5. A CASE STUDY IN ENVIRONMENTAL STATISTICS Simon Pack Shell Research Ltd., Sittingbourne, Kent
5.1 Introduction
Applied statisticians need a variety of skills to function effectively. My recent involvement with the development of an international ecotoxicological testing guideline will serve to highlight some of the skills that I think statisticians need to have. There are obvious implications for those involved in the teaching of statistics.
5.2 Background
Pesticide residues, the quality of drinking water, toxic waste and oil spills are just some of the environmental issues of current widespread concern. Media news reports frequently cover these topics, both raising awareness and maintaining public interest. Correspondingly there has been an increase in pressure on governments in recent years to act to protect both local and global environments.
It is clearly important that decision makers are well informed. For advice and assistance they often turn to industry, academia and independent research organisations for help in monitoring, quantifying and assessing the impact of man's activities. Part of this effort necessarily involves a considerable amount of scientific experimentation and at a considerable cost.
Statisticians have a central role to play in this enterprise since it is not only vital that unambiguous and reliable data are collected, analysed and interpreted but also that this is done in a cost-effective way. It is the policy of Shell companies to conduct their businesses with proper regard for the conservation of the environment, seeking continuous improvements in their environmental performance. In support of this, Shell Research is constantly looking at ways of eliminating or minimising any detrimental effects of Shell's operations. It also provides experimental assessments in the areas of ecotoxicology and environmental fate.
Ecotoxicology is generally concerned with possible toxic effects on wildlife, both terrestrial and aquatic. The physical, chemical and biochemical breakdown and movement of chemicals in soil and water are involved in studying a chemical's environmental fate.
5.3 The Case Study
Routine ecotoxicology experiments to quantify effects are usually carried out following national or international testing guidelines. One international body producing guidelines is the OECD (Organisation for Economic Co-operation and Development). Recently they have placed increasing emphasis on the use of statistics in the area of aquatic toxicology, ie effects on fish and other aquatic organisms. An example of this was the involvement of statisticians in the development of a new guideline designed to assess chemical effects on fish growth. Statisticians from Shell Research (Sittingbourne, Kent) and the Water Research Council (Medmenham, Berkshire) were asked by OECD to investigate all the statistical aspects of the test.
The Fish Growth Test is very simple in concept. Separate groups of juvenile fish are exposed to the test chemical in their tank water at one of a range of concentrations. The fish are weighed at the start and end of the experiment (28 days). Significant reductions in growth, relative to unexposed control fish, may then be concluded to be evidence of chemical effect. Concentration-related trends in growth reduction are also sought.
The statistical investigations covered many issues including the definition of an experimental unit and the question of pseudo-replication. This is important since fish are usually held together in tanks. Therefore the tank is, strictly speaking, the experimental unit. The standard problems of the level of replication and power of the statistical analyses were also studied. This was facilitated by a large amount of data that had been collected by collaborating companies and organisations following a trial of an early draft of the test guideline. Thus the primary sources of variability could be identified and quantified. The type of analysis was also important. The two main options are analysis of variance, followed by treatment-control comparisons, and modelling of the concentration-effect curve.
The findings and recommendations of the statisticians were presented in a set of reports and at a multinational meeting of aquatic toxicologists and those involved in the test guideline production. Many of the recommendations have been incorporated into a revised guideline that will ultimately be adopted by OECD member countries.
5.4 Conclusions
This project did not involve very advanced statistical methods. The most important aspects were the interactions with non-statisticians, the discussions, the negotiations and the ways in which the findings were presented.
The key point for those involved in the teaching of statistics is that technical ability is not sufficient to make someone an effective applied statistician. Technical ability makes up only half of the job.
A good breadth and depth of knowledge is, of course, necessary since many problems cannot be conveniently 'pigeon-holed' into one standard statistical procedure and often a number of techniques need to be tried or combined. Another attribute for success is the ability to develop a working understanding of a subject area. This enables meaningful discussions to take place at a reasonably common level between the statistician and the 'client'. This is very important if a satisfactory and practical solution is to be found. In science this means, at least, being able to get to grips with scientific principles and jargon. A real interest in the subject matter clearly helps enormously.
Perhaps the most important facet of the job is the possession of good interpersonal skills. This means being able to write clearly and concisely, with the correct amount of statistical detail for the intended readership and having the ability to present work orally to statisticians, technical experts, managers or any other non-statistical audience. Less obvious is the need to be able to listen and question effectively so as to find out about all relevant aspects of a piece of work. Influencing and negotiation skills should also be part of the statistician's 'tool kit'.
The above points may only all be met by a 'statistical giant'! The challenge is how to try and equip statisticians with as many of these skills as possible. The interpersonal side is hardly covered at all by schools and universities at present and on-the-job training and coaching must suffice. Collaboration between the 'providers' of statisticians and prospective employers must be the best way forward. Any initiatives in this area deserve support from all sides.