Results from the Scottish Agricultural Census: June 2023 Methodology Report

Data Collection

The June Agricultural Census is conducted annually by the Scottish Government’s Rural and Environmental Science Analytical Services division (RESAS). Data are requested from all holdings who submitted a Single Application Form (SAF) in the previous year, together with some other large businesses that would not be eligible for support payments. A sample of smaller holdings which didn’t submit a SAF or who didn’t return a form for the previous census were also sent a census form.

Data for the June census is collected from three sources:

• Land data were extracted from the Single Application Form (SAF) database for around 22,800 holdings that are claiming under the Basic Payment Scheme (BPS).
• All holdings within our sample were sent a census form covering land, livestock and labour. Holdings completing a SAF were not expected to complete section 3 (land area) of the census form.
• All cattle data (including data on cattle breeds) were collected from the CTS administrative source. Farmers are required by law to register any births, deaths and movements of their cattle to CTS who maintain an online database. More information can be found online here. By using this data source we effectively have 100 per cent coverage, even for those smaller holdings that were not selected for inclusion in the census.

RESAS categorises all agricultural holdings in Scotland as either a “Main” holding or a “Min” holding. A holding is a “Main” if it meets certain thresholds such as a certain area of land or a certain amount of animals. All “Main” holdings in Scotland are included in the census. The smaller holdings are only surveyed once every four years to reduce the burden on the smaller farms. Holdings which have not received a form in the past four years are included in the sample.

There are 44,698 known agricultural holdings in Scotland and within the census 33,215 were selected to complete a census form (the majority of these online). Of those who received a request to complete the census 18,269 provided a return, resulting in a 55 per cent response rate.

Table 1 below shows the response rates for different types of farms. The first column shows the percentage of holdings that responded. Due to the fact that some larger farms account for a larger proportion of overall activity or land area in Scotland, the second column shows what percentage of land area, livestock or workforce each category represents. In most cases this is higher than the percentage of holdings, indicating that there is better coverage in the census of larger holdings.

Table 1 presents the percentage cover of holdings, land area, livestock and workforce for different farm features

A downloadable copy of this table can be found here.

Online Data Collection

Since 2017 respondents have been able to complete their census online, through the Rural Payments and Services website. This year, all those in the sample who had completed the June Census online the previous year or had an account with the Rural Payments and Services online portal were sent an email notification to complete the survey. All other census recipients received paper forms. In total, 11,792 SAF holdings returned online data, 84 per cent of SAF census returns or 51 per cent of the SAF sample. In addition, we received 1,284 online non-SAF returns. For future surveys RESAS will continue to improve the online response rate.

Accuracy

Data undergo several validation processes as follows; (i) checking for any obvious errors on the paper and online census forms upon receipt, (ii) auto-checking and identifying any internal inconsistencies once loaded onto the initial database, (iii) auto-checking for any sudden changes in comparison with previous annual returns and other holdings, and (iv) assessing farm-level trends between years and any trends or switches in item areas or quantities that look unreasonable.

If necessary, farmers are contacted to ensure data are correct. Additional quality assurance is provided at the later stages by utilising expert knowledge within the Scottish Government and the agriculture industry.

Treatment of non-response

In Scotland the registered details of the 44,698 agricultural holdings are used to maintain a holding-level dataset of agriculture for statistical purposes. This provides a virtually complete coverage of agricultural activity in Scotland. However, please note that:

• as noted above, we do not conduct a full census as this would place an unnecessary burden on farmers.
• for the holdings that are surveyed, not all farmers return data to us.
• gaps in our holding-level data set are ‘maintained’ by producing estimates.

Labour, occupier working time, occupier gender, and legal responsibility figures are rolled forward using the most recently returned data within the last ten years of Census (i.e. 2012 onwards). Holdings which had not returned either within that timeframe were estimated to be zero for the relevant census items. Legal responsibility responses were also rolled forward for holdings which provided a form but did not answer this question.

Multiple imputation was used for missing crop and livestock data, excluding cattle and poultry. This involves replacing missing data not with a single value, but with multiple values drawn from the estimated distribution of what the missing value was expected to be. This is based on a statistical model linking current values of a census item to other information about the agricultural holding, primarily what the values were in previous years when the census data were collected. Multiple imputation has the advantage that it accounts for the uncertainty in imputation, and this can be indicated by giving confidence intervals for the reported summaries.

Our imputation was based on predictive mean matching as implemented in the Multivariate Imputation by Chained Equations (MICE), package for R. The predictive mean matching method is based on finding similar observations where data are not missing, and drawing samples from these. It therefore avoids making any inappropriate assumptions, such as Normality of distributions, about the data. This does multiple imputation using fully conditional specification implemented by the Multivariate Imputation by Chained Equations (MICE) algorithm as described in Van Buuren and Groothuis-Oudshoorn (2011).

 Census items were only imputed when the holding had returned a Census or a SAF within the last ten years of Census (i.e. 2012 onwards). Holdings which had not returned either form within this timeframe were estimated to be zero for the relevant census items.

Table 2 at the end of this guidance provides a measure of uncertainty around the imputation values created for each item total.

Estimating for summary values

This year (2023) we reduced the level of detail requested within the land use section of the census form (Section 3). We requested summary figures for crop area. This was done to reduce farmer burden. Individual crops can be identified directly from the SAF data set, this accounts for more than 80% of the arable land area in Scotland. For farms not completing a SAF application the summary values were apportioned to individual crops based on their last return. This methodology will be reviewed over the coming years to determine accuracy over time.