FAO STATISTICAL DEVELOPMENT SERIES
PROGRAMME FOR THE WORLD CENSUS OF AGRICULTURE 2000
Food and Agriculture Organization of the United Nations
GUIDELINES FOR TAKING A CENSUS OF AGRICULTURE
This chapter presents basic principles for planning and carrying out an agricultural census. It provides a check list of major aspects rather than an exhaustive detailed description. Steps covered include the work plan from which budget and staffing estimates are generated.
In addition to budgeting and staffing the chapter includes information on tabulations, the questionnaire, cartographic preparation, and the importance of frames (list and area).
The chapter continues with information on pretesting, pilot censuses, data collection, instructions on training and monitoring the work of enumerators and their supervisors. A summary of data processing and dissemination issues is provided, concluding with notes on related research activities.
3.1 A detailed workplan should be one of the first census activities. The workplan defines the detailed activities in a logical sequence and includes staffing and resource estimates for each phase of the census process. Budget estimates, staffing requirements, equipment and supply needs will all be derived from the work plan and planning and monitoring will be based on it. The workplan should be built with a realistic time schedule for each activity in chronological sequence and show the relationship between the various phases of the operation.
Budget and expenditure control
3.2 An agricultural census budget should be prepared covering all census phases from the preparatory work to publication of results and installation of all data in a useroriented data system. The budget should show, for each fiscal year, permanent and temporary personnel required, salaries and wages, travel costs and expenditures for acquisition and operation of machinery, stationery and other supplies, and office space, communications, transport etc. with provision made for unforeseen expenses. The agency legally responsible for executing the census should be empowered to reallocate resources in case of unforeseen difficulties, especially during data collection and data processing.
3.3 The budget for data collection, processing and dissemination should show the volume of work to be performed, performance rates and measurable costs of all activities in the work programme. The budget should be reviewed periodically and work accomplishments compared to budget expenditures. Corrective action should be taken when necessary.
3.4 Census publicity is very important. Experience shows that inadequately informed and hence uncooperative citizens may jeopardize the entire census. The purpose of publicity is to bring the census to the attention of all agricultural holders, or at least to one family member thereof. The scope and coverage of the census should be clearly explained in the national and local press to familiarize people with the type of questions to be asked. The publicity programme should explain the uses to which the data will be put, particularly for development planning and formulation of agricultural policies; making the census meaningful to people, and highlighting, in particular, the possible benefits to farmers, thus establishing confidence between people and census authorities. The confidential treatment of data collected and the need for accurate replies should also be emphasized. Includedin the publicity is a special need to stress women's involvement both as holders and in holding operations and the importance of their support in participating in the census.
3.5 The best means of achieving effective publicity will vary from country to country. In many countries, press, cinema, radio, television and posters are used. In others, the publicity has been targeted at local religious and community leaders, business associations, labour groups, women and familial organizations and public service organizations. The cooperation of such organisations has proved to be the key to success. School publicity programmes may be efficient because schoolchildren are apt to pass information to and influence their families. The national or local census committee can also play an important role in the publicity campaign. In some countries, committees have been organized in villages to read and explain census publicity material to illiterate farmers. The use of audiovisual aids under such conditions may be helpful.
3.6 Senior administrative and professional staff need to be highly skilled and qualified, recruited from personnel familiar with agriculture, census methods and procedures, and government work. Supervisory personnel for enumerators can be recruited or borrowed from government agencies or local sources, such as statistical and agricultural extension services and educational organizations. Such personnel need to be intimately familiar with local conditions, customs, transportation systems, dialects and other relevant facts. Effective enumerators should be recruited independent of sex; but in some societies it may be necessary to use female enumerators to interview female holders.
3.7 Enumerators are best recruited from the localities in which they will work. They should undergo simple tests designed to measure their ability to read and apply instructions, understand maps, communicate easily with people, enter information on questionnaires accurately, and perform simple arithmetical operations.
3.8 Successful enumerators are tactful and resourceful in handling problems that arise when meeting and talking with holders and others; their actions and attitudes should gain the respect and confidence of those they encounter. They must be willing and able to work fulltime, without engaging in other activities, until the job has been completed. They must work carefully and diligently and always maintain required records.
