ISI - International Statistical Institute
Newsletter Volume 26, No. 2 (77) 2002

News of Members

 

In Memoriam 
 

Deceased Members 

The ISI regrets to announce the death of our colleagues: 
  Born  Elected  Deceased
Prof. Em. Bernard Benjamin  1910  1964 15 May 2002 
Prof. Maurice S. Bartlett 1910  1949   2002
Prof. Dr. Indra Chakravati 1928 1969  2002
Prof. Tore Elon Dalenius  1917 1957 2002
Prof. Dr. Andi Hakim Nasoetion 1932  1972  14 March 2002
Prof. Dr. Burton T. Oņate 1921 1971 24 March 2001
Jeanne E. Griffith 1950 1992 2002

 

  Maurice S. Bartlett (1910-2002)

Maurice Bartlett was born in 1910 in Stoke-on-Trent. He read Mathematics at Cambridge obtaining a First Class degree in 1932. In his final year as a student he went to lectures by J.Wishart, who had recently derived the distribution that bears his name. In one lecture Wishart suggested that the distribution might perhaps be derived by characteristic functions and the next day Bartlett, Wishart once told me, gave him the proof. This formed Bartlett's first paper (jointly with Wishart). After one year as an Assistant Lecturer at University College London (UCL) Bartlett worked for four years in the agricultural research station of Imperial Chemical Industries before returning to Cambridge in 1938 as a Lecturer in Mathematics.

During these years he published pioneer papers on the 2 by 2 by 2 contingency table, on the geometry of the linear model, on transformations in analysis of variance, on estimation problems in factor analysis, on spatial aspects of design in field trials, as well as a remarkable group of papers on foundational problems in statistical inference and the role of the likelihood. In one of these he gave the test of homogeneity of variance that bears his name, essentially introduced marginal and conditional likelihood and investigated the adjustment to the likelihood ratio test that is also called after him.

After military work during the War he returned briefly to Cambridge before becoming the first Professor of Mathematical Statistics at University of Manchester. Subsequently he held chairs at UCL and at Oxford, retiring from there rather early, partly on medical grounds.

In the period from 1945 there followed major papers on, among other topics, time series, multivariate analysis, stochastic processes, point and spatial processes and likelihood derivatives. He used theoretical models to explain why measles epidemics had a two-year periodicity in some cities and not in others. He wrote also about the connection between probability and quantum mechanics and about the nature of probability. His last major paper read to the Royal Statistical Society when he was 80 dealt with the relation between chaos and stochastic processes.

In 1955 there appeared the first of several editions of Introduction to stochastic processes and its applications, one of the most remarkable books in our subject. It covers quite concisely stochastic processes in discrete and continuous time, and statistical inference in such processes, including time series as a special case of a stochastic process. The book had been completed some years earlier, partly on the basis of lectures he gave at University of North Carolina, Chapel Hill, but Bartlett characteristically waited some years in the unrealized hope that a colleague would complete a promised contribution on stochastic processes in quantum theory. The book is full of deep insights and it may well be that not all of them have yet been fully absorbed. 

In 1961 he was elected F.R.S., in 1980 to Honorary Membership of ISI and in 1992 to be a Foreign Associate of the (US) National Academy of Sciences. In 1988 three volumes of his papers were published, with commentary, by University of Manitoba Press, under the editorship of R.G.Stanton and others.

To say that his lectures and papers were and are not always totally easy to understand would be an understatement. His concern was ultimately with the application of statistical ideas to serious subject-matter topics. Thus, for example, I am not aware that any of his mathematical papers contained regularity conditions for the precise validity of the statements made. I do not think this was the source of the difficulty of the papers. This lay rather in the depth and density of the ideas being set out combined with a precise but concise style of writing which means that every sentence, and sometimes every phrase or even word, has to be savoured before proceeding to the next.

