ICSA - WordPress.com · 2016-05-31 · ICSA Bulletin Volume 24/2, July, 2012 ISSN 2226-2393...
54
Transcript of ICSA - WordPress.com · 2016-05-31 · ICSA Bulletin Volume 24/2, July, 2012 ISSN 2226-2393...
ICSABulletin
Volume 24/2, July, 2012ISSN 2226-2393
Editorial Staff
Editor-in-Chief
Contributing Editors
Haodao FuYihui Xie
Editorial Assistant
Gong-Yi [email protected]
Executive Committee
PresidentIvan S. F. [email protected]
Past PresidentNaisyin [email protected]
President-ElectMing-Hui [email protected]
Executive DirectorShu-Yen [email protected]
TreasurerLynn [email protected]
Consulting Editor
Fang [email protected]
Cover Design: Kan Wu
Contents of this issue:
Looking Back and Looking Out . . . . . . . . . . . . . . . . 42
From the 2012 President, ICSA . . . . . . . . . . . . . . . . 43
From the Executive Director, ICSA . . . . . . . . . . . . . . 44
Reader’s Feedback: On History of ICSA . . . . . . . . . . . 44
Candidates for 2012 Election of ICSA Officers . . . . . . . . 45
ICSA Member Meeting and Annual Banquet at JSM 2012 . 53
ICSA Financial Report, January – June, 2012 . . . . . . . . . 54
Report from the Program Committee . . . . . . . . . . . . . 55
Highlights of 2012 ICSA Applied Statistics Symposium . . 56
Recipients of Student Paper Awards and Travel Grants . . 59
Report From OICSA . . . . . . . . . . . . . . . . . . . . . . 62
2012 Report on the ICSA Journal Statistics in Biosciences . 63
New Papers in ICSA Journals . . . . . . . . . . . . . . . . . 64
New Fellows of ASA and IMS . . . . . . . . . . . . . . . . . 68
People News . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
In Memory of My Mentor Pao-Lu Hsu . . . . . . . . . . . . 70
Academic Achievements of Professor Pao-Lu Hsu . . . . . 74
Statisticians Supporting Late-stage Clinical Development
at Merck . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
Bringing Statistical Innovation to Oncology Drug Devel-
opment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Reproducible Research: Notes from the Field . . . . . . . . 88
Making Reproducible Research Enjoyable . . . . . . . . . . 89
Upcoming Events . . . . . . . . . . . . . . . . . . . . . . . . 91
Professional Opportunities . . . . . . . . . . . . . . . . . . . 91
Editorial July 2012 Vol.24/2
Looking Back and Looking OutJun Yan
The first Pao-Lu Hsu Prize, an award to be pre-sented every three years by the ICSA, will be an-nounced this August. The award is to “an individ-ual under the age of 50, who makes influential andfundamental contributions to any field of statis-tics and probability, and exemplifies Hsu’s deep in-volvement in developing statistics and probabilityresearch with significant impact on education.” Forthe younger generation, who was Pao-Lu Hsu?
I heard Hsu’s name for the first time in early1990s in a mathematical statistics course taught byProf. Yaoting Zhang, who mentioned his professor,Hsu, with great passion and deep respect. It wasnot until about ten years later when I learned moreabout Hsu from a short article in “Leading Person-alities in Statistical Sciences: From the SeventeenthCentury to the Present”, edited by Johnson andKotz (1997), after I started teaching. The short ar-ticle made me want to know even more about him.In column “Looking Back”, we have two memo-rial articles about Prof. Hsu and his impact fromboth personal and academic perspectives. The firstis by Dihe Hu, and the second by Jiading Chenand Zhongguo Zheng; all were students and col-leagues of Hsu. I would suggest reading more arti-cles collected at a website (http://www.math.pku.edu.cn/teachers/Hsu/articles.htm) in memoryof Prof. Hsu, hosted by his home department, theDepartment of Probability and Statistics, PekingUniversity. Interestingly, you will find that Hsuwas not only a top scientist, but also versatile intraditional Chinese arts and music.
For those who are curious about how exactlystatisticians play an important role in clinical anddrug development, we have two articles in the“Statisticians at Work” column. Jerald Schindlerand Yang Song tell about statisticians supportinglate-stage clinical development at Merck. Pan-durang Kulkarni, Nathan Enas, and Yanping Wangshare their experience in bringing statistical inno-vation to oncology drug development at Eli Lilly,and illustrate what “advanced today, routine to-morrow” is, something we statisticians look outproudly. I am very happy to see more articles fromour members on statisticians at work, perhaps inresponse to the first article in this column in the Jan-uary issue. In fact, the first article in this column byRuberg and Fu (2012) is cited here — another rea-son to write for the ICSA Bulletin.
Reproducible research and what it meansfor statisticians have gained much public atten-
tion in the news recently. In column “BlogSpot”, we republish an article on field experi-ence in reproducible research by Roger Peng, oneof the three biostatistics professors behind blog“Simply Statistics” (http://simplystatistics.tumblr.com/). In column “R ‘ R’ Us”, our columneditor Yihui Xie tells how to make reproducible sta-tistical research enjoyable with his R package knitr,a redesign of Sweave.
On ICSA business, we have: reports from Pres-ident Ivan Chan, Executive Officer Shu-yen Ho,Treasurer Lynn Kuo, and program committee chairTianxi Cai; biosketches of candidates of 2012 ICSAofficers; highlights of the 2012 ICSA Applied Statis-tics Symposium from the co-chairs Mingxiu Huand Tianxi Cai; report of winning entries of stu-dent paper awards and travel grant from JianhuaHuang and Siva Tian; report from the office ICSA;and report from the co-editors-in-chief of Statisticsin Biosciences. The Joint Conference of the ICSAand the International Society for Biopharmaceuti-cal Statistics (ISBS) is announced to be held duringJune 9–12, 2013, in Bethesda, Maryland.
Getting ready for the JSM 2012? Our JSM LocalChair, Ronghui (LiLy) Xu has put together a nice,detailed list of information about ICSA membermeeting and annual banquet at JSM 2012, includingtransportation, banquet menu, additional recom-mended restaurants, and Chinese restaurants nearthe Convention Center. Websites of the restaurantsare printed and also clickable if you view the elec-tronic version with a computer or a hand-held de-vice such as iPad/iPod or a smart phone.
Prof. Dayue Chen, Chairman of the Depart-ment of Probability and Statistics at Peking Univer-sity offered tremendous help in soliciting the twomemorial articles on Prof. Hsu. Contributing edi-tor Dr. Haoda Fu organized the two articles in col-umn “Statisticians at Work”. Contributing editorYihui Xie kept writing articles in the “R ‘ R’ Us” col-umn. My volunteer assistant Gong-yi Liao refinedthe ICSABul LATEX package and helped to design theback-cover. Dr. Lihan Yan from the FDA handledthe mailing labels. My sincere thanks to all of them!
Enjoy the rest of the summer and see many ofyou in San Diego!
Jun YanEditor-in-chief, ICSA BulletinAssociate ProfessorDepartment of StatisticsUniversity of Connecticut
42 © www.icsa.org
July 2012 Vol.24/2 From the ICSA Executives
From the 2012 President, ICSAIvan S. F. Chan
Dear friends and ICSA mem-bers,
Wow, summer is here withus already! As I am writ-ing this article, I have just re-turned home from the ICSAApplied Statistics Symposiumat Westin Water Front Hotel inBoston. This year’s sympo-sium featured three wonder-
ful keynote speakers, Dr. Richard Simon fromNCI, Professor Andrew Lo from MIT, and Profes-sor Bradley Efron from Stanford University. Therewere multiple short courses offered on the first day,and over 100 high-quality scientific sessions overthe next three days covering a wide range of top-ics such as financial statistics, statistics practice inlaw, statistics genetics, statistics in clinical trialsand regulatory applications, The symposium wasvery well attended with >550 participants fromacademia, industry and government. In addition,the two social events, Boston harbor cruise on Sun-day evening and Conference Banquet on Mondaynight were all sold out. Professor Stephen Blythfrom Harvard University gave an insightful speechat the banquet on the importance of statistics in fi-nancial management. Finally, the symposium wasalso highlighted, unexpectedly, by being in thesame hotel with Mr. President Obama! Overall, Ithink this symposium is very successful and showsthe growth and capability of our association. Iwould like to express my deepest gratitude to thesymposium organizing team, led by Mingxiu Hufrom Millennium, Tianxi Cai from Harvard Univer-sity, and Hongliang Shih from Millennium.
The membership of ICSA has grown rapidlyfrom just under 500 a few years ago to over 1300currently. Last year, the constitution and by-lawswere revised to reflect the current state of the so-ciety and allow further expansion of the member-ship, especially via sections of special interests andlocal chapters. I am very excited to announce thatthe first ever chapter of ICSA, the Canada Chap-ter, was born on June 23, 2012. Kudos to Profes-sor Grace Yi of Waterloo University, who spent alot of time and efforts in developing the gover-nance structure and by-laws of the Canadian chap-ter. Many thanks also go to our Shu-yen Ho, ourExecutive Director, for working closely with Grace
and the Executive Committee and Board of Direc-tors in finalizing the chapter by-laws. I believeICSA is in a strong position to support the growthof local chapters in providing membership benefitsto local members.
Another piece of excellent news to share is theestablishment of ICSA-Springer Book Series in May2012. This is the first of its kind of collaboration be-tween ICSA and a publisher to publish a series ofbooks on different fields of statistics. We are for-tunately to have Professor Jiahua Chen from Uni-versity of British Columbia serve as the Editor-In-Chief of this book series. Together with the ICSAsponsored/co-sponsored journals, Statistica Sinica,Statistics in Biosciences, and Statistics and Its Inter-face, this book series will enhance the benefits tomembers and increase the influence of ICSA in theinternational statistical community. If you are inter-ested in writing a book on statistics, please give theICSA-Springer Book Series a serious consideration.
ICSA is continuing to explore ways of establish-ing collaborations with other societies in the future.The planning of the first joint symposium betweenICSA and the International Society for Biophar-maceutical Statistics (ISBS) is well underway, withthe date and venue set for June 9–12, 2013 at theBethesda North Marriott Hotel & Conference Cen-ter, Bethesda, Maryland, USA. The 2013 ICSA Inter-national Conference, to be held from December 20–23, 2013 in Hong Kong, will be co-sponsored by theAmerican Statistical Association (ASA), Institute ofMathematical Statistics (IMS), and International So-ciety of Bayesian Analysis (ISBA). We also have hada discussion with ENAR about a possible collabora-tion and submitted an ICSA-sponsored invited ses-sion proposal for the 2013 ENAR spring meeting.In addition, ICSA is planning to host the AppliedStatistics Symposium jointly with the Korean Inter-national Statistical Society (KISS) in 2014 and withthe Graybill Conference in 2015.
If you are planning to attend the Joint Statisti-cal Meetings (JSM) in San Diego (July 29 to Au-gust 2, 2012), please note that ICSA will hold theannual Members Meeting on August 1 (Wed) at5:30 pm in San Diego Convention Center. Dur-ing the Members Meeting, the 2012 election resultswill be announced and an award ceremony will beheld for this year’s ICSA Distinguished Achieve-ment Award and Outstanding Service Award. Inaddition, the first ICSA Pao-Lu Hsu Award recipi-ents will be announced, with the official award cer-
© www.icsa.org 43
From the ICSA Executives July 2012 Vol.24/2
emony to be held at the 2013 International Confer-ence in Hong Kong. An ICSA banquet will followthe members meeting at a local restaurant.
I look forward to seeing you at the JSM andwish you a summer with productivity and fun.
Ivan S. F. Chan2012 President, ICSASenior DirectorLate Development StatisticsMerck Research Laboratories
From the Executive Director, ICSAShu-yen Ho
Dear ICSA Members,Over the past six months,
ICSA enjoyed membershipgrowth and additional con-nections with other statisticalinstitutions and organizationsthrough co-sponsorship andco-organizing future meet-ings. For the membershipgrowth, currently we have ap-
proximately 1340 active members and 40% amongthose are permanent members. For the connectionswith other organizations, please refer to our web-site www.icsa.org and other reports on this bulletinfor more information.
The 2012 Applied Statistical Symposium wassuccessfully held in Boston recently with over 550participated. A report on the symposium can befound in this bulletin. It is worth noting thatthis symposium website was developed and main-tained by the home office staff and the online reg-istration was enhanced and streamlined to betterserve symposium registrants. We plan to continueto build on this success for all future symposia andwe welcome your feedback and suggestions.
The 2012 first ICSA board meeting was held onJune 23, 2012 during the Applied Statistical Sym-
posium. In that meeting, the birth of Canada Chap-ter was announced, the board approved ICSA 2012awards, including the first PL Hsu award, and alsoapproved the candidates for 2012 ICSA elections.In addition, the ICSA journals, book series, and fu-ture symposia and international conferences werealso updated and all were on track and in goodprogress.
The annual ICSA members meeting will be heldon August 1, 2012, 5:30 PM at Hilton Bayfront Ho-tel San Diego (room Indigo 202) during the upcom-ing JSM, where the 2012 awards and election resultswill be announced. We look forward to your partic-ipation in this members meeting. Immediately afterthe members meeting, ICSA welcome all membersto join a banquet in a local Chinese restaurant. De-tails are included in this bulletin and will also beavailable at the ICSA table in JSM.
As always, your support of and participation inICSA programs and activities will be important forthe continued success of ICSA. Your ideas and sug-gestions will be greatly appreciated.
Have a great second half of the Dragon year!
Shuyen Ho, Ph.D.ICSA Executive Director (2011-13)Director, Statistics and ProgrammingGlaxoSmithKline
Reader’s Feedback: On History of ICSADr. Nancy Lo, Mathematical Statistician from Na-tional Oceanic & Atmospheric Administration andformer Executive Director of ICSA (1997), wrote:“It is good to see an article about early days ofICSA. FYI, I wrote an article on the past, the presentand the future of ICSA published in January, 1997,the 10th anniversary of ICSA. This article was alsopublished in Encyclopedia of Statistical Science, Vol3, John Wiley & Sons.”
The citation of the article is:Lo, Nancy (1997): “The past, the present and thefuture of International Chinese Statistical Associa-tion”, in Encyclopedia of Statistical Science, Vol 3,John Wiley & Sons.
We thank Dr. Lo for pointing us to this ref-erence. The reference has been archived at http://www.icsa.org/References/index.html.
Feedbacks from readers are always welcome.
44 © www.icsa.org
July 2012 Vol.24/2 ICSA Business
Candidates for 2012 Election of ICSAOfficersCandidates for 2013 PresidentElect
Lu, Ying
[PRESENT POSITION] Professor, Division of Bio-statistics, Department of Health Research and Pol-icy, Stanford University; Director, US Veterans Af-fairs Cooperative Studies Program Palo Alto Coor-dinating Center, Palo Alto, CA, USA.[FORMER POSITION] Professor, Department ofRadiology and Department of Epidemiology andBiostatistics, University of California, San Fran-cisco, CA, USA.[DEGREES] Ph.D. in Biostatistics (1990), Universityof California, Berkeley, CA, USA; M.S. in AppliedMathematics (1984), Shanghai Jiao Tong University,Shanghai, China; B.S. in Mathematics (1982), FudanUniversity, Shanghai, China.[FIELDS OF MAJOR STATISTICAL ACTIVITIES]Dr. Lu’s research interests include clinical trial de-sign, evaluation and validation of medical diagnos-tic tests, medical decision making, and statisticalapplications in osteoporosis, radiology, oncologyand other disease areas.[PUBLICATIONS] Dr. Lu has published over 210peer-reviewed research papers in statistical andmedical journals. His statistical works were pub-lished in Biometrics, Statistics in Medicine, Statis-tics in Biopharmaceutical Research, Medical Deci-sion Making, Journal of Biopharmaceutical Statis-tics, Contemporary Clinical Trials, Statistical andProbability Letters, and Mathematics Biosciences.He also published in clinical journals, such as Jour-nal of the American Medical Association (JAMA),Proceedings of the National Academy of Sciences(PNAS), Journal of Bone and Mineral Research, Ra-diology, Cancer, Osteoporosis International, andNeuroimage. He co-edited Advanced MedicalStatistics with Professor Ji-Qian Fang, which waspublished in both English and Chinese.[ICSA ACTIVITIES AND OFFICES HELD] Dr. Luis a lifetime member of the ICSA. He currentlyserves as the Chair of the ICSA Publication Com-mittee (2012-2013). In the past, he served as theChair of the ICSA Program Committee from 2008-2011. He co-chaired the 2009 ICSA Applied Statis-
tics Symposium in San Francisco, CA; chaired theshort-course committee for the 2004 ICSA AppliedStatistical Symposium, San Diego, CA; and orga-nized ICSA local activities for the 2002 JSM in SanFrancisco.
[RELATED PROFESSIONAL ACTIVITIES] Dr. Luis an elected fellow of the American Statistical As-sociation. He was the recipient of the 1990 EvelynFix Memorial Award for excellent dissertation fromthe Department of Statistics, University of Califor-nia, Berkeley. He also received the 2003 Health-star Osteoporosis Medical Research Award by theChinese Development Foundation for Science andTechnology for his research accomplishment in thestandardization of hip BMD, osteoporosis diagno-sis, quality control and quality assurance for den-sitometry in major osteoporosis clinical trials, andhis teaching, consulting, and promoting of osteo-porosis research in China.
Dr. Lu has been actively involved in the sta-tistical and research communities. He served asthe Program Chair for the WNAR 2006 AnnualMeeting. He was a member of the WNAR Re-gional Committee during 2007-2009, the planningand organization committee of the 1st Pacific CoastStatisticians and Pharmacometricians InnovationConference (PaSiPHIC), and the scientific programcommittees for many statistical and clinical con-ferences and symposiums. Dr. Lu has servedas a reviewer and study section member for NIHgrant applications since 2001; a member of the FDAAdvisory Panel on Peripheral CNS Diseases from2007-2011; a member of the American Joint Com-mittee on Cancer (AJCC) Statistical Task Force on9th Edition of AJCC Tumor Staging from 2006-2009,and the AJCC Molecular Modeler Group since2007. He also served as the only statistical mem-ber in the International Committee for Standard-ization in Bone Measurement (1995-2000) and theCommittee on Standards in Bone Measurementsof the International Society of Clinical Densitom-etry (2005). He is a member of the external advi-sory board of Tufts Medical Center – Cancer Center,the external advisory board of the UCLA Neuro-Oncology SPORE program, and the internationaladvisory board of Shanghai Jiao Tong UniversitySchool of Bioengineering. In addition, Dr. Lu con-tinuously served as an ASA San Francisco Chap-
© www.icsa.org 45
ICSA Business July 2012 Vol.24/2
ter officer from 1995-2010, including twice as theChapter President. Dr. Lu is a member of the ed-itorial board of the Journal of Molecular Diagno-sis, a statistical co-editor of the Shanghai Archiveof Psychiatry, and co-editor for an ICSA special is-sue on clinical trials for Statistics and Its Interface(SII). He has been a reviewer for 22 statistical andclinical journals.
[STATEMENT] I thank the Nomination and Elec-tion Committee for the honor of being selected asa candidate for the president of the ICSA. In thepast 20 years, ICSA has been an integral part of myprofessional life. Through ICSA activities, I havemet so many outstanding statisticians and receivedtremendous help and support. I appreciate everyopportunity to return the favor by serving our or-ganization. If I were given the opportunity to serveas the president of the ICSA, I would work withour colleagues and listen to their ideas and initia-tives that promote the continued growth and en-hance the impact of the ICSA.
The success of an organization depends on itsvibrant members. Under the leadership of pastpresidents, ICSA has grown into an association of1,000 members. It is both an exciting challenge andan opportunity to continue to foster the growth ofthe association. We can attract the most talentedstatisticians not only through our professional ser-vices but also by encouraging members to haveownership of the association through their initia-tives and activities. I would like to work with mem-bership committee to expand sections and chaptersso members can cultivate their specific regional in-terests and objectives. In addition, we want to ex-pand our membership to areas such as MainlandChina and Europe. We can host international webi-nars to provide cutting edge continuing educationcourses to potential new members as outreach tostatisticians in those areas. We also want to takefull advantage of social media web tools, such asGoogle+ to hold large multi-person video meet-ings; LinkedIn to establish a professional presenceand increase prominence by having our memberslist ICSA in their profiles; or other social mediumthat could provide free recruitment and promotionfor the association.
Our organization would not be able to functionwithout the generosity of hard working volunteers,officers, and directors who devote their time andeffort to enable the ICSA thrive. Many excellentideas and best practices have been developed overthe years by our volunteers. If I am elected, I willwork with committees and volunteers to establish
operational manuals, document the best practice,and develop tools for the betterment of the ICSA. Iwill also work with the ICSA Board of Director toimprove ICSA infrastructure to ensure that our vol-unteers get the help they need and reduce the bur-den from administrative tasks, thus allowing mem-bers to focus their energy on creative activities andpromoting statistical sciences.
Compared to other well-established statisticalassociations, ICSA is unique in its truly interna-tional representation. It is an organization madeup of individuals of all race, creed, color, sex, andnationality. Its membership comes from all overthe world. In the past, ICSA has had bi-annualinternational conferences in Asia and annual ap-plied statistical symposium in the North Americathat bridges international exchanges. However, ourrecent past efforts have been focused on scholarlyactivities. The ASA has demonstrated the benefitsof promoting better understanding and interest instatistics in the general public through its K-12 sta-tistical education program. The ICSA is in a privi-leged position to exercise its influence on K-12 sta-tistical education in Asia. These and other similarinitiatives can help our passionate members inspirethe next generations of statisticians. I hope to chan-nel all the incredible talents of this association tomake a positive impact in the world.
ICSA is 25 years old this year. This is a beautifulage - full of energy and imagination. I am certainlylooking forward to working with our organizationregardless of the election results. LetâAZs work to-gether to make ICSA the most successful organiza-tion.
Shen, Wei
[PRESENT POSITION] Senior Director, GlobalStatistics and Advanced Analytics, Eli Lilly andCompany.[FORMER POSITIONS] Research Advisor –Globalhealth outcomes (2010-2011), Head of Statistics,Late Phase and commercialization (2003-2009),Principal Research Scientist (2001-2003), SeniorStatistician (1996-2000), Eli Lilly and Company.[DEGREES] Ph.D. in Biostatistics, University ofMinnesota, 1996; M.S. in Biostatistics, University ofMinnesota, 1994; B.S. in Mathematics, University ofMinnesota, 1991.[FIELDS OF MAJOR STATISTICAL ACTIVITIES]Dr. Shen’s research interests include statisticalmethods and application in clinical trials, Bayesianand empirical Bayes methods, joint modeling of
46 © www.icsa.org
July 2012 Vol.24/2 ICSA Business
longitudinal and time-to-event data, survival anal-ysis, non-parametric methods, and health out-comes research.