3.9 A questionnaire is the medium for recording, in a standardized manner, the data obtained in censuses and surveys. Development of the census questionnaires is an important and exacting task in census preparation. Final data quality depends largely on the questionnaire design and the selection and training of enumerators.
3.10 Constraints during enumeration and the required format for tabulation of the data must be kept in mind when designing the questionnaire. It must use unambiguous concepts and definitions easily understood and clearly explainable by enumerators to the respondents. The questions must be simple and plainly phrased. The questionnaire aims to provide a standardized interpretation of the meaning of census items and data to be collected. Countries using computer assisted interviewing techniques, i.e. hand held computer devices for field enumeration, will find planning the questionnaire even more demanding.
3.11 The questionnaire must be prepared sufficiently in advance of commencement of the enumeration to permit adequate pretesting, and enumerator training; also the early finalization of the questionnaire is required in order to finalize the tabulation programme and start programming for data processing. A major objective of pretests is to ascertain deficiencies in the questionnaire and any problems holders have in responding to it so that revisions can be made prior to enumerator training.
3.12 The design of the questionnaire crucially affects the tabulation programme. Experience shows that often data are recorded on the questionnaire in such a way that they cannot be readily extracted for processing and tabulation, involving extra expense in time and resources. Countries are recommended, therefore, to prepare the tabulation programme concurrently with the final stages of questionnaire design.
3.13 It is necessary, during initial consideration of the tabulation programme, to decide upon the number of tabulations to be produced at various levels of aggregation according to administrative units and agro-ecological zones. Few countries can expect the tabulation programme to provide statistics for every village or commune. Further limitations on area and zone levels are imposed when the census is taken through sample enumeration.
3.14 Another important tabulation programme consideration is the choice of class boundaries and/or size criteria to be adopted for classification purposes. Explanations in Chapter 6 provide guidance on this subject. Countries are encouraged to adopt for each characteristic the classes set out in this Programme so as to produce internationally comparable results. Countries that expand the agricultural census scope beyond the items given in Chapter 5 may find it useful to adopt a size classification based on area irrigated, total value of products sold or consumed by the holding, or on value added, depending on relevant data collected in the census.
3.15 There is increased interest in tabulating certain characteristics of holdings by various types of farming. Countries where farm typology is established may find it useful to obtain the same set of cross tabulations for each type of farming in order to make comparative studies.
3.16 The overall tabulation programme must be assessed in terms of resources available to process the data to avoid undue delays. Some prioritization of tabulations may be necessary.
3.17 Enumerating all, or a large sample of, agricultural holdings in a country, without omission or duplication, in a short period of time, requires utmost attention to all the details. The exact delineation of each enumeration area is necessary and each enumerator should be provided with an enumeration area map showing the exact boundaries. In addition, supervisors and local offices should have copies of these maps for their respective enumeration areas.
3.18 Census authorities should investigate availability of cartographic resources at an early stage of census planning. Sources include maps, topographic charts, aerial photographs or satellite images.
3.19 Where maps, satellite imagery, or aerial photographs are not available and the agricultural census is undertaken on a sample enumeration basis, a complete list of villages or other identifiable geographic units should be prepared in advance. The list should include, as far as possible, complementary data on size of villages or units such as agricultural population or people engaged in agriculture, population of ethnic groups, total area and agricultural land area, main crops and agricultural practices, and facilities including water availability for irrigation and agricultural machinery. These data are useful for stratification purposes which will improve the efficiency of the sample design.
3.20 The listing of all holdings within each enumeration area is another important and difficult agricultural census operation. This refers to screening the entire population within the area with a short questionnaire requesting information on area cultivated, animals kept and responsible person(s) in order to arrive at the list of agricultural holdings to be enumerated. Lists of holdings or holders available at administrative offices are frequently incomplete and out of date and unsuitable for census enumeration. Listings from a population census taken shortly before the agricultural census are animportant source to provide a first draft listing that can be used in the screening to identify holders. Countries without maps or independent sources will be well advised to include a few screening questions on the population census questionnaire. Some countries lacking the above sources may have to prepare for each selected enumeration area a new listing of households and holders within households in order to identify the holdings.
