1. Introduction
2. Research process
3. Study protocols
4. Objectives
5. Study design
6. Data management
7. Data analysis
8. Reporting
9. Collaboration
10. Related Reading

1. Introduction

Applied biometrics plays a key role in the development of research strategies. This becomes truer now than ever before as research projects become more and more complex. Part of the reason is that the nature of agricultural research is changing. As discussed by Lynam in Rowlands (2000),

"Stakeholders now recognise that the role of agriculture is much more than food production. It provides income and employment and is an integral part of rural livelihoods. Agriculture is recognised as having impacts on the natural resources not only of farmers but also of others, both nearby and distant."

so that

"The focus has moved beyond farmers' plots to households, villages and larger scales, and social sciences have to be as important as biophysical sciences in understanding key processes."

With the increasing multi-disciplinary approach to agricultural research, to which other disciplines such as environmental and social science are increasingly being added, the discipline of 'biometrics', the statistical or quantitative study of biology, is becoming increasingly important. Indeed, one could say that biometrics is now an essential discipline within the makeup of a multi-disciplinary research team. 'Biostatistics' is an alternative word, commonly used in the medical field. 'Econometrics' is the statistical study of economics. 'Statistics' is a more general term that encompasses all three.

The discipline of biometrics (or biostatistics) contributes to the overall field of 'research methodology' that is applied in the design and analysis of research studies. Indeed the term 'research methods specialist' is now sometimes used instead of the word 'biometrician' because of the fear that the subject 'biometrics' can sometimes instill in people. As a subject, however, research methods deals with other aspects of study design that are not statistical in nature, and hence is the broader than that based on statistics alone.

As a compromise we have called this resource 'Biometrics & Research Methods Teaching Resource'. Whilst the primary focus is in the teaching of statistics, the amalgamated name provides the opportunity to deal with some aspects of research methodology that may lie slightly outside a traditional course in applied biometrics.

A research investigator, whatever his/her discipline, needs to have some understanding of the principals of study design and methods of data analysis for getting reliable results from research. Some knowledge of research methods is, therefore, essential.

A research investigator can be any one of a number of people working in a range of development and research fields: scientist, biologist, agriculturalist, extension worker (or even in some cases the biometrician himself/herself!). In order to maintain a constant nomenclature in this Teaching Resource to cover all these categories we use the word 'researcher'.

Each member of a multi-disciplinary team needs to have some involvement from the beginning, whether it be at the project proposal writing stage or at the beginning of the research project itself once funding has been approved. This means that clearly defined plans with clearly stated milestones need to be established so that all members know what is expected of them. This is where 'research strategy' comes in.

Every research project will encompasses a range of activities in the research process - namely from study design to data management, to data exploration to statistical modelling and, finally, to the reporting of results. These five stages each require important research method inputs all of which need to be considered at the outset when putting together a research strategy.

So what is a research strategy? It is the activity that needs to be undertaken to ensure that there are adequate resources available to complete the study in the time available, to make sure that the approach to the design of the study is the appropriate one to achieve the study's objectives, that suitable softwares are available to manage and analyse the data, that sensible sets of data are collected to ensure that analysis will allow the required information to be extracted, and so on.

A research strategy can evolve and is not necessarily cast in stone. Things can go wrong; for example, unexpected field conditions (drought or flood) may cause a study to fail. Thus, the original strategy may need to be thought through again and revised. Case Study 6, which describes the different studies used to evaluate the benefit of providing smallholder dairy farmers in Kenya access to credit to allow increased feeding of concentrates to cows in early lactation, demonstrates how research strategies can change.

Alternatively, execution of exploratory analysis may suggest a different approach to formal statistical analysis than had originally been envisaged. This will cause a modification to the original strategy that had been proposed for the statistical analysis of the data.

Other case studies also describe aspects of research strategy.

Case Study 1 discusses the steps that were taken in deciding the best approach to underake an initial study of the cattle husbandry practices of the Orma people in eastern Kenya, whilst, at the same time, getting estimates of levels of milk production.

Case Study 5 describes how a research strategy was developed to plan a step-by-step approach for determining from a large number of Napier grass accessions a list of 'best-bet' forages to be grown by smallholder farmers.

Case Study 11 addresses some of the fundamental questions that needed to be addressed before embarking on a livestock breed survey in Swaziland. Many of the lessons learned from a similar survey are described in Chapter 6 of Rowlands et al. (2003) that illustrates the various issues that need to be considered when planning a survey.