[PUBLICATIONS] Dr. Shen has published over 30manuscripts in mainstream statistical and medicaljournals, including Journal of Royal Statistics So-ciety – Series B, The American Statistician, Statis-tics in Medicine, Journal of Computational andGraphics Statistics, Journal of BiopharmaceuticalStatistics, Health Services and Outcomes ResearchMethodology, Medical Care, Journal of Urology,Journal of Rheumatology, Journal of Bone and Min-eral Research, Obstetrics and Gynecology, Osteo-porosis International, Arthritis and Rheumatism,Journal of the American College of Cardiology, andOral Oncology. In addition, Dr Shen has con-tributed to over 80 oral and poster presentations atmajor statistical and medical conferences.
[ICSA ACTIVITIES AND OFFICES HELD] Dr.Shen is a life member of ICSA. He currently servesas a member of the ICSA nomination committee.Dr. Shen was elected to the ICSA board in 2007,and he served on the ICSA Board of Directors from2008 to 2010. Dr. Shen co-chaired the 2010 ICSAApplied Statistics Symposium in Indianapolis. Hehas served on various ICSA committees, includingICSA local event committee (2000), membershipcommittee (2004 âAS 2007), committee of promot-ing academic-industrial collaboration (2009-2010),and nomination committee (2012). Dr Shen hasbeen involved in organizing invited sessions atICSA Applied Statistics Symposium and ICSA In-ternational Conference.
[RELATED PROFESSIONAL ACTIVITIES] Dr.Shen is an elected member of the International Soci-ety for Biopharmaceutical Statistics (ISBS) Confer-ence Program Committee. He has served on thescientific program committee for the 1st and 2ndISBS international conferences in Shanghai (2008)and Berlin (2010). Currently, he serves on the sci-entific program committee for the 2nd Joint Bio-statistics Symposium in Beijing (2012) and the jointISBS/ICSA statistics symposium in WashingtonDC (2013). Dr Shen has organized and chaired ses-sions at major statistical conferences, including theJoint Statistical Meetings, FDA/Industry StatisticsWorkshop, and ISBS international conferences. Dr.Shen has served as a reviewer for several journals,including Biometrics, Statistics in Medicine, TheAmerican Statistician, Computational statistics anddata analysis, Journal of Biopharmaceutical Statis-tics, Quality of Life Research, Medical Care, andJournal of clinical epidemiology.
[STATEMENT] It is a great honour to be nominatedas a candidate for the president of ICSA. I’d liketo thank Ivan Chan (ICSA president) and Sue-JaneWang (Chair of the ICSA nomination committee)for their encouragement and support.
My first involvement with ICSA came in 2000,when the joint statistical meetings (JSM) were heldin Indianapolis and I served as a member of thelocal event committee. Over the years I was for-tunate to serve ICSA and its members in variousways, including sitting on the ICSA board of direc-tors (2008 to 2010), participating in various ICSAcommittees, and co-chairing the 2010 ICSA appliedstatistical symposium in Indianapolis.
With its outstanding leadership and highly en-gaged members, ICSA has become one of the pre-mier statistical organizations in the world. As a life-time member of ICSA, I am very excited to witnessand be a part of the growth of ICSA. I have seenwhere we have been, and I am passionate about theopportunity to lead ICSA to reach the next level.
If elected, I plan to focus on the following areasto make ICSA a stronger and more influential orga-nization:
1) Grow the influence of ICSA through ex-panded collaborations. In this age of informationexplosion and social networking, growth will comethrough effective collaborations. Over the years,ICSA has developed strong partnerships with sev-eral sister organizations, including the AmericanStatistical Association. I will continue these effortsto expand our partnerships with a broad range oforganizations, particularly in the area of member-ship drive and sponsorship of conferences. Whileserving on the ICSA committee of “promotingacademic-industrial collaboration” (2009 to 2010),I have worked closely with ICSA past-presidentXuming He to introduce ways of productive col-laborations between academic and industrial statis-ticians. As our membership from industry contin-ues to rise, I will support initiatives to strengthencollaborations among industry, academia, and gov-ernment by leveraging information and network-ing. Finally, collaborations should not be limitedby where we are located. ICSA, as an interna-tional statistical organization, has the responsibilityto lead collaborations across the globe. Given mostof our members are US-educated with Chinese-origin, we have a unique opportunity to bridgethe knowledge gap between US and China/Asia.As the demand of statistical talents continues toexplode, future talents are expected to increas-ingly come from China/Asia. Through our rich
© www.icsa.org 47
ICSA Business July 2012 Vol.24/2
network and educational programs, ICSA is posi-tioned to raise the supply and prominence of next-generation of statisticians in China/Asia to meetthe global growth of statistical applications. I willlead ICSA to increase our efforts in scientific collab-orations, educational programs, and outreach ef-forts in China, Asia, and Europe.
2) Transform ICSA through scientific leader-ship and operational excellence. ICSA has alreadyestablished its scientific leadership in the statisti-cal discipline through its flagship journal Statis-tica Sinica, and its annual applied statistical sym-posium. I will lead ICSA to strengthen our sci-entific leadership position, by propelling StatisticaSinica into a top-tier journal in the internationalstatistics community, and making our newly estab-lished journal Statistics in BioSciences one of thefast growing journals in the field of applied statis-tics. Given the growing importance of statisticalapplications outside of North America, I will workwith the program committee to increase of the fre-quency of ICSA international conferences and pos-sibly add regional conferences. Our leadership notonly lies in our scientific work, but also in the qual-ity and skills of our members. After all, our mem-bers are our most important assets. With so manyyoung talents joining ICSA every year, we owe ourmembers a huge responsibility of career growthand leadership development. ASA, under its cur-rent president, Robert Rodriguez, has launched aseries of effort to improve the leadership skills ofits members. For members of ICSA, we must pro-vide opportunities for growth and character build-ing: âAc I will create a task force focusing on men-toring and leadership development. âAc I will ad-vocate inclusion of leadership development in ourpublications and conferences, to enable our mem-bers to become successful scientists and leaders inthe 21st century.
Last but not least, with our fast growing mem-bership and size of our conferences, our opera-tional burden continues to increase. Thanks to pastICSA executive director and president-elect Ming-hui Chen, the Office of ICSA started its operationin 2007. In order for ICSA to function well as a pre-mier professional organization, we are at a tippingpoint to establish a right operational structure sothat ICSA can sustain its growth in the next decade.I will lead efforts to expand ICSA office by addingpermanent and professional staff. This is not to di-minish the support of our volunteers. Indeed, ouroperational excellence will allow us to deliver bet-ter services to our members and enable our vol-unteers to focus on strategic and scientific leader-
ship of our association. By providing a strong op-erational framework, our scientific leadership maygrow through the spark of innovative thinking andcollaboration.
Thank you for giving me the opportunity toshare my ideas and passion about the future ofICSA. I am excited and committed to the missionof ICSA. Inspired by many dedicated ICSA mem-bers and leaders, I am humbled by this opportunityto serve ICSA as its president. I ask for your sup-port and invite you to join me in our journey to takeICSA to the next level together.
Candidates for 2013 BiometricsSection Chair
Liu, Aiyi
[PRESENT POSITION] Senior Investigator, Bio-statistics and Bioinformatics Branch, EuniceKennedy Shriver National Institute of Child Healthand Human Development, National Institutes ofHealth.[FORMER POSITION] Investigator, Biostatis-tics and Bioinformatics Branch, Eunice KennedyShriver National Institute of Child Health and Hu-man Development, National Institutes of Health;Assistant Professor, Department of Biostatistics,Georgetown University Medical Center.[FIELDS OF MAJOR STATISTICAL ACTIVITIES]Research interests include general statistical the-ory and methods, sequential and adaptive methodsin clinical trials and biomedical research, statisti-cal methodology for diagnostic biomarkers, semi-parametric and nonparametric methods for multi-dimensional data, genome-wide association stud-ies, and group testing methodology.[SELECTED PUBLICATIONS] Author of about100 publications including 66 peer-reviewed pa-pers in statistical journals including Biometrics,Biometrika, Biostatistics, Statistica Sinica, andStatistics in Medicine.[ICSA ACTIVITIES] Permanent member of ICSA;Member, Planning Committee, ICSA 2005 AppliedStatistics Symposium; Session organizer and chairon “Pooling Biospecimens: Methodologies, Appli-cations and Limitations”. ICSA Applied StatisticsSymposium, 2005; Session organizer and chair on“Sequential Methods for Evaluation of DiagnosticBiomarkers”. ICSA Applied Statistics Symposium,2006; Member, Board of Directors, ICSA, 2007-2009; Session organizer on “Dealing with Measure-
48 © www.icsa.org
July 2012 Vol.24/2 ICSA Business
ments Subject to Limits of Detection in Epidemi-ology Studies”. ICSA Applied Statistics Sympo-sium, 2007; Member, Program Committee, ICSAApplied Statistics Symposium, 2011; Co-chair, Pro-gram Committee, 2013 ICSA Applied StatisticsSymposium.[RELATED PROFESSIONAL ACTIVITIES] Mem-ber, Committee on Award of an Outstanding Statis-tical Application, the American Statistical Associa-tion (ASA), 2011-2013; Associate Editor, Journal ofStatistical Planning and Inference, 2009-2012; GuestEditor for “Mathematical and Statistical Methodsfor Diagnoses and Therapies”. Philosophical Trans-action of the Royal Society A, Number 1874, Vol-ume 366, July 2008.
Liu, Guanghan (Frank)
[PRESENT POSITION] Director, Clinical Biostatis-tics, Merck Research Laboratories, North Wales,PA.[FORMER POSITION] Biometrician, Sr. Biome-trician, and Associate Director (1995-2005), Clini-cal Biostatistics, Merck Research Laboratories, BlueBell, PA, Postdoctoral Research Fellow (1994-1995),Department of Biostatistics, The Johns HopkinsUniversity, Baltimore, MD; Research Statistician(1993-1994), Graduate School of Education, Univer-sity of California, Los Angeles, CA[DEGREES] Ph.D in Statistics, University of Cali-fornia, Los Angeles, CA, 1993; M.S. in Mathemat-ics, University of California, Los Angeles, CA, 1990;M.S. in Statistics, East China Normal University,China, 1987; B.S. in Mathematics, East China Nor-mal University, China, 1984.[FILED OF MAJOR STATISTICAL ACTIVITIES]Research in clinical trial design and data analysisincluding longitudinal data analysis; missing datamethodology; analysis of survival data; Bayesianmethods in clinical trials specifically in assessingprobability of success and in handling missingdata.[SELECTED PUBLICATIONS] Published morethan 30 papers in statistical journals. Recent pub-lications including Pharmaceutical Statistics (2012,2011), J of Biopharm Stat (2011a, 2011b), Statistics inMedicine (2010, 2009a, 2009b, 2008), ContemporaryClinical Trials (2008).[ICSA ACTIVITIES] Permanent Member of ICSA;Served as Board of Director (2007-2009), Member ofPublication Committee (2012), Member of organiz-ing committee for the ICSA Applied Statistics Sym-posium (2008) and Special Issue Editorial Board for
publishing special issues at Statistics in Biopharma-ceutical Research (2009).[RELATED PROFESSIONAL ACTIVITIES] Activemember of ASA and ENAR, organizing and/orchair sections and making presentations; Co-chairDIA scientific working group on Bayesian meth-ods for missing data handling; Referee for sta-tistical journals including Biometrics, Statistics inMedicine, Journal of Biopharmaceutical Statistics,Pharmaceutical Statistics, Statistics in Biopharma-ceutical Research, and The American Statistician.
Candidates for Directors of theICSA Board (2013–2015)
Zhang, Zhengjun
[PRESENT POSITION] Associate Professor, De-partment of Statistics, University of Wisconsin-Madison.[FORMER POSITION] Assistant Professor, De-partment of Statistics, University of Wisconsin-Madison (2005-2010). Assistant Professor, Depart-ment of Mathematics, Washington University inSaint Louis (2002-2005).[DEGREES] Ph.D. in Statistics, University of NorthCarolina at Chapel Hill, 2002; Ph.D. in Manage-ment Engineering, Beijing University of Aeronau-tics & Astronautics, 1996; M.S. in ComputationalMathematics, Academia Sinica, Beijing, 1993; B.S.in Computational Mathematics and Software, Yun-nan University, Kunming, 1986.[FIELD OF MAJOR STATISTICAL ACTIVITIES]Statistics of extremes with application to finance,insurance and environmental sciences; time series;risk management; medical statistics, etc.[SELECTED PUBLICATIONS] Refereed papers arepublished in major statistical and medical sciencejournals, including Annals of Statistics, Journal ofBanking and Finance, Journal of Time Series, Ad-vances in Econometrics, Journal of Applied Prob-ability, Extremes, Journal of Statistical Planningand Inferences, Insurance: Mathematics and Eco-nomics, Annals of Family Medicine.[ICSA ACTIVITES] ICSA student paper award andtravel fellowship, 2002; Organizer of invited talksessions for ICSA applied statistics symposium;Lifetime member of ICSA.[RELATED PROFESSIONAL ACTIVITIES] Asso-ciate Editor of Journal of Business and EconomicStatistics starting 2012, and Associate Editor of
© www.icsa.org 49
ICSA Business July 2012 Vol.24/2
Journal of Korean Statistical Society since 2010;NSF Review Panelist; Organizing committee, sci-entific program committee, and local committee ofabout ten international conferences, symposiums,and workshops on statistics; organizer and chairfor many other invited sessions for JSM, IMS meet-ings, and CFE meetings worldwide.
Zhang, Wei
[PRESENT POSITION] Regional Head of Bio-metrics and Data Management, Asia/MENA,Boehringer Ingelheim.[FORMER POSITION] Senior Principal Biostatisti-cian, Boehringer Ingelheim[FILED OF MAJOR STATISTICAL ACTIVITIES]Research interests include adptive designs, non-inferiority trials, doubly censored data, and sur-vival analysis.[SELECTED PUBLICATIONS] Published morethan a dozen of papers in statistical and medicaljournals.[ICSA ACTIVITIES] Member of ICSA; Member ofExecutive Committee for ICSA 2011 Applied Statis-tics Symposium[RELATED PROFESSIONAL ACTIVITIES] Servedas Vice President of ASA Connecticut Chapter; Pro-gram Chair of 9th Annual ASA Connecticut Chap-ter Meeting, Program Co-chair of 6th Annual ASAConnecticut Chapter Meeting; Chair of a topic-contributed session for 2010 JSM; Chair of an in-vited session for 1st International Symposium onBiopharmaceutical Statistics.
Ji, Yuan
[PRESENT POSITION] Director of Cancer Infor-matics, Center for Clinical and Research Informat-ics, NorthShore University HealthSystem.[FORMER POSITION] Associate Professor, Depart-ment of Biostatistics, The University of Texas M.D.Anderson Cancer Center.[FILED OF MAJOR STATISTICAL ACTIVITIES]Research interests include Bayesian Inference, ge-nomics (next-generation sequencing and other highthroughput biotechcology), integromics, networkmodels, and adaptive designs for clinical trials.[SELECTED PUBLICATIONS] Over 50 papers andmanuscripts ranging across a variety of jour-nals, including JASA, Biometrics, Bioinformatics,JNCI, Lancet Oncology, Clinical Trials, Statistics inMedicine, and Statistica Sinica.
[ICSA ACTIVITIES] Member of ICSA; Organizersof the ICSA invited sessions in ICSA annual meet-ings (2010-2012), including a session at the ICSA in-ternational workshop at Guangzhou in 2010.[RELATED PROFESSIONAL ACTIVITIES] ASAHouston Chapter President, 2006, NIH Study Sec-tion Ad Hoc Members, NIH R01 Awardee (2008).
Wang, Ming-Dauh
[PRESENT POSITION] Principal Research Scien-tist, Statistical Science, Eli Lilly and Company.[FILED OF MAJOR STATISTICAL ACTIVITIES]Research and application of statistics in pharma-ceutical setting; interests include clinical trials,Bayesian inference, adaptive design, biomarker in-ference.[SELECTED PUBLICATIONS] Published papers instatistical and medical journals, including Statisticsin Medicine, Biometrical Journal, and Journal of theAmerican Medical Association.[ICSA ACTIVITIES] Member of ICSA; Treasurer of2010 ICSA Applied Statistical Conference.[STATEMENT] Having long benefited from the ser-vice of ICSA and started to give back, I am feel-ing the desire and burden to serve more as a Boardmember. If elected, I will strive along with otherBoard members to bring about fuller realization ofICSA objectives.
Fu, Haoda
[PRESENT POSITION] Senior Research Scientist,Group Leader, Global Statistics, Eli Lilly and Com-pany[DEGREES] Ph.D. in Statistics, University of Wis-consin – Madison, 2007. B.S. in Probability andStatistics, Nankai University, 2002.[FIELDS of MAJOR STATISTICAL ACTIVITIES]Dr. Fu’s research interests include developingand applying Bayesian methods in pharmaceuti-cal research. He published 21 manuscripts andcovers areas in: Bayesian adaptive design, in-direct and mixed treatment comparison, copulamethods for joint modeling with applications inoncology and diabetes areas, Bayesian decisionrule for drug development go/no-go decision, andBayesian method for drug safety evaluation.[RELATED PROFESSIONAL SERVICES AND AC-TIVITIES] Dr. Fu is actively involved in pro-fessional activities. He is a member of ICSA,
50 © www.icsa.org
July 2012 Vol.24/2 ICSA Business
ENAR and ASA. He was selected as ENAR Re-gional Advisor Board member in 2009 and cur-rently serves the ICSA membership committee. Dr.Fu chaired numerous invited and contributed ses-sions in ENAR, JSM, and MBSW (Midwest Bio-pharmaceutical Statistics Workshop) conferences.He also served as reviewers for multiple statis-tical and medical journals, including statistics inmedicine, journal of biopharmaceutical statisticsetc.[STATEMENT] It is my great honor to be nomi-nated as a candidate for the director of ICSA. Thankthe Nomination and Election Committee for thisopportunity. During the past 5 years, I was sograteful to have the chance to closely work withmany talented statisticians in Lilly, including WeiShen and Yongming Qu who are very active in ourICSA communities. I was also very glad to havemany external collaborations. Through these col-laborations, I feel that there is a need to improve theconnection between pharmaceutical research andcurrent methodology research in academia. Al-though there are many statisticians working in in-dustry, they may not be well represented in differ-ent statistician associations. I hope that my involve-ment can continue our ICSA tradition to bring di-versified opinions, also to promote our ICSA in in-dustry.
Lo Huang, Mong-Na
[PRESENT POSITION] Professor, Department ofApplied Mathematics, National Sun Yat-sen Uni-versity, Kaohsiung, Taiwan.[FORMER POSITION] Director of University Li-brary, National Sun Yat-sen University(2003-2007),Chairman of Department of Applied Mathematics,National Sun Yat-sen University (1997-1998), As-sociate Professor of General Education, NationalSun Yat-sen University (1984-1987), Associate Pro-fessor of Department of Applied Mathematics, Na-tional Sun Yat-sen University, (1987-1990). Assis-tant Professor of Department of Mathematics, Oak-land University, U.S.A. (1983-1984).[FIELDS of MAJOR STATISTICAL ACTIVITIES]Research interests include Experimental designs,Industrial statistics, Environmental statistics.[SELECTED PUBLICATIONS] Published 51 papersin statistical journals including Biometrika, En-vironmetrics, Metrika, Sankhya, Statistica Sinica,Statistics in Medicine, Journal of Multivariate Anal-ysis, Journal of Statistical Planning and Inference,Computational Statistics and Data Analysis.
[RELATED PROFESSIONAL SERVICES AND AC-TIVITIES] Associate Editor for Statistica Sinica,Metrika, Journal of the Chinese Statistical Associa-tion, Executive Managing Editor for Journal of DataScience, Member of the Board of Review Commit-tee of National Science Council, Taiwan. Mem-ber of the Council Committee of National ScienceCouncil, Taiwan. Committee Member of the Boardof the Chinese Institute of Probability and Statisticsof Supervisors.
Tseng, George C.
[PRESENT POSITION] Associate Professor, De-partment of Biostatistics, University of Pittsburgh(Primary appointment). Associate Professor, De-partment of Human Genetics, University of Pitts-burgh (Secondary appointment). Associate Pro-fessor, Department of Computational and SystemsBiology, University of Pittsburgh (Secondary ap-pointment).[FILED OF MAJOR STATISTICAL ACTIVITIES]Research interests include statistical methodologyand inference of high-throughput genomic dataanalysis, genomic meta-analysis, genomic integra-tive analysis, machine learning and clustering anal-ysis.[SELECTED PUBLICATIONS] Published 25 ma-jor papers in statistical and bioinformatic journalsincluding JASA, Biometrics, Annals of AppliedStatistics, Bioinformatics, Nucleic Acids Researchand BMC Bioinformatics.[ICSA ACTIVITIES] Member of ICSA; Invitedspeaker of Annual ICSA Applied Statistics Sympo-sium (2006, 2008, 2010, 2012).[RELATED PROFESSIONAL ACTIVITIES] Asso-ciate editor for BMC Medical Genomics and Jour-nal of Computational and Graphical Statistics.
Lu, Wenbin
[PRESENT POSITION] Associate Professor, De-partment of Statistics, North Carolina State Univer-sity[FORMER POSITION] Assistant Professor, Depart-ment of Statistics, North Carolina State University[FIELDS of MAJOR STATISTICAL ACTIVITIES]Research interests include statistical methods forclinical trials and personalized medicine, sur-vival analysis, nonparametric/semi-parametric in-ference, model and variable selection methods andstatistical genetics.
