E.F. Allan and R.D. Stern
Statistical Services Centre, University of Reading, Harry Pitt Building
Whiteknights Road,Reading, UK
E-mail: e.f.allan@reading.ac.uk and r.d.stern@reading.ac.uk
Agriculture students at the University of Reading take courses in biometry. There are separate courses for undergraduate and postgraduate students, and they consist of two hours of lectures and two hours of practical work per week during two 10-week terms. This year, for the first time, the postgraduate course has been taught mainly by staff in the Statistical Services Centre (SSC), who all regularly provide consultancy support to their clients during the course of their work. The course was attended by approximately 70 students following programmes in agriculture, horticulture, wildlife management and vegetation surveys.
Previously the course took a fairly traditional approach to statistics. We did not believe that it equipped the postgraduate students adequately with the biometric skills necessary for research work in a life sciences environment. Consequently, we changed the content radically. This paper summarises the changes we have proposed and our initial experiences of teaching `relevant biometric skills' to research students.
Until recently the material on our university course for postgraduate agriculture students could be described as conventional. Descriptive statistics, the Normal distribution and the use of t, Chi-square and F-tests have all been taught in a fairly standard way. Further topics included the analysis of simple designed experiments such as randomised blocks, factorial experiments and split-plot designs and how to do regression analysis. Practical classes either involved tutorial sessions with hand calculators or hands-on computer work
We felt that this format was of limited use to students, as they were not being exposed to the types of situations to which they would find themselves later. Even within their MSc or PhD courses it did not prepare them sufficiently for their subsequent research projects. Too much time was spent on the mechanical application of statistical methods so that students were not:
These views parallel those described by Stern and Allan (this proceedings) in relation to the training strategy for scientists in Guinea. (Similar experiences have been described by Akundabweni (this proceedings) in the teaching of biometrics to MSc students in Nairobi.) In the courses at the University of Reading we have been trying to make the same type of change for university students who are just starting their research career. Our new approach is built around the general objective of introducing the students to the topics needed for research methods. Where appropriate we have included in our teaching materials adaptations of course notes designed by the International Centre for Research in Agroforestry (ICRAF) (see Stern and Allan).
One major change to the course content is to put greater emphasis on the planning of research investigations. With the current trends in agricultural research we believe that students need to know not just the principles of experimental design, namely blocking, replication and randomisation, but also the ideas and concepts of surveys and other field studies, the steps in identifying research objectives and selecting the type of study that best meets these objectives and the decision on which treatments to apply and measurements to make.
We are trying to link together different components of the course. For example, the description of the different types of measurements that can be collected links to the need for more advanced tools to be able to analyse a wide variety of types of data, not merely data from a normal distribution. Another new component is to give prominence to the importance of a good data entry and management strategy. We have found that a researcher's lack of knowledge of how to manage his/her data has contributed to his/her inability to conduct a full and satisfactory analysis.
Topics covered previously remain useful. But the further techniques required in research investigations are also being introduced. A key topic is the collection of data at multiple levels (households within villages, plants within plots, quadrats within sampled areas etc.). In such situations the course addresses the concepts and approaches to data analysis and interpretations of results, rather than the theory. The methods are illustrated using appropriate statistical software.
During the first term (the term just ended) we cover:
During the first half of next term, the agriculture students will follow a different programme from those specialising in wildlife management and vegetation survey. For example, the agriculture students will have sessions on:
These sessions cover topics such as how to lay out an experiment for increased precision, and how to deal with complexities such as repeated measurement data or data collected over multiple levels. The objective is to raise student awareness of the important issues in these different topics and to provide guidance on some simple but sensible approaches to the analysis.
The two groups of students are brought together again for the last quarter of the course. Then, within the framework of modern modelling, students are introduced to the general linear model and shown how these modelling principles can be extended to the analysis of non-normal data.
The practical sessions on the course are varied. Some use statistical games, such as (a) The Rice survey and the Tomato experiment, where the objectives are to encourage sensible strategies for designing studies, and (b) To the Woods and Mice which respectively illustrate the benefits of stratification and blocking. Students also review published papers for the clarity and appropriateness of the statistical content and carry out practical work using Excel, Minitab and Genstat.
On the computing side we make the assumption that students are computer literate, and in particular that they have met Excel before. Most of our agriculture students are from abroad and, where this is not the case, they attend extra sessions in computer literacy organised within the university.
The Statistical Services Centre has recently produced a number of Good Practice Booklets for the Department for International Development (DFID), UK to provide guidance to their researchers in the field. These have been written to describe concepts, approaches and methods of design and analysis in a non-mathematical way. Some of these booklets have been used within our course materials, in particular:
At the half-way point both positive and negative comments have been received. Many staff and students in agriculture have been positively excited by the course, but those in wildlife management and vegetation survey feel that the material is less relevant to them. This may be addressed in term two, which provides more specialisation for the different groups. Also, despite removing much of the theoretical aspects of analysis from the course, students still find statistical analysis difficult to understand. We need to consider how to overcome these problems in future years.
Computers are still not sufficiently integrated into the course structure. We would like to have practical sessions which incorporate any mix of instruction, study design, data collection and analysis and with computers available to assist when required. For this we need lecture/practical rooms with both computers and space for other activities. Currently this is not the case, but we are hopeful that it will change shortly.
After only one term we believe, for the reasons given above and from the reactions we have received, that our new approach to the course is appropriate. It aims to meet the changing approaches to agricultural research as described by Lynam (this proceedings) and achieve the realisation that `Biometrics is an exciting scientific field in which statistical practice and methodology go hand in hand with research in agriculture, biology, medicine etc.' (Janssen, this proceedings).
