Small Business Bonus Scheme: evaluation

Appendix F: NDR – IDBR data linkage exercise (provided by the Scottish Government)

Executive summary

• A dataset of 130,861 non-domestic properties derived from the 2018 Non-Domestic Rates (NDR) Valuation Roll and Billing System Snapshot were matched to the Inter-Departmental Business Register (IDBR).
• In total, 47% of all properties were matched successfully, using a variety of name and address information.
• The dataset included both SBBS recipients and non-SBBS recipients. The match rate for SBBS recipients was identical to the overall match rate of 47%.
• Properties with a higher rateable value were more likely to match to the IDBR than those with a lower rateable value.

Background

This report describes the process and results of linking data on Small Business Bonus Scheme (SBBS) relief recipients to the IDBR. The data-linkage exercise was carried out to gain a better economic understanding of the relief recipients as part of the SBBS Review. The linked data will be used by the contractor, the Fraser of Allander Institute, to describe the types of businesses in receipt of SBBS rates relief, and allows (through the IDBR reference number) linkage to other economic datasets.

To gauge the feasibility of the data-linkage exercise, we first worked with statisticians in the Office of the Chief Economic Advisor (OCEA) to link a sample of 1,000 SBBS recipients to the IDBR. The initial match rate of the sample was 42%. After further review of the data we deleted a number of incorrect matches. Most of these resulted from matching to the wrong name or address information in the sample, for example where a match was based on proprietor name and address even though the ratepayer was the tenant. The final match rate of the matched sample was 34%, which was considered sufficient to proceed with the full data-linkage exercise which was carried out by the ONS. We used the results of the matched sample to inform the specification of the full data-linkage exercise, which led to additional name and address information being included. As shown below, this contributed to a higher match rate in the full data-linkage exercise.^[51]

Non-domestic rates data

We selected all SBBS relief recipients from the most popular property core descriptions to avoid including property types that we knew were unlikely to be on the IDBR, e.g. public toilets and sewage tanks. We also limited our selection to those core descriptions that had at least 100 relief recipients. A full list of all core descriptions included in the dataset is provided in Section F.1. In total, 86% of all SBBS relief recipients were included in the dataset for matching. We also included a number of properties with the same core descriptions and with a rateable value (RV) below £20,000 which were not in receipt of any kind of NDR relief, to allow the contractor to compare SBBS relief recipients to non-recipients.

Our NDR data provides a number of different name and address variables. To allow all of these to be used in the data-linkage exercise, we split the data into four datasets that used different combinations of the name and address variables. This resulted in the following datasets:

1. Business name & property address (130,861 observations)

2. Ratepayer name & ratepayer address (97,362 observations)

3. Business name and proprietor, tenant or occupier address (97,648 observations)

4. Ratepayer name & property address (97,398 observations) ^[52]

Dataset 1 is larger than the others because it includes non-SBBS relief recipients.

Analysis

ONS matched all four datasets separately to the IDBR. Matches were searched for on any record on the IDBR (VAT, PAYE, local unit, reporting unit or enterprise). For each dataset, we received three files with results: matches, non-matches and multiple matches. We merged these with the original rates datasets. For each match, ONS provided an enterprise reference number as well as all Scottish local unit numbers associated with that enterprise number. Due to the research question at hand, we only focused our analysis on the enterprise-level results, although the final dataset contains a count variable that indicates the number of Scottish local units associated with each enterprise reference number.

ONS also included matches to historical IDBR records. On advice from colleagues in OCEA, we decided to exclude any matches to firms that had ceased trading prior to 1 January 2017.

For each matched property we received a match score. ONS only included those matches with a match score of 84% or higher.

As it was possible for properties to match to different enterprises across the four datasets, we used the following selection criteria to decide which matches to keep:

1. If only one match across all four datasets, use that reference number.

2. If matched to the same reference number in all datasets where there is a match, use that reference number.

3. If matched to different reference numbers across the datasets, use the reference number that occurs most frequently.

4. If matched to different reference numbers across the datasets and no reference number occurs more frequently than another, use the reference number with the highest match score.

There were no cases in which several reference numbers had the same frequency, and the same match scores, so these four criteria were considered sufficient to select an enterprise number to retain for all matches.