Bartlett was a large man, somewhat austere at first meeting but intrinsically of great kindness. He married quite late in life Sheila Chapman and the synergy between Maurice's apparent austerity and Sheila's ebullience was wonderful to see and moving to recollect. She died in 1998. At a small party for his 90th birthday held in his apartment overlooking the sea at Exmouth he was very frail physically but alert mentally, interested in and enquiring about current scientific activities. Up to the end his very small handwriting remained totally clear and he answered technical questions about his earlier work with precision.

He was one of the truly great figures of 20th century statistics.

D.R. Cox
Nuffield College
Oxford

Top

Indra Mohan Chakravarti (1928-2002)

Indra Mohan Chakravarti was born on April 8, 1928 in Nawabganj in the province of Bengal, in a region which is now a part of Bangladesh. He was one of the top ten among 35,000 students in the Calcutta University Matriculation Examination in 1944. He was attracted to the newly created undergraduate Statistics Honors program in Calcutta's Presidency College, where he distinguished himself by serving a first class in both the B Sc (Hous) and M Sc degrees in Statistics.

Indra began his research career at the Indian Statistical Institute, in 1952, where he worked under the guidance of C.R. Rao and received his PhD in 1958. During 1959-1964, he visited the University of North Carolina at Chapel Hill, the Case Institute of Technology in Cleveland, Ohio, the University of Geneva in Switzerland and the University of Aberdeen in Scotland. Indra came back to Chapel Hill in the fall of 1964, where since 1968 he had been Professor of Statistics. He continued to travel to Switzerland and France during his tenure at Chapel Hill.

Indra's research in the 1950's was in the broad areas of: (i) sample surveys and finite population sampling, (ii) statistical inference and (iii) the design and analysis of experiments. In the late 1950's, his association with Professor R.C. Bose was a turning point in his career. It reversed his interest in combinatorial mathematics, the design of experiments and error correcting codes. Beginning with the mid 1960's, Indra did his best work primarily in these areas. After R.C. Bose retired from Chapel Hill in 1971, Indra took over as the Department's expert in these subjects.

During 1969-2001, Indra supervised the PhD dissertation of 15 advisees, all in Chapel Hill and almost all in the area of combinatorial mathematics and design of experiments. He has nearly 50 publications in this area.
He maintained his interest in scientific matters almost till the end of his life. He took a deep interest in the current emphasis on bioinformatics and genomics, was planning to offer an advanced graduate level course in the coming Fall on the design and analysis of bioinformatics studies with feedback from classical combinatorics and the design of experiments; his death left that task unfulfilled.

Indra is survived by his only son Xavier. His wife Monique died in 1993.

P.K. Sen and G. Kallinpur

Top

Tore Elon Dalenius (1917-2002)

The statistical profession lost one its prominent members when Professor Tore E. Dalenius passed away in January, 2002, at the age of 84. He grew up in Sweden where he also received most of his education. His university studies covered, besides Statistics, also disciplines such as Political Science and Economics. Among his teachers were prominent members of Sweden's scientific, intellectual, and political life at the time, including political scientist Herbert Tingsten and economist (later Nobel laureate) Gunnar Myrdal.

Heeding the advice of one of his mentors to learn more about the practical aspects of statistics, he gathered such experience in the early 1940's. One of his employers was the Swedish Gallup Institute. 

His early experiences with practical survey work explain why, even in his subsequent academic career, he never lost sight of, and never trivialised, the more mundane aspects of surveys.

He also studied in the United States. In 1947-48, he was at Cornell University, where his main teacher was Philip J. McCarthy. While there he entered into contact with noted social scientists, such as Paul Lazarsfeld and others.

At the time, the Swedish Ph.D. degree was a pinnacle reached only at a mature age after long lasting effort and a voluminous thesis. Dalenius' doctoral thesis from 1957 is entitled Sampling in Sweden. It analyses the development and the gradual but slow inroads in Sweden of scientific sampling. The notions of efficient randomised sample selection were still in their infancy when Dalenius began his career in the 1940's. One must bear in mind that this was not long after the theoretical breakthrough that Neyman's famous 1934 paper had provided, especially for practising optimal allocation of stratified samples. In Sweden of the 1950's, Dalenius contributed greatly to promoting the philosophy and the practice of efficient sampling design. 