Case Study 10 describes the statistical issues that need to be considered before embarking on a study of the impact of an intervention - what approach to use for defining appropriate baseline values against which the results of the intervention can be compared, how the data resulting from the monitoring might be analysed, etc.

Case Study 16 shows how agronomic, nutritional, morphological and quality characteristics of taro that are of interest to subsistence farmers, processors and consumers alike were taken into account when designing a series of experiments to determine taro growth in relation to yield, corm crisping quality, mineral content and storability.

2. Research process

Studies are rarely done in isolation. One study builds upon another. Each study is composed of a set of different stages that can be broadly summarised into: Study design, Data management, data analysis (in the form of Exploration & description followed by some form of Statistical modelling) and finally Reporting of the results. The other five teaching guides follow this research process. Results obtained from one study, even before it is written up, will help to define the next (Figure 1). The figure shows how plans can be initiated as soon as some preliminary results are known. On other occasions the researcher may prefer to wait until the results are written up and firm conclusions from the study are available before deciding what to do next.

Figure 1. Diagram showing how one study leads to the next.

By the beginning of a project the person leading the research will have studied the literature in relation to the particular field of interest in order to provide a starting point. He/she will then need to develop with his team a research strategy on how to go about the research and the type of studies that will needed. For example, an initial study may be required to fill gaps in knowledge, to confirm previous results or to provide baseline data within the environment in which the project is planned.

Research goals can change during the course of a project. For example, data that had been collected over a number of years from village cattle in Ethiopia as part of the research described Case Study 10, and used to provide a baseline against which the impact of tsetse control could be judged, were originally collected for a different purpose. This was to provide data on trypansomosis in susceptible breeds of cattle to compare with similar sets of data being collected in West Africa for breeds that tolerated the disease.

Two publications by Dolan (1998) and Bealby et al. (1996) provide end-of-project reports (the first in relation to investigations into levels of trypanotolerance of Orma Boran cattle in Kenya, and the second in relation to studies on alternative methods of control of trypanosomosis in goats in Zambia). Each describes series of studies that took place during the research. On reading these publications the reader will be able to see the systematic way in which the research was done, how one study led to another and how the results obtained from one sometimes influenced the design of the next.

The first step in the research process, of course, is the preparation of a funding proposal itself. The researcher needs to have pretty good ideas at this stage as to the research process and the types of studies that will be planned. This is the first stage in which the biometric issues need to be addressed. The person providing this input, whether he/she be a biometrician or not, needs to be a key person throughout the whole research process - from assisting in proposal writing, to protocol preparation, through study design, and all the way to reporting.

This person could be one of the researchers in the team who has good research method skills, or alternatively a professionally trained biometrician. Such a person needs to bring the special attributes of objectivity and ability to think laterally around a problem. He/she will need to be able to look at the detail of what is being proposed, and to force others to rethink study plans when he/she believes that proposals will not meet the intended objectives.

The biometrician's skills in analysis and interpretation of results will help to ensure appropriate conclusions are drawn for each study. Skilled interpretation can influence future research strategy and recommend areas where further evaluation might be useful. Where possible, opportunities should be taken to introduce additional hypotheses that might be tested in a subsequent study. Sometimes the results from a study may be sufficiently inconclusive that it is decided to repeat the study. Even then it may be possible to design a new study that not only replicates the previous study but, at the same time, also addresses other questions.

3. Study protocols

Every study needs a protocol to be written that gives the background to the research (perhaps with a literature review) and the study aims and objectives. It also needs to give details of how the study is to be designed, where it is to be done, its size, the treatments to be applied, measurements to be made, methods proposed for the management of the data and the types of statistical analysis to be undertaken.

Examples of study protocols are included in Case Study 13 and Case Study 5. Case Study 13 describes the protocol for an experiment to compare the performance and meat quality characteristics of three indigenous breeds of goats in Ethiopia. Case Study 5 includes a protocol developed to set up an experiment to compare the performances of different accessions of Napier grass when fed to sheep.

The discipline of biometrics can help to ensure that protocols are clear and unambiguous and likely to provide a study with a good chance of leading to a successful and statistically significant conclusion.

It is also important at the planning stage to estimate the amount of time required to execute the study, including data handling and statistical analysis. Deadlines and milestones also need to be established and written down so that the study can be completed 'on time'. When preparing an initial project proposal for funding purposes it is important to include sufficient funding in the budget for statistical support - this is often forgotten.

