• 837/90 (concerning statistical information to be supplied by the member states on cereals production);
• 959/93, 2197/95, 296/2003 (concerning statistical information to be supplied by member states on crop products other than cereals);
• 93/16 (on statistical surveys of milk and milk products);
• 93/23 (on the statistical surveys to be carried out on pig production);
• 93/24 (on the statistical surveys to be carried out on bovine animal production);
• 93/25 (on the statistical surveys to be carried out on sheep and goats stocks).

Furthermore, some additional topics were surveyed by a specialized section of the questionnaire as shown in the following list:

• 1997 - fruit trees production;
• 1998 - environments;
• 1999 - rural development.

Since 2003 survey, an approach closer to local administrations’ purposes has been used; sample design has been determined considering accuracy on variables of local interest and some items were introduced in the questionnaire in order to take into account some local needs.

- the reference of the national legal base of the FSS survey (Act, Government Decree, etc.)

These categories of holdings have been also included in the survey:

• Agricultural holdings managed by non-profit and public entities;
• Agricultural holdings managed by industrial, good and services enterprises;
• Holdings with livestock only for reproductive goals, breeding of horses and poultry hatchery;
• Agricultural holdings without agricultural land (exclusively zoo-technical ones);
• Zoo-technical holdings which use pasture and meadows belonging to Municipalities and/or other public/private entities;
• Common lands;
• Holdings with NACE code rev. 2 number 01.61, which have the task of good status maintenance of land.

The thresholds which have been applied for 2013 DIFFER FROM THE ONES OF ART.3  ANNEX II OF THE REGULATION.

The national survey covered also holdings below 1 ha and other thresholds from Annex II of Regulation, thus providing a 100% coverage with respect to the 2013 reference list, which was built as described in point 2.5.

2.6.5. The number of holdings in the population covered by the records transferred to Eurostat, according to the EU definition of a holding and, if different from the EU definition of a holding, according to the national definition

Number of holdings in the sampling frame, according to the data sent to Eurostat and according to the whole data in the national coverage

Coverage of Eurostat data, in terms of number of holdings, UAA and LSU, according to the source data (data in the sampling frame)

We used the threshold of 1 hectare UAA.

More precisely the holding must meet at least one of the following conditions:

• UAA>=1 ha;
• open fresh vegetables>=0.50 ha;
• tobacco>=0.50 ha;
• covered fresh vegetables>=0.10 ha;
• covered flowers>=0.10 ha;
• cattles>=10 heads;
• pigs>=50 heads;
• breeding sows>=10 heads;
• sheeps>=20 heads;
• goats>=20 heads;
• poultry all>=1000 heads.

When the address of the HH lies within 5 km from that of the holder’ residence and it is in the same municipality, but the unit cannot be located through the information collected  by holding headquarter section of the questionnaire, then the address registered for the holder’ residence has been considered.

Please indicate the reference periods/dates of the main groups of national characteristics. (new) Please provide justifications if the reference periods/dates from the Regulation 1166/2008 are not respected.

The FSS provides a point-in-time stocktaking of legal status, management system,  total and utilized agricultural area of holding and livestock with the reference date  of 1 December 2013.
Information on land use, labour force and other gainful activities related to the farms refer to the agricultural marketing year (1st November 2012 to 31st October 2013).

Data on the professional status of the holder, his/her family and related, on the head of farm, agricultural skills of farm manager refer to the 12 months prior to 31st October 2013.
Characteristics concerning rural development support refer to the  three years (2011,2012,2013).

3.a The steps of the survey organisation and the starting and ending time of each step.
This information could help countries in the future planning of the activities.
As guidelines, the steps can consist of the following. Please adapt to the national situation if needed.

1. definition of survey objective and requirements:
  1.1. formation of workgroups for survey organisation;
  1.2. consultation of users;
  1.3. set-up objectives, target population, statistical units, classifications, precision requirements etc.;
  1.4. survey promotion.

2. survey design:
  2.1. set-up organisation of the survey (e.g. detailed timetable, specification of resources, costs estimation);
  2.2. definition of the survey variables;
  2.3. design of the sampling frame and sampling procedures;
  2.4. design of data collection procedures (e.g. questionnaire design, selection of data collection modes etc.);
  2.5. design of data processing procedures (e.g. CATI/CAPI/CAWI input programmes etc.);
  2.6. pilot survey organisation and execution.

3. data collection:
  3.1. sampling frame construction and sample selection;
  3.2. recruitment of interviewers;
  3.3. training of interviewers;
  3.4. fieldwork;
  3.5. evaluation and assessment of fieldwork.

4. data processing and validation:
  4.1.data entry and data coding:
  4.2. data validation (at record level);
  4.3. data correction and imputation.

5. data compilation:
  5.1. weight calculation and estimation;
  5.2. calculation of derived variables;
  5.3. calculation of quality indicators (e.g. non-response rates, relative standard errors, coverage errors, bias etc.);
  5.4. aggregation and tabulation;
  5.5. validation of aggregated data.

