Brazil Agricultural Census 2006: Innovations and Impacts Octávio Oliveira Instituto Brasileiro de Geografia e Estatística Flávio Bolliger Av. Republica do Chile, 500 – 7º andar Antônio Florido Rio de Janeiro, 20031-170, Brazil 1. Abstract

In its first agricultural census of the new millennium, Brazil’s Institute of Geography and Statistics (IBGE) is innovating data collection in 2007 by introducing electronic questionnaires into Personal Digital Assistants (PDAs) integrated with Global Position System (GPS) to achieve geo-referenced information on all of the approximately five million farms in the country, covering 8,514,877 km² in 5,564 municipalities.

Real-time edits and automatic jumps performed during enumeration will ensure a faster and more reliable interview. The PDA’s Bluetooth® technology allows direct data transmission to IBGE´s central mainframe by each of the seventy thousand enumerators on a weekly basis. This eliminates the cost and time of transporting, scanning or keying data from paper questionnaires and allows supervisors to immediately address problems while data are being collected.

The final results of the 10th Brazilian Agricultural Census are expected to be released at the end of 2008. The database of geo-referenced farm addresses will provide a comprehensive list for future surveys. In addition, an area frame, based on IBGE´s enumeration sectors, will be constructed. Presently, IBGE does not conduct sample surveys in agriculture, so these programs are expected to result in a great improvement in Brazilian agricultural statistics.

This paper covers the main innovation in Brazil’s agricultural census, their motivations and consequences. Data quality improvements and new horizons offered by geo-referenced agricultural data at the farm level, including sample surveys, are also discussed.

2. Introduction

Agricultural censuses have been conducted in Brazil since 1920, on a decennial basis until 1970 and quinquennially thereafter. Since its establishment in 1936, the Brazilian Institute of Geography and Statistics has had the responsibility for carrying out the agriculture census. From 1970 through 1985, the reference dates, concepts and structure of the agriculture censuses were similar, ensuring international comparability for 1970, 1975, 1980 and 1985 census data. An economic crisis in Brazil led to the postponement of the agriculture census until 1996, when the agricultural census was performed simultaneously with a basic enumeration of the population. However, in order to reduce costs, the definition of an agricultural holding was changed so that only commercial holdings were included in the agriculture census. To allow adequate integration of the data collected for agriculture with the population data collected at the same time, reference dates for the agriculture data also had to be modified, representing a break in the historical series and introducing difficulties in data comparability with past agriculture censuses.

The mid 90’s brought a more open economy and economic stability to Brazil, which led to significant changes in the agricultural structure of the country. The significant increase in the use of technically advanced farm management practices and increased investments in modern equipment and machinery by the country’s producers since 1996 has not, until now, been measured by an agricultural census. Grain yields increased from 69 million tons in 1996 to 118 million tons in 2005 (+71%), while harvested area increased only 38% (from 33 to 46 million ha). This higher level of productivity has allowed Brazil to assume a position as one of the world leaders in the production of soybeans.

Since the 1996 Census of Agriculture was conducted, various government programs have been implemented to provide credit to agricultural producers and to bring about agrarian reform. In addition, there has been an increase in land used for the cultivation of crops as producers move into frontier areas that were previously forested or used for grazing of livestock. The importance of knowing the results of such programs and actions was an important argument for the necessity of the 2006 Census of Agriculture.

2

This paper covers Brazil’s experience in carrying out its 2006 Agricultural Census, and was written during the data collection phase; thus some information is preliminary and the discussion presented here reflects initial impressions of the overall process. We look at the main innovations, both operational and contextual, introduced in this census. We will discuss the motivations and consequences of our decisions, especially those associated with the use of PDAs and GPS coordinates. Within the limitations of this article, we examine the difficulties found in field operations and the solutions used to overcome them. The paper also looks at the impacts the census results will have on IBGE´s ability to construct list and area sampling frames for follow-on surveys.

