Notes on the MEANS Workshop held in Glasgow

20th February 1997

Participants Phil Crook (ODA), Kirsty Davidson (Napier University), Glyn Davis (University of Teeside), David Edelman (Royal Bank of Scotland), Laurence Ewan (Quintiles), Anne Hawkins (RSS Centre for Statistics), Lynne Holton (Astra Clinical Research Unit), Peter Holmes (MEANS Project), Iain Knox (Royal Bank of Scotland), Cecilia MacIntyre (Scottish House Condition Survey), Zamalia Mahmud (University of Strathclyde PhD student), John McColl (University of Glasgow), Maria Melling (The Scottish Office), Bob Peacock (Astra Clinical Research Unit), Trevor Ringrose (University of Aberdeen), Prof Peter Sprent (Retired), Stuart Young (University of Glasgow).

Notes

The meeting was organised by John McColl of the University of Glasgow and held in the Melville Room of the Gilbert Scott Building. He and Peter Holmes welcomed the delegates and thanked them for coming.

The main purpose of the meeting was to obtain insights into the problems of matching education, assessment and employment needs in statistics from the different perspectives of those attending the seminar.

Peter Holmes and Anne Hawkins summarised the work done so far by the project and drew attention to the newsletter. Each delegate then gave a short presentation which was followed by discussion.

Glyn Davies spoke from his experience as senior lecturer in IT and Quantitative Methods within the school of Business Management. Students there generally do not like number crunching so the emphasis now is on the ability to take and use data, use commercial software and then communicate to non-statisticians. Students work on individual and group reports - if courses requiring this are optional then students do not choose them. They have tried to find out what employers want and do not get much demand for higher level techniques. Most statistics at Teeside is done in non-specialist courses. There is a fear of the subject - do we need a new name? In discussion it was suggested that the department might get feedback from students who had graduated; they and staff from the 'home' departments might be used on the courses.

Zamalia Mahmud is doing research on statistical education in Scotland. She sees problems of students coming on courses from a wide variety of backgrounds with a general dislike of the subject which they see as largely irrelevant. There is a need to educate employers on the role of statistics.

Stuart Young is responsible for the electronic side of the CTI in Statistics and is list manager for Allstat. He sees the need for universities to take a problem based approach to teaching and to put statistics as part of a problem solving process. Increasingly, with growing computer power, there is a need to emphasise interpretation rather than calculation. Statistics only stands with its applications. He queried the value of the 10 week service course - the main thing that people should learn from this is to consult a statistician.

Employers may need educating to employ statisticians, and not think they can do it all with their computing packages. Students need to learn how to use packages - though they should be able to transfer from one used at university to a different one in industry if necessary. This would be helped if there were an element of programming in the university course. This would also promote logical thought.

Discussion. If we are to teach appreciation of statistics to user departments then we need staff who can do it. Such people may or may not be found in each department and it is unreasonable to expect staff in the statistics department to have a deep knowledge of all the contexts. It is not possible to divorce appreciation completely from understanding and the ability to do some statistics. We need to discourage the idea that a package will solve all statistical problems. The timing of service courses is important; in many cases they are too early. Reference was made to John Whitehead's courses in medical statistics run at the University of Reading. He also goes to specific companies and gives tailor made in-house courses.

Lynne Holton recently graduated from the University of Glasgow and all her work is now connected with clinical trials. She works as a lone statistician on a day to day basis, dealing with experts in many other disciplines. She thinks a good theoretical background is essential in the undergraduate course. This she had. It was the practicalities of the job that were missing. These include: discussion with others; having to use SAS; communication skills; why statistics is useful to solve problems. In her undergraduate course she did two pseudo-real problems with other departments. These were valuable in giving insight. They were not real enough; they were still too tidy and paid insufficient attention to the non-statistical issues.

As a statistician she needs to understand the biology but not make biological/clinical decisions.

Statistics has problems with a bad press. The 10 week service course does not help here. Statisticians need to learn to communicate with non-statisticians; they need more real problems as part of their undergraduate course.

Discussion. Could we use external people to bring in real problems? Could the statistician become involved in, say, the biology practicals? Could the statisticians from PSI become involved in pharmacy courses? There are real practical difficulties in building cross-subject co-operation into undergraduate practicals. There are also associated assessment problems - but the experience itself is very important.

Kirsty Davidson started work in the Scottish Office which required versatility and many transferable skills. She is now at Napier University lecturing mainly on service courses. The major problems here are motivation, attendance and modularization. Statistical methods need to be a core part of the course - if they are optional they will be avoided. Areas of good practice include written communication; group projects with peer assessment; case studies; operational research output interpreted; 5 minute interviews on interpreting computer output. She has had students visit the Scottish Office and the General Registrar's Office to help students learn the practicalities. She has also had external speakers from pharmacy etc. The problem of motivation is strongly linked with assessment. Visits have to be written up as an assessed report. She uses real life problems e.g. Napier customer satisfaction for catering. 2000 interviews were given to MSc students to check, analyse etc. Another project was on performance indicators. There was a landscape project with employers and graduates; they come in to assess students. Business students use Excel. They use TLTP STEPS material in teaching. They teach research and consultancy on MSc courses.

