Introduction
General methods available for PSR household level data collection
References
Les enquêtes sur les ménages dans la recherche sur les systèmes pastoraux
Barbara E. Grandin¹ and Solomon Bekure²
¹Anthropologist;
²Team leader and Agricultural Economist, Arid Zones (Eastern and Southern
Africa) Programme, ILCA, Kenya
This section of the workshop deals with household level data collection in pastoral systems research (PSR). It first briefly reviews differences between crop production and pastoral production which affect research possibilities. It then briefly reviews basic types of methods that may be used to collect household level data. Finally for three important areas of household level data collection in pastoral production systems (i.e. livestock transactions, household budgets, and labour use) it discusses specific problems as well as special techniques which are available for data collection.
Farming vs. pastoral systems research
De Haan briefly mentioned some important differences between cropping systems and pastoral systems of production. Important differences also exist in research for these systems. It would be useful to reiterate some of these, particularly as they affect the collection of household level data.
Pastoral systems are marked by high mobility which causes difficulties in sample selection and long-term survey work. Logistic problems can largely determine possible sampling procedures both in terms of the general location studied and specific households within that location. Many areas are largely inaccessible to researchers on the ground, or would be so at a prohibitive cost. Given the high mobility, it is difficult to ensure that the same families can be followed throughout a long-term study. It will be more cost effective, ceteris paribus, to study more households in fewer communities than a few households in a large number of communities.
Pastoral systems are marked by the communal sharing of major resources such that the focus on the household as a unit of production is often more problematic with pastoralists than with farmers. Generally, in order to understand pastoral management practices and strategies, far more attention must be paid to the relationships between households in a system.
Pastoral systems, unlike cropping systems, are marked by the need for continuous inputs as well as the production of et least some continuous outputs. Farming systems research (FSR) studies tend to concentrate on the agricultural growing season, which can be as short as a few months. Thus in a fairly short period of time it is possible to observe all the major inputs to and outputs from crop production. This is not possible with pastoral production systems, which have daily outputs and inputs, the nature of which can change dramatically in the course of an annual cycle. It is not possible to observe these different inputs and outputs in a short period of time. (Parenthetically it might be noted that farming systems based on certain tree crops would resemble pastoral systems in this respect). If the research suffers from time constraints, these will be far more seriously felt in PSR. In the study of pastoral systems, at least the major seasons and their transitions (Swift, 1982; Grandin, 1982) must be identified and considered. If they cannot be observed more reliance must be put on the ability of the producer to recall and report events.
Pastoral systems are marked by long cycles of drought and post-drought recovery that are much less important in systems of annual cropping. More attention needs to be paid to long-term cycles in PSR, first in order to locate the population under study in a cyclical time frame and subsequently to be able to evaluate the long-term suitability and effects of proposed interventions.
Although it is a research difference rather than a production system difference, it is important to note that testing raises more serious problems in PSR than FSR. Some of these problems are due to factors such as mobility which have already been mentioned. Other relate to the virtual impossibility of simulating pastoral conditions on research stations.
Lastly, PSR must differ from FSR because it suffers from a dearth of reliable data on which to build. The bulk of research in agricultural production in Africa has been on crop production; even within farming systems little attention has been paid to the livestock sub-sector. Gilbert et al (1980) noted with regard to the initial stages of FSR, "The poorer the data and informational base is, the more research at this stage becomes an art rather than a science and the more on-farm studies are needed to describe and diagnose the areas' characteristics and constraints". Yet as Eicher and Baker (1982) noted, "In summary, research on the behaviour of livestock herders in Africa is about at the same point where research was on the economics of crop production some 20 years ago - many assertions and a sparse supply of facts".