3. Brazilian Agricultural Census Characteristics

The 2006 Agricultural Census in Brazil is a complete enumeration of all agricultural holdings in the country. It accounts for all 27 federation units, divided into 5,564 municipalities and covering 8,514,877 km², in both rural and urban areas. Enumeration is done at the level of the agricultural holding, as defined by FAO, i.e., the economic unit dedicated to agricultural production under single management, regardless area size, title, legal form, production purposes or value. The reference period for previous censuses has been the prior calendar year, except in 1996 when 1995-1996 agricultural year was used. There are no list frames available for use in data collection, thus census coverage depends on the complete canvassing of the enumeration area by each enumerator. The time allotted for data collection was 90 days, during which time face-to-face interviews with all agricultural producers within the Census Enumeration Area were to be completed. Enumerators were hired on a temporary basis to collect data in their own communities, and they were paid according to the number of interviews conducted or units registered. The minimum educational requirement for enumerators was completion of elementary school.

4. CENSOS 2007: An integrated operation

The 2006 Agricultural Census in Brazil was conducted as an integral part of a larger undertaking, known as CENSOS 2007. In addition to the agriculture census, CENSOS 2007 included two other major undertakings. A population count (with very basic demographic information collected) accounted for the people and households in rural and urban areas. Update of the national addresses list, by including rural addresses and updating urban ones is the final endeavor included in CENSOS 2007.

The population count covered only municipalities with populations of less than 170,000, with the exception of 21 others with higher populations. These municipalities are in states which have only one or two municipalities with more than 170,000 residents, and to exclude them from the count would severely limit the value of that state’s data. The questionnaire used for the population count asked for only five items for each household member: Name, Relationship to Head of Household, Age, Gender and Migration.

The questionnaire for the agriculture census, on the other hand, has more than 360 possible entries. Determination of the core questions to be included in the report form was based on recommendations by FAO (FAO, 2006). In addition to core items, the 2006 Brazilian Agriculture Census included other variables requested for special study or for the development of follow-on surveys. In the end, the questionnaire asks producers for information on irrigation, aquaculture, livestock, farm labor, crops, agricultural practices, price and values of agricultural products and land, educational level of farmer, electrical energy use, organic production, investments, income and expenses, debts, and others. Because of the detail of many of the questions, they were asked only of respondents who reported a certain minimum amount of an item. For example, the distribution of cattle by age would only be asked of agricultural holdings having at least 50 animals; production practices in permanent crops were inquired only for those holdings having at least 20 trees per species. The electronic questionnaire was designed so that the respondent would only be asked detailed questions for an item if he met the threshold for that item, otherwise the instrument would skip to the next appropriate question. Thus, while all agricultural holdings, livestock, production and trees were registered, most detailed information was asked only of holdings that either specialized in an item or were significant producers of that item. This manner of collecting detailed

3

data from these large, or specialized, farms is similar to the use of census supplementary modules recommended by FAO (FAO, 2006).

The third part of the CENSOS 2007 project is the construction of a National Address List for Statistical Purposes (CNEFE, in Portuguese). Along with the agriculture census, it will provide the names and geo-referenced coordinates for all agricultural holdings in the country. In addition, the population count will enumerate all households in the rural and selected urban areas, providing addresses and geo-referenced coordinates. Finally, non-residential structures (churches, hospitals, government buildings, etc.), will be identified and listed with their corresponding addresses/geo-reference coordinates. It is important to note that all geo-reference coordinates were obtained exclusively in rural enumeration sectors, and the system automatically inhibited this function for urban sectors.

Data collection for CENSOS 2007 was scheduled from mid-April through August, 2007. More than 5 million agricultural holdings were enumerated. More than 34 million urban households were visited and 105 million people (about 60% of whole estimated population) were registered. Staffing for the field operations consisted of some 5,500 local agents, 11,000 supervisors, and 68,000 enumerators. The operation comprised 162,770 enumeration sectors from the existing 249,068 in territorial database. The total budget was R$ 538,000,000 - approximately US$ 269,000,0001.

Why IBGE decided to use Personal Digital Assistants (PDA) in CENSOS 2007 The decision to conduct the address list, agricultural census and a population count in a single, integrated data collection effort was, in fact, the only way to accomplish our goals with the resources available to IBGE. To carry out separate operations would have increased the costs by up to 40%.

