Institut national des statistiques et des études économiques (STATEC), Luxembourg

The institutes involved in the data collection process are:

- STATEC, as coordinator and compiler for the micro-data;

- Service d'Economie Rurale (SER), as provider for both direct survey and administrative sources data.

FEAGA - Le Fonds européen agricole de garantie

IACS - Integrated Administration and Control System

LSU - Livestock unit

MAFEA - Database of the Ministry of agriculture on the payments made to farmers and other beneficiaries eligible for FEAGA or FEADER

SER - Service d'économie rurale

UAA - Utilised agricultural area

Excerpts in French from the modified law as of April 18, 2008 on the revision of rural development support:

(1) Au sens de la présente loi, la notion d'exploitant agricole couvre l'ensemble des activités des agriculteurs, viticulteurs, éleveurs, arboriculteurs, horticulteurs, pépiniéristes, jardiniers, maraîchers, pisciculteurs, sylviculteurs et apiculteurs.

(2) Par exploitation agricole, on entend une unité technico-économique à caractère agricole gérée distinctement de toute autre, disposant d'un ensemble de moyens humains et matériels, et comprenant en propriété ou ayant à sa disposition permanente, notamment par voie de location, tous les moyens de production nécessaires permettant d'en assurer une gestion indépendante, dont notamment les bâtiments, les machines et les équipements et exploitant au minimum 3 hectares de terres agricoles ou 0,10 hectare de vignobles ou 0,50 hectare de pépinières ou 0,30 hectare de vergers ou 0,25 hectare de maraîchages.

(3) Par association d'exploitations agricoles, on entend la fusion de deux ou plusieurs exploitations agricoles. (...)

(4) Par entreprise, au sens de la présente loi, on entend un ensemble de moyens humains et matériels concourant, sous une direction économique, à la réalisation d'un objectif économique."

The survey is carried through at the headquarters of the holding. It is considered that the whole surface of the agricultural holding resides in the municipality where the headquarters are located, even if the surfaces are all in another municipality or outside the national borders. This is also true concerning geolocalisation.

Even though land is measured at the reference date April 1 2016, the measurement covers the crop year is the 12 months period covering the whole cultivation period (from seed to harvest) of the main arable crops cultivated in Luxembourg. The cultivation period lasts from autumn 2015 to autumn 2016. It lasts from  September 1, T to August 31, T+1.

Labour characteristics refer to the period of 12 months preceding the reference date and rural development characteristics refer to the 3 years period from 2014 until 2016 included.

The reference date is fully in line with the administrative source data collection.

The printed questionnaires were available in the end of March 2016. SER managed the whole data collection. Surface and animal figures (except bovine livestock) proceed from the application forms for subsidies, which were sent to all agricultural holdings. Bovine livestock figures proceed from the identification and registration system SANITEL. Data on organic farming exist in the Organic farming register and data on rural development are provided by the Ministry of Agriculture MAFEA database (payments supporting rural development). These data were merged and made available to STATEC as a single source to minimise the statistical burden on agricultural holdings. SER covered the characteristics related to other gainful activities, labour force and agricultural production methods through a direct survey.

In the mid of April 2016, the agricultural holders received the SER direct survey questionnaire (reference date April 1, 2016) at the same time as the SER/IVV application forms for subsidies (reference date April 1, 2016). The web forms (provided by STATEC's IT department) were ready to use before the paper questionnaires were sent. The survey questionnaires were sent back by the agricultural holdings to IVV (winegrowers) or SER (all other agricultural holdings) during the period covering April until September 2016.

As a transitory solution, data entry of survey questionnaires was done by SER staff as from July 2016 onwards using the web form.

In autumn 2016, the administrative sources obtained through SER were combined with the survey data using an official common identifier to form a combined micro data set. Micro data validation was performed at STATEC to check the accuracy and plausibility of both the survey and the administrative data. Analysis was also performed at a national aggregate level with historical data. Moreover, SER assisted the data validation by providing feedback on the aggregated data at national level for the purpose of the production of economic accounts of agriculture.

While data entry was achieved in the end of 2016, the data validation process was only completed by the end of the year 2017, this delay being mainly caused by both the partial overhaul and late availability of Eurostat's transmission format for quite a few characterstics (notably OGA variables).

The final target population comprised 1 965 agricultural holdings for the reference year 2016 defined in conformity with the Regulation and was obtained from the database of agricultural holdings managed by SER.

