Module 5: Characterising crop–animal systems using participatory research techniques

D Pezo¹, C León-Velarde² and T Soedjana³

1. ILRI, c/o IRRI DAPO Box 7777 Metro Manila, The Philippines
2. ILRI, c/o CIP. Apartado 1558 La Molina, Lima 12, Peru
3. Center for Agricultural Library and Technology Dissemination (ICALTD), Ministry of Agriculture, Jl. Ir. H. Juanda 20, Bogor 16122, West Java, Indonesia

5.1 Performance objectives

5.2 Introduction

5.3 Hypotheses formulation

5.4 Type and source of information

5.5 Techniques used to collect primary data

5.6 Methods of analysis

5.7 Flow diagrams: A tool for the characterisation of crop–animal systems

5.8 Outputs of the characterisation process

5.9 References

5.1 Performance objectives

Module 5 is intended to enable you to:

• List the steps required to conduct a systems characterisation study.
• Formulate hypotheses to be used as a basis for the design and analysis of a study on the characterisation of crop–animal systems.
• Define the statistical procedures to be used to analyse data collected in characterisation studies, considering the nature of the attributes under study.
• Discuss the potential difficulties in data collection and ways to prevent them.
• Describe the potential sources of primary information for the characterisation of smallholder crop–animal systems.
• List the techniques used to collect primary information for the characterisation of crop–animal systems, and establish the differences between them.
• Discuss the relevance of using participatory approaches in research oriented to solve the problems of smallholders.
• Identify eight basic principles that guide the participatory rural appraisal (PRA) technique.
• List the actions to be taken for planning data collection using PRA.
• Explain the tools used to characterise community/village using PRA.
• Prepare relevant and appropriate questionnaires for static and dynamic surveys.
• Identify the elements of a crop–animal system and the type of interactions that may occur. Incorporate them in a flow diagram that represents the system.
• Differentiate between qualitative and quantitative flow-diagrams of crop–animal systems, and construct them based on the information collected in the field, through observation, discussion with farmer groups and one-to-one interviews.

• List the products to be obtained after conducting a systems characterisation study.

5.2 Introduction

Characterisation is one of the initial phases in any project using a systems research approach. Sometimes researchers forget that characterisation is also a research activity, not just a data collection process, therefore the scientific method has to be applied. It means that the research team should start reviewing the literature and/or any other source of information to formulate researchable hypotheses, collect and analyse data to test the hypotheses, and prepare the appropriate inferences based on those tests.

The following general steps have been suggested for the characterisation of market-oriented dairy systems in sub-Saharan Africa (Rey et al. 1993), but could also apply to crop–animal systems in South-East Asia:

• Formulate questions and hypotheses to be answered or tested in relation to perceived problems in the systems under study.
• Determine the analytical methods to be used to answer the questions and to test the proposed hypotheses.
• Define the data needed to respond to those questions and to test the proposed hypotheses.
• Identify the sources of data and the most appropriate data collection procedures for the different sources.
• Where primary data is deemed necessary, design the survey protocol, including the sampling design, the field instrument to be used to gather information, and the analytical procedures to be applied.
• Analyse the data collected to test the proposed hypotheses.
• Draw conclusions based on the hypotheses tested.

A logical sequence of steps for the choice of analytical methods, data and data sources, methods of data collection, and sampling procedures is shown in Figure 5.1 (Jabbar et al. 1997). All these steps are discussed in detail in following sections.

5.3 Hypotheses formulation

Research aimed at solving problems requires a good understanding of the most relevant problems in the systems to be characterised and their possible origin (Jabbar et al. 1997). Hypotheses can be formulated based on the information gathered during site description (see section 4.4), and also from the review and analysis of secondary information, but always within the limits delineated by the conceptual framework of the project (Rey et al. 1993) (see section 4.3).

Even with a conceptual framework, the complexity of crop–animal systems in terms of components and interactions may lead to a long list of hypotheses. Moreover, this could be complicated by the multidisciplinary character of the research team, as people with different backgrounds may have different perceptions of the problems and their causes. Therefore, consensus is needed to define and prioritise researchable questions and hypotheses to be tested.

The team must define hypotheses carefully considering that they have to be tested using the information collected in the field. Many hypotheses arrived at are not researchable given the time, resource and data constraints; these should be deleted. Some examples of hypotheses that can be formulated when studying crop–animal systems are shown in Box 5.1. Hypothesis need to be defined at the outset of the project to avoid the costs of changing research plans after they are already under way or repeating activities to obtain missing information (Mullins et al. 1994).

5.4 Type and source of information

The questionnaires used to collect primary information in characterisation studies frequently contain too many questions; this problem is more acute when more complex systems are under study. Each member of the research team tends to include as many questions as possible in his/her field of expertise, and it is not uncommon that the inter-phases between components are not properly covered. Furthermore, researchers tend to collect more information than needed, with the hope that the use and relevance of

Source: Jabbar et al. (1997).

Figure 5.1. Steps for the characterisation of crop–animal systems.

such data will appear during the analysis, or may be used for future objectives (Jabbar et al. 1997). In many cases the target communities have been already surveyed by other projects or programmes; therefore, information must be available somewhere, and could be reviewed by the research team. If that is the case, researchers should avoid repeating questions included in previous studies, unless there is evidence that responses could have changed with time.

Box 5.1. Some examples of hypotheses formulated for characterisation studies working with crop–animal systems Animal component Cut and carry is practiced by the majority of farmers Land allocation to forage production is positively related to farm size and security of land tenure The extent of use of crop residues is negatively correlated to the length of the growing period The use of tree foliages as feed is not a common practice Ruminants and monogastric animals are complementary, since both do not compete for feed resources Native animal breeds are preferred by smallholders because of their adequate reproductive performance even under feed scarcity conditions Endoparasite infestation is greater in exotic breeds than in native animals Utilisation of veterinary services for curative interventions is greater when improved breeds are used Crop–animal interactions The main purpose of raising cattle is for draft, but the use of draft animals is declining Farmers do not use efficiently the excreta produced by their animals Farmers do not like dwarf sorghum genotypes, although they produce high grain yields, because of their poor stover production The number of cattle/buffaloes maintained in the farm is directly associated to the availability of crop residues Farmers raising animals with higher genetic potential have a larger area devoted to the production of improved forages Supply, processing and marketing Middlemen are the most common means of cattle trading Those farmers with better access to the market practice more intensive feeding systems Lack of knowledge and facilities for processing/handling milk on-farm is perceived by farmers as the major constraint to increase milk production Socio-economic aspects Farmers give greater importance to crops than animals, therefore the former has preference in labour allocation The relative contribution of animals to the total farm product is higher in those farms with greater capital and better endowed in land resources. Specialisation is considered by smallholders as undesirable because it is perceived as too risky Household security and the nutritional status of children is enhanced by incorporating milking animals into crop–animal smallholder systems Access to credit is a major constraint for increasing the animal component in the smallholder crop–animal systems Smallholders' risk-bearing ability is expressed by multi-commodity found on the farm as they are risk avert individuals Smallholders are not maximising profit generated by their activities on the farm, but rather maximising utility or satisfaction of all farm family members Gender issues Women are more open than men to improved forage technology, because they are the ones responsible for its collection and distribution to the animals Supplemental income associated to increases in animal production are disproportionately retained by the men of the household Women and children are the ones managing animal species, but very seldom they benefit of extension activities Households with young children may give priority to adequate food crops and the demands for women's labour Households with older children at home and more labour upon which to draw may take on more labour demanding activities Temporary or permanent migration may leave a high proportion of female-headed households, with less available labour and more limited access to resources for production

 Some shortcomings of ignoring the recommendations cited in the previous paragraph are (Mullins et al. 1994; Jabbar et al. 1997):

• Respondent-fatigue from long interviews leading to poor data quality and unwillingness to participate in subsequent surveys.
• Surveyor-fatigue, which may affect his/her attitude towards the informants (e.g. farmers and middlemen) and the quality of the data collected. This is particularly important when the recommended strategy for enumerators is not to use a questionnaire or take notes in front of the informants and their families to facilitate a more relaxed environment during the interview.
• Overloading the survey instrument with irrelevant questions, and omitting key information.
• Inefficient use of resources when analysis eventually considers only a fraction of the data collected.

All the data collected should be relevant and useful to answer specific questions or to test the hypotheses, and must be compatible with the analytical methods chosen. However, all data may not be necessarily analysed in the way it was collected. In some cases, the data need to be transformed to a common unit of measure (e.g. the farmer may recall the amount received from animal sales, but does not know the weight). In addition, some of the variables may be integrated into a compound attribute (e.g. the production of forage per unit of land can be estimated based on the number of animals fed, the amount given per day, and the number of days animals are fed).