© www.icsa.org 51
ICSA Business July 2012 Vol.24/2
[SELECTED PUBLICATIONS] Published 34 papersin statistical journals including Biometrika, JASA,Statistica Sinica, Biometrics, Biostatistics and Life-time Data Analysis.[ICSA ACTIVITIES] Permanent member of ICSA;organize an invited session for the ICSA AppliedStatistics Symposium in 2007, 2008 and 2011.[RELATED PROFESSIONAL SERVICES AND AC-TIVITIES] Organize invited and topic contributedsessions for the JSM and ENAR meetings. Fac-ulty fellow for the SAMSI research program onrisk analysis, extreme events and decision theoryin 2007–2008. Guest editor for a special issue ofJournal of Probability and Statistics. Associate ed-itor for Biostatistics, Biometrics, Journal of Statisti-cal Theory and Practice.
Chang, Yuan-Chin Ivan
[PRESENT POSITION] Research Fellow, Instituteof Statistical Science, Academia Sinica, Taipei, Tai-wan.[FORMER POSITION] National Chengchi Univer-sity, Taipei, Taiwan; Associate Research Fellow,Institute of Statistical Science, Academia Sinica,Taipei, Taiwan; Deputy Director, Institute of Sta-tistical Science, Academia Sinica, Taipei, Taiwan;Supervisor, Statistical Computing Lab, Institute ofStatistical Science, Academia Sinica, Taipei, Tai-wan.[FILED OF MAJOR STATISTICAL ACTIVITIES]Research interests include Sequential Analysis andIts Applications, Generalized linear models andClassification[SELECTED PUBLICATIONS] Published papersin statistical journals including Annals of Statis-tic, Statistica Sinica, Biostatistics, Biometrics, Bio-metrical Journal, Psychometrika, Journal of Sta-tistical Planning and Inference, Neurocomputing,Bioinformatics, Statistics and Probability Letters,Metrika, Computational Statistics and Data Anal-ysis, Journal of Educational and Behavior Statisics.[RELATED PROFESSIONAL ACTIVITIES] Admin-istration Services: International Association of Sta-tistical Computing Asian Regional Section BoardMember (ISAC-ARS BoD, since 2009); Member ofLegal Affairs Committee, Ministry of Examination,Examination Yuan of ROC (2009); Advisory Com-mittee: Academia Sinica Computing Center (2005- 2006); Committee of Art and Culture Activity:
Academia Sinica (2000 - 2007); Supervisor of Com-puting Lab., Inst. of Statistical Science, AcademiaSinica.
Organizing Committee member of workshops andconferences: 2011 Taipei Symposium and 7th IASC-ARS Conference Dec. 16–19, 2011; Statistics Sci-ence Camp 2008, 2009; Machine Learning Sum-mer School, 2006; Statistics and Machine LearningWorkshop I to V (2004 to 2006, 2008, 2009); FirstInternational Workshop in Sequential Methodolo-gies, 2007.
Editorial Services: Associate editors: Journal ofChinese Statistics Association; Sequential Analy-sis; Journal of Japanese Society of ComputationalStatistics. Guest Editor: Statistica Sinica – Spe-cial Issue on “Data Mining and Machine Learning”Managing Editor: Statistica Sinica Statistics; Con-sultant: Taiwanese Journal of Psychiatry.
Shi, Jian Qing
[PRESENT POSITION] Reader in Statistics, Leaderof the Group of Applied Statistics and Probability,Postgraduate Selector/Tutor in Statistics, Schoolof Mathematics & Statistics, Newcastle University,UK.[DEGREES] Ph.D in Statistics, The Chinese Univer-sity of Hong Kong, 1996.[FORMER POSITION] 02/2008–03/2008 VisitingFellow, Isaac Newton Institute for MathematicalSciences, Cambridge University. UK. 2002–2012Senior Lecutrer/Lecturer in Statistics, School ofMathematics and Statistics, University of Newcas-tle. Newcastle, UK.[FIELD OF MAJOR STATISTICAL ACTIVITIES]Nonparametric functional data analysis, Incom-plete data and model uncertainty with applicationsin medicine, Covariance structural analysis and la-tent variable models, Statistical diagnostics.[SELECTED PUBLICATIONS] Dr. Shi has pub-lished over 40 refereed papers in major statisticaljournals, including JRSSB, Biometrika, Biometricsand Statistica Sinica .[ICSA ACTIVITES] Member of ICSA.[RELATED PROFESSIONAL ACTIVITIE] Asso-ciate editor of JRSSC (Applied Statistics, 2010-2013), Guest AE for JRSS discussion paper, RCUKMathematics Prioritisation Panel member, APTS(Academic for PhD Training in Statistics) executivecommittee member.
52 © www.icsa.org
July 2012 Vol.24/2 ICSA Business
ICSA Member Meeting and AnnualBanquet at JSM 2012Ronghui (Lily) Xu
2012 JSM will take place on July 28 — August 2at the San Diego Convention Center, California.ICSA member meeting and the annual banquet arescheduled on Wednesday, August 1st. The banquetwill be held at Jasmine Chinese Seafood Restau-rant, 7:00pm — 10:00pm, 4609 Convoy St. #A,San Diego, CA 92111, (858) 268-0888, http://www.jasmineseafood.com/jsr/html/home.html.
Transportation Transportation will be arrangedafter the member meeting at the Convention Cen-ter. Banquet tickets and price (including trans-portation) will be available at the ICSA ConferenceBooth.
Banquet menu
1. Peking Duck2. Sauteed Chicken & Shrimp with Pine Nuts3. Sauteed Seafood in Phoenix Nest4. Supreme Seafood Soup with Bean Curd5. Lobster with Ginger and Scallions6. Braised Garden Greens with Mushrooms7. Peking Pork Chops8. Twin Flavored Rock Cod Fillet9. House Special Fried Rice
10. Dessert of the Day
Additional Recommended Restaurants ConvoyStreet (where the banquet takes place) is whereAsian businesses concentrate in San Diego. Hereare some additional recommended restaurants, allare Chinese except otherwise specified (some aresmall and popular on the weekends, so you mightwant to call first):
• Emerald Restaurant, 3709 Convoy St., SanDiego, CA 92111, (858) 565-6888, http://
emeraldrestaurant.com/
• China Max, 4698 Convoy St. #101, SanDiego, CA 92111, (858) 650-3333, http://
www.chinamaxsandiego.com/
• Spice City, 4690 Convoy St., San Diego, CA92111, (858) 278-1818, http://spicycity.
menutoeat.com/
• Tofu House [Korean], 4646 Convoy St., SanDiego, CA 92111, (858) 576-6433, http://
convoytofuhouse.menutoeat.com/
• Pho T Cali [Vietnamese, near Convoy],7351 Clairemont Mesa Blvd., San Diego,CA 92111, (858) 565-6997, http://www.
photcalisd.com/
• Spicy House, 3860 Convoy St. #105,San Diego, CA 92111, (858) 278-5883,http://www.sandiegochinesepress.com/
yellowpage/san_diego_chinese_business.
php?business_id=3
• Chin’s Restaurant, 4433 Convoy St., SanDiego, CA 92111, (858) 499-8964, http://
www.yelp.com/biz/chins-restaurant-san-diego#
query:Chin%27S%20Convoy
• Great Plaza Buffet (if you are in the Pa-cific Beach area), 1840 Garnet Avenue, SanDiego, CA 92109, (858) 273-6868, http://
greatplazabuffet.com/
Chinese Restaurants Near the Convention Cen-ter Chinese restaurants near the conventioncenter in downtown San Diego (also calledthe Gaslamp area) can be found at: http:
//www.yelp.com/search?find_desc=chinese+
food+downtown+gaslamp&find_loc=San+Diego%
2C+CA
Ronghui (Lily) Xu, Ph.D.JSM Local Chair, ICSAProfessorDivision of Biostatistics and BioinformaticsDepartment of Family and Preventive Medicineand Department of MathematicsDirector, CTRI Design, Biostatistics and EthicsUniversity of California, San Diego9500 Gilman Drive, Mail Code 0112La Jolla, CA 92093-0112Phone: 858-534-6380Fax: 858-534-5273Email: [email protected]
© www.icsa.org 53
International Chinese Statistical Association Profit and Loss
Jan 1, 2012 through June 30, 2012 Balance, Dec 31, 2011 $133,632.12
Income Advertising Fee $100.00 Membership fee $8,860.00 Total Income $8,960.00
Expense Miscellaneous IMS Bulletin, two ½ page ads for Pao-Lu Hsu Award $540.00 Paypal Service charge $220.88 Accountant (2011 Tax Form) $525.00 Jasmin Seafood Restaurant (JSM Banquet Deposit) $1,000.00 Five Star Tours (JSM Banquet Bus) $1,392.00 Total Miscellaneous $3,677.88
Postage and Delivery Formost $2,722.43 Statistica Sinica (20-4, 21-1, 21-2, 21-3, & 21-4) $1,101.00 Total Postage and Delivery $3,823.43
Printing and Reproduction January ICSA Bulletin $3,669.00 Total Postage, Printing, and Reproduction $7,492.43
Total Expense $11,170.31
Net Ordinary Income $-2,210.31
Net Other Income Interest income from CD $1,162.00
Net Income $-1,048.31
July 2012 Vol.24/2 ICSA Reports
International Chinese Statistical Association Balance Sheet
Jan 1, 2012 through June 30, 2012 ASSETS Checking/Savings Checking $80,847.51 CD $51,736.30 TOTAL ASSETS $132,583.81 LIABILITIES & EQUITY
Equity Opening Balance Jan 1, 2012 of ICSA $133,632.12 Jan – Jun 2012 Net Income -$1,048.31 Total Equity $132,583.81 TOTAL LIABILITIES & EQUITY
$132,583.81
Report from the Program CommitteeTianxi Cai
ICSA Program Committee
1. The ICSA is collaborating the Eastern NorthAmerican Region (ENAR) of the InternationalBiometrics society to have one invited sessionat the 2013 ENAR annual conference dedi-cated as a special session organized by theICSA. This year, the invited session proposalwill be submitted as a session on the ICSAsponsored journal, SIB.
2. The program committee is working withthe 2014 applied symposium chair DongseokChoi to initiate discussions on preparing forthe symposium and exploring the possibilityof letting Korean International Statistical So-ciety to co-sponsor the symposium.
3. The ICSA Canadian chapter has been for-mally founded on June 23nd, 2012. ProfessorGrace Yi from the University of Waterloo will
serve as the founding president of the newchapter.
Past events in 2012
1. The ICSA co-sponsored the 19th IMS/ASASpring Research Conference (SRC) on Statis-tics in Industry and Technology: Enablingthe Interface between Statistics and Engineer-ing, which took place successfully at Har-vard University on June 12–15, 2012. Thegoal of the conference was to promote inter-disciplinary research in statistical methods inengineering, science and technology and tostimulate interactions among statisticians, re-searchers in the application areas, and indus-trial practitioners.
2. ICSA 2012 Applied Statistical Symposiumwas held successfully in Westin Waterfront,Boston, Massachusetts between June 23 toJune 26, 2012. Detailed information on this
© www.icsa.org 55
ICSA Reports July 2012 Vol.24/2
event can be found in a separate report in thisbulletin.
3. JSM will take place in August at San Diego,California. There will be an ICSA membermeeting from 5:30 PM to 7:00 PM on August1st and a banquet following the meeting atthe Jasmine Seafood Restaurant. ProfessorRonghui (Lily) Xu, UCSD Biostatistics is theChair for the local committee. Please contacther at [email protected] if you have anyadditional suggestions/comments.
Year 2013
1. ICSA 2013 Applied Statistical Symposiumwill be held in Bethesda, MD. Drs. YiTsong ([email protected]) and Aiyi Liu([email protected]) are the Chairs of theICSA organization committee. The ICSABoard has approved a proposal by the or-ganization committee to join the 2013 sym-posium with the International Conference ofthe International Society for Biopharmaceuti-cal Statistics (ISBS). Details of this event willbe forthcoming.
2. The 2013 (Ninth) ICSA International Con-ference will be held December 20–23, 2013at Lam Woo International Conference Cen-tre, Hong Kong Baptist University. For moreinformation, please contact Professor LixingZhu ([email protected]) at Hong Kong Bap-tist University. Details will be developed andannounced.
Year 2014
1. ICSA 2014 Applied Statistical Symposiumwill be held in Portland, Oregon. Ifyou would like to help, please contact Dr.Dongseok Choi ([email protected]).
Year 2015
1. ICSA 2015 Applied Statistical Sym-posium will be held in Fort Collins,Colorado. If you would like tohelp, please contact Dr. Naitee Ting([email protected])or Professor Haonana Wang([email protected]).
If you would like to have ICSA co-sponsorshipfor statistical conferences and meetings, pleaseuse the website http://www.icsa.org/meetings/
co-sponsorship/index.html to submit your ap-plication for co-sponsorship.
The program committee would appreciate com-ments and suggestions to improve ICSA programs.Please send your inputs to Professor Tianxi Cai([email protected]).
Tianxi Cai, Ph.D.,Chair, ICSA Program Committee (2012–2013)Associate Professor of BiostatisticsDepartment of BiostatisticsHarvard University
Highlights of 2012 ICSA AppliedStatistics SymposiumMingxiu Hu and Tianxi Cai
The 21st ICSA Applied Statistics Symposium wassuccessfully held from June 23 to June 26, 2012, inthe Westin Boston Waterfront Hotel, located in thebeautiful seaport district of Boston, Massachusetts.International Society for Biopharmaceutical Statis-tics (ISBS) and American Statistical Association(ASA) co-sponsored the conference. There were 552conference participants, 140 short course attendees,and 18 student volunteers from Harvard Univer-sity, Brown University, University of New Hamp-shire, and University of Chicago. Over 400 papers
were presented in 103 scientific sessions. The twosocial events, the Boston Harbor Sunset Cruise andthe Banquet, were both sold out with 200 and 250participants, respectively. We would like to apol-ogize to those who wanted to participate in theseevents but could not due to the limitation of capac-ity.
Three keynote speeches were delivered byworld renowned speakers. Dr. Richard Simonof National Cancer Institute presented “On theRoad to Personalized Genomic Medicine in Oncol-ogy,” Professor Andrew Lo of MIT discussed “BigData, Systemic Risk, and Financial Crises,” and
56 © www.icsa.org
July 2012 Vol.24/2 ICSA Reports
Professor Bradley Efron of Stanford University pre-sented “Model Selection, Estimation, and BootstrapSmoothing.” In addition, Dr. Steve Blyth of Har-vard University delivered the banquet speech.
ICSA President Dr. Ivan Chan gives the openingremarks (photo by Hongliang Shi).
Symposium co-chair Dr. Mingxiu Hu receives anappreciation of sponsorship from ICSA president,
Dr. Ivan Chan (photo by Hongliang Shi).
All seven short courses were well attended withan average of 20 attendees per course. The fivehalf-day short courses were: “Regulatory Experi-ences versus Consulting Experiences in Planning
and Implementing Adaptive Clinical Trials” byDrs. Sue-Jane Wang of FDA and Cyrus Mehta ofCytel, “Causal Inference from Observational andRandomized Studies with Treatments that Varyover Time” by Drs. James Robins and Eric Tchet-gen of Harvard University, “Analysis of Biomark-ers for Prognosis and Response Prediction” byDr. Patrick Heagerty of University of Washing-ton, “Statistical Machine Learning in Modern DataAnalysis” by Dr. Hao Helen Zhang of North Car-olina State University, and “Comparative Effective-ness ResearchâATIntroduction for Statisticians” byDrs. Constantine Gatsonis of Brown Universityand Sharon-Lise Normand of Harvard University.The two full-day short courses were: “HybridBayesian Adaptive Clinical Trial Designs” by Dr.Peter Thall of MD Anderson Cancer Research Cen-ter, and “Analysis of Genome-Wide Sequencing As-sociation Studies” by Drs. Xihong Lin of HarvardUniversity and Yun Li of University of North Car-olina at Chapel Hill.
Symposium co-chair Dr. Tianxi Cai introduces akeynote speaker (photo by Hongliang Shi).
This year we increased the number of StudentTravel Awards from 3 to 4 in addition to J.-P. HsuPharmaceutical and Regulatory Science StudentPaper Award. The winners of this year’s compe-tition were Yingqi Zhao from University of NorthCarolina at Chapel Hill (J. P. Hsu Student PaperAward), Peisong Han from University of Michigan,Xueying Chen from Rutgers University, Sy HanChiou from University of Connecticut, and OliviaYueh-Wen Liao from Stanford University. We are
© www.icsa.org 57
ICSA Reports July 2012 Vol.24/2
Symposium participants on Boston Harbor Sunset cruise (photo by Hongliang Shi).
thankful to the Student Award Committee for host-ing this competition and for selecting these excel-lent student papers.
The symposium received financial supportfrom 17 organizations: Millennium/The TakedaOncology Company, the International Society forBiopharmaceutical Statistics (ISBS), American Sta-tistical Association, Abbott, Amgen, BoehringerIngelheim, Brightech, Celgene, Cytel, Genentech,Pfizer, Novartis, Vertex, Eli Lilly, Mitsubishi TanabePharma, Bristol-Myers Squibb, and Sanofi Aven-tis. We are sincerely grateful to these organizationsfor their generous support and to the Fund-raisingCommittee for their successful efforts.
The committee is considering a proposal bySpringer to publish the first proceeding of the ICSAApplied Statistics Symposium. If you are inter-ested in publishing your papers in the proceedingor have any questions, please contact us.
We would like to extend our genuine gratitudeto all the volunteers, speakers, meeting partici-pants, sponsors, and exhibitors. It is all because ofyou that this symposium becomes successful. Weare especially grateful to the organizing commit-tee members and many other volunteers for con-tributing numerous hours in preparing the confer-
ence, and to the programming committee membersand session organizers for assembling the keynotespeeches, the short courses, and the 103 scientificsessions.
Hope you enjoyed the conference and your visitto Boston! To share a few fine moments, several pic-tures taken by Hongliang Shi are presented in thisissue of the ICSA Bulletin.
Mingxiu Hu, Ph.D.Co-Chair, ICSA 2012 AppliedStatistics SymposiumHead of Biostatistics & StatisticalProgrammingMillennium: The Takeda OncologyCompany
Tianxi Cai, Ph.D.,Co-Chair, ICSA 2012 AppliedStatistics SymposiumAssociate Professor of BiostatisticsDepartment of BiostatisticsHarvard University
58 © www.icsa.org
July 2012 Vol.24/2 ICSA Reports
Recipients of Student Paper Awards andTravel GrantsJianhua Huang and Siva Tian
The student travel award committee met on April18th, 2012. The committee received 20 submissionsfor this year’s competition. All papers are of highquality and worthy of consideration for an award.It was difficult for the committee to make the de-cision. After careful considerations, the committeeselected five papers for an award because they con-tain extraordinary innovations and they togethercover different applied areas. Each awardee re-ceived $400 cash award after the conference regis-tration and free registration for one short course.All awardees were invited to present their papersat the 2012 ICSA Applied Statistics Symposium.
The committee selected the paper
• “Estimating Individualized treatment rulesusing outcome weighted learning” by YingqiZhao (University of North Carolina, Advi-sors: Donglin Zeng and Michael Kosorok)
for the Jiann-Ping Hsu Pharmaceutical and Regu-latory Sciences Student Paper Award.
The other four papers selected for a travelaward are:
• “Efficient Estimation For Missing OutcomeData With Surrogate Using Conditional Em-pirical Likelihood” by Peisong Han (Univer-sity of Michigan, Advisors: Lu Wang and Pe-ter X.-K. Song)
• “Biomarker-Based Adaptive Accrual Designsfor Confirmatory Oncology Trials” by OliviaYueh-Wen Liao (Stanford University, Advi-sor: Tze-Leung Lai)
• “Semiparametric Multivariate AcceleratedFailure Time Model with Generalized Esti-mating Equations” by Sy Han Chiou (Univer-sity of Connecticut, Advisor: Jun Yan)
• “A Split-and-Conquer Approach for Analy-sis of Extraordinarily Large Data” by XueyingChen (Rutgers University, Advisor: MingeXie)
We would like to take this opportunity of thankall the committee members:
• Jianhua Huang (Texas A&M University)Chair
• Siva Tian (University of Houston) Co-chair
• Xuelin Huang (MD Anderson Cancer Center)
• Hao Liu (Baylor College of Medicine)
• Peng Wei (UT Health Science Center)
The abstract of the winning papers are as fol-lows.
Authors: Yingqi Zhao, Donglin Zeng, A. John Rush,Michael R KosorokTitle: Estimating Individualized treatment rules us-ing outcome weighted learningAbstract: There is increasing interest in discover-ing individualized treatment rules for patients whohave heterogeneous responses to treatment. In par-ticular, one aims to find an optimal individual-ized treatment rule, which is a deterministic func-tion of patient specific characteristics maximizingexpected clinical outcome. In this paper, we firstshow that estimating such an optimal treatmentrule is equivalent to a classification problem whereeach subject is weighted proportional to his or herclinical outcome. We then propose an outcomeweighted learning approach based on the supportvector machine framework. We show that the re-sulting estimator of the treatment rule is consistent.We further obtain a finite sample bound for the dif-ference between the expected outcome using the es-timated individualized treatment rule and that ofthe optimal treatment rule. The performance of theproposed approach is demonstrated via simulationstudies and an analysis of chronic depression data.
Authors: Peisong Han, Lu Wang, Peter X.-K. SongTitle: Efficient Estimation For Missing OutcomeData With Surrogate Using Conditional EmpiricalLikelihoodAbstract: We consider parametric regression wherethe outcome is subject to missingness, yet some sur-rogate is available. Missing at random mechanismis assumed. Under conditional mean structuremodel, we propose a conditional empirical likeli-hood (CEL) method for estimation and inference.The proposed method requires no construction ofestimating equations. We study CEL-based inverseprobability weighted (CEL-IPW) and augmentedinverse probability weighted (CEL-AIPW) estima-tors in details. CEL-AIPW estimator possesses
© www.icsa.org 59
ICSA Reports July 2012 Vol.24/2
Jiann-Ping Hsu Pharmaceutical and Regulatory Sciences Student Paper Award recipient Yingqi Zhao withTianxi Cai and Karl Peace (photo by Hongliang Shi).
double robustness property, and attains the semi-parametric efficiency bound when missing mecha-nism and conditional mean of the outcome givensurrogate and covariates are both correctly mod-eled. CEL-IPW estimator is consistent if the miss-ing mechanism is correctly modeled. Numericalimplementation is discussed, asymptotic distribu-tions are established, and superior ïnAnite sampleperformance compared to some widely used esti-mators is demonstrated through simulation exper-iments. Data collected from an intervention studyof adolescents of parents with HIV are analyzed asapplication.