Item non-response was addressed using cold-deck imputation, hot-deck imputation and, to a minor extent, ratio imputation as well as deductive imputation – the type of imputation applied depending on the nature of the variables to be imputed. Imputation was also extended to unit non-response for regulatory reasons. Meanwhile, the transmission format, i.e. the conversion program of national variables into the characteristics defined in the Regulation was updated in spring and autumn 2017.

In December 2017, the first batch of micro-data established in conformity with the farm structure characteristics (including rural development) defined in the Regulation was transmitted to Eurostat via the Edamis platform.

- STATEC as a coordinator to ensure the achievement of the data collection;

- SER as the data provider and the national producer of economic accounts of agriculture.

SER is responsible for all the variables. Nonetheless, the cooperation and coordination goes beyond the data collection. Both actors meet regularly every year.

The type of frame was a list with administrative source data.

The initial frame referred to March 2016. The time frame of the administrative source data collection was April 2016. The final updated frame for FSS 2016 was in October 2017. 

The administrative data sources used are the following:

- Integrated administration and control system (IACS);

- Bovine register (SANITEL);

- Database of the Ministry of agriculture on the payments made to farmers and other beneficiaries eligible for FEAGA or FEADER (MAFEA);

- Organic farming register.

IACS:

EU-legislation: Regulation (EU) 1306/2013 of the European Parliament and the Council on the financing, management and monitoring of the common agricultural policy;

National legislation: Règlement grand-ducal du 30 juillet 2015 portant application au Grand-Duché de Luxembourg de règles communes relatives aux paiements directs [...] et au soutien au développement rural.

Time reference: crop year for crop productions, 1st April 2016 for livestock.

Please note that the crop year is the 12 months period covering the whole cultivation period (from seed to harvest) of the main arable crops cultivated in Luxembourg. It lasts from  September 1, T to August 31, T+1. As the crop year goes over 2 civil years, by convention the year under which the information is stored in IACS is the year of harvest;

Updating of the source: annually.

SANITEL:

EU-legislation: Regulation (EC) 820/97;

National legislation: Règlement grand-ducal du 22 avril 1999 portant mesures d’application du règlement (CE) 820/97.

SANITEL is a permanently updated database. The bovine livestock can be derived from SANITEL at any date.

MAFEA:

EU-legislation: Regulation (EU) 1306/2013 of the European Parliament and the Council on the financing, management and monitoring of the common agricultural policy;

National legislation: Règlement grand-ducal du 30 juillet 2015 portant application, au Grand-Duché de Luxembourg de règles communes relatives aux paiements directs [...] et au soutien au développement rural;

This database is fully integrated with the IACS database.

Time reference: an extraction of the payments made to farmers can be derived from MAFEA. For FSS, it is checked whether there was a payment or not for at least one of the 3 preceding harvest years;

Updating of the source: permanent.

Organic farming register:

EU-legislation: Regulation (EC) 834/2007.

This register is permanently updated and is fully compatible with IACS (same identification of the units). The information on the status of the holdings concerning organic farming at a reference day (1st of April) is uploaded in IACS yearly. The register of agricultural holdings certified or under certification in organic farming is managed by the Administration des Services Techniques de l’Agriculture (ASTA), an administration of the Ministry of agriculture.

The statistical department of SER has the possibilities to specify its statistical needs with the SER team in charge of the administrative source.

SANITEL: The reporting unit is the holder of the animals. A permanent mutual updating between IACS and SANITEL concerning the list of reporting units (holding/holder) exists.

There are no differences in the definition of the characteristics between administrative source and FSS as the statistical division of SER is involved in the preparation of the data collection process from the beginning on.

SANITEL:

The characteristics requested in FSS are calculated/deducted from the characteristics registered in SANITEL per individual animal (date of birth, sex, age of first calving, date of slaughtering, etc).

Organic farming register:

A list of organic farms is extracted from this register. For the units within the scope of FSS, all the characteristics from other administrative sources and survey results are considered as “organic”. This is possible because in Luxembourg, a holding, if it seeks cofinancing through subsidies, can only be 100% organic or not at all. Hence, there are no holdings with partial organic production.

Organic farming register: This register also covers farms that don’t belong to the scope of FSS, so there is over-coverage. 

SANITEL covers all units holding bovine animals.

