Planning and Managing Agricultural and Ecological Experiments

Peter Johnstone

• Publisher: Routledge
• Year: 2013
• ISBN:
  9781317856399
• Edition: 1
• Topic: Physical Sciences
• Subtopic: Geology & Earth Sciences

A text addressing the essential issues required to undertake satisfactory comparative agricultural and ecological experiments. It offers an integrated presentation, with the focus strongly placed on the planning and execution of experiments.

Experimental intentions

1

1.1 NATURE OF EXPERIMENTS

An experiment is a deliberate action or procedure, frequently referred to as applying a treatment, which is undertaken with the intention of provoking a response which the experimenter measures. However, the measured response is often difficult to interpret. This is because most experimental measurements are made up of two components. The first component is the real or true response to the applied treatment or combination of treatments. That is, it is the response if the treatment were precisely applied and the response was measured without error on experimental material which was absolutely representative of the population to which the results were to apply. The second component is uncontrolled variation and is the composite effect of the three potentially perturbing influences which are assumed to be absent from the true response. In a field experiment the uncontrolled variation is the composite effect of imprecision in applying the treatment, the effect of environmental factors such as uneven fertility, and the effect of imprecision in measurements. Similarly, in an animal experiment the uncontrolled variation is the variation in the individual’s ability to respond, together with imprecision in applying the treatment and subsequent measurements. The uncontrolled variation masks the true response to the applied treatment in the same way as haze masks the true identity of landscape features. Consequently an observer is able to measure the overall effect, which is the sum of the two components, but is unable to measure the true response directly.

Experiments in which the magnitude of the real or true treatment responses are comparable to the magnitude of the uncontrolled variation are of particular importance and are the subject of this book. These experiments typically come from agriculture, ecology, medicine and industry. Statistical design techniques, when combined with statistical analytical techniques, enable an experimenter efficiently to separate and estimate the magnitude of the two components and so facilitate the drawing of defensible conclusions from the experimental measurements.

The classical and fundamental ideas of statistical design involve replicating the experimental treatments, allocating the treatments to the experimental material at random and minimizing the influence of uncontrolled variation. These will be referred to throughout the book, although Chapter 5 concentrates rather more on some specific issues concerning these aspects.

1.2 EXPERIMENTAL INTENTIONS

Before sensible experimental design and analysis strategies can be decided upon it is necessary to be specific about the purpose of a proposed investigation. One of the most eloquent pleas for this course of action was from Wilson (1974):

There can be no greater waste of time and money than to embark on intricate programmes of applied research before one has a clear idea of the precise course of action one intends to embrace before, and I emphasise ‘before’, a single piece of data has been collected.

It is surprising how often this step is deftly avoided, although many words may be used to create the impression that it has been dealt with satisfactorily. It is important at the initial stages of an investigation to obtain a clear statement of intent, as many bewildering design questions are simply answered by referring back to these statements.

In order to amplify the idea consider the following example.

EXAMPLE 1.2.1

Many years ago a series of agricultural fertilizer experiments were stated to have had two intentions:

1.  To delineate more closely the soil/climate/production conditions suitable for the effective use of a direct application of reactive phosphate rock (RPR) as a maintenance phosphate fertilizer.

2.  To provide an initial screening of the agronomic performance of a variety of alternative RPR-based fertilizers relative to water-soluble fertilizers.

These non-specific intentions need to be translated into specific actions in order to undertake a focused investigation. The first step in this translation is to obtain a more specific statement of the intentions.

For example, it is necessary to define what is meant by ‘the effective use of a direct application of reactive phosphate rock (RPR) as a maintenance phosphate fertiliser’. This will require making a decision concerning a measure of effectiveness. Is it to be the performance of a flock of sheep or herd of cattle, or some more direct measure of the productivity of a crop or a property of the soil? In either case, is the quality as well as the quantity to be compared? In order to say a fertilizer is effective its performance will need to be compared with a known standard. Consequently, it will be necessary to consider how the chosen measure of performance might be compared and the probability of detecting an important difference if it exists. For example, it might be decided to compare mean production at a specific application rate or, alternatively, the parameters of a response curve. Of course it is unlikely that the two fertilizers would produce exactly the same responses no matter how precisely they are measured. In view of this, how close are the measures of performance required to be before the two materials are to be considered to be equally effective?

As far as concerns the logistics of running the experiment, it will be necessary to consider what variates need to be collected, as the variates to be compared might not be the variates which can be directly observed. For example, if interest is in dry matter production over a specific interval of, say, four months, it will be necessary to cut and measure the production of the grown green herbage more frequently and this will need to be specified.

The ‘soil/climate/production conditions’ need close scrutiny. How many combinations are of interest? Is it to be an extreme and a midpoint of each combined in all combinations – that is, 27 combinations – or is it to be a random sample of all farms throughout the country, or a province? Are there specific combinations which are particularly important and only those combinations need to be investigated? The possibilities are infinite!

Perhaps the second requirement of the experiment, which is to provide an initial screening of the agronomic performance of a variety of alternative RPR-based fertilizers relative to water-soluble fertilizers, can be satisfied by listing the alternative candidates and asking if a subset would provide the required information. The chosen subset might be based, for example, on the solubility of the fertilizers in a standard solvent.