These estimates need to be realistic. So often, especially in surveys (see, for example Case Study 6 and Case Study 11), the length of time needed for data management and analysis is grossly underestimated. It is instructive to look at Tables 6.1-6.3 in Chapter 6 of Rowlands et al. (2003) which set out the various steps needed in the planning and execution of a livestock breed survey.

4. Objectives

Clear and precise objectives are essential for study execution and analysis. It is through these objectives that hypotheses are formulated for evaluation during statistical analysis. Researchers often have difficulty in defining study objectives. It is therefore important that this topic be addressed during research methods teaching.

Every case study within this Teaching Resource includes a section that sets out the objectives for the study. These are not all perfect and when studying a particular case study students could be asked to comment on their suitability.

For example, Case Study 13 sets out to

These are fairly broad and perhaps should have been more specific in describing the actual breeds to be compared, the ages at which they were to be studied and the type of feeding regime to be employed.

Objectives set out in Case Study 5, however, provide more specific details in describing a series of studies to evaluate the agronomic and morphological characteristics of different Napier grass accessions.

The case study sets out an overall objective:

To identify and select a short list of Napier grass accessions from the ILRI Forage Gene Bank suitable for growth under different environmental and climatic conditions.

and individual objectives for each of four studies:

These objectives specify the locations where the studies will be done and provide some details of how the studies are to be designed. Hence, they are a little more detailed than those given for Case Study 13.

At the opposite extreme, however, consider how poorly an objective might have been written for Case Study 6, which evaluates the benefits in increased overall lactation performance achieved by allowing smallholder dairy farmers extra credit when cows calve, namely:

To evaluate the potential for reallocation of concentrates in the feeding of dairy cattle

There are a number of things are missing.

A somewhat better statement that appears in Case Study 6 is:

To investigate the potential for shifting to early lactation the allocation of concentrates fed to cows by smallholder dairy farmers in the central highlands of Kenya by arranging access to credit from the dairy cooperatives to which they sell their milk. It is assumed that the low milk yields currently found among cows owned by smallholders is due to poor feeding of concentrates in early lactation when cash is not available.

When preparing an objective it is helpful to set out the hypothesis that will need to be formally evaluated. Statistical inference is built around the concept of the 'null hypothesis'. Thus, the null hypothesis in the example above could be:

Advancing credit to smallholder farmers in Kiambu district in Kenya as soon
as their cows calve has no effect on overall lactation yield.

The statistical test to be applied will assess the probability of this null hypothesis being true or not. The smaller the probability that the test gives for the null hypothesis to be true, the more likely is the hypothesis false. Thus, when applied to the above example, rejection of the null hypothesis will mean that reallocation of concentrates does in fact improve lactation yield.

Sometimes hypotheses are developed during a study and after data have been generated. It is important to distinguish these from those formulated at the beginning of the study. In such cases the researcher will need to be more cautious, and possibly set the probability levels of rejection of a null hypothesis a little lower than when dealing with a hypothesis formulated at the beginning.

Animal experimentation often involves studying species that are not the primary target species. Thus, for example, mice are often used as a laboratory or model species for research in cattle. The reasoning is that mice are cheaper and, since their generation intervals are shorter than in cattle, more rapid progress can be made. It must always be borne in mind, however, that results obtained for mice will apply only to a certain population of mice and may not necessarily be applicable to cattle. This must be clearly stated when setting objectives.

Case study 5 provides an example of this when an experiment to compare livestock productivity when fed different Napier grass accessions was done in sheep when cattle are the primary target species to which Napier grass is fed.

5. Study design

The researcher's first task, once a research proposal is approved, is to plan the first study. Biometric inputs are essential to ensure that sample size is adequate to meet the study objectives, and that an appropriate study is planned that makes optimal use of resources.

It is also important at this stage to make sure that study can be done with the resources available, and to be able to define a time frame within which the study can be completed. Once again, Chapter 6 of Rowlands et al. (2003) gives various tips on the requirements that need to be taken into account when planning a survey.

The process in planning a study often becomes an iterative one, involving an assessment of objectives and resources, and usually needs more time for discussion than is often appreciated. The biometrician, for example, may propose a scaling down of objectives, for example, a reduction in the number of treatments. Alternatively, the researcher may agree to a biometrician's recommendation to increase the size of the study beyond that originally considered.