6. data analysis

7. data dissemination

Calendar of major FSS 2013 operations

In the preparatory phase a lot of meetings with the bodies involved (Regions) have been done, in order to discuss the organization and the content of the survey. 

Training for FSS 2013
The aim of the FSS training was to transfer methods and organization for FSS operation to all the network: Contents (FSS organization, Questionnaire, Definitions, Regulation, etc.), Methods (techniques of statistical data gathering), Role (FSS operators function).

Training beneficiaries were:
- territorial Responsible persons and Coordinators (Regions/Provinces);
- interviewers.

The developed idea was a “fall” training. It consisted in a training for trainers organized in two steps. The first one was a training section between Istat  and Territorial Responsible (Regions/Provinces); the second one was a training  between Territorial Responsibles and the interviewers.

The contents of the training were focussed on:

1. QUESTIONNAIRE

- contents and definitions;

- how to compile it.

2. INFORMATICAL SYSTEM FOR DATA AQUISITION AND MONITORING

- how to create and monitor the network;

- how to register data on the electronic questionnaire.

Training instruments: classic classroom session supported by slide and handbooks, and simulation to show how to use the informatic tools supporting the survey, surfing the net.

Help desk for the FSS 2013 network

A dedicated mailbox has been set up in order to send specific questions on the survey.

E-mail spa@istat.it

The Survey Management System  (SGR)

In order to support the survey network in conducting the various steps of the FSS survey, an information technology system has been implemented. More specifically, a dedicated application based on the use of web technologies has been set up, enabling data collection and the monitoring of the various data processing phases. The website set up ensures maximum data security during the data transmission and storage phases, in compliance with the National Statistical Institute’s standard rules. The management system can be seen as a distributed workflow system in which each operator can work independently, following a clearly defined procedure. This design moreover has had to provide for the management of recycling in production processes (delete of questionnaires, changes of status, reactivation of check, etc.) to prevent any problems from becoming such that Istat can only work on them manually. This operating procedure has produced benefits in terms of timeliness, data quality and costs.

The system includes over 50 functions grouped by type and organized into 5 macro-areas:
QUESTIONNAIRES – includes all functions strictly connected to the survey (recording of the interview, data entry);
OPERATORS – enables the survey network and user profiles to be defined and the units in the FSS 2013 list to be assigned to a specified enumerator;
SUMMARY REPORTS – includes a set of survey progress monitoring reports;
SUMMARY FORMS – includes all functions for primary variables data collection and a summary of the primary variables necessary for publication of provisional data;
UTILITIES – includes a set of network support functions spanning the entire survey process.

After 2010 Census a coverage survey was driven which estimated number and UAA of holdings.

The current frame is the one as described in 2.6.2.

Please describe the sampling design according to the following structure. This structure aims to increase the clarity and comparability of information between countries.

• coverage thresholds, 2 groups:

1. if the holding had a UAA greater or equal to 1 ha, or it reached at least one threshold as stated in Annex II of Regulation 1166/2008;

2. if the holding did not meet those thresholds, but only Census thresholds.

• regions (only NUTS2).

Each sub-domain so determined was then stratified, according to UAA size, Livestock equivalent units (LSU), and OTE classes (9). OTE is the Italian acronym for typology. In particular UAA and LSU were divided into 5 classes each (the last one being a take-all stratum), whose cutting thresholds were determined by means of R package "stratification".  This leads to the stratification in the annex.
Strata collapsing has been applied as a second step not to allow empty strata or too small strata, by joining adjacent categories of the categorised continuous variables.

Please note that the package "stratification" has been independently applied to each NUTS2 region.

Additionally, no NUTS3 level has been utilised because sample size would have been probably increased, since applying stratification by  UAA(5)xLSU(5)xOTE(9) in each of the more 100 NUTS3 level would have led to many small strata, and the overall dimension would have been increased.   Furthermore, a stratification by NUTS 3 level in order to get reliable estimates would have required to impose a set of expected CVs at NUTS 3 level similarly to Annex IV. Having Italy over than 100 NUTS 3, this work would let explode the overall sample and may be too costful.

The holding should have a UAA above a specific threshold or LSU above a certain threshold.

Thresholds have been determined by means of the package "stratification" (Kozak method).

• Package "Stratification" (to determine strata boundaries and take-all strata)
• Package "Sampling Strata" (Bethel algorithm to compute optimal allocation)

• Data collection (for data on rural development)
• Data control and correction (see 3.1.d.3, row on validation of survey data)
• Data imputation.

