I have held several major and minor consulting appointments.

At the University of North Carolina I held a visiting appointment in the L.L. Thurstone Psychometric Laboratory directed by Lyle Jones. My duties were initially to teach the ANOVA part of the statistical methods course for graduate students. Lyle then asked me to teach a graduate course on Test Theory. I said I did not know anything about test theory and he said "Good, then you will be able to teach a fresh course. The current text book by Harold Gulliksen (1950) is very much out of date. Fred Lord and Mel Novick are writing a new modern book on test theory - if you ask them I'm sure they will be happy to give you chapter drafts to teach from." I had already visited the Educational Testing Service ETS in 1965 after an invitation to visit Princeton from John Hartigan, who had finished his PhD there and had given me a Bayes theory course in my third year. I wrote to Fred, and he sent three large volumes of the draft chapters, on condition that we provided him and Mel with feedback. I went through them with the students, and we all learnt a great deal from this, though unfortunately we did not get to Allan Birnbaum's fundamental chapters on item response models, which I did not learn about till much later.

We were critical of the complex and cumbersome notation for random variables and suggested a simpler version, and made a lot of other detailed comments and corrections (which I still have) on the first nine chapters. Fred was very pleased (though he did not take up the notational suggestions) and acknowledged all the contributors who used the early chapters in the preface of the published book. He told me much later that our comments were the most detailed and helpful of any they received. This began a long connection with ETS.

I was also a consultant for Harriet Rheingold, who was the eminent Director of the Child Development Laboratory (http://psychology.unc.edu/gift/hr/front.html). I was happy to do this as part of my appointment, but Harriet insisted on paying me 10% on top of my salary, a concept that was new to me and (obviously) welcome. However I soon found that this gave Harriet the right to require my help on any, and many, problems of data analysis that came up in her Lab. One I recall struggling with was to explain to her staff the age-period-cohort confounding issue in longitudinal survey analysis. As a new PhD and with little practical experience in statistical modelling, I found this a formidable task.

I very much enjoyed working with her, however - she was a remarkably warm and friendly person as well as a formidable research worker. The Psychometric Lab was also a warm and friendly place, due much to Lyle's personality. He offered to extend my visa so we could stay longer, but we decided to return to Australia in 1967.

Macquarie University 1969-76

I moved from the Statistics Department at the University of New South Wales to a Senior Lectureship at Macquarie, a joint appointment in the School of Behavioural Sciences (comprising Psychology, Sociology and Anthropology) and the School of Economic and Financial Studies (comprising Economics, Accounting, Business Administration, Actuarial Science, Demography and Statistics). Although I got on very well with the statistics staff at UNSW (who remain friends to this day), the evening teaching and the car-pooling for the long trip to Kensington from West Pymble (where three of us lived) did not suit me, and when Macquarie (a short bus-ride away in North Ryde) advertised for a statistician in Psychology I applied, on the basis of my 18 months at Chapel Hill and a general interest in psychology. I did not apply for a corresponding position in Statistics as I did not know the new development there (Macquarie began teaching in 1967 when I was in the US). The outcome of my application was a joint appointment (the only one at Macquarie, then or later). At the time the two schools occupied neighbouring floors in one of the large Macquarie buildings, and I had a room in both schools.

My main commitment was to Psychology. In Statistics I taught several courses in Applied Statistics and Linear Models (the latter a joint course for third-year statistics and psychology students, discussed in Teaching interests). The first mathematical statistics course had a heavy combinatorial probability component, needed for the Institute of Actuaries course which the Macquarie syllabus in Actuarial Science followed. This had a discouraging effect on other students, and the enrolment in second and third courses in statistics was generally low during my appointment (though Don McNeil's later appointment to the chair of statistics changed this dramatically).