In previous Agricultural Censuses, data collection was performed using paper questionnaires, and data were later keyed2. The use of PDAs offers some real savings by eliminating the costs of printing and handling of questionnaires, and the keying of data from paper questionnaires to a computer system. It also enables a single data collection instrument to be used by the same enumerator on all three operations, and provides interactive edits, ensuring that the data is clean. In addition, the integrated Global Positioning System (GPS) built into the PDAs provides coordinates for the rural addresses. This allows us to build a national address list that can be used for statistical surveys and to examine the possibility of moving away from a traditional demographic census to a series of continuous measurement surveys. Obviously, the decision to use PDAs also was a strategic one to move the institution forward in the area of modern procedures and use of new technologies, and to prepare IBGE to use them in future survey work.

Choosing the device Selection of the device was made after a process of public bidding, with the winner based on meeting required specifications at the lowest cost. The device selected was the MIO P550 B (Figure 1) with the following characteristics:

• Samsung processor 400 MHz

• Operating System - Windows Mobile® 5.0 in Portuguese

• Display - 3.5″ TFT with touchscreen/65k colors/QVGA-240×320 pixels

• Memory - On-board 2GB flash memory with 64MB SDRAM, SD/MMC/SDIO slot, and a SD card 256 Mb

• GPS - 20 channel SiRFstarIII

• Connectivity - USB v1.1, Bluetooth V2.0, built-in WLAN

1 Using an exchange rate of R$ 2.00/US$. 2 At the end of the 1996 Agricultural Census, in 1998, a small group started studies to adopt PDA use in data collection in the next census, scheduled to be in 2001. A prototype of the electronic questionnaire was developed, but the project was interrupted after the programmer death. The proposed alternative was to collect data in paper questionnaires and digitize data through high-speed scanners and Optical Characters Recognition – OCR (IBGE, 2003).

Figure 1. PDA used in

CENSOS 2007 data collect.

4

802.11b+g module

• Microphone - Built-in microphone

• Speaker/Headphone - Built-in speaker / 2.5mm earphone jack

• Battery - 1900mAh rechargeable Li-ion battery

• Dimensions - 4.15″x2.4″x0.6″(126×73x17.6mm)

• Weight - 6 oz (170g)

• Plastic protector case, water and shock resistant, with elastic handle to prevent drops.

82,000 devices were acquired in 2006 and 2007, including 2% for substitution stocking, in two lots. The average price was R$ 1.190,65 (US$ 595).

Software All data collection and supervision software was developed by IBGE in-house, using the Visual Studio platform in Microsoft Operations Manager 2005 environment and Microsoft SQL Server 2000, with assistance of Microsoft Brasil consulting. In addition, GEOPAD application was installed to view, navigate and consult maps and GPS orientation.

Data collection software automatically saved questionnaires onto an SD memory card, avoiding data loss in case of device crash or freeze-up. Updated versions of the software were installed automatically when an enumerator connected his PDA to the central server to transmit data.

5. Operational innovations

Infrastructure With the adoption of PDAs, the collection, editing and transmission of data, along with field supervision were totally revolutionized. The initial plan for transmitting data back to headquarters was to use any available phone line, including cell phones. However, due to the high cost involved in securing lines that would guarantee the integrity and confidentiality of data transmissions, this alternative had to be abandoned. Instead, secure communication lines were made available in CENSOS 2007 data collection stations. 5,597 data collection stations were established across the country, with 1,106 of them computer-equipped (with microcomputer, broadband access and modems) and 4,491 which were not computer-equipped (only modem and phone line available). IBGE created partnership agreements with public entities such as townships, the army and the postal service in order to access their phone lines, computers and Intranet connections to complement IBGE’s own field network. Broadband access (considered as providing higher speed than obtained via modem, 56 kbs), which had been limited to 380 local offices until 2007, was extended to all 534 IBGE offices after installing radio antennas for internet connection. 700 antennas were installed provide broadband service to the IBGE offices and also to establish computer-equipped data collection stations in remote areas where the existing telephone system was inadequate for data transmission.

In the computer-equipped data collection stations, census managers had access to a system with personal information on all enumerators and supervisors (including the serial number of the PDA assigned to that person), the census enumeration areas assigned to each enumerator, and the enumerators assigned to each supervisor.