The need for household level data in PSR
PSR is aimed at increasing the productivity of pastoralists and improving their welfare. As much resource management, inputs, and outputs occur on the household level, data must be collected at this level to understand these important production parameters. Additionally, in the final analysis interventions generated by PSR have to be acceptable to and adopted by individual pastoral households. A good understanding of the perceptions, goals, strategies, and decision-making of individual pastoral households is essential if researchers are to effectively identify constraints and design and test interventions which must be implemented at the household level. (This does not negate the need to pay attention to extra-household relationships as noted above).
Although PSR will inevitably differ from FSR, there are general methods used in FSR (and generally in survey research) which can be used in a PSR framework. It is not in the scope of this paper to review these methods in detail. A number of excellent texts are available for that purpose (Collinson, 1972; Byerlee et al, 1980; Kearle, 1976; Bernstein, 1979). Nevertheless, it would be useful to list basic techniques for gathering household level data and evaluate some of their advantages and disadvantages particularly with respect to criteria used in FSR. Within a pastoral equivalent of FSR, the first questions would be: what exactly do we need to know and what is the most cost-effective way of obtaining the necessary data? The answer to these questions will depend on the nature and phase of the research project itself, and will determine the specific data and methods required.
Types of data collection techniques which might be used in a PSR context can be usefully scaled along two dimensions: formality and frequency. They range from informal, open-ended interviews normally done by the researcher himself, to highly formalised pre-coded questionnaires normally administered by a trained enumerator. In terms of frequency, techniques may require either single visits, limited visits, or multiple visits which may be continuous throughout the study period, or may use seasonal sub-samples. Multiple visit surveys themselves may vary considerably in terms of the frequency with which the data are collected from each producer.
Informal surveys
Informal surveys play a major role in the early stages of FSR. They are conducted by the researchers themselves without the use of enumerators, questionnaires, or random samples, and hence are rapid and qualitative. In the early stages of research, informal surveys provide critical background information on the production system, and form the basis for later, formal questionnaires. Even with the wealth of data available on cropping systems, Byerlee et al (1980) noted: "It is our experience that researchers rarely spend sufficient time on this rewarding and essential task." In PSR, informal surveys will need to play an even more important role.
Although informal techniques are often used primarily in initial surveys, it is important to note that they can be and should be more frequent. As Byerlee et al (1980) noted "(this) type of informal researcher-farmer dialogue... should be a continuous process through all phases of the programme". Informal surveys can elicit types of information that are impossible to collect with formal questionnaires. These include information on producers' strategies, decision-making, and social aspects of the production process. Sensitive information is also more likely to be provided in informal contact with a researcher than with an enumerator. Generally, informal surveys give the researcher the opportunity to interact directly with the producer and explore issues as they arise, rather than in a predetermined way.
Informal surveys by themselves have several disadvantages. Firstly, as the producers are usually chosen opportunistically, the representativeness of the sample may be a problem, especially in an area where there is significant heterogeneity among producers. Secondly, as informal surveys are qualitative, their validity is more likely to be questionned by decision-makers and planners.
Case studies may be seen as a variant of the informal survey in which a carefully selected sample of households are chosen for intensive study, including frequent visits by the researchers themselves. Case studies are particularly valuable in providing data on the full range of production parameters for the sample households, including data on goals and decision making, which are impossible to glean from formal questionnaires. Through case studies, linkages in the system are more likely to be understood.
Formal surveys
Formal surveys rely upon the administration of precisely designed questionnaires by enumerators. Formal surveys have the advantage of significantly expanding the number of households which can be included in any study. They provide standardised and quantifiable data, although there are inherent limitations in the type of data that can be collected. While sampling errors are easily decreased through formal surveys, measurement or observation biases inevitably increase. Formal surveys also have the disadvantage of requiring significant inputs for training and supervising enumerators, for coding, data preparation and analysis.
The amount of measurement error in formal surveys depends upon many factors, including the quality of the enumerators and the questionnaires, the type of data to be collected, the frequency of the visits, and the cooperation of the respondents.
Enumerators and questionnaires
The use of enumerators necessarily introduces biases in the data collected. While it is impossible to eliminate the 'enumerator effect', steps should be taken to minimize it through the proper selection, training and supervision of enumerators as well as attention to details of questionnaire design.