Psychology was popular with students and classes were large. Statistics was initially handled by Jeff Sutton and Pat Michie; Jeff soon transferred this to me and Pat. I was a general consulting resource for psychology staff and postgraduate students in experimental and survey design. The strong experimental group under George Singer initially dominated but this gradually changed to social psychology over the period of my appointment, especially when Singer moved to Latrobe University in Melbourne, taking several of his research staff. I built up a stock of good study designs and their data which was useful in examples and tests. As the University grew the consulting load increased and Ian Waterhouse, the first head of the School, accepted the need for additional consulting appointments. June Crawford moved from Physics at Sydney University to provide much-needed computing support, and enrolled for a PhD in psychology. Ross Homel was the second statistician appointment, and the consulting group expanded to four by the time I left in 1976. These statistics appointments in Psychology caused considerable strain with the statisticians in Economics, and the limited possibilities for promotion, and recurring bouts of illness, led to my departure. This would have occurred earlier had it not been for the great support from, and great personal relationships with the psychologists, which made this period professionally and personally rewarding.

University of Lancaster 1976-86

My three-year appointment (1976-9) as Social Science Research Council Professorial Fellow in Statistics Applied to the Social Sciences was remarkable for the freedom to do what I thought was important. This guarantees that you work harder than under any other kind of appointment. The basis of this appointment was to improve the applications of statistics in the social sciences in the UK. (A second SSRC appointment a few years later, of John Fox at the City University, had the same remit.)

Early discussions with the social scientists on unbalanced cross-classifications, which were the basic meat of all survey analysis, led to the major discussion paper in JRSSA 1978, referred to in Research_interests. I began a seminar series in statistical modelling for the social science departments, and after a short period, and with help and encouragement from Joe Whittaker, switched from OMNITAB to GLIM, which was then in its second release, for my own research.

This proved of great importance, as following the three years of the Fellowship appointment the University appointed me as permanent Professor of Applied Statistics and established the Centre for Applied Statistics (CAS) under my direction. At the same time the SSRC funded a large research programme under my direction on EM algorithm applications to incomplete data problems, with computations to be developed in GLIM, and a user-friendly GLIM manual to be produced. Over the period 1978-87, I published 26 papers alone or with the research programme visitors and research staff, and Lancaster and UK staff.

• The 1980 Applied Statistics paper with David Clayton establishing the Poisson kernel of the likelihood for right-censored exponential and Weibull distributions (the same results were derived independently and at the same time by Ted Holford in 1980 Biometrics and Nan Laird and Donald Olivier in 1981 JASA).
• The 1981 JRSSA paper with John Hinde and Dorothy Anderson on the analysis of data on teaching styles and pupil progress, which combined latent class and variance component models in a single analysis.
• The 1981 Psychometrika paper with Darrell Bock on the development of an EM algorithm for fitting the 2PP item response model.
• The 1986 JRSSA paper with Nick Longford setting out the efficient analysis of multi-level educational data and giving guidelines for the design and analysis of multi-level educational studies.

The dissemination of modern model-based analysis was developed through intensive CAS short courses in generalized linear modelling applications, with the SSRC research staff of John Hinde and Dorothy Anderson (later replaced by Nick Longford), and Brian Francis as computing support for the CAS. Our experience from these very successful courses was incorporated in the (also successful) book Statistical Modelling in GLIM, with John, Dorothy and Brian, though this was completed (and published 1989) well after the programme had finished.

Tel Aviv University 1990-91

I took over the direction of the Statistical Consulting Unit of the Department of Statistics at Tel Aviv after the previous director resigned due to ill-health. The department had 1.5 tenured consulting positions funded by the Medical School, and many part-time consultants, and nearly all the consulting was for this School. Consulting work from non-Medical School staff included several which led to research subjects for MSc or PhD students. Estimating the size of the bat population in the Tel Aviv Central Bus Station was one which presented interesting problems in capture-recapture analysis; Tzippora Schochet did her MSc on this subject under my direction.

My main consulting work however was outside the Statistical Consulting Unit, with Ruth Zuzovsky in the School of Education. Over several years we published joint papers on the multi-level analysis of Israeli national test data and SISS and TIMSS data international comparative data, in D. Willms and S. Raudenbush (1991), School Effectiveness and School Improvement 1 (1990), School Effectiveness and School Improvement 5 (1994) and International Journal of Educational Policy, Research and Practice 1 (2000).

