IBGE (Bolliger et al., 2012) has adopted the concept of the MSF for the household survey system, which is based on census EAs. This MSF is being used for the System of Integrated Household Surveys, in which all individual household surveys use the same frame.
IBGE later initiated a National System of Agricultural Establishment Sampling Surveys (Sistema Nacional de Pesquisas por Amostragem de Estabelecimentos Agropecuarios - SNPA in Portuguese) with a view to also develop an MSF for agricultural surveys, using the multiple frame approach by combining census EAs as the area frame and a register of farms. The MSF was designed to be based on the 2006 agricultural census, but also uses information from the 2010 population census. The 2006 agricultural census enumerated 5.2 million agricultural holdings, while the 2010 population census enumerated only 2.6 million. It is plausible that many small holdings were identified as households but not as farm establishments in the population census. To build the area frame, it was decided to map the EAs from the 2006 agricultural census into those used for the 2010 population census. Attempts were also made to develop a list of agricultural establishments, by combining agricultural census records with those from the Central Registry of Enterprises and administrative records provided by employers to the Ministry of Labor and Welfare. This turned out to be difficult; it was not possible to identify a large number of units that were present in more than one register, despite the use of advanced linkage methods.
Finally, given the above discrepancies between the 2006 agricultural census data and the 2010 population census data, and the other difficulties mentioned, in 2013 IBGE decided to postpone the implementation of the National System of Agricultural Surveys and wait for the next agricultural census, that was planned for 2016 (Bolliger, 2014). However, the uncertainty regarding the date of the Agricultural Census because of budgetary reasons, led IBGE to study alternative methodologies for building a master sampling frame for implementing SNPA.
One of the alternatives currently under consideration is to use administrative records and build the MSF by developing a list frame. This will restrict the SNPA’s target population to the set of units of the agricultural establishments that are formally registered by the public federal or state administration; this will be possible thanks to the advancements observed in the country’s administrative records. This list will exclude agricultural establishments whose primary function is leisure, housing or livelihood. It will also exclude producers for whom agriculture is a secondary activity, one for which the holders do not deal with the government for funding or technical assistance, nor do they engage in marketing products to the point of seeking to observe the provisions of movement of goods. It is assumed that the impact of excluding those units will be limited since the current scope of the administrative records – in particular, the Pronaf register, which is directed towards family farms and now has almost five million active records. In this scenario, the SNPA’s sample frame would be a list of rural producers. The Agricultural Sample Frame formed by the union of the main sample frames maintained by the federal and state governments that register agricultural producers, updated periodically, would form a List Frame, focused on SNPA’s ordinary surveys.
This approach is expected to present some advantages, including the possibility of annual updating of the system, independence from the agricultural census, and a low implementation cost. On the other hand, the target population will be more restricted in its scope and coverage. It will not be possible to integrate it with the agricultural census, and the data quality will depend on the quality of external registers, with all the risks relating to the difficulty of accessing these registers as their content changes over time.
Another possibility under consideration is to use the results of the new Project of Land Use and Coverage developed by IBGE’s Geosciences Directorate. The project’s main goal is to monitor the changes in the use and coverage of land for the entire national territory, at regular intervals (every 2 years), from the acquisition and processing of MODIS images. In the Project, six categories and classes of land use and coverage are included, with the information provided through a territorial grid for statistical purposes. The incorporation of land use and coverage data in the territorial grid for statistical purposes makes it possible to obtain area data on the categories of land use and coverage for each square. The use of an area frame with land use and coverage information for stratifying and selecting squares requires the use of mobile devices and GPS to identify the field limits. This option also presents some advantages, including the possibility of defining segments with the same size, selecting segments that are multiples of 1 x 1 km2, and the opportunity for biannual updates independent of census data. Limitations include a lack of visually identifiable boundaries, a greater variability of the number of establishments in each segment, and less detailed information than that provided by the census20.
The use of both alternatives require surveys, studies and field experiments to evaluate their viability and efficiency. At the time of writing, the review is still underway; therefore, the concept of master frame for agriculture is not yet operational in Brazil.