His thesis also contains important theoretical contributions, notably for determining optimal stratum boundaries for stratified simple random sampling. He wrote several papers about this problem; the Dalenius-Hodges rule is still often used and referred to.

He spent fruitful periods at the US Bureau of the Census in Washington, permitting him to observe and participate in the survey methodology research and development of the group around Morris Hansen. He often spoke admiringly about the high calibre of the scientific and professional environment he found there, and about Hansen's inspiring leadership. 

Throughout his career he was to entertain a lively exchange of ideas with professional and scientific circles in various countries beyond Sweden, especially in the United States.

In 1951 Dalenius was named head of a newly established survey unit at Sweden's national bureau of statistics, Statistics Sweden. Over the years, this unit was to carry out many surveys of a national scope. These surveys often used sample selection via a master sample of geographically defined units.

In 1965 he was named Professor of Statistics at the University of Stockholm. The post was designated as one "with specialisation in Official Statistics", the first of its kind in Sweden. It stipulated a time sharing with Statistics Sweden. He retired from this post in 1983. However, starting early in the 1970's, he spent long periods of leave of absence in the United States, mainly at Brown University. He continued to lead broad and important research projects, some with funding from the Bank of Sweden Tercentenary Foundation. One of these dealt with errors in surveys, another with confidentiality in surveys.

He founded the Swedish Statistical Association in 1962 and was one of the initiators of the International Association of Survey Statisticians (IASS), whose president he was in 1979-80.

Dalenius demanded strict adherence to clearly defined quality requirements for survey work. Some of these demands may seem unrealistic in today's hardened survey climate. He deplored the "high" nonresponse rates in many modern surveys. It should be kept in mind that 40 years ago, a 5% nonresponse rate was considered high for important surveys.

As survey participation declined in the 1980's and 1990's, he was alarmed that surveys with high rates of nonresponse were even allowed to continue and to deliver published results. He argued that attempting to "correct", after the fact, for nonresponse was of questionable merit. Statistics, or numbers, derived from some surveys were in his opinion next to useless and should "belong in the waste basket" rather than be published. His main concern was the different varieties of bias, the squares of which represent unmeasured, or immeasurable, contributions to the total Mean Squared Error.

The notion of total survey error came early to him. Already in the late 1940's, he had realised that a survey is typically subject to errors of many kinds and that a "total survey error theory" was needed to measure and control, in an optimal or near-optimal fashion, the different sources of error. Even though similar ideas may have evolved simultaneously in a few other places, it is clear that he was a precursor for our modern view on survey error as a multifaceted concept.

He knew that although sampling error received much attention, nonsampling errors, especially the measurement errors, could completely dominate the total error of a survey estimate. He was aware, though, of the difficulties, both theoretical and practical, in the objective measurement of all errors, and this still remains an obstacle for today's survey statisticians.

The title of a paper of his from 1977 illustrates the colourful and striking language that he was fond of using: "Strain at a gnat and swallow a camel or, the problem of measuring sampling and non-sampling errors." He states: "The sampling error plays the role of the gnat, sometimes malformed, while the non-sampling error plays the role of the camel, often of unknown size and always of unwieldy shape. There are some signs today that the situation just discussed is worsening: non-response rates have increased significantly in recent years and may become even higher." The decades that followed would confirm his fears.

It was the duty of the survey statistician, he argued, to actively and effectively combat nonresponse; high rates were in many cases a consequence of permissiveness or defeatism on the part of managers and others responsible for the survey process. He emphasised the importance of establishing a fruitful contact with the respondent so as to gain co-operation.