Specific sources are: 
1. the Integrated Administration and Control System - IACS^[1](AGEA: Agency for the Disbursement in Agriculture) – year 2013
The most significant and complete source is the IACS. The database managed by AGEA has been set up in accordance with the EC n. 885/2006 that, under the Common Agricultural Policy, acts in the coordination and execution of payments to support farmers. The IACS has been realized in order to record, verify and control data. The core of this system is made of files containing information on data that each agricultural holding is obliged to present for any aid application. In the database there are many available information; they can be divided into two main groups:
1) identification data of the farmer or the agricultural holding: Unique Code of Agricultural Holdings (CUAA code) that corresponds to the fiscal code of the holder. The CUAA code is mandatory whenever a relation with the Public Administration is undertaken. Holder’s name, permanent address or place of residence, VAT number if present. Dates of inscription and updating.
2) territorial data: agricultural parcels of the holding; information on the use of each parcel (crops, livestock); hectares invested by type of product (cadastral area and agricultural area utilized for farming).
If the holder is not the owner of the parcels, it is recorded the identification code of landholders and the type of contract that links farmer to landholder.

2. the System for the Identification and Registration of Bovine Animals and other species (AA.ZZ.) – year 2013,
The System for the Identification and Registration of Bovine Animals and other species are registers in an archive (AA.ZZ) managed by the Ministry of Health. Recorded units concern animals and their holders with the scope to preserve public welfare. The covered animal species are bovines, pigs, sheep and goats, poultry, equines. The national database of Bovine Animals realized in accordance with the EC n. 1760/2000 for the setting up of a System for the Identification and Registration of Bovine Animals, is the only fully operative register up to date.
[1] COUNCIL REGULATION (EC) No 1782/2003 of 29 September 2003 establishing common rules for direct support schemes under the common agricultural policy and establishing certain support schemes for farmers.

GINO has simplified the questionnaire compilation by:

•        automatic calculating arithmetical operations;

•        reporting errors in data input, displaying the appropriate message;

•        displaying additional messages on mouse-over of words or phrases requiring a brief explanation.

In addition, to simplify the compilation of the questionnaires still further the system has allowed various steps in processing the questionnaire: saving it as a draft, which enables users to enter data without worrying about their accuracy, or saving a final version which entails the activation of control rules, and final sending. Obviously the holder might choose to skip these partial saves and send the questionnaire directly.

The data of this table refers to the coverage for which data is sent to Eurostat.

From the 2010 Census list, which recorded 1 620 884 holdings, by applying the minimal thresholds according to Regulation 1166/2008, only 1 138 218 holdings were fullfilling the requirements, thus ensuring at frame level a coverage of 97,93% for UAA and 99,60% for LSU. After the survey the number of holdings in the population covered by the records transferred to Eurostat is 1 010 328.

42 530.

Please note that in case of split of holdings and/or merging of holdings between the sample this sum will  never be equal. So it is not 2=3+4+5

There were at the beginning 42 151 sampled units, and we observed 379 new holdings (348 respondent, 10 were just starting their business so they were temporary inactive during the reference year, 20 had a UAA below the thresholds, 1 could not be found because the enumerator couldn' t get the address)

The equality follows:

42 151=37 807-348+749+2418+282+1 243

a. fell below set thresholds during the reference period: 769 (20 of them were new holdings from splits, 76 were of those were not even more agricultural holdings but solely for mushrooms and forestry )

b. have the activities ceased during the reference period:  2 428 (10 were new holdings that were temporary inactive)

c. do not exist at the selected address (rather holdings with unknown eligibility status, because the address could be wrong or missing but we cannot be sure if the unit exists and is eligible or not)  283 (1 was a new holding for which the enumerator couldn' t get the address)

Number of holdings with ceased activities who sold their land to pre-existing holdings: 1 208

Assuming your assumptions 0.

In case of a refusal it may be that the holding is not eligible, at least with unknown eligibility status.

where 348 are new activities, following split or merging

1. Use of administrative sources (for example Bovine animals Register) for micro and macro data checks;
2. Use of Forward Search methods for the outliers detection;
3. Use of score functions to prioritize records to be manually reviewed in order to identify and treat potentially influential errors;
4. Adoption of techniques that minimize the number of changes, particularly for the treatment of not influential random errors;
5. Mix of different imputation methods as nearest neighbour approach or model based imputation.

The strategy adopted is the outcome of different simulation studies, carried out for identifying the most suitable methodologies to solve problems of missing, invalid or inconsistent values.

E&I during the data gathering phase
In order to improve data quality during the data collection phase a controlled data entry system was developed (the electronic questionnaire GINO). In order to prevent and correct fatal errors and missing values, especially for the items related to farms identification and localization, a set of checking rules has been implemented in the data entry system. The system analyzed at unit level (micro-editing check) the coherence between the answers referring to related topics with the aim to find missing or inconsistent values.
Two types of edit rules have been used:

1. Fatal edit rules, to underline errors and force the respondent or the interviewer to restore data correctness;
2. Query edit rules, to highlight the need to do further investigation on the information gathered.
   
