Agricultural census 2020 – how to reduce costs and burden? The European Statistical System approach

2.1.1 Thresholds and frame extension

The previous regulation on farm structure statistics required all member states to cover holdings above certain physical thresholds. In a few cases, these thresholds were too high to achieve sufficient coverage of UAA and LSUs, and some EU countries collected data from almost all their farms. IFS sets lower physical thresholds, but if the requirement of achieving 98% UAA and LSU coverage is fulfilled, EU countries can apply to raise the physical thresholds, so as to reduce the number of holdings from which data are collected. So the IFS thresholds are flexible and can even be raised above the ones set in the previous regulation if the 98% coverage requirement is met (R1).

Under IFS, if countries cannot achieve the required 98% coverage when applying the thresholds, they should extend the frame to include a sample of small farms that fall below these thresholds (see Fig. 1); this sample should collect only core variables (not all variables, as was the case under the previous FSS regulation) (R2). The frame will be extended to include small farms in 2020 only (not in 2023 and 2026). This contrasts with the previous regulation, which required small farms to be surveyed in each reference year (R3). In addition, the frame extension can be applied to a sample in 2020, whereas a full census was required in 2010 under the previous regulation (R4). The smallest farms are surveyed only once a decade because they produce mainly for themselves; it is assumed that they are fairly homogeneous and change relatively little over time. This can be proven by the fact that the sampling rate in the stratum of small farms is lower or much lower than in the stratum of large farms. In other words, small farms are undersampled and large farms oversampled as shown by Eurostat’s analyses of datasets. In order to meet a given relative standard error, statisticians thus need fewer smaller farms than larger farms in the sample because smaller farms are more homogeneous while larger farms are more atypical and account for the largest part of the variance.

In addition, each individual small farm can be unstable, but small farms in their entirety change relatively little over time. Large farms are the drivers of agricultural evolution and they account for the largest share of variables’ totals. Their behaviour thus determines overall agricultural evolution. In contrast, small farms’ importance lies mainly in the social dimensions of agriculture and rural areas. This change will be beneficial to EU countries with many small farms, such as Bulgaria and Romania. It represents a departure from the previous system, under which all farms were included in the populations covered by the former censuses and sample data collections.

Under IFS, modules are generally collected on holdings above given physical thresholds (main frame, see Fig. 1). However, the relevant population is further reduced for particular modules (R5). In 2020 and 2026, for example, the ‘Animal housing and manure management’ module will only be collected from farms with livestock, which account for between 20% and 80% of all farms, depending on the EU country concerned.

2.1.2 Source agnosticism

EU national authorities can use a variety of methods and sources to collect data, provided that they meet the necessary quality requirements. These methods and sources are:

S1: statistical surveys: census (a) or sample (b), S2: administrative data sources, and S3: other methods and innovative approaches.

For the census year, the IFS regulation allows samples to a greater extent than the previous regulation. While most variables were collected by census in 2010, the IFS regulation states that a sample-based approach can be used for all modules’ variables in 2020 (R6). Censuses and sample data collections can be implemented through surveys (S1a and S1b), by making use of administrative registers (S2), or through other approaches (S3).

Administrative sources (S2) based on EU regulations and where a certain control is implemented can be used without any justification. The previous regulation explicitly mentioned fewer such administrative sources and required prior information on the methods and the quality of the data from additional administrative sources. IFS explicitly mentions more administrative sources, thereby reducing the number of cases in which prior information on their use and quality is required (R7). The administrative sources which are regulated to support the CAP are the organic farming registers, the vineyard register, the integrated administration and control system, the systems for the identification and registration of bovine, ovine and caprine animals and the administrative sources associated with specific rural development measures. For example, in the 2020 agricultural census, the data for the ‘Rural development’ module on farms benefiting from rural development measures are expected to be collected entirely from administrative registers, while the variables on livestock in the core are expected to be collected from the animal registers.

Other methods and innovative approaches (S3) refer to modelling, expert estimates, remote sensing, and so on. Some of the variables in the ‘Animal housing and manure management’ module can be estimated using models. For example, the annual average number of animals in each category can be estimated using models based either on the number of animals raised, divided by the number of livestock raising cycles per year, or on a combination of the number of places and the number of empty days. To give another example, the quantity of manure produced by a given category of livestock can be estimated on the basis of the number of animals under certain types of management.

