THE ISBA BULLETIN

Vol. 17 No. 3 September 2010

The ocial bulletin of the International Society for Bayesian Analysis

AMESSAGE FROM THE membership renewal (there will be special provi- PRESIDENT sions for members who hold multi-year member- ships). Peter Muller¨ The Bayesian Nonparametrics Section (IS- ISBA President, 2010 BA/BNP) is already up and running, and plan- [email protected] ning the 2011 BNP workshop. Please see the News from the World section in this Bulletin First some sad news. In August we lost two and our homepage (select “business” and “mee- big Bayesians. passed away on Au- tings”). gust 6, and Arnold Zellner passed away on Au- gust 11. Arnold was one of the founding ISBA ISBA/SBSS Educational Initiative. Jointly presidents and was instrumental to get Bayesian with ASA/SBSS (American Statistical Associa- started. Obituaries on this issue of the Analysis tion, Section on Bayesian Statistical Science) we Bulletin and on our homepage acknowledge Juli- launched a new joint educational initiative. The an and Arnold’s pathbreaking contributions and initiative formalized a long standing history of their impact on the lives of many people in our collaboration of ISBA and ASA/SBSS in related research community. They will be missed dearly! matters. . . Continued on page 2.

ISBA Elections 2010. Please check out the elec- tion statements of the candidates for the upco- In this issue ming ISBA elections. We have an amazing slate ‰ A MESSAGE FROM THE BA EDITOR of candidates. Thanks to the nominating commit- *Page 2 tee, Mike West (chair), Renato Martins Assuncao,˜ Jennifer Hill, Beatrix Jones, Jaeyong Lee, Yasuhi- ‰ 2010 ISBA ELECTION *Page 3 ro Omori and Gareth Robert! ‰ SAVAGE AWARD AND MITCHELL PRIZE 2010 *Page 8 Sections. ISBA now has a Objective Bayesian ‰ (ISBA/OB). The new section was enthu- ARNOLD ZELLNER 1927-2010 Section *Page 9 siastically approved by the board. The section of- ficers are Jim Berger (chair), Dongchu Sun (pro- ‰ JULIAN BESAG 1945-2010 gram chair), Jaeyong Lee (secretary) and Bru- *Page 10 nero Liseo (treasurer). Thanks to the founding ‰ ANNOTATED BIBLIOGRAPHY members for taking the initiative to launch this *Page 12 new section. The long running series of bienni- ‰ SOFTWARE HIGHLIGHT al O’Bayes workshops is one of the most exciting *Page 15 regular Bayesian meetings. It was embarrassing ‰ that this happened without ISBA. Thanks to the STUDENTS’ CORNER *Page 17 founding members of the new section to bring this active area of Bayesian research into ISBA. ‰ NEWS FROM THE WORLD We welcome the new section and invite members *Page 18 to join by simply checking a box in the annual ISBA Bulletin, 17(3), September 2010 BAYESIAN ANALYSIS - A MESSAGE FROM THE EDITOR

MESSAGE FROM THE PRESIDENT, Continued Savage Prize and Mitchell Prize Nominations: from page 1. ... One of the most visible ISBA initiatives is the We hope to eventually jointly sponsor short Savage Prize, awareded every year for the best courses and other activities in keeping with our Bayesian dissertations. Please consider nomina- shared mission of promoting Bayesian analysis. ting outstanding recent Ph.D. theses! Any ISBA We would welcome any other interested profes- member can submit nominations. The perhaps sional societies, including in particular other na- biggest award is the invitation of all finalists to tional societies to join the initiative. Marina Van- present their work in a super prominent session nucci (SBSS) and Peter Hoff (ISBA) have kindly at the upcoming JSM meeting. The Mitchell Pri- agreed to serve as representatives of the two so- ze is awarded for outstanding applied Bayesian cieties to coordinate any upcoming activities. papers. Please consider nominating eligible pa- pers that you enjoyed reading! Please see the no- te elsewhere in this Bulletin and our homepage www.bayesian.org (click Prizes”) for details.s

AMESSAGE FROM THE EDITOR butions. Other regretable losses are those of the always inspiring Prof. David Blackwell who pas- sed away on July 8 and Prof. who Manuel Mendoza passed away on August 7. An obituary for Prof. [email protected] Blackwell, by Robert Sanders, can be found at the UC Berkeley website. As for Prof. Nelder, a beau- During the last quarter we all have witnessed tiful obituary by Prof. David Hand can be found how the Bayesian Universe has continued its ex- at Imperial College’s website pansion. Our regular activities have been gro- In any case, life continues and this issue inclu- wing; new initiatives were announced and the des most of the usual sections. I might call your future appears full of exciting challenges. It must attention to the information regarding the forth- recognized however, that some sad news have al- coming ISBA elections. We have an impressive so arrived with the end of the summer. list of nominees and this material will help you As our President has mentioned above, Prof. to make your decision. Arnold Zellner passed away on August 10. Four As always, I want to encourage all members of days before, on August 6, Prof. Julian Besag also ISBA to contribute to the Bulletin with their sug- passed away. This issue includes a note on each gestions, manuscripts and announcements. Plea- one of these distinguished colleagues. We are de- se do not hesitate to contact me or any member eply thankful to Prof. Siddharta Chib, Prof. John of the Editorial Board.s Geweke and Prof. Peter Green for these contri-

BAYESIAN ANALYSIS - A MESSAGE FROM THE EDITOR

UPDATE FROM BA analysis of large datasets using mixture models, by Ioanna Manolopoulou, Cliburn Chan, and Mi- Herbie Lee ke West. Discussions by Fabio Rigat and Nick Editor-in-Chief Whiteley provide some additional ideas and di- [email protected] rections for thought. The remainder of the issue contains six other fine articles in areas of Bayesi- The September issue of Bayesian Analysis fea- an computation and modeling, and an applicati- tures a discussion paper on the practical Bayesian on to hydrology.s

Content 2 www.bayesian.org ISBA Bulletin, 17(3), September 2010 2010 ISBA ELECTION

2010 ISBA ELECTION

Candidate Statements first on nonparametrics and the BISP series on Merlise Clyde stochastic processes). [email protected] The President-Elect will be in charge of IS- BA2012 in Japan: it is a great opportunity to ex- The 2010 ISBA nominations committee, chai- pand ISBA activities well beyond its stronghold red by past-President Mike West, has assemb- (North America, Western Europe and Australia). led an outstanding slate of candidates for Pre- Our focus should be in promoting local chap- sident Elect, Treasurer, and four open Board of ters, regional workshops, cooperation with lo- Director positions. Biographical information and cal statistical societies, courses by leading Baye- candidate statements appear below, as well as sians, joint research programmes (e.g. looking for on the ISBA website. ISBA will open electronic opportunities offered by the European Union) voting via the ISBA website beginning October in Asia, Africa, Eastern Europe and Central and 15th, with elections closing November 15th. All South America (without forgetting our strong- current members will be emailed instructions for hold...). Those activities should have continuity voting prior to the election. in time, being, possibly, coordinated by an ad hoc committee. Although deeply rooted in my home- President 2012 (President Elect 2011, Past Pre- lands (Italy and European Union), where I am sident 2013) Research Director at the Italian National Rese- arch Council and Faculty in a Ph.D. programme Fabrizio Ruggeri (CNR-IMATI Milan, Italy) in Pavia, I am a typical example of globalisati- on: I got my M.Sc. at Carnegie Mellon and my I would like to thank the Nominations Com- Ph.D. at Duke in the USA, I am Adjunct Faculty mittee for inviting me to run, for the third ti- at New York University and, this fall, Visiting Fa- me, for ISBA President. I am very flattered since culty at George Washington University, besides three independent committees, whose members being Faculty in a Ph.D. programme in Algeria are deeply involved in ISBA activities, thought of and having given many courses in Italy, Spain, me for such a prestigious job. Now I have just to Brazil and Chile (even in Spanish). I am also ve- convince ISBA members I am fit for it... ry honoured of being an Advisory Editor of the I believe ISBA has the mission to promo- Chilean Journal of and having strong te sound statistical methods and applications, connection with Latin America, besides coopera- worldwide and across disciplines, in the socie- tion in Eastern Europe and a recent one with Tai- ty, in teaching and in the research. Here are my wan. Finally, I am Fellow of the American Stati- plans to contribute to it, if elected, and some in- stical Association. formation on me. ISBA should continue its policy towards a I am an ISBA member since the first meeting in greater cooperation with scientists in other fields, San Francisco in 1993 and I served ISBA in many through contacts with their organisations, inter- ways: Board member, member of three Nomina- disciplinary workshops, invited papers in Baye- tions Committee, four times member of the Sa- sian Analysis, groups within ISBA, plenary talks vage Prize Selection Committee (and chair of the and courses. It is not only an issue of outreach latest one), once member of the De Groot Prize but also of continuity of the cooperation over Selection Committee, Chair of ISBA2004 Scienti- time. I am devoting many efforts in spreading fic Committee, founding (and current) Editor of sound Bayesian methods especially among engi- Bayesian Analysis and its first Production Editor, neers and, more recently, biologists, epidemiolo- Editor of ISBA Bulletin (and responsible for its gists and MDs. My research interests go from ro- transformation into its current format), one of the bustness, nonparametrics, wavelets and stocha- main proposers of the Pilar Iglesias Fund, Chair stic processes to reliability, project management, of the recent Editors Search Committee and Chair and, more recently, biostatistics and biology. I of ISBA endorsed workshops (on robustness, the have been ENBIS (European Industrial Statisti-

