6. Accuracy and reliability |
Top |
|
There are some very big horticultural companies in Finland. Profitability of the horticultural enterprises have been last years better than agricultural enterprises (https://portal.mtt.fi/portal/page/portal/economydoctor/farm_economy/timeline/profitability_ratio_by_production_type) This means that horticultural enterprises have been able to invest and expand their production during last year. We have checked the data and it is right. We get agricultural and horticultural areas and crops from administative registers which reliability is very high. Value of horticultural production is over 10 % of total returns of agricultural and horticultural production. |
6.1. Accuracy - overall |
Main sources of error |
The main sources of errors are sampling errors and measurement errors. However these are not big problems because we have total of the areas and number of animals. There are also one farm which has classification problem. This farm has been classified as horticultural farm and we have used horticultural SO for it. This caused very high SO for the farm. However agricultural area of the farm is real and right. |
|
6.2. Sampling error |
Method used for estimation of relative standard errors (RSEs) |
The results were estimated with SAS software. Variances of the characteristics collected on the sample survey were estimated using the CLAN software developed by Statistics Sweden (see annex, also available at http://www.amstat.org/meetings/ices/2000/proceedings/S09.pdf). |
Annexes: 6.2. CLAN software |
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 |
Relative standard errors are in annex. Even if we collected the main crop and livestock characteristics for each population unit, but carried out a sample survey for other characteristics, we reported at Eurostat's request the relative standard errors (RSEs) for the main crops and livestock characteristics, as if they were collected based on the sample, as well. The purpose is to illustrate the overall quality of the sample. For the applicable cases, the RSEs are 0 as we conduct a census for crop and livestock characteristics.
However assuming a sample survey for crop and livestock characteristics, there are the following 'non-compliance' cases: • breeding sows in NUTS2 region FI1C • poultry in NUTS2 regions FI19 and FI1C. |
Annexes: 6.2.1-1. Relative Standard Errors |
6.3. Non-sampling error |
see below |
6.3.1. Coverage error |
1. Under-coverage errors |
Registers are updated annually in Finland, so under-coverage does not pose a significant problem. The risk for under-coverage is very small, because practically all farms that have significant agricultural production apply for subsidies. Since the 2013 the threshold to the farm has been SO 2000 €. |
2. Over-coverage errors |
During the data collection we collected data from the farms which had stopped production. Administrative registers were also used to find out farms which had stopped farming. We used also a web-survey and telephone interviews to remove any instances of over-coverage, that is, non-functioning farms (sold, combined, or production ceased). Therefore, there is no longer any over-coverage in the final sample frame for the FSS 2016. SO of the farms was calculated for all the farms in the frame. This could be done because all the information needed to the SO calculation are from the administrative registers. Thus there was no farms in the survey whose SO is under the threshold. We have not adjusted the weights afterwards. |
2.1 Multiple listings |
There was no multiple listing in the frame. Each farm is only once in the frame. |
3. Misclassification errors |
There are no wrongly classified units. Classification variables are from registers. |
4. Contact errors |
We have access to the Population register centre's database. Contact data was found using the social security number of the farmer. |
5. Other relevant information, if any |
There is no other relevant information. |
|
6.3.1.1. Over-coverage - rate |
Over-coverage - rate |
There where just a little over-coverage in the sample (about 1%). The sample was re-weighted after the number of the over-coverage units was found out. |
|
6.3.1.2. Common units - proportion |
[Not requested] |
6.3.2. Measurement error |
Characteristics that caused high measurement errors |
The most important administrative source of data for farm structure statistics is Integrated Administration and Control System (IACS), where the date from farm subsidy applications is recorded. Farmers almost invariably fill in their subsidy applications meticulously, as they may otherwise face sanctions. Errors in land areas and livestock figures are usually minor and result from misunderstandings, lack of time, or inaccurate data entry. Information from other animal registers (bovine, pig, sheep and goat) is used as a source of animal number data. Farmers must inform the record keeper of any changes in their farm’s animal numbers by the due date. These registers are therefore largely comprehensive. Farmers found questions concerning their labour force and the farm’s other business activities quite difficult. Calculating working hours retrospectively was a problem, as most farms do not keep an account of working hours. In these cases, calculating the annual number of hours spent on farm work was sometimes challenging. In Finland, agricultural workers – and livestock farmers in particular – work more than 1 800 hours per year, that is, more than one person-year. In previous surveys, forestry work may have been partially included in farm work. However, from 2005 onwards, the number of hours spent on forestry work has been a separate item in the questionnaire. Even now, the classification of certain tasks is open to various interpretations. In some cases, it is not always clear at what point farm or horticultural production becomes further processing, that is, other business activity. Other questions for which farmers’ responses may contain measurement errors include irrigation, arable farming, horticulture, and livestock production. As this information may not be directly obtainable from registers, farmers may find it difficult to provide completely accurate information. This does not, however, have a significant effect on the final results. |