He maintained that nonresponse could be nearly eliminated in many surveys if only interviewer training, supervision, workload and encouragement were handled with skill and with assiduous effort. He would mention examples of surveys where he had been in charge and where, in the final analysis, nonresponse was very low.
To measure a phenomenon in society was, as I interpret him, more than matter of expressing it by a number, a point estimate. If you are unable to keep under strict control the often-considerable uncertainty of that number, you have in effect measured nothing, and little or no knowledge has been gained by the survey. Needless to say, his insistence on the highest standards was not always "politically correct" among administrators and managers in the world of surveys, or, more generally, among those persons whose statistical conscience was perhaps less sensitive or less developed than his own.

He challenged the profession to face the question: If the foundations of survey sampling, as he and many of his generation had learned them, were indeed solid ground on which to build, then one should not depart from these foundations, one should not permissively tamper with them.

At an early stage he noted the importance of relevance errors, that is, the hard-to-quantify imperfections arising out of the failure of a survey to address the "correct" variables and concepts. He noted, quite appropriately, that concepts which were adequate and of interest 30 years ago in a given statistical program were not necessarily so at present.

He recognised the importance of confidentiality and the need for active measures for protecting the privacy of respondents, and he wrote extensively about those issues. Sweden was one among several European countries where the 1980's brought heated public debate about issues of privacy and confidentiality of data. For example, the Swedish media argued that implementing a proposal for an essentially register based census in 1985 would significantly increase the risk for invasion of privacy. At the request of Statistics Sweden, Dalenius prepared in 1987 an extensive manual entitled Controlling Invasion of Privacy in Sweden.

Although he himself contributed considerably to theory and generally admired the mathematical/theoretical advancements in survey sampling, he was at the same time worried about the incompleteness of the education in survey methodology, in Sweden and elsewhere. 

In his own teaching, he did what he could to emphasise the total aspect of surveys, including those parts that find support in the behavioural sciences, such as issues relating to measurement and questionnaires. Developments over the last few decades have done him justice in this regard, in that in several countries now have teaching programs which promote a balanced view of the various steps in the statistical production chain.

Some would perhaps call Dalenius an idealist, implying that in today's hard survey climate one must "be realistic". To me, as to many others, he was one who, aided by his vast knowledge of the field, held high the banner of survey quality. He inspired us to do likewise.

Carl-Erik Särndal

Top

Jeanne E. Griffith (1950-2002)

Jeanne Griffith, who was elected a member of the ISI in 1992, and a Council member for the term 1999-2003 - a position she had to decline due to her health, left this world last August.

Her outstanding contribution to official statistics had been recognised in her country a few months before, when she was presented with the "Roger Herriot Award for Innovation in Federal Statistics" in Washington, on June 4, 2001. She developed her professional life around interests emerging from sociology, in which she majored at the College of William and Mary, and which she pursued at the University of Pennsylvania, George Washington University, and Johns Hopkins University through the doctorate level. 

She practised her skills as a statistician and demographer in numerous Federal positions of growing responsibility at the Bureau of the Census, the Department of Health, Education and Welfare, the Office of Management and Budget, the Congressional Research Service, the Department of Education, and the National Science Foundation. She served as Associate Commissioner, and later as Acting Commissioner of Education Statistics, at the Department of Education, and as Director of Science Resources Studies at NSF.

It seems also most appropriate that the ISI pay its own tribute to Jeanne Griffith's dedication to international co-operation with these few lines. When she was Associate Commissioner at the US National Center for Education Statistics, she was for some of us a major force in the INES project of OECD education indicators efforts to produce the now popular "Education at a Glance" report. The first edition of this now annual report was issued in September 1992 ; a few days later, Jeanne attended an IAOS biennial meeting in Ankara where statistics on education were on the agenda for the first time. She also delivered a very lively paper on the subject during the ISI 1995 session in Beijing, in the frame of a meeting I had the privilege of organising.

One of the major achievements among her international works is certainly the "Third International Mathematics and Science Study - TIMSS": the results of this unprecedented international education study have given statisticians and researchers, as well as educators and policy makers, entirely new concepts of what is possible to examine and learn from international education comparisons.

The ISI as well as her many colleagues in the US and around the world will miss her skills, her dedication, her kindness.