Collecting and editing the data Data collection was performed entirely by use of handheld personal data assistants. It was expected that using the PDAs would provide the following benefits:

• Real-time edits and correction of respondent or interviewer error at time of interview;

• Ensure that all mandatory fields are completed at time of interview;

• Smoother and faster interviews, due to programmed jumps, eliminating the pauses and hesitations as enumerators turn pages and look for the next appropriate question on a paper questionnaire;

• More consistent interviews - because of the programmed skip pattern, the relevant questions asked of each respondent are based on answers provided, and regardless of the enumerator, the sequence of questions will be the same, given the same set of responses;

5

Figure 2. Example of a register

screen: permanent crop

production.

• The ability to monitor data collection in quasi real-time in all municipalities, providing better job management, especially in cases where corrective actions during data collection were needed;

• Eliminating the handling and transportation of high volumes of paper questionnaires;

• Elimination of the time required for data keying from paper forms and the errors that result from that operation,

• Provide GPS coordinates for agricultural holdings, households, and other buildings.

The 24 pages and 1470 fields in the original paper questionnaire for the 2006 Agricultural Census were replaced with 206 screens and 928 fields for the electronic questionnaire (Figure 2). It was discovered during testing that the complete set of edits could not be loaded into PDA, as the system became too slow and unstable, due to the number of variables involved. Only basic edits on important values, keying errors in mandatory fields and the questionnaire skip pattern were programmed into the PDAs. The complete set of edits was run on the data after it was transmitted to IBGE’s central server.

The GPS feature was used not only in registering addresses, but also for enumerator orientation in the field. Using GEOPAD application and digital maps, it was possible for the enumerator to verify if a location was within the census enumeration area. Municipality limits and census enumeration areas were digitized in vector data and, with captured GPS coordinates, structured in Geographic Information System (GIS) environment (Fortes et al., 2007).

An Intranet forum was created to provide support to the field supervisors. By logging on in one of the computer-equipped data stations, supervisors could raise questions or concerns about field activities, and have those questions or concerns addressed by headquarters. These questions and answers were available for all field staff to see.

As for the data collection itself, the devices performed well, with very few cases of malfunctions, losses, and theft. Losses of equipment due to these problems were estimated to be less than 1%. The plastic case and strap were essential in providing protection against impact from falls and bumps, resistance to moisture, and inhibiting robberies.

Despite increasing the scope of the agriculture census from 1996, interview time was less than previous censuses. Median interview time ranged from 40.8 to 67.3 minutes, according to Federation Unity, depending on the complexity of agricultural activity.

Data transmission The transmission of data to the central server in headquarters was performed directly from the same handheld devices used for data collection. Enumerators at data collection stations that were not computer-equipped used the built-in Bluetooth® feature to access the modems and send the data over phone lines. Enumerators in computer-equipped data collection stations used USB cables to connect and download data directly via the intranet. Data were periodically transmitted without the need to wait until an enumeration area was completed – enabling centralized monitoring of the data collection, and allowing review and analysis to begin sooner.

Supervision and remote electronic communication All supervision tools, which in the past had been maintained in paper, were replaced by automated systems. The practice of supervisors visually checking paper questionnaires was replaced by reviewing data in the enumerator’s PDA directly, and from reports and alerts generated by the system (Figure 3).

6

Figure 3. Summary of data collection for an enumeration area using SIGC. Supervisor alerts for the agricultural census,

population count, and address list are shown in upper part. Primary data for the enumeration area are listed, along with

information on the enumerator(s) and the beginning and ending dates of data collection.

The flow of administrative, technical and operational information for the census was controlled by a centralized system called, “Data Collection Management Indicators System” (SIGC, in Portuguese). Data received from the field were aggregated, providing information at enumeration area level, which fed SIGC.

This system had a set of indicators, calculated at the census enumeration area level, that would highlight discrepancies and suspect data, which could be reevaluated in the field, as needed. The system was designed to make performance and quality indicators available via the Web interface, and depending on the circumstances, automatic messages were generated and transmitted to supervisors’ and enumerator’s PDAs when they connected to the system. Supervisors’ responses to these messages were maintained in the system.

One of the biggest advantages of continuous transmission of data during data collection was that problems could be identified immediately, and the enumerator could return and verify or correct the information that caused concern. Inconsistencies in the data were identified by the edit program, and a SAS® routine generated an electronic version of the questionnaire that had the inconsistent data, which could be reviewed by analysts in headquarters. The headquarters analysts could then correct the problem there, or if necessary, redirect the case to the local supervisor.

