I am sorry to announce that John Nelder died on Saturday 7th August in Luton & Dunstable Hospital, where he was recovering from a fall. John was very active even at the age of 85, and retained the strong interest in our work – and statistics generally – that we will all remember with deep affection. On 13 May 2010 I took him to the Numerical Algorithms Group’s 40th Anniversary Celebration near Oxford, where he was pleased to catch up with many old friends from his time on their Technical Policy Committee. Then, on 4 July 2010, he entertained Youngjo Lee, Yudi Pawitan, Mike Kenward, James Roger and myself to lunch – and to some challenging statistical discussions. However, he was becoming increasingly frail and it was a shock but perhaps, in retrospect, not a surprise to hear that he had died peacefully in his sleep.
If you would like to leave messages of condolence below, I will pass them on to his family.
John Ashworth Nelder was one of the most influential statisticians of his generation, whose work will continue to have an important and widespread effect on statistical analysis.
John was born on 8 October 1924 in Dulverton, Somerset, UK. He was educated at Blundell’s School and at Sidney Sussex College, Cambridge where he read Mathematics (interrupted by war service in the RAF) from 1942-8, and then took the Diploma in Mathematical Statistics.
Most of John’s formal career was spent as a statistician in the UK Agricultural Research Service, later renamed Agricultural and Food Research Service (AFRC), and now Biotechnology and Biological Sciences Research Council (BBSRC). His first job, from October 1949, was at the newly set-up Vegetable Research Station, later renamed National Vegetable Research Station (NVRS), and now Horticultural Research International, Wellesbourne. Then, in 1968, he became Head of the Statistics Department at Rothamsted, and continued there until his first retirement in 1984. The role of statistician in AFRC was very conducive for John, not only because of his strong interests in biology (and especially ornithology), but also because it allowed him to display his outstanding skill of developing new statistical theory to solve real biological problems. So, for example at NVRS, John developed the theory of general balance to provide a unifying framework for the wide range of designs that are needed in agricultural research (see Nelder 1965). Then, at Rothamsted, he developed the theory of generalized linear models with the late Robert Wedderburn, to overcome the problems of analysing response variables like counts and proportions that do not come from Normal distributions; see the citation classic Nelder & Wedderburn (1972) or the book by McCullagh & Nelder (1989).
This idea of directing statistical research at real biological problems began with the two earlier Heads of Statistics at Rothamsted, R.A Fisher and F. Yates, to whom John became such a worthy successor. However, John emphasized an important additional aspect, namely that the new theory should be implemented in widely-distributed statistical software to enable it to become widely used in practice.
The initial aim for John’s first statistical program, GenStat, was to provide analysis of variance for generally balanced designs. The underlying ideas took shape in 1965–1966 when John visited the Waite Institute of the University of Adelaide to work with Graham Wilkinson, who was then on secondment there from CSIRO (Commonwealth Scientific and Industrial Research Organization). More intensive development began in 1968 when John joined Rothamsted, and the wider statistical and computing expertise available at Rothamsted allowed him to develop GenStat as a truly general-purpose statistical system. GenStat continues in widespread use today, and is distributed by VSN International to users in more than 120 countries. I was honoured to take over leadership of the GenStat development in 1985, after John’s retirement from Rothamsted, and glad that John continued as an enthusiastic (although sometimes critical!) user. GenStat embodied John’s originally-very-novel view that statistical programs should provide a programming environment for the development of new methodology. The procedure structure introduced in 1987 allowed the resulting new programs to be added to the system as new commands. A good example is the suite of procedures for Lee & Nelder’s hierarchical generalized linear models (see below).
John’s other two major contributions to statistical computing came about while he was Chairman of the Royal Statistical Society’s Working Party on Statistical Computing (1967-1984). The first, in 1968, was the Applied Statistics Algorithms, which aimed to support good computing practice by providing implementations of the basic building blocks of a statistical program. Later much more complicated techniques were added, and the publication of an algorithm for a new piece of methodology became an equally valid (and perhaps more effective) way of registering a new idea. The second contribution was the program GLIM which first appeared in 1974, with 4 further releases up to the final GLIM4 in 1993. This implemented Nelder & Wedderburn’s generalized linear models, and led to a dramatic improvement in the quality of statistical analysis allowing unsatisfactory approximate analyses, such as those involving the angular transformation of percentage data, to be discarded. It had an immense influence on the new generation of practical statisticians. For many it provided their first experience of analysing data interactively. It encouraged them to think about each data set, instead of directing it at a black box with a request for “statistics all”. It provided opportunities to investigate a rich set of models, and good diagnostics to assess which one would be most appropriate.
John retired from Rothamsted in 1984 at the age of 60, but continued his research at Imperial College (of Science, Technology, & Medicine, London) where, since 1972, he had been a Visiting Professor. He retired from Imperial College in October 2009. His first task there was to lead the GLIMPSE project (Nelder 1991), which was funded by the UK Government’s Alvey programme to produce a knowledge-based front-end for GLIM. The GLIMPSE system provided advice on data validation, data exploration and model selection. However, it seems to have been intended more as a guide for experts, than as a system to provide expert help to novices, and it never achieved the widespread use that it deserved. However, it contained many very interesting and far-sighted ideas and, when it was released in 1989, it was one of the first statistical expert systems to be made available commercially – and perhaps one of the few to deliver what the originators had promised.