SANITEL: The reporting unit is the holder of bovine animals. A link with the reporting unit in IACS (agricultural holding) exists. There are regular exchanges of information and permanent mutual updating between IACS and SANITEL concerning the list of reporting units (holding/holder).

MAFEA: This register is permanently updated and is fully compatible with IACS (same identification of the units). See article 4 of Regulation (EU) 1307/2013.

Organic farming register: This register is permanently updated and is fully compatible with IACS (same identification of the units). See article 4 of Regulation (EU) 1307/2013.

Common identifiers: the same across all sources.  Linking was performed using the common identifier.

Electronic data for web questionnaires (16% of the target population).

- The survey questionnaire was sent by SER at the same time as the subsidy application package. The plan was to avoid agricultural holdings having to delve twice into their reporting systems and thus to accelerate the response.

- Reminders were managed by SER for the reference year 2016. In the end, the unit non-response was only 5.24%.

No special priority was given to important agricultural holdings.

In case of a census 1=3+4+5

Only for sample survey, in which case 2=3+4+5

This item is a subset of 3.

4>4.1+4.2

5=5.1+5.2

5.1>=5.1.1+5.1.2

6=5.2+5.1.2+4.2

As far as web forms are considered, tests were integrated in them to make impossible missing, inaccurate or implausible items as far as possible. The results of web forms were moved to an excel file and checked manually. In order not to render the 2016 coding too burdensome, these checks were performed ex-post, i.e. after the coding stage. Typically, the tests are written using statistics software syntax (e.g. SPSS).

As the survey FSS data did not include any surface and animal figures, arithmetic checks and ratio edits were not necessary when entering the data, but the administrative data (i.e. surfaces and animals) were checked for arithmetic, ratio and coherence issues at the integration with the survey data.

Same as above, the tests were written in statistical software.

The tools used for data validation were the Eurostat validation programme for micro-data implemented using SPSS.

Survey data and administrative data were handled by SER. The data integration as well as any data processing thereafter was performed by STATEC.

- measurement errors, in particular agricultural production methods;

- non-response (5%) and the related unit imputation errors;

- item imputation errors;

- asynchronous administrative sources (minor impact).

- inconsistent data concerning OGA and working time;

- difficulty to measure agricultural production methods, as the data collection is separate from the official grant forms.

Reasons for unit non-response

It appears that smaller agricultural holdings which receive less or no agro-environmental funding are more likely not to respond to the survey. Another explanatory factor are holdings who are classified as specialist horticulture and specialist permanent crops - these farm types respond relatively less often to the survey than other farm types. While agro-environmental payments were found to significantly explain non-response, taking into account holding size and farm type was also important to avoid any unwanted and uncontrolled side effects during imputation.

Treatment of unit non-response

First of all, unit non-response is minimised by 2 stage reminder policy, one early on after the launch of the survey (May / June of the reference year), another one upon reception of the first set of administrative data (somewhere in autumn of the reference year). In some minor cases, big agricultural holdings are contacted by telephone to encourage response. Once unit non-response remains confirmed, the fall-back strategy is imputation. Re-weighting is not an eligible strategy because of the census type survey.

The questionnaire design has been optimised since 2011 to minimise the item non-response and in the worst case to at least identify item non-response. For example, before asking about agricultural production methods, a yes/no question would capture whether or not there are any fields on which production methods could be applied. A non-response on this question or a yes-answer on this question without any subsequent answers would trigger item non-response analysis.

Reasons for item non-response

Item non-response by agricultural holdings is by far less common. Though agricultural production methods variables is typically burdened by such problems.

Another rare form of item non-response are responses invalidated by the statistical producer due to internal inconsistencies. Such responses would then be set to missing and imputed by the relevant strategy.

Treatment of item non-response

At the stage of data entry, we distinguished between two scenarios:

1) the questionnaire was insufficiently filled out for an agricultural holding in the target population: we contacted the agricultural holding either by phone or by mail to request the remaining information;

2) the questionnaire contained a few minor item non-responses. If the missing item was a structural characteristic (e.g. legal status, holding manager, etc.) and the prior year questionnaire was available, the information was directly manually imputed. This procedure was not performed for characteristics that are likely to change every year (e.g. labour force, other gainful activities, agricultural production methods, etc.).

After that, imputation procedures serve as fall-back strategy.

The strata were defined so that they would best explain non-response. Various combinations of variables available in administrative sources, provided by SER and thus for all agricultural holdings in the target population, were tested against unit non-response using the logit classifier.

