Every ten years, the agricultural census allows a great deal of data to be collected on all French farms. A key operation of agricultural statistics, the census is also one of the most important statistical operations carried out in France.

The census covers all farms, whether large or small and whether agriculture represents the main or secondary activity of the farmer. In order to ensure full coverage of the agricultural world, more than 500,000 units are thus surveyed.

For each farm, more than 900 pieces of data are collected on cultivated areas, livestock, labour, production methods and marketing methods, together with on-farm product diversification and processing activities. The census thus provides an accurate and exhaustive snapshot of the agricultural world and its diversity, both in metropolitan France and in the French overseas departments.

Agriculture is constantly adapting, echoing changes in society. The sector is using digital technologies, diversifying its marketing methods and markets, reviewing its practices to protect the environment, developing quality labels, etc. The whole reality of professionals in the agricultural sector is changing. Similarly, the agricultural census has had to evolve and the 2020 version set the stage for several structural innovations.

The majority of data is now collected online or by telephone. All communications with the respondents have been reviewed. The component with additional data on labour and livestock buildings is collected for a sample, for which the sampling method has been optimised. The interviewers have been trained online. These innovations have facilitated the continuation of collection despite the health situation imposed by Covid-19.

Due to their exhaustiveness, census data are important references for all stakeholders in the agricultural world: farmers, professional organisations, trade unions, researchers, public authorities, etc. They make it possible to analyse French agriculture and its developments and to measure its weight within the European Union. Agriculture is indeed a strategic sector, not only as an economic sector, but also for its role in food independence, in the face of population growth and climate change. In this context, the results of the agricultural census will contribute to the direction and evaluation of agricultural and food public policies, implemented at regional, national and community levels.

The census data are also used to quantify measures linked to sustainable development and agri-environmental policies and to evaluate policies on economic support for farms. For example, the results of the previous census were used to simulate the changes in the classification rules of territories eligible for the Natural Constraint Compensation Payment (Indemnité compensatoire de handicaps naturels – ICHN). This aid supports farmers in territories where the production conditions are more difficult than elsewhere, as a result of natural or specific constraints.

The agricultural census makes it possible to evaluate not only the state of agriculture, but also its position and evolution, by comparing the results with those of previous censuses or of other European countries. This need for information about the agricultural sector far exceeds the national framework.

The FAO, the United Nations Food and Agriculture Organization, has established a global programme and developed concepts and methods to support the coordinated performance of agricultural censuses around the world (Ouvrir dans un nouvel ongletFAO, 2020).

Within the European Union, the Regulation on Integrated Farm Statistics (IFS) organises the performance of the census in the 27 Member States (Eurostat, 2020).

The collection periods are harmonised. While a common basis of information is to be sent to Eurostat by all countries, each of them may also decide to add questions to meet national needs.

Farm Structure Surveys (enquêtes sur la Structure des exploitations agricoles – ESEA) are carried out between the ten-yearly censuses to update the data (Figure 1). The next ESEAs will be carried out in 2023 and 2026.

Figure 1. A comprehensive, modular European system, spread over time

The European Regulation also distinguishes between two types of data: a common core that must be completed exhaustively for each farm and eight modules that can be collected from samples. In 2020, the three modules to be collected focused on labour, rural development and animal housing. In 2023, the module on animal housing will be replaced by four modules: irrigation, soil management, machinery and equipment, and orchards.

Member States must be able to guarantee the quality of the data collected, but they remain free to choose how they obtain the information requested: questioning via the Internet, the telephone or in person, the use of administrative sources, etc. Online collection has been developed in many countries (Austria, Germany, Spain, etc.).

The organisation of the 2020 agricultural census was affected by the Covid-19 pandemic in most European Union Member States. The latter delayed several activities, notably the preparation of the collection tools, the dissemination of information to interviewees, the recruitment of interviewers, the training of staff and the collection of the data themselves. However, the health situation has also paved the way for a modernisation of the collection process and working methods. Several countries, including France, have thus made greater use of administrative data sources and the collection of questionnaires online and by telephone.

However, in the case of a census, the primary concern is to ensure its exhaustiveness.