University of Newcastle-upon-Tyne 1997-2000

I directed the new Statistical Consultancy Service for University staff at Newcastle 1997-2000 with Ross Darnell and Sarah Roberts as consultants. Sarah retired from ill-health during this period and was not replaced: Irit Aitkin and Rob Foxall worked unpaid part-time. The strong engineering and applied science departments provided a number of interesting problems, and discussions with the chemical engineering statistics group led to my supervision of Rob Foxall's PhD in statistics on neural network models Statist. and Computing (2003). Work with Louise Parker in Epidemiology led to two major papers in The Lancet 354 (1999) and J. Roy. Statist. Soc. A 165 (2002) on the effect of radiation on stillbirths in workers at the Sellafield nuclear plant. Consulting work with the group in geography led to a joint paper on the analysis of spatial geographic data Geographical and Environmental Modelling 3 (1999), and I part-supervised the DEd research of a student in Education on the effect of OFSTED inspections on school performance: Brit. Educ. Res. J.29 (2003).

The University expected the Consultancy Service to be self-funding after its initial three years, but this was not achieved. The University already had a well-established commercially-oriented Industrial Statistics Research Unit which was self-funding, but this did not provide advice to University staff, which is inherently less well-funded. My tenure as Director ended and the service was restructured as a commercially oriented service; it eventually closed.

American Institutes for Research 2000-2002

Following my term as Director of the Consultancy Service I took a leave of absence from Newcastle to take up an appointment in Washington DC as Chief Statistician for the Education Statistics Services Institute (ESSI), an arm of the American Institutes for Research which provides consulting and assistance to the National Center for Education Statistics (NCES) of the US Department of Education. My main function was to provide high-level statistical advice to NCES, which I did through personal consulting and two series of brownbag lunch seminars on the model-based approach to large-scale survey analysis. NCES ran the very large NAEP (National Assessment of Educational Progress) surveys which had a complex survey design and a very unusual spiralling balanced incomplete block design (due originally to John Tukey) of test items into test booklets; this was used because the test could not take up substantial time, but the number of items both within and among the school subjects being assessed was very large.

With considerable help from Jon Cohen, the Chief Scientist of AIR, I mastered the different language, algebra and philosophy of the "pure Neyman" approach to survey sampling through the repeated sampling distribution of the sample selection indicators. It was clear that this approach was inconsistent with any model-based approach (even one which did not specify any restrictive parametric model), and this raised my interest in alternative Bayesian approaches which used only the multinomial model and conjugate Dirichlet prior for inference about finite population parameters. It was striking that the full Bayesian analysis using this approach had been given in 1968 by Jon Rao and H.O. Hartley, and reinvented by Don Rubin as the Bayesian bootstrap in 1981. It seemed to have never been implemented practically. I later extended the Bayesian bootstrap approach very generally to handle clustering, stratification and regression problems, using only the multinomial/Dirichlet model and prior, and gave several examples J. Official Statistics 24 (2008).

My seminars were well received by the large majority of ESSI and NCES staff, but my modelling philosophy brought me into conflict with the current analysis paradigm and philosophy in NCES. The University of Newcastle would not extend my leave beyond two years and so I returned to Newcastle in 2002. The Acting Commissioner of NCES funded two research projects which I worked on in Newcastle, and subsequently this contract work expanded in a different direction, to examine the model-based analysis of the NAEP surveys. This continued in Melbourne with Irit Aitkin after my early retirement from Newcastle, and led to a series of major research reports, joint with Irit, on the full model-based analysis of these surveys (1986 and 2005 primary mathematics surveys were used as real examples). The reports have been integrated into a major monograph on the model-based approach to these large-scale surveys: Murray and Irit Aitkin Statistical Modeling of the National Assessment of Educational Progress Springer 2011.

A summary of each project, and its importance for the programme of research we were following, can be found here.