Instructions and training for enumerators
3.21 Country experience indicates that it is essential to provide instruction manuals and extensive training for census enumerators to standardize procedures, secure a common understanding of tasks to be performed and provide a reference guide during enumeration. Instruction manuals and enumerator training programmes are essential because census data quality depends primarily upon the enumerators. There is a risk that all efforts and investments made for the census will fail if adequate training is not given. Instruction manuals should contain detailed explanations of procedures for conducting the enumeration, interview techniques, guidance on how to handle major and frequently encountered problems (such as uncooperative holders), and examples of properly completed questionnaires. Preparation of these instruction manuals is a high priority task and should be the responsibility of persons with a thorough knowledge of census design, holders' characteristics and interviewing techniques and with wide experience in preparing such manuals. Training should be carefully organized and conducted within an appropriate time schedule.
3.22 Enumerator training should cover:
Instructions and training for supervisors
3.23 The supervisors' work in overseeing enumerators' work and assisting them to solve problems encountered is essential to the census success. The supervisor's presence and inspection of enumerators' work helps prevent carelessness, and facilitates error detection and correction while enumeration is in progress. Supervisors should encourage enumerators to perform acceptable work, ensure they complete work assignments on time and help enumerators to promote holders' cooperation. Supervisors should follow and record enumeration progress, and take appropriate action when work is not performed in accordance with instructions or according to a prescribed time schedule. Good supervision is a proven and worthwhile investment; one supervisor for a reasonable number of enumerators is fully justified by improvement in work quality ensuring data accuracy and completion of work on schedule. The best supervision is achieved by constantly working in the field with enumerators. The supervisor should be present at several initial interviews, to detect deficiencies and take immediate remedial action. Subsequent regular visits should be organized to observe at least one interview and inspect a sample of completed questionnaires for completeness and internal consistency. When the enumerator has completed one work phase in a locality, the supervisor must review the enumerator's work, ensuring that all households have been accounted for, all holders interviewed and all questionnaires properly completed. Omissions must be detected and visited and unsatisfactory interviews may need to be repeated, if necessary with the supervisor's assistance.
3.24 In view of their important role, supervisors should be selected preferably among candidates having field experience in similar activities and undergo an intensive training programme combined with field work. This training programme should include the full training course provided to enumerators (see paragraph 3.22 above) and, in addition, supplementary training on the following subjects specific for the work of supervisors:
3.25 Omissions in the list of holdings require special supervisory attention. In many countries, significant listing errors result from difficulties in identifying households along the enumeration area borders. If the map or locality sketch map does not clearly distinguish boundaries by natural features, the supervisor should check carefully the accuracy of listings on the enumeration area perimeter.
Pretesting and pilot census
3.26 Pretesting involves enumerating a very limited number of holdings. The pretest collects evidence, through interviews and/or objective measurement techniques, of the adequacy of various census procedures. Pretesting alternative census methodologies, the questionnaire, and enumerators' instructions are vital and no census should be undertaken without it. Pretests must be performed exactly as prescribed for the main census enumeration. Some purposes of pretesting are to:
3.27 Quality control records of enumerators' activities during the pretest should be kept, preferably by supervisors or staff members without training responsibilities. Such records should include the number and type of errors made by each enumerator and the time required for each operation or part of the training programme. Summaries of time used and problems recorded during pretesting should also be prepared. These records and written suggestions from supervisors and staff members participating in the pretest provide the basis for revising the questionnaire or instructions to enumerators.
3.28 Pretesting may involve a pilot census which is a pretest of sufficient size and design to fully test procedures, forms, and systems to be used in the census. It is designed to provide information on all phases of a programme from the effects of pre-publicity to tabulation of data. A pilot census is a final test of the census programme, used to detect and correct any weakness in the programme before the actual census or sample enumeration is conducted. If considerable changes are made in the programme after a pretest, a second smaller test may be required.
3.29 In an agricultural census, data are collected through interview and/or by mail. In the interview technique, the enumerator visits the holding, interviews the respondent and records the responses on the questionnaire. Interviews may be supplemented by observations or measurements carried out by the enumerator. Interviewing is the method adopted in most developing countries where postal services are not fully developed and the literacy rate is low.