Content 3 www.bayesian.org ISBA Bulletin, 17(3), September 2010 2010 ISBA ELECTION cians) President, I was Editor-in-Chief of Ency- ductory courses from a Bayesian perspective are clopedia of Statistics in Quality and Reliability the exception rather than the rule. It is also rare (a major reference in the field) and I am Editor- to find sessions or even talks on teaching Bayesi- in-Chief of Applied Stochastic Models in Busi- an statistics at the undergraduate level, either at ness and Industry. I edited two books on Bayesi- teaching conferences (e.g., out of over 110 sessi- an Robustness and two forthcoming books are on ons at the recent 8th International Conference on in Stochastic Processes (as co- Teaching Statistics, I found only one session on author) and Statistics in Healthcare Practice (as teaching Bayesian statistics) or at Bayesian con- co-editor). ferences (e.g., I do not recall any such talks oc- curring at the ISBA meetings I have attended). Young researchers are a top priority for ISBA. We should think of some form of active participa- If we believe that Bayesian thinking is the key tion in ISBA activities, besides offering courses, for scientific quantitative reasoning, and that the awards and travel grants (and expanding them 21st century belongs to Bayesian statistics, why as much as possible). We should discuss and are we then not teaching it first, literally, that is, decide about forms of deeper involvement by when future scientists, policy makers, educators, young researchers: an autonomous group, parti- etc. are first exposed to serious quantitative re- cipation in some ISBA committees, organisation asoning? Two common responses have been (1) of some events devoted to themselves? As Edi- there is already too much material for introduc- tor of the ISBA Bulletin, I created the Students’ tory level statistics courses and students are al- Corner, run by students to promote their fellows’ ready too busy, and (2) Bayesian statistics is very works. As Chair of ISBA2004 Scientific Commit- difficult to teach at the introductory level. tee, I worked hard with Pilar Iglesias to bring ma- ny young people to Chile and, later, I was one of For (1), I would argue that it is exactly becau- the promoters of the fund named after her to sup- se the introductory courses are filled with mes- port travel grants for students. sages that are inconsistent with Bayesian reaso- ning that we, at the very least, need to provi- Last but not least, ISBA has its well establis- de students with an opportunity to be exposed hed activities, including worldwide conferences, to Bayesian thinking before their minds get ”too chapters and groups, Bayesian Analysis, Bulle- busy” with materials we think they should learn tin, and awards. They have been run very well so less or not at all. For those of us who enjoy occa- far and they strongly depend on ISBA finances, sional DIY (do it yourself), we understand well so wealthy because of a wise administration in how building from scratch often takes signifi- the past. Continuation and improvement of such cantly less effort than trying to remodel an ”old successful scientific and financial management house.” are the primary goals of the Board and, first of all, of the President. New ideas are very welco- As for (2), some recent experiences have taught me and I will invite and strongly rely on inputs me that it is entirely possible to teach Bayesian from members, as I did when I reshaped the ISBA thinking even before formally introducing proba- Newsletter or I chaired the Editors Search Com- bility. For the past two summers, I have taught mittee. ”Vital Statistics for Medical and Life Sciences”, a study-abroad course that my ”happy team” (a Xiao-Li Meng (Harvard U, USA) group of Harvard graduate students and I) has put together. The very first lecture starts with By 2012, ISBA will be leaving its teenage years the question of what is the chance of a wo- behind. By any measure, ISBA has had a fabulous man with positive screening test result actual- teen experience. Just take a look at the society’s ly having breast cancer, given the knowledge of website, or attend any of its conferences. We do the breast cancer prevalence and the specifici- so well on so many fronts, and with so much fun ty and sensitivity of the screening test. Every- and energy, just like teenagers. The use of ”tee- thing was described in terms of simple percen- nagers” is more than a metaphor here, because tages and illustrated with tree diagrams and Po- the future of ISBA, say in its second 20 years, de- werPoint animations. All students are undergra- pends literally on those who are teenagers now. duates, and the vast majority of them have no Yet this is a group that we have essentially ne- prior exposure to statistics or probability. Yet by ver reached out to — undergraduate-level intro- the end of the first lecture, they have learned the

Content 4 www.bayesian.org ISBA Bulletin, 17(3), September 2010 2010 ISBA ELECTION thinking process of going from prior to posterior, certainly have highly qualified friends and col- as well as the power of such reasoning, especial- leagues here at Booth. One such friend ly because of the unexpectedness of the answer. (and, unfortunately, former colleague) is Carlos My best ”teaching moment” was when a student Carvalho. I am also counting on my friends and came to me after the class and said ”This is just former Treasurers, whose expertise and experi- weird,” referring to the fact that the answer see- ence I will constantly and humbly seek. med to be substantially too small compared to his ”prior intuition”, yet he had to agree that it is the I am currently Associate Professor of Econo- right answer because of the ”evidence” presented metrics and Statistics at the University of Chi- in the lecture, especially in multiple ways. What cago Booth School of Business. My research in- a great demonstration of Bayesian reasoning at terests concentrate on Bayesian modeling and work! computation, with particular emphasis on lon- gitudinal, spatial and dynamic factor analysis, One of the wittiest puns of Morris DeGroot is time-varying covariance models, dynamic mo- ”Bayesians know where the bar is” (if you don’t dels, Markov chain Monte Carlo and sequential get the pun, email Ed George). We need many Monte Carlo methods. I published the book on more teenagers who know the right bar befo- MCMC for Bayesian inference with Dani Gamer- re they reach legal drinking age. Certainly, the- man (Chapman & Hall/CRC Press) and will pu- re is no society more fitting than ISBA to lead blish in 2012 a book on Bayesian Econometrics the collective and organized effort international- with Robert McCulloch (Wiley). My research pa- ly in providing Bayesian statistics education at pers have been published in Statistical Science, the undergraduate level. Efforts here range from Biometrics, Statistica Sinica, Journal of Time Se- organizing pedagogically passionate and experi- ries Analysis, Statistics and Computing, Jour- enced scholars to brainstorm and design effective nal of Statistical Planning and Inference, Com- teaching and learning strategies, to collecting in- putational Statistics and Data Analysis, to na- formation on the problems and challenges faced me just a few. I advise or advised around 20 at different institutions across different countries, MSc and PhD students since 2000. I am currently and to providing a variety of sample curricula, Associate Editor for Bayesian Analysis. Additio- teaching materials, and teacher training tailored nal information can be found on my website at correspondingly. If I am given the honor to serve http://faculty.chicagobooth.edu/hedibert.lopes. as ISBA President, I will give the highest priori- ty to fostering such endeavors and to integrating Mike Daniels (U of Florida, USA) I am very hap- them into ISBA’s on-going effort in building and py to be nominated for the Treasurer of ISBA. I enhancing the future force of the Bayesian army have served ISBA in a variety of ways over the to meet the ever-increasing need for principled years: Member of the Savage Thesis Award Com- quantitative reasoning in every walk of life. mittee in 2007-2008 and 2008-2009 and Member of the Mitchell Prize Committee in 2009-2010. I Treasurer 2011-2013 also served as the program chair for the Secti- on on Bayesian Statistical Science of the ASA in Hedibert Lopes (U of Chicago, USA) 2009.