   

E&I to release the final data
For the final figures the E&I process has been structured in different sequential steps.
In order to detect errors, during this stage, a list of more than 1000 consistent and not redundant edit rules has been used. These edit rules, referring to single variables or to the expected relations among them, correspond to logical and/or mathematical constraints that must be satisfied simultaneously by the values of an individual record (within-record edit rules), or by the aggregation of different records included in the same subset of analysis (between-records edit rules).
The complexity of the constraints has been managed by dividing the variables into subsets that are treated in different E&I steps.
In each E&I stage, every single step has been separately processed when the subsets of variables were unconnected (the edit rules didn’t affect variables belonging to different subsets) so that the order of the processing runs was not of influence. Otherwise, when the subsets of variables were connected (there were edit rules defined on variables of different related subsets), during the performance of a single procedure, all the variables imputed in the previous runs were maintained fixed.

The following steps have been repeated:

Automatic error detection:

(a) Micro-editing: error detection at unit level, according to the whole set of checking rules.
For each responding unit, the minimum number of values to be modified, for restoring the situation of correctness has been identified. The detection of errors has been performed conditionally to the path followed in the compilation of the questionnaire (relative minimum change), according to the answers given by the units without errors (donors) closest to the wrong unit. This phase has been performed jointly for the qualitative and quantitative variables.

(b) Macro-editing: data validation, by analysing the aggregates describing the structure of agriculture system (e.g. the total number of farms collected, the Utilised Agricultural Area, area invested in the major crops, etc.).

Design weights were corrected for unit-non response by multiplying them with the inverse of the response rate within each strata. When needed, if the response rate was lower 2/3 or 70%, strata were collapsed and then design weights were recomputed as the inverse of inclusion probabilities, then multiplied with the new inverse of the responding rate.

Response homogeneity groups were considered as coincident with strata (or collapsed ones).

For non-significant or non-existent characteristics, you may repeat the information sent to Eurostat according to art. 7 par. 3 of Regulation 1166/2008. You can also attach the relevant file to this section using the "Add file" button below.
The overall answer to this item should provide information on:
-the list of characteristics non-significant and the list of characteristics non-existent from the EU list of characteristics [1];
-the reasons i.e. the prevalence or physical thresholds;
-the source(s) of information used (for the prevalence or physical thresholds);
- (new) how are non-significant or non-existent characteristics marked in the dataset transmitted to Eurostat.
(new) In addition, please specify whether non-significant characteristics are reported under the headings of other characteristics (as in the case of some countries). If yes, please specify which those other characteristics are and please indicate if the Standard Output of those other characteristics is recalculated considering the inclusion of the non-significant characteristics.

The following characteristics have not been included in the FSS because they are non-existent (NE) in Italy:
- energy crops of which on set-aside area;
- genetically modified crops;
- raisins.
In fact,
- The Regulation (EC) 1782/2003 under which energy crops of which on set-aside should be provided, has been abolished.
- the national legislation (D. Lgs. n. 224/03 and D. Lgs. n. 70/05) allows the GMO cultivation in the national territory only for research purposes and the Regulation (CE) n. 73/2009 repealing the Regulation (CE) n.1782/2003  stated that energy crops on set-aside are not more existing in Italy.

Non-significant characteristics have been collected under their own headings.

See attached file Non-existent and not significant characteristics in FSS 2013.

[1] See Annex III of Regulation (EC) 1166/2008 of the European Parliament and of the Council on farm structure surveys and the survey on agricultural production methods and repealing Council Regulation (EEC) 571/88.

Please provide a brief general assessment on the main sources of error (e.g. sampling errors, measurement errors etc.)

Since the FSS 2013 is a sample survey, data are affected by the sampling error, especially for minor variables for which the precision were not established in defining the sample size (since not included in Annex IV of Reg. 1166/2008).

Concerning non-sampling errors (arised during the planning, conducting, data processing and final estimation stages), Istat carried out a sample survey using a CATI technique aimed to evaluate the quality of the data collected. The sample survey is based on replicated measurement on the same units interviewed by 2013 FSS. See item 6.3. for further discussion.

Please describe the method used for estimation of RSEs. You can annex a document with the description of method and formulae applied, using the "Add file" button.

The estimates and the corresponding estimated standard errors are derived using the theory of stratified random sampling.
Domain totals and the associated estimated standard errors are derived using standard formula for domain estimation as reported in Sarndal et al (1992), Section 10.3.

More specifically as weights were corrected only for non-answer, and ceased activities as well as out of target holdings were considered as respondents, they were considered as observed 0s for the computation of RSE. This leads to higher values of RSE in comparison to the ones computed on the unique set of respondents and active holdings.

The cases where precision requirements are applicable are identified with the information provided in section 6.2.a. For those cases, the requirement is that the estimated RSEs are below the thresholds stipulated in Annex IV "Precision Requirements" of the Regulation 1166/2008. However, in some of these cases, estimated RSEs might be above the thresholds. In the latter cases, please provide justifications.

Concerning livestock, most of the reasons arise from the fact that we stratified according to LU, so strata which were uniform or close in terms of LU, after years didn' t show the same characteristics, as a result, there turned out to be more variability, than expected.