Authors: Olivia Yueh-Wen LiaoTitle: Biomarker-Based Adaptive Accrual Designsfor Confirmatory Oncology TrialsAbstract: Recent scientific advances in disease biol-ogy have brought us into a new era in drug devel-opment. The newfound heterogeneity of diseaseshas driven the pharmaceutical industry toward thedevelopment of personalized therapy. However,with these new advances come new challenges. Us-ing conventional clinical trial designs for targeted
therapies exposes some patients who do not bene-fit from the treatment to unnecessary side effects.The use of conventional designs often also leads toa failure to identify existing treatment effects dueto dilution in the overall population. It is thereforecrucial to devise appropriate patient accrual crite-ria when designing confirmatory clinical trials oftargeted therapies. However, typical phase II stud-ies usually do not have enough data to support thedetermination of proper patient accrual criteria, es-pecially when the treatment effect in the resistantpatients is highly uncertain. In this paper, we pro-pose an adaptive accrual design that can effectivelyselect the population that benefits from the treat-ment and efficiently implements early stopping cri-teria (for futility and efficacy). The operating char-acteristics of the proposed design are demonstratedby simulation studies. Some potential issues withthese kinds of trials and the proposed methodologyare also discussed.
Authors: Sy Han Chiou, Junghi Kim, and Jun YanTitle: Semiparametric Multivariate AcceleratedFailure Time Model with Generalized Estimating
60 © www.icsa.org
July 2012 Vol.24/2 ICSA Reports
EquationsAbstract: The semiparametric accelerated failuretime model is not as widely used as the Cox relativerisk model mainly due to computational diïnCcul-ties. Recent developments in least squares esti-mation and induced smoothing estimating equa-tions provide promising tools to make the accel-erate failure time models more attractive in prac-tice. For semiparametric multivariate acceleratedfailure time models, we propose a generalized esti-mating equation approach to account for the mul-tivariate dependence through working correlationstructures. The marginal error distributions canbe either identical as in sequential event settingsor different as in parallel event settings. Some re-gression coefficients can be shared across marginsas needed. The initial estimator is a rank-basedestimator with GehanâAZs weight, but obtainedfrom an induced smoothing approach with com-putation ease. The resulting estimator is consis-tent and asymptotically normal, with a variance es-timated through a multiplier resampling method.In a simulation study, our estimator is up to threetimes as efficient as the initial estimator, especiallywith stronger multivariate dependence and heaviercensoring percentage. Two real examples demon-strate the utility of the proposed method.
Authors: Xueying Chen, Minge XieTitle: A Split-and-Conquer Approach for Analysisof Extraordinarily Large DataAbstract: If there are extraordinarily large data, toolarge to fit into a single computer or too expensiveto perform a computationally intensive data anal-ysis, what should we do? To deal with this prob-lem, we propose in this paper a split-and-conquerapproach and illustrate it using a computationallyintensive penalized regression method, along witha theoretical support. Consider a regression set-ting of generalized linear models with n observa-tions and p covariates, in which n is extraordinar-ily large and p is either bounded or goes to ∞ ata certain rate of n. We propose to randomly splitthe data of size n into K subsets of size O(n/K).For each subset of data, we perform a penalized re-gression analysis and the results from each of the
K subsets are then combined to obtain an overallresult. We show that the combined overall resultstill retains all the desired properties of penalizedestimators such as the model selection consistencyand asymptotic normality under mild conditions.When K is less than O(n1/5), we also show thatthe combined result is asymptotically equivalent tothe corresponding analysis result of using the entiredata all together, assuming that there were a supercomputer that could carry out such an analysis. Inaddition, the split-and-conquer approach involvesa random splitting and a systemic combining. Wedemonstrate that the approach has an inherent ad-vantage of being more resistant to false model se-lections caused by spurious correlations, and wefurther establish an upper bound for the expectednumber of falsely selected variables and a lowerbound for the expected number for truly selectedvariables. Furthermore, when a computational in-tensive algorithm is used in the sense that its com-puting expense is at the order of O(na), a > 1,we show that the split-and-conquer approach cansubstantially reduce computing time and computermemory requirement. The proposed methodologyis demonstrated numerically using both simulationand real data examples.
Jianhua Huang, Ph.D.Chair, Student Award Committee2012 ICSA Applied Statistics Sym-posiumProfessorDepartment of StatisticsTexas A&M University
Siva Tian, Ph.D.Co-Chair, Student Award Commit-tee2012 ICSA Applied Statistics Sym-posiumAssistant ProfessorDepartment of PsychologyUniversity of Houston
© www.icsa.org 61
ICSA Reports July 2012 Vol.24/2
Report From OICSALili Yu
Dear ICSA members,It has been almost six years since the central of-
fice of ICSA (OICSA) was established. The centraloffice has taken on more and more tasks over theseyears. I would like to take this opportunity to in-troduce the OICSA and thank all parties for theirsupport and help.
OICSA was established in 2006 by the ICSAExecutive Committee at the request of Dr. KarlPeace to memorialize the importance of ICSA tohis late wife, Dr. Jiann-Ping Hsu (http://jphcoph.georgiasouthern.edu/about/hsu), who was ac-tive in and provided great contribution to the ICSA,particularly in the early years of its growth. Thepurpose of the OICSA is to assist the ICSA to func-tion more efficiently, appropriately and profession-ally, so that ICSA members are better served. Cur-rently, OICSA provides the following services: (1)membership enrollment, renewal and update, (2)necessary assistance to the Executive Director, (3)ICSA Bulletin distribution, (4) communication andassistance to different parties within ICSA, such ascommittees and members, for various of tasks, (5)webmaster services, including but not limited toposting documents on web and maintenance of thewww.icsa.org and ICSA server, (6) emailing of theICSA e-newsletter, (7) helping ICSA members ac-cess on-line issues of Statistica Sinica and Statisticsin Biosciences as needed, (8) hosting office hours, (9)processing and posting job lists on ICSA website,(10) helping the local committee for ICSA sympo-sium, and (11) hosting the ICSA informational deskat Joint Statistical Meetings. The office will con-tinue enriching services as needed to better servethe entire ICSA.
OICSA is incubated in the Jiann-Ping Hsu Col-lege of Public Health (JPHCOPH) at Georgia South-ern University. Currently, ICSA supports a Bio-statistics graduate student at OICSA. Three grad-uate students have worked at this position so far:Chunfeng Ren, Adam Chen, and Jingxian Cai.Their hard work on a daily basis makes great con-tributions to OICSA and ICSA. To ensure smoothoperation at OICSA, some professors and staff pro-vide unremunerated help. Lili Yu (Assistant Pro-fessor), Karl Peace (Professor) and Charles Hardy(Professor and former Dean of JPHCOPH) super-
vise OICSA. Ruth Whitworth, the IT support inJPHCOPH, provides great help on webmasteringfor the ICSA website. The office team members en-joy serving the ICSA and try hard to provide betterand better service to this association.
Office of ICSA at Jiann-Ping Hsu College of PublicHealth, Georgia Southern University.
The office’s growth and continued success de-pends on help from all parties. Here, we wouldlike to thank the presidents for their support. Spe-cial thanks go to the Executive Directors, Ming-HuiChen and Shu-yen Ho, for their advice, help andunderstanding. Thank Simon Gao for his prompthelp on IT-related issues. In addition, thanks go tothe Editors of the ICSA Bulletin, Fang Yu and JunYan; the ICSA treasurer, Lynn Kuo; the committeechairs; and all members for their kindness, cooper-ation, and input.
To provide excellent service to all members,OICSA would like to hear from our members aboutsuggestions or ideas on improving the work andfunctionality of the office. Please feel free to con-tact and discuss with us any issues you may haverelated to OICSA.
Best regards,
Lili YuOICSAAssistant ProfessorDepartment of BiostatisticsJiann-Ping Hsu College of PublicHealthGeorgia Southern University
62 © www.icsa.org
July 2012 Vol.24/2 ICSA Reports
2012 Report on the ICSA JournalStatistics in BiosciencesXihong Lin, José Pinheiro and Hongyu Zhao
Statistics in Biosciences (SIBS), the ICSA appliedstatistical journal, was launched in 2009, in partner-ship with Springer. It aims at development and ap-plication of statistical methods and their interfacewith other quantitative methods, such as computa-tional and mathematical methods, in biological andlife science, health science, and biopharmaceuticaland biotechnological science. SIBS has published 7issues, with two issues a year. The editorial boardconsists of 20+ associate editors.
In the last three years, with the support of theSIBS editorial board, the ICSA leadership and itsmembers, the journal has flourished. By the endof 2011, the number of submissions has steadilyincreased. The number of downloads of the pub-lished papers from the Springer SIBS sites has dou-bled. SIBS has online subscriptions by 154 institu-tions/libraries worldwide. A total of 4,794 institu-tions worldwide have online exposure to the jour-nal as part of an online deal with Springer (con-sortia, multi-site licenses, and site licenses). TheICSA members can have online access to SIBS freeof charge through the ICSA website.
In addition to journal subscriptions, Springeralso sends out Table of Contents (ToC) alerting. In2011 Springer sent out a total of 9.3 million ToCalerts to over 682,000 subscribers. Researchers caneasily register for this free service on the journal’shomepage. The ToC Alerts inform readers when anew issue is available online.
With the strong support of the editorial board,SIBS offers fast and high quality reviews. The av-erage review time from submission to final accep-tance of a new paper was about 5 months in 2011.The average time from acceptance of a paper to itsappearance online first at the Springer SIBS websiteis about 1 month.
An application of SIBS into the SCI indexdatabase is currently under review, and an appli-cation of SIBS into PubMed is underway.
SIBS has published several special issues oncutting edge areas in health sciences, including“Statistical Methods in Comparative Effectiveness
Research” (June 2011), and “Statistical Methodsfor Network Analysis” (July 2012). Several addi-tional special issues are underway, such as issueson “Statistical Methods for Analysis of Sequenc-ing Data” and “Statistical Methods for Personal-ized Medicines.”
SIBS has held invited journal sessions at ma-jor statistical meetings such as JSM, ICSA AppliedStatistical Symposium, and the ICSA InternationalConference. An invited SIBS session at the 2013ENAR has recently approved by the 2013 ENARprogram committee.
For submission and more detailed informationabout SIBS, please visit http://www.springer.
com/statistics/life+sciences,+medicine+
%26+health/journal/12561
Xihong Lin, Ph.D.Co-Editor-in-ChiefStatistics in BiosciencesProfessorDepartment of BiostatisticsSchool of Publich HealthHarvard University
José Pinheiro, Ph.D.Co-Editor-in-ChiefStatistics in BiosciencesSenior Director, BiostatisticsJohnson & Johnson PharmaceuticalResearch and Development, LLC
Hongyu Zhao, Ph.D.Co-Editor-in-ChiefStatistics in BiosciencesProfessorDepartment of Epidemiology andPublic HealthYale University
© www.icsa.org 63
New Papers from ICSA Journals July 2012 Vol.24/2
New Papers in ICSA JournalsStatistica Sinica
Statistica Sinica endeavors to meet the needs of statisticians faced with a rapidly chang-ing world. It publishes significant and original articles that promote the principled use ofstatistics along with related theory and methods in quantitative studies, essential to mod-ern technologies and sciences. It is published quarterly in January, April, July and October.
Volume 22, Number 3, July 2012
http://www3.stat.sinica.edu.tw/statistica/J22N3/22-3.html
Effect of heavy tails on ultra high dimensional variable ranking methodsAurore Delaigle and Peter Hall
Principal component analysis in very high-dimensional spacesYoung Kyung Lee, Eun Ryung Lee and Byeong U. Park
Regularization and model selection with categorial effect modifiersJan Gertheiss and Gerhard Tutz
Standardization and the group Lasso penaltyNoah Simon and Robert Tibshirani
Variable selection in partly linear regression model with diverging dimensions for right censored dataShuangge Ma and Pang Du
Robust combination of model selection methods for predictionXiaoqiao Wei and Yuhong Yang
On model selection strategies to identify genes underlying binary traits using genome-wide associationdata
Zheyang Wu and Hongyu ZhaoNew efficient and robust estimation in varying-coefficient models with heteroscedasticity
Jie Guo, Maozai Tian and Kai ZhuSemiparametric pseudo likelihoods for longitudinal data with outcomedependent nonmonotone nonre-sponse
Deyuan Jiang and Jun ShaoTuning parameter selection for penalized likelihood estimation of Gaussian graphical model
Xin Gao, Daniel Q. Pu, Yuehua Wu and Hong XuThe strength of statistical evidence for composite hypotheses: Inference to the best explanation
David R. BickelA switching Markov chain Monte Carlo method for statistical identifiability of nonlinear pharmacokineticsmodels
Seongho Kim and Lang LiHigh dimensional exponential family estimation via empirical Bayes
Omkar Muralidharanα-stable limit laws for harmonic mean estimators of marginal likelihoods
Robert L. Wolpert and Scott C. SchmidlerIsomorphism examination based on the count vector
Chang-Yun Lin and Shao-Wei ChengMinimal dependent sets for evaluating supersaturated designs
Arden Miller and Boxin TangJackknifed Whittle estimators
Masanobu Taniguchi, Kenichiro Tamaki, Thomas J. DiCiccio and Anna Clara Monti
64 © www.icsa.org
July 2012 Vol.24/2 New Papers from ICSA Journals
Sparse paired comparisons in the Bradley-Terry modelTing Yan, Yaning Yang and Jinfeng Xu
Volume 22, Number 2, April 2012
http://www3.stat.sinica.edu.tw/statistica/J22N2/22-2.html
Clustering high dimension, low sample size data using the maximal data piling distanceJeongyoun Ahn, Myung Hee Lee and Young Joo Yoon
Bayesian wavelet-based curve classification via discriminant analysis with Markov random tree priorsFrancesco C. Stingo, Marina Vannucci and Gerard Downey
Analysis of cohort survival data with transformation modelKani Chen, Liuquan Sun and Xingwei Tong
Model checking techniques for assessing functional form specifications in censored linear regression modelsLarry F. León and Tianxi Cai
Profiled forward regression for ultrahigh dimensional variable screening in semiparametric partially linearmodels
Hua Liang, Hansheng Wang and Chih-Ling TsaiExtended BIC for Small-n-Large-P Sparse GLM
Jiahua Chen and Zehua ChenAdaptive semi-varying coefficient model selection
Tao Hu and Yingcun XiaA unified variable selection approach for varying coefficient models
Yanlin Tang, Huixia Judy Wang, Zhongyi Zhu and Xinyuan SongLarge sample properties of the scad-penalized maximum likelihood estimation on high dimensions
Sunghoon Kwon and Yongdai KimOn the derivatives of the trimmed mean
Subhra Sankar Dhar and Probal ChaudhuriNonparametric estimation of a smooth density with shape restrictions
Mary C. MeyerQuantile regression for competing risks data with missing cause of failure
Yanqing Sun, Huixia Judy Wang and Peter B. GilbertA covariance regression model
Peter D. Hoff and Xiaoyue NiuA finite mixture model for working correlation matrices in generalized estimating equations
Lili Xu, Nan Lin, Baoxue Zhang and Ning-Zhong Shi
Bias-robustness and efficiency of model-based inference in survey samplingDesislava Nedyalkova and Yves Tille
Bayesian finite population imputation for data fusionJerome P. Reiter
Bayesian inference from composite likelihoods, with an application to spatial extremesMathieu Ribatet, Daniel Cooley and Anthony C. Davison
Methods for identifying influential variables in an out-of-control multivariate normal processChia-Ling Yen and Jen Tang
Minimal sufficient confounding information among main effects and two-factor interactionsJianwei Hu and Runchu Zhang
Optimal designs for two-level factorial experiments with binary responseJie Yang, Abhyuday Mandal and Dibyen Majumdar
© www.icsa.org 65
New Papers from ICSA Journals July 2012 Vol.24/2
Statistics in Biosciences
Statistics in Biosciences (SIB) is published twice a year in print and electronic form. Itaims at development and application of statistical methods and their interface with otherquantitative methods, such as computational and mathematical methods, in biological andlife science, health science, and biopharmaceutical and biotechnological science.
Volume 4, Number 1, May 2012
http://www.springerlink.com/content/1867-1764/4/1/
Special Issue “Methods for Analysis of Graphs and Networks and Their Applications in Biosciences”An Efficient Optimization Algorithm for Structured Sparse CCA, with Applications to eQTL Mapping
Xi Chen and Han LiuA Two-Step Penalized Regression Method with Networked Predictors
Chong Luo, Wei Pan and Xiaotong ShenNetwork Based Prediction Model for Genomics Data Analysis
Ying Huang and Pei WangAdaptive Thresholding for Reconstructing Regulatory Networks from Time-Course Gene Expression Data
Ali Shojaie, Sumanta Basu and George MichailidisA Latent Eigenprobit Model with Link Uncertainty for Prediction of Protein–Protein Interactions
Xiaoyu Jiang and Eric D. KolaczykA Bayesian Approach to Pathway Analysis by Integrating Gene–Gene Functional Directions and Microar-ray Data
Yifang Zhao, Ming-Hui Chen, Baikang Pei, David Rowe and Dong-Guk Shin, et al.New Approaches to Principal Component Analysis for Trees
Burcu AydÄsn, GÃabor Pataki, Haonan Wang, Alim Ladha and Elizabeth Bullitt, et al.Frequent Pattern Discovery in Multiple Biological Networks: Patterns and Algorithms
Wenyuan Li, Haiyan Hu, Yu Huang, Haifeng Li and Michael R. Mehan, et al.Surveying Hard-to-Reach Groups Through Sampled Respondents in a Social Network: A Comparison ofTwo Survey Strategies
Tyler H. McCormick, Ran He, Eric Kolaczyk and Tian Zheng
Statistics and Its Interface
Statistics and Its Interface is an international statistical journal promoting the interface be-tween statistics and other disciplines including, but not limited to, biomedical sciences,geosciences, computer sciences, engineering, and social and behavioral sciences. The jour-nal publishes high-quality articles in broad areas of statistical science, emphasizing sub-stantive problems, sound statistical models and methods, clear and efficient computationalalgorithms, and insightful discussions of the motivating problems.
Volume 5, Number 2, 2012
http://www.intlpress.com/SII/SII-vol-5.php#SII-5-2
Special issue — Spatial Statistics
Bayesian areal wombling using false discovery ratesPei Li, Sudipto Banerjee, Alexander M. McBean and Bradley P. Carlin
Variogram estimation in the presence of trendNikolay Bliznyuk, Raymond J. Carroll, Marc G. Genton and Yuedong Wang
66 © www.icsa.org
July 2012 Vol.24/2 New Papers from ICSA Journals
Dynamical random-set modeling of concentrated precipitation in North AmericaNoel Cressie, Renato Assuncao, Scott H. Holan, Michael Levine, Orietta Nicolis, Jun Zhang and Jian
ZouSpatial analysis linking landscape features and genetic population structure in cougars (Puma concolor) inthe northern Rocky Mountains
David C. Wheeler and Lance A. WallerAdditive hazards regression and partial likelihood estimation for ecological monitoring data across space
Feng-Chang Lin and Jun ZhuLikelihood-based estimation of spatial intensity and variation in disease risk from locations observed witherror
Dale L. Zimmerman, Peng Sun and Xiangming FangNonparametric estimation of the dependence of a spatial point process on spatial covariates
Adrian Baddeley, Ya-Mei Chang, Yong Song and Rolf TurnerGeneral Submissions
Power of the Cochran-Armitage trend test when exposure scores are based on empirical quantiles of expo-sure
Huilin Li and Mitchell H. GailA modified Bartlett test for linear hypotheses in heteroscedastic one-way ANOVA
Jin-Ting Zhang and Xuefeng LiuRandom threshold for linear model selection, revisited
Merlin Keller and Marc Lavielle
Volume 5, Number 1, 2012
http://www.intlpress.com/SII/SII-vol-5.php#SII-5-1
A review of statistical methods for protein identification using tandem mass spectrometryOliver Serang and William Noble
Protein identification problem from a Bayesian point of viewYong Fuga Li, Randy J. Arnold, Predrag Radivojac and Haixu Tang
Spectral library searching for peptide identification in proteomicsHenry Lam
Bayesian false discovery rates for post-translational modification proteomicsYan Fu
Permutation methods for testing the significance of phosphorylation motifsHaipeng Gong and Zengyou He
Analyzing LC-MS/MS data by spectral count and ion abundance: two case studiesThomas I. Milac, Timothy W. Randolph and Pei Wang
Generalized linear and mixed models for label-free shotgun proteomics and Supplementary MaterialsMatthew C. Leitch, Indranil Mitra and Rovshan G. Sadygov
Protein quantitation using iTRAQ: Review on the sources of variations and analysis of nonrandom miss-ingness
Ruiyan Luo and Hongyu ZhaoProtein structural model selection based on protein-dependent scoring function
Zhiquan He, Jingfen Zhang, Yang Xu, Yi Shang and Dong XuStatistical methods for proteomic biomarker discovery based on feature extraction or functional modelingapproaches
Jeffrey S. MorrisConstructing human phenome-interactome networks for the prioritization of candidate genes
Yong Chen, Wangshu Zhang, Mingxin Gan and Rui Jiang
© www.icsa.org 67
People News July 2012 Vol.24/2
New Fellows of ASA and IMS2012 ASA Fellows
Mousumi Banerjee, University of Michigan, AnnArbor, Michigan For outstanding and sustainedresearch, collaboration and mentoring involvingstatistical methods, theory and design for clinicaltrials, and for service to the profession.
Sudipto Banerjee, University of Minnesota, Min-neapolis, Minnesota For theoretical, method-ological and applied research in spatiotempo-ral statistical modeling, especially as applied toproblems in environmetrics, ecology, occupationalhealth, agriculture and economics, for professionalwork at the local and national levels and for edito-rial service to the profession.
Dongseok Choi, Division of Biostatistics, OregonHealth & Science University, Portland, OregonFor contributions to statistical methods in bioinfor-matics and spatial modeling, service to the statisti-cal profession and to biostatistics education.
Scott R. Evans, Harvard School of Public Health,Wrentham, Massachusetts For excellence in thedesign and analysis of clinical trials, consulting andeducation in clinical research, methodological de-velopment, and for outstanding leadership and ser-vice to the ASA and the statistics profession.