Sources of information need to be carefully identified and chosen, combining quality attributes of reliability and accuracy, while considering access and easiness of data acquisition (Jabbar et al. 1997). It is also important to differentiate between sources of secondary and primary data. Among secondary data are published materials, official statistics, GIS maps, and grey literature (documents with limited circulation).

The most direct source of primary data for the characterisation of crop–animal systems are the farmers involved in this type of activity and members of their families however, other key informants should also be approached. In this context, a key informant is an individual who is knowledgeable about the subject and is willing to share his/her knowledge with the research team (Jabbar et al. 1997). The perceptions and views about a particular problem or issue may differ between farmers and key informants and among key informants. A chain of informants would be therefore useful to get a complete picture of a given problem. Members of a chain of key informants may be the community/village leaders and authorities, government officers (i.e. agricultural and natural resource management research and extension and health service personnel), teachers, NGO representatives, farmer association leaders, milk collection plant or slaughter house managers, middlemen, retail shop managers etc.

5.4.1 Procedures used for data collection

The procedures used for data collection are a function of the type of data sought and the sources of information. Some of the procedures commonly used are discussed below:

Group interview

This involves an open-ended discussion, usually conducted with a group of people who share resources or activities. This procedure is particularly useful to collect qualitative information on situations and events that may help to understand changes with time for a given attribute or set of characteristics within a community. It also helps researchers understand the perceptions regarding policy environment, socio-economic conditions, and interactions between people and systems.

Informal survey

This is systematic, but semi-structured activity carried out in the field by a multi-disciplinary team, to get a rapid appraisal of the conditions in a given area and to determine the systems practised (Farrington and Martin 1987; Rey et al. 1993). It is more appropriate for understanding rather than quantifying a given situation, but could be complemented by a small-scale, focused verification survey to improve reliability (Mullins et al. 1994).

Formal survey

A formal survey is usually a one-to-one questionnaire-based set of interviews to a sample of respondents who are considered representative of the population under study. This type of procedure is recommended when valid statistical inferences are sought, and quantitative data are required.

Case study

This is a detailed study of a limited number of units, selected as representatives of the target group(s) relevant to the issue under consideration, but not necessarily representative of the whole population (Jabbar et al. 1997). For example, a case study could be carried out with a group farmers selected among those who apply animal excreta as fertilisers to document, quantify and define in detail the variation in the use of this resource by smallholders who practise crop–animal systems in a given area. Using case studies, it is possible to collect quantitative data directly and frequently, since a limited number of farmers are involved.

5.5 Techniques used to collect primary data

The characterisation of smallholder crop–animal systems, frequently found in poor farming conditions (i.e. low and erratic rainfall, weak infrastructure, poor soils, hilly topography and high variability over time and space), is not an easy task (Sagar and Farrington 1988). It is a continuous process, with clear objectives that can be met using diverse techniques, namely rapid rural appraisal, static and dynamic surveys, case studies etc. Participatory approaches must be used in all stages of characterisation if the research team wants to properly identify farmers' goals, objectives and problems (Farrington and Martin 1987). In this way the research agenda is established based on real problems identified by the farmers, through the facilitation of the research team, rather than extracting information from the farmers to respond the concerns of researchers (Rhoades 1987).

5.5.1 Participatory rural appraisal (PRA)

Since the early years of the farming systems research approach, researchers have tried to acquire new information and hypotheses about rural life and rural resources, as well as to understand the farmers' reasoning and points of view. For that purpose, the tools for ethnographic and applied anthropology research were used. The technique applied at that time was rapid rural appraisal (RRA), a type of informal survey conducted on site, by a multi-disciplinary team which not only gathered the information, but also analysed it.

Participatory rural appraisal (PRA) is derived from the RRA, but in this case the research team works as a facilitator for local people to conduct their own analysis, planning, and even for them to take action to change their conditions. PRA is a move from letting people participate to letting them take the command of their own processes. PRA transforms the old dependency roles and recognises farmers and local people as active analysts, planners and organisers.

Participatory rural appraisal constitutes a first step in a process of interaction with farmers throughout the research cycle (Farrington and Martin 1987), which will allow primary data to be collected for the definition of a research agenda oriented to tackle the problems faced by smallholders. The aim is to identify the range of farm resources and physical environment, production priorities and practices in a specified area, through the active participation of farmers and key informants.

Rationale for and principles of PRA

Since the earliest stages of agriculture, farmers have been involved in technological experimentation. Many things we now take for granted in agriculture (e.g. planting seeds, use of buffaloes as draft animals and cheese production) are the outcomes of farmers' inventiveness. Farmers have been active actors in the process of selecting, consciously observing, manipulating and experimenting with plants, animals, natural resources and tools to improve the outputs of agriculture. Rhoads (1987) indicated that rather than being conservative, bound by tradition and simple, good farming requires:

• experimenters, risk takers, innovators
• intensifiers and diversifiers
• colonisers or pioneers
• addicts to new information
• practitioners of great common sense
• social and economical rational beings.

Participatory agricultural research, and PRA as part of it, is built on the recognition that farmers have those capabilities. Therefore, any research efforts driven by problem solving and oriented to benefit smallholders should have them as participants in the process. Farmers' decisions on production and consumption are determined by their attitudes and perceptions, goals and preferences, which in turn reflect their natural, cultural and socio-economic circumstances. PRA allows researchers to properly identify those aspects, since through this technique groups of farmers freely express their thoughts and concerns about the problems they face regarding their systems and the environment under which they work, identifying limitations and possibilities they have in common (CIAT-FSP 1995). Also, the research team facilitates farmers to establish priorities, defining research and/or development activities that will contribute to improve their living conditions.

The following are some of the basic principles of PRA.

Systematic group learning process. The focus of PRA is on the mutual commitment of all participants (i.e. farmers, researchers and agricultural officers) to best understand the complexity of the farming systems practised in a given community to find solutions to the problems detected.

Context specific. The PRA approach is flexible enough to be adapted to suit any new set of conditions and/or actors.

New roles for experts. The PRA methodology is concerned with the transformation of existing activities and practices to improve the livelihood of local people, i.e. the research agenda is `demand-pull', rather than `technology push'.

Interdisciplinary team. PRA requires a team consisting of members with different skills and professional backgrounds, which will approach the problems from different viewpoints with the interaction among them resulting in new and deeper insights.

Diversity and sources of information. PRA accuracy is achieved by drawing diverse information from different sources; however, the information collected needs cross checking.

Use of appropriate tools. The tools applied in PRA must be clear, self evident and simple, appropriate to local conditions and open to modifications suggested by the local people.

Flexibility and informality. Plans and survey methods used in PRA are only structured to start with, but should be revised, supplemented, detailed, adapted and modified as the fieldwork proceeds.

Rapid progressive learning. PRA is a way of learning from local people, eliciting and using their criteria and categories, and finding, understanding and appreciating their indigenous knowledge.

Community participation. PRA is based on the discussions of farmers, groups, designed to facilitate them to identify their own problems. Most of the activities are done jointly with other community members who can contribute in data analysis and interpretation. Projects dealing with natural resource management cannot achieve much involving only individual farmers, since most of the decisions and actions related to the use and conservation of natural resources involve/affect the whole community, and even neighbouring communities.

Preventing biases. PRA should actively seek for wide participation in all activities. The views, perceptions, preferences and goals of those members of the community frequently ignored in other activities (e.g. women, elderly and poor) must be sought.

Optimal ignorance. PRA should avoid unnecessary details and over-collection of data, hence should focus on what is relevant. However, it is usually difficult to define which information should be ignored and how much inaccuracy is tolerable before being criticised.

On-the-spot analysis/on-site presentation. In PRA, data collection takes place in the communities and the analysis of the information gathered is an integral part of the field work. The results of the study are evaluated by the entire team (at least in a preliminary stage) and discussed with all previously involved in focus group discussions.

The application of PRA results in several benefits for the community and the research team involved in this activity. Among these are:

• The community as a whole (including those members frequently ignored in other processes) is involved in the analysis of their conditions, gaining confidence to state and assert their community priorities, to present proposals, and to take action, leading to effective and sustainable participatory programme.
• The diversity and value of indigenous technology is documented and recognised as the basis for change, encouraging the community members to become actors in the research process, applying their innovative capabilities, along with the research team.
• The approach establishes the basis for community participation throughout all stages of the research process (identification and appraisal of problems, design and testing of alternatives, and monitoring and evaluation).
• The community is given an opportunity to explore the risks, costs and benefits of different options, and the research team to learn the rationale farmers apply in these types of decisions.
• Results in the identification and prioritisation of research issues relevant to the community, and sets the basis for participatory research aimed at solving problems.
• Contributes to the development of partnerships and solidarity among all community members and between these and the research team, creating a culture of open learning in all participants.