|
6.3.3. Non response error |
1. Unit non-response: reasons, analysis and treatment |
There were two main sources of the FSS data. First source was administrative registers and the second source was questionnaires. Administrative registers include all the farms so there were actually not at all no responding to the variables which are from the registers. Instead there was some non-response in the questionnaires. About 1 600 farms had either refused to respond, or a response had not been received for some other reason (illness, farmer not reached, etc.). Farms and horticultural enterprises that did not respond were left out and remaining farms were re-weighted. We have not run an analysis of non-response. However we suppose that there is no significant difference between non-response and respondents' characteristics. |
2. Item non-response: characteristics, reasons and treatment |
Almost all respondents provided complete information. There were, however, rare exceptions when the farmer did not supply all the required information. For example, some agricultural workers failed to disclose their year of birth and/or gender. In the case of farmers and their spouses, this information was, however, largely available in the Farm Register or IACS customer records. Information on other members of farmers’ families was obtained from the Population Register. Some data on working hours was also missing. These gaps were filled in using a comparable person’s average working hours. For example, if the working hours for a milk-cattle farmer’s wife were missing, the average working hours of a milk-cattle farmers’ wife was used instead. Missing information was so rare that it was dealt with on a case-by-case basis, and case-specific discretion was also used in individual cases. Discretion was used in, for example, information on a farm’s livestock numbers and the farmer’s employment outside the farm. Missing information about other business activities was obtained using corresponding information from the 2013 and 2010 Farm Structure Survey. |
|
6.3.3.1. Unit non-response - rate |
Unit non-response - rate |
Non-response rate was 0 % because we got about half of the data from administrative registers and there is data on every farm. Partial non-response was 10 %. This means that 10 % of farmers didn't answer to the questionnaire. |
|
6.3.3.2. Item non-response - rate |
Item non-response - rate |
There are no characteristics which had high non-response. |
|
6.3.4. Processing error |
1. Imputation methods |
The imputation method used varied depending on the amount of background information available for the variable in question. For example, IACS data on the farmer or farmer’s spouse could be used to fill in missing data about a farm’s labour force. The most common imputation method was to fill in a missing data item using an average obtained from similar farms, or to substitute information on a missing farm with data from a similar farm that had filled in the questionnaire. |
2. Other sources of processing errors |
Due to numerous controls and checks, data processing errors are extremely unlikely. However, there was the potential for error when data from various registers were combined. As all the registers use the same farm identification code, combining register data was relatively trouble-free. There is also a small chance of a processing error occurring when information is modified to fit the format specified by Eurostat. It is sometimes challenging to modify the data obtained from questionnaires so that it matches the variables used in the structure survey. Various errors can take place when information is reformatted. For example, labour force data were collected as working hours and then changed to person-years as required by Eurostat. However, Eurostat’s validation process is highly comprehensive and the potential for errors is minimal. |
3. Tools used and people/organisations authorised to make corrections |
Corrections were the responsibility of the Luke's researcher in charge of the Farm Structure Survey. All corrections/changes were made by order of the researcher. Erroneous values were searched by looking at minimum and maximum values and outliers, and by cross-checks between different characteristics. Once the electronic response period had closed, any missing information was collected via telephone interview. |
|
6.3.4.1. Imputation - rate |
Imputation - rate |
The main variables are from registers. The number of replaced values of the main variables is zero (and the rate is zero). |
|
6.3.5. Model assumption error |
[Not requested] |
6.4. Seasonal adjustment |
[Not requested] |
6.5. Data revision - policy |
Data revision - policy |
The revision policy allows revisions and corrections of the data, after it is published. |
|
6.6. Data revision - practice |
Data revision - practice |
The data collected during the FSS was delivered to Eurostat as a single file. The information was validated by Eurostat, which sent Luke a list of errors and items to be checked. Luke then carried out the necessary changes and corrections. If any errors are later detected or specified, a revised file will be sent to Eurostat. Part of the data were published nationally. Once the data have passed Eurostat’s validation process, a final version will be published. Any corrections to published national data will be made according to the recommendations of the Official Statistics of Finland. |
|
6.6.1. Data revision - average size |
[Not requested] |