Discussion. In the pharmaceutical industry they have to use validated packages in the investigation, hence the use of SAS. Excel would not be good enough.

John McColl lectures at the University of Glasgow. There is a difference between specialist statistics courses and service statistics courses. Statistics and mathematics graduates have a wide range of possible careers. The main ones are research, medical, actuarial, government, pharmacy and teaching. The specific requirements in these differ so the department needs to look at the general skills needed. So they look for a strong theoretical foundation. Statistics graduates need to do much more in statistics than, say, the psychologist. But it is also important to emphasise the applied side, so they do project work, laboratory work and work in teams to solve problems. In some specialist areas they have developed specialist courses. Psychology includes one third of a year in statistics. They use a computer package, emphasise collection, design and interpretation. There is a problem that no text books take this emphasis. Universities should be teaching generally how to use a statistical package. The employer is responsible for the specialities.

Discussion. What should be included in the theoretical foundation course. How much is proof important. For example should all be able to prove the Central Limit Theorem or should they know in practice when the CLT will give good enough results?

Laurence Ewan As an employer he is looking for a basic understanding of statistics and the ability to explain simple concepts. He wishes his graduates to know about design of experiments and modelling and data-fitting. They need to be computer literate with word processing skills and know about systems and network communications. They will work with a group which includes IT people and the clients. They need good reporting skills and commonsense. They need to be able to think round a problem and use their time well. Over the years students seem to be graduating with fewer programming skills. (Fortran was good for teaching logical thinking.)

Group work modules have been found useful by young employees. Work placement students fit in well. Laurence did a degree including practical statistics. The group work was assessed; the overall mark was shared within the group. A logbook was kept of the group's meetings and this was seen by the lecturer.

Students need to be taught to look at the data before analysis and plot before any inferences. Any training on the job has to have a short term pay off or the experience will be used to get promotion elsewhere.

Discussion. In some areas, such as pharmacy, experience and a Masters degree are expected for the first job!! Large companies find it easier to put on their own training schemes than do small companies. One large employer was reported as looking for lifetime development.

Maria Melling works for the Scottish Office in a group of four statisticians and five assistant statisticians. Assessment on entry requires the candidate to explain things to statisticians and show they can understand data. They are assessed for quick thinking and the ability to hold an argument. A statistician needs to be flexible to move round different departments. They are expected quickly to pick up the background in each department and also quickly to take on responsibility. Maria also gets involved with policy decisions. At times there is a need to dig to find what colleagues actually want - this was not part of her training. You may have to cope with people who do not like your conclusions, so at times there is a need to stand up for professional integrity and here a code of practice can help A weakness in her own undergraduate course was that she was not taught SAS or any package. There are in-service courses within the Civil Service to fill gaps and develop new skills.

Discussion. At times you may have to hold out for what you can NOT say from the data as well as what you can. A good course in consultancy should do this.

Cecilia MacIntyre works for Scottish Homes as a statistician with one assistant. When advertising for the assistant they were looking for intelligence, fast thinking, ability to do some boring work sorting data files, working together in a team to produce a report, not easily side-tracked. They found someone with a mathematics/statistics degree who had done an external project one Christmas.

Before working for Scottish Homes, Cecilia had lectured to medical students. She had to use students to help with projects in the summer each taking about two weeks and using a statistical package.

Although she had a practical component in her MSc she learned properly how to deal with real (unclean) data when she started work. She learned a lot by absorbing ideas from colleagues.

Employers need to appreciate that statisticians have skills to offer. It would be helpful for undergraduate statisticians to see themselves as part of the whole research process.

Discussion. John McColl gives his students 3-4 months outside placement work between the third and fourth years. It is possible that the Office of National Statistics could use summer students like this. David Edelman used something like this scheme. Cecilia MacIntyre has a large amount of data which could be used; it would be helpful to have such people looking at it. The Government Statistical Service runs a sort of probationary year which has similarities to university work placement schemes.

David Edelmen What employers are looking for depends on whether they are employing for a job or a career. If for a job, and the person is working in a team, then an employer can cope with weaknesses if they can be developed. Employers may be looking for numerate/scientific skills not just statistical.

It is essential in an undergraduate course for students to look at data. As an undergraduate he had 6 hours of labs for two years which was very helpful. In the assessment exercise he was given 3 hours to say what he would do to analyse a set of data and was then given 3 days in which to do it. Any change in plan had to be justified. He also had to collect his own data which taught him a lot.

Report writing needs to be taught separately from assessment and could be part of an undergraduate course, thought it is not clear at which stage. It should not be optional.