3.30 Using the cheaper mail approach, the questionnaire with explanatory notes is mailed to the holder, usually with a stamped preaddressed return envelope. Reminders may be necessary to increase response rate.
3.31 Interview and mail techniques are sometimes used to complement each other. Interviewers are sent to a sample of holders who have not responded even after receiving reminder letters. Sometimes data are collected from registered holders and institutional holdings by mail and enumerators are sent to other holders. Each country must decide its own technique, based on local conditions and available resources.
3.32 Early preparation Data processing is an important phase of the agricultural census requiring adequate planning and preparation. Such planning and preparation should include hardware acquisition, personnel training, participation of data processing experts in questionnaire design, and writing computer programmes prior to enumeration. Specifically, a detailed tabulation plan and instructions for manual and computer data editing should be finalized early enough to make possible an efficient organization of data processing. Since the quantity of data to be processed is large, insufficient preparation causes long delays in obtaining census results.
3.33 Modern basic guidelines are oriented to electronic data processing. Countries without computer facilities may need to further restrict the census scope and/or limit the number of units covered. A combination of manual and computer processing may be appropriate for a few countries. For example, a number of preliminary data processing operations, including preliminary editing of questionnaires and preparation of district or provincial totals for some important census items, may be undertaken manually by qualified field staff in each administrative district or province. However, with the rapid advances in lowcost powerful small computers, each country undertaking an agricultural census should obtain some type of appropriate computer capability. It is one of the most effective resource expenditures a statistical organization can make, beyond basic organization, staffing and training. Computer processing equipment may also be available in provincial statistical offices, which allows for data processing to be decentralized.
3.34 Whereas data processing details will depend upon the equipment and software packages available for the purpose, a number of operations are common. The most important of these are briefly described below.
3.35 Maintaining control of questionnaires - Controls should be established to ensure that questionnaires are received from every enumeration area and every enumerator. Questionnaires should be grouped so that such controls are simple and not too timeconsuming. Questionnaires for large or special holdings within a province or state might be batched together. All questionnaires for one enumeration area covered by a single enumerator should also be batched together. Records need to be kept of the flow of batches of questionnaires through the various processing steps and should be checked periodically to detect delays, misplacement of questionnaires, etc.
3.36 Checking for enumeration completeness - Questionnaires for large holdings must be checked against the complete list of such holdings and action should be taken to obtain missing questionnaires. Questionnaires for each enumerator must be checked against the holding list for the enumerator's area, and a satisfactory explanation sought for missing questionnaires. Adequate field organization, including supervision of enumerators' work and questionnaires received in each district and province, will appreciably reduce work in the central office.
3.37 Checking for questionnaire completeness - A visual check should be made to ensure that each questionnaire has entries in essential sections (land use, livestock, etc.). Questionnaires lacking essential entries should be referred back to the originating office for action. Questionnaires for large or special holdings should be reviewed for completeness by professional staff. Again, efficient control of field work before completed questionnaires are sent to the central office is essential. The speed and accuracy of data processing will be greatly influenced not only by the accuracy but also the legibility of the questionnaires sent in. Enumerators must be trained to write clearly.
3.38 Verifying office processing - Some verification of data processing work should be performed. Complete or sample verification of data entry and other routine operations is important. Training of processing staff must take into account the time needed by data entry operators or clerical personnel to acquire the skill and practice required in order to perform the work at relatively stable quality and acceptable error levels. Complete verification is costly and does not detect all errors, although computer editing has advanced the possibilities of such complete checks. Once a data processor produces work of satisfactory quality, verification of a sample of the work is sufficient to ensure quality standards are maintained. Similarly, controlling errors at the work unit level can be achieved by verifying small samples, accumulating verification results and comparing cumulative results with acceptable standards. If a work unit does not meet the required standards complete verification of the output is necessary with correction of errors and retraining of staff. Normally, when work units meet quality standards, errors detected in sample verification processes are not corrected.
3.39 When the census is based on sample enumeration, verification of the data at each process is even more important than for complete enumeration. Complete verification of data is preferred.