Gareth James, University of Southern Califor-nia, Los Angeles, California For outstanding,impactful research contributions on contemporarystatistical theory, methods and applications inthe areas of functional data analysis and high-dimensional variable selection, for excellence inteaching, and for conscientious service to the pro-fession.
A. James O’Malley, Harvard Medical School,Boston, Massachusetts For novel use of Bayesianstatistics, multivariate-hierarchical modeling,causal inference and social network analysis tosolve problems in health policy and health servicesresearch, for improving evaluation of treatmentsand quality of health care, and for leadership inhealth policy statistics.
Liang Peng, Georgia Institute of Technology, At-lanta, Georgia For significant research in extremevalue theory, nonlinear time series and nonpara-metric statistics, with an emphasis on applicationsto actuarial science and risk management.
José C. Pinheiro, Janssen R&D, Raritan, New Jer-sey For the development of novel methods andsoftware in mixed-effect models and dose find-ing; for influential efforts in statistical consulting;for important contributions to statistical educationthrough book authorship and training; and for out-standing leadership in service to the profession andthe ASA.
Abdul J. Sankoh, Vertex Pharmaceuticals, Depart-ment of Biometrics, Cambridge, MassachusettsFor fine research contributions to the field of bio-statistics, and the enhancement of clinical trial de-sign and data analysis.
Thaddeus Tarpey, Wright State University, Day-ton, Ohio For influential contributions to statis-tical research and applications, particularly in theareas of multivariate analysis and for excellencein teaching and dissemination of statistical knowl-edge.
Colin O. Wu, Office of Biostatistics Research,Division of Cardiovascular Sciences, NationalHeart, Lung & Blood Institute, Bethesda, Mary-land For innovative and fundamental research inmathematical statistics, especially in methods forlongitudinal data analysis; for extensive contribu-tions to innovative design and analysis of clinicalstudies in cardiovascular disease and hematologi-cal disorders and broad impact on aplastic anemiapatient care and ethics of clinical trial execution.
Kelly Hong Zou, Pfizer Inc., New York, New YorkFor outstanding contributions to receiver-operatingcharacteristic methodology, particularly in the ar-eas of nonparametric and parametric transforma-tion classification methods; for key contributions todiagnostic medicine and medical imaging research;and for innovative designs
68 © www.icsa.org
July 2012 Vol.24/2 People News
2012 IMS Fellows
David Banks, Duke University For contributionsto bootstrap analysis, network analysis and adver-sarial risk analysis, as well as important contribu-tions to applications. For extraordinary service tothe profession including a term as editor of JASA.
Daniel Francis Heitjan, the University of Penn-sylvania For significant contributions to the the-ory and methodology of inference from incompletedata; for outstanding applications in cancer, cardio-vascular medicine, health economics, and smoking
cessation research; for distinguished editorial ser-vice.
Samuel Kou, Harvard University For influentialand pioneering contributions to stochastic model-ing and statistical inference in biophysics, and toMonte Carlo, Bayesian and nonparametric meth-ods.
Tapabrata Maiti, Michigan State University Forsignificant research contributions in small area in-ference, inference for mixed models, and Bayesianmethodology for panel-count data.
People NewsBoard of Regents Announces Karl E.Peace Recipient of Hall of Fame AlumniAward
The Jiann-Ping Hsu Collegeof Public Health of GeorgiaSouthern University (GSU)biostatistics professor andGeorgia Cancer Coalition Dis-tinguished Cancer Scholar,Karl Peace, Ph.D., was pre-sented with the 2012 Univer-sity System Board of Regents’Hall of Fame Award on March31, 2012 in Atlanta. Peace was one of only three re-cipients to receive the honor.
The award was established by the Board of Re-gents to honor those who exemplify superb leader-ship and support of higher education in the stateof Georgia. Recipients are nominated by their almamater and are selected by an external panel basedon their outstanding accomplishments and contri-butions to their institution.
ASA Announces the Karl E. Peace Award
The Karl E. Peace Award for Outstanding Statis-tical Contributions for the Betterment of Society,established in 2012, recognizes statisticians whohave made substantial contributions to the statisti-cal profession and to society in general. The awardwas established by Christopher K. Peace, son of
Karl E. Peace, on behalf of the Peace family to honorthe life work of his father.
Jeff Wu to Deliver the Deming Lecture atJSM 2012
Professor Jeff Wu from Geor-gia Tech has been selected todeliver the Deming Lecture atthe JSM on Tuesday, Jul 31,4:00 p.m. on “Quality Evo-lution and Revolution: FromAutos and Chips to Nano andBio”.
Bin Yu Elected IMSPresident-Elect
Bin Yu, Chancellor’s Professorat the Department of Statisticsand the Department of Electri-cal Engineering & ComputerScience, University of Califor-nia, Berkeley, has been electedPresident-Elect of IMS (Inst ofMathematical Statistics). TheIMS is a leading internationalprofessional and scholarly so-ciety devoted to the development, dissemination,and application of statistics and probability. TheInstitute currently has about 4,500 members in allparts of the world.
© www.icsa.org 69
Looking Back July 2012 Vol.24/2
Editorial: The first ICSA Pao-Lu Hsu Award will be announced in August, 2012. We have organized two articlesin memory of Prof. Hsu: one from a personal perspective by Prof. Dihe Hu, a former student and colleague ofProf. Hsu, now Professor Emeritus of Mathematics, Wuhan University; the other from an academic perspectiveby Jiading Chen and Zhongguo Zheng, Professors Emeritus of Statistics from Prof. Hsu’s home department atPeking University. The first article was based on a Chinese version written in 2010 at the occasion of the 100thbirthday of Prof. Pao-Lu Hsu. Knowing that the ICSA Bulletin is interested in an memorial article on Prof. Hsuin coordination with the ICSA Pao-Lu Hsu Award, Prof. Hu happily agreed to rewrite a shorter version in English.The full Chinese version is available at the webpage of the Department of Probability and Statistics, School ofMathematical Sciences, Peking University, http://www.math.pku.edu.cn/teachers/Hsu/Hudihe-Hsu-100.pdf. More memorial articles on Prof. Hsu can be found at http://www.math.pku.edu.cn/teachers/Hsu/
articles.htm. We are very gratefully to Prof. Dayue Chen, Chairman of the home department of Prof. Hsu atPeking University, for his tremendous help in soliciting the two articles and proofreading them.
In Memory of My Mentor Pao-Lu HsuDihe Hu
In this article I intend to introduce some life sto-ries and academic contributions of Professor Pao-Lu Hsu who was worldwide famous as one ofthe founders of mathematical statistics and the fa-ther of probability theory and statistics in China, toshow my deep respect to him.
Apt. #8 of Courtyard #3 in TongHouse
If you enter Peking University from the west en-trance and go straight to pass a stone arch bridge,you can see a square lawn with an ornamental col-umn on each side. Just opposite to the lawn isthe Administration Building of Peking University,which is a traditional Chinese building from theearly 20th century with beautiful craft works andChinese style paintings. Going southeast from theAdministration Building for about 500 meters, youwould see a group of one story houses, collectivelyknown as the Tong House, with big willow trees,bamboos and lilac trees scattered around. Some ofthese houses were courtyards, consisting of a cen-tral court and rooms around, which were quite typ-ical in old Beijing. During 1950–1970s, most of thedistinguished professors in Peking University livedin Yannan Garden (燕南园), Yandong Garden (燕东园), Jingchun Garden (镜春园), and so on. Com-pared to these residential areas, the Tong Housewas much less known. In fact Apt. #8 of Courtyard#3 in Tong House (佟府丙八号) was simple andhumble. Prof. Hsu lived here from the early 1950sto 1970 until he passed away. This apartment had
four rooms in two columns. On the right hand ofthe entrance there was a four-square-meter-kitchenand at the end of the lobby there was a small stor-age room. Ms. Zhang Jing-Zhao (张景昭), andher family lived in the two rooms on the right ofthe lobby. Prof. Hsu lived in the other two rela-tively smaller rooms on the left. The one closer tothe entrance served as his living room, which was14 square meters large, and the other one was hisbedroom. Fortunately this bedroom had its ownbathroom.
His living room was arranged in this way: onthe east wall hung a blackboard, along the northwall were two ceiling-high bookshelves, along eachof the west and the south walls laid an armchair,in the middle of the living room was a squarecoffee table, and the rest furniture were severalwooden stools and two thermos bottle with bam-boo shells. This living room was actually multi-purposed: when Prof. Hsu held seminars, it wasused as a classroom; when the faculty of probabil-ity and statistics gathered, it served as a meetingroom; when Prof. Hsu looked for math literatures,it was a library; when Prof. Hsu dined, it was a din-ing room; only when he had visitors or companies,it was used as what it was designed to be, livingroom.
Prof. Hsu had very simple meals each day:three bottles of milk, two or three dishes of meat orvegetable for lunch and supper. His meals were al-ways quite light, since he used to have pulmonaryand stomach trouble.
Ms. Zhang Jing-Zhao was in the Math Depart-ment in the National Southwestern Associated Uni-versity for her college education when Prof. Hsuwas a professor there. She was asked to take care of
70 © www.icsa.org
July 2012 Vol.24/2 Looking Back
the daily life of Prof. Hsu when they shared theapartment. She hired a housekeeper for the twofamilies and she sometimes came to dine with Prof.Hsu. Since there was no extra room, the house-keeper had to come every morning and leave everyevening.
Prof. Hsu seldom worked in his living room,he spent most of his working time in his bedroom.Leaning on his bed, he held a square board withscratch paper on it to write papers or lecture notes.Like many mathematicians, he often worked atnight. Every time he felt hungry in the middle ofhis work, he had a piece of chocolate. During theperiod of 1960–1962, even a piece of chocolate wasnot available.
Prof. Hsu had little leisure time. When he felttired, he took a short nap or listened to the radio.I even remember the brand of his radio, Panda,which was popular in 1950s. Listening to the radiowas almost the only way for him to know what wasgoing on in the world. And he also relied on theradio to enjoy Peking Opera or Kunqu Opera, oneof the oldest extant forms of Chinese opera. I heardthat Prof. Hsu was very knowledgeable on Chinesetraditional literature and arts. He could even per-form Kunqu Opera as good as a professional actorwhen he was young.
In the 1950s, Prof. Hsu usually took a breakin summer and winter vacations for one week ortwo. Considering his health condition, the univer-sity sometimes suggested that he should have hisvacation abroad, but he always insisted that oneweek in a hotel in the downtown Beijing wouldbe a wonderful treat for him. In his vacations, henormally went to listen to Peking Opera, KunquOpera, concerts or visit some of his relatives indowntown. But after middle of 1960s, Prof. Hsudid not have any vacations at all, he stayed homealmost all the time.
It would be a surprise for many people that sucha prestigious scholar, who had made great contri-butions to Chinese academic society, lived such ahumble life. Now the Tong House does not existanymore. If it had still been there, we would havehad opportunities to recall his life stories there. Theold time has gone with the wind, the only thing wecould do is to keep all our respect and memories inour mind.
Noble and Grace
Prof. Hsu was born into a scholarly family in thecity of Hangzhou. There were seven famous schol-
ars among his ancestors back in Qing Dynasty andthe family was once bestowed an inscribed boardwith “seven sons acknowledged in the royal court”(七子登科), by one of the emperors. Prof. Hsu ofcourse received authentic education in Chinese lit-erature and history when he was young. Later hemoved to Beijing to study in Huiwen High School(which used to be one of the most famous mission-ary schools in Beijing) and then had his college ed-ucation in Tsinghua University. At the age of 26he went to England to study at University CollegeLondon under the supervision of Prof. Neyman.The training and education he received during thisperiod had not only set a solid academic founda-tion for him but also remolded his mind with thethoughts of democracy and science.
It has always been well acknowledged in Chi-nese academic society that Prof. Hsu had made sig-nificant contribution to mathematics and statistics.In 1948, Prof. Hsu was among the first academi-cians of the Academia Sinica for his outstandingand pioneering work. Actually he was a scientistwith remarkable background and accomplishmentin other fields too, such as history, literature and soon. He always insisted that a scientist should talkand write not only with accuracy but also with im-peccable grammar and beautiful rhetoric. I learneda lot about how to write a mathematical paper withbeautiful Chinese when he helped me go throughmy manuscripts. He applied this principle as wellwhen he translated some math terminologies fromEnglish to Chinese. In 1963, he taught us generaltopology in the seminar. His lecture note was justan excellent example for us. In fact he only usedabout 30 thousands Chinese characters to presentthe main content of his course. Even in daily lan-guage he insisted accuracy. Once a relative of hissaid that he studied “arithmetic” in Tsinghua Uni-versity, he immediately pointed out that it shouldbe “mathematics”. Of course he was not so seriousall the time, he had a good sense of humor as well.
In the 1950s we had to follow the Soviet Unionin every aspect. And in each working unit therewere some Soviet experts to instruct the Chineseto do things in the right way according to whatthey did at home. At that time there was a womanprofessor in Mechanics from Leningrad University(now St. Petersburg) working in the Department ofMathematics and Mechanics at Peking University.Following their rules each year we had to maketeaching and research plans for the probability andstatistics group. Prof. Hsu was the director of thegroup and I was his secretary. When I did not knowhow to write the research plan, he told me how to
© www.icsa.org 71
Looking Back July 2012 Vol.24/2
get around this kind of nonsense: “The papers Ijust finished will go for the plan for next year andwe will certainly finish our plan successfully.” Likeother professors he did not always agree with thoseSoviet professors, but he never said anything meanor sharp against them.
From 1950s to 1960s Peking University was theleading research institute of probability and statis-tics in China. In the seminars held by Prof. Hsu,young people were encouraged to ask any ques-tions and even to challenge professors, which wasunusual for Chinese at that time. His personalityand the atmosphere in the seminar attracted manyyoung people from other institutes and universi-ties, which nurtured and promoted the communityof probability and statistics in China a lot. We werereally impressed by his broad knowledge. He wasnot only remembered for his academic accomplish-ments, but also for his character and his integrity.
Foresight and Sagacity
In 1930s Kolmogorov set the mathematical foun-dation for probability, so that it became a branchof mathematics and developed quickly. Almostat the same time in England, statistics, especiallybiometrics, medical statistics, agriculture statisticsand industrial statistics, made rapid progress andwere widely applied in practice. We all know itis very important in inferences to choose statisticsand their distributions and analyze the errors instatistics. During the first few years after he re-ceived his PhD, Prof. Hsu did a series of pioneer-ing work on statistical inferences and multivari-ate analysis. He introduced the theory on matrix,the theory on functions and the measure theory tostatistics, which enhanced the mathematical foun-dation for statistics and gradually led to the forma-tion of mainstream mathematical statistics. It canbe seen that Prof. Hsu was one of the founders ofmathematical statistics.
When he came back to Peking University in1947 from the United States, although the academicsociety in China was not so active and internationalexchanges were not so convenient either, Prof. Hsutried very hard to lead a group to the frontiersof probability and statistics and catch up with themainstreams. He realized that the classical limittheory in probability and the theory on stochasticprocesses were two active branches, therefore apartfrom the effort on statistics, he led two groups ofyoung faculty members to study the limit theoryof independent random variables and Markov pro-
cesses respectively. Prof. Hsu himself had alsodone important work on the limit theory. Unfor-tunately, some of the work remained unpublishedbecause of poor communications with the latestprogress, and mathematicians in other countriespublished similar works first.
In 1956 we read and discussed “Limit distribu-tion of the sum of independent random variables”by Gnedenko and Kolmogorov under the supervi-sion of Prof. Hsu. He thought that we should payattention to Donsker’s work on invariance princi-ples and Prokhorov’s work on advanced limit the-ory, and that there had to be interesting problemson those topics. At the same time, he tried to ar-range Prokhorov to visit Peking University to in-troduce his recent work.
In order to promote research on stationary pro-cesses, Prof. Hsu first asked the department tosend Jiang Ze-Pei (江泽培) to visit Moscow StateUniversity for three years. Having done beautifulworks on stationary processes during his visit tothe Soviet Union, Prof. Jiang came back to led aseminar on stationary processes, and then super-vised students to work on time series, just as Prof.Hsu planed long time before.
Prof. Hsu came to realize very early that the-ories on Markov processes would be one of themost active topics in probability. At that time, po-tential theory in probability and semigroup the-ory of the operators for Markov processes were atthe frontier of probability theory. He gave sequen-tial lectures on those topics and chose related pa-pers for us to read. He brought us to the fron-tiers of these areas in one or two years. Apart fromsending young faculty members abroad, he also in-vited Prof. Dynkin to Peking University to givelectures. Unfortunately when Dynkin visited us in1958, China was in the “Great Leap Forward”, andwe hardly had enough time to make benefit fromhis lectures because of those endless political meet-ings. Although not every one of his plans was car-ried out smoothly, the development of probabilityand statistics in China had made great progressesbecause of Prof. Hsu’s effort and foresight.
In 1956 a 12-year master plan for the develop-ment of sciences was worked out and probabil-ity and statistics was one of the three branches inmathematics which were given priority to supportand develop. Of course Prof. Hsu was in the lead-ing position to promote the research and the educa-tion of probability and statistics in China. He tookthe following five actions:
1. Training qualified teachers for universities.He first chose 54 senior students from
72 © www.icsa.org
July 2012 Vol.24/2 Looking Back
Peking University, Sun Yat-sen Universityand Nankai University to form a special class,joined by some instructors from universitieswhich were in need of faculty in probabil-ity and statistics. In the mean time, he in-vited Profs. Wang Shou-Ren (王寿仁) andZhang Li-Qian (张里千) from the Instituteof Mathematics of the Chinese Academy ofSciences, Profs. Zheng Zeng-Tong (郑曾同)and Liang Zhi-Shun (梁之舜) from Sun Yat-sen University to give lectures systematicallyon Measure Theory, Limit Theory in Probabil-ity, Markov Chains, Mathematical Statisticsfor that special class. I was a member of thatclass.
2. Making outlines for curriculums and writingtextbooks. Prof. Hsu set the core coursesfor students majored in probability and statis-tics: Measure Theory, Limit Theory in Proba-bility, Stochastic Processes, and MathematicalStatistics. In addition, students could chooseone or two courses according to their researchinterests from the followings: Markov Pro-cesses, Stationary Processes, Game Theory,Queuing Theory, Sampling Theory, StatisticalExperimental Design and so on.Before 1956 the Department of Mathematicsand Mechanics at Peking University only of-fered a basic course in probability theory, andthe textbook was “Probability” by the Sovietmathematician Gnedenko. Prof. Hsu thoughtthat we should write some textbooks basedon the materials used in the seminar, and thatsome classical textbooks in the west should betranslated into Chinese. Under his supervi-sion “An Introduction to Probability Theoryand Its Applications” by Feller was first trans-lated. Unfortunately, before his plan had beenfully carried out, the Great Cultural Revolu-tion began and some of the textbooks in theplan never appeared.
3. Inviting foreign scholars to give lectures.Since at that time we could not communicatewith mathematicians in the West, Prof. Hsumade a list of probabilists and statisticians inthe East Europe, including Eugene Dynkinand Yuri Vasilevich Prokhorov from the So-viet Union, and Marek Fisz and KazimierzUrbanik from Poland. In 1957 Prof. Fiszvisited us and gave lectures on multivariateanalysis and sampling theory, and Prof. Ur-banik introduced generalized stochastic pro-cesses to us. As mentioned before, Prof.Dynkin visited Peking University in 1958 and
talked on several active topics on stochasticprocesses.
4. Organizing seminars. The faculty of Prob-ability and Statistics at Peking Universitywere divided into three groups: mathemati-cal statistics, Markov processes, and station-ary processes. The first two groups were un-der the supervision of Prof. Hsu himself, andthe last one was mainly led by Prof. Jiang Ze-Pei. Prof. Hsu also held a seminar on Walddecomposition and sequential analysis whichwere related to stationary processes.
5. Promoting setting up a journal in probabil-ity and statistics. In 1950s, there were only afew mathematical journals in China. So it wasquite difficult for young researchers to pub-lish their papers. Prof. Hsu strongly insistedthat we should have a journal in probabilityand statistics and he even planed to fund thejournal with his own deposit. But the controlfor publication was very strict at that time.One could hardly have the approval evenfor an academic journal. By the time we fi-nally started the Journal of Applied Probabil-ity and Statistics (http://aps.ecnu.edu.cn/EN/volumn/current.shtml), Prof. Hsu hadleft us for more than ten years.
Everlasting Memory
Prof. Hsu’s charisma and integrity deeply affectedme throughout my life and his mentoring on math-ematics was greatly beneficial to me.
One day in 1958 Prof. Hsu suggested to me thatI should translate “An Introduction to ProbabilityTheory and Its Applications” by Feller. I was notquite sure whether I could handle the task, for I my-self at that time had not finished reading the bookand I only learned English for three years in highschool (Russian was the only foreign language wecould learn in college). Nevertheless, Prof. Hsupromised me that he would always be there if Ineeded any help in my translation. Finally, I signeda contract with the Academic Press of China. Af-ter I finished the translation, he checked my draftfrom the beginning to the end and wrote commentswhile he read it. When he noticed that I alwaystranslated “now” into Chinese as in “nowadays”or “right now”, he told me that “now” in Englishhad other meanings and I should decide which onewas the most suitable by checking the context care-fully. Another detailed comment was that I some-times used a full stop when I should use a comma.
© www.icsa.org 73
Looking Back July 2012 Vol.24/2
Of course encouragement was the main tone. Onecomment in Chapter Nine of the book said that mytranslation for “a random variable” was very good:“accurate in mathematics and elegant in Chinese”.
In 1959 the Division of Probability and Statis-tics planned to offer a course on the limit theory ofprobability to students majored in probability. Prof.Hsu wanted me to teach the course. He suggestedthat I should first write the outline of the coursebased on the notes of our seminar on limit theoryof independent random variables and “ProbabilityTheory” by Loéve. So I did, and he discussed theoutline with me. Every young faculty member inthe division found him always fully supportive.
In summer 1960, the Ministry of Higher Edu-cation organized a group of specialists in variousfields to promote the progress of these fields in uni-versities nationwide. Probability and statistics wasamong those with high priority, so our divisionwas required to send two instructors to that group.Prof. Jiang Ze-Pei was sent to Zhengzhou to lectureon stationary processes and I to Guiyang to teachprobability theory to college teachers. By 1960 I hadonly three years teaching experience and had neverfaced students with different backgrounds and in-terests before. Again Prof. Hsu told me how tohandle the level of the course and how to explainconcepts and results clearly and straightly to thosewho were not specialists in this field.