Planning and implementation of PRA

PRA needs to be carefully planned given that many actors are involved, in many cases it could be the first time the research team has approached a given community and errors at the outset may affect future partnership relations. The following paragraphs present some ideas for planning, which could be used as a checklist of actions to be taken during the planning phase, recognising that this could be incomplete and specific criteria can be added depending on the conditions under which this technique is going to be applied.

Integrating an interdisciplinary team. The composition of the interdisciplinary team will vary depending on the nature of the project and as well as the human resources available in the participant institution(s). For a project dealing with crop–animal systems within an eco-regional perspective, the minimum research team should include an animal scientist, an agronomist, a natural resource management specialist, and an agricultural economist or a social scientist. If the institution does not have one of those areas of knowledge represented in its staff, it should look for inter-institutional co-operation.

Defining site(s). Although one outcome of the PRA activity could be the definition of the site(s) where the project is going to take place, the general characteristics of the sites considered as candidates for the project should be defined beforehand. If that is the case, a pre-selection of sites, based on secondary information and key informants, should be done (see Module 4 for details). Once the site(s) has (have) been identified, the researchers must clarify the demand for innovation, and whether the area is more homogeneous or heterogeneous in terms of its environmental and socio-economic conditions.

Identifying target group(s) The identification of the target group(s) is a key step since they constitute the partners and beneficiaries of the whole research effort. For the selection of target groups, it is important to consider that some of the options developed in the research process can be useful for a given group of farmers (recognising that there are homogeneous groups among them), but also for the whole community. Target grouping divides the heterogeneous population of farmers into more homogeneous subgroups (recommendation domains) based on the type of farming systems they practice and the resources they have. These have to be taken into account for group discussion activities, since perceptions, attitudes and expectations of different members of the community may differ, and some of them may not express freely those in the presence of members of a different subgroup.

Clarifying goals and objectives for the intended PRA activities. The goals and objectives of the PRA have to be clearly defined and shared by the research team before starting any activity in the field. They must be congruent with the ones stated in the project. To avoid false expectations and eventual frustration, all partners in the PRA (e.g. extension officers, local authorities and farmers) should be informed of the purposes of the project and the PRA, before becoming involved in these activities.

Defining of tools to be used. PRA includes a diversity of tools, such as field visits, group discussions, one-to-one interviews, participatory mapping, construction of a flowchart of problem-cause interactions etc. However, for all of them, the research team needs to define in advance detailed methodologies and the expected results.

Identifying of sources of information and techniques used for collection. All partners do not have equal access to the information sought. Therefore, an effort to identify who can provide the most reliable information for a given topic/subtopic, and how to get it, must be defined by the research team. This is important when responses need to be cross-checked.

Planning field visit(s). The field visit(s) should be arranged through the partners living in the site (i.e. local authorities, government officers and NGO representatives). The researchers should prepare a list of activities and information to be collected in each visit and the sources of information and techniques to be applied (Table 5.1). Details on how to prepare the field visit and guidelines on how to conduct the field-work are shown in Boxes 5.2 and 5.3, respectively.

Table 5.1. Sources of and techniques to obtain information in participatory rural appraisal.

Note: SSI = semi-structured interviewing
MM = mobility map
VP = village map
DRD = daily routine diagram
VM = village meeting
LA = livelihood analysis
T = transect
PR = preference ranking
SC = seasonal calendar
MR = matrix ranking
TT = time trends
HP = historical profile
Source: Soedjana (1998).

Box 5.2. Some guidelines to prepare the field visit in participatory rural appraisal Before going to the field: Decide on which site(s) or community(ies) are included in the field visit. Consider representativeness (cultural and geographical) of the recommendation domain, and accessibility Establish contacts with district authorities and community leaders Clarify goals and objectives of the PRA field visit to local authorities and community leaders (what they can and cannot expect) Ask the village leader to introduce the research team to the rest of the community, when they arrive Make sure that the community knows when the research team will be arriving and how long will stay Ask for the date and time that will be more convenient for all local people (men and woman) to schedule the fieldwork Make initial arrangements with the community for interviews and other PRA related activities in terms of time schedule, food and accommodation. These should be later informed to all members of the research team Design a PRA research plan Design the PRA techniques to be used Form sub-groups among the research team, each will have a team leader. Proper consideration should be given to mix people based on language, gender, professional experience and personality Divide the community in sectors and distribute responsibilities among subgroups to do the survey in each sector One person should assume the responsibility to supply materials Enumerators or interviewers need to study the questionnaire. It will help to conduct the interview as a casual conversation, if possible not taking notes in front of the farmers, as a way to prevent inhibition from them In the field: Start the fieldwork by obtaining broad background information from the community leader and other key informant. This will form the basis of knowledge for further inquiries. A community leader or a well-respected member should introduce the research team to the farmers participating in the interviews and/or focus group discussions. In some cases the extension agent could also play that role. During the meeting, introduce the objective of the PRA to the community. Start with a problem and solution ranking exercise. Afterwards, make arrangements for group and one-to-one interviews, as well as for the rest of the activities The team splits into groups of 2 or 3 persons, to start applying the different procedures (e.g., interviewing, observation, transect, mapping). The team meets during evening hours to discuss preliminary findings and to reassign tasks. Local people should be invited to these feedback meetings On the last day, the team analyses and summarizes the findings and discusses conclusions with the community

Box 5.3. Guidelines for carrying out the fieldwork in participatory rural appraisal Plan the PRA fieldwork to be carried out in 3 to 5 days Organise small teams with a good mix of age, gender, personality and professional background, if possible include local people in each team Each small team must choose its leader Before going to the field, discuss with the research team the secondary information collected in advance (secondary data) Based on the characteristics, knowledge and experience of community members and key informants, identify and select carefully those who must be approached by the research team Start the fieldwork with something simple, such as direct observation, interviews with key informant, village mapping, transect etc. Improve the quality of interview information by combining with direct observation Keep a checklist to remind team members of important issues that need further inquiry Only ask necessary questions, those that will provide information you can not collect by other means Fieldwork has to be adapted to the community member's pace, meeting them when at their convenience. Do not impose your schedule If the research team has not had experience in PRA, it is highly recommended they to interact with other colleagues who had done it, and do some practice work The group should be able to review the list of topics and questions to be asked after the first round interviews. Identify additional information needed Agree with each subgroup on a plan for every day's fieldwork, based on analysis of the information collected The group team leader review fieldwork daily, going over notes, evaluating achievements and the methods applied. Discuss mistakes, lessons learned and what needs to be changed Refine, modify and experiment with different techniques to respond to the changing focus of the PRA as the research team goes along. Remember, not all communities are the same, but also there are difference between groups within a community If time allows, have second and third meetings with the same people to get deeper understanding Use analytical techniques to summarise findings and to compare information from different informants on the same topics Discuss your analysis with key informants to confirm and cross check your findings and conclusion Weigh the relative importance of the information gathered The team leader, but also the whole research team should be critical with themselves Stay overnight in the community to bring you closer and good interaction with its members Rethink how and in what form the results will be used Show interest in learning from the people, respect the community members and their knowledge. Ask for advise and be sensitive Meet as a team to prepare diagrams and charts, and to summarise the main findings Hold participatory assessment meetings with the community to review and discuss the findings, as well as to agree jointly on further steps to be taken

Preparing a list of topics to be discussed with the focus groups. The information to be collected in the discussions with the focus groups must be defined by the research team, based on the hypotheses formulated and data needs. The team should define the topics and sub-topics to be covered, and the relevance of each. 

Preparing group meetings. To be effective at gathering the information required and efficient at using the time assigned to the meeting(s) with partners, the team should consider the following points as guidelines (CIAT-FSP 1995):

• Carefully select the participants and invite them personally or in writing, indicating clearly the purposes of the meeting. Local authorities, community leaders, government officers and/or NGO representatives should be involved in these contacts.
• The site, date and time must be the most convenient for the community (not necessarily for the research team).
• Elaborate a specific plan for the meeting. It should include the activities to be developed and the time allocated to each activity, identify responsibilities for each member of the research team, and define the type of products expected from each activity and from the whole meeting. The flow chart of events shown in Figure 5.2 can help planning the meeting.
• Prepare the required aids and forms for recording the information obtained at the meetings. 

Conduction of the meetings. The members of the research team function as facilitators in the group meetings. Their responsibilities are therefore to guide discussions, assure wide participation, promote the coverage of all topics and subtopics previously defined, and summarise discussions when needed. Some suggestions on how facilitators can stimulate and promote participation of farmers in the discussions are listed in Box 5.4.

Source: Adapted from CIAT-FSP (1995).

Figure 5.2. Flowchart of events to be considered in a meeting with farmers for diagnosis of crop–animal systems, using the participatory rural appraisal technique.