Honor was (and still is) my feeling when I I also have relevant experience to serve as Tre- browsing through the list of past ISBA Treasu- asurer given my holding the same position in rers. Bayesian statistics has done many things for ENAR (of IBS) for 2010-2011 and in dealing with me and I am privileged to be nominated to one financial issues in my capacity as the chair of the more of ISBA’s offices. I have been a member of Department of Statistics at the University of Flo- the Savage Thesis Award for a couple of years, rida since 2008. I will use my experience to help ISBA bulletin editor for another few years and keep ISBA in the best financial situation possible. currently serve as board member, which included the chair of the ad-hoc membership committee I am currently Professor and Chair in the De- for one year. I have also helped the foundation partment of Statistics at the University of Flori- of ISBRA, the 10-year old Brazilian Chapter of IS- da. My research interests focus on Bayesian me- BA. thodology for longitudinal data and missing da- ta and methodology for estimating dependence. I have no prior experience as Treasurer, but I I recently published a book (w/ Joe Hogan) on

Content 5 www.bayesian.org ISBA Bulletin, 17(3), September 2010 2010 ISBA ELECTION

Bayesian methodology for missing data in lon- king experience being officer of SBSS and ICSA gitudinal studies (title: Missing Data in Longi- would help increase the membership in the socie- tudinal Studies: Strategies for Bayesian Mode- ty and promote the collaboration between ISBA ling and Sensitivity Analysis, with Chapman & and other societies, and my research experience Hall/CRC Press). My research papers have be- as a Bayesian would help promote the en published in JASA, Biometrika, Biometrics, value and use of Bayesian statistics in interdisci- Biostatistics, JCGS, JMVA, among other venues. plinary research. I am an Associate Editor for JASA, Biometrics, and Statistics & Probability Letters. I am current- Chris Holmes (U Oxford, UK) ly working on collaborative projects in the areas I am professor of Biostatistics at the Universi- of aging, genomics, infectious diseases (malaria ty of Oxford. My research interests surround ap- and flu), weight management, and muscular dys- plications and methods development for Baye- trophy. For more information on me, please visit sian analysis of high-dimensional data sets ari- my website, http://www.stat.ufl.edu/ mdaniels. sing in genomic and genetic . This Board of Directors 2011-2013 (4 openings, 8 has lead me to investigate Bayesian nonparame- candidates listed randomly) tric methods, including a recent book with Hjort, Mueller and Walker, as well as recent work in Dongchu Sun (U Missouri, USA) Bayesian computation on graphics cards. I’ve ac- ted as Associate Editor for The Annals of Stati- Candidate Statement Pending, please see the stics and The Annals of Applied Statistics. I have ISBA website for up-to-date information. also served on the last two Savage Award com- mittees. I have organized workshops and invi- Ming-Hui Chen (U Connecticut, USA) ted sessions on Bayesian statistics and their ap- plications in genetics. I have published papers in I am delighted to be a candidate for ISBA’s the mainstream statistics journals and also in ma- board of directors. I am Professor and Director chine learning, bioinformatics and genetic epide- of the Statistical Consulting Services in the De- miology. I would be honoured to serve on the IS- partment of Statistics at University of Connec- BA board. ticut in Storrs, Connecticut, USA. Over last 20 years, I have been working on methodological Judith Rousseau (U Paris Dauphine, France) and application aspects of Bayesian statistics, in- cluding the Bayesian analysis of categorical da- I am professor at Universite´ Paris Dauphine ta, survival data, missing data, and genomic da- and at the ENSAE (Paris, France). My research ta, the development of Monte Carlo methods for interests include interractions between Bayesian Bayesian computations, and the Bayesian design and frequentist methods and understanding the of clinical trials. I have taught short courses on impact of using a prior with a more recent focus Bayesian computations, Bayesian Survival ana- on nonparametric models; they include also mo- lysis, and missing data analysis at various mee- del choice and tests and in a more applied per- tings and universities including JSM’s, ENAR’s, spective prior elicitation. I am currently associate WNAR, and ICSA symposiums. I have served as editor of the Australian and New-Zealand Jour- an Associate Editor of Bayesian Analysis since nal of Statistics and of the Annals of Statistics. I 2004. I was the program chair of the Section on have also been a member of the prize commit- Bayesian Statistical Sciences (SBSS) in 2005 and tee for the last two years and last year I have ac- the Publications Officer of SBSS for 2008-2009. I ted as Chair of the prize committee. I have be- also served on the Board of Directors of the In- en a member of the scientific committee for diffe- ternational Chinese Statistical Association (ICSA) rent ISBA meetings and have organized Bayesian (2004-2006). Currently, I am the Executive Direc- sessions at meetings. I have published papers in tor of ICSA (2007-2010). In 2009, ISBA and ICSA mainstream statistical journals and in computa- signed the Memorandum of Understanding on tional journals. I would be honoured to serve on the co-sponsorship of future ISBA or ICSA mee- the ISBA board. tings. In March 2010, I also served as the Associa- te Chair of the executive committee for the confe- Hajime Wago (Kyoto Sangyo U, Japan) rence titled ”Frontiers of Statistical Decision Ma- king and Bayesian Analysis”. If elected, my wor- I am delighted to be a candidate for ISBA’s

Content 6 www.bayesian.org ISBA Bulletin, 17(3), September 2010 2010 ISBA ELECTION board of directors. I am currently Professor of January 2011 and to submit your papers to BA! Econometrics and Statistics at Kyoto Sangyo Uni- versity in Kyoto, Japan. Over the past two deca- Besides having a vision of where Bayesian sta- des and more I have been working on various tistics should go (which I’m sure is shared by al- aspects of Bayesian analysis, including high fre- most everybody who is on this list) I would like quency financial data and spatial criminal and to share with you my thoughts on how to help business data, computational aspects such as move in the right directions. I want to be very MCMC methods, particle filtering and simulati- concrete so here is a TODO list (I was tempted to on based inference, modelling for the state space write a WISH list but, in name of concreteness...) time series analysis, and macroeconometric mo- delling (Dynamic Stochastic General Equilibri- 1. know thyself (gnothiˆ seauton,´ nosce te ip- um (DSGE) Model). I have organized 15 confe- sum): use Bayesian statistics to make infe- rences and workshops on theory and application rence and prediction on who we are as a of Bayesian statistics. Recently, I have also publis- community, what is it that we do well and hed four MCMC related Bayesian books in Japa- less well, and what we should do to impro- nese to promote Bayesian ideas on useful empiri- ve. To be concrete: have a thematic compe- cal analysis. tition where people are encouraged to use their favorite new modeling and inferenti- As a new era of globalization intensifies, Japan al tool to analyze ”Bayesian data”, which is statistical society will attempt to expand its acti- not data that is coherent and consistent but vities across national boundaries and to consoli- it is data that Bayesian have generated (he- date friendship and partnership with re are a few examples: feed to your favorite in other Asian societies. text classifier the abstract books of Valencia meetings, study the time-space model of lo- Regarding my contribution for ISBA, I served cation of Bayesian conferences, study the ti- a member of the founding board (1992-94) and I me series of ”Bayesian papers” in statistical remember the intense discussion on whether ha- journals ...) ving our constitution and publishing the journal of ISBA. 2. divide and conquer (divide et impera): which are the areas where we are good and Bayesian method is getting more popular in Ja- the ones were we can improve? Divide: ha- pan and Asia, so it is the best time for us to hold ve small meetings like MCMSki, Bayesi- the next ISBA2012 meeting in Kyoto. I am hosting an nonparametrics, Prior elicitation, Clini- it as the chair of Japanese organizing committee. cal trials, Genomic and proteomic ... Con- I am confident the ISBA meetings have given and quer: spread the voice around! Provide fi- will give a stimulus occasion for local young re- nancial support for invited talk at thematic searchers and students. ”non-Bayesian” conferences (JSM does not If elected, my experience would help increase count, but the annual meeting of seismic the membership in the Society, and help promo- forecast: an area where (Bayesian) statisti- te the value of Bayesian thinking in statistics and cians have little to say, would be the perfect the applied areas. example) 3. do ’model criticism’ about ISBA activities: Antonietta Mira (U Insubria, Italy) these days we get a ”costumer satisfaction” I am professor of statistics both at the U. of questionaire for pretty much everything we Insubria (Italy, where the Bayesian approach is do, what about doing one on ourselves? Af- well established) and at the U. of Lugano (Swit- ter all statisticians are ”our costumers” and zerland, where there are very few universities we offer them conferences, journals, bulle- that have Bayesian courses at graduate or un- tins, websites ... for all these activities we dergraduate level: I am helping spread the voice could have an on-line survey to get a fee- around). I have been on the scientific committee ling of how we are doing and get inputs of the joint MCMSki I, II and III (co-chair) and on what and how we could improve (the- I am co-editor of Bayesian Analysis. And I take se surveys would provide genuine ”Baye- this opportunity to invite you to Salt Lake City in sian data” to be analyzed along the spirit of point 1)

Content 7 www.bayesian.org ISBA Bulletin, 17(3), September 2010 SAVAGEAWARD AND MITCHELL PRIZE 2010

