4. STATISTICS IN THE REAL WORLD Ann Gould Shell Research Ltd, Sittingbourne, Kent
4.1 Introduction
There is a dearth of good statisticians in the job market; recruiting gives us great difficulty at Shell. I am sure that a large part of the problem lies in the fact that students, both at school and at university, are unaware of the range of interesting work there is to be done: people - perhaps even some teachers - think of statistics in industry as piles of numbers added together or, if you're lucky, averaged. They probably have little idea of the interaction with scientists which is involved in good consultancy work or the challenges that complex experiments and studies can present.
My objective in this paper is to demonstrate how varied and interesting statistics in industrial research can be. I will tell you about the Mathematics and Statistics group at Shell Research and the facilities we have at our disposal; then I will give you a broad picture of the range and variety of work we cover. My colleague, Simon Pack, gives in his paper an example of a rather interesting problem he has dealt with recently.
The group at present (time of writing: September 1992) at Shell Research consists of ten permanent members of staff and a sandwich student. The staff all have at least a first degree in statistics; most have an MSc; four have a PhD. We work together effectively, always feeling that we can knock on one another's doors when we need to discuss a problem. We have regular group meetings where a member of the group will present some recent work and lively discussion usually ensues.
4.2 Tools of the trade
We work both on mainframe computers (where large databases are involved) and on PCs. The statistical package we use most is SAS but sometimes GLIM or NAG routines are more appropriate. We have several in-house packages, some of which are written in GENSTAT, for routine tasks. This takes care automatically of the masses of analyses of variance that inevitably need to be dealt with. For word processing and presentation work we are moving to MicroSoft windows products.
4.3 Science at Sittingbourne
The laboratory was established to discover and develop crop protection chemicals: insecticides, fungicides and weed killers. Originally these were oil by-products but nowadays they are solely synthetic.
The process of development of a new compound involves firstly the chemists. They synthesise many compounds for testing. Statistics can help them even at this early stage with computer-aided molecular design. Using a set of compounds whose effects are known, mathematical relationships can be found which help to guide the chemist to make compounds to do particular jobs. This analysis is particularly challenging. There are very many quantities used to describe the compounds, so complex multivariate methods are used.
The next stage is known as screening. The compounds are applied to plants, insects and fungi to see if they will have the desired effect. Several thousand compounds are screened each year and the design of the screens has to be very efficient as well as effective in finding the right compounds: a type of operations research problem.
If a compound appears to be useful, a whole range of tests for safety have to be done. These look for toxicity and carcinogenic potential in many organisms, both in the laboratory and in the field. Toxicity can be very complex, particularly in the environment. A good example of the kinds of tests that are done is one involving a rodenticide. The pellets are distributed in farmyards in the hope that rats and mice will eat them. However should the rodent then be eaten by an owl, it may be that the bird is also poisoned. There are many statutory tests that have to be done to ensure that all these effects are investigated. At every stage the statistician is involved in the design and analysis of the experiment.
The other main branch of research at Sittingbourne is in chemical additives for petrol, diesel and automotive (e.g. brake and gear) fluids. The overall approach is similar to that used for agro-chemicals but the tests for efficacy are rather different and often involve multifactorial experiments. Various experimental designs are used.
4.4 Conclusion
I have given you a brief overview of our work and I hope that I have put Simon's paper in context. I also hope that I have shown what a wide variety of work goes on at Shell Research and that you will pass the great news on to your students. Perhaps they will go on to study statistics with enthusiasm and then come and work with us.