ve come to realize that in 1883 that Galton coined the name for the subject of Eugenics, while planning to improve the quality of the human race. Now he may well have had good motives, of course, but what happened because of his work is really quite traumatizing. Florence Nightingale, was instrumental in setting up the Department of Applied Statistics at UCL in 1911. She died in 1910, but she was instrumental in appointing Karl Pearson, to the chair rather than a more theoretical statistician of the Galton school. When I was at UCL, I was taught to revere Karl Pearson and Ronald Fisher. Fisher was professor of Eugenics at UCL and so like Pearson and of course Galton before him, was associated with the Galton Laboratory. A lot of Fisher’s work in Genetics concerned Eugenics. A lot of Pearson’s work in Statistics which was published in Biometrika concerned Eugenics. In other words, comparisons were made of the attributes of different ethnic groups. I don’t know whether their motives were good or not but what came out of the subsequent Eugenics movement around the western world during the 20th century is absolutely terrifying, for example, forced sterilizations, racial discrimination, the CIA mind control program MK Ultra, and genocide. Karl Pearson’s son Egon continued the collaboration with the eugenicists in the Galton Lab after his father’s death. When I was at UCL there was still a professor of Eugenics called C.A.B Smith, and there was a mysterious lady who walked between the Lab and the Statistics Department every day. And the University archivist was a statistician who’d published several joint papers in Eugenics! All of those terrible consequences! They originated from Galton’s conception and the work of the two Pearsons and Fisher. I find it quite disturbing to think this.
PERSPECTIVES FOR THE FUTURE
Diego We are arriving to the last part of this interview. How would you advise young academic statisticians who are just starting their career?
Thomas: The first thing is to feel confident in your own ideas and not to be pushed over by anybody else’s preconceptions. When talking to your Ph.D. supervisor or your senior colleagues, you should always be prepared to assert that you think this might be a better way to do it and to discuss it rather than just be told what to do. That’s what is very important; you’re responsible for your own originality. The other thing, I might sound very traditional on this but it’s very important to develop the relevant mathematical expertise to the subject. Dare I emphasise that you need to develop the ability to do the algebra and advanced calculus yourself, rather than using Mathematica to do it, so that you can manipulate and play with it, and that includes using Laplacian approximations or Taylor series expansions, completing the square, and whatever. Certainly, in America it is important to have expertise which I never really possessed, but it is good to be able to prove asymptotic theorems for yourself. It’s the American tradition. But please remember that even saddle-point accuracy is an asymptotic criterion; it doesn’t imply finite n accuracy. Don’t get lost in the asymptotics as many have who have gone before! Asymptopia isn’t the Utopia which the Berkeley school make it out to be. Computational skills are also important. You should be prepared to do your own computer programming. I think that it is important to veer away from Bayesian computer packages since you never know what is really going on inside them. You are totally responsible for the numerical results which you publish, and you should develop your own talents while producing them. It’s also important to take time out to get acquainted with basic statistical principles. You could start off with the latest edition of the book by Box, Hunter and Hunter, Statistics for Experimenters. It is important to realize that the quality of the data helps determine the quality of the conclusions. It is important to avoid small, unreplicated studies which yield apparently significant conclusions. You should try to replicate your results as many times as possible. Please do not selectively report results or fudge or shuffle your data in order to please your superiors, or to get a paper published. Give due, even slightly overgenerous, credit to others when credit is due even when this affects the apparent originality of your work. You get back what you give out. I believe that there are people in the higher echelons of Society who attempt to control Statistics while trying to control ordinary people everywhere. It is important that you try to assert yourself while attempting to report the most honest conclusions which you can from the data. The same data set can yield entire different conclusions (e.g. about outliers) depending on where it’s coming from. Think how the data was collected when you analyze it. You might decide to be interdisciplinary like me or you could decide to focus on one or two areas of application. So it’s essential to maintain the honesty and integrity of the statistical paradigm. If done badly, Statistics can be used to confuse the population. Bad statistics can used to control ordinary people with misinformation. I fear that it may already be like that. One thing happening is that there is a huge expansion of Statistics in Big Data analysis. Lots, lots more people from different disciplines are regarding themselves as statisticians. This is in a sense quite tricky because they then have different objectives. Please don’t clean the data before you’ve scrutinized it in relation to its real-life background. You may be ridding yourself of the key messages in the data.
I very much fear the current strangely indecisive situation in the world today. I believe that good Statistics can be used to help change the world for the better. It can for example substantiate suspicion as to what is actually going on. Please be sure to use your expertise in Statistics fairly and wisely.
Diego: Thank you Thomas that’s an inspiration for all of us. Many thanks for this interview.
We would like to thank Diane Ruiz and Thomas Tallis for helping us with this interview.
Diego and Diane Ruiz
No comments:
Post a Comment