Disaggregating UK local authority-
level gross value added to lower
levels of geography:1998 to 2020
Office for National Statistics, UK
Blessing Chiripanhura
Presentation structure
• Why subnational statistics?
• Illustration: Wales
• Basic requirements for subnational statistics development
• Disaggregating UK local authority level GVA to small areas
• Pre-processing of apportioning datasets
• Problems of producing subnational data
• Statistical disclosure
• Detecting and treating statistical disclosure
• Outliers
• Key points to remember
Source: ONS Census 2021 data visualisation
Why subnational statistics?
• There is a realisation that national (or higher-level geographical areas) data
are too aggregated and tend to mask differences within countries.
• Subnational data provides useful and detailed contextual data that improves
our understanding of the social and economic outcomes within regions.
Policy makers use such data for resource allocation to where there is
greatest need.
• Communities / regions are inter-dependent in many ways and such
proximity creates synergies for growth and development. Lack of
subnational statistics reduces our understanding of the synergies.
Basic principles for subnational statistics development
• There is need for a pre-existing national framework within which
development is constrained e.g. National Accounts framework
• Secure access to administrative and other proxy data
• Existence of an up-to-date business register
• Devise an approach for dealing with complex business operations
operating across multiple sites
• Select the appropriate geographical level
• We target the lower-layer super output area (LSOA) | data zone (DZ)| super output area level (SOA)
Disaggregating data to lower levels of geography
Subnational Statistics & Analysis
• In developing the UK gross value statistics for small areas, we use
a simple apportionment method.
• We apportion using administrative and bespoke data sources
• We apportion the VAT turnover of large and complex
enterprises to local units using employee counts at each site
• This approach assumes equal productivity, which may not
be the case
Pre-apportionment data processing
• Data engineering:
• Data linkage – VAT Turnover data + IDBR
• We start with data engineering to allocate VAT turnover to enterprises’ local units. This is
achieved by matching VAT turnover records to the Inter-Departmental Business Register
(IDBR) to create a new dataset that is used to allocate VAT turnover to business sites.
Granular GVA
Using VAT
turnover from
HMRC linked to
IDBR to get local
units for
businesses with
multiple sites
Apportion
turnover based
on employment
at each site
Key assumption
that employees
contribute
equally to the
productivity of
the business
regardless of site
or
characteristics
We apportion to LSOA, DZ and SOA, which we call ‘the building blocks’
We apportion at Section level:
reminder of the sections…
Section Description
A AGRICULTURE, FORESTRY AND FISHING
B MINING AND QUARRYING
C MANUFACTURING
D ELECTRICITY, GAS, STEAM AND AIR CONDITIONING SUPPLY
E WATER SUPPLY; SEWERAGE, WASTE MANAGEMENT AND REMEDIATION ACTIVITIES
F CONSTRUCTION
G WHOLESALE AND RETAIL TRADE; REPAIR OF MOTOR VEHICLES AND MOTORCYCLES
H TRANSPORTATION AND STORAGE
J INFORMATION AND COMMUNICATION
K FINANCIAL AND INSURANCE ACTIVITIES
L REAL ESTATE ACTIVITIES
M PROFESSIONAL, SCIENTIFIC AND TECHNICAL ACTIVITIES
N ADMINISTRATIVE AND SUPPORT SERVICE ACTIVITIES
O PUBLIC ADMINISTRATION AND DEFENCE; COMPULSORY SOCIAL SECURITY
P EDUCATION
Q HUMAN HEALTH AND SOCIAL WORK ACTIVITIES
R ARTS, ENTERTAINMENT AND RECREATION
S OTHER SERVICE ACTIVITIES
T ACTIVITIES OF HOUSEHOLDS AS EMPLOYERS; UNDIFFERENTIATED GOODS-AND
SERVICES-PRODUCING ACTIVITIES OF HOUSEHOLDS FOR OWN USE
Apportioning local authority level GVA to building blocks level
• We start by apportioning GVA for all sections (except O, P, Q, T and part of L) using VAT turnover data:
a)
Building block Section 𝑖𝑖 VAT turnover
LA SUM of Section 𝑖𝑖 VAT turnover
∗ LA Section 𝑖𝑖 GVA = Building block Section 𝑖𝑖 GVA
(where 𝑖𝑖 = all other sections except O, P, Q, T, and part of L (68.2IMP))
This gives the building blocks GVA for all sections with VAT turnover data.