In addition to their suitability to conduct interviews, the enumerators selected should, to the extent possible, have the same ethnic background and if possible be known to the target population. This facilitates their acceptance by the latter and enhances their ability to explain the questions in terms and concepts the respondents understand. The latter is very important for standardisation, and will be improved upon even more if the questionnaire format is translated into the local language.
Great attention should be paid to developing the questionnaire, to be sure it taps the precise information required, to phrase it so that it enhances the recall of the respondent, and to standardise it to lessen "enumerator effects". Evidence suggests that the more precise the questions, the more likely the respondent is to recall and relate the answer. On the whole this is more tedious for both the enumerator and respondent, but if fatigue can be avoided, it is far superior in the accuracy of the data produced.
The accuracy of enumerators is enhanced by training them so that their understanding of the terms used in the questionnaires and the procedures for administering them is standardised. In training enumerators, the ILCA Kenya team asked them as a group to translate the proposed questionnaires into the Maa language. Considerable debates followed which were instructive to both the enumerators and the researchers. Once the questionnaires were standardised and pretested, role playing was used for further training, with the researchers trying to approximate all conceivable problems. Lastly enumerators were observed while they administered the questionnaires to non-sample households. If at all possible, one should start training more than the required number of enumerators so that at the end of training one will have a better selection on the basis of their observed performance.
If a questionnaire is at all complicated, the enumerators will still be learning in the initial phases of administration. Close supervision will be required in checking the responses so that errors (either in format design or administration) can be corrected before they are made repeatedly. The first set of data collected by the enumerators should be analysed as quickly as possible to assess its quality.
One cannot over-emphasise the high degree of vigilance required to ensure good quality data in survey work. (For a detailed discussion of quality control and correction of enumerator bias using analysis of variance, see Zarkovich, 1966).
Data type
Not all events are equally well remembered. In terms of probability of-good recall, events can be scaled along two dimensions: frequency and regularity. Events which are regular are less likely to be remembered individually, but producer estimates of their occurrence are likely to be adequate. Irregular events which occur rarely (such as livestock transactions) are likely to be remembered individually, but frequently occurring irregular events will pose great difficulty in recall and will necessitate high frequencies of questioning. Labour inputs and expenditures fall in the latter category.
Although this section concentrates on recall, it must be noted that information on certain types of events (e.g. treatment of animals) might best be gathered by actual observation rather than questioning. However, for most data, observation is far more time consuming and consequently far more expensive than recall.
Frequency of visits and end effects
The frequency of visits needed depends to a large extent on the type of event to be recalled (or observed). The frequency will also depend upon the depth of data desired on the event in question. For example, if detailed questioning about decision-making is desired, greater frequency would be helpful.
All surveys based on recall suffer from "end effect", i.e. the difficulty of specifying the limits of the period for which information is requested and the tendency of respondents to include events from outside this reference period. In multiple-visit surveys, the frequency of the visits will normally define the reference period, which is normally the time lapsed since the last visit of the enumerator. End effect problems occur with the first interview as the end point of the period of reference is often unclear. It is strongly recommended that the data from the first interview be excluded from analysis as it is invariably over-reported. Single-visit surveys suffer particularly from end affect problems unless the period can be clearly specified. Collinson (1972) advocates limited visits for certain data collection and stresses the need to use time frames that fit a local natural or cultural cycle.
Single-visit formal surveys. Single-visit formal surveys form an integral part of FSR methods as currently advocated by CIMMYT (Byerlee et al, 1980). Their primary role is to quantify and verify the hunches which have emerged from the initial informal survey. Single-visit surveys are most appropriate for collecting information on variables (such as herd structure) for which a single objective measurement will suffice. They may also be used to collect general data on a relatively small number of variables particularly when the essential parameters of the production system are already well understood. For well remembered and registered events (e.g. livestock transactions) single-visit surveys may be able to elicit time-depth data on actual events. For non-registered events, single-visit surveys rely more on the producer's ability to estimate, based on general experience, rather than recall actual events; as such, they produce averages rather than indicate variability, and are inadequate when time series data are required.