Michel Euriat*
INSEE - France

*With the help of Katherine K. Wallman (OMB-USA) and quotations from Emerson J. Elliott, former U.S. Commissioner of Education statistics.

Top

Cedric Smith (1917-2002)

Born during the Great War, Cedric Austen Bardell Smith was committed to Quaker pacifism, mathematics, and statistical genetics. As a Cambridge undergraduate he was one of the four members of the Trinity Mathematical Society which attacked arcane questions. They solved the problem called "Squaring the Square", demonstrating that a square is composed of smaller squares, all of different sizes, and applied this to electrical networks with varying resistance. He wrote amusing stories, poems and songs, usually about mathematics or mathematicians, under the pen-name of Blanche Descartes.

An attempt much later by a graduate student to contact the distinguished lady herself uncovered two facts: the inquiry received a prompt reply in French, with a London postmark, and Smith's full name anagrams to "U.R. Blanche Descartes, Limit'd". This quirky humour lasted throughout Smith's life, but never so light-heartedly as in his undergraduate years.

As a conscientious objector he spent the Second World War portering in a hospital. From1946 until his retirement he worked at the Galton Laboratory, University of London, where he became Weldon Professor of Biometry. His early work with J.B.S. Haldane applied the likelihood approach to human linkage, which he extended in highly original papers for the next half-century. Among Smith's achievements is the most powerful test for mimic loci, which produce what appears to be the same disease but are located in different chromosome regions and act in different ways. With James Renwick he pioneered sex-specific analysis based on the observation that chromosomes recombine at different points in male and female meiosis. Smith's 1953 paper introduced autozygosity mapping based on co-inheritance of a rare disease gene and close markers in relatives. The method lay fallow for thirty years, waiting for the molecular markers that have made it invaluable.

Smith's work was especially influential in the United States, which at the time he joined the Galton Laboratory had begun to develop the computers of which Charles Babbage had dreamed a century earlier. This stimulated applications that were unthinkable to Smith's predecessors in Britain, who dominated statistical genetics in the pre-computer era. Smith generously attributed the successes of the early computer period to his American competitors, who less characteristically gave him precedence. The truth is that the two currents were so intermixed and the rivalry so friendly as to baffle a historian of science. Smith's best ideas were incorporated into genetic mapping whereby hundreds of disease genes were localised as a necessary first step to sequencing, characterisation, and attempts to ameliorate their effects. This has been a precious tool for clinical genetics and the impetus for the Human Genome Project.

Although these contributions are best known, they are only a part of Smith's scholarship. He contributed to many of the classical topics in statistical genetics, including segregation ratios in family data, kinship, population structure, assortative mating, genetic correlation, and estimation of gene frequencies. The latter had wide application, but Smith's role has not been recognised by mathematicians. With Ceppellini and Siniscalco he introduced the method of gene counting in 1955. It gives maximum likelihood estimates that converge more slowly but more reliably than methods requiring an information matrix. This principle was presented at the Royal Statistical Society in 1976 by Dempster and others as the EM algorithm. Characteristically Smith contributed to the discussion without mentioning that he had priority of more than 20 years. The method is now used to advantage for many missing-data problems in which some of the observations are mixtures of discrete probabilities.

Smith's eminence in mathematics was recognised by membership in the Royal Statistical Society and the ISl. A large book has been written on "Squared Squares", the name under which the new branch of combinatorics founded by Smith and his small coterie is known in pure mathematics. Human geneticists, and especially genetic epidemiologists, appreciated his contributions to genetic statistics made when that field was nobbled by government policy that protected British computers. For a score of years the sciences that needed competitive computing were stifled, and many of their practitioners changed disciplines or countries. In that difficult period Smith continued his research, oblivious to practical problems. Fortunately his work was valued and the international effort benefited.

Until his wife's death Smith delighted his friends with a Christmas letter commenting on peculiarities of news, railway stations, and other English curiosities. He is survived by his son and remembered with affection by colleagues who overlapped his tenure of the Weldon Professorship and active retirement.

Newton Morton

Top

Back to Home Page

Back to Newsletter