4. make sure that when you google ISBA the IMS. (especially in some languages like italian) the first entry is not the Illinois State Bar As- Rosangela Loschi (UF Minas Gerais, Brazil) sociation! I am professor and head of the Department of Statistics at Universidade Federal de Minas Ge- In compiling the above TODO list I have tried rais, Brazil. I received my Ph. D. in Statistics in to be ’nonparametric’ in the sense of thinking 1998 from the Universidade de Sao˜ Paulo, Bra- about ISBA taking on new things, not bound by zil. I am currently AE for the Chilean Journal of hard constraints: this is my vision on where Baye- Statistics. My research interests include, among sian statistics should go and the spirit that should others, clusters and change points identification, inform our decisions. survival analysis, flexible models using exten- ded classes of asymmetric distributions. I have (York U, Canada) H´el`eneMassam published papers in Statistical journals, such as, I am a Full Professor in the Department of Ma- J. of Multivariate Analysis, Computational Stati- thematics and Statistics at York University in To- stics and Data Analysis, Lifetime Data Analysis, ronto, Canada. My research area is in graphical Biometrical J., Test, and correlated areas as Com- models and Bayesian inference. My recent rese- puters and Operations Research and European J. arch has focused on the development of prior dis- Operation Research. I have been involved with tributions for the covariance parameter in Gaus- the foundation of ISBRA - the Brazilian Chapter sian models and for loglinear parameters in hier- of ISBA - in 2000, and served as ISBRA executive archical loglinear models as well as on the geo- secretary twice (2000-2002, 2006-2008). I was one metry of discrete hierarchical models. My works of the organizers of the I Latin American Meeting tends to be of a theoretical nature but always on Bayesian Statistics in 2002, which was consi- with a view to improve methods widely used in dered as an ISBA Regional meeting, of three Bra- applications. I have published papers in the An- zilian Meetings on Bayesian Statistics and some nals of Statistics, Biometrika, Bernoulli and other other local meetings. I also served on two ISBA mainstream journals. I am an Associate Editor of nominate committees, the last one in 2009. It is a Bayesian Analysis. I have been a member of the great honor to be nominated as a candidate for Statistics NSERC Selection Committee, the IMS the ISBA office. If elected, I hope, I can help to Fellows Selection Committee and the IMS Presi- promote ISBA in South America and increase the s dent Selection Committee. I am also a Fellow of number of membership in our society.

SAVAGE AWARD AND MITCHELL PRIZE 2010

CALL FOR NOMINATIONS The Mitchell Prize is named after Toby J. Mit- chell and was established by his friends and col- Peter Muller¨ leagues following his death from leukemia in [email protected] 1993. The Mitchell Prize is awarded in recogni- The Savage Award, named in honor of Leo- tion of an outstanding paper that describes how nard J. ”Jimmie” Savage, is bestowed each year to a Bayesian analysis has solved an important ap- two outstanding Bayesian Ph.D. theses. The Sa- plied problem. The Prize is jointly sponsored by vage Prize is one of our most visible activities. the Section on Bayesian Statistical Science (SBSS) The prize is jointly administered by ISBA and of the ASA, the International Society for Bayesi- the ASA Section on Bayesian Statistical Science an Analysis (ISBA), and the Mitchell Prize Foun- (SBSS). ders’ Committee. Deadline is Friday, Oct 29. Advisors, de- Deadline is Dec 31, 2010. A paper partment chairs or any ISBA or SBSS member may be nominated by an author or any may nominate a dissertation for the prize. Plea- member of ISBA or SBSS. Please see se see http://bayesian.org/awards/Savage.html http://bayesian.org/awards/MitchellPrize.html for details. for submission details. Content 8 www.bayesian.org ISBA Bulletin, 17(3), September 2010 ARNOLD ZELLNER 1927-2010

ARNOLD ZELLNER 1927-2010

OBITUARY on of HGB Alexander Professor in the Business School, where he remained Siddhartha Chib & John Geweke until his retirement in 1996. [email protected] It was at the University of Washington that [email protected] he developed the seemingly unrelated regressi- Arnold Zellner, co-founder of ISBA and its first ons model, a model that was the basis for nume- president, passed away at his home in Chicago rous applications and subsequent developments on August 10, 2010 at the age of 83. One of the in the analysis of multivariate outcomes. His con- world’s most prolific and influential econome- tributions to Bayesian econometrics and stati- tricians, publishing more than 200 articles and stics, which would be the theme of his research 22 books and monographs, he made pioneering for the balance of his academic career, began with contributions to Bayesian statistics, the analysis a paper published in 1964 with his University of of simultaneous equation models and time series Wisconsin colleague George Tiao. analysis. Zellner’s landmark book, An Introduction to He was born on January 2, 1927 in Brooklyn, Bayesian Inference in Econometrics, was publis- New York. He attended Harvard University on a hed by John Wiley and Sons in 1971. This book scholarship, earning a Bachelors degree in Phy- pioneered the field of Bayesian econometrics. Pu- sics in 1949. Following two years of military ser- blished at a time when few econometricians utili- vice, he attended graduate school at the Univer- zed Bayesian statistics, it demonstrated how pro- sity of California, Berkeley, where he earned a blems of concern to economists, and researchers Ph.D. in Economics in 1957. more broadly, could be tackled from a Bayesian Zellner held faculty appointments in the De- perspective. The book had a large impact on the partment of Economics at the University of Wa- subsequent growth of the field. shington (1955-1960) and the University of Wis- Zellner’s hallmark energy, enthusiasm and or- consin (1961-1966). In 1966 he took up the positi- ganizational skills are reflected most significant-

Content 9 www.bayesian.org ISBA Bulletin, 17(3), September 2010 JULIAN BESAG 1945-2010 ly in two organizations that he founded. The on- the Journal of Business and Economic Statistics going Seminar on Bayesian Inference in Econo- in 1983. He was Chair of the Business and Eco- metrics and Statistics (SBIES), launched in 1971, nomics Section in 1982 and went on to be elec- was a key forum for many developments in Baye- ted the Association President in 1991. The annual sian econometrics and statistics. Following up on Zellner Thesis Award in Business and Economic his belief that Bayesian methods were useful in Statistics recognizes both these contributions and all scientific endeavors, Arnold helped to found his influence on scholarship in econometrics. He ISBA in 1991. He was its first president, a privile- was an elected fellow of the International Stati- ge of which he remembered fondly. stical Institute as well as the American Statistical Zellner was instrumental in the founding of Association. the Journal of Econometrics in the early 1970’s. Arnold leaves a legacy of warmth, creativity Due in no small part to his vigorous leadership, and enthusiasm for the Bayesian profession and serving as editor until his passing, the Journal is life in general that will long live in the minds now a major publication in the field. Zellner al- of those who knew him. He will be greatly mis- so made outstanding contributions to the Ameri- sed.s can Statistical Association, where he inaugurated

JULIAN BESAG 1945-2010

Photo courtesy of Larissa Stanberry.

OBITUARY years of ill-health. His contributions to the dis- cipline of statistics are profound. While he did Peter Green not describe himself as a card-carrying Bayesi- [email protected] an, his originality and creativity in areas that became central to modern Bayesian modelling Julian Besag died on 6 August 2010, in hospi- and computation have been, and continue to be, tal in Bristol, UK, following surgery after some of far-reaching consequence: Julian was using

Content 10 www.bayesian.org ISBA Bulletin, 17(3), September 2010 JULIAN BESAG 1945-2010

”full conditionals” in modelling and computati- interacting systems and his role as one of the very on from the mid-70s! early proponents of Markov chain Monte Carlo methods for fitting statistical models. In particu- Julian’s research work had authority and great lar, he was well ahead of his time in recognising originality. He seldom wrote the last word on a the duality between the conditional specification subject, but was there at the start of many of the of stochastic models and the construction of al- key developments in modern stochastic model- gorithms for such models using these conditional ling. He did not write a great deal, but his work specifications. was very deliberate, and densely and painsta- Julian was an inspiring and challenging men- kingly written; he had very high standards over tor to younger researchers; his career shows a what he put his name to. string of papers written jointly with more juni- The general area where he made the biggest or people, most of whom have since developed impact is the conditional modelling of spatial sy- into independent researchers with substantial re- stems. Together with the parallel, independent, putations of their own. work on interaction in contingency tables by Dar- Julian was born in Loughborough, UK, on 26 roch, Lauritzen and Speed, Julian Besag’s work in March 1945, and after his mother died brought the 70s and early 80s laid the foundations for the up by German-speaking grandparents. He began entire contemporary tradition of highly structu- studying engineering at Cambridge but moved red stochastic systems. This strategy for building to the to study stati- complex global models through local specificati- stics, obtaining his BSc in 1968. He then spent a ons guaranteed to be self-consistent has made a year as a research assistant to Maurice Bartlett huge impact on stochastic modeling in many are- in Oxford before obtaining a lectureship at Liver- as of science, medicine and technology, and sti- pool. He moved to Durham in 1975, where he be- mulated important work on statistical inference came a professor in 1986. He was a visiting pro- for such systems, and on their probabilistic theo- fessor at the University of Washington in Seattle ry. during 1989-90 and, after a year back in Newcast- What began in his early work as a quest for fle- le, returned to Seattle long-term. He officially re- xible and mathematically-sound models for eco- tired from Seattle in 2007 but remained an emeri- logical phenomena, resulting in the notion of mo- tus professor. He returned to the UK, and during delling spatial systems as Markov random fields 2007-09 he held a chair at the . led on directly to the idea of MRFs as a ge- At his death in 2010 he was a visiting professor at neric model for interacting systems, with Julian both Bath and Bristol. himself responsible for adapting the models and Outside his working life, Julian’s passions we- the methodology to agricultural field trials, ar- re for mountaineering (including a 14000’ peak chaeological problems, image analysis and disea- in the Alps scaled on his honeymoon with Vale- se mapping, and to hierarchical versions of such rie), hockey (while he was at Liverpool he had a models under the Bayesian paradigm. trial for the Welsh national team), running, and, The second major plank to his research was in- after moving to Seattle, sailing. He lived on the novatory work on inferential methods for spatial water in Seattle and owned two ocean-going sail- systems, and their computational implementati- boats. His personal relationships - emotional, so- on. Especially noteworthy are the notion of pseu- cial and professional - were often stormy and ve- dolikelihood in interacting systems, as a com- ry demanding for all concerned, but he was an putationally tractable alternative to the true li- unforgettable character, who will be much mis- kelihood, the ’iterated conditional modes’ algo- sed.s rithm, important contributions to the algebra of