Above all the help from Prof. Hsu, his mentor-ship on research benefited me most. He himselftaught Markov chains with countable state spacesin the seminar and directed me to read “MarkovChains with stationary transition probabilities” byK-L Chung. He brought my attention to the pa-per “An invariance principle for certain probabil-ity limit theorems” by Donsker which got me inter-ested in the limit theory of stochastic processes anda few of my early papers were on this topic.
Since he left London for China, Prof. Hsu hadworked for several decades at Peking Universityand devoted himself to the development of prob-
ability and statistics in China. So many young re-searchers and students benefited from his help andmentoring. It was in one afternoon in the late au-tumn of 1969 that I met Prof. Hsu for the last time.It was just in the middle of that Great Cultural Rev-olution. I saw him sitting on the ground in the cam-pus. When I went up to say hello, I came to knowthat there was a public political meeting to be heldin Building 18 to criticize and denounce some wellestablished professors and he was ordered to re-ceive “re-education” there. Since he was too weak,he had to take a break on the way to that meeting.I left him speechless. A few days later I went to alabor camp in south China, along with most mem-bers of the faculty in Peking University, to receivemy “re-education”. I stayed in that labor camp inJiangxi Province without knowing anything hap-pening in the university until one day some newcomers told me that Prof. Hsu had passed away onDec. 18th, 1970. Near his death bed was a Parkerpen he had used for decades, and piles of some un-finished drafts. He worked till his last breath andnobody ever knew how much he suffered.
My memory of Prof. Hsu will be with me for-ever.
Acknowledgements
My daughter, Hu Xiao-Yu, who is also a probabilist,has made contributions to prepare the English ver-sion of this article. I am also grateful to the carefultyping of this work by Mr. Zhang Zhi-Yang.
Dihe HuProfessor Emeritus of MathematicsSchool of Mathematics and Statis-ticsWuhan University
Academic Achievements of ProfessorPao-Lu HsuJiading Chen and Zhongguo Zheng
In 2010, we celebrated the 100th birthday of Profes-sor Pao-Lu Hsu, a world-class statistician and the
founder of probability and statistics in China. Hewas a member of the Chinese Academy of Sciences,and a Professor of Rank One at Peking University.With reverence we recall his contributions to prob-
74 © www.icsa.org
July 2012 Vol.24/2 Looking Back
ability and statistics in China.Professor Hsu was born on September 1, 1910,
in Beijing, though his forefathers were natives ofHangzhou, Zhejiang Province. He was from aprominent intellectual family. In his childhood,he received solid training in both traditional Chi-nese and modern western cultures (Editorial boardfor “Pao-Lu Hsu Memorial Collection”, 2010). Hegraduated from Tsinghua University in 1933, ma-joring in mathematics. After his graduation, heworked at Peking University as a teacher. In themeantime, he published a joint paper with Tsai-hanKiang (Jiang Zehan) on the numbers of nondegen-erate critical points, which showed his solid mathe-matical foundation and research capability. In 1936,he went to University College London, and spentfour years studying mathematical statistics. Dur-ing this period, with his strong mathematical skillscombined with advanced statistical ideas, he wrotea series of remarkable papers. He earned his Ph.D.in 1938 and Sc.D. in 1940. From London, he re-turned to China accepting a professorship in theDepartment of Mathematics, Peking University. In1945, he went to the United States, visiting the Uni-versity of California at Berkeley, Columbia Univer-sity and the University of North Carolina at ChapelHill. In 1947, he returned to Beijing and thereafterhe was engaged in teaching mathematics at PekingUniversity for more than 20 years. On December18, 1970, he died in his home on the Peking Univer-sity campus.
Professor Hsu set the highest standard for hisresearch work. “The value of a paper is not deter-mined by its publication. Instead, it is validatedwhen it is cited frequently by others later on”; “Idon’t like to get famous because my papers arepublished on the journals with good reputation, Iprefer that a journal builds its reputation becauseof my papers” (Editorial board for “Pao-Lu HsuMemorial Collection”, 2010). Owing to his highstandard, his published papers have profound im-pact in statistics.
Professor Hsu’s main research areas were math-ematical statistics and probability theory. Besides,his works on matrix theory and integral transfor-mation were excellent. He was the first Chineseinternationally recognized in the area of probabil-ity and statistics. In 1979, in honor of Hsu’s 70thbirthday, Anderson et al. (1979) wrote on his lifeand work in the Annals of Statistics. (They had Hsu’sbirth year as 1909, which should be 1910.) Each ofthem reviewed Hsu’s research in detail from dif-ferent aspects: E. L. Lehmann (who was a mem-ber of the National Academy of Science) reviewed
on statistical inference (Lehmann, 1979), T. W. An-derson in multivariate analysis (Anderson, 1979),and K. L. Chung in probability (Chung, 1979). In1981, the book entitled “Hsu Pao-Lu Collected Pa-pers” (in Chinese, Hsu, 1981) was published by theScience Press of China, with a preface by ProfessorTsai-han Kiang and Professor Hsio-Fu Tuan (bothwere members of the Chinese Academy of Science),who highly praised Professor Hsu for his contri-butions to the development of the field of proba-bility and statistics in China. Chung (1983) editedthe collection “Pao-Lu Hsu Collected Papers”, pub-lished by Springer-Verlag in 1983, including al-most all Hsu’s papers, with those in Chinese trans-lated into English. In Johnson and Kotz (1997), anedited volume entitled “Leading Personalities inStatistical Science from the Seventeenth Century tothe Present”, Professor Hsu was the only Chinesestatistician among the 114 people who had greatinfluence in the development of statistical sciencesince early 17th century, along with I. Newton, C. F.Gauss, P. S. Laplace, R. A. Fisher, J. Neyman, A. N.Kolmogorov, and others.
We introduce Hsu’s academic achievementsfrom 10 aspects.
(1) The Behrens–Fisher Problem The first paperof Hsu in statistics was on the Behrens–Fisher prob-lem (Hsu, 1938a). Let X1, X2, ..., Xn and Y1, Y2, ..., Ymbe samples from N(µ1, σ2
1 ) and N(µ2, σ22 ) respec-
tively, where σ1 and σ2 are unknown. The problemis to test the null hypothesis H0 : µ1 = µ2. (Whenσ1 = σ2, it is solved by the t-test. The challengingpoint here is that σ1 and σ2 are not necessary equal).Hsu considered the class of statistics
U = (X− Y)2/(A1S21 + A2S2
2),
where
X =1n
n
∑i=1
Xi, Y =1m
m
∑i=1
Yi,
S21 =
n
∑i=1
(Xi − X)2, S22 =
m
∑i=1
(Yi − Y)2,
and A1 and A2 are two constants. When A1 =A2 = (m + n)/[(n + m− 2)nm] , U is just the Stu-dent’s t-statistic u1, and when A1 = 1/(n(n− 1))and A2 = 1/(m(m − 1)), U is the Behrens–Fisherstatistic u2. Hsu found the series expansion forthe density of U and utilized it to study the powerfunction of the rejection area of {U > C} for someC > 0. He showed that the power function re-lies only on the parameters θ = σ2
1 /σ22 and λ =
© www.icsa.org 75
Looking Back July 2012 Vol.24/2
(µ1− µ2)2/( 1
n σ21 +
1m σ2
2 ). This is an exact instead ofasymptotic result, which Scheffé (1970) describedas “a model of mathematical rigor”. Hsu’s mainconclusion, obtained by a combination of his ana-lytical study and some numerical work, was that,for λ = 0 and varying θ, neither u1 nor u2 con-trol the rejection probability (except when m = n).But the statistic u2 is less sensitive to variation ofθ. Today, the most commonly used method to solvethis test problem is still to utilize the rejection area{U > C}. Owing to Hsu’s work, in the Behrens–Fisher problem, the method utilizing the rejectionarea {U > C} is called “Hsu’s Method”.
(2) Optimal estimator of variance Hsu (1938c)dealt with the optimal estimator of variance in theGauss-Markov model. Let
y = Aβ + ε
be a linear model, where y = (y1, ..., yn)T (here-inafter MT denotes the transpose of matrix M),β = (β1, ..., βp)T , A is a known n× p matrix (rankp), ε = (ε1, ..., εn)T with independent componentssatisfying
Eεi = 0, Eε2i = σ2, Eε4
i = αiσ4 (i = 1, ..., n),
where σ > 0 is parameter and αi > 0, i =1, . . . , n, are constants independent of σ. Let Q =Q(y1, y2, ..., yn) be a quadratic form of y1, y2, ..., yn.Hsu considered the class of quadratic form Qwhich satisfies the following condition (i) unbiased,i.e. EQ = σ2 for all β and σ2; and (ii) the varianceof Q is independent of β. A quadratic form Q inthe class is said to be an optimal quadratic estima-tor of σ2, if the variance of Q reaches the minimumwithin the class of quadratic forms.
Let Q = yTΛy, where Λ is a n × n symmet-ric matrix. Hsu obtained a necessary and sufficientcondition for Q to be an optimal quadratic estima-tor of σ2. Let M = I− A(AT A)−1 AT , where I is theunit matrix. Let τ1, τ2, ..., τn be the values at whichthe quadratic function
F = ∑i,j
µijτiτj
takes the minimum value under the constraint∑n
i=1 miiτi = 1, where
µij =n
∑k=1
(αk − 3)m2kim
2kj + 2m2
ij
(i = 1, ..., n; j = 1, .., n), mij are elements of matrixM = (mij)n×n, αi = σ−4Eε4
i (i = 1, ..., n). A nec-essary and sufficient condition is that the matrix Λ
has the following form
Λ = MDτ M,
where
Dτ =
τ1 0 · · · 00 τ2 · · · 0...
.... . .
...0 0 · · · τn
.
Let
S20 =
1n− p
‖ y− Aβ ‖2
where β is the least square estimate of β. Hsushowed that a necessary and sufficient conditionfor S2
0 to be an optimal quadratic estimate of σ2 is
(n− p)n
∑i=1
(αi − 3)miim2ik = mkk
n
∑i=1
(αi − 3)m2ii.
This part of Hsu’s work is regarded as the ori-gin of the large literature on the optimal quadraticestimators of variance and variance components.
(3) Small Sample Inference For the small sam-ple inference, Hsu was concerned with testing uni-variate and multivariate linear hypotheses, andparticularly, with power properties of the tests ofthese hypotheses. Linear hypotheses are hypothe-ses about linear relations between the parametersof the model. Hsu (1938b) obtained the power func-tion of Hotelling’s T2-test and pointed out that,under nonnull hypothesis the distribution of T2
is the distribution of the ratio of a noncentral χ2
to an independent central χ2 variable. He alsoshowed that the test is in certain sense locally mostpowerful. Hsu formulated the general multivari-ate linear hypothesis in its canonical form. He ob-tained the nonnull distribution of the likelihood ra-tio statistic when the covariance matrix is known.For the case of unknown covariance matrix, let θibe the nonzero roots of the associated determinan-tal equation. Hsu (1940) considered test statisticsW = ∏(1− θi) and V = ∑ θi/(1− θi) and pointedout that their asymptotic power are the same whenthe sample size tends to infinity.
In small sample inference, probably the mostimportant result of Hsu’s work is the finding andproving first optimum property for the likelihoodratio test of the linear hypothesis (Hsu, 1941). With-out loss of generality, we present the result in thecanonical form.
76 © www.icsa.org
July 2012 Vol.24/2 Looking Back
Let the joint density of Y1, Y2, ..., Ym, Z1, ..., Zn be
p(y1, ..., ym, z1, ..., zn)
=(√
2πσ)−(m+n) exp{− 12σ2 [
m
∑i=1
(yi − ηi)2 +
n
∑i=1
z2i ]},
where η1, ..., ηm are m arbitrary real numbers, σ isan arbitrary positive number, and all numbers areunknown. The problem is to test the hypothesis
H0 : η1 = η2 = ... = ηn1 = 0,
where n1 is a positive integer less or equal to m. Let
F =∑n1
i=1 y2i
∑n1i=1 y2
i + ∑ni=1 z2
i,
W0 = {(y1, ..., yn1 , z1, ..., zn) : F ≥ Fα},
where Fα is the 1− α quantile of F under the nullhypothesis, i.e. Fα is determined by the equationP(F ≥ Fα|H0) = α, where F = ∑n1
i=1 Y2i /(∑n1
i=1 Y2i +
∑ni=1 Z2
i ).When W0 is used as the rejection area in the test
problem, it is shown that the power function of thetest has the form of β0(λ), where
λ =1
2σ2
n1
∑i=1
η2i .
Hsu obtained the following result. Supposethat the rejection area W is the set of points(y1, ..., ym, z1, ..., zn) satisfying the conditions: thelevel of W is α, and the power function of W relieson the parameters only through the parameter λ,i.e. the power function has the form of β(λ). Then,β(λ) ≤ β0(λ) for all λ > 0.
In other words, for the problem of testing thenull hypothesis H0, the rejection area W0 has themaximum power function within the class of rejec-tion areas with level α whose power function re-lies on the parameter λ only. This is the first resultabout the optimal property of F-test, and the as-sociated theorem was called “P. L. Hsu Theorem”(Mann, 1949). This work initiated two lines of de-velopments. On the one hand, Hsu’s work was ap-plied to the multivariate problem, Hotelling’s T2
and the multiple correlation coefficient (Simaika,1941). On the other hand, Hsu’s paper offereda new method for obtaining all the similar testsand was formulated by means of the concept ofcompleteness by Lehmann and Scheffé (Lehmann,1979).
(4) Multivariate Analysis From 1938 to 1945, Hsupublished several papers on the forefront of the de-velopment of the theory of multivariate analysisand obtained several exact or asymptotic distribu-tions of important statistics.
A crucial element of multivariate theory is thedistribution of the sample covariance matrix S. LetX1, X2, ..., XN be a sample from p dimensional nor-mal population N(0, Σ), then
A , (N − 1)S =N
∑α=1
(Xα − X)(Xα − X)T
has the so-called Wishart distribution W(Σ, N− 1).Hsu (1939a) derived the density function of theWishart distribution based on algebra and analy-sis by mathematical induction. Anderson (1979)praised that Hsu’s proof was the most elegantamong all proofs available.
Hsu (1939b) obtained the joint distribution ofroots of certain determinantal equation which is afundamental result in multivariate analysis. LetA and B be independent and from Wishart distri-bution W(Σ, m) and W(Σ, n) respectively, wherem ≥ p, n ≥ p and p is the size of the matrix Σ. Letθ1 ≥ θ2 ≥ ... ≥ θp be the roots of the determinantalequation
|A− θ(A + B)| = 0.
After complicated calculations, Hsu proved thatthe joint density of θ1, ..., θp equals to a constanttimes
p
∏i=1
θ12 (m−p−1)i
p
∏i=1
(1− θi)12 (n+p−1)
p
∏i=1
p
∏j=i+1
(θi − θj).
Now suppose that the matrices A and Σ are parti-tioned into
A =
(A11 A12A21 A22
), Σ =
(Σ11 Σ12Σ21 Σ22
),
where A11 and Σ11 are matrices with size p1 × p1,and A22 and Σ22 with size p2 × p2. Sample andpopulation canonical correlation are defined as theroots of the following equations respectively,∣∣∣∣ −λA11 A12
A21 −λA22
∣∣∣∣ = 0,∣∣∣∣ −λΣ11 Σ12
Σ21 −λΣ22
∣∣∣∣ = 0.
Hsu obtained the asymptotic distribution for thenormalized sample canonical correlation (Chung,1983, pp. 142–149).
Let A and B be independent random matrices,where A has a noncentral Wishart distribution andB has a central Wishart distribution. Hsu consid-ered the joint distribution of the roots of the follow-ing determinantal equation |A − φB| = 0. Under
© www.icsa.org 77
Looking Back July 2012 Vol.24/2
certain conditions, he obtained the asymptotic dis-tribution of the roots (Chung, 1983, pp. 129–140).
Hsu’s result about random matrix (for example,matrices A and B mentioned above) is one of thefirst results of random matrix in modern researchhistory. In the preface of the book “ Random Matri-ces”, the author Madan Lal Mehta took Hsu (1939b)as one of the two first important papers in the ran-dom matrix theory.
To test the independence of identically dis-tributed random variables X1, X2, ..., XN , Hsu con-sidered the statistic T = Q/S, where
Q =N
∑i,j=1
aij(Xi − X)(Xj − X), S =N
∑i=1
(Xi − X)2.
Let aij rely on the sample size N. Under certainconditions, Hsu obtained the asymptotic expansionfor distribution function of T as N tends to infinity(Chung, 1983, pp. 224–228).
Besides, Hsu investigated the asymptotic be-havior of function f (u1, . . . , uk) when the samplesize tends to infinity, where u1, . . . , uk are indepen-dent vector sample means, and f is a smooth multi-variate function. By applying the central limit the-orem to the means, and by the Taylor expansion off (·), Hsu obtained the result that the limit distribu-tion is normal or the distribution of weighted sumof squares of normal random variables. We shouldpoint out that Hsu’s method (by using central limittheorem and Taylor expansion) is a general methodwhich is well known today in the large sample re-search area and is called the ∆-method.
(5) Probability Theory and Applications Hsu’simportant results in the probability theory andtheir applications were derived by his capability ofproficient manipulation of characteristic functions.He was a true virtuoso in the method of charac-teristic function (Chung, 1979). His result in Hsu(1945) is a crucial improvement of Berry’s result.Suppose that ξ1, ξ2, ..., ξn are i.i.d. random variableswith mean zero and variance one. Let
ξ =1n
n
∑i=1
ξi, η =1n
n
∑i=1
(ξi − ξ)2
and denote by Φ(x) the standard normal distribu-tion function. An important problem in both the-ory and application is the convergence rate of thenormalized ξ and η towards the standard normaldistribution. Let
Fn(x) , P(√
nξ ≤ x).
Cramér showed that
Fn(x) = Φ(x) + ψ(x) + R(x),
where ψ(x) and R(x) rely on the distribution of ξ1and limn→∞ R(x) = 0. Berry obtained the formulathat for all x,
|Fn(x)−Φ(x)| ≤ Aβ3n−12 ,
where β3 = E|ξ1|3 and A is an absolute constantwhich relies neither on n nor on the distribution ofξ1.
Hsu (1945) extended Berry’s method to give asimpler proof of Cramér’s result on the asymptoticexpansion of Fn(x). Furthermore, instead of con-sidering the sample mean ξ = 1
n ∑ni=1 ξi, Hsu inves-
tigated the asymptotic property of sample varianceη = 1
n ∑ni=1(ξi − ξ)2. Let
Gn(x) = P(√
n(η − 1)√α4 − 1
≤ x)
,
where α4 = Eξ41. Under the condition that α6 =
Eξ61 < ∞ and α4 − 1 − α2
3 6= 0 (α3 = Eξ31), Hsu
proved that for all x
|Gn(x)−Φ(x)| ≤ A√n
( α6
α4 − 1− α23
) 32,
where A is an absolute constant. Hsu also obtainedthe asymptotic expansion of Gn(x) under the con-dition Eξ2k
1 < ∞ for some k > 3, and that the re-minder of the expansion has an upper bound.
By the way, influenced by Hsu (1945), Chenet al. (1985) extended Hsu’s result to estimate of σ2
in the linear model, and obtained a series of impor-tant results.
(6) Complete Convergence of Series of i.i.d. Ran-dom Variable Hsu and Robbins (1947) dealt withthe complete convergence of series of i.i.d. randomvariables, another important contribution to prob-ability theory. Let {ξn, n ≥ 1} be i.i.d. series withcommon mean µ and finite variance. They provedthat for arbitrary ε > 0,
∞
∑n=1
P(| 1n
n
∑k=1
ξk − µ| ≥ ε) < ∞.
This result strengthens the classical Strong Law ofLarge Numbers. When this property holds, therandom series 1
n ∑nk=1 ξk is called complete conver-
gence to the common mean µ of ξi. Hsu and Rob-bins (1947) further conjectured that the condition
78 © www.icsa.org
July 2012 Vol.24/2 Looking Back
of finiteness of the variance of ξ is also a necessarycondition for the complete convergence to hold.Two years later, the famed mathematician P. Erdösproved the conjecture.
(7) Central Limit Theorem Around 1940, a chal-lenging problem was to find a solution of the mostgeneral form of the Central Limit Theorem, whichdrew the attention of many famed mathematicians,such as Levy, Feller, Kolmogorov and Gnedenko.Hsu was a competitor and the competition showedthat he was also on the peak. Hsu (1968) was Pro-fessor Hsu’s manuscript which Hsu mailed to K.L. Chung in 1947. In this paper, Hsu indepen-dently obtained the necessary and sufficient condi-tion under which the row sums of a triangular ar-ray of infinitesimal random variables, independentin each row, converges in distribution to a given in-finitely divisible distribution. Despite the fact thatGnedenko obtained the same result in 1944, Hsu’smethod is direct and has its own trait. When K.L. Chung translated the book “Limit Theorems ofSums of Independent Random Variables” by B. V.Gnedenko and A. N. Kolmogorov in 1968, he de-cided to include Hsu’s paper in the book as Ap-pendix III.
(8) Characteristic Functions Hsu was an expertin manipulating characteristic functions. He usedcharacteristic functions as a tool to obtain distribu-tion of certain random variables, to calculate thepower function in a test problem, and to deter-mine the limit distribution of series of random vari-ables. He also obtained some important propertiesof characteristic functions.
Let F(x) be the distribution function of randomvariable X. The characteristic function of F(x) isdefined as f (t) =
∫ +∞−∞ eitxdF(x). Hsu (1951) ob-
tained the necessary and sufficient condition, interms of the property of its characteristic functionon certain interval (−δ,+δ) where δ > 0 is a smallnumber, for the finiteness of the βth absolute mo-ment of the corresponding distribution, i.e.
Mβ(F) =∫ +∞
−∞|x|βdF(x) < ∞.