7

Another tool that was used during the data collection period integrated the GPS information of the census enumeration areas and the agricultural holdings contained in the census database with Google Earth®, enabling visualization of both the enumeration areas and the location of the holdings within the enumeration areas (Figure 4). This provided a means to ensure that the holdings were in the correct enumeration area.

Call Center Another important planned innovation is the use of data verification through phone contact with farm operators (Computer Assisted Telephone Interview – CATI) using a call center. This system will be implemented to correct inconsistent data that remain after data collection is completed, and its use will be limited to large agricultural holdings. Data problems with small agricultural holdings will be solved by imputation methods under evaluation in IBGE.

6. Innovations in data tabulation and dissemination

In addition to the traditional census tabulations by territorial levels, 2006 Brazilian Agricultural Census will tabulate data by special areas as follows: watersheds, biomes (groups of similar ecosystems), rural settlements, indigenous reservations, environmentally protected areas and forest reserves.

Geo-referenced maps of watersheds and biomes were integrated with census enumeration area maps, making it possible to publish, for example, the amount of soybean production in the Amazon Rainforest, the use of pesticides in a specific watershed, etc. Information from the administrative registers of indigenous villages and rural settlements was entered into the appropriate census enumeration area registers. The registers of rural and indigenous settlements include information from a question in the census questionnaire specific to agricultural holdings in these communities.

Census data will be released in several different volumes, the first providing general information. This will be followed by volumes providing data based on particular themes: aspects of family farms; agriculture and the environment; labor and diversity in rural areas; and agricultural technology and production. Geographic details will only be published in electronic media.

7. Difficulties encountered in implementing innovations One of the major concerns during census planning was whether IBGE’s staff and enumerators would accept, and adapt to, the new technology. Indeed, a considerable number of experienced workers, accustomed to collecting data with paper questionnaires and the associated methods of supervision, had a difficult time adapting to the implemented innovations. Other factors which had to be overcome from the standpoint of acceptance and use of the PDAs by field staff included:

Figure 4. Data collection view in Google Earth®. Yellow line represents census enumeration area limits. Dots are

enumerated holdings with corresponding GPS coordinates. In this view all collected points were obtained

correctly inside enumeration area. Red arrow indicates a unit selected for detailed view, shown in the figure on

right. Clicking the dot enables the user to view some associated data, as well.

8

bearing responsibility for an expensive piece of equipment, inexperience with automated devices, and generally low education levels in some regions.

While it was fairly simple to develop training and data collection procedures using the new technology, the exact methods that the supervisors would use to control the field work proved difficult to define and test, especially in field conditions. The procedures were completed just before the actual training began, and so, without the benefit of testing them out in a pilot census, there was no way to know how well they would work.

Major difficulties experienced in data collection:

• Initial data loading into PDA – Data from each census enumeration area and respective enumerator information, along with software, were supposed to be transferred into each of the 82,000 PDAs upon its first connection with the central server. This transfer was not made in the time originally allotted, as the demands on the central server overloaded the system, preventing the data from being loaded into the PDAs. Without the PDAs, data collection could not start. Valuable time was lost while the problems were solved, which included changing some of the server hardware. After a few weeks, data collection finally started. However, this lost time, and the anxiety it created among the field staff, resulted in some enumerators quitting, and necessitated training new enumerators.

• Transmission of collected data – Soon after data collection began, field personnel started to transmit the data to IBGE’s central server in Rio de Janeiro. During this operation, it was discovered that the poor quality of some telephone lines made it impossible to access our network. Moreover, a number of antennas installed for Internet broadband access became misaligned due to strong wind or did not function well under rainy conditions, and there was not enough technical support to ensure all antennas remained functional. These difficulties required the adoption of alternative processes to transmit information, through CDs, pen-drives and FTPs. Of the total data transmissions, 60.5% were via desktop (VPN and IBGE net), 38.7% via modem/phone, and only 0.8% through alternative processes3.