Content 11 www.bayesian.org ISBA Bulletin, 17(3), September 2010 ANNOTATED BIBLIOGRAPHY

ANNOTATED BIBLIOGRAPHY

COMPUTATIONAL INFERENCE of low-, mid-, and high-level models are set FOR INVERSE PROBLEMS out, as well as an example in ultrasound imaging. As the authors note, the issue of Tiangang Cui physical modelling for the likelihood is not [email protected] considered. Colin Fox [email protected] Many statistical papers on inverse problems Inverse problems occur when indirect obser- highlight the high computational cost required vations d are made of a physical system x, typi- in evaluating the forward map, and hence like- cally by scattering energy of some kind, and we lihood. However, understanding the structure of want to infer properties of the unknown true sy- the forward map is critical for efficient compu- stem. That includes nearly all physical imaging tation. Specifically, the forward map necessarily systems, with well-known examples being x-ray has an unbounded range of sensitivities, hence imaging, CAT scanning, and remote sensing. posterior distributions are supported on a re- The forward map F that models the determi- latively low-dimensional manifold with correla- nistic mapping x d is the subject of theoretical tion coefficients that equal 1 to within machi- 7! physics, though often requires further physical ne precision. These properties give the extreme modelling of instrumentation and data model- sensitivity that inverse problems display to both ling as practiced in statistics. Computer imple- measurement and modelling errors, and make mentation of these models is usually the realm of believable inference from measured data a chal- specialists in computational methods, with effi- lenging task. cient implementation being essential when com- prehensive Bayesian inference is planned. A sui- • Fox, C., Tan, S., & Nicholls, G. K. (1997) In- table probabilistic model for this modelling and verse Problems, lecture notes for ELEC 404, observation process is then available at http://elec.otago.ac.nz/w/ images/1/19/ELEC404 Inverse Problems.pdf d=F (x) + b(x) + ✏. This graduate course has been taken up by Here, b and ✏ denote the model bias and the engineering groups around the world, as measurement noise, respectively. The model bi- an accessible introduction to inverse pro- as may arise from various sources: numerical er- blems. These notes cover the basic structure ror in computer implementation of the forward and difficulties arising in inverse problems, model, spatial discretization of the unknown pa- regularization methods, simple probability rameters, inappropriate assumptions in the ma- models, and sample-based inference. thematical model, etc. The system can often be represented in terms In common with most of Bayesian statistics, of spatially distributed parameters x (for exam- computational inference for inverse problems is ple, coefficients in a partial differential equation a rapidly developing research field, with remar- modelling energy propagation), so representati- kable work appearing each year. We have expli- ons and prior models are usefully drawn from citly excluded many exciting advances that we spatial statistics. can forecast, and instead focus on the best tech- nology available today for performing inference • Hurn, M. A., Husby, O., & Rue, H. (2003) in practical inverse problems. The basic structure “Advances in Bayesian Image Analysis”, should be familiar to all; posterior inference im- Highly Structured Stochastic Systems, eds plemented by random walk Metropolis-Hastings Green, P. J., Hjort, N., & Richardson. S, Ox- (MH). However, sophisticated advances in trea- ford University Press, 302-322. ting model error and utilizing model reduction A great overview of modern Bayesian have vastly improved the efficiency and quality image modelling and analysis. The notions of inference.

Content 12 www.bayesian.org ISBA Bulletin, 17(3), September 2010 ANNOTATED BIBLIOGRAPHY

Model bias libration using High Dimensional Output”, Journal of the American Statistical Association, A major advance in recent years has been the 103(482), 570–583. development of frameworks for treating bias in Implosion of steel cylinders are analyzed. complex computer models. These Gaussian pro- To deal with the functional model output, cess (GP) based approaches allow estimation of singular value decomposition is used to model bias, and also the possibility of fast infe- construct the basis vector for the model out- rence via GP surrogates to the forward map. The- puts. Then, a different set of basis vector are re are difficulties in high dimensional settings, chosen for the bias function, based on what since then building GP models is expensive. In is known about actual physical process and the context of inverse problems, these methods potential deficiencies in the simulator. have proved a practical route to improving on biased and overly confident estimation that re- • Higdon, D., Lee, H., & Holloman, C. (2003) sults from artificially simplified forward maps. “Markov chain Monte Carlo-based approa- • Kennedy, M. C., & O’Hagan, A. (2001) ches for inference in computationally inten- “Bayesian calibration of computer models” sive inverse problems”, Bayesian Statistics 7, (with discussion), Journal of the Royal Stati- eds Bernardo, J. M., Bayarri, M. J., Berger, J. stical Society: Series B, 63, 425–464. O., Dawid, A. P., Heckerman, D., Smith, A. F. M., & West, M., Oxford University Press, In this milestone paper, the model parame- 181–197. ters x are considered as a combination of unknown calibration parameters and con- A Metropolis couple scheme is implemen- trollable inputs. By collecting field obser- ted to analyze the permeabilities of a sub- vations at various controllable inputs, and surface flow system. Since measurements pre-evaluating the forward model F at a can only be observed for a single set of con- collection of model parameters, Gaussian trollable inputs, the framework of Kenne- process (GP) surrogates to F are construc- dy and O’Hagan (2001) is not applicable. ted, with the bias b represented by a se- Hence the nature and size of the model bias parate GP model. Thus, the posterior dis- can only be judged by knowledge about the tribution is a joint Gaussian distribution physical process. conditioned on field observations and pre- evaluated model outputs. Estimating mo- We note that in practice the forward model del parameters and hyperparameters of the is often ‘corrected’ after examination of residual GP models facilitates predictions and ana- structure, also using modelling judgements. lysis of uncertainty in the forward model. • Bayarri, M. J., Berger, J. O., Cafeo, J., Garcia- Proposals and adaptation Donato, G., Liu, F., Palomo, J., Parthasa- rathy, R., Paulo, R., Sacks, J., & Walsh, D. • Nicholls, G. K., & Fox, C. (1998) “Prior mo- (2007) “Computer Model Validation with delling and posterior sampling in impe- Functional Output”, Annals of Statistics, dance imaging”, Bayesian Inference for Inver- 35(5), 1874–1906. se Problems, Proc. SPIE 3459, 116–127. The framework of Kennedy and O’Hagan Several proposal distributions built from (2001) is extended to analyze vehicle sus- multiple ‘moves’ are applied to sampling pension systems where the model output in electrical impedance tomography, un- is a time series. Wavelet decomposition is der various prior models. Demonstrates the used to decompose the model output, with computational gain available by exploiting model output, bias function and field ob- the structure of the inverse problem when servations transformed into the decompo- designing proposal distributions. sed basis. Hence, the framework of Kenne- dy and O’Hagan (2001) can be expressed • Green, P. J., & Mira, A. (2001) “Delayed independently on each basis. rejection in reversible jump Metropolis- • Higdon, D., Gattiker, J., Williams, B., & Hastings”, Biometrika, 88, 1035–1053. Rightley, M. (2008) “Computer Model Ca- The reversible jump formalism is presented

