Estimating populations by age, sex, deprivation and ethnic group

Introduction

For some applications we need populations broken down by deprivation and ethnic group, as well as by age and sex. For example, when exploring or presenting data for factors which vary with age, sex, deprivation and ethnic group, but specifically where these are not independent effects, such as when the effect of deprivation on the factor might vary across ethnic groups.

Routine mid-year estimates only give us updated estimates by age, sex and small area (giving us deprivation quintile or decile) but do not include ethnic group. The census, once every 10 years, gives us much more detailed breakdowns by ethnic group, but very granular breakdowns are not possible due to the risk of disclosure.

Synthesising various views of census data allows us to estimate very granular populations quite accurately. Because the population ages by one year each year, we can ‘age on’ estimates for 2021 to subsequent years, constraining to mid-year estimates to account for deaths and migration.

This page describes, step-by-step, the algorithm developed and used by OHID to estimate populations by age, sex, deprivation and ethnic group, for use in a range of analytical contexts.

Data sources

The following three data sources have been used to construct the estimates, and are referred to in the method section by their numbers:

  1. Census population data broken down by:
  • Sex
  • Age group (5-year age bands up to 85+)
  • Deprivation quintile
  • Ethnic group (19 groups)

This is a bespoke analysis, commissioned from ONS.

  1. Census population data broken down by:
  • Sex
  • Single year of age (up to 90+)
  • Ethnic group (19 groups)
  • Lower tier local authority (LTLA)

This can be downloaded from the ONS Create a custom dataset tool. Disclosure control processes necessitate downloading the dataset for all persons as well as broken down by sex, in order to estimate sex-specific populations for some LTLAs. This is described in Appendix 1.

  1. Mid-year population estimates for 2021 and subsequent years, broken down by:
  • Sex
  • Single year of age (up to 90+)
  • Lower layer super output area (LSOA)

These are published tables from ONS.

Method

The analysis consists of two stages and is carried out separately for males and females:

  1. Construct population estimates for 2021 by ethnic group (19 groups), age (single year of age up to 90+) and LSOA, constrained to ONS 2021 mid-year estimates

  2. Estimate populations for subsequent years in the same breakdowns, constrained to ONS mid-year estimates for each year

Stage 1 steps

1.1 Using Source 1, for each 5-year age band, for each deprivation quintile, calculate the proportion of the population in each ethnic group

1.2 Using Source 3 (2021), aggregate the population counts into 5-year age bands and deprivation quintiles (using the lookup table from LSOA to deprivation quintile)

1.3 Combine the tables from steps 1.1 and 1.2 to estimate population counts for each ethnic group, 5-year age band and deprivation quintile, by applying the proportions from step 1.1 to the counts from step 1.2

1.4 Using Source 3 (2021), for each 5-year age band, for each deprivation quintile, calculate the proportion of the population in each LSOA

1.5 Combine the tables from steps 1.3 and 1.4 to estimate population counts for each ethnic group, 5-year age band and LSOA, by applying the proportions from step 1.4 to the counts from step 1.3

1.6 Using Source 2, for each ethnic group, for each 5-year age band, for each LTLA, calculate the proportion of the population in each single year of age

1.7 Using the table from step 1.6, within each ethnic group, within each 5-year age band, attribute the proportions from each LTLA to every LSOA within that LTLA

1.8 Combine the tables from steps 1.5 and 1.7 to estimate population counts by ethnic group, single year of age and LSOA, by applying the proportions from step 1.7 to the counts from step 1.5. The 85+ age band is split into 85, 86, 87, 88, 89 and 90+ at this point.

Figure 1: Flowchart showing the stage 1 calculation process

Stage 2 steps

2.1 For each year after 2021, age on the single year of age LSOA, detailed ethnic group population by the appropriate number of years, replicating the age zero population. When aging on the oldest age groups, add the 90+ populations to the relevant aged-on single year populations. For example, for 2022 each SYOA is aged on by a single year:

  • n0(2022) = n0(2021)
  • n1(2022) = n0(2021)
  • n2(2022) = n1(2021)

and so on for each year of age up to

  • n89(2022) = n88(2021)
  • n90+(2022) = n90+(2021) + n89(2021).

For 2023 estimates n0(2021) would be replicated for n0(2023), n1(2023) and n2(2023). n1(2021) would become n3(2023), and so on.

2.2 Using the table from step 2.1, for each single year of age, for each LSOA, calculate the proportion of the population in each ethnic group

2.3 Combine the table from step 2.2 with the published MYEs (Source 3) for the target year (for example 2022, 2023, etc) to estimate population counts by ethnic group, single year of age and LSOA for the target year, by applying the proportions from step 2.1 to the MYEs

Figure 2: Flowchart showing the stage 2 calculation process, for 2022 as an example

The resulting population estimates may be aggregated into broader ethnic groups, age groups, higher geographies and/or deprivation quantiles as required.

Appendix 1 – Pre-processing of census population data

The following additional steps were taken to prepare census population data from the ONS Create a custom dataset tool for use in stage 1:

  • Records with the ethnic group category “Does not apply” (code –8) were removed.
  • Sex specific populations were not available for 6 LTLAs due to disclosure control applied by ONS. For these areas, population data for persons were extracted and divided by 2 to provide sex-specific estimates. The affected areas were: Allerdale (E07000026), Copeland (E07000029), Eden (E07000030), Torridge (E07000046), West Devon (E07000047), Richmondshire (E07000166), Ryedale (E07000167).
  • Populations for one LTLA (Isles of Scilly (E06000053)) were also unavailable at person level. Therefore, data for Isles of Scilly were excluded from all stages and are not available in the final table.


Page last updated: July 2026