This means it has to do with the use of a non-updated list, and this effect becomes stronger with poultry and pigs (the number is more variable by nature).

Concerning pasture excluding rough grazing, the sample was built to achieve precision on pasture all. For this reason we also computed RSE for total pasture.

The remaining cases which are exceeding 5% (olives in ITI2 or ITF3 or oilseed crops in ITF2) show most relevant differences in totals as well, thus achieving a higher variability in terms of actuals RSEs, as compared to theoretical CVs, but always keep values from 5% to 7/8% which still represent very good estimates.

Concerning some discrepancies on the identification of non-compliance:  initial weights were corrected only for non-answer by direct factors, while ceased activities as well as out of target holdings were considered as respondents, this implies that they were considered as observed 0s for the computation of RSEs. This leads to higher values of RSE in comparison to the ones computed on the  set of respondents and active holdings. For this reason, for instance, cereals in ITI2 in our calculation is 5.26% while in EU procedure is 4.96%.

If comparison with another source or consistency study is made, please give a brief description of the source used and the differences observed which can be proof of bias.
(new) Please also consider here bias risks associated with non-response by assessing the distribution of non-response across holdings' categories.

In a sample survey, such FSS 2013, two main kinds of errors could be encountered: sampling errors, and measurement (or response) error. The non-sampling error is a function of many factors: organisational aspects of the survey, the behaviour of a plurality of individuals or institutions.
A post-FSS surveys, in particular a re-interview survey on a sub-sample of units have been done, in order to certify the statistical quality of the information collected by the FSS 2013 (see item 6.3.2.c). Main results are summarised in the file annexed to this item.

Please provide the value of the over-coverage rate.
The over-coverage rate is the proportion of units accessible via the frame which do not belong to the target population (e.g. holdings with ceased activities still included in the frame).

The Re-interview Survey
This survey was carried out thorough a re-interview of a sample of agricultural holdings already interviewed by the FSS 2013 survey, in order to measure the response error due to the respondent and/or enumerator. Among the different possible techniques, Istat adopted the re-interview of a sample of the FSS 2013 survey agricultural holdings, collecting information on a limited number of variables already collected within the FSS 2013 questionnaire, through a short questionnaire. The procedure consisted on the comparison between the two answers given by the same unit on the selected questions. In all cases of differences between the two responses, the reconciliation procedure was applied. In this procedure, the respondent (the holder) has to choose the right one, between the two different response given to the FSS and to the re-interview survey.
The survey was carried out on a sample of about 3 000 holdings selected with a one-stage stratified sampling from the FSS 2013 frame. The territorial domain was the national one. The survey was conducted in July 2014. The re-interview was carried out, through a Computer Assisted Telephone Interviewing (CATI), by an external specialized company.
Istat transmitted to the specialized company the sample of the holdings to be re-interviewed (5 300 units in total – 3 000 for the basic sample and 2 300 for the substitutions) and together they planned the management software of the survey. Istat trained the telephone interviewers and also prepared (and sent) the informative letters to the holders in order to inform them about the survey. The variables selected from FSS 2013 questionnaire and used for the re-interview telephone questionnaire were:

- Total Arable land
- Total permanent crops
- Permanent grassland, Pastures and meadows; 
- Utilized Agricultural Area (UAA),
- Total holding land;
- Livestock (bovine, buffalo, sheep, goat, pigs)
- Labour force (workers –family and non- family).

The interviewer did not know the response of the FSS 2013 until to the moment in which the provided value to the telephone survey is different to the FSS 2013 value. At this moment, the FSS 2013 value appeared on the monitor (in fact the value shown did not coincide in the established range) and the interviewer asked to the respondent to reconcile the data.
The reconciled value represents the “true value” and it allows the use of models to estimate the response error.
The controls included in the questionnaire permitted the preservation of specific constraints between the aggregate variables (for example UAA) and their components (Arable land; Permanent Crops, Kitchen gardens; Permanent grassland, Pastures and meadows).
Ultimately, 3 000 complete interviews (eligible units) have been conducted by the external company and all strata are represented. There were non cases of item non-response.

See table as annex in 6.3 for relative % estimated bias.

6.3.3.a (new) Unit non-response: reasons and treatment
Please specify the reasons for unit non-response and  how the unit non-response was accounted for. Unit non-response can be accounted for by e.g. re-weighting, imputation.

We have considered different causes for non-responses:

1- absence of the holder (and of anyone else be able to answer to the interview);

2- wrong address;

3- refusal;

4- other (illness, judicial measures, etc.).

The unit non-responses have been corrected by re-weighting (according to the strata they belong to).

6.3.3.b Item non-response: reasons and treatment
Please mention any characteristic(s) having higher item non-response rate together with the reasons of the item non-response. This information is important and will be useful for the organisation of future surveys.
Please also specify how the item non-response was accounted for. Item non-response can be accounted for by e.g. re-weighting, imputation.

