6. Accuracy and reliability |
Top |
|
|
6.1. Accuracy - overall |
Main sources of error |
The methodology of the survey, thorough review of the data and comparison with administrative data, previous surveys and other sources guarantee sufficient precision of key indicators as well as compliance with the EC 1166/2008 requirements. However, measurement errors may occur in the case of some indicators that are not verifiable by administrative and other sources (see 6.3.2 Measurement error manure export). |
|
6.2. Sampling error |
Method used for estimation of relative standard errors (RSEs) |
Information about the method used for the estimation of relative errors is presented in the annex. |
Annexes: 6.2. Method for estimation RSE |
6.2.1. Sampling error - indicators |
1. Relative standard errors (RSEs) - in annex
2. Reasons for possible cases where precision requirements are applicable and estimated RSEs are above the thresholds |
RSEs are below thresholds. |
Annexes: 6.2.1-1 Relative standard errors |
6.3. Non-sampling error |
See below |
6.3.1. Coverage error |
1. Under-coverage errors |
Frame under-coverage could be considered as non-existent as the frame has been updated on the basis of all available sources. |
2. Over-coverage errors |
These units were included in the frame but did not belong to the target population. Mostly these units had finished their agricultural activity or their agricultural activity had decreased below the threshold. This was detected during data collection and therefore does not cause errors in the survey results. To correct over-coverage and prevent over-estimation of the number of holdings and other characteristics, recomputing of the design weights for the sampled units (by considering a corrected population) is undertaken. |
2.1 Multiple listings |
Multiple listings may occur when several persons from one holding exist in the frame (have registered themselves in administrative registers). They are discovered during the survey and the questionnaire is completed only once for the whole holding. They are treated as errors in the population and sample; thus, when the information is received, the design weights are changed accordingly. |
3. Misclassification errors |
All units have been classified on the basis of the newest available information. |
4. Contact errors |
Contact data have been updated on the basis of all available sources. |
5. Other relevant information, if any |
Not available. |
|
6.3.1.1. Over-coverage - rate |
Over-coverage - rate |
The frame over-coverage was 330 units (1.9% of the frame). 330 is the estimated over-coverage in the population, taking into account that the population was 17,026 units and the total number of holdings according to the final results was 16,696. Therefore, 17,026 – 16,696 = 330 units. |
|
6.3.1.2. Common units - proportion |
[Not requested] |
6.3.2. Measurement error |
Characteristics that caused high measurement errors |
The most difficult question was M_6_3$T Total produced manure exported from the holding. The reason is that holders do not have information about these quantities, and if they try to give an estimate, these estimations may not be accurate. The data were checked using the livestock coefficients of produced manure per livestock head (developed by the Estonian University of Life Sciences). It was assumed that it is not possible to export more manure than is produced. In total, data of 15% of holdings who have exported manure were corrected. Information about other major measurement errors is not known. The questionnaires were carefully designed, they were in electronic format and included several arithmetical and logical checks. A part of the information was collected by trained interviewers. |
|
6.3.3. Non response error |
1. Unit non-response: reasons, analysis and treatment |
There were a few holdings (with unknown eligibility status) that did not complete the questionnaires. Reweighting was used to compensate for non-response. Unit non-response was small and there were no strata with much larger non-response than the average. Therefore, the results of FSS 2016 can be regarded as good. |
2. Item non-response: characteristics, reasons and treatment |
In general, the design of the questionnaire did not enable item non-responses. Still, in the case of cattle, sheep and goats, the questionnaires were prefilled with data of 1 July and respondents were informed that there is no need to correct them if the data in the register is correct and they accept that we take the data of 1 September (reference date) directly from the register. 43 holdings had in their questionnaires only prefilled administrative data and their all other data can be regarded as item non-responses. In case of item non-responses, the data of the Farm Structure Survey 2013 or Agricultural Census 2010 were used for imputation. If these data were missing, the hot-deck imputation (nearest neighbour) method was used for imputation. |
|
6.3.3.1. Unit non-response - rate |
Unit non-response - rate |
During data collection, no information was received for 97 units (1.5%) with unknown eligibility status. Reweighting was used to compensate for that, as, according to the IACS register, these units were active. There were no holdings within the totally surveyed stratum for which data were missing (including utilised agricultural area and livestock data from administrative sources). There were cases when the holding accepted the administrative data but did not give further information, but this is not unit non-response and has been treated as item non-response. Therefore, unit non-response within the eligible units is 0. |
|
6.3.3.2. Item non-response - rate |
Item non-response - rate |
In general, the design of the questionnaire did not enable item non-responses. Still, in the case of cattle, sheep and goats, the questionnaires were prefilled with data of 1 July and respondents were informed that there was no need to correct them if the data in the register is correct and they accept that we take the data of 1 September (reference date) directly from the register. The rates of respondents who chose this possibility were the following: 93% of respondents with cattle, 94% of respondents with sheep and 90% of respondents with goats. 43 holdings had in their questionnaires only prefilled administrative data and their all other data can be regarded as item non-responses. So, the highest item non-response rates were 0.7% (for the data of manager, for example). Some other examples: 0.3% for full time regular male and female employees, 0.6% for crop rotation and up to 0.5% for soil cover. |
|
6.3.4. Processing error |
1. Imputation methods |
If possible, data from the Farm Structure Survey 2013 or Agricultural Census 2010 were used. If these data were missing, the hot-deck imputation (nearest neighbour) method was used for imputation. Administrative data were also used, if available – these were mostly used to prefill the questionnaires, but there were cases when the relevant administrative data for some units were received later. |
2. Other sources of processing errors |
The data were checked and edited with the help of special data editing software. In case of errors, imputation was used. Due to several arithmetical and logical checks, it is practically impossible for processing errors to exist in the data. |
3. Tools used and people/organisations authorised to make corrections |
During data processing, the new editing software was used. The people authorised to make corrections included permanent data collection and processing staff. Two persons were also hired temporarily, one of them has worked on the previous Farm Structure Surveys and therefore has great experience. |
|
6.3.4.1. Imputation - rate |
Imputation - rate |
There were cases when the holding accepted the administrative data but did not give further information, but this is not unit non-response and has been treated as item non-response. Therefore, the imputation rates are the same as the item non-response rates presented in item 6.3.3.2 Item non-response error. |
|
6.3.5. Model assumption error |
[Not requested] |
6.4. Seasonal adjustment |
[Not requested] |
6.5. Data revision - policy |
Data revision - policy |
There are no planned revisions of published data. |
|
6.6. Data revision - practice |
Data revision - practice |
There are no planned revisions of published data. |
|
6.6.1. Data revision - average size |
[Not requested] |