Appendix 4: Survey methodology
Sampling and data collection
Using the June 2021 Agricultural Census (13) (Tables 21 & 23), two samples were drawn. The first sample was taken from holdings with grassland (Table 18), the second from holdings growing fodder crops (Table 19). For the purpose of sampling, the country was divided into 11 land–use regions (Figure 37). The sample was stratified by these regions and also according to holding size. The holding size groups were different for grassland and fodder crops (Tables 18 & 19) and were based on the total areas of crops grown on the holding. Holdings were chosen at random within each of these strata, with the numbers of holdings selected being proportional to the total area of crops grown. Sample sizes for each stratum were based on area rather than number of holdings, so that smaller holdings did not dominate the sample.
The survey period for pesticide applications to grassland was from August 2020 to August 2021. For fodder crops, the survey period covered pesticide applications during the 2021 growing season, including any post-harvest applications following the 2020 harvest through to the end of harvest in 2021. As well as recording treatments applied directly to the crop, land preparation treatments prior to sowing the crop were also collected.
For holdings in the fodder sample, an introductory letter was sent to farmers followed up by a telephone call. The majority of information was gathered during this telephone call, although some holdings required a subsequent telephone call or email. When necessary, data were also collected from consultant agronomists, contractors and seed merchants. In addition to information about fodder crops, pesticide use data were also collected for grassland crops grown on holdings selected in the fodder sample. In total, data were collected from 182 fodder holdings. These 182 holdings collectively grew 16 per cent of the census fodder area. Details of the distribution of the fodder sample can be found in Table 19.
Postal questionnaires were sent to holdings selected in the grassland sample. This postal survey supplemented the grassland data collected during the fodder survey. This combined dataset ensures that the proportionately large areas of grassland grown in Scotland are adequately represented in the survey. Of 1,298 questionnaires sent out there were 438 (34 per cent) useable responses (Table 27). Details of the distribution of the grassland sample can be found in Table 18. The grassland sample represented six per cent of the total grassland grown in Scotland and two per cent of the total rough grazing.
For both samples, the data collected included the area of grassland and/or fodder crops grown, selected agronomic information and a record of the area and weight of all pesticide applications. Holdings that were not able or not willing to provide data were replaced with alternative holdings from the same region and size group where possible.
National pesticide use was estimated by ratio raising. This is a standard statistical technique for producing estimates from a sample. It is the same methodology used by the other UK survey teams and has been used for all historical datasets produced by the Pesticide Survey Unit, allowing comparability over time. The sample data were multiplied by raising factors (Tables 24 & 25). These factors were calculated by comparing the sampled area to the areas recorded in the 2021 Agricultural Census within each region and size group. Grassland is raised only by region, size groups are not taken into account. An adjustment (Table 26) was made for each crop within each region by applying the raising factors to the sample area of each crop grown and comparing this with the census area. This adjustment modifies the estimate to take into account differences in composition of crops encountered in the sample and those present in the population. A second adjustment was necessary for some crops which were present in the population but were not encountered in the sample in some strata.
Changes from previous years
There have been no changes in methods or presentation between 2021 and the previous report in 2017. However, as described in the 2017 report(3), the following change should be noted when making comparisons with the 2013 data:
In 2015 there was a change in the census definition of temporary and permanent grass on the Single Application Form (SAF). This change is explained in clause 4.8 of the 2017 Scottish agriculture census(17). From 2015, temporary grass relates to whether it has been reseeded in the last five years, whereas previously it related to how long it had been used for grass. The new definition only includes land that is included in a holding's crop rotation. This means changes in grass one to four years and grass over five years between 2013 and subsequent surveys do not solely represent genuine changes in land use but include differences in the way crop data has been recorded. As sampled areas are raised to census areas this will also influence our estimates of pesticides used. Whilst it is not possible to quantify the impact of these changes, data users should be aware of them when making comparisons of total pesticide use over time. To aid interpretation, the trends section presents pesticide usage information in relation to crop area grown as well as overall estimates of use.
Data quality assurance
The dataset undergoes several validation processes as follows; (i) checking for any obvious errors upon data receipt (ii) checking and identifying inconsistencies with use and pesticide approval conditions once entered into the database (iii) 100 per cent checking of data held in the database against the raw data. Where inconsistencies are found these are checked against the records and with the grower if necessary. Additional quality assurance is provided by sending reports for review to members of the Working Party on Pesticide Usage Surveys and other agricultural experts. In addition, the Scottish pesticide survey unit is accredited to ISO 9001:2015. All survey related processes are documented in Standard Operating Procedures (SOPs) and our output is audited against these SOPs by internal auditors annually and by external auditors every three years.
Main sources of bias
The use of a random stratified sample is an appropriate survey methodology. A stratified random sample, grouped by farm size and region, is used to select holdings used in this survey. Sampling within size groups is based on area rather than numbers of holdings, so that smaller size groups are not over-represented in the sample. The pesticide survey may be subject to measurement bias as it is reliant on farmers/growers recording data accurately. As this survey is not compulsory it may also be subject to non-response bias, as growers on certain farm/holding types may be more likely to respond to the survey than others. Reserve lists of holdings are held for each stratum to allow non-responding holdings to be replaced with similar holdings.
Experience indicates that stratified random sampling, including reserves, coupled with personal interview technique, delivers the highest quality data and minimises non-response bias.
There is a problem
Thanks for your feedback