The percentage of non-responding holdings is 3.2%, as the ratio of non respondent units out of elegible units (according to Eurostat definition).

Please provide the ratio of the in-scope (eligible) units which have not responded to a particular item (characteristic) to the in-scope (eligible) units that are required to respond to that particular item (characteristic). Please provide this rate for characteristics with high item non-response.

6.3.4.b Completion/correction methods applied
These can consist of follow-up interviews, imputation, re-weighting, use of other data sources etc. Please describe.

The non-sampling errors have been identified and treated by an Editing and Imputation System (E&IS), preserving as much as possible the collected information.
The Editing and Imputation System identified outliers for the primary variables. After the manual review the remaining errors in the primary variables have been treated at central level by automated procedures managed by Istat. To solve these inconsistencies, the same E&I methods have been applied for the treatment of the secondary variables.

The treatment of errors was:
(a) Detected outliers and influential errors have been manually reviewed by experienced staff. In particular, the most relevant units have been re-contacted or clerically edited, by using score functions to prioritize micro data review in selective editing. For this purpose, the available auxiliary information from statistical or administrative registers has been used.
(b) Random errors, having a lower impact on data dissemination have been treated by automatic methods.
The treatment of not influential random errors has been based on the minimum change approach. Once the minimum number of values to be modified for restoring the situation of correctness has been identified at unit level, the most appropriate imputation method has been adopted for each subset of related variables (see next item).

As a whole, the imputation process has been a combination of the following methodologies:
(a)   deductive imputation, if the values to impute are uniquely determined by the values assumed by other variables;
(b)   rule based imputation (based on deterministic rules like "if-then");
(c)   nearest neighbour imputation;
(d)   model based imputation (preferred for the imputation of the continuous variables);
(e)   interactive imputation.

Particularly, the imputation of missing non linearly dependent data concerning the different types of crops has been performed through conditional Copula functions in continuous variables. This procedure is the result of several tests conducted by Istat, in cooperation with researchers from the University of Bologna (Bianchi et al., 2009; Di Lascio and Giannerini, 2012; Nelsen, 2006; Sklar, 1959). This new approach has allowed to preserve the variables distribution, as the missing values imputed have been randomly chosen from the conditional distribution of missing values, given the observed values.

The main features of these models are:

1. the first one is a general ILP model for data editing and imputation with minimum change (minimum number of variables to change) and minimum difference between the original value and the imputed value;
2. the other one is a specialised model focused on balance-edits (constraints which checks that a total equals the sum of its parts).

In order to treat the continuous variables, mostly for the labour force section, the software DIESIS has been used, both for the detection and the imputation of the inconsistent values.
DIESIS system (Bruni et al., 2001) is a software developed in C ++ language, and was implemented to treat the demographic variables in 2001 Population and Housing Census. In DIESIS, the detection of errors is based on the Integer Linear Programming, solved by applying Branch and Cut methods (Bruni et al., 2001). The DIESIS system allows to deal both with qualitative and quantitative variables simultaneously. The ‘first donors then fields' algorithm has been used for the treatment of the labour force variables.
For the interactive imputation, a special Java web application (CORRINT) has been developed. For each section of the questionnaire, different users interfaces have been implemented to insert the new value for the single item to be corrected. The same edit rules implemented in the data collection system have been integrated in the module, thus combining data entry and data editing. During this stage, if the new value failed an edit rule, an alert appeared describing the failed rule, in order to find the best solution to restore the coherence at unit level. The process ended when all edit rules were satisfied and the questionnaire was saved.

For FSS preliminary data are not published. Only final estimates are disseminated.

The quality of the preliminary results have been quite high as confirmed by the discrepancies with the final data shown in the file attached Comparison between preliminary and final data.

The data transmission to Eurostat has taken place after 5 months after the deadline forecasted by Regulation (EC) 1166/2008. Concerning the national release, the delay respect to the scheduled data  was of 4 months. As a consequence also the pre-established schedule for national release was delayed.

1. holdings over 1 ha UAA threshold and over the thresholds of other physical characteristics than UAA in Annex II of Regulation 1166/2008;
2. holdings under the above-mentioned thresholds and above the 2010 thresholds,

the coverage of the records sent to Eurostat covers only 1.

The following table refers to 2013 estimates in both cases.

• Legal status  (Institution or Municipality that runs common land)
• Management system  (Other form of management)
• Land ownership  (ownership)
• Land use – Section III
• Labour force - Section IX

Section V on livestock has not been compiled because the number of heads will be counted in the questionnaires of the farmers using the not allotted common land.  
The agricultural holdings using not allotted common land have not to declare the area on land ownership and on land use but they have filled in the following part of the questionnaire

• Livestock – section III

SECOND CASE - Common land allotted
The area of the common land allotted is recorded by each beneficiary farm therefore is not anymore considered as common land but as land of the beneficiary farm. In the land ownership  this area has been indicated as rented or in free use in accordance with the kind of formal agreement between the Institution/Municipality and the farm. The “free use” is a particular form of contract were the holder get to use the land in "loan for use "but the land remains in the property of the owner.