• Completion of questionnaires – The agriculture census questionnaire took quite a while to complete, due its size and complexity and the characteristics of the PDAs. It was a dynamic process, in that the next screen to appear on the screen was based on the answers to previous questions. At times, this process was slower than hoped for. To make matters worse, enumerators would sometimes get impatient and navigate to the next screen as the data entry program was still processing the data, and the program would freeze. This would require the enumerator to reboot the PDA and then pick up the interview at the point where the PDA had locked-up.

• Software updates – Various software versions were released during the data collection period, forcing enumerators to return to computer-equipped data collection centers to update their software, downloading it from the central server. With the existing problems in data collection, along with insufficient information about the fixes introduced in the new releases, each new version of the software increased the anxiety of the field staff.

• Delays in updating data on management system – The Data Collection Management Indicators System (SIGC) was designed to perform an extensive set of data edits, and process that data simultaneously with administrative information. Problems encountered in transmitting the data from the field to headquarters resulted in long delays in updating the system, preventing the SIGC to be used as a real-time management tool. Since this SIGC was the primary tool for managing the field operations, this setback seriously compromised the quality of supervision of data collection. The alternative implemented, of having supervisors review the data collected directly from the enumerators’ PDAs, required enumerators to stop work until their PDAs were returned to them. This became a major source of dissatisfaction among the supervisors, and in the end, was not widely used. The absence of paper reports as a back-up, the fact that 80% of the data collection centers did not have a computer link to the Intranet, aggravated the situation.

3 As of 08/21/2007, data collection not completed.

9

A parallel management system had to be developed and implemented, being fed with essential information by field staff via web interface and a toll-free phone line.

These various problems added to the inherent difficulties of conducting any census operation, and led to tremendous work overloads for supervisors and coordinators, as well as discouraging enumerators. As the attrition rate for enumerators increased, a new round of testing, hiring, and training of replacement enumerators had to be completed. About 4,000 new enumerators were hired, trained, and allocated to various municipalities.

For the electronic data collection and management systems conceived for the CENSOS 2007 to work effectively, it is crucial that the data flow through the system smoothly. This means the PDAs must be able to adequately process the questionnaire and data collected without slowing down the interview process. Management reports must be updated in real time, or at least daily, which requires an efficient means of transmitting data from the field to headquarters. Having all data collection stations equipped with broadband internet access would have helped enormously. An important lesson learned is the wisdom of establishing independent systems for the flow of the administrative data and collected data from the field to headquarters, and from headquarters back to the field. It should be noted that the training devoted to the use of the PDAs, especially operating configurations, data transmission, and new resources such as GEOPAD, were not as effective as hoped. A more effective training in the technological aspects of the operations could possibly have alleviated some of the problems that were encountered.

8. Impacts

IBGE’s structure and surveys As a result of the need to update the infrastructure used for data collection with the new technology, all net of IBGE’s field offices, consisting of 27 state offices and almost 530 local offices, gained Internet broadband access. The acquisition of personal data assistants for data collection in CENSOS 2007 will impact IBGE’s data collection procedures long after the end of the census. The proposal is to gradually move from paper-based data collection to electronic means, with gains in both speed and quality. The 2007 National Household Sample Survey (PNAD 2007) will be the first continuous survey to change data collection methodology after CENSOS 2007. Other surveys will change as time goes by, culminating in the 2010 Population Census, which will be the largest data collection by electronic devices to be performed by IBGE.

The use of PDA’s was an important step, for both IBGE and Brazil, into the digital age. Roughly 80,000 IBGE employees and temporary enumerators were trained in the use of this technology and gained extensive experience using it. Additionally, millions of people were exposed to the technology, in itself important when we consider that only 13.7% of households in the country had internet access in 2005 (IBGE, 2007).

In agricultural statistics CENSOS 2007 produced an exceptional data set to use as a frame for future sample surveys of agricultural holdings. Combining the information from the address list (CNEFE) and the 2006 Agricultural Census provides IBGE an Agricultural Holding and Farmers List, which will allow it to conduct efficient sample surveys of agricultural holdings, based on location and type of production. It will be the first automated list including all agricultural producers in the country, with identities and characteristics of each holding. In addition to other information on the list, GPS coordinates of the agricultural holdings is also recorded. This position data, besides providing an additional address component, is important to guide field work, and can be employed in the eventual design of an area frame.