Content 13 www.bayesian.org ISBA Bulletin, 17(3), September 2010 ANNOTATED BIBLIOGRAPHY

in a form that we use to evaluate the Ha- • Higdon, D., Reese, C. S., Moulton, J. D., stings’ ratio for subspace moves. The de- Vrugt, J. A., & Fox, C. (2009) “Posterior ex- layed rejection algorithm is also discussed. ploration for computationally intensive for- ward models”, Technical Report LA-UR 08- Tuning of proposals is often required to gi- 05905, Statistical Sciences Group, Los Alamos ve statistically efficient implementations, though National Laboratory, to appear in The Hand- can be very time consuming for the high dimen- book of Markov Chain Monte Carlo, eds Meng, sional representations that are typical in inverse X.-L., Gelman, A., & Jones, G., CRC press. problems. The advent of adaptive methods gave Several implementations of the MH are re- a practical way to automate that tuning process. viewed and tested on a synthetic 2D elec- trical impedance tomography problem. In- • Haario, H., Saksman, E., & Tamminen, terestingly, the multivariate Gaussian pro- J. (2001) “An adaptive Metropolis algo- posal used by the adaptive Metropolis al- rithm”, Bernoulli, 7, 223–242. gorithm gives poor mixing, while simple This adaptive Metropolis algorithm uses single-site update shows acceptable mi- 2.832 xing. the multivariate Gaussian N(0, d ⌃) as proposal, where d is dimensionality of the parameter space, with the covariance ma- The following two algorithms provide useful trix ⌃ estimated from past samples of the black-box sampling tools for low dimensional chain. Ergodicity is established by an argu- mid-level representations of fixed dimension. ment akin to lace work. • Ter Braak, C. J. F. (2006) “A Markov Chain • Haario, H., Laine, M., Mira, A., & Saks- Monte Carlo version of the genetic algo- man, E. (2006) “DRAM: Efficient adaptive rithm differential evolution: easy Bayesian MCMC”, Statistics and Computing, 16, 339– computing for real parameter spaces”, Sta- 354. tistics and Computing, 16, 239–249. Combines the delayed rejection and adap- A population MCMC algorithm is desi- tive Metropolis algorithms. gned by combining the genetic algorithm differential evolution and the random walk • Roberts, G. O., & Rosenthal, J. S. (2007) MH algorithm. “Coupling and Ergodicity of Adaptive • Christen, J. A., & Fox, C. (2010) “A gene- MCMC”, Journal of Applied Probability, 44, ral purpose sampling algorithm for conti- 458–475. nuous distributions (the t-walk)”, Bayesian Establishes simplified regularity conditions Analysis, 5(2)263–282. that are sufficient to ensure ergodicity of Two points in state space are maintained adaptive MCMC algorithms, namely, the and used to design moves for the t-walk diminishing adaptation and bounded con- that is close to affine invariant. vergence conditions. Now, anybody can de- sign a convergent adaptive MCMC.

• Roberts, G. O., & Rosenthal, J. S. (2009) Efficient MCMC “Examples of Adaptive MCMC”, Journal of One major difficulty in MCMC sampling for in- Computational and Graphical Statistics, 18(2), verse problems is the computing cost of each ite- 349–367. ration in the MH. To make a fast inference, a Based on the above ergodicity conditions, reduced order model can be employed to mi- several adaptive MCMC algorithms are de- mic the behavior of the posterior distribution. As signed including the adaptive Metropolis we already mentioned, GP models are possible with Gibbs algorithm and the Regional ad- candidates but limited by the dimensionality of aptive Metropolis algorithm that tunes ran- the parameter space. Alternatively, we can use dom walk proposals to match an ‘optimal’ the knowledge of the physical modelling of the acceptance rate. These two algorithms po- forward map to build a reduced order model, tentially can be applied in a high dimensio- which could provide reasonable numerical accu- nal context. racy and feasible computing speed.

Content 14 www.bayesian.org ISBA Bulletin, 17(3), September 2010 ANNOTATED BIBLIOGRAPHY

• Christen, J. A., & Fox, C. (2005) “MCMC A useful summary of convergence results using an approximation” Journal of Compu- for MCMC, with simple window estima- tational and Graphical Statistics, 14(4), 795– tors for the variance in the sample mean. 810. • Wolff, U. (2004) “Monte Carlo errors with This paper presents our preferred method less errors”, Computer Physics Communicati- for taking advantage of computational effi- ons, 156(2), 143–153. ciencies available from state-dependent ap- proximations, while still sampling from the Introduced a -method to estimate the va- correct posterior distribution. An example riance of the mean estimator, and the inte- is presented of inference in an inverse pro- grated autocorrelation time, by choosing an blem using a cheap approximation based optimal window size that minimizes the er- on a local linearization of the forward map. ror of the Monte Carlo error for a given se- quence of correlated samples. MATLAB co- • Kolehmainen, V., Tarvainen, T., Arridge, S. de is available on the author’s website. R., & Kaipio, J. P. (2010) “Marginalization of uninteresting distributed parameters in inverse problems”, International Journal for Selected applications Uncertainty Quantification, 1, accepted. Develops the enhanced error model, eva- We list here, without annotation, papers that de- luated over the prior distribution. Presents monstrate exemplary modelling and inferential an application in which the forward map solution of substantial inverse problems. is approximated by a coarse numerical mo- del, with statistics of the numerical error • Haario, H., Laine, M., Lehtinen, M., Saks- between the coarse model and the forward man, E., & Tamminen, J. (2004) “Markov map used to construct an accurate reduced chain Monte Carlo methods for high di- order model of the posterior distribution. mensional inversion in remote sensing”, Journal of the Royal Statistical Society: Series • Cui, T. (2010) “Bayesian calibration of geo- B, 66(3), 591–608. thermal reservoir modelling via Markov chain Monte Carlo”, Ph.D. thesis, Depart- • Cornford, D., Csato, L., Evans, D. J., & Op- ment of Engineering Science, the Universi- per, M. (2004) “Bayesian analysis of the ty of Auckland. scatterometer wind retrieval inverse pro- blem: Some new approaches”. Journal of the Presents detailed development of the ad- Royal Statistical Society. Series B, 66(3), 609– aptive delayed-acceptance Metropolis Ha- 652. stings (ADAMH) algorithm, with exten- sive performance verification and large- • McKeague, I. W., Nicholls, G. K., Speer, K., scale examples. This algorithm represents & Herbei, R. (2005) “Statistical inversion our best current technology for correct and of South Atlantic circulation in an abyssal complete sampling in inverse problems neutral density layer”, Journal of Marine Re- using low-level representations. A milesto- search, 63(4), 683–704. ne for geothermal model calibration, allo- wing automatic calibration of large-scale • Watzenig, D., & Fox, C. (2009) “A review models for the first time. of statistical modeling and inference for electrical capacitance tomography”, Mea- surement Science and Technology, 20(5), doi: Convergence diagnostics 10.1088/0957-0233/20/5/052002.

• Roberts, G. O. (1996) “Markov chain con- • Lipponen, A., Seppanen, A., & Kaipio, J. P. cepts related to sampling algorithms”, in (2010) “Reduced-order estimation of non- Markov Chain Monte Carlo in Practice, eds stationary flows with electrical impedance Gilks, W. R. and Richardson, S., & Spiegel- tomography”, Inverse Problems, 26(7), doi: halter, D. J., Chapman & Hall, 45–57. 10.1088/0266-5611/26/7/074010.s

Content 15 www.bayesian.org ISBA Bulletin, 17(3), September 2010 SOFTWARE HIGHLIGHT