For this case no information was collected, which means we weren' t able to identify these holdings.

The reason beyond that is that we don' t ask, when there is free use, any information about whom is giving the free use.

So the records which are identified as common land are only of the first type

CHOICE OF TENURE:

As a preliminary rule to assign the tenure, we ask the holding what kind of management system they adopt:

a. direct management by farmer

b. management with salaried workers

c. other kind of management system (which may be mainly common land, public municipalities) where the enumerator was supposed to specify which other form was

If case is c. then the tenure is classified as shared farming.   In other cases a. and b. we classify as owner farming, land which may be of his own property or given as free use.

Tenant farming (case c) is about land which is rented.

Classification of tenure for common land follows the same rule.

(a) a public administration or public body (Municipalities or Commons) whose area is common land;

(b) a public administration or public body.

Maybe the question is not well posed, and there have been some cases in which even b) should have been classified as common land.

In some cases for instance we observed b) (which is indeed a correct answer ) and common land use was specified when answering to management system question.

Unfortunately only case a) was classified as common land.

In the holder residence section of the questionnaire (pre-printed with information derived from Census, eventually corrected by the holder in case of errors or omissions) the information considered (among others) for AH location purpose has been: address, municipality name and Istat code, province name and Istat code.
In the holding headquarter (HH) section the information collected refers to address and to cadastral polygon (map sheet or parcel) in which the HH falls. For information on cadastre, in particular: cadastral section, cadastral map sheet and cadastral parcel, depending on kind of cadastre used in each specific territorial area, were collected. None of these information were pre-printed, as it is the case for the location of the holder’ residence. The information referring to the distance between the localization of the holding headquarter and the holder residence (whether it falls within 5 km - in a straight line - or not) was collected also in this questionnaire section. It has to be underlined that the holding headquarter section had to be filled in by farmer only if the HH location was different from the holder residence.
Depending on which information was collected, two main different approaches have been identified to locate HH: the address and/or the cadastral map.
In the data process flow, the address location was considered the most accurate information so, for location purpose, it was treated at first, while geocoding through cadastral map was considered when the first option was not feasible. 
Another option was to consider the information on the distance between the holding headquarter and the holder’s residence. When the address of the HH lies within 5 km from that of the holder’ residence and it is in the same municipality, but the unit cannot be located through the information collected with the information collected by holding headquarter, then the address registered for the holder’ residence has been processed.
The procedures adopted to couple information collected through the questionnaire and the enumeration areas were different.
Addresses were coupled with enumeration areas in two different ways:
i) through address processing with a commercial software, named Egon that, among other things, normalize the street or road description followed by geo-referencing and geo-coding;
ii) through deterministic matching procedure with two national street archives derived from two specific and surveys (reference year 2010) realized for the preparation of the population census.
The cadastral information declared in the questionnaire are matched, instead, with geocoded cadastral maps that have been in advance processed and matched with enumeration areas through a spatial join.In all cases it is guaranteed that the holding is in the same/correct municipality. Municipalities are "comuni" in the Italian language - which correspond to LAU2.
For more details on the application of the two different approaches, please refere to Census NMR.
In case AHs were located in more than one municipality, only the information on land use and livestock rising characteristics related to the HH’s municipality was considered for string assignation.
Strings used as linkage key have been defined as follows:
A) the ‘long string’ was based, for land use on: arable land, vineyard, permanent crops, kitchen gardens, permanent grassland, wooden permanent crops, farm forest, other land uses. These variables have been classified into new different variables whose values have been defined by the classes values. In particular, for permanent grassland and for farm forest (wooded area other than short rotation coppices) the classes are: equal to zero (value equal to 0), higher than zero and lower than 500 ares (value equal to 1), higher/equal than 500 and lower/equal 2000 ares (value equal to 2) and higher than 2000 (value equal to 3), and for kitchen gardens only presence (value equal to 1) or not (value equal to 0) has been considered. Arable land, vineyard, permanent crops, wooded area (short rotation coppices), and other land uses values are classified into 4 classes defined as: equal to zero (value equal to zero), higher than zero and lower than 200 ares (value equal to 1), higher/equal than 200 and lower/equal 900 ares (value equal to 2), higher than 900 ares (value equal to 3). For livestock, based on: bovines, buffaloes, pigs, sheep-goats, poultry and other livestock, only presence or not has been considered, for each specific species;
B) similar procedure has been adopted for the construction of the “medium string”. In this case, for land use: a specific digit refers to presence or absence of each specific macro-category; for livestock: a specific digit refers to presence or absence of at least one of the raised species;
C) in the short string: two digits define the presence - or not - of i) at least one kind of land use mentioned above and ii) at least one kind of livestock species.
The procedure separates donors (geo-referenced AH) from receiving units (non-geo-referenced AH), thus the latter were linked to the previous ones taking into account the municipality code and the AH string value described above. The linkage procedure considered first the long string than the following ones.
Ones the linkage was established, since the receiving unit could have more than one donor units, located in different census enumeration, it has been chosen as donor the AH that minimizes the sum of the absolute difference between the true specific value of the single variable used for the strings construction. Then the census enumeration area centroide coordinates in which the specific AH chosen is located were transferred from the donor unit to the receiving unit.