A census must be exhaustive. This is why the first issue to resolve is the statistical universe of this operation. All farms are surveyed in the agricultural census, both in metropolitan France and in the French overseas departments. However, only one person on each farm is invited to respond to the survey: it is usually the farm manager, even in cases where agriculture is not their main activity.

Some of the units surveyed will, ex post, prove to be outside the coverage of the agricultural census: to ensure that only farms exceeding a certain economic size are included in the results, rules are applied on the basis of the characteristics of the farm recorded during the census (cultivated area, nature of the production, etc.) (Box 1). These rules are based on thresholds, defined at European and national levels. The thresholds used at national level are generally lower than those used at European level. They are mostly based on those in effect at the time of the previous census (2010) in order to allow continuity in the statistical series. The European regulation on agricultural statistics (IFS, see below) thus sets a minimum threshold for the utilised agricultural area (UAA) of 5 ha, compared with the French threshold of 1 ha.

In theory, compliance with these conditions ensures that the farm selected is an economic stakeholder capable of participating in a commercial transaction based on its production.

However, how can the list of statistical units that constitute the coverage of the collection be established on an ex ante basis?

During the 2010 census, the Department of Statistics and Foresight Analysis (service de la Statistique et de la prospective – SSP) called on the 36,000 municipalities which needed to verify the lists of farmers within their territory established by the regional departments (SRISE). This work, although useful, was however subject to interpretation errors in relation to the concept of farming and omissions; in short, it was of very heterogeneous quality.

In order to avoid this shortcoming, the statistical universe of the 2020 census was deliberately established using a broad base of already available information. Starting with the Business Register Identification System (SIRENE), which is managed continuously on the basis of declared start-ups and cessations, it was supplemented with the administrative files of the agricultural sector: declarations under the Common Agricultural Policy (CAP), Computerised Vineyard Register, animal movement files and the Agricultural Workers’ Mutual Benefit Fund (Mutualité sociale agricole – MSA) and using the thematic surveys carried out by the Department of Statistics and Foresight Analysis (SSP).

A great deal of work was then done to update the farm contact details, in particular the email addresses and telephone numbers. Thus a directory, the Farms Sampling Frame (base de sondage des exploitations agricoles – BALSA), was established and renewed in 2021 to ensure that, ultimately, the agricultural census data make it possible to validate the universe of farms and that the integration of administrative data is performed automatically on an annual basis. This makes the census a key element for other agricultural statistics operations.

Since its inception, the agricultural census has been collected in France in person by interviewers. For the 2010 edition, the interviewers were for the first time provided with computers to monitor the data entered in real time. However, it was the implementation of true mixed-mode data collection that would constitute the main innovation in the 2020 agricultural census.

The overhaul of the collection method had two objectives. First, online collection should facilitate the feedback of data: the service is open 24/7, meaning that those using it can complete the questionnaire at a time that suits them best, according to the work schedule imposed by running their farm. It should also make it possible to lighten the burden on the Official Statistical Service.

Specifically, in 2020 and for metropolitan France, the Agricultural Statistical Service chose to use three collection methods, online, by telephone and in person (Figure 2):

Data was collected between October 2020 and May 2021.

Figure 2. Two mixed-mode protocols for two levels of detail in the data collected

Almost 400,000 units were invited to respond to the questionnaire securely online. To do this, they received a letter in advance that contained a username, a password and a link to the collection site, as well as contact details for assistance: a freephone telephone number and a contact email address. After a certain amount of time with the respondent not connecting to the site, reminders, first by mail then by telephone, were issued to encourage a response to the online questionnaire. Finally, in a third phase, the collection was conducted by telephone.

Two survey institutes were selected to perform the collection by telephone. Each service provider was responsible for six regions in metropolitan France, and needed to conduct the protocol in successive rounds of a maximum duration of three months, in order to spread out the activity of their telephone platform. The service providers also issued SMS reminders to non-respondents for whom the mobile phone number was available.

The telephone platforms were linked to a dedicated information system, making it possible to optimise the call rate and automate tasks such as call pick-up. These tools have enabled the French Official Statistical Service to cut the cost of collecting the units for the common core of questions.