John’s other major activity at Imperial College was his collaboration with Youngjo Lee to develop the theory of hierarchical generalized linear models (HGLMs); see the papers by Lee & Nelder (1996, 2001, 2006) and the book by Lee, Nelder & Pawitan (2006). The 1996 and 2006 papers were presented as “read papers” at meetings of the Royal Statistical Society; it is impressive to note that John was 81 years old when he and Youngjo presented the 2006 paper. HGLMs aimed to provide satisfactory methods of analysis for non-Normal data when there is more than one source of random variation. John viewed generalized linear models as a way of liberating statisticians from the “tyranny” of the Normal distribution, and was a little bemused to see this same tyranny reestablished in methods that were devised initially to extend generalized linear models. These generalized linear mixed models (GLMMs) catered for additional random variation by adding additional Normally-distributed random effects into the linear model of the generalized linear model. John and Youngjo’s new HGLMs extended the methodology to include the beta-binomial, gamma and inverse-gamma distributions, and showed that the conjugate HGLMs (namely binomial GLM with additional beta-binomial random effects, or Poisson with gamma, or gamma with inverse gamma) had attractive advantages in their mathematical theory, computing algorithms and philosophical interpretation. HGLMs can be fitted very efficiently by two interlinked generalized linear models. So we have access to a familiar repertoire of model checking techniques, and can base our choice of error distributions on the data rather than on prejudice or software limitations. Furthermore the analysis can still be carried out interactively – always a very important consideration for John.
With John’s many achievements in statistics, it is important not to forget his other interests. He shared a keen interest in gardening with his wife Mary (nee Hawkes), whom he met and married in 1955 while he was at NVRS; they have a son Jan and a daughter Rosalind. John and Mary were also keen birdwatchers, and were two of the three finders of Britain’s first Siberian Thrush (Zoothera sibirica); see Andrew, Nelder & Hawkes (1955). John was very proud of his other paper in British Birds (Nelder 1962), which gave a rigorous statistical assessment of the implausibility of the “Hastings Rarities” and provided convincing evidence for their subsequent removal from the British List. Finally he was a very keen musician and a virtuoso piano player, and his musical soirees at his house in Redbourn will be remembered by the attendees with lasting pleasure.
John received many honours during his career. He had a DSc from University of Birmingham, and received an honorary DSc. from Universite Paul Sabatier, Toulouse, in 1981. He was also elected a Fellow of the Royal Society in 1981. He was President of the International Biometric Society from 1978-1979, and was made an Honorary Life Member in 2006. He was President of the Royal Statistical Society from 1985-1986, and was awarded Guy Medals of the Society in Silver in 1977, and in Gold in 2005. He wrote three books and over 120 papers in statistical and biological journals, including two citation classics: the Nelder & Wedderburn (1972) paper on generalized linear models already mentioned, and his paper written with Roger Mead while at NVRS describing their now very widely-used adaptive simplex optimization algorithm (see Nelder & Mead 1965).
More important perhaps is his statistical legacy of general balance, generalized linear models, hierarchical general linear models – and GenStat – which will keep him always in our thoughts.
Andrew, D.G., Nelder, J.A. & Hawkes, M. 1955. Siberian Thrush on the Isle of May: a new British bird. British Birds, 48, 21-25.
Lee, Y., & Nelder, J.A. (1996). Hierarchical generalized linear models (with discussion). Journal of the Royal Statistical Society, Series B, 58, 619-678.
Lee, Y., & Nelder, J.A. (2001). Hierarchical generalized linear models: a synthesis of generalised linear models, random-effect models and structured dispersions. Biometrika, 88, 987-1006.
Lee, Y. & Nelder, J.A. (2006). Double hierarchical generalized linear models (with discussion). Appl. Statist., 55, 139-185.
Lee, Y., Nelder, J.A. & Pawitan, Y., (2006). Generalized Linear Models with Random Effects: Unified Analysis via H-likelihood. CRC Press, London.
McCullagh, P. & Nelder, J.A. (1989). Generalized Linear Models (second edition). Chapman & Hall, London.
Nelder, J.A. (1962). A statistical examination of the Hastings Rarities. British Birds, 55, 283-298.
Nelder, J.A. (1965a). The analysis of randomized experiments with orthogonal block structure. I Block structure and the null analysis of variance. Proceedings of the Royal Society, Series A, 283, 147–162.
Nelder, J.A. (1965b). II Treatment structure and the general analysis of variance. Proceedings of the Royal Society, Series A, 283, 163–178.
Nelder, J.A. & Mead, R. (1965). A simplex method for function minimization. Computer Journal, 7, 303-333.
Nelder, J.A. (1991). GLIMPSE, a knowledge-based front end for GLIM. In: IMA Volume in Mathematics and its Applications 36: Computing and Graphics in Statistics (Ed. A. Buja & P.A. Tukey), pp. 125–131. New York: Springer Verlag.
Nelder, J.A. & Wedderburn, R.W.M. (1972). Generalized linear models. Journal of the Royal Statistical Society, Series A, 135, 370–384.
Payne, R. (2004). Algorithms, data structures and languages — the computational ingredients for innovative analysis. In: Methods and Models in Statistics — In Honour of Professor John Nelder, FRS. (Ed. N. Adams, M. Crowder, D.J. H and & D. Stephens), 95-118. London: Imperial College Press.
Senn, S. (2003). A Conversation with John Nelder. Statistical Science, 18, 118–131.