Identifying problems. One of the expected outputs of PRA is the identification of biological and socio-economic factors or inefficiencies in the use of resources, which restrict the productivity and sustainability of the systems under study. Therefore, the research team should look for mechanisms to assure their identification by the different groups participating in the activity.

Preparing an inventory of relevant indigenous knowledge. Indigenous knowledge (IK) embodies the capacity of farmers to interpret processes under the local environment. It has been accrued over time and is a critical aspect of the community culture, tradition and way of life. In case local options to solve a given problem are available, these should form the basis for the design of technological interventions, which have a greater likelihood of being accepted, adopted and maintained by farmers. However, as IK has mostly been ignored by the scientific community, provisions should be taken to gather, document in a comprehensive manner and analyse the available indigenous technologies. PRA is just the starting point for this inventory effort.

Setting up a forum for participatory diagnosis and technology development. The purpose of such forum is to bring together farmers, researchers, extension workers and other stakeholders and let them classify and prioritise jointly the problems identified in different groups, analyse the indigenous knowledge options and develop the research agenda.

Tools used in PRA

Several tools are used for systems characterisation in PRA. Besides the field visits and group discussions, whose methodologies have been already discussed, some complementary tools used when working with the groups in the community are: diagrams and maps, seasonal calendars, long-term calendars, flowcharts of problem-cause interactions and transects (CIAT-FSP 1995).

Diagrams and maps. Diagrams and maps are simple schematic devices, which present information in a condensed and readily understandable visual form. They summarise data in such a way that it can be used at all stages of the site description study, for research and development planning, joint discussions and analyses of information. They are useful in helping identify problems or opportunities in specific areas, time periods or activities and in clarifying issues under discussion between the research team and community members. The most widespread diagrams and maps commonly used include: the village mapping, transects, seasonal calendar, time trends, historical profiles, mobility maps, daily routine diagrams, livelihood diagrams and flow diagrams.

Source: CIAT-FSP (Unpublished data).

Figure 5.3. Seasonal cropping calendar for an upland rainfed site in Leyte, The Philippines.

Source: Palacpac (1994).

Figure 5.4. Cropping pattern and feed resources flow model for a crop/animal system in a subhumid agro-ecological zone of the Philippines

Mapping is usually an entry-point activity to establish rapport between the project and the community. Asking the group of farmers to draw a map of their village, identifying the most relevant features as perceived by them, is a strategy that makes them to feel confident. In some cases it will be necessary to give orientation for them to include all elements that are needed, e.g. roads, waterways, hills, major land-use systems, houses and main buildings, but the team should also pay attention to those features that are important for the community. The map enables the research team to get an overview of the area, and can be used later in the group discussions as a support tool for analysis and planning. Mapping also facilitates interactions among community members and can help create group consensus and identify actual or potential conflicts in land use and tenure.

Mapping can be done on the ground or on paper. Mapping on the ground has many advantages, such as everybody can see it and suggest modifications and these can be made easily. Such a map usually contains more information than a drawing on a paper and can be expanded, because space is very seldom a limiting factor. Mapping on paper has the advantage that it can be kept for other group activities, but most frequently farmers feel shy participating in the drawing because of lack of practice. One alternative is that a member of the research team prepares a drawing on paper after the farmers finish the exercise on the ground, and asks for comments or corrections.

Seasonal calendar. A seasonal calendar is an important tool for the characterisation of crop–animal systems; it gives an idea of land use over time, interactions between crops and animals, seasonal demand for labour, variations in feed availability etc. The methodology applied for the preparation of the seasonal calendar is the same as that described for mapping. These calendars must represent the common practice, but also illustrate variations, which can be used as topics for discussion trying to obtain explanations on why practices differ. The variations might be indicative of innovation (Farrington and Martin 1987) or just a consequence of differences in availability of resources. In such case, it would be desirable to identify in the  maps of the village prepared by farmers where those variations occur. A seasonal calendar prepared in a rapid rural appraisal activity conducted in a community, which is considered representative of the rainfed uplands in the Philippines is shown in Figure 5.3. Changes in the use of feed resources during the year can be super-imposed on this such as in the example shown in Figure 5.4, although it does not correspond to the same site shown in Figure 5.3.

Source: Gabunada et al. (1998)

Figure 5.5. Problems of the livestock component as identified by farmers using participatory diagnosis in the rainfed uplands, Leyte, The Philippines.

Transect. A transect is another tool used in PRA. The research team and local people (farmers and key informants) walk through the land used by farmers, and observe, ask, listen, identify different zones, seek problems and possible solutions. These transects can follow a loop, a straight line or a winding route, trying to cover all the topographical variations in the village, and what informers and the research team consider key features (CIAT-FSP 1995). A transect is used to identify the different agro-ecological zones and land uses represented in the village, observe indigenous technologies applied, and differences or innovations, which could represent problems and opportunities in the different zones. If this exercise is repeated asking old members of the community for previous uses, it could serve as a reference for changes in land use and availability and on technological innovations.

The product of this exercise is a graphic representation of the landscape, indicating the different forms of land use practised. It could also have annotations on resource characteristics (e.g. topography, soil type, water sources and natural vegetation), agricultural components (e.g. crops, animals, trees, forages and aquatic plants), and farming practices, indigenous technology applied, problems and opportunities in each sector of the landscape.

Problem-cause interactions flowchart. The construction of flowcharts illustrating problem-cause interactions is another tool used in participatory diagnosis and is also relevant for the definition of the research agenda. The technique used for this is a `brainstorming exercise', in which farmers are asked to identify problems in their systems. It may help if this work is done by component (crops and animals), so problems related to the interactions will be identified either under one of the component or both. Once the brainstorming is over the research team should look for consensus to simplify the list before the group is invited to vote to prioritise the problems. Each participant could identify 2 or 3 problems he/she considers the most important and, based on this, a ranking is established. Some of the problems identified could be intermediaries to a major problem therefore the last exercise is oriented to establish those relationships, to define the causes. The results obtained using this tool in a community in the rainfed uplands of the Philippines are illustrated in Figure 5.5 (Gabunada et al. 1998).

Ranking and scoring. The techniques used for ranking and scoring options are analytical instruments, which allow management of information regarding preferences and perceptions of farmers with respect to problems and opportunities in the systems they manage. These also provide an insight on individual or group decision-making processes, and help identify the criteria people use to select items or activities. They are particularly valuable for determining how different or similar the perceptions and beliefs of individuals or groups within a community are. Ranking also helps to illustrate why different people have diverse criteria for making decisions and judgements about choices of technologies or any other things relevant to the farmer. Among the techniques used for this purpose are: preference ranking, pair-wise ranking and matrix ranking.

1. Preference ranking compares each individual problem against the others until they are ranked from highest to lowest. Preference ranking can be used to quickly identify problem areas, priorities and preferences of individuals and compare them with the assessments of others by assigning scores of 5 (most favourite) to 1 (least favourite). Each interviewed individual makes his/her own ranking using these scores and once all are registered, the `average' ranking is computed. An example of how data is tabulated and computations are made is shown in Table 5.2. Another option could be to rank different technological options using the same scoring criteria; in this case each respondent to give out laud the reason for ranking each option in that way (CIAT-FSP 1995).

Table 5.2. Application of the preference ranking technique to evaluate a given technological option in terms of the problems associated with it.

2. Pair-wise ranking produces a matrix by comparing each item with the other; the respondent defines whether a given option is better or worse than the one it is compared with. Pair-wise ranking should not have more than five options otherwise it becomes tedious (Ashby 1992) since with 5 treatments 15 comparisons will be made. To make pair-wise comparisons, each option is identified and annotated on a card; two cards are shown to the interviewee at the same time and he/she indicates his/her preference. Frequently the interviewee is asked for the reasons for his/her preference. These procedures are repeated until all possible combinations have been used and the results are tabulated.

An example of pair-wise ranking, including information on how computations are made to rank option is shown in Table 5.3. The example is based on the results obtained after interviewing five co-operators who compared five treatments in pairs. In each cell is the number of interviewees who preferred the option listed in the row, when compared with the one listed in the column. For example, when S1 was compared with S2, three informants said S2 was better than S1 and the other two said the opposite. Then in the cell defined by column S2 and row S1 there is a 3, whereas in the cell defined by column S1 and row S2, there is a 2 in parenthesis. Also, all five informants said that option S2 was better that S3 hence there is a 5 in the cell defined by column S3 and row S2, and a zero (0) in the cell corresponding to column S2 and row S3.

Table 5.3. Summary of hypothetical results obtained applying pair-wise ranking to different land-use options.

S1= upland rice; S2 = strip cropping; S3 = alley cropping; S4 = hillside; S5 = conventional practices.