Hsu (1954) dealt with the problem of identifying acharacteristic function, i.e. to find a condition un-der which the values of a characteristic functionon the interval (−∞, ∞) are determined by the val-ues of the characteristic function on a small interval(−δ, δ) with some δ > 0. It is related to the momentproblem, i.e., to find a condition under which the
distribution function is determined by its moments.Gnedenko found a counterexample that two differ-ent characteristic functions coincide in a small in-terval (−δ, δ), where δ is a positive number. Hsuclassified characteristic functions into two classes.For a characteristic function of the first class, it isfully determined by its values in the neighborhoodof zero. The second class is called by U. A char-acteristic function belongs to the class (U) if it isequal to another characteristic function in a neigh-borhood of zero without being equal to it identi-cally. Hsu gave three subclasses of the class (U).The first subclass is the simplest. However it in-cludes all the known counterexamples. The sec-ond subclass consists of those characteristic func-tions of some stable distributions. The third sub-class is composed of characteristic functions whosecorresponding distribution function F(x) has den-sity p(x) of the following form:
p(x) = O[exp(− |x|ψ(|x|) )], |x| → ∞;
where ψ(x) has one of the following forms:
(ln x)λ, (ln x)(ln ln x)λ, ... λ > 1.
(9) Matrix Theory Hsu was an expert in apply-ing matrices as a tool to solve mathematical prob-lems. He obtained several theorems in matrix the-ory. Hsu (1955a) studied the transform from squarematrix A to square matrix B, where all the elementsof the matrices are complex numbers, A → B =PA(P)−1, where matrix P is nonsingular squarematrix, and P is the conjugate matrix of P. If Aand B can be transformed to each other, then Aand B are called similar. Hsu obtained the canon-ical form under the sort of transformations for allmatrices and a necessary and sufficient conditionfor two matrices to be similar.
Hsu (1955b) studied a transform from a ma-trix pair (A1, A2) to another pair (B1, B2), where Aiand Bi are matrices of the same size, with complexnumber as their elements. If (A1, A2) can be trans-formed into (B1, B2) through the following equa-tions
A1 → B1 = PA1Q, A2 → B2 = PA2Q,
where P and Q are nonsingular square matrices,then the two pairs are called equivalent. Hsu foundthe canonical form for the equivalent pairs anda necessary and sufficient condition under whichtwo pairs are equivalent.
© www.icsa.org 79
Looking Back July 2012 Vol.24/2
In a long article (38 pages in the original Chineseversion and 54 pages in the English version), Hsu(1957) investigated profoundly the property of jointtransformation of Hermitian matrix and a symmet-ric (or skew symmetric) matrix. Let A1 be a Her-mitian matrix (i.e. AT
1 = A1), and A2 be a symmet-ric matrix (or skew symmetric). The transformationfor the pair (A1, A2) is
A1 → B1 = PA1(P)T , A2 → B2 = PA2PT ,
where P is some nonsingular matrix, and the trans-pose of P is denoted by PT . If the two pairs canbe transformed to each other through these trans-forms, then the two pairs are said to be congruent.By a careful derivation and complex calculation,Hsu obtained the following conclusions:
(i) Let A1 be a Hermitian matrix, A2 be a sym-metric matrix. The canonical forms of the pairswere found (there exist 7 different forms totally).A necessary and sufficient condition for two pairsto be congruent was obtained.
(ii) Let A1 be a Hermitian matrix, A2 be a skewsymmetric matrix. The canonical forms of the pairswere found (there exist 8 different forms totally). Anecessary and sufficient condition for two pairs tobe congruent was obtained.
These results in Hsu’s papers are important con-tributions to the matrix theory. The arguments inthe proof exhibited Hsu’s skills at manipulatingmatrices and his attention to details
(10) Markov Process Hsu (1958) investigated thedifferentiability of probability transition function ofa purely discontinuous homogeneous Markov pro-cess on the Euclidian space. Let X be the n dimen-sional Euclidian space, and F be the Borel σ-fieldin the space X and p(t, x, E) be the probability tran-sition function of a purely discontinuous homoge-neous Markov process on X, i.e., for x ∈ X, t > 0and E ∈ F , p(t, x, E) is the conditional probabilityof the event when the process is at state in the setE at time s + t under the condition that the processis at x at time s. Hsu proved the differentiabilityof probability transition function p(t, x, E) with re-spect to the variable t. Hsu also derived severalintegral equations for the differential of p(t, x, E)which was a generalization of the results of Austinfor the probability transition function on the dis-crete space. Hsu’s method was more elementarythan Austin’s and Hsu’s results were sharper.
In addition to the 10 aspects of his results, Pro-fessor Hsu led several seminars at Peking Univer-
sity in 1957–1966. Under his guidance, the partic-ipants obtained valuable results especially in thearea of experimental design and order statistics.Some results were published in the journals by penname “Ban Cheng”. Ban (1964b) deals with par-tial balanced incomplete block design (PBIB de-sign). For certain design parameters, Ban Chengobtained the condition of existence for the PBIB de-sign with m-associate classes and constructed thedesign. Ban (1964a) investigated the limit distri-bution of order statistics. Let X1, ..., Xn be i.i.d.random variables with common distribution func-tion F(x) and their order statistics be denoted byξ(n)1 ≤ ... ≤ ξ
(n)n . Ban (1964a) proved that under
certain conditions, the series of normalized statis-tics ξ
(n)kn
(kn → ∞, kn/n→ λ ∈ [0, 1)) has one of thefollowing distributions as the limit distribution:
φ1(x) =1√2π
∫ x
−∞e−
t22 dt,
φ2(x) =
0 x ≤ 01√2π
∫ α ln x+β−∞ e−
t22 dt x > 0, α > 0
and
φ3(x) =
1 x ≥ 01√2π
∫ α ln |x|+β−∞ e−
t22 dt x < 0, α > 0.
The conditions for the common distribution func-tion F(x) to be in the domain of attraction ofφi(x), i = 1, 2, 3, were also obtained respectively inBan (1964a).
Above we summarized Prof. Hsu’s major sci-entific achievements. Most of his papers were col-lected in Chung (1983), the edited volume of “Pao-Lu Hsu Collected Papers”, which also reflectedhis high standard in research quality and scientificspirit. This kind of spirit is especially important forscientists today.
Bibliography
Anderson, T. W. (1979). Hsu’s work in multivariateanalysis. Ann. Statist. 7, 474–478.
Anderson, T. W., Chung, K. L., and Lehmann, E. L.(1979). Pao-Lu Hsu 1909–1970. Ann. Statist. 7,467–470.
Ban, C. (1964a). The limit distribution of orderstatistics. Acta. Math. Sinica. 14, 694–714.
Ban, C. (1964b). Partial balanced incomplete blockdesign. Adv. Math. (China) 7, 240–281.
80 © www.icsa.org
July 2012 Vol.24/2 Looking Back
Chen, X., Chen, G., Wu, Q., and Zhao, L. (1985).Theory of Parameter Estimation in Linear Models.The Science Press of China, Beijing.
Chung, K. L. (1979). Hsu’s work in probability.Ann. Statist. 7, 479–483.
Chung, K. L. (editor) (1983). Pao-Lu Hsu CollectedPapers. Springer-Verlag, New York.
Editorial board for “Pao-Lu Hsu Memorial Collec-tion” (editor) (2010). Pao-Lu Hsu Memorial Collec-tion. Peking University Press, Beijing.
Hsu, P.-L. (1938a). Contribution to the theory of“Student’s” t-test as applied to the problem oftwo-samples. Statist. Res. Mem. 2, 1–24.
Hsu, P.-L. (1938b). Notes on Hotelling’s generalizedt. Ann. Math. Statist. 9, 231–243.
Hsu, P.-L. (1938c). On the best unbiased quadraticestimate of the variance. Statist. Res. Mem. 2, 91–104.
Hsu, P.-L. (1939a). A new proof of the joint productmoment distribution. Proc. Cambridge Philos. Soc.35, 336–338.
Hsu, P.-L. (1939b). On the distribution of roots ofcertain determinantal equations. Ann. Eugenics.9, 250–258.
Hsu, P.-L. (1940). On generalized analysis of vari-ance. Biometrika 31, 221–237.
Hsu, P.-L. (1941). Analysis of variance from thepower function standpoint. Biometrika 32, 62–69.
Hsu, P.-L. (1945). The approximate distribution ofthe mean and variance of a sample of indepen-dent variables. Ann. Math. Statist. 16, 1–29.
Hsu, P.-L. (1951). Absolute moments and character-istic function. J. Chinese Math. Soc. (New series) 1,257–280.
Hsu, P.-L. (1954). On characteristic functions whichcoincide in a neighborhood of zero. Acta. Math.Sinica. 4, 21–32.
Hsu, P.-L. (1955a). On a kind of transformations ofmatrices. Acta. Math. Sinica. 5, 333–346.
Hsu, P.-L. (1955b). On a kind of transformations ofmatrix pairs. Acta Sci. Natur. Univ. Pekinensis. 1,1–16.
Hsu, P.-L. (1957). Simultaneous transformation ofa Hermitian matrix and a symmetric or skew-symmetric matrix. Acta Sci. Natur. Univ. Pekinen-sis. 3, 167–209.
Hsu, P.-L. (1958). Differentiability of probabilitytransition function of a purely discontinuous sta-tionary Markov process on the Euclidian space.Acta Sci. Natur. Univ. Pekinensis. 4, 257–270.
Hsu, P.-L. (1968). A general weak limit theoremfor independent distributions. Appendix III in“Limit Theorems of Sums of Independent Ran-dom Variables” by B. V. Gnedenko and A. N. Kol-mogorov, translated by K. L. Chung. Revised edi-tion, Addison-Wesley.
Hsu, P.-L. (1981). Hsu Pao-Lu Collected Papers. TheScience Press of China, Beijing.
Hsu, P.-L. and Robbins, H. (1947). Complete con-vergence and the law of large numbers. Proc. Nat.Acad. Sci. U.S.A. 33, 25–31.
Johnson, N. L. and Kotz, S. (editors) (1997). Lead-ing Personalities in Statistical Science from the Sev-enteenth Century to the Present. Wiley and Sons.
Lehmann, E. L. (1979). Hsu’s work on inference.Ann. Statist. 7, 471–472.
Mann, H. B. (1949). Analysis and Design of Experi-ments. Dover, New York.
Scheffé, H. (1970). Practical solutions of theBehrens–Fisher problem. J. Amer. Statist. Assoc.65, 1501–1508.
Simaika, J. B. (1941). On an optimum property oftwo important statistical tests. Biometrika 32, 70–80.
Jiading ChenProfessor Emeritus of StatisticsPeking [email protected]
Zhongguo ZhengProfessor Emeritus of StatisticsPeking [email protected]
© www.icsa.org 81
Statisticians at Work July 2012 Vol.24/2
Statisticians Supporting Late-stageClinical Development at MerckJerald S. Schindler and Yang Song
Overview of Late DevelopmentStatistics at Merck
At Merck, the Biostatistics and Research Deci-sion Sciences (BARDS) organization has a proudrecord of providing comprehensive analytical andmethodological expertise to support Merck’s strate-gic goals. BARDS scientists are skilled in a rangeof technical and scientific disciplines including re-search design methodology, statistical theory andmethods, mathematics, computer science, epidemi-ology, outcomes research and health economicmodeling. The organization is comprised of var-ious functional areas which collectively are orga-nized to provide support across all divisions ofMerck spanning product discovery through man-ufacturing and marketing.
• Late Development Statistics (LDS) providesstatistical support for late-stage clinical de-velopment, generally starting from Phase II.In certain therapeutic areas such as oncologyand vaccines, LDS also supports Phase I de-velopment.
• Early Development Statistics provides statis-tical support for early-stage clinical develop-ment, and support of data analysis and ex-perimental design for pre-clinical and non-clinical studies.
• Epidemiology supports the scientific study ofdisease, endpoints and patient outcomes toevaluate the benefits and risks of Merck prod-ucts.
• Health Economic Statistics supports the de-velopment and application of quantitativescientific methods to demonstrate the eco-nomic value of Merck products to customers.
• Scientific Programming provides scientificprogramming to biostatisticians, epidemi-ologists, and health economists across theBARDS organization.
The LDS department within the BARDS orga-nization consists of over 100 statisticians globally.Administratively, LDS is organized into 7 researchgroups, 5 in the US, 1 in Europe and 1 in theAsia-Pacific region, which is the newly established
group based in Beijing. To better support clinicaldevelopment projects, LDS statisticians are also or-ganized by therapeutic areas (or franchises), which,to some extent in the US, align with the researchgroups.
Statisticians in LDS provide statistical supportfor the full life cycle of a clinical study. It startswith study planning, a key part of which is protocoldevelopment. The statistician will collaborate veryclosely with the clinical team to understand the ob-jectives of the study and provide input on the studydesign, including sample size. The statistician willwrite certain sections of the protocol, especially thesection on statistical planning and analysis. Cer-tain studies, especially those pivotal Phase III stud-ies intended for regulatory submission, will needmore activities in the planning, including discus-sion with regulatory agencies, scientific input fromexperts or key opinion leaders, and the setup ofoversight committees for the study, e.g. data mon-itoring committee and endpoint adjudication com-mittee. Merck LDS statisticians have a long tradi-tion of providing very strong support and leader-ship in these activities. After study planning, thereare a few activities that need statistical support forstudy initiation. The statistician will support thepreparation of randomization schedules for patientenrollment and the setup of IVRS (Interactive VoiceResponse System) to facilitate enrollment if appli-cable. The statistician will also support the devel-opment of CRF (case report form) to ensure the datapoints that are needed for statistical analysis can becompletely and accurately collected. Sometimes,statisticians also participate in investigator meet-ings to help orient the physicians and study coor-dinators who participate in the study on the studydesign and objectives and train them on the con-duct of the study. When a study is ongoing, statis-ticians would support medical monitoring of thestudy safety and any interim analyses of the studyif applicable. At the end of the study, statisticianswill provide complete analysis of the study and ap-propriate statistical interpretation of the analysis.In collaboration of the clinical team, LDS statisti-cians will further compile the analysis results andwrite the clinical study report.
In addition to providing statistical supportfor clinical studies, Merck LDS statisticians takepride in providing support of clinical develop-
82 © www.icsa.org
July 2012 Vol.24/2 Statisticians at Work
ment strategies. With our quantitative expertise,we work with clinical and regulatory colleagues todevise the development strategy for clinical pro-grams. Often times, we are able to evaluate differ-ent development strategies quantitatively and rec-ommend the optimal one. In fact, the LDS statisti-cian in the team is a key contributor of the clinicaldevelopment plan.
The statistical support LDS statisticians provideis also important to the success of a regulatory sub-mission. Working closely with regulatory and clin-ical colleagues, we help develop the submissionstrategy, provide timely and quality response toquestions from regulatory agencies, and engage indiscussions with the agencies regarding the benefitand risk of the product. For some submissions tothe US FDA, a public advisory committee meetingis required to get independent expert advice on thebenefit risk of the product. LDS statisticians playa vital part in the preparation of such meetings. Alot of expertise, experience and efforts are neededfrom the statisticians to help understand and inter-pret the clinical data from different angles.
Within the LDS department and more gener-ally within the BARDS organization, there are a lotof interactions and collaborations among statisti-cians on various statistical and scientific topics. Tostandardize the approach to important, commonly-encountered statistical technical issues across thelate and early development statistics departments,a committee called ELSTIC (Early / Late-stageStatistics Technical Issues Committee) coordinatesthe development, review and approval of guidancedocuments. There are various working groups un-der the oversight of the ELSTIC on a broad range oftopics, e.g. longitudinal data analysis, multiplicity,time-to-event analysis, safety analysis, and adap-tive designs. There are also other working groupsto work on some special topics such as decisionanalysis and Bayesian methods and practice. Thereare regular seminar series where statisticians canvolunteer to present interesting topics and sharetheir experience in projects. To promote sharing ofexperience on clinical development projects, LDS
statisticians conduct peer review on important de-liverables of the projects. For example, TDRC (TrialDesign Review Committee) provides input on clin-ical study design when the protocol is still at con-cept stage.
Merck statisticians, in LDS and in the broaderBARDS organization, have a reputation of beingan industry leader in statistical innovation. Merckstatisticians are well known for their scientific ex-cellence in adaptive designs, Bayesian methods,and other areas, e.g. sample size estimation andvaccine clinical trial designs, with widely quotedpublications. Our statisticians are encouraged todo statistical research, especially that originatingfrom clinical development projects and that addingvalue to Merck’s R&D pipeline. Every year at theJoint Statistical Meetings, statisticians from Merckmake a large number of presentations. In fact,Merck is one of the pharmaceutical companies, ifnot the pharmaceutical company, that have thelargest number of presentations. Within Merck,LDS statisticians also take leadership on innova-tion to improve efficiency in drug development atMerck. For example, LDS statisticians take leader-ship in a cross-functional initiative to increase theuse of adaptive design clinical trials in clinical de-velopment strategy and develop an infrastructureto support the design and conduct of adaptive de-signed clinical trials.
Experience in Oncology Drug De-velopment — Yang Song
Jerry has provided an overview of the LDS depart-ment. As an LDS statistician that has been work-ing on oncology projects, I would like to share myexperience in supporting oncology drug develop-ment at Merck. I am currently in one of the 5US LDS research groups and I am excited to havelearned that I will soon be assigned to the LDS re-search group in the Asia-Pacific region for a periodto support the building of the organization in theregion.
© www.icsa.org 83
Statisticians at Work July 2012 Vol.24/2
I started to work as a Merck LDS statisticiansupporting oncology over 5 years ago. At that time,Merck had a fairly young and small oncology fran-chise, compared to industry leaders in oncology.There was just a very small team of LDS statisti-cians supporting oncology. Over the past a fewyears, Merck oncology has been expanding with amuch stronger pipeline. It feels very fulfilling to bepart of the growth.
At Merck oncology, LDS statisticians play a keyrole in supporting the clinical development strate-gies. One of Merck oncology’s focuses is targetedtherapies. We are interested in identifying the pa-tients who are most likely to respond to particu-lar drugs based on the molecular characteristics oftheir tumors. To support the strategy, LDS statis-ticians researched into not only statistical methodsbut also portfolio management approaches to eval-uate different ways of developing a target therapy.Another focus of Merck oncology is novel combina-tion therapies. There are a variety of complexitiesfor combination development in all phases of clini-cal trials, from dose finding to providing confirma-tory data for regulatory approval. LDS statisticianscollaborate closely with clinical and regulatory col-leagues and engage in discussions with external ex-perts and regulatory agencies to advance the devel-opment of combination therapies.
Cancer clinical trials have some unique charac-teristics. Clinical proof of concept and sometimeseven regulatory approval is based on a surrogateendpoint. LDS statisticians provide pivotal sup-port in understanding the relationship between thesurrogate endpoint and the clinical endpoint. Acommonly used surrogate endpoint is often a timeto event endpoint, e.g. progression-free survival,assessed by regular disease assessments, e.g. CTscans every 8 weeks. Such data are interval cen-sored in nature in that the true event time is onlyknown to be within an interval between disease as-sessments. A scientific understanding of the datarequires an understanding of the interval censorednature of the data and statistical analysis consistentwith generation of the data. LDS statisticians ad-vocate the use of interval censored methods. Wehave evaluated the statistical properties of differentmethods. For registration studies, we have also dis-cussed with regulatory agencies on the use of meth-ods that take into account the interval censored na-ture of the data.
To improve efficiency to support cancer clinicaltrials, LDS statisticians look for ways to standardizethe process for certain deliverables. For example,we have partnered with statistical programmers to
standardize safety and even some efficacy tables forclinical study reports. Another example is that wehave been pushing to use standardized approachesfor dose finding Phase I studies, based on method-ology developed by researchers at MD AndersonCancer Center and our own customization.
I have enjoyed working on oncology drug de-velopment. In working on early phase studies, itis fun to speak to clinical colleagues to understandthe biology of the disease, scientific rationale for thetherapy, and their clinical insight, sometimes basedon experience in individual patients. It has not onlyenriched my knowledge about cancer but also al-lowed me to collaborate in a more productive andcustomer focused way. In later phase studies, sub-stantial statistical input is needed and can be crit-ical as complex issues such as subgroups definedby biomarker, surrogate endpoint, interim analy-sis for futility and/or efficacy arise. It is fulfill-ing to be able to provide statistical guidance andknow that my work would make a difference incancer patients’ lives. It is particularly fulfilling towork on regulatory submissions. I was fortunateto be the statistical lead of the worldwide regula-tory submission of a cancer product. It was inten-sive cross-functional collaboration effort to preparethe documentation for submission. Communica-tion is really important not only for internal coor-dination of submission activities but also for inter-action with regulatory agencies. It was interest-ing to experience different levels of statistical ex-pertise from regulatory agencies and scientific ex-perts worldwide. I worked in the sponsor team forthe advisory committee meeting, which was part ofthe interaction with the US FDA for their review ofthe submission. During the preparation, we con-ducted a large number of analyses in response topossible questions from the committee. It was in-tense and yet fulfilling to collaborate with clinicaland regulatory colleagues to translate a questioninto statistical inquiries, perform the analysis, in-terpret the results, and form a response. There aremany examples of challenging questions. One ex-ample is how to evaluate and present the treatmenteffect size for interval censored data. We had a lotof discussion on how to put appropriate clinical in-terpretation for the abstract concept of hazard ra-tio. Another example is the analysis and interpre-tation of patient reported outcome data. Expertshave been more and more focusing on how can-cer patients feel in the evaluation of cancer thera-pies. Yet, such data collected in clinical trials oftenhas a design limitation which provides very little,if any, data after patients are off therapy. This lim-
84 © www.icsa.org
July 2012 Vol.24/2 Statisticians at Work
itation creates informative missing data and posesgreat difficulty in analysis and interpretation. Mystatistical colleagues and I were at the center of thestage in tackling such difficult issues. In address-ing regulatory questions and particularly in prepar-ing for and attending the advisory committee meet-ing, I felt a great sense of responsibility to presentthe data with both correct statistical interpretationand understandability for experts with less statis-tical training. After all, the interaction between thesponsor and regulatory agencies on the appropriateunderstanding of the data matters most to cancerpatients.
Jerald S. Schindler, Ph.D.Head of Late Development StatisticsBiostatistics and Research DecisionSciences (BARDS)Merck Research LaboratoriesMerck & Co., Inc.
Yang Song, Ph.D.Associate DirectorLate Development StatisticsBiostatistics and Research DecisionSciences (BARDS)Merck Research LaboratoriesMerck & Co., Inc.