20 The information from images (on the use of land alone) are usually more aggregated, divided into few categories. These may provide only a more general view of the land use. On the other hand, census results can provide much more detailed and specific information on land use, thus enabling a better distinction between different crops, forest plantations and pastures. Furthermore, census information can go beyond land use. An area frame stratification based on census data can take into account activities such as small livestock, and adopt different measures such as production value, sales and others.
However, IBGE has experience with building a sampling frame for agriculture. This experience derives essentially from the Harvest Forecasting and Monitoring Survey (PREVS in Portuguese), initiated by IBGE in the mid-1980s and implemented up to 1993 with the financial support of the Banco International por la Reconstruction et lo
Desenvolvimento (BIRD) (Bolliger, 2014). The PREVS basically followed the methodology then adopted by NASS/
USDA in its June Survey and December Survey. Its main objectives were to provide statistical information on harvests through objective data collection and use of probability sampling methods that allow confidence intervals to be computed for the final results.
The survey followed an area sampling frame design, stratified according to land use and systematically selected in a single stage, with equal probability and with no substitutions.
The area sample frame was constituted by strata of land use, established according to the rate of cultivated land or by the predominance of crops, divided in Counting Units (CUs); these were subdivided in Area Segments, the survey’s sampling unit.
The main material used included (i) Statistical municipal maps generally at a scale of 1:100,000, prepared for census purposes and showing each municipally divided into EAs, (ii) Political maps and Land use maps; (iii) Topographic
maps, generally at scales 1:50,000 and 1:100,000, established using information from IBGE’s agricultural census, the
Agricultural Production Systematic Survey, Municipal Agricultural Survey, and the Municipal Livestock Survey; (iv)
Satellite imagery TM/Landsat-V on paper, at scales 1:100,000 and 1:250,000; (v) Planimetric coordinates, produced
at the same scale as the TM/Landsat images, used for visualization and geographical location of the interpreted patterns and (vi) Aerial photographs, covering the selected segments with photo enlargements to 1:10,000 used for the annual field data collection.
The field data collection for each survey round was undertaken by a total of 20 supervisors and 200 enumerators. Common difficulties on conducting the fieldwork were reported, including: problems in locating the selected segments; deficient roads and paths which hindered access in several cases; difficulties in locating respondents; difficulties in determining the existence of the household/headquarters or, alternatively, verifying whether the largest part of the establishment was within the segment, if the producer lived in the city.
This area frame was complemented by a list frame, that consisted of a relatively small number of Special Holdings based on the 1985 agricultural census information and that accounted for a large percentage of the total variable. This was updated every year, making it a multiple frame.
Despite the reported challenges encountered in the field work, this survey was successful, since it produced acceptable results (66 percent of crop estimates had a CV between 15 and 25 percent). However, it was ultimately discontinued. It is interesting to analyse the reasons why this technically sound survey was discontinued. Among these, a shortage of financial resources, personnel and institutional support are mentioned. However, beyond these, key additional factors were the lack of national coverage of the survey, the lack of timeliness in providing results (a crucial factor for forecasting) and the country’s longstanding tradition of agricultural statistics production based on subjective surveys As in many countries, a survey using sound statistical methods was abandoned for the Agricultural Production Systematic Survey, which with much less effort and cost, has been providing for over 40 years monthly estimates of planted area, harvested area, medium yield and production for over 30 products and for all the Federation Units, even with the generation of such quantitative information through subjective surveys being methodologically condemned21. In normal years, when the Harvest Forecasting and Monitoring Survey results were being launched, the estimates from the Systematic survey were already found to be consolidated and agreed between users and producers (Bolliger, 2014).
Among the lessons learned are the following:
• The population census did not contribute to the development of the master frame.
• Despite the use of advanced record linkage methods, the linkage of administrative records with other list frame records was not satisfactory. This also holds true for the US.
• When designing a survey, care should be taken to ensure that the methods used are in line with its objectives. • Building a master frame for agriculture is much more challenging than for household surveys.
In summary, the lessons learnt from Brazil indicate that when selecting a viable option for building an MSF for agriculture, all relevant factors should be considered, including cost and resources, material and information available, institutional support and correspondence between the methods adopted and the survey objectives. This reinforces the need for guidelines on building and using an MSF to take into account factors beyond sampling and frame construction, such as measurement methods, timeliness and resource requirement.