It is important when designing a study to look both backwards, to consider what has been done before, and forwards, to consider what might be investigated next. Study design is a major component of the overall research strategy and the biometrician's contribution to study design will be greatest when he/she understands the overall strategy.

A clear understanding of the conclusions that might be drawn at the end of the study is also important at this stage. The study needs to be relevant to the problem in hand and appropriate to the population for which its eventual impact is intended. Basic questions need to be asked. In relation to the survey described in Case Study 11, the following questions were asked before proceeding with the design of a nationwide livestock breed survey:

Sooner or later study design may need to cater for some form of participation by farmers. Participation can vary from simple participation, to farmer management of a trial, to joint researcher/farmer trial design. The research strategy needs to decide when an element of farmer participation is required and what form it should take.

Case Study 1, Case Study 10 and Case Study 16 illustrate different ways in which projects can benefit from farmer participation.

6. Data management

Data collection, management and processing are integral parts of the research process, and, to ensure efficient and timely data analysis, these need to be planned ahead. It must be clear from the outset who is to be responsible for entering the data and into what type of data system (e.g. a spread sheet or a relational database system) the data should be entered.

Sometimes it will be possible to use an existing computer package, e.g. Excel or a statistical software package such as SAS or GenStat. However, when studies become larger and more complex (for example, Case Study 6 , Case Study 10 and Case Study 11) it is likely that a data base system will need to be written (in Access, say) to handle the data as they are collected.

Researchers and biometricians need to know about data base design, but the programming of a comprehensive database system, such as for Case Study 11, will usually need to be left to a database expert.

It is important that researchers know how databases are constructed, and how data can be stored in them, for this will help them to understand whether or not a database specification meets the research objectives. By understanding the research strategy a biometrician can anticipate future research requirements and so ensure that any data collection and storage procedures written for the current study can also apply to future studies that may use a similar database structure.

When planning a research programme it is essential that the team leader ensures that the research team possesses sufficient data handling skills. Often the data entry and management process last much longer than expected. As a rule of thumb an initial estimate of the time anticipated can often be doubled. Case Study 6 shows how the volume of data that was collected was difficult to handle, and how the delays in data entry interfered with study milestones.

Decisions on the procedures for the archiving of the data that will take place at the end of a project should also form part of the research strategy; the documentation procedures need to be defined at the start. Archiving can be a boring part of research, but is, nevertheless, one of the responsibilities of the researcher. Data belong to a researcher's institution, not to the researcher. The researcher, therefore, must ensure that his/her data are archived in such a way that they can be made available to anyone else in the institution.

Well documented, archived studies can be useful to another researcher and, alongside a general literature review, help him/her in the design of a future study. Alternatively the data may be interrogated, possibly along with other data sets, as described in Case Study 7. This case study analyses archived meteorological data to answer questions on appropriate timing of crop planting in Zambia.

Fifteen of the case studies in this Teaching Resource contain sets of data with a documentation file attached. These illustrate different ways of storing and documenting data.

Studies can involve repeated measurements in time (see, for example, Case Study 10). In such cases the data management system not only needs to facilitate the way the final analysis of the data is conducted, but also needs to be designed so that it can provide interim reports as appropriate. Indeed the timing of an intervention depends on the analysis of previous data. The resources needed to ensure sufficient utilisation of data, both during and at the end of a study, need to form part of the research strategy.

There are important data ownership issues concerned with any research project, especially those that cut across different institutions and involve multiple partners. These issues, such as the extent to which individual researchers are allowed to access the database, need to be addressed at the outset.

7. Data analysis

Data analysis can form, along with data management, a significant part of a study, and the time involved can again often be underestimated. It is important that data analysis is not delayed too long after the data have been collected, since the next study to be planned may depend on the results of the current study.

In simple cases the form that data analysis is likely to take will be clear at the outset. An outline of the analysis and the statistical models to be fitted, together with the tools and software to be used, should be described in the study protocol (see, for example Case Study 13 and Hanson and Fernadez-Rivera (2006) referenced in Case Study 5).

As studies become more complex it becomes more difficult to specify precisely the form that statistical treatment of the data will take; some exploratory analysis may therefore be necessary to study the patterns displayed within the data. This makes a strategy to be followed in the analysis somewhat more difficult to define at the beginning. Nevertheless some thought needs to be given at the start of a study to describe, in general terms, what approach is likely to be taken in the handling of the data.