While in 2010, the survey covered both:

•  holdings over 1 ha UAA threshold and over the thresholds of other physical characteristics than UAA in Annex II of Regulation 1166/2008;

•  holdings under the above-mentioned thresholds and above the 2010 thresholds,

in 2013, the survey covered only the holdings under the first case.

For a reasonable interval of what the estimates should be see 8.1.d.3 and 8.1.d.4.

Please indicate whether the FSS statistics are reconcilable (i.e. can be combined) with those obtained through other data sources or statistical domains.

Other data sources can be for example administrative data, crop production surveys, animal surveys, labour force surveys, National Accounts.
If you run comparisons, please give a brief description of the results of these comparisons and possible adjustment made to FSS data. If not, please indicate why not.

8.3.a Comparisons at micro level

8.3.b Comparisons at macro level

The whole process of the Editing and Imputation System (E&IS) has been monitored by the analysis both of the data distributions and of the performances of the scheduled editing steps. By computing a set of tables, at regional level, the final results of the procedures have been compared with the available statistical and administrative sources.

For the administrative sources used for the evaluation of the results, see item 3.1.d.1.

The following statistical sources have been used for the evaluation of the results:
- 2010 Census
- Animal surveys
The comparison  with other sources (administrative and statistical) shows a general coherence of the results.

Here a selection of the comparison for some relevant items:

Data dissemination of FSS 2013 survey has been carried on  by two channels:

- a digital publication on the Istat website (Farm structure survey) containing some tables and comments on  the principal phenomenon

- tables on the web site, concerning the main structural characteristics of the Italian agriculture at NUTS 2 level (25 tables on: general characteristics, land use, irrigation, livestock, employment, time series)

Concerning Regular and ad-hoc publications in which data are made available to the public:

The publication Farm structure survey has been published on 5^th September 2015 on the web and available for the users.
The tables have been published in September 2015.

Data released to Eurostat on May 2015 and on November 2015 (Rural development) have not been published.

Farm structure survey: http://www.istat.it/en/archive/167694

Tables: agri.istat.it

Tables concerning the main structural characteristics of the Italian agriculture at NUTS 2 level have been published on the web site (25 tables on: general characteristics, land use, irrigation, livestock, employment, time series) at the address: agri.istat.it

• Atkinson A. C., Riani M. (2000), "Robust Diagnostic Regression Analysis", Springer, NY.
• Bianchi G., Di Lascio F.M.L., Giannerini S., Manzari A., Reale A., Ruocco G. (2009), "Exploring copulas for the imputation of missing nonlinearly dependent data". Proceedings of the VII Meeting Classification and Data Analysis Group of the Italian Statistical Society (Cladag), Editors: Salvatore Ingrassia and Roberto Rocci, Cleup, p. 429-432. ISBN: 978-88-6129-406-6.
• Bruni R., Reale A., Torelli R. (2001), "Optimization Techniques for Edit Validation and data Imputation", Proceedings of the Statistics Canada Symposium 2001/ “Achieving Data Quality in Statistical Agency: a Methodological Perspective” /XVIIIth International Symposium on Methodological Issues.
• Di Lascio F.M.L., Giannerini S. (2012), "A Copula–Based Algorithm for Discovering Patterns of Dependent Observations". Journal of Classification, 29(1), p. 50-75.
• Luzi et al. (2007). EDIMBUS. "Recommended Practices for Editing and Imputation in Cross-Sectional Business Surveys", August 2007.
• Nelsen R.B. (2006), "Introduction to Copulas (2nd edn)". Springer.
• Riani M., Perrotta D., Torti F. (2012), FSDA: "A MATLAB toolbox for robust analysis and interactive data exploration", Chemometrics and Intelligent Laboratory Systems, in press doi 10.1016/j.chemolab.2012.03.07 .
• Sklar A. (1959), “Fonctions de repartition a n dimensions et leures marges”, Publications de l’Institut de Statistique de L’Universitè de Paris, 8, 229–231.

Please indicate if there is any co-ordination between surveys to avoid the situation that some farms have to answer multiple questionnaires with the same kind of questions.

Please mention if micro-data are also disseminated and if yes, the confidentiality provisions that are applied.

Procedures can include controlled rounding, cell suppression, aggregation of disclosive information, aggregation rules on aggregated confidential data, primary confidentiality with regard to single data values etc. Main reference: Handbook on Statistical Disclosure Control (2007).

We published data at aggregated level (regional) and we did not meet cases with problems of confidentiality.