3.40 Countries where provincial offices are involved in the processing will have some of the functions listed above. The necessary controls must be rigidly applied in provincial offices as in the central office.
3.41 Computerized data processing requires various routines to have been decided in advance: precoding of questionnaire items; types of corrections to be made during questionnaire verification; coding of the data; correcting errors detected during data entry operations; and the tabulation plan. This advance work requires experienced agricultural statisticians to work with computer system analysts and programmers during the census planning operations.
3.42 The success of data processing depends on:
3.43 Preparing computer programmes Considerable time is required to write computer programmes for tabulation, error identification and error correction. Available software packages should be carefully studied for their possible installation and use to perform these activities. After the selected software packages are installed or the computer programmes prepared, they should be tested with data from pretest surveys and pilot census. Computer printouts should be run to identify errors and corrections. If errors remain uncorrected additional specifications are required.
3.44 Once census questionnaires are received a further test of the computer programmes should be carried out using a sample of, say, 100-500 questionnaires. The resulting tabulations should be compared with a manual summary and against each other to ensure consistency across sets of tables.
3.45 Error detection and correction by computer include measures to deal with missing and impossible entries which are beyond normal ranges in the values entered and data inconsistencies. Generalized edit and imputation algorithms are used to impute values to replace erroneous entries. The design of data entry, editing and correction programmes should be based on a set of tests and procedures defined jointly by statisticians and data processing specialists. Caution must be exercised to ensure that these procedures do not falsify data. Parameters used for editing such as minimum and maximum acceptable values should be carefully determined, based on pilot census and/or other independent data sources in order to avoid eliminating valid entries. If missing data is insignificant, it may be preferable to tabulate them under holdings not reporting rather than imputing values. A record should be retained of the number of cases imputed or changed.
3.46 Some common types of automatic error detection include:
3.47 Census data processing requires time, often more than a year, to complete. Steps should be taken to obtain priority data in advance of other results by planning the tabulation programme in two or more phases. Even earlier preliminary estimates of high priority data may be obtained by tabulating a sample of the census data as a first stage. Advance tabulations can be supplemented by manually compiled aggregates of main census items. Enumerators may prepare enumerated holding aggregates on summary sheets; these can be further aggregated by administrative areas in districts and provinces, while corresponding national and other totals may be prepared and issuedby the central office. It is important, however, that procedures should be developed for processing all the data. If much of the data remain unprocessed after some years the effort that went into the data collection will be deemed to have been wasted and the policy analysts or planners needing such data will be less likely to support future census efforts.
3.48 Census processing no longer requires main-frame computers. Micro-computers are now very effective for data processing of large data files and have a number of advantages over large mainframe computers. First, they are less costly and are physically easy to transport and install. Second, they are user friendly and a wide range of software applications are tailor-made for use on them. Third, it is possible to dedicate microcomputers totally to census work, whereas this is generally impossible with a mainframe computer serving many users for different purposes with priorities often given to non-census applications.
3.49 The number of micro-computers or work stations required and compatibility with other available equipment need to be carefully planned. A microcomputer system compatible with larger centralized computers can be used to decentralize and speed-up data entry and checking for later processing on a central computer. Prospects of using the micro-computer system for projects following the census will add value to the purchase of a system. Other important factors to consider before acquiring a system include ensuring adequate software is available, ensuring service and support for the system is available, and that the power supply is sufficiently reliable to avoid the risk of damage to the computerized data files.
3.50 Despite precautionary measures taken in the course of data processing, some errors will remain and be incorporated into census tables. The effect of these errors may be considerable and all tables should be systematically reviewed before publication, to eliminate or at least minimize the effects of errors. Various methods can be used to carry out this review. An essential procedure is a consistency study of census data compared with statistics on the same subject available from previous censuses and surveys. Agreement between old and new data obviously does not establish the accuracy of either. The same applies for internal consistency checks. However, if there are no major discrepancies, the data can be released. Should large discrepancies be discovered, further investigation is necessary and verification of original data may be required to ascertain the error source. Another useful step involves breaking down basic aggregates by various administrative units and comparing the subaggregates with analogous data and statistics from other sources, including professional judgement.