Eurostat also allows novel data sources such as Big Data if NSIs can make use of them. At the moment, there is still room to increase the use of and improve the quality of administrative sources in the ESS. NSIs and Eurostat are currently focusing more on these sources than on Big Data because they are more structured and less complex, their representativeness and coverage are often better known, they have a manageable volume, and their types of errors are typical so that classical statistical methods can be used to process the data and measure quality aspects.

Such methods reduce the administrative burden on national authorities and the costs they incur, but also the effort that farmers have to make to recall and estimate this information (R8).

Under IFS, in order to allow member states to flexibly choose the source and reduce the burden of data collection further, information on the ‘Machinery and equipment’ (in 2023), ‘Orchard’ (in 2023) and ‘Vineyard’ (in 2026) modules may be based on the year directly preceding or following the reference year, as long as it reflects the situation in the reference year (R9). This information is assumed to be slow-changing. Previously, all data had to be collected for the reference year. NSIs are obliged to respect the provisions set in the EU regulations and make sure that the information in these modules refers to the situation in the reference year. Instruments to ensure this include trainings for interviewers, post-enumeration surveys or cross-checks against variables available in administrative sources for the reference year in question.

While the increased use of S2 and S3 reduces the overall burden, it poses challenges with respect to assessment of data quality. Data quality is affected by both sampling and non-sampling errors. Only data collections based on samples are affected by sampling errors. In this case, the theory for probability samples allows for the sampling design to control for this type of error, and the IFS regulation sets precision targets for certain variables collected on a sampling basis. However, it is impossible to control for non-sampling errors or to target them in advance, because they happen in a non-controlled manner for every statistical process. They cancel each other out or add to or multiply each other, depending on the specific data collection and the specific context. No sound theory is available to predict them. This is more problematic in cases S1a, S2 and S3, where non-sampling errors have a big impact on overall data quality. After the data collections are carried out, multiple indicators can be calculated for each source of non-sampling errors associated with the different steps in a statistical process, while a unique synthetic indicator, such as the mean square error, is not computable. Thus, synthetic quality indicators cannot be set in a regulation such as IFS. However, national authorities are required to describe in quality reports a set of indicators that are both quantitative and qualitative. Quality assessment for multisource data is under continuous development in ESSnet projects. These are networks of several ESS organisations which aim to provide results that will be beneficial to the whole ESS, such as the ESSnet on Quality of Multisource Statistics.

The move towards source agnosticism improves subsidiarity, enabling EU member states to choose the most cost-effective sources for themselves according to their national conditions and needs. It also increases flexibility, thanks to greater openness to future data sources such as precision farming data.

2.1.3 Reduced number of variable breakdowns

The 2010 agricultural census covered a total of 273 variables, collected from all EU farms.

In the 2020 census, the core will comprise 184 variables and will be supplemented by 30 variables in the module ‘Labour force and other gainful activities’, 15 in the module ‘Rural development’, and 70 in the module ‘Animal housing and manure management’, totalling 299 [5]. These are maximum numbers, as member states may transmit fewer variables if particular items do not exist or are not significant in the country concerned.

The above information refers to the absolute numbers of variables in 2020. Burden reduction should be assessed throughout 2020–2026, as the data for the modules will be collected less frequently in the new decade (R10). Moreover, much burden reduction will be achieved in 2023 and 2026 by taking the previous regulation on permanent crops into account as well: in 2023, IFS will collect a maximum of 470 variables, including the ‘Orchard’ module, whose predecessor regulation contained over 650 variables, while in 2026 a maximum of 350 variables are allowed, including the ‘Vineyard’ module, whose predecessor regulation contained almost 900 variables (R11).

In addition, the core and module system makes it possible that fewer farms will have to provide data on all variables in a census or sample year. This reduces the burden on individual farms (R12).

Furthermore, the same list of variables and definitions, as well as common quality standards and data transmission deadlines, will be used in all EU member states, increasing interoperability and reusability so as to reduce the costs and burdens of data collection (R13).