SOFTWARE HIGHLIGHT

RECENT DEVELOPMENTS IN THE (with permission), representing another progres- BUGS SOFTWARE sive move to take the project forwards. A finance hub has been set up so that the project can now receive donations and an international steering Dave Lunn, committee will decide on future directions, such MRC Biostatistics Unit as how to spend donated funds to best serve the Cambridge, UK user community. [email protected] In terms of developing the capabilities of the soft- In recent years the main thrust of the BUGS pro- ware, here at the MRC Biostatistics Unit in Cam- ject [1, 2] has been towards focussing develop- bridge we have been focussing on translating so- ment efforts on the open-source incarnation of me of the more specialised WinBUGS develop- the software, OpenBUGS. The main difference ments into OpenBUGS, such as templates for im- between OpenBUGS and WinBUGS lies in the plementing your own specialised functions and behaviour of the ’expert system’, which chooses distributions (WBDev [6]), and reversible jump the various sampling methods for updating un- capabilities (Jump [6]). Mostly we have been im- known variables in the model. WinBUGS defines plementing and developing facilities for model- one algorithm for each possible classification of ling dynamical systems -models involving diffe- sampling distribution whereas there is no limit rential equations- and combining these effective- to the number of algorithms that OpenBUGS can ly with other models. We are also looking at im- make use of. The OpenBUGS approach provides plementing truncated distributions correctly, so much greater scope for experimenting with con- that there is a clearer distinction between censo- temporary MCMC methods and has thus been ring and truncation [7]. Meanwhile, aside from adopted as the way forward for the project. At his important work on improving the expert sy- the time of writing, OpenBUGS already supports stem, implementing new MCMC methods, and considerably more sampling methods than does general refinement of the system, Andrew Tho- WinBUGS, such as several new types of block- mas has been working on methods for sequenti- updater and some ’population MCMC’ methods al importance sampling and parallelising multi- [3]. ple Markov chains across multi-core processors, leading to ”OpenSIS” and ”MultiBUGS”, respec- Another reason for the shift towards OpenBUGS tively. Also here at the Biostatistics Unit, Chris is the open-source nature of the software, poten- Jackson has made several important contribu- tially facilitating developments from many indi- tions, including techniques for visualising the viduals. As the software has matured, from so- uncertainty conveyed by posterior distributions mewhat experimental into a stable and reliable [8], current work-in-progress on Bayesian non- package, much behind-the-scenes work has led parametrics, and collaborating with Uwe Ligges to an appropriate infrastructure for supporting (Technische Universitat¨ Dortmund) and others such developments. The source code now resides to produce a new version of BUGS-R (an inter- on SourceForge [4], an internet repository spe- face for controlling OpenBUGS from within R, cifically designed for such projects. Anyone can aka BRugs). Vijay Kumar (Gorakhpur Universi- download the source code and apply for ”devel- ty) has contributed numerous specialized distri- oper” status if they have code to contribute. In butions, such as various extensions of the Wei- addition a wiki page has been developed [5] to bull distribution. And last, but by no means least, facilitate, for example, distribution, disseminati- a team led by Neal Thomas (Pfizer, US) and Bill on, discussion and bug tracking, and also to act Gillespie (Metrum Institute, Connecticut) have as a general BUGS resource, providing FAQs and developed a suite of validation programs for the instructions on how to get the software running software, to prove that it is actually working cor- on Linux, or how to begin developing the soft- rectly! ware yourself, say. As a wiki this has the faci- lity to be edited from anywhere and by anyone OpenBUGS is now considered to be at least as re-

Content 16 www.bayesian.org ISBA Bulletin, 17(3), September 2010 STUDENTS’ CORNER liable and efficient as WinBUGS 1.4.3 (no jokes future directions (with discussion). Statistics in please) across a wide range of test applications. Medicine 28:3049–3067, 2009. WinBUGS will always be available but is no lon- ger being developed, and so we are urging users [2] D. J. Lunn, A. Thomas, N. Best, and D. Spie- to make the switch. Please take a look anyway, by gelhalter. WinBUGS -a Bayesian modelling fra- visiting the wiki page [5]; comments are always mework: concepts, structure, and extensibility. very welcome. Statistics and Computing 10:325–337, 2000. [3] C. J. F. Ter Braak. A Markov chain Monte Carlo version of the genetic algorithm Differential Evo- Acknowledgements lution: Easy Bayesian computing for real parame- ter spaces. Statistics and Computing 16:239–249, Thank you to everyone involved in the project, 2006. there are too many to mention here! The followi- ng people have been instrumental in getting the [4] http://sourceforge.net project to where it is today: Andrew Thomas, Da- vid Spiegelhalter, , Bob O’Hara, Neal [5] http://www.openbugs.info Thomas, Bill Gillespie, Nathan Coulter, Uwe Lig- [6] http://www.winbugs-development.org.uk ges, Larry Gould, Chris Jackson and Steve Miller. Many thanks also to all the users of the software, [7] https://www.jiscmail.ac.uk/cgi-bin/webad whose patience, enthusiasm and innovation is al- min?A2=ind0910&L=BUGS&F=&S=&P=5643 ways very much appreciated. [8] C. H. Jackson. Displaying uncertainty with shading. The American Statistician 62:340-347, References 2008.s

[1] D. Lunn, D. Spiegelhalter, A. Thomas, and N. Best. The BUGS project: Evolution, critique and

STUDENTS’ CORNER

Call for Dissertation Abstracts research, but it may potentially lead to collabo- Luke Bornn rations with future colleagues. In addition, you [email protected] are providing an important service to the Bayesi- an community by giving established researchers Recent Ph.D graduates, having your disserta- a taste of the interests of young researchers. Fa- tion abstract published is as simple as emailing culty, please encourage your students’ participa- it to the email address above. Publishing your tion.s abstract will not only provide exposure for your

Content 17 www.bayesian.org ISBA Bulletin, 17(3), September 2010 NEWS FROM THE WORLD

NEWS FROM THE WORLD

CALL FOR ANNOUNCEMENTS 66th Annual Deming Conference on Applied Statistics, Atlantic City, NJ. 5-10th December, Sebastien Haneuse 2010. [email protected] The purpose of the three-day conference is to provide a learning experience on recent de- velopments in statistical methodologies in twel- I would like to encourage those who have any ve three-hour tutorials. Attendees receive bound announcements or would like to draw attention proceedings of the presentations. The conference to an up-coming conference, to get in touch with is followed by two parallel short courses on (1) me and I would be happy to place them here. Bayesian Adaptive Clinical Trials by Prof. Brad Carlin, University of Minnesota and Scott Berry, Berry Consultants; and (2) SAS for Mixed Models Announcements by Profs. Ramon Littell, University of Florida and Walter Stroup, University of Nebraska. The con- ference makes the books on which the conference 2012 ISBA World Meeting Planning has alrea- is based available for sale at about a 40% dis- dy begun for the 11th ISBA World Meeting, to count. be held in June 2012 in Kyoto, Japan. See the The full program and online registration June 2010 issue of the ISBA Bulletin for the an- is available at http://www.demingconference. nouncement and more information http://www. com. bayesian.org/bulletin/1006.pdf. Adapski III: Advances in Monte Carlo, Park Ci- I would like to encourage those who have any ty, UT, 3-4th January, 2011. announcements or would like to draw attention Following an enthustiastic reponse to the ear- to an up-coming conference, to get in touch with lier editions of the workshop in 2005 and 2008, me and I would be happy to place them here. this workshop is intended to provide an updated snapshot of the methodological and theoretical advances in Monte Carlo methods with an em- Meetings and conferences phasis on adaptive Monte Carlo methods in the broad sense (adaptive MCMC, adaptive popula- tion Monte Carlo, and various breeds of adaptive Eighth ICSA International Conference, importance sampling amongst others), that is al- Guangzhou, China. 19-22nd December, 2010. gorithms that attempt to automatically optimise The conference will be held at Guangzhou Uni- their performance to a given task. versity, located in Guangzhou city in Guang- The workshop will consist of 4 half-day sessi- dong Province in Southern China and has a the- ons on 3rd and 4th January and one or two po- me ‘Frontiers of Interdisciplinary and Methodo- ster sessions and will be held at The Canyons. logic Statistical Research’. The conference is co- There will be breaks on both afternoons in order sponsored by the American Statistical Associati- to allow both informal discussions and relaxation on, the Chinese Association of Applied Statistics, (skiing!). the Chinese Society of Probability and Statistics, Additional information can be found at and the Institute of Mathematical Statistics, and http://www.maths.bris.ac.uk/%7Emaxca/ International Society for Bayesian Analysis (IS- adapskIII/. BA), and jointly organized by the College of Ma- thematics and Information Science, Guangzhou MCMCSki III: Markov Chain Monte Carlo in University. Theory and Practice, Snowbird, UT, 5-7th Janua- Please note that the deadline for early registra- ry, 2011. tion is the 1st Nov. Additional information can be A central theme of the conference will be Mar- found at http://www.icsa2.org/Intl_2010/. kov chain Monte Carlo (MCMC) and related me-