The cumulative time taken for the online and telephone questionnaire, as determined by the service providers, was 20 to 25 minutes on average, compared with 50 minutes for collection in person. This is easily explained by the difference in the length of the questionnaire: when conducted in person, the survey included the questions of the three modules, in addition to the common core of questions.

The service providers achieved a return rate of 91%. This rate varies mainly depending on the degree to which the contact details were known at the time of collection. All of the units interviewed by the service providers for collection online and by telephone were classified into five groups, from group 1 for whom all contact details were completed (landline telephone number, mobile phone number, email address) to group 5 for whom only the postal address was known.

Almost 70,000 farms with their registered office in metropolitan France were drawn from a representative sample and then interviewed on the basis of the more detailed questionnaire referred to as “complete” because, in addition to the common core of questions, it includes the three modules for 2020. Sampling has led to this type of collection being spread to all farms in Corsica and the French overseas departments.

Collection in person and reminders for non-respondents to the common core of questions were carried out by the network of interviewers of the Ministerial Statistical Office for Agriculture. This network has been around for many years and is composed of around 850 people, many of whom are from the agricultural world (farmers or former farmers and farmers’ partners, in particular), meaning they have a high level of skill and a certain degree of legitimacy when conducting surveys in person. It was bolstered by almost 300 people recruited for the purposes of the census.

Despite the situation linked to the Covid-19 pandemic, the interviewers emphasised the quality of the welcome they received from farmers, “people who are passionate about their work and who like to talk about it”.

Accompanied by a manager from the Regional Statistical Service, particularly during their first interviews, the interviewers received a new kind of training, which constituted the second major innovation of the 2020 census.

Until 2019, all interviewer training of the entire agricultural survey system was provided by the Regional Departments of Statistical and Economic Information (SRISE), which were themselves trained by the survey managers at national level (French Official Statistical Service).

There were several problems raised by these training processes. Failings were sometimes found in the transmission of instructions to regional trainers and then to interviewers. In addition, they created a heavy logistical burden of needing to find almost 80 training locations and manage the travel of interviewers. In retrospect, it is assumed that these difficulties would have been particularly accentuated due to the restrictions on travel and meeting in groups imposed by the health crisis in 2020.

There was an initial experiment in the online training of interviewers carried out in 2019, in relation to the Teruti survey. The result was the good participation by the network of interviewers, a majority of whom expressed satisfaction, ending with a positive opinion of the tools developed and an appreciation of not having to travel far from home. The main difficulty stemmed from poor or even non-existent internet access.

Buoyed by these lessons, the team in charge of the agricultural census designed online training in partnership with members of the regional trainers network. The participation of the Regional Departments of Statistical and Economic Information (SRISE), as part of a working group responsible for the design of the media, was particularly important for the success of this operation. The introduction of online training is indeed a major change in the relationship between the survey managers in the Regional Departments of Statistical and Economic Information (SRISE) and the network of interviewers.

This training was created with a structure based on that of the census questionnaire. The training platform reproduces the graphical appearance of the questionnaire and the response applications so that interviewers can easily link the collection to the concepts presented in the training. For each part, an introductory video describes the points that will be covered in the modules that follow. Each module consists of an animated slideshow with comments, presenting the concepts to be learnt. Finally, there are quizzes to enable interviewers to test their understanding and memory of the instructions.

In order to facilitate the subsequent reuse of the training modules, the decision was made not to mention the specific characteristics of the agricultural census, such as the literal wording of each question or the growing year for example, in the comments. Thus, some modules may be reused in the context of future surveys conducted by the French Official Statistical Service, thereby ensuring consistency in the concepts handled throughout agricultural statistics.

The training programme is adapted to each type of questionnaire. Specific modules have thus been developed for the French overseas departments, the question processes of which have been adapted (see Box 2).

An individual account has been created for each interviewer. Thus, the Regional Departments of Statistical and Economic Information (SRISE) survey managers, who previously formed the network of trainers in the regions, were able to track the training activity of all the interviewers under their responsibility. While 1123 interviewer accounts were created, only 943 interviewers actually completed the online training. An offline solution was therefore offered to those who had Internet access problems. The media were also used by the service providers to train their own telephone platform operators.