To compute the total score assigned to S1, sum all the values in the corresponding row. For example, the score for S1 equals 3+4+3+3 = 13, and for S4 is 2 + 1 + 1 + 4 = 8. After doing all the computations, the option with the highest score is ranked as number 1. In the case of the example (Table 5.3) the most preferred option is S2, with 16 points, followed by option S1 with 13 points. Although the table does not include the reasons for the preferences, these could be also analysed to identify why the option was ranked in a given position.

3. Matrix ranking compares items or technological options against selected criteria that the same interviewees considered relevant for judging those. Matrix ranking can help the research team visualise and determine the relative importance of a set of problems and opportunities for management of crop–animal systems, land use, cropping patterns, natural resources and others.

When matrix ranking is applied, there should not be too many options and criteria for evaluation, since the responses of informants may become mechanical due to fatigue (Ashby 1992). In matrix ranking scores are assigned based on the relative suitability of the option to each criterion. The highest score (or the first rank) is given to the option, which fits the best according to the criterion under consideration, and the lowest to the one most poorly suited or absolutely unsuitable.

Special care needs to be taken in defining the way questions are formulated in relation to the criteria used. For example, if the criterion refers to the demand for cash to implement an option-assuming it is a scarce resource-the question should emphasise less demand for cash then the highest score will be obtained by the option, which requires the least; otherwise the scoring may be misleading. Once all criteria have been considered and all informants interviewed, the data are summarised adding the scores obtained by each option, and the option preferred will be the one with the highest score.

An example of using this tool to evaluate five dry-season feeding strategies for ruminants, using five criteria previously identified by evaluators is given in Table 5.4. Alternatives were ranked based on all criteria but using one at a time. As the computation is made by simply adding the scores obtained using each criterion, the same weight is given to all and this may not be the rationale when decisions are taken. To overcome this problem the scoring can be arranged so that the researcher can identify not only the option evaluators prefer, but also the most relevant criterion for them. To do this evaluators are asked to distribute a total score (e.g. 100) among all cells resulting of the combination of options and criteria. Doing this gives a weighted evaluation of options, considering the relative importance interviewees give to different criteria. Although this procedure seems more complicated, it has been successfully used with farmers in different developing countries, frequently using a simple way to allocate scores (e.g. small stones or grains) in a two-way table.

Table 5.4. Example of the results obtained after applying the matrix ranking method to evaluate different dry season feeding strategies for ruminants.

Note: Evaluators were asked to distribute a maximum of 100 points among different options and criteria.
a. Higher score to those options that require less external inputs. The criterion used by farmers is that cash is not easily available, especially during the dry season.
b. Lower score to those activities that demand more labour. The criterion used by farmers is that any new activity in this regard will compete with other activities they are already involved in.
c. Higher score to those options with less demand for additional equipment or infrastructure.
d. Lower score to those options which require more technical support.

The same example as Table 5.4 is shown in Table 5.5 but after evaluators considered a relative weight for each criterion. In this way it was possible to identify the criterion evaluators gave more importance to (e. effectiveness to prevent liveweight losses was the one with the highest score), but also the ranking of options changed due to the different weight evaluators gave to each criteria. When each criterion was considered independently, feeding only crop residues was the option selected (Table 5.4), whereas the use of these plus tree foliages was the preferred alternative when the relative importance of each criterion was also taken into account (Table 5.5).

Table 5.5. Results obtained applying the matrix ranking method for the evaluation of different dry season feeding strategies for ruminants, identifying the criterion evaluators that give the highest importance.

Note: Evaluators were asked to distribute a maximum of 100 points among different options and criteria.
a. Higher score to those options that require less external inputs. The criterion used by farmers is that cash is not easily available, especially during the dry season.
b. Lower score to those activities that demand more labour. The criterion used by farmers is that any new activity in this regard will compete with other activities they are already involved in.
c. Higher score to those options with less demand for additional equipment of infrastructure.
d. Lower score to those options which require more technical support.

5.5.2 Structured surveys

Static or cross-sectional survey

The static survey is a questionnaire-based survey covering a sample of farmers who are representative of the community or communities under study or a sample of individuals involved in a given activity the research team is interested in (e.g. milk processing and marketing; rice production). It is called `static' because it refers to the conditions of the farms/communities at a given point in time and the people who are interviewed respond based on what they recall or what is the actual practice in their system. The static survey is usually recommended as a follow-up of the PRA to get information that can be quantified and therefore used for valid statistical inferences with respect to the hypotheses formulated.

The correct application of the static survey requires the use of appropriate sampling procedures, such as randomness in the selection of sampling units and the use of a sample size sufficiently large to make valid inferences (Jabbar et al. 1997). The selection of a representative sample generally requires a comprehensive sampling frame, which is a complete list of the population of farmers from who the research team wishes to collect information. For example, if the study deals with crop–animal systems, then an updated list of all farmers practising these systems in a community should be obtained. This might be available at the local government offices, extension agencies, co-operatives etc. If this list is available samples can be chosen at random otherwise a list could be prepared based on interviewing key informants.

If an adequate or reliable sampling frame cannot be developed with this procedure, samples may be selected using the `select as encountered' procedure within the geographical area under study (Houseman 1975). The research team may identify routes, choose some to be sampled, and then select at random which farmers will be interviewed within those routes (a sort of nested sampling). If there are hypotheses related to different strata among the population (e.g. based on resource availability), provisions should be made to assure this criterion is considered when choosing samples (e.g. stratified sampling design).

Another important element for the static survey is the development of an appropriate questionnaire for data collection. Although the research team can use previous instruments as a reference, these must be adapted to the objectives of the project, the hypotheses formulated and the local conditions.

The static survey is complementary to PRA therefore information already obtained by PRA should not be asked again in the static survey unless specific issues need cross checking or quantification is needed. Furthermore, when designing the questionnaires the research team has to prioritise the topics to include. The exploration of a wide range of issues should be avoided since some questions may be irrelevant to the objectives of the project or may cause the quality of information collected to decline due to informant or interviewee fatigue. For the definition of which variables to include in the questionnaires, the reader should refer to the recommendations included in Table 4.3.

Additional guidelines for designing a questionnaire for data collection are listed below (Lapar 1999).

Closed vs. open questions. There are no restrictions in the type of questions to be included in the survey instrument. They could be either focused to a set of options `closed question' or give the person being interviewed the opportunity to respond according to his/her perception on a given issue `open question'.

An example of a closed question is when do you feed rice straw to your animals? Alternative options to answer it are: (a) never, (b) only during the dry season, (c) when other activities do not allow me to harvest fresh forage and (d) the whole year. An example of open question to complement the previous one could be how do you provide rice straw to your animals? In that case, each farmer can describe his/her own practice.

Leading questions. The questionnaire should not include questions that push or encourage farmers to respond positively or negatively to a particular issue. This aspect must be also emphasised to the interviewers so that they can avoid this type of questioning when they collect primary information. For example, in the topic illustrated in the previous paragraph, a leading question could be `animals eating only rice straw lose weight, what do you use to supplement it?' Another leading question in this respect could be one in which some supplements commonly used in rice straw-based diets are listed. These types of questions may prevent researchers discovering the diversity of technologies practised in the community, some of which may form the basis for new technological options.

Specificity of questions. To facilitate the reliability and/or focus the answers of the farmers, questions should be specific and use local expressions and common units of measurement. For example, if researchers are interested on the average milk production per cow, it is difficult for a farmer to recall those figures, and many may not be even familiar with the word average. There are indirect but specific ways to get this information. One is to ask the farmer to identify his/her best cow, and ask about the level of milk produced by it in the peak of lactation, and at the end of the lactation period. The same can be done for the one with the lowest production. Another option is to ask for the total milk produced by the herd in the best and the most difficult times of the year, and how many cows are milked during those periods of the year, and based on that information the enumerator can estimate the average.

Order of questions. The questionnaire should follow a logical order, starting with the most general aspects of the farm, to specifics details regarding the crop and animal enterprises. Sensitive questions should be asked at the end of the interview. However, as it is desirable to conduct the interview as a friendly and relaxed conversation, the interviewer should not interrupt the farmer if he/she is giving information that was supposed to be obtained later. The interviewer should take note of the data provided and continue on that topic until all the information needed is collected.

Review, testing and training. The questionnaire has to be reviewed by all the members of the research team, but should also be reviewed by other colleagues with experience in these matters. It should first be tested with a small sample of farmers, and the research team should check beforehand how answers are encoded. Interviewers need to be trained not only in the questionnaire itself and the way to encode answers, but also in best ways to approach farmers. To achieve this the interviewers/encoders should be preferentially from the same areas/communities under study, since they will generally be more readily accepted and trusted by the local people, and will be familiar with the way farmers like to be approached.