Generally, the requirement to translate the intention of a proposed scientific investigation into specific actions which need to be completed, soon leads to the need to answer important experimental design questions. These are:

1.  What is the experimental unit?

2.  What is a suitable selection of experimental sites?

3.  What measurements have to be made?

4.  How is the measured response to be described?

5.  What comparisons need to be made?

6.  What is the probability of detecting real differences between treatments? That is, what is the power of the comparisons?

7.  What experimental layout is suitable?

Techniques which help experimenters answer these seven questions will be dealt with in subsequent chapters. However, it is only after the answers have been provided that experimental intentions can be translated into a work plan. This details specific tasks which need to be successfully completed in order to obtain the required experimental information. The preparation of a work plan is an essential part of the experimental process as it is necessary for assessing resource requirements and subsequent allocation. The exact methodology followed in this step does not matter so much as actually making a focused attempt at preparing such a plan. In fact, methodologies for translating intentions into a work plan are well practised in many fields of endeavour and many suggestions as to how it might be done are available to people requiring guidance. Basically, it involves translating an intention into a written document which details how and when specific tasks are to be undertaken and completed, and by whom.

1.3 JUGGLING INTENTIONS AND RESOURCES

In translating the intentions into specific actions, the physical and financial requirements for the conduct of the experiment soon become apparent. It is often found that the resources required are far in excess of those immediately available. Modifications to the intentions, and attempts to obtain more (rarely less) resources, are a time-consuming but essential part of the process. This iterative process of assessing the resources required for an experiment and modifying the intentions often goes through many cycles before the requirements of the experimenter and the available resources are matched. It is always possible they will never be matched. In this case it is worth considering the enterprising and courageous step of redirecting resources to or from other proposed experiments in order to do a single satisfactory experiment.

EXAMPLE 1.3.1

Suppose in Example 1.2.1 there is a requirement to reduce the resources required. In order to achieve this, the types of questions which may be asked include: is it possible to achieve the stated intentions with a subset of the available RPR-based fertilizers? Is it possible to limit the number of variates required to be observed without compromising the intention of the experiment? Is it possible to modify the intentions of the experiment and still find out something which is valuable? Is it possible to extend the time available for the experiment and reduce the size of each year’s experimental programme? Is it possible to forgo the conducting of another experiment in order to satisfactorily complete this one?

1.4 FACILITATING THE PROCESS

The planning of complex experiments which involve a number of collaborators can be a time-consuming and contentious process. The purpose of a complex experiment is sometimes not clearly understood by all participants and, in addition, they usually have useful ideas they want to contribute. Some see the experiment as an opportunity to pursue a related but alternative investigation. Others simply misunderstand, sometimes due to a lack of fundamental knowledge. These ambiguities cause major difficulties which are becoming more important with present-day emphasis on multi-disciplinary, multi-site and multi-manager approaches to scientific endeavour. Many of the problems can be quickly resolved or at least ameliorated by a skilful facilitator who is knowledgeable in experimental design. The facilitator can be the project manager, a statistician or anyone else who has the necessary skills and is able to focus the attention of collaborators onto important issues and so speed the planning process to a successful and harmonious conclusion.

A facilitator also has a role in ensuring the subsequent management of the experiment is satisfactory. Meetings with facilitators can be very successful team-building experiences. As is argued in Chapter 6, this has a very positive effect on the quality of experimental data.

1.5 THE UNEXPECTED

Despite careful planning it is unlikely that an experiment will be conducted without something unexpected occurring. All experimenters hope the something unexpected will lead to an important discovery like Fleming’s discovery of penicillin. However, it is far more likely that something will go seriously wrong. Hurlbert (1984) has called it ‘demonic intrusion’. When faced with such occurrences it is unlikely that exorcism or human sacrifices will be as helpful as vigilance and a readiness to act sensibly and quickly in order to minimize the loss of data or the compromising of experimental intentions.

Selection of experimental material

2

2.1 IDENTIFICATION OF EXPERIMENTAL UNIT

Once the intentions of an experiment are clear, the experimental unit can be identified. It is the population entity about which some knowledge is sought. More formally, it is the entity about which hypotheses are to be formulated and inferences are to be made. It requires the experimenter to identify the population about which the results of the experiment are to apply.

It is important that each experimental unit has its allocated treatment applied independently of all the other experimental units. In this context ‘independent’ means that all the sources of uncontrolled variation are free to be fully expressed on each individual experimental unit. This ensures that the expression of uncontrolled variation is not conditional on what happened on any other experimental unit. As well as applying the treatment independently of the other experimental units it is necessary to ensure the response remains independent of the influence of treatments applied to other experimental units for the duration of the experiment.

There are two types of experimental units. The first is typically a single plant or animal. The second is typically a group of plants or animals. In order to distinguish between the two types of experimental units consider the following situations.

If inference is to be made about a single plant or a single animal, the experimental unit should be the single plant or single animal. For example, if inference is to be made about hormone dynamics within a plant or animal, individual treatments are independently applied to individual plants or animals. The individuals are then kept in such a way as to ensure the response to the treatment in any one of them is not influenced by the response in the others.

If inference is to be made about a field of plants or a group of animals then the experimental unit is the field of plants or the group of animals. For example, if inference is to be made about yields of wheat cultivars when grown as a field crop, the experimental unit should be a field of wheat. The individual plants within the field do not behave independently of each other as they compete with each other for light and nutrients. If inference is to be made about growth of animals when maintained as a group then the experimental unit is the group of animals. Just as individual plants within a field do not behave independently, individual animals within the group do not behave independently of other members of the g...