Additional checks

To assess the quality of the match we carried out some further checks. We first inspected a number of matches manually to see if they were as expected. This did not reveal any unexpected matches.

We also reviewed where multiple properties matched to the same enterprise. Across the matched dataset, in the vast majority of cases less than ten properties matched to a single enterprise (99.4%). Most enterprises (82.4%) matched to a single property.

We checked cases where more than 50 properties had matched to the same enterprise. Examples of a large number of properties matching to the same enterprise and properties still receiving SBBS relief include local authorities, time-share companies and a large business operating as a franchise.

We also matched the results back to the pilot to see if results were similar. Of those cases matched in the pilot, around 60% had the same enterprise reference number in the full-data match. Around 40% of the properties in the pilot yielded a different result in the full-data match. Three quarters of these had a match in the pilot but not in the full data linkage; one quarter was matched to a different reference number in the pilot than in the full data linkage exercise, with about three-quarters not matching to the IDBR.

Results

Through linking the matched datasets back to the original sample we are able to calculate the match rate of the data-linkage exercise. Table F.1 presents the match rate across the four datasets as well as the overall match rate for the whole sample, which is 47%.

Note: Business Name and Proprietor, Tenant, and Occupier (PTO) Address (PTOA) are derived from the PTOA dataset, Property Address comes from the Valuation Roll. Ratepayer Name and Ratepayer Address come from the Billing System Snapshot. See Annex B for more detail.

Dead records are IDBR records with a death date before the 1st of January 2017 (i.e. those that had ceased trading prior to this date).

The IDBR excludes sole traders or partnerships with no employees and an annual turnover of less than the VAT threshold. This would suggest that smaller businesses (which are less likely to be on the IDBR) are more likely to be excluded from the matched sample. Comparing the mean RV of matched and non-matched cases did in fact show that the mean RV for properties that matched to the IDBR is around £2,000 higher than the RV of properties that did not match, a statistically significant difference. This indicates that properties with a higher RV are more likely to match to the IDBR. This is in line with the findings of the smaller sample matching exercise.

Table F.2 shows the match rate by RV quintile. This shows that as RV increases, properties are more likely to match to the IDBR. Table F.3 shows that properties that were part of a chain were more likely to match to the IDBR than those which occupied a single site. Once again this illustrates that larger businesses were more likely to be matched than smaller ones.

The match rate for SBBS and non-SBBS relief recipients was identical, at 47%. In total, 46,040 SBBS properties were matched, 47% of all SBBS properties (35% of all properties included in the sample).

The tables below compare the matched and non-matched properties by local authority, property core description and property class.

Property class and core descriptions are classifications used by the Scottish Assessors to describe the type of a property. Properties are classified into one of 20 classes, which describe the general type such as 'shop' or hotel', and into one of 202 core descriptions, such as 'retail shop' or 'hostel'.

F.2 Variables included in data matching

The variables included in the data matching were derived from three different datasets: the Billing System Snapshot, Valuation Roll and PTOA dataset.

Billing System Snapshot

The Billing System Snapshot is collated annually and holds information used by local authorities to keep track of payments and reliefs. The Ratepayer name and address are derived from this dataset.

Valuation Roll

The Valuation Roll is maintained by Scottish Assessors and hold information on all non-domestic properties. The Property address variables come from this dataset.

PTOA dataset

Finally, the PTOA dataset holds at least one Proprietor, Tenant, or Occupier record for each Valuation Roll entry. There can be a tenant or occupier in addition to a proprietor, so this dataset may hold multiple name and address variables for each property. The occupier status variable identifies whether the property is in occupation by the Proprietor, Tenant, or Occupier, or if the property is vacant. This variable is also used to derive the Business Name variable, which uses the Proprietor name in case the property is vacant or occupied by a proprietor, the Tenant name in case the property is occupied by a tenant and the Occupier name where there is an occupier.

Variable: Business name - Source: PTOA dataset

Variable: Ratepayer name - Source: Billing System Snapshot

Variable: Property address - Source: Valuation Roll

Variable: Ratepayer address - Source: Billing System Snapshot

Variable: PTOA address - Source: PTOA dataset