Bringing Statistical Innovation toOncology Drug DevelopmentPandurang M. Kulkarni, Nathan H. Enas, and YanpingWang
In an earlier ICSA Bulletin article, our colleaguesRuberg and Fu (2012) from Lilly Global StatisticalSciences shared our journey on bringing statisticalinnovation to life in drug development. To buildon that, we would like to share with you some ofthe statistical innovation we have brought to theOncology drug development and how we are mak-ing what is “advanced today, routine tomorrow”.Though there are many topics we could highlight,we are focusing here on two important topics thathave major impact on the development of Oncol-ogy molecules. The first one has to do with the typeof designs, and the second concerns the likelihoodof success of the designs. We provide an overviewof these two topics and share how we have goneabout advancing drug development using analyti-cal approaches.
Reducing Single Arm Trials byEfficiently Enabling RandomizedTrials
In oncology clinical drug development, clinical tri-als used to test for drug efficacy have often been de-signed without a concurrent control group. These“single arm” studies rely on historical controls inorder to assess the magnitude and significance of
the new drug candidate. However, single arm tri-als (SAT) have severe inferential limitations, owingto bias in patient selection and evaluation, makingthe comparison to historical controls highly unre-liable. On the other hand, randomized controlledtrials (RCT) are the “gold standard” for evaluatingnew drug candidates, but are also more costly andtime-consuming than single arm studies.
From an inferential perspective, SATs rely on a”control arm” that is completely constructed fromhistorical patient outcomes (i.e., 100% “borrowing”from historical control), whereas traditional RCTsdo not rely at all on historical patient outcomes(i.e., 0% “borrowing” from historical control). Oneof the arguments used for not conducting random-ized trials is the extra cost and time associated withRCTs compared to the SATs. However, this costcontainment has the risk that we are all too famil-iar with the SATs, and the Oncology communityhas come to realize this. Therefore, to balance thetwo approaches, we need to offer an alternativethat can be used whenever possible. For instance,Bayesian methods can help model the continuumbetween SATs and RCTs based on the amount of“borrowing” of historical control information uti-lized within a prospective clinical trial. This contin-uum is the domain of the so-called Bayesian Aug-mented Control (BAC) design. This design beginswith the RCT design, but uses Bayesian modelingto enable borrowing from historical control patientoutcomes collected in prior clinical trials (borrow-ing x% depending on the appropriateness of histor-
© www.icsa.org 85
Statisticians at Work July 2012 Vol.24/2
ical data to the current trial). Thus, in cases wherethere is very good historical information availableon the control, we could design an RCT that allo-cates much smaller number of patients to controlarm than to the treatment arm. Cost of this RCTshould be much less than the standard RCT whileit is somewhat greater than SAT. On the other hand,if the historical data is not robust then one shouldborrow less. As one borrows less the cost of thisBAC design would be closer to the standard RCTand considerably more than SAT. Even in thesecases, we strongly recommend using RCTs becauseobviously it is clear that you do not have any goodhistorical data to compare against and hence a sin-gle arm study would provide you with no cleardecision about whether to proceed to Phase 3 andwhat outcome to expect in Phase 3. There are spe-cial circumstances where SAT may make sense. Forexample, if there are no accepted treatment optionsand definitive efficacy is expected (e.g., completetumor response). In such cases, it is likely that anSAT may be acceptable and act as the pivotal studyfor provisional regulatory approval.
The BAC methodology has now been op-erationalized and simulations are made pos-sible through the use of software calledFACTS (http://www.smarterclinicaltrials.com/what-we-offer/facts/) and supported byour Advanced Analytics group of the Lilly GlobalStatistical Sciences function. We have utilized thismethodology in numerous oncology trials, the firstof which is completing in the near future. Initiallythis was met with caution and uncertainty. How-ever, as clinicians understood the details and be-came more comfortable with the approach, and wewere able to quickly simulate many scenarios using
the FACTS software, we have been able to applythis innovative approach on a regular basis. Thus,what was advanced a few years ago is routine now.
As with any Bayesian design, this model affordsresearchers greater control over important designparameters than frequentist designs, but with thisincreased control comes increased responsibility tochoose valid historical data and model specifica-tions. Nevertheless, when credible and relevanthistorical data exist for the prospective control ther-apy, this design has become a method of choice forPhase II cancer trials at Lilly. Therefore, a full RCTshould be the first choice, and when valid histor-ical information exists, then BAC should be givenserious consideration.
Understanding and Increasingthe Probability of Study Success(PrSS)
Oncology clinical trials can be very lengthy, costly,and hard to enroll due to limited patient pool. Fur-ther, most often these trials are dealing with treat-ing patients with potentially life ending cancers.Yet, only 34% of Phase 3 oncology drug trials withresults reported from 2003 through 2010 were suc-cessful (Sutter and Lamotta, 2011). Therefore, a crit-ical question that study sponsors always ask beforeinvesting in a clinical study is how likely is it thatthe study is going to be successful.
Traditional statistical power does not providea reliable answer to this question — most Phase 3studies are powered at 80% or higher, but the suc-cess rate of Phase 3 studies is much lower than 80%.
86 © www.icsa.org
July 2012 Vol.24/2 Statisticians at Work
This is so because power is the probability of suc-cess (i.e., achieving statistical significance) at an as-sumed effect size. The assumed effect size is of-ten based on regulatory, payer and/or marketingrequirements or needs and may not be supportedby available evidence or reflect the true treatmenteffect. Occasionally, the effect size is estimatedfrom available evidence, but the estimate is typi-cally treated as a fixed constant without any vari-ability. In short, not utilizing available data or notutilizing the data appropriately is the underlyingreason that power does not reliably measure theprobability of study success (PrSS).
Several authors have proposed to use the socalled “assurance”, “expected power”, or “aver-age success probability” to quantify PrSS (O’Haganet al., 2005; Chuang-Stein, 2006). The idea is to useavailable data to derive the distribution of the un-known true treatment effect and then average thepower function of the new study over the distri-bution. The resulting average is nothing but theso-called predictive power, a hybrid of frequen-tist and Bayesian concepts (Dmitrienko et al., 2005;Spiegelhalter et al., 2004). The PrSS is based onavailable data and also accounts for the uncertaintyin our current knowledge on the treatment effect,thus representing an improvement over power inquantifying the likelihood of success for the newstudy. Often times, the distribution of treatmenteffect and/or the design of a new study are quitecomplex, but simulation studies are very helpful inestimating the PrSS. In Lilly Oncology, we startedto institute this thought process to evaluate everyPhase 3 study for its PrSS, once again enabled byour Advanced Analytics group. Initially, this wasdifficult and harder to consume by the customers,as they are accustomed to the traditional conceptof “power”. However, because of its intuitive ap-peal, we have been able to make this idea routinethrough education and use in all Phase 3 trials. Wehave demonstrated to the teams that PrSS helps im-prove study designs and inform clinical planning.This approach also has been extremely helpful toevaluate the utility of interim analysis and the ul-timate PrSS. Therefore, simulating the PrSS by uti-lizing historical data from “proof of concept” andother trials as appropriate is a routine benefit thestatisticians should provide to their clinical teamsto help make better decisions and to improve thesuccess of Oncology drug development.
In summary, with the focused effort and leader-ship from Oncology Statisticians and the AdvancedAnalytics group, we have been able to turn statisti-
cal innovation into routine practice.
Bibliography
Chuang-Stein, C. (2006). Sample size and the prob-ability of a successful trial. Pharmaceutical Statis-tics 5, 305–309.
Dmitrienko, A., Molenberghs, G., Chuang-Stein,C., and Offen, W. (2005). Analysis of Clinical TrialsUsing SAS: A Practical Guide. SAS Institute Inc.,Cary, NC.
O’Hagan, A., Stevens, J., and Campbell, M. J.(2005). Assurance in clinical trial design. Phar-maceutical Statistics 4, 87–201.
Ruberg, S. and Fu, H. (2012). The Advanced Ana-lytics Hub at Eli Lilly. International Chinese Statis-tical Association Bulletin 24, 22–25.
Spiegelhalter, D. J., Abrams, K. R., and Myles, J. P.(2004). Bayesian Approaches to Clinical Trials andHealthcare Evaluation. Wiley, Chichester.
Sutter, S. and Lamotta, L. (2011). Cancer drugs haveworst phase III track record. Internal MedicineNews.
Pandurang M Kulkarni, Ph.D.Senior Director & Global Head ofLilly & ImClone Oncology Statis-ticsGlobal Statisticcal Sciences & Ad-vanced AnalyticsEli Lilly & Co
Nathan H. Enas, M.S.Research AdvisorGlobal Statistical SciencesEli Lilly and Company
Yanping Wang, Ph.D.Principal Research ScientistGlobal Statistical SciencesEli Lilly and Company
© www.icsa.org 87
Blog Spot July 2012 Vol.24/2
Reproducible Research: Notes from theFieldRoger Peng
Editorial: “Simply Statistics” is a blog of threebiostatistics professors at Johns Hopkins University(Jeff Leek, Roger Peng, and Rafa Irizarry), who arefired up about the new era where data are abun-dant and statisticians are scientists; see http://
simplystatistics.tumblr.com/. Dr. Roger Pengwrote on reproducible research on November 6, 2011.A slightly edited version is published here with permis-sion.
Over the past year, I’ve been doing a lot oftalking about reproducible research. Talking topeople, talking on panel discussions, and talkingabout some of my own work (a YouTube videois available at http://www.youtube.com/watch?v=aH8dpcirW1U). It seems to me that interest in thetopic has exploded recently, in part due to some re-cent scandals, such as the Duke clinical trials fiasco(see, e.g., a very good summary in The Economist,http://www.economist.com/node/21528593).
If you are unfamiliar with the term “repro-ducible research”, the basic idea is that authors ofpublished research should make available the nec-essary materials so that others may reproduce to avery high degree of similarity the published find-ings. If that definition seems imprecise, well that’sbecause it is.
I think reproducibility becomes easier to de-fine in the context of a specific field or application.Reproducibility often comes up in the context ofcomputational science. In computational sciencefields, often much of the work is done on the com-puter using often very large amounts of data. Inother words, the analysis of the data is of compa-rable difficulty as the collection of the data (maybeeven more complicated). Then the notion of repro-ducibility typically comes down to the idea of mak-ing the analytic data and the computer code avail-able to others. That way, knowledgeable peoplecan run your code on your data and presumablyget your results. If others do not get your results,then that may be a sign of a problem, or perhapsa misunderstanding. In either case, a resolutionneeds to be found. Reproducibility is key to sci-ence much the way it is key to programming. Whenbugs are found in software, being able to reproducethe bug is an important step to fixing it. Anyone
learning to program in C knows the pain of dealingwith a memory-related bug, which will often ex-hibit seemingly random and non-reproducible be-havior.
My discussions with others about the need forreproducibility in science often range far and wide.One reason is that many people have very differentideas about (a) what is reproducibility and (b) whywe need it. Here is my take on various issues.
• Reproducibility is not replication. There’s of-ten honest confusion between the notion ofreproducibility and what I would call a “fullreplication”. A full replication doesn’t ana-lyze the same dataset, but rather involves anindependent investigator collecting an inde-pendent dataset conducting an independentanalysis. Full replication has been a fun-damental component of science for a longtime now and will continue to be the primaryyardstick for measuring the plausibility of sci-entific claims. I think most would agree thatfull replication is preferable, but often it issimply not possible.
• Reproducibility is not needed solely to pre-vent fraud. I’ve heard many people empha-size reproducibility as a means to preventfraud. Journal editors seem to think this isthe main reason for demanding reproducibil-ity. It is one reason, but to be honest, I’m notsure it’s all that useful for detecting fraud. Ifsomeone truly wants to commit fraud, thenit’s possible to make the fraud reproducible.If I just generate a bunch of numbers andclaim it as data that I collected, any analysisfrom that dataset can be reproducible. Whiledemanding reproducibility may be useful forferreting out certain types of fraud, it’s not ageneral solution and it’s not the primary rea-son we need it.
• Reproducibility is not as easy as it sounds.Making one’s research reproducible is hard.It’s especially hard when you try to do it afterthe research has been done. In that case it’smore like an audit, and I’m guessing for mostpeople the word “audit” is NOT synonymouswith “fun”. Even if you set out to make yourwork reproducible from the get go, it’s easy
88 © www.icsa.org
July 2012 Vol.24/2 R ‘ R’ Us
to miss things. Code can get lost (even witha version control system) and metadata canslip through the cracks. Even when you’vedone everything right, computers and soft-ware can change. Virtual machines like Ama-zon EC2 and others seem to have some poten-tial. The single most useful tool that I havefound is a good version control system, likegit (http://git-scm.com/).
• At this point, anything would be better thannothing. Right now, I think the bar for repro-ducibility is quite low in the sense that mostpublished work is not reproducible. Even ifdata are available, often the code that ana-lyzed the data is not available. So if you’republishing research and you want to make itat least partially reproducible, just put whatyou can out there. On the web, on github(http://github.com/), in a data repository,wherever you can. If you can’t publish the
data, make your code available. Even thatis better than nothing. In fact, I find read-ing someone’s code to be very informativeand often questions can arise without look-ing at the data. Until we have a better in-frastructure for distributing reproducible re-search, we will have to make do with what wehave. But if we all start putting stuff out there,the conversation will turn from “Why shouldI make stuff available?” to “Why wouldn’t Imake stuff available?”
Roger PengAssociate ProfessorDepartment of BiostatisticsBloomberg School of Public HealthJohns Hopkins University
Making Reproducible ResearchEnjoyableYihui Xie
It is hard to convince people to think about repro-ducible research (RR). There are two parts of dif-ficulties: (1) tools used to be for experts only and(2) it is still common practice to copy and paste.For some statisticians, RR is almost equivalent toSweave (R + LATEX). I love LATEX, but LATEX is stillhell to many people. I had an experience of teach-ing Sweave in a stat-computing class at Iowa StateUniversity, and I can tell you their horrible faces af-ter I taught them LATEX in the first half of the class.I will never do that again.
But RR is really important. I recommend youto watch this video if you have not heard of theDeception at Duke to see how improper data pro-cessing killed patients: http://www.cbsnews.com/video/watch/?id=7398476n, then you should feelguilty when you copy and paste as a statistician.I fully respect the seminal work of Sweave, but inmy eyes, it is really a half-done project which didnot make much progress in the past few years. Isuggested a few features to the R core team, whichwere often rejected. I understand that R is too big tomake substantial changes. As a useR, you always
have the right to vote by packages, so I wrote theknitr package to fully implement what I thoughtwould be a good engine for RR with R.
The basic ideawas the same: tomix code and text to-gether, then compilethe whole documentwith code being exe-cuted, and you get areport without copy-
ing/pasting anything since the code will faithfullygive you results. The design was very differentfrom Sweave, however: knitr is not restricted toa specific format like LATEX any output formatis possible, including HTML, Markdown and re-StructuredText. I will ignore LATEX in this article,although it took me much more time to work onthan other formats.
I use Github extensively and learned mark-down there. For those who are not familiar withmarkdown, it is an extremely simple language andyou can learn it in five minutes at most: http:
//en.wikipedia.org/wiki/Markdown. It was al-most trivial for me to add support for markdown
© www.icsa.org 89
R ‘ R’ Us July 2012 Vol.24/2
in knitr, so we can mix R code and markdown texttogether and compile reports quickly. That was thebeginning of the story.
Later, the developers of RStudio (http://www.rstudio.org, the IDE, integrated development en-vironment, of R) saw the work of knitr and decidedto add support to it. First we finished the workwith Sweave documents, which was painful but re-warding (well, that is LATEX). Before that I had fin-ished adding the knitr support in LyX – an excel-lent front-end of LATEX and RR became enjoyablesomehow, but only enjoyable for me and perhapsalso some other LyX users. We could write LATEXeasily and click the button to get a PDF report fromLyX, which was quite handy (http://yihui.name/knitr/demo/lyx/).
After the Sweave work was done, I suggestedmarkdown to RStudio developers, and fortunatelythey listened. The progress was fast; soon we hada format named R markdown in RStudio. That waswhen I believed RR became accessible to the gen-eral audience.
And suddenly a golden glow de-scended on me, and all my sins werewashed away. . .
Many people seem to have been waiting for a sim-ple format like R markdown for a long time. Theonly thing you need to do for a reproducible reportis to write code and text. When you write in LATEXthere are tons of rules to remember like which char-acters need to be escaped, or how to write a back-slash or tilde, whereas in markdown, you feel likewriting emails.
JJ Allaire (one of the RStudio authors) and Iwere invited to give a talk (http://yihui.name/slides/2012-knitr-RStudio.html) at useR! 2012on RR a few days ago, and we successfully con-vinced quite a few people to RR and R markdown.One of my points was that RR should be made en-joyable. If people suffer from tools all the time,there is no hope for RR to become the commonpractice. To ask people to go to the right way, wejust need to make the right way easier than thewrong way (one smart guy in the audience saidthis after we gave a talk to the Twin Cities R UserGroup). Chris Fonnesbeck, an instructor in Bio-statistics at Vanderbilt University, decided to com-pletely ban Word documents in his Bios301 thisFall. I admire his courage, and I am evil to be happyto see Word die, but I will be happier if the studentscan see why Word sucks and how knitr/RStudio/Rmarkdown can make things much easier and morebeautiful. As I proposed at useR! 2012, we should
really start to train students to do their homeworkassignments in a reproducible manner before theydo research in the future. This is not hard now.
Kevin Coombes and Keith Baggerly are the twoheroes (and detectives) who revealed the Dukescandal, which I mentioned before. They have beentrying to promote Sweave, and I was thrilled atuseR! 2012 that Kevin used one slide to introduceknitr in his invited talk. I was also excited whenKeith told me R markdown was cool and he wasgoing to use it in his reports.
There are many other features in knitr whichmake RR enjoyable. For example, code is high-lighted by default so that plain text will not becomepain text; for users who do not care about codingstyles, their code will be automatically reformattedwith the formatR package to make ugly code morereadable (Martin Maechler does not like this but heis an R expert and knows how to format R code);figures will never exceed the page margin in LATEXoutput; you do not have to use dirty tricks in or-der to get multiple figures per chunk; . . . In all, weget beautiful reports by default, although the beautyhere is highly opinionated.
It is always enjoyable when we can embracethe web, where we have lots of fancy technologies.Markdown can be easily translated into HTML, sowe can build web applications with knitr as well.Two examples:
1. Rpubs.com (http://rpubs.com): you canpublish your reports to this website (hostedby RStudio) freely from RStudio, and you cansee there have already been a couple of nicereports (just forget about emailing ugly Worddocuments back and forth)
2. An OpenCPU demo: http://public.
opencpu.org/apps/knitr/ (you do not needanything but a web browser, then you cancompile a report in the cloud)
You can see what other people have been doingwith knitr at http://yihui.name/knitr/demo/
showcase/. Let’s stop the old habit of copy andpaste. Let the code speak, and in code we trust.
Yihui XiePhD studentDepartment of StatisticsIowa State [email protected]
Tel: (+1) 515-203-2465http: // yihui. name
90 © www.icsa.org
July 2012 Vol.24/2 Professional Opportunities
Upcoming EventsJoint Statistical Meetings 2012
July 28 — August 2, 2012San Diego, CA, USAhttp://www.amstat.org/meetings/jsm/2012/
index.cfm
Computational Advertising Summer Pro-gram
August 6 — August 17, 2012Research Triangle Park, NC, USAhttp://www.samsi.info/
Data-Driven Decisions in HealthcareProgram Opening Workshop
August 26 — August 29, 2012Research Triangle Park, NC, USAhttp://www.samsi.info/
Statistical and Computational Methodol-ogy for Massive Datasets Program Open-ing Workshop
September 9 — September 12, 2012Research Triangle, NC, USAhttp://www.samsi.info/
IAOS 2012 Official Statistics: GettingYour Messages Across
September 12 — September 14, 2012Kiev, Ukraine
http://iaos2012.ukrstat.gov.ua/
BASS XIX 2012
November 5 — November 9, 2012Savannah, GA, USAhttp://bass.georgiasouthern.edu/
The 68th Deming Conference on AppliedStatistics
December 3 — December 7, 2012Atlantic City, NJ, USAhttp://demingconference.com/
ASA Conference on Statistical Practice2013
February 16 — February 18, 2012New Orleans, Lousiana, USAhttp://www.amstat.org/meetings/csp/2013/
index.cfm
29th European Meeting of Statisticians
July 20 — July 25, 2013Budapest, Hungraryhttp://ems2013.eu/site/index.php?page=en/
Home
Professional OpportunitiesFor details and contacts about all posts, see http:
//www.icsa.org/job/index.html.
Head of Department of Applied Mathe-matics, The Hong Kong Polytechnic Uni-versity
Experience in fund-raising will be an additional ad-vantage. Terms of appointment and remunerationpackage are negotiable and highly competitive.
Faculty Position in Statistics or Econo-metrics, Central University of Financeand Economics (CUFE)
School of Statistics at CUFE invites applications forfull-time tenure-track positions of all ranks (Assis-tant, Associate, and Full Professor) in all areas ofstatistics or econometrics to begin in the fall of 2012.Salary and benefits are competitive and commen-surate with qualifications and experience.
© www.icsa.org 91
Membership Application & Renewal Form
Name (Last) (Middle) (First)
(English)
(Chinese)
Affiliation:
Gender (optional):
Address
Office Address:
City:
State: Zip Code: Country:
Email: Telephone: FAX:
Home Address:
City:
State: Zip Code: Country:
Email: Telephone: FAX:
Education
Degree: Year Graduated:
University:
Professional Occupation & Title
Occupation: Title:
Membership Fees Regular (US$40)
Student (US$10)
Permanent (US$600)
Spouse (50%)
Donations
Total Amount Paid: US$
Statistical Area of Interest (circle all applicable): A: Agriculture B: Business / Economics
C: Computing / Graphics D: Education E: Engineering F: Health Sciences
G: Probability H: Social Sciences I: Biostatistics N: Theory & Methodology
Would you like to join the Biometrics Section (please circle your answer)? Yes No
Please Make Check Payable to: I.C.S.A. Mail This Form & Fees to: ICSA c/o Shuyen Ho, Statistics and Programming GSK, GlaxoSmithKline, 5 Moore Drive, PO Box 13398, Research Triangle Park, NC 27709, USA
![[PPT]PowerPoint-Präsentation - Das größte deutsche …data.lustich.de/powerpoints/o/2393-maenner-t-shirts.pps · Web viewTitle PowerPoint-Präsentation Author Oliver Last modified](https://static.fdokument.com/doc/165x107/5ba0ca2c09d3f2857a8d826a/pptpowerpoint-praesentation-das-groesste-deutsche-data-web-viewtitle-powerpoint-praesentation.jpg)