Consequently, based on our analysis conducted in 2010 and reconfirmed in 2016, the strata defined on the basis of the following ancillary information were found to sufficiently explain unit non-response:

- UAA size class: less than 10 ha, at least 10 ha ;

- 1st digit of the farm type code in reference to the typology defined in Commission Regulation (EC) No 1242/2008 ;

- agro-environmental payments (rural development support): yes, no.

The strata were used both for item non-response and then for unit non-response.

The imputation strategies for item non-response

Item non-response was addressed before unit not response.

A few characteristics required deductive imputation in case of item non-response. These were mainly the characteristics related to the manager in case of group holdings and legal persons but also some national characteristics.

Random hot-deck imputation was performed by predefined strata. The procedure is used for any characteristics for which there is no prior year data and no deductive imputation strategy.

The imputation strategies for unit non-response

At the stage of data production, we used automated rules to impute:

Cold-deck imputation was used for labour force and related characteristics. The cold-deck contained fresh 2015 survey data. Data available in 2015 were then directly imputed without any other adjustment for a given agricultural holding. This imputation procedure concerned 4% of the target population in terms of number of units;

A few characteristics required deductive imputation in case of item non-response. These were mainly the characteristics related to the manager in case of group holdings and legal persons but also some national characteristics.

Random hot-deck imputation was performed by predefined strata. The procedure was used for all characteristics other than those imputed by cold-deck or deductive imputation.

The first version of micro-data were transmitted to Eurostat in December 2017.

For reasons of consistency with economic accounts of agriculture in Luxembourg, the definition of the Annual Work Unit (AWU) is, as in 2010, defined as follows: “a person is considered working full-time with an average of 8 hours a day during 275 days (2 200 hours a year). Persons with less than 15 years or with more than 80 years are excluded. Persons aged between 15 and 18 years as well as persons aged at least 65 years declared as full-time working in the survey have been transformed into part-time working using fixed coefficients, which is in line with national economic accounts of agriculture recommendations.

The 2013 NMR thresholds '1 000 laying hens or 1 000 other poultry' are replaced by '1000 poultry' in item 2.6-2 of the present 2016 NMR. Please note that even though the 2016 survey form specifies a 1 000 threshold for either laying hens and other poultry, the actual implementation is applied on administrative data with a single threshold of 1 000 poultry. Any missing survey data are dealt with appropriate imputation strategies.

The 2016 threshold of 10 breeding sows was not mentioned in the FSS 2013 questionnaire, however, during Eurostat FSS validation phase, it was added to the national implementation. Consequently, there is no change between the FSS microdata 2013 and the FSS microdata 2016 for this threshold.

The data are reconcilable with other data sources. However, due to multiple identifier types existing in the various sources, the micro-data linking through such identifiers is not always possible or flawless.

However, perfect reconciliation is possible with administrative sources obtained through SER due to the use of a common identifier.

http://www.statistiques.public.lu/stat/ReportFolders/ReportFolder.aspx?IF_Language=eng&MainTheme=4&FldrName=2&RFPath=7274

https://agriculture.public.lu/de/agrarstatistik/structures-agricoles.html

Metadata and publications:

http://www.statistiques.public.lu/en/methodology/methodes/enterprises/Agriculture/agriculture/index.html

https://agriculture.public.lu/de/agrarstatistik/structures-agricoles.html

https://agriculture.public.lu/de/agrarstatistik/structures-agricoles.html

- table design: very detailed tables are avoided (e.g. data published by municipality) unless the data is not confidential (e.g. number of agricultural holdings, total UAA, etc.). On a national level, almost any data (i.e. low occurrence variables) can be published. Problems arise with national data broken down by one or more spanning variables (farm type, size class, municipality, etc.). The table design usually results in a compromise between relevance and confidentiality;

- cell suppression: any cell that contains confidential data according to a sensitivity rule (n,k) or a minimum frequency rule are suppressed due to primary confidentiality. Any cell that is needed to protect one or more primary confidential cells are suppressed due to secondary confidentiality. In practice, suppression comes down to aggregation for one-dimensional tables and to flagged cells for multi-dimensional tables – linked tables are also accounted for. To guarantee the proper protection of tabular data, the exact parameters used to apply the above rules are kept confidential. Calculations are performed with tau-Argus and verified ex-post.