Exploratory data analysis has several functions. For example, it can involve exploration of the distribution of data in order to judge the appropriate statistical methodology that should be applied. It can be used to spot outliers in the data. The more complex a study the greater can be the scope for data errors. Decisions can be made ahead of time to determine a system of ground rules for the actions to take when outliers are detected. This can reduce elements of subjectivity.

Exploratory data analysis also provides an opportunity to explore patterns in the data (see, for example, Case Study 3). It enables the possible influences of other variables (covariates) that have been measured (for example, gender, age) to be determined.

Data analysis can often involve a team approach - the researcher may be able to conduct some simple, exploratory analyses to familiarise himself/herself with patterns in the data so that he/she can then provide the biometrician with suggestions for the more formal statistical analysis. Sometimes it may be possible to use a subset of the data to try out the statistical analysis in order to save time later. Whatever approach is adopted it is important that some plans are made at the initiation of a study on the analyses that are expected and how they might be done. 

8. Reporting

Appropriate methods for disseminating results, both during and at the end of the study, need to be considered at the start. Decisions need to be made on who needs to have reports and in what form they should be presented.

Mostly, results will be presented in the form of papers in scientific journals or talks at conferences or to colleagues. But reports will also be required at regular intervals for donors, for the researcher's institution and for project partners. Such reports cannot be delayed.

When working with farmers it will be important to provide feedback in the form of simple reports: they may be important in ensuring future cooperation. Farmers are often likely to be the ultimate beneficiaries of agricultural research, and so it is important to explain to them the results of any research in which they have participated. Imaginative forms of dissemination may be appropriate, for example in the form of plays or cartoon drawings (see, for example, Case Study 7).

Decisions on co-authorships are best made at the start. Not to do so, and to leave this decision until later, can sometimes lead to conflict. The senior author will generally be the one who writes the report, but others will be expected to contribute. Anyone included as an author needs to be able to defend his/her particular aspect of the work.

Different institutions collaborating in a project may have their own formal systems for publication review. If not, an internal report reviewing process within the research team could be valuable.

9. Collaboration

The foregoing has demonstrated the important contribution that biometrics play in the research process. Every researcher needs to endeavour to develop some research method skills of his/her own in order to improve the overall quality of study design and analysis. A person's level of knowledge will vary and depend on his/her general aptitude for and interest in the subject.

Not everyone, however, can be expert in every discipline and so every member of a research team needs to recognise the expertise of others. Even when a researcher has excellent research method skills, he/she will often find it fruitful to share his thoughts with a professional biometrician. Because of the biometrician's wider exposure to different fields of work, he/she will often think of other ideas or solutions.

A biometrician must get involved in research projects. When a researcher comes along with a query, the biometrician must take the opportunity to learn more about the project and to see other ways in which he/she can help. Once the biometrician starts to be brought in at the research planning phase it is then that his /her contribution becomes most effective. Some thoughts on biometric collaboration are discussed by Rowlands (2005).

To be an effective applied biometrician one needs to have certain skills that may not always be taught in an applied biometrics course. One needs, for example, to be able to write well, and know how to write up results concisely. One also needs to be able critically review other work. This latter ability is addressed in Case Study 6. R. Coe in the second paper in Rowlands (2000) looks at the role of the biometrician amidst the changing agricultural research focus.

An applied biometrician needs to be approachable, to get on easily with people, and to be able to clearly express statistical concepts and explain results of statistical analysis. He/she also needs to have the confidence to challenge a researcher on the suitability of a particular study design, on the interpretation of a research result, etc.

Finally, as for any research discipline, the biometrician needs to be a good team player and be able to work and respect the views of others.

The purpose of this Teaching Resource is to provide opportunities, through a case study and 'hands-on' approach, for the student to develop research method skills and, in doing so, recognise that not only is biometrics a subject that can be enjoyed but one that is essential in biological research.

10. Related reading

Participants who reviewed this Biometrics & Research Methods Teaching Resource at a workshop held at the University of Cape Town recommended the following text books:

Ford, E. David. 2000. Scientific Method for Ecological Research. Cambridge University Press, Cambridge.

O'Leary, Zina. 2004. The Essential Guide to Doing Research. Sage Publications.

For a summary of some of the contents of these teaching guides the reader is referred to the paper by the late Harvey Dicks: A guide to good research planning.

The reader is also referred to the following Good Statistical Practice Guide written by Statistical Services Centre University of Reading:

Statistical Guidelines for Natural Resources Projects