• Next, we apportion the GVA of sections O, P, Q, T and part of L.
b) Sections O, P and Q: Building block Section 𝑖𝑖 employment
LA SUM of Section 𝑖𝑖 employment
∗ LA Section 𝑖𝑖 GVA = Building block Section 𝑖𝑖 GVA
(where 𝑖𝑖 = Sections O, P and Q)
c) Section T: Building block population
LA total population
∗ LA Section T GVA = Building block Section T GVA
d) Section L (68.2IMP): Building bock dwelling stock
LA total dwelling stock
∗ LA Section L 68.2IMP GVA = Building block Section L: 68.2IMP GVA
Checking the apportioned GVA
• After apportioning all industries to building blocks level, we calculate the building block total GVA.
That is, sum across industries to calculate total GVA for each building block:
Total building block GVA = ∑[Building block Section 𝑖𝑖 GVA] where 𝑖𝑖 = Sections A to T.
• Next, we perform a global check of the sum of all building blocks GVA, which must equal the local
authority GVA we started with.
∑Total building block GVA in LA = Total LA GVA
• This summation must equal the Local Authority GVA for a given year. We apply this method for the
years 1998 to 2021 to produce GVA time series for each building block.
Points to note about LSOA, DZ and SOA data
• Customisable, flexible, bespoke geographies for analysis.
• Not constrained by published geographies.
• Focus on transport routes, hubs of industry, cutting across pre-defined
geographical boundaries.
• Individual LSOA|DZ|SOA should not be compared with one another
but aggregated to build larger areas for analysis – used as building
blocks.
• We are not able to produce confidence intervals for out series –
because of the nature of the datasets involved in apportionment.
12
Users with local
knowledge may ‘guess’
the GVA of dominant local
enterprises/businesses
We must apply disclosure
‘treatment’ to adjust
values
In small geographical areas, there is a perceived risk of ‘disclosure’ of
economic statistics:
Statistical disclosure and treatment
13
Stage 1 –
Identification
Identify building
blocks in which
the largest
industry
represents 80%
or more of total
GVA
Identify the
turnover share
of the largest
business
enterprise in
that industry in
the building
block
If a business
enterprise
represents over
80% of total
GVA in the
building block,
it is disclosive.
Testing for statistical disclosure
Treatment options: Suppression
Adding noise to data
Averaging / combining blocks
14
Stage 2 -
treatment
Identify all
building blocks
that fall within
the same MSOA,
IZ or DEA as the
disclosive
building block.
Establish
‘bounds’ which
determine the
minimum
amount the GVA
needs to be
adjusted for the
building block to
become non-
disclosive
Select a building
block within the
same MSOA, IZ
or DEA that is
closest to one of
the bounds
Average the two
building blocks
We must choose building blocks to pair within the same MSOA, IZ or DEA so we maintain the correct local authority totals.
MSOA = middle-layer super output area (larger building block)
IZ = intermediate zone (larger building block)
DEA = district electoral area
Treating statistical disclosure
Sensitivity analysis: Changing the threshold
Cut off points/thresholds LSOA/DZ/SOA
count, 1998-2020
Percentage of
disclosive
LSOA/DZ/SOA
LSOA/DZ/SOA
count, 2012-2020
Percentage of
disclosive
LSOA/DZ/SOA
75%+ 3720 8.7 2611 6.1
79%+ 2833 6.6 1962 4.6
80%+ 2657 6.2 1831 4.3
81%+ 2477 5.8 1706 4.0
85%+ 1757 4.1 1205 2.8
90%+ 1053 2.5 725 1.7
Upper limit: treated LSOA/DZ/SOA = 2657 x 2 = 5314
As a proportion of all LSOA/DZ/SOA: 12.5%
Uniquely disclosive LSOA/DZ/SOA: 1998 to 2020 and 2012 to 2020
Outlier detection and treatment
• Not all building blocks series are smooth – some have outliers
• Driven by VAT turnover data
• Caused by the disclosure treatment.
• We have developed a tool to detect outliers, and to plot the
series to aid adjustment.
• We generate a list of adjustments for each local authority
• Adjustments are re-apportioned in each local authority, to
LSOA|DZ|SOA that were not adjusted.
• Sum of all building blocks GVA in LA must equal LA total GVA
Key messages
Building blocks GVA can allow us to create flexible geographies for analysis
Subnational administrative institutions value subnational statistics because they
informs local policy formulation decision making
Subnational statistics methods continue to development and improve
- Disaggregating UK local authority-level gross value added to lower levels of geography:1998 to 2020
- Presentation structure
- Slide Number 3
- Why subnational statistics?
- Basic principles for subnational statistics development
- Disaggregating data to lower levels of geography
- Pre-apportionment data processing
- We apportion at Section level: �reminder of the sections…
- Apportioning local authority level GVA to building blocks level
- Checking the apportioned GVA
- Points to note about LSOA, DZ and SOA data
- Statistical disclosure and treatment
- Testing for statistical disclosure
- Slide Number 14
- Sensitivity analysis: Changing the threshold
- Outlier detection and treatment
- Key messages