3.51 Table review should not delay publication: knowing the tabulation programme, existing statistics can be studied beforehand. Thus, a quick review of the tables is possible as they are produced. Results of quality checks made during and after field enumeration can be fully utilized in the final evaluation of census tables. Quality checks may reveal major gaps in the enumeration, certain categories of holdings or particular areas of the country. In such an eventuality tabulations to be released may need to exclude such categories or qualify their entries with cautionary footnotes pending further examination leading to correction in the final census report.
Quality checks and post enumeration surveys
3.52 Nonsampling errors may arise from numerous sources. The census frame or list of holdings may be incomplete or inaccurate; the wording of questions ambiguous or misleading; enumerators may introduce their own biases; respondents may not really know the true answer, or cannot recall the data requested, and others may consciously answer incorrectly; field work may be inadequately organized or supervised; enumerators may lack specific training or unsatisfactory standards may have been used for their selection; completed questionnaires may be lost. Specifically the following error types commonly occur during field work and need to be kept in mind during field and office checks:
3.53 Efforts to reduce errors arising from all field work sources and data processing stages have been emphasized in the earlier sections of this chapter. One further stage may be required, namely the carrying out of sample enumerations as quality checks during or just after the main census enumeration. Statisticians have an obligation to their profession and to data users to undertake these checks. Such post-enumeration checks may represent the only serious attempt to obtain evidence of census methodology deficiencies, types of errors occurring, and magnitude of such errors. Such evidence provides a concrete basis for overall improvement in survey methods and the elimination or reduction of errors and biases. The publication Quality of Statistical Data (FAO 1966), contains detailed information on the control of non-sampling errors.
3.54 Census taking uses public resources and the published results represent the public return of a major product from this expenditure. Primary tables easily summarized should be published as early as possible. Immediately following this, a schedule of publication results should be organized with wide dissemination of these results.
3.55 A general census report, prepared by professional staff, may be issued in several volumes. The report should include, in addition to statistical tables, all information that might be useful to better understand and evaluate the data. Details of organizational and administrative aspects of the census should also be included as they may be useful in preparing and implementing future censuses. The report may also include material on objectives, legal authority and administration, scope and coverage, essential definitions, concepts and classification, assessment of reliability of results, copies of questionnaires, summary of main instructions for enumerators and supervisors, data collection methods, data processing and tabulation methods, description of administrative and agro-ecological zones used, and comparisons with statistics from prior censuses or other sources. The report should also provide all relevant sample design details where samples were used, particularly those discussed in the United Nation's publication on preparation of sample survey reports (UN 1964).
3.56 The dissemination programme including the publication list is as important as other components of census operations. Availability of computers and feasibility of storing primary data permit utilization of results in a variety of ways in addition to those included in main census publications. The micro data within a census file contains a wealth of information awaiting further use. Short reports with graphics of data reflecting changes in structure or new trends can be very important in exposing the value of information in the data base. The data dissemination plan should consider the need to get basic data in the hands of data users as early as possible.
3.57 When preparing the reporting and data release schedules the following should be kept in mind.
3.58 Well conducted research studies furnish, for each alternative data collection technique, an assessment of corresponding cost and resource requirements, needed qualifications of enumerators, relevant difficulty of different procedures, etc. The quality of census results depends heavily upon techniques used in data collection. The design of a sample enumeration requires basic information about the population being sampled. The appropriate number of units to be included in a sample enumeration cannot be determined without first estimating variances in important population characteristics. Pilot research studies are necessary to obtain this information when it is not available from other sources.
3.59 Another important reason for establishing a research programme is the need to monitor development of the national food and agricultural statistics programme of which the census is a component. On the one hand, the contribution of the census to this programme depends on relative development of other data sources, such as various agricultural surveys and administrative records. On the other hand, the census should provide a frame for specialized surveys on agricultural holdings essential for planning these surveys.
3.60 Pilot research studies must be conducted under field operating conditions, otherwise the research results have inadequate bearing on the real situation. Research results must be adequately and properly analyzed and, finally, the best course of action must be implemented to ensure a successful agricultural census. Qualified statisticians with research training and experience are a prerequisite for guaranteeing that these criteria are satisfied.