In order to ensure good response rates online, there are several levers available, particularly in relation to improving reminder letters and emails, the telephone support and the website used for the collection.

This approach is part of the actions led by the Inter-ministerial Directorate for Public Transformation on the simplification of administrative documents (Ouvrir dans un nouvel ongletDITP, 2021). These actions are based on the lessons learned in the area of behavioural science, developed in particular by D. Kahneman, D. Ariely, C. Sunstein and R. Thaler. This work has shown, firstly, that there is cognitive bias in decision-making and, secondly, the predictive nature of these decision biases: an individual may not necessarily make the most rational decisions, but their decision is not surprising. In fact, most often, individuals use an automatic and fast “system” that is linked to emotion rather than a slower reflexive method. It is then possible to apply an analytical approach to use these cognitive biases with the aim of getting an individual to behave in the desired manner. This is the principle behind Nudge theory, which will be used to try and boost responses online.

Based on this approach, the letters sent to respondents were reviewed in relation to both substance and form, following a working group requesting the opinion of a selection of farmers.

The experience first showed the advantage of increasing the personalisation of the letters and of using the first person. Letters are therefore addressed individually and signed by the head of the Ministerial Statistical Office, because identifying the sender is also important. The official nature of the letter provides reassurance and so re-formatting was therefore necessary to comply with the government’s communication charter (Figure 3).

Figure 3. Application of principles of Nudge theory to the agricultural census letters

More encouraging phrasing was also sought. The respondent adopts the “good” behaviour when they have a minimum amount of guidance. The prominence has been calculated to draw attention to the important elements, which needed to be easily identified, in particular the three steps of connecting to the questionnaire. The visuals used (three steps, three icons) helped highlight the desired message: “The agricultural census, it’s quick and easy”. A distinction is therefore drawn between the new version of the letter and the versions previously used by more graphic and airy visuals and more informative wording.

Finally, elements were added during the site survey to show that many other farmers had already responded: this meant that pressure to conform to social norms helped to trigger a response from those who had not yet responded. Other improvements were made to make the freephone number more visible and to add a contact postal address, as some farmers want to send a letter; in short, these efforts were made to make the operation as easy as possible for the respondents.

In theory, due to its exhaustive nature, a census does not use a sampling method. However, to retain the possibility of collecting responses from the majority of farmers online, it was important that the questionnaire not be too long or too complex. In addition, the new European regulation made it possible to switch part of the collection to sample-based surveys. This is why it was decided that the questions included in the supplementary modules (on livestock buildings and labour) should not be asked to all farms but only to a sample. This first sample for an agricultural census was created to produce representative results at departmental level.

In terms of the volume of data collected and the number of farms interviewed, the sample for the 2020 census is comparable to a Structure survey (ESEA), an interim survey conducted twice between each census (see above), the latest edition of which was relatively recent (2016). Rather than replicating the samples of the ESEAs, the decision was made, in the context of the 2020 census, to conduct a full review of the sampling method, along two lines:

The objective of the 2020 agricultural census is to describe farm production, with the cultivated areas and livestock, as well as the main production factors used in agriculture, in particular the employed labour force and the land use method. Questions also relate to engagement in specific approaches (quality and environmental approaches), diversification of activities and marketing arrangements for products.

The 2020 questionnaire is based on information already known elsewhere, for example by pre‑completing the cultivated areas based on “CAP” declarations (see below). Indeed, in order for online collection to be successful, it was essential to reduce the response burden for farmers and to reduce the size of the questionnaire. Two sources were used in particular:

Matching between the units surveyed in the census and the administrative sources was carried out based on the establishment number, the SIRET. This identifier is sometimes completed incorrectly in the sources used, which requires a thorough check during the compilation of the census launch file.

The imputation of CAP data instead of the farmer’s declaration has also led to three types of problems. First, differences between the crop classifications used in CAP data and that used in the agricultural census require additional questions to be asked for certain crops. For example, crops declared under the CAP as “Potatoes for fresh consumption” must be split between “Early or new potatoes”, “Ware or mid-season potatoes” and “Potato plants”. Differences in classifications are particularly marked for grasslands, feed legumes and fruits. A medium-term strategy aimed at better linking the CAP classification with the census questionnaire (which is also the result of a European regulation) should iron out the differences.