An example of a questionnaire used for household survey to study crop–animal systems practised by smallholders in five countries of South-East Asia can be found in Appendix 5.1. The corresponding encoding guidelines used for that questionnaire are given in Appendix 5.2.

Dynamic or longitudinal survey. A dynamic or longitudinal survey is a detailed case study or set of case studies, each in a small number of farms, carefully selected as representative of the target group(s) relevant to the issue under consideration, but not necessarily representative of the whole population (Mullins et al. 1994). It is useful for documenting variations in time (along a year or between years), in practices applied at the farm level, and to analyse why these are practised, to understand complicated interactions among components, but is also valuable for monitoring variations throughout the year in quantitative attributes.

Dynamic surveys require frequent interactions between farmers and members of the research team; therefore the farmer should be clear of the purpose of those visits and the potential benefits he/she can derive from the study. Co-operators must be periodically informed of the results obtained in their farms. Although the research team would not want to influence changes in farm management with this type of survey, it is quite difficult to prevent these to occur as a result of the interactions between the farmers and the team, and this may eventually bias the results of the study.

The selection and training of the technicians and the strategy used by them to approach farmers and collect information is critical for the accomplishment of the objectives, since in this case the quality of the information collected is more important than the number of farmers covered. Therefore the number of farms a surveyor manages should be such that he/she can visit each at least biweekly, and spend ideally half a day in it, preferentially helping the farmer in his/her routine activities, but at the same time collecting information.

Technicians responsible for dynamic diagnosis often have to take direct measurements of the amount of inputs applied and the products obtained. Frequently, there are conflicts of schedule among farms, either because a relevant practice is applied at the same time in more than one farm (e.g. the planting season tends to be short in subhumid agro-ecological zones), or because an activity, such as harvest, takes more than one day (Valdivia 1990). Also, there are events that are not under the control of the farmer or the technician (e.g. dates of calving and mating or date and quantity applied of a given agrochemical). For all these, it is useful if one member of the household is trained to fill out simple farm records (Farrington and Martin 1987).

A voluminous amount of information is collected in a dynamic survey and therefore provisions must be taken to code and input it as frequently as possible. This will help not only to prepare partial reports for the partner farmers, but also to detect inconsistencies that can be corrected early and not after all the fieldwork has been completed (Valdivia 1990).

Special efforts need to be devoted to the definition of the encoding system. Sometimes data has had to be input twice due to the incomplete definition of data codes (e.g. the time factor was not considered, even though it was a variable collected repeatedly in the same farm). Most statistical analysis packages have a built-in data entry and management mechanism but these are usually only suitable when a small volume of data is involved (Jabbar et al. 1997) which is not the case in a dynamic survey. It is therefore more convenient to use a data management package and to check if the database management system to be used can be interfaced with the statistical analysis package planned to be used. Several database management packages are available. Some used in the past were dBase and Panacea, but more recently Excel, Access and Lotus are the most commonly used. Excel and Access are part of the Microsoft Windows package; therefore archives can be easily transformed from one system to the other.

5.6 Methods of analysis

The methods of analysis to be applied are a function of the research objectives, the type of data collected and the hypotheses formulated. The statistical procedures to be used for the analysis of data need to be identified before data are collected in the field. The opposite is usually the case with researchers approaching statisticians to ask which analytical procedures could be applied to a set of data already obtained.

As characterisation entails quantification, simple analytical tools such as descriptive statistics (percentage, mean, mode, median, standard deviation and coefficient of variation) will often be enough to quantify the descriptors; however, to do so and to perform other statistical analyses it is necessary to consider the diversity of data collected. Some attributes are categorical (e.g. type of feeds used to supplement the basal diet) or even of binary nature (e.g. provide supplements or not, pregnant or non-pregnant and preferred or non-preferred), and others are continuous (e.g. performance indicators such as milk production and liveweight gain). Based on the nature of the attributes and the hypothesis to be tested, appropriate procedures for statistical analysis must be defined.

5.6.1 Reduction of the number of variables

Household surveys usually result in the collection of too many variables that eventually hinders the ability of the research team to , interpret and understand the multiple relationships existing among variables within households and among households. This problem is exacerbated when there are more components in the farms under study and more diverse backgrounds are represented in the research team. Therefore, before starting to analyse data the research team should reduce the number of variables.

A priori, the research team can determine whether a given attribute is relevant and/or whether reliable information can be obtained through the household survey technique. If the answer is no, then that variable should not be included in the survey questionnaire. Once data collection is completed, the research team can revisit the list of variables and check for their relevance and/or reliability. Other options to reduce the number of variables a posteriori are to:

• use statistical procedures which may result in building indexes which summarise relevant variables and discard those that provide limited information (e.g. principal component analysis)
• build indexes with a specific meaning, incorporating several variables that have some degree of association (not necessarily statistical association).

5.6.2 Statistical analysis for determining baseline status

The determination of the baseline status entails of descriptors that characterise a population at a given point in time, therefore simple analytical tools can be used, but the appropriateness of the method will be a function of the type of data collected. Some attributes are categorical or even of binary nature (e.g. provide supplements or not, pregnant or non-pregnant, preferred or non-preferred), and others are continuous (e.g. performance indicators such as milk production, live-weight gain).

For categorical data, results can be expressed either as frequencies or percentages and this can be applied to the whole sample or to separate subsets of the sample. An example of this on the use of rice straw for animal feeding is shown in Table 5.6. If the variables are continuous, then measures of central tendency (e.g. mean, mode and median) and dispersion (e.g. standard deviation, standard error, coefficient of variation and range) for the whole sample or subsets may be appropriate (Steel and Torrie 1980). An example of baseline productive performance parameters for the cattle component in a sample of dairy farms is shown in Table 5.7.

Table 5.6. Baseline information on the use of rice straw for animal feeding in a sample of farms covered by a household survey.

Table 5.7. Baseline performance parameters for the cattle component in a sample of farms covered by a household survey.

5.6.3 Testing of hypotheses

As indicated in Section 5.3, the characterisation process involves testing hypotheses using the information collected during surveys. However, the procedure used for this purpose will be a function of the type of variable (categorical or continuous), the number of classes (two or more) and the nature of the hypothesis.

Testing of hypotheses with categorical data

Many of the questions included in a household survey refer to a categorical rather than a continuous characteristic (e.g. two villages) the researcher could use either the normal approximation analysis or the chi square criterion for a 2 × 2 or four-fold contingency table (Steel and Torrie 1980), as illustrated in Box 5.5. The same procedure could be applied to a greater number of classes, but making independent comparisons each with one degree freedom, or even grouping some of them, using the principle of the additivity of . Details on how to compute this can be found in any statistics text.

Testing hypotheses with continuous data

The analytical option to be used for the analysis of continuous data could be either a simple t-test in case only two classes are considered and analysis of variance with an f-test for more than two classes. The analysis of variance is also used when more complex treatment structures are evaluated (e.g. comparisons among systems within sites).

As data sets based on household surveys are usually unbalanced (i.e. different number of observations in each class), least square procedures are frequently used for the statistical analysis of household survey information. However, in many situations the data available may lead to biases in the estimation of parameters (Draper and Smith 1981) or the tests of hypotheses (León-Velarde and Quiroz 1994). You should therefore consult a statistician during the planning stage and also for the analysis and interpretation of data.

The problems most commonly faced when analysing data using least square procedures are the following.

Lack of normality. One of the basic assumptions for the analysis of variance is that data are normally distributed, with mean equal to zero and variance equal to one. However, in many cases the observations collected have a skewed distribution and it is not uncommon for the standard deviation to be greater than the mean. This affects the test of significance of the parameters and the estimation of confidence intervals for the parameter estimates. Under these circumstances data transformations based on the analysis of the residuals should be used.

Heterogeneity of variances. Another assumption in the analysis of variance is that all observations have a common variance, and the ordinary least square procedure applied for this variance gives the same weight to each observation. Again, the solution to this problem is transformation of data based on the distribution of residuals.

Co-linearity. Many attributes used for characterisation could be correlated and this can contribute to increase the variance and may change the magnitude and even the sign of a given parameter estimate. One way of detecting if variables are correlated is to estimate the correlation matrix and look for the association between each pair of variables; however, some association cannot be detected using this procedure. For this reason you should run a principal components analysis to develop new variables that integrate those that are correlated (León-Velarde and Quiroz 1994). This is discussed later in this paper.

Correlated errors. This problem is mostly faced in dynamic surveys where data are collected repeatedly in the same units. Under these circumstances variables tend to present correlation between their residual values, which in turn diminishes the precision of estimates. This could even invalidate any test of significance of the estimated parameters. This problem can be overcome by using time series analysis and generalised least squares (GLS) procedures.