The flexibility and efficiency of statistical sampling from a list frame is well known. However, a continuously updated list is necessary to ensure ongoing quality of the sample results. Because of this, maintaining and updating the list is of paramount importance, and resources must be permanently allocated to this vital activity. The use of external sources, such as cooperatives, official agencies, and other institutions which have information about agricultural holdings and farmers, is one way to keep the lists current. To successfully manage this effort, IBGE will need to

10

forge ahead into new partnerships with other institutions, and assume new tasks in its lead role in the national statistics system.

The census enumeration areas used in the agriculture census provide an existing, defined division of all land area of the country. Combined with the information gathered in the agriculture for each enumeration area, this is the necessary information to create a precisely stratified area frame. Such an area frame would enable IBGE to develop sample surveys to accurately and efficiently measure such items of interests as area planted in particular crops, yield forecasts, some livestock, and final production estimates. This area frame would also provide an objective measurement of the completeness of the list frame.

The statistical infrastructure represented by the development of both a comprehensive list frame and exhaustive, well stratified area frame, opens the door to a number of possibilities that can impact Brazilian statistics and methodology of our surveys. Currently, ongoing programs to provide statistics on Brazilian agricultural are based primarily on indirect investigations collecting subjective data. Information is obtained from agricultural specialists and people with knowledge of existing conditions within one or more municipalities. In general, collecting precise information on individual holdings is not done in the survey programs. The results of the 2006 Agricultural Census provide an opportunity to change this situation. A continuous national agricultural survey system, using probabilistic sampling to ensure the highest possible accuracy of the results, has been a goal for many decades. In addition, it offers the opportunity to broaden the previous goals: the new sample surveys being planned will provide information not only for the major agricultural activities, but include separate estimates for enterprise and the household production, based on National Accounts recommendations. CENSOS 2007, in particular the 2006 Agricultural Census, gives us the tools to develop such a system, providing accurate estimates using list frame, area frame, and multiple frame surveys.

9. Conclusions

Despite being plagued by enormous difficulties presented by a short time frame and technical problems, the 2006 Agricultural Census represented a great leap ahead, both technologically and in terms of data quality, for an operation of such magnitude. The procedures for data collection and supervision were critical, and suffered as a result of the need to quickly adapt to new processes made necessary by technical limitations of the PDAs and the existing infrastructure. Introduction of such a revolutionary change in procedures should have been thoroughly tested with a pilot census, but unfortunately, we were given neither the time nor budget to conduct a pilot. This technology will be extended to other IBGE’s surveys and, in conjunction with the National Address for Statistical Purposes List, will allow us to collect data of higher quality in less time. After fixing the supervision and data collection problems experienced during the CENSOS 2007, and with more time to plan, 2010 Demographic Census will be conducted in a similar way.

10. References

FAO. 2006. World Census of Agriculture 2010. Available in: http://www.fao.org/es/ess/census/default.asp. Consulted in: 08/02/2007.

FORTES, L.P.S. et al. (2007). GIS-based census mapping approaches: Brazil experience. In: United Nations Expert Group on Contemporary Practices in Census Mapping and Use of Geographical Information Systems. Proceedings... 29 May - 1 June 2007, New York.

IBGE (2003). Metodologia do Censo demográfico 2003. Série Relatórios Metodológicos, v.25, Rio de Janeiro. IBGE: Rio de Janeiro, 2003.

IBGE (2007). Pesquisa Nacional por Amostra de Domicílios 2005 - Acesso a Internet e posse de telefone móvel celular para uso pessoal. Rio de Janeiro. IBGE: Rio de Janeiro, 2003.

SYSTEM of national accounts 1993. Rev. 4. New York: United Nations, 2004. Chap. IV Institutional Units and Sectors. Available in: <http://unstats.un.org/unsd/sna1993/tocLev8.asp?L1=4&L2=6>. Access: sep. 2007.

11

11. Acknowledgments The authors acknowledge Word Bank and USDA-NASS for financial support in their participation in ICAS-IV, Mr. Michael Steiner and Mr. Roger Beinhart (USDA-NASS) for their suggestions, and IBGE personnel contributing to the paper: Roberto Augusto Soares Pereira Duarte, Heleno Ferreira Mansoldo, Márcio Tavares Fernandes and Wolney Cogoy de Menezes.