Content 18 www.bayesian.org ISBA Bulletin, 17(3), September 2010 NEWS FROM THE WORLD thods and applications in the 21 years since the Smith. These papers contain two central ideas in publication of Gelfand and Smith (1990, JASA), the theory and practice of modern Bayesian sta- the paper that introduced these methods to main- tistics, namely Hierarchical Models (Lindley and stream statisticians. The conference will also fea- Smith, Journal of the Royal Statistical Society, Se- ture three plenary speakers (Nicky Best, Mike ries B, 1972) and Markov chain Monte Carlo (Gel- Newton, and Jeff Rosenthal) and six invited ses- fand and Smith, Journal of the American Statisti- sions from internationally known experts cover- cal Society, 1990). This conference will showca- ing a broad array of current and developing sta- se many key advances that continue to be made, tistical practice. As with the first joint IMS-ISBA using these ideas, by many established and youn- meeting in Isla Verde, Puerto Rico,and the second ger researchers. and third joint meeting in Bormio, Italy, nightly Research involving hierarchical models and poster sessions will offer substantial opportunity MCMC continues to grow at an astonishing ra- for informal learning and interaction. te, spanning a broad spectrum of topics in me- Additional information can be found at http: dicine, engineering, scientific computation, busi- //madison.byu.edu/mcmski/ ness, psychology, bio-informatics, computatio- nal physics, graphical models, neural networks, Fourth Annual Bayesian Biostatistics, Houston, geosciences, and public policy. This explosion of Texas. 26-28th January, 2011. Bayesian ideas is, in part, the result of papers au- Current and prospective users of Bayesian bio- thored or co-authored by Adrian Smith. statistics are invited to join others with similar in- Additional information can be found at http: terests for a three-day conference sponsored by //afmsmith.com/index2.htm. the Department of Biostatistics at The Universi- ty of Texas MD Anderson Cancer Center. Resear- International Research Conference on Bayesian chers are invited to submit abstracts for conside- Learning, Istanbul, Turkey. 15-17th June, 2011. ration as contributed presentations at the confe- In all domains of research, a major part of the rence. All topics in Bayesian biostatistics are ac- problem that needs to be solved involves the task ceptable, including clinical trial design and ana- of managing the uncertainty inherent in the pro- lysis, health policy, epidemiology, health econo- blem. In that instance, Bayesian Learning pro- mics, clinical decision making, comparative ef- vides a powerful methodology to researchers, fectiveness, and bioinformatics. Submit abstract enabling them to reach effective decisions in light title; text of 150 words or less; authors, affilia- of evidence. With its ability to incorporate pri- tions, and contact information via e-mail to Ly- or knowledge to the inference process, Bayesi- dia Davis ([email protected]). Submissi- an Learning appeals to researchers for both of its on Deadline: October 1, 2010. Applicants will be theory and applications. notified of the Program Committee’s decision by Interested individuals from academic and November 1, 2010. business worlds are invited to come together to All participants are encouraged to display discuss and communicate on challenging issues a poster at the conference poster session related to theory of Bayesian learning and app- and discussion. If interested, send an abstract lications in finance & accounting, general mana- (up to 300 words) to Lydia Davis (lbda- gement, marketing, organizational behavior and [email protected]) by November 15, 2010. production & operations within the historic and Online registration and additional informa- mystic environment of Istanbul while we will ce- tion about Bayesian Biostatistics Conference lebrate the 310th birthyear of Rev. Bayes. 2011 is available at http://biostatistics. Additional information can be found at http: mdanderson.org/BBC2011. //marc.yeditepe.edu.tr/yircobl11.htm. Conference in Honour of Professor Adrian F. M. Smith on Hierarchical Models and Markov Chain Monte Carlo, Heraklion, Greece. 2-5th Ju- Short courses and workshops ne, 2011. In Bayesian statistics, two influential papers in the latter part of the 20th Century –reprinted Workshop on Bayesian Inference for Latent in Volume III of Breakthroughs in Statistics, Gaussian Models with Applications, Zurich, Springer-Verlag–were co-authored by Adrian Switzerland. 2-5th February, 2011.

Content 19 www.bayesian.org ISBA Bulletin, 17(3), September 2010 NEWS FROM THE WORLD

Latent Gaussian models have numerous ap- laborations and partnerships that will channel plications, for example in spatial and spatio- efforts into pending problems and open new di- temporal epidemiology and climate modelling. rections for further study. O-Bayes2011 will also This workshop brings together researchers who serve to further crystallize objective Bayes me- develop and apply Bayesian inference in this thodology as an established area for statistical broad model class. One methodological focus research. is on model computation, using either classi- Additional information can be found at http: cal MCMC techniques or more recent determi- //www.sfs.ecnu.edu.cn/Obayes2011/index. nistic approaches such as integrated nested La- html. place approximations (INLA). A second theme of the workshop is model uncertainty, ranging from model criticism to model selection and mo- Eight Workshop on Bayesian Nonparametrics, del averaging. Veracruz, Mexico. 26-30th June, 2011. Hravard Rue will give an INLA tutorial on the The workshop aims at presenting the latest first day. Further confirmed invited speakers are developments on Bayesian nonparameteric stati- Renato Assunc¸ao,˜ Gonzalo Garc´ıa-Donato, Alan stics, covering a wide range of theoretical, metho- Gelfand, Finn Lindgren, Douglas Nychka, Chri- dologic and applied areas. The meeting will be stopher Paciorek and Stephen Sain. Contributed structured in 4 tutorials on special topics, a series talks and a poster session complete the four-day of invited and contributed talks and contributed program. posters sessions. For those interested this event Additional information can be found at http: will be preceded by the Mexican Workshop on //www.math.uzh.ch/bilgm11. Bayesian Statistics (TAMEB) which will feature a day of introductory courses (in Spanish) to Baye- 2011 International Workshop on Objective sian statistics. Bayesian Analysis, Shanghai, China. 11-15th Ju- Scientific committee: David B. Dunson, Subha- ne, 2011. shis Ghosal, Jim Griffin, Nils L. Hjort, Michael I. Following earlier meetings on objective Jordan, Yongdai Kim, Antonio Lijoi, Ramses H. Bayes methodology the principal objectives of Mena, Peter Muller,¨ Luis E. Nieto, Igor Pruenster, OBayes2011 are to facilitate the exchange of re- Fernando A. Quintana, Yee W. Teh and Stephen cent research developments in objective Bayes G. Walker methodology, to provide opportunities for new Additional information can be found at http: researchers to shine, and to establish new col- //www.bnpworkshop.org/.s

Content 20 www.bayesian.org ISBA Bulletin, 17(3), September 2010

Executive Committee Board Members:

President: Peter Muller¨ 2010–2012: Siddhartha Chib, Arnaud Past President: Mike West Doucet, Peter Hoff, Raquel Prado President Elect: Michael I. Jordan 2009–2011: David Dunson, David van Dyk, Katja Ickstadt, Brunero Liseo Treasurer: Gabriel Huerta 2008–2010: Sylvia Fruhwirth-Schnatter,¨ Lur- Executive Secretary: Merlise Clyde des Inoue, Hedibert Lopes, Sonia Petrone

Program Council

Chair: Alex Schmidt Vice Chair: Igor Pruenster Past Chair: Herbie Lee

EDITORIAL BOARD

Editor

Manuel Mendoza [email protected]

Associate Editors

Interviews Students’ Corner Donatello Telesca Luke Bornn [email protected] www.stat.ubc.ca/ l.bornn/ ⇠ [email protected] Annotated Bibliography Beatrix Jones News from the World www.massey.ac.nz/ mbjones/ Sebastien Haneuse ⇠ [email protected] http://www.centerforhealthstudies.org/ ctrstaff/haneuse.html Software Highlight [email protected] Alex Lewin www.bgx.org.uk/alex/ [email protected]

Content 21 www.bayesian.org 945HM_MBAISBA_SP_AD_4C:945HM_MBAISBA_SP_AD_4C.qxd 8/26/10 4:41 PM Page 1

Bayesian Analysis for Population Ecology Ruth King, Byron J.T. Morgan, Olivier Gimenez, and Stephen P. Brooks Catalog no. K10597, January 2010, 456 pp. ISBN: 978-1-4398-1187-0, $79.95 / £49.99 Applied Bayesian Hierarchical Methods Peter D. Congdon Catalog no. C7206, May 2010, 604 pp. ISBN: 978-1-58488-720-1, $89.95 / £57.99 Time Series Modeling, Computation, and Inference Raquel Prado and Mike West Catalog no. C9336, May 2010, 368 pp. ISBN: 978-1-4200-9336-0, $89.95 / £57.99 Bayesian Model Selection and Statistical Modeling Tomohiro Ando Catalog no. K11633, May 2010, 300 pp. ISBN: 978-1-4398-3614-9, $89.95 / £63.99 Statistical Inference An Integrated Bayesian/Likelihood Approach Murray Aitkin Catalog no. C3436, June 2010, 254 pp. ISBN: 978-1-4200-9343-8, $89.95 / £57.99 Bayesian Ideas and Data Analysis An Introduction for Scientists and Statisticians Ronald Christensen, Wesley Johnson, Adam Branscum, and Timothy E. Hanson Catalog no. K10199, July 2010, 516 pp. ISBN: 978-1-4398-0354-7, $69.95 / £44.99 Bayesian Adaptive Methods for Clinical Trials Scott M. Berry, Bradley P. Carlin, J. Jack Lee, and Peter Müller Catalog no. K11217, July 2010, 323 pp. ISBN: 978-1-4398-2548-8, $89.95 / £57.99 Bayesian Modeling in Bioinformatics Edited by Dipak K. Dey, Samiran Ghosh, and Bani K. Mallick Catalog no. C7017, September 2010, c. 466 pp. ISBN: 978-1-4200-7017-0, $89.95 / £57.99