5.6.4 Multiple and partial regression and correlation analyses

Household surveys include a wide range of variables, some of which could be correlated, while some could be used for the prediction of others. Computer packages make it relatively easy to run multiple correlation analysis, and this is normally how household survey data are . However, this analysis determines the degree of empirical association among a set of variables, therefore a high correlation value between two variables does not mean that there is a causality relationship between them (Gomez and Gomez 1984).

Multiple regression is also a procedure frequently used to analyse household survey data, but the research team needs to identify the adequate dependent and independent variables and rationalise the basis for running the analyses. Stepwise procedures for multiple regression analysis are commonly used for this purpose (Rawlings 1988). The advantage of applying such techniques is that these can help reduce the number of independent variables used for the prediction of a given dependent variable, dropping from the equation those that make a small contribution to the coefficient of determination (R2). However, after dropping variables it is necessary to repeat the analysis because the equation contains partial regression coefficients, whose magnitude varies depending of the presence of other terms in the regression equation.

5.6.5 Multivariate analyses

Multivariate analyses are tools for the classification and typification of households in the benchmark site based on their common responses to the set of variables in the questionnaire. In this module three procedures used in multivariate analysis are discussed, namely principal components, cluster and discriminant analyses.

Principal component analysis

Principal component analysis is a procedure used to present a set of variables in terms of a smaller, more manageable set of linear combinations of the variables, which retains as much of the information of the original set as possible. Furthermore, principal component analysis is a tool for controlling co-linearity problems among different variables, a frequent problem in household survey data sets.

Mathematically, principal components are linear combinations of the original variables (X1, X2, X3, .... Xn), which can be represented as:

Y₁= a₁ X₁+a₂X₂+....+a_n X_n

Each principal component has a maximum variance and satisfies the condition of, i.e. in other words, the sum of the coefficients equals one (1). All principal components are orthogonal among them, i.e. the correlation between each pair of principal components is zero.

Although p principal components are required to reproduce the total variability, often a small number of them (k principal components) can account for much of this variability. If so, there is almost as much information in the k components as there is in the original p variables. The k principal components can then replace the initial p variables, and the original data set, consisting of n measurements on p variables, is reduced to one consisting of n measurements of k principal components (Johnson and Wichern 1982). The latter are the ones used for subsequent analysis, such as cluster analysis.

An example of the results obtained after running principal component analysis to a household survey data set resulting from a research project which worked with dual-purpose cattle production systems in Mexico (Anderson and Santos 1997) are shown in Tables 5.8 and 5.9. The Eigen values for the principal components identified and the relative contribution of each to account for the total variability are also shown Table 5.8. Notice that the first two principal components explain 99.86% of the total variability, whereas the contribution of the others is negligible; therefore, it is advisable to concentrate only on the first two principal components.

Table 5.8. Eigen values of the covariance matrix for a set of livestock farms surveyed.

Source: Anderson and Santos (1997).

Figure 5.6a. Average distance between the last two conglomerates formed as a function of the number of

Table 5.9. Description of the first five Eigen vectors related to each principal component in a set of livestock farms.

Source: Anderson and Santos (1997).

Figure 5.6b. Changes in R² as the number of clusters varies.

The variables included in the first five principal components are listed in Table 5.9, and those that make the greater contribution to each principal component are highlighted. Based on this, the most important attributes to discriminate among farms are stocking rate, use of concentrates, animal records and seasonal mating (included in PC1), and the total farm area, pasture area, the use of technical assistance and off-farm work (included in PC2). Based on the coefficient assigned to the elements of each principal component new variables can be estimated.

Cluster analysis

Clustering is the grouping of objects based on their similarities. In household surveys, clustering of farms is a mechanism used to identify and groups of households that are similar but quite different from those belonging to other clusters. Each group of farms within a cluster may constitute a recommendation domain, so that specific technological recommendations are suitable for all members of the cluster.

Variables that are largely the same for all households have little clustering power, whereas those manifesting differences from one household to another are more likely to induce strong distribution. Cluster groupings depend on the selection of variables. Cluster groups could be poorly defined when a relevant variable is ignored, or by including a non-relevant one. Furthermore, the scale of the variables could influence the cluster composition (i.e. variables with a larger scale will dominate the conformation of clusters). Therefore, to give a uniform importance to each variable you should standardise all variables before running cluster analysis. Standardisation of a variable is done by subtracting its mean and dividing by its standard error.

There are different procedures (algorithms) for measuring the distance between classes, which is the basis for defining clusters, but based on our own experiences we recommend the Ward procedure (SAS 1998), which estimates the variance within each cluster and tries to minimise it.

One important decision in cluster analysis is to define the adequate number of groups (clusters). The distance among groups becomes greater as the size of the clusters increases (Figure 5.6a), but also increases the variability within clusters. For each variable it is possible to estimate the sum of squares within groups (Wk) and the total sum of squares (Tk), from which the multiple correlation coefficient (R2) can be estimated as:

The multiple correlation coefficient decreases with the reduction in the number of clusters; the researcher needs to decide on how much reduction in R2 he/she is prepared to accept. In the example shown in Figure 5.6b, it seems reasonable to accept from three to five clusters as there is an important decline in R2 when only two clusters are considered. The output from cluster analysis includes the grouping of units in the different clusters and the statistics for the different attributes used for clustering. It is advisable to run analyses of variance for the different variables, using clusters as a class variable, to detect potential differences among groups.

An example of the results of using cluster analysis for the information gathered in a household survey involving 100 farms practising crop–animal systems in the Red River Basin of Vietnam is shown in Table 5.10. Six clusters were identified: two grouping high income households, with and without special activities (e.g. fruits, garlic production, cattle production, fish farming and income from off-farm labour); two clusters with medium income households, with and without special activities, but did not include animal production activities; a fifth cluster grouped all low-income farms, which do not have special activities; and finally a sixth cluster that covers those farms, that do not fit in to the previous groups, which have the highest income but are strongly influenced by commercial swine operation.

Considering the components of the farming systems represented by the different clusters, the level of income and availability of resources, it is clear that the same technological options do not necessarily apply to all farms. For example, even though in all clusters there is a pig component present in the farms, alternatives demanding high cash expenditures may be adequate for clusters 1 to 4, but not for cluster 5. A more detailed analysis of other attributes and farmers' expectations and concerns within each cluster will

Figure 5.7. Simplified representation of a crop–animal system.

give a better indication of the type of interventions to be proposed. The research group should also decide on which cluster(s) to target in their research efforts.

Table 5.10. Clusters identified in a household survey involving smallholder crop–animal systems in the Red River Basin of Vietnam.

Source: Chan-Dung (1993).

Discriminant analysis

Once clusters have been defined with a sample of households, other households not used for the analyses can be assigned to any of the cluster groups based on the similarities they have in attributes that characterise each cluster. As each unit may have attributes in common with more than one cluster, then using mathematical procedures (i.e. Mahalonobis distance estimates) it is possible to determine which group each unit is more likely to belong to, with a probability factor associated with this decision. The greater the probability level, the more the unit is a typical representative of the cluster. An example of the outputs after applying discriminant analysis to a set of farm is shown in Table 5.11. Notice that farms 1, 2 and 3 have been assigned to clusters 4, 5 and 2, respectively, since these showed the highest probabilities to belong to each one of those.

Table 5.11. Example of the outputs of discriminant analysis classifying farms within clusters.

 5.7 Flow diagrams: A tool for the characterisation of crop–animal systems

Any system can be defined in terms of its structure (the elements present) and function (how the system works). The elements of a system are inputs, components, interactions, outputs and limits. Figure 5.7 gives the simplest graphic representation of a crop–animal system in which all its elements are generally identified. Under the general category of inputs are: the elements needed for primary production (e.g. light, water, temperature and plant nutrients), feed supplements, medicines, fertilisers, pesticides, fuel etc. The components are crops (including forages) and animals (all species managed in the farm). The interactions are the ways in which components relate to each other, i.e. the functional relationships between crops and animals. Outputs are the products from crop (e.g. grains, tubers and residues) and animal origin (e.g. milk, meat, eggs and hives) and services derived from the animal component (e.g. draft and hauling). The limit refers to the boundaries within which those components function.

The relationships between components can be of different types: direct chain, cycle chain, competition and complementation. When the products of one component are used by another component they are related to each other in a direct chain, but if at the same time a product of the second component is used by the first, it is a cycle chain. If two components make use of a third component, there is a competitive relationship, or competition between the first two components; and if two components are used by a third component, the first two are complementary. Illustrations of these relationships among components in crop–animals systems are shown in Figure 5.8.

Figure 5.8. Types of relationships between components.