Discussion. John McColl reported that at the University of Glasgow, as part of the statistics course, four times a year students had to do a practical, report on it and receive feedback. Report writing skills were developed in the context of the data with not too much detail. The Government Statistical Service course for Africans insists on report writing, including an appropriate level of approximation and a proper use of English. Eurostat has produced a report on how to write a statistical report for press releases. It really would help if universities trained people to write clearly.

Bob Peacock works in clinical research. The unit employed its first statistician in 1986 and as it grew he joined in 1991 from the Open University with a multidisciplinary background. His MSc was largely useless - his job was a big learning experience. Clinical Research is to assist drug development through to drug licensing and is all heavily legislated. A university course needs to raise the idea of constraints like this. The context is of randomised, controlled clinical trials and often the non-statistical aspects are difficult. For large data sets he uses SAS because it is validated software. The data analysis is followed by a report. Protocols have to be reviewed and he has to advise on design, sample size, power, analysis and plans. There is both a statistical report and a clinical report. He has to collaborate with statistical and medical colleagues for presentation. He also has to follow up queries. More and more nowadays recruitment is from graduates with an MSc in medical statistics and two to three years experience. They look for someone who can communicate, write and program and who also has interpersonal skills. Recent recruitment has been from the University of Reading and de Montfort University. Some come through in-house promotion.

Trevor Ringrose lectures at the University of Aberdeen. Aberdeen has a placement programme for four days a week for the first term in the fourth year. This counts as 2 to 3 courses. All is done in the local area and supervised from the department. Suppliers have included BIOSS, medical centres, hospitals. Students are given feedback during their period of placement. They also have team bonding exercises at weekends. Modularisation is posing problems of slicing up knowledge into too small slices. It is difficult to use many real examples in a course - especially when the course is tied to specific methods. Should data analysis be separated from methodological courses? Students can learn by having to explain. Could we do this in an undergraduate course by having groups of, say, five explain to each other. We might have to disentangle this from assessment.

Logical thought is an important of transferable skills; students pick up some during the course but do need practice. They also had help in framing a question statistically and in seeing the importance of asking the right questions. One possibility for service courses is to have students read papers and present to others.

Discussion. The practicalities of the 4 day a week placement were questioned. It works reasonably well because there are only a few students involved.

Phil Crook works for the Government Statistical Service. His degree was in Pure Mathematics, Statistics and Geography followed by an MSc in Statistics. This had an emphasis on Bayesian statistics which was not much use in the GSS. His first job in the Ministry of Defence was modelling the Navy Stock Control System. He employed students there to do some of the work.

He them moved to the Seychelles as one of two statisticians. For 6½ years he did all the non-economic statistics. He designed the questionnaire, duplicated it, went out and trained enumerators to administer it. He learned how to deal with practical problems in small area statistics such as when the tomato production figures depended entirely on the sole non-cooperative tomato grower.

He then moved to working on manpower statistics in the navy and is now with ODA developing statistics projects for Developing Countries.

In a statistician he is looking for numeracy, literacy, logical thinking based on a statistical background and the ability to get into the mind of other professions. Possibly a background in philosophy and language would help. [He would have found it useful if, as a student, he had been made to write more and speak in public.] Statisticians need to be able to look at data, the shape of the data and need to do some hand calculations. It is not possible at the undergraduate level to cater for all possible careers. ODA is now working in a more interdisciplinary way and statisticians have to show they have skills to offer others. GSS recruitment is about to be based on competencies and entrants will be assessed against competency skills. The GSS does have a University Liaison Service.

Iain Knox recently started work for the Royal Bank of Scotland. As an undergraduate he would not have chosen statistics but his course included it. He then did an MSc which covered most areas of statistics and included report writing and dealing with people. By specialising in statistics he found there were many fields to choose to enter. University courses therefore have to concentrate on giving general skills.

In the Bank he has to work quickly to sharp deadlines. Early on he had to learn the basics of SAS and handling large data sets. His boss is looking for someone who thinks clearly. His area of work is on behavioural scoring systems, how they work and how the model is derived. He monitors the system giving feedback. He has needed general skills, including consultancy skills to find out genuinely what the customer wants.

Discussion. If students don't learn theory at the university it will not get learned. You may only appreciate the significance of theory after you start work.

Peter Sprent is now retired. Universities can make theory more attractive by broadening courses to include applications. Regression should include some examples with dirty data so that students can learn to deal with it. Simple examples can be used to show what is going on; maybe bring in ideas rather than detail. Student awareness can be raised by using interesting examples. The difference between statistically significant and practically important is not sufficiently emphasised; neither is the problem of the power of a test and its effect on significance. Paradoxes can be used to make people think.

His own background was probably too theoretical and he had to develop practical skills in an applied statistics job on fruit trees which posed particular problems for design of experiments. The good basic grounding enabled him to pick up the specific skills.

As an external examiner he has seen university courses which differ markedly from each other. One is highly theoretical and one highly practical. But it is not obvious which is better when it comes to picking up new techniques.


These pages are maintained by Peter Holmes. Please email peter.holmes@ntu.ac.uk with any comments or corrections.