Multiple-visit formal surveys. Multiple-visit formal surveys have several advantages over single-visit surveys. They provide time series data, are more likely to reflect actual events rather than estimations, benefit from learning on the part of both the enumerator and respondent. However, as the number of visits is increased, given the same amount of resources ,the number of households must be de creased. Again, careful attention must be paid to the nature of the data required in order to decide between 'measurement and sampling biases
In literate societies, it is possible for respondents to keep written records of the variables under study. This is almost impossible in pastoral systems research. However, Swift (1981) believes it may be possible to find, hire, and train local residents to collect time series data, with a minimum input from senior researchers.
When time series data are desired, decisions must be made as to whether the recording will be continuous or non-continuous. For example, rather than recording continuous expenditures, it might be possible with care to subsample within the important seasons.
Respondent cooperation and bias
Whatever the information collected, it is necessary to distinguish between the respondent's ability and his willingness to respond. The producer's ability to respond to the questionnaire is largely a function of survey design: whether the questions are well phrased and solicit information which it is possible for him to remember. Multi-visit surveys appear to have a learning component, so that some improvement in recall occurs, provided that it is not offset by respondent fatigue'.
A respondent's willingness to respond, however, is a function of his general level of cooperation as well as the sensitivity of the data solicited. FSR researchers frequently note value of early interventions in increasing the general level of cooperation. Nevertheless, there are still likely to be social/psychological factors which encourage inaccurate, and especially selective, reporting. These may be due to a desire for greater prestige, fear of shame, or a cultural taboo on certain topics. Case studies and participant observation are particularly useful to evaluate and adjust for such reporting biases.
Having briefly reviewed types of data collection methods for household level research, this session now turns to specific topics and methods. We will discuss, in turn, livestock transactions, household budget, and labour studies.
Bernstein, R. 1979. Design and management of agricultural research: A guide for agricultural researchers. Draft Paper, IRRI, Los Banos.
Byerlee, D. and Collison, M. 1980. Planning technologies appropriate to farmers - Concepts and procedures. CIMMYT.
Collinson, M. 1972. Farm management in peasant agriculture. Praeger, New York.
Eicher, C. and Baker, D.C. 1982. Research on agricultural development in sub-Saharan Africa: A critical survey. MSU Inter national Development Paper 1. Department of Agricultural Economics, Michigan State University, East Lansing. p. 168.
Gilbert, E.H., Norman, D.W. and Winch, F.E. 1980. Farming systems research: 4 critical appraisal. MSU Rural Development Paper 6. Michigan State University, East Lansing. p. 146.
Grandin, B.E. 1982. Time allocation and labour inputs on a Maasai group ranch: Preliminary findings from Olkarkar. Manuscript. ILCA, Nairobi.
Kearle, B. (ed) 1976. Field data collection in the social sciences. Agricultural Development Council.
Swift, J. 1981. Rapid appraisal and cost-effective participatory research in dry pastoral areas of West Africa. Agric. Admin. 8(6):48-5-492.
Swift, J. 1982. The start of the rains. Research Memo 2, Institute of Development Studies, University of Sussex, Brighton.
Zarkovich, S.S. 1966. Quality of statistical data. FAO, Rome.
Résumé
La recherche sur les systèmes pastoraux a pour objectif d'accroître la productivité des éleveurs et d'améliorer la qualité de leur vie. Etant donné que s'effectuent dans le ménage des échanges inter-scetoriels et des opérations de gestion, il conviendrait de recueillir des données à ce niveau pour comprendre ces importants paramètres de production. Les interventions engendrées par la recherche sur les systèmes pastoraux doivent être acceptables pour les éleveurs et adoptées par les ménages pastoraux. Pour que les chercheurs puissent identifier de manière effective les contraintes et mettent au point puis testent les interventions qui doivent être introduit-es au niveau du ménage, il faudrait qu'ils acquièrent une connaissance approfondie de l'opinion, des objectifs et de la stratégie des ménages pastoraux ainsi que du mécanisme de prise de décision.