The second difficulty is that the coverage of the CAP varies considerably depending on the crops, which requires compensation through additional surveys.

Finally, the data declared under the CAP sometimes vary greatly with the data declared by the farmer. A comparison between the CAP data and the data declared in the 2016 ESEA shows that between 10% and 20% of farms, depending on the crops, have an absolute difference of more than 5% between the CAP declaration and the survey. However, this finding does not prejudge the quality of one source compared to another. In the survey, farmers tend to round off the areas and sometimes confuse crop seasons, giving the areas of the previous or next season.

For the 510,000 units interviewed, 494,000 questionnaires were collected, which is a return rate of 97% across French national territory, in both metropolitan France and the French overseas departments. The collection of the 2020 agricultural census is therefore a great success (Box 3). This has allowed for five innovations that affect all stakeholders in the collection operation:

This result therefore supports the Official Statistical Service in most of the choices made in advance. First, the flexibility of the online solution has proven to be well-suited to the activities of farmers. In the end, only one in five responses was obtained by telephone or by mail.

Assistance by specially trained operators was essential for good participation online. In addition, the diversity of media for reminders (letter, email, SMS, telephone) enabled the collection to be managed efficiently.

The reminder by telephone by the network of agricultural interviewers, after the work of the service providers, paid off, as 63% of the 35,000 non-respondents were recovered. The link between the two collection methods is therefore effective.

The telephone reminder and telephone interview phases are complementary and can be carried out simultaneously, managing the focus in each of the phases in accordance with the priorities. During the reminder phase, farmers who cannot or do not wish to respond online must be able to be interviewed immediately by telephone. During the interview phase, a person who claims to want to respond online will not respond by telephone, even if the interviewer is persistent.

The duration of the questionnaire varies according to the characteristics of the farm and the response method, but it remains within acceptable standards:

This 2020 edition also reveals a set of areas in need of attention or improvements to be considered for future collections.

A key element in ensuring a good response rate is having all the contact details possible for the unit being surveyed. Manually seeking out additional contact details can improve the participation rate by a few points. This is indeed costly, as it is carried out by a survey manager, but it should remain less expensive than collection in person.

In self-administered mode, several short and simple questions are sometimes preferable to a single long and complex question, particularly online when the screen is read quickly. Built-in controls make it possible to ensure good data quality, but they must be well calibrated given the very wide variety of situations encountered.

Some farmers did not feel that the agricultural census applied to them and therefore did not respond to the questionnaire without prompting. This is the case, for example, in respect of certain wine growers or beekeepers. A notification that includes them more specifically should be developed for the next agricultural census.

In addition to the support solutions proposed during the collection, a voice response unit could provide information on the next steps and explain possible solutions for responding.

The organisation decided upon used three data entry applications (two for the service providers and one for the Official Statistical Service). This created a very high burden in terms of specification, monitoring development and ensuring correct functionality for each of them. The creation of a single application for all stakeholders involved in the collection operation now seems more optimal and, in any case, the possibility should be examined.

The context of all the work carried out in three years, from preparation to the end of the collection, was obviously impacted by the health crisis. This did not prevent the smooth running of the protocol: the 2020 census, with its range of new features, resulted in a very effective coverage of the agricultural universe by the end of 2021.

The dissemination of the results is accompanied by a sixth innovation. The first results are presented using data visualisation on all media, particularly on smartphones. It uses techniques that combine graphics and explanations: the significant parts of the visuals are highlighted or zoomed in on as the user scrolls down the page (Ouvrir dans un nouvel ongletOfficial Statistical Service, 2021). The aim of this method of dissemination is to reach all audiences, first and foremost farmers.

Ouvrir dans un nouvel ongletRèglement (UE) 2018/1091 du Parlement européen et du Conseil du 18 juillet 2018 concernant les statistiques intégrées sur les exploitations agricoles. In: site EUR-Lex. [online]. [Accessed 10 December 2021].