A flow diagram is a tool used to represent systems graphically. It could be helpful in the characterisation process since the diagrams are built based on the information collected from field visits, group discussions, one-to-one interviews and other procedures. Symbols are used in flow diagrams to represent the elements of the system. You should use some conventional symbols to enable sharing of information with other research groups. Some of the symbols proposed by Hart (1980), which are widely accepted and based on the circuits language also applied in ecology are shown in Figure 5.9.

Adapted from león-Velarde and Quiroz (1994).

Figure 5.9. Symbols used for the construction of system's flow Diagrams.

The construction of flow diagrams is done in two phases. It starts with the qualitative representation of the system and is later complemented with quantitative data. The first step is to define the limit of the system. Next, identify the household subsystem, including the family nucleus and other members of the household and the production resources. The third step is to identify the agro-ecosystems (e.g. individual crops or multiple cropping arrangements and animal components), including information on the area used for each enterprise. Once all the components have been identified, the inputs can be listed along with the functional relationships among inputs, the household subsystem and the different agro-ecosystems. The last step could be the identification of the outputs of the system.

A crop–animal system drawn based on the information collected through field visits, group discussions and one-to-one interviews with members of a community is shown in Figure 5.10. The same crop–animal system is shown in Figure 5.11 using the symbols proposed by Hart (1980).

Figure 5.10. Graphic representation of a typical crop–animal system based on information collected in field visits and group discussions with the community.

The quantitative flow diagram is constructed using as a basis the qualitative diagram already described. In a quantitative diagram are included yields and prices of products produced and the amount of inputs used and their costs, for each agro-ecosystem or enterprise. In addition, any other source of income for the household is incorporated in the flow diagram. With this information it is possible to estimate the costs and gross income for the whole farm and for each individual enterprise (León Velarde and Quiroz 1994). The quantitative flow diagram for the same crop–animal system illustrated in Figures 5.10 and 5.11, but ranges for income and cost are shown in Figure 5.12. These ranges consider the variability observed for the different farms included in the static survey, which were used as the basis for the construction of the flow diagrams.

 Figure 5.11. Flow diagram representing the same crop–animal system shown in Figure 5.9.

Figure 5.12. Quantitative flow diagram for the typical crop–animal system represented in Figure 5.9.

5.8 Outputs of the characterisation process

The analysis and interpretation of the qualitative and quantitative data collected using the different procedures proposed for characterisation should result in the following outputs:

• An identification of the main systems , their resources (land, labour, and capital) and how they are managed. To describe these you must not only obtain position statistics (e.g. mean, median and mode) but also an estimate of how diverse these systems are (e.g. standard deviation and coefficient of variation). All these serve as a basis to define the recommendation domains where the results of the research effort could impact.
• An understanding of the biophysical, and institutional conditions under which the systems function, why they are managed in the way they are, what their limitations are and where the opportunities lie. In part, the hypotheses tested (depending on how they were formulated) lead to this output.
• A set of benchmark information needed for the ex ante evaluation of the biological feasibility, economical viability and social acceptability of potential interventions, and for the ex post analyses of the research and development efforts applied along the whole project.
• An inventory of the indigenous technological knowledge available in the communities under study and the rationale and opportunities for this knowledge to be used as a basis for the design of improved technological options.
• A definition of researchable issues to be addressed during the on-farm research planning activities and some ideas of the potential interventions to be tested.

5.9 References

Anderson S. and Santos J. 1997. De sistemas de doble propósito a nivel de la vaca, el hato y el productor. In: Lascano C.E. and Holmann F. (eds), Conceptos y Metodologías de Investigación en Fincas con Sistemas de Producción de Doble Propósito. Consorcio Tropileche, CIAT (Centro Internacional de Agricultura Tropical), Cali, Colombia. pp. 191-212.

Ashby J.A. 1992. Manual para la Evaluación de Tecnología con Productores. Proyecto IPRA. CIAT (Centro Internacional de Agricultura Tropical), Cali, Colombia. 101 pp.

Chan-Dung L T. 1993. Analysis of activities of farm households in a commune in Vietnam using multivariate statistical techniques. MS thesis, UPLB (University of Philippine Los Baños), Los Baños, The Philippines. 206 pp.

CIAT-FSP (Centro International de Agricultura Tropical-Forages for Smallholders Project) 1995. Training modules for farmer participatory research. CIAT- FSP Los Baños, The Philippines. (v.p.)

Draper N.R. and Smith H. 1981. Applied regression analysis. 2nd Edition. Wiley, New York, USA. 709 pp.

Farrington J. and Martin A. 1987. Farmer participatory research: A review of concepts and practices. Agricultural Administration (Research and Extension) Network Discussion Paper 19. ODI (Overseas Development Institute), London, UK. 88 pp.

Gabunada F.G. Jr. Balbarino E.A. and Obusa A.P. 1998. Farmer participatory research on forage in Matalon, Leyte. In: Stur W. Owen J.A. Kerridge P.C. Horne P.M. and Hacker J.B. (eds), Proceedings of the Second Regional Meeting of the Forages for Smallholders Project held at the Chinese Academy of Tropical Agricultural Sciences, Danzhou, Hainan, China 19–24 January 1997. Forages for Smallholders Project Technical Report 2. CIAT Working Document 173. CIAT (Centro Internacional de Agricultura Tropical), Cali, Colombia. pp. 59–79.

Gomez K.A. and Gomez A.A. 1984. Statistical procedures for agricultural research. 2nd Edition. IRRI (International Rice Research Institute). Wiley, New York, U.S.A. 680 pp.

Hart R.D. 1980. Agroecosistemas: Conceptos básicos. CATIE (Centro Agronómico Tropical de Investigación y Enseñanza), Turrialba, Costa Rica. 211 pp.

Houseman E.E. 1975. Area frame sampling in agriculture. USDA-SRS Series 20. USDA-SRS (United States Department of Agriculture-Statistical Reporting Service), Washington, DC, USA 79 pp.

Jabbar M.A. Tambi E. and Mullins G. 1997. A methodology for characterizing dairy market systems. Market-Oriented Smallholder Dairy Research Working Document 3. ILRI (International Livestock Research Institute), Nairobi, Kenya. 62 pp.

Johnson R.A. and Winchern D.W. 1982. Applied multivariate statistical analysis. Prentice Hall, Englewood Cliff, New Jersey, USA. 594 pp.

Lapar M.L. 1999. Crop–animal systems research in South-East Asia: Data collection for benchmark site characterisation. In: Devendra C. (ed), Proceedings Planning Workshop on the Crop–Animal Systems Project held at IRRI, Los Baños, Philippines on 1–4 June 1999. ILRI (International Livestock Research Institute), Los Baños, The Philippines. pp. 16–19.

León-Velarde C. and Quiroz R. 1994. Análisis de sistemas agropecuarios: Uso de métodos bio-matemáticos. CIRNMA (Centro de Investigación en Recursos Naturales y medio Ambiente), La Paz, Bolivia. 238 pp.

Mullins G. Rey B. Nokoe S. and Shapiro B. 1994. A research methodology for characterising dairy product consumption systems. Market-Oriented Smallholder Dairy Research Working Document 2. ILCA (International Livestock Centre for Africa), Addis Ababa, Ethiopia. 40 pp.

Rawlings J.O. 1988. Applied regression analysis: A research tool. Wadsworth and Brooke-Cole Statistical/Probability Series, Davis, California, USA. 553 pp.

Rey B. Thorpe W. Smith J. Shapiro B, Osuji P. Mullins G. and Agyemang K. 1993. Improvement of dairy production to satisfy the growing consumer demand in sub-Saharan Africa: A conceptual framework for research. Market-Oriented Smallholder Dairy Research Working Document 1. ILCA (International Livestock Centre for Africa), Addis Ababa, Ethiopia. 13 pp.

Rhoades R.E. 1987. Farmers and experimentation. Agricultural Administration (Research and Extension) Network, Discussion Paper 21. ODI (Overseas Development Institute), London, UK. 17 pp.

SAS (Statistical Analysis Systems). 1988. SAS/STATTM user's guide: Release 6.03 edition. SAS Institute Inc. Cary, North Carolina, USA. 1028 pp.

Sagar D. and Farrington J. 1988. Participatory approaches to technology generation: From the development of methodology to wider-scale implementation. Agricultural Administration (Research and Extension) Network Paper 2. ODI (Overseas Development Institute), London, UK. 50 pp.

Steel R.G.D. and Torrie J.H. 1980. Principles and procedures of statistics: A biometrical approach. McGraw Hill, New York, New York, USA. 633 pp.

Valdivia R. 1990. El de producción familiar: Caracterización. In: II Seminario Taller Enfoque y Análisis de Sistemas Agropecuarios Andinos. Proyecto de Investigación de Sistemas Agropecurios Andinos (INIAA-PISA). Puno, Perú. Serie Didáctica. Material de Enseñanza 4. Chapter III. pp. 1-14.