Les types de techniques de collecte de données qui doivent être utilisées dans le cadre d'une recherche sur les systèmes pastoraux peuvent être déterminés selon deux critères: la forme et la fréquence. Ils varient des interviews informelles à bâtons rompus effectuées normalement par le chercheur lui-même, aux questionnaires précodés de type très classique, remplis normalement par un enquêteur qualifié. En ce qui concerne la fréquence, les techniques peuvent faire appel à des visites uniques, à des visites limitées ou multiples qui peuvent être continues pendant toute la période de l'étude ou à des sous-échantillons saisonniers.. Les enquêtes à visites multiples peuvent elles-mêmes varier en ce qui concerne la fréquence de la collecte des données auprès de chaque producteur.
Les enquêtes informelles jouent un rôle important dans la phase initiale de la recherche sur les systèmes d'exploitation agricole. Elles sont effectuées par les chercheurs eux-mêmes sans la participation des enquêteurs et sont par conséquent rapides et de qualité élevée. Lors de la phase initiale de la recherche, les enquêtes informelles fournissent des informations importantes sur le système de production et constituent la base des questionnaires de type classique qui seront élaborés après. Les enquêtes informelles comportent elles-mêmes de nombreux inconvénients. Etant donné que les producteurs sont généralement choisis de manière opportuniste, la représentativité de l'échantillon peut poser un problème, notamment dans un domaine où il y a beaucoup d'hétérogénéité entre les producteurs. Etant donné que les enquêtes informelles sont qualitatives, leur validité est beaucoup plus sujette à caution pour les décideurs et les planificateurs.
Les enquêtes de type classique se fondent sur la distribution de questionnaires conçus de manière précise par les enquêteurs. Les enquêtes de type classique ont l'avantage d'accroître de manière significative le nombre des ménages qui peuvent être inclus dans une étude. Elles fournissent des données harmonisées et quantifiables, quoiqu'il y ait des limites inhérentes aux types de données susceptibles d'être recueillies s'il est vrai que les erreurs d'échantillonnage diminuent facilement dans :les enquêtes de type classique, par contre, les distorsions en ce qui concerne les mesures ou les observations augmentent immanquablement. Les enquêtes de type classique ont également l'inconvénient d'exiger des ressources importantes pour la formation. et la supervision des enquêteurs en vue. du codage, de la préparation et de l'analyse des données. La quantité des erreurs de mesure dans les enquêtes de type classique est fonction de plusieurs facteurs, y compris la qualité des enquêteurs et des questionnaires, le type des données à rassembler, la fréquence des visites et la coopération des enquêtés.
L'utilisation des enquêteurs introduit des distorsions au niveau des données collectées. S'il est impossible d'éliminer l'effet de l'enquêteur, des mesures doivent être prises quand même en vue de les minimiser grâce à une sélection appropriée, à la formation et à la supervision de l'enquêteur de même qu'à la vigilance dans la conception des divers éléments du questionnaire.
La fréquence des visites nécessaires dépend du type de l'événement dont il faut se rappeler. La fréquence dépend également de la "profondeur" des données désirées sur les événements en question.
Le rôle primordial des enquêtes de type classique à visite unique consiste à quantifier et à confirmer les impressions qui émergent des enquêtes informelles initiales. Les enquêtes de type classique à visites multiples ont plusieurs avantages sur les enquêtes à visite unique. Elles fournissent des données sur les séries chronologiques et permettent en général de recueillir des informations sur des événements qui ont effectivement eu lieu plutôt que sur des estimations. Ces données permettent d'apprendre beaucoup de choses sur l'enquêteur aussi bien que sur l'enquêté
