CIU Logo
 
GLA 2023-based Household Projections
Ben Corr, GLA Demography
May 2025
 
 
 
 

1 Background

This document provides a brief overview of the draft GLA 2023-based household projections, currently being developed for use in the London SHMA and other parts of the London Plan evidence base. While projections based on alternative methods may ultimately be included in the SHMA, the scope of this document is limited to projections based on a modelling framework broadly similar to that which has been used to produce official household projections for England for more than two decades.

Household projections have long been a key input to the process set out in the National Planning Policy Framework (NPPF) for the objective assessment of overall housing need in Local Plans. However, recent changes to the NPPF dictate that housing targets now be based on the existing number of dwellings and local affordability measures. The role of household projections in the next SHMA and London London is therefore expected to be different to previous versions of these.

For the 2017 SHMA, the overall housing need figure was calculated as the sum of projected future growth in households and the existing backlog of need. Projected households disaggregated by type were then used in a ‘Net-Stock’ model to help determine the required unit mix of additional housing. For the next SHMA, it is expected that London’s overall housing need will be set at 88,000 homes per year in line with the new NPPF methodology, and household projections will primarily serve to help inform the required characteristics of the additional housing.

This dual approach to assessing overall and specific needs adds complication to the design of the projections. Previously, the overall number of households and the number of households by type that fed into the different stages of the SHMA were naturally consistent with one another. For the next SHMA, the overall housing target represents a step change in delivery that does not align with the results of the trend-based approaches typically used in the past. Extrapolating forward past changes in household formation and applying these to trend-based population projections would likely yield annualised growth in the range of 10 to 30 thousand households per year, far below the 80+ thousand consistent with delivery of the target number of homes. Such a disparity in the overall numbers would make use of a Net-Stock approach to determining detailed housing requirements impractical. As such, the projections presented here include variants based on scenarios of future household formation intended to be more consistent with a greatly increased rate of housing completions.

2 Methodology

While the models used to produce official household projections for England have evolved over the last 20 years, they have remained conceptually similar.

In summary:

Each of these steps are performed by local authority, age group, sex, and household type.

The number and type of households projected by the model are dependent on:

The flow diagram below illustrates the 2-stage projection process used in the 2018-based Subnational Household Projections (SNHP) published by the Office for National Statistics. The model used to produce the draft projections presented here follows this same basic structure, with variants based on different population projections and different methods for projecting forward household formation rates.

3 Limitations of the approach

3.1 Data availability

Possibly the most obvious problem faced during development of ONS’s 2016- and 2018-based projections was that of limited data on which to base the model - key inputs to which being the census estimates of households by type and characteristics of reference persons that are used as the basis of estimated and projected Household Representative Rates. The models used for the projections were based on just two data points for the past estimates, because:

  • The census is the only source of this information, with no equivalent data available for intercensal periods.
  • Census data on the basis of the current ‘Household Representative Person’ definition is only available from 1991 onwards. Prior to this, data is only available on the now defunct ‘Head of Household’ definition (the 2012- and 2014-based projections published by DCLG were based on a combination of HRP and Head of Household data to allow a longer series of estimates to be used, but this approach was not without problems)
  • Analysis of the data from the 1991 census during the development of the 2012-based and 2016-based projections determined that it suffered from too many quality issues for it to be used in the models.

Further data limitations relate to the available breakdowns of household types in available tables. The data used as the basis of projections in this note is disaggregated into only six different household types.

3.2 Cohort effects

The models project future representative rates for a given age group of HRP based on past estimates of the rates for that same age group. The projected rates for each age group are therefore independent of the rates of every other age group. In reality, the proportion of the people of a given age that represent a household is not independent of the proportion of people ten years younger that represented a household ten years earlier. That is to say: household formation is subject to age, period and cohort effects, but the model does not capture the last of these. While this was understood during the development of the current models, the lack of data available limited the options to develop a model that better accounted for cohort effects.

3.3 Dependent children

The household types in the model include ‘Household with one dependent child’ and ‘Household with two or more dependent children’. As with all other household types, these are projected forward by extrapolating representative rates and applying these rates to the adult population. The projected number of households with dependent children would therefore be entirely independent of the number of children in the population if it were not for an additional step (omitted from the model diagram above) in which the minimum number of children consistent with the projected households are compared to the number of dependent children assumed in the projected population. If this minimum number is higher then an adjustment is applied to the number of households.

This approach is somewhat inelegant and liable to give questionable results as birth rates change over time. Instinctively, it would seem that a better approach would be to more directly link the number of households with children to the number of children in the population.

4 Household formation

Patterns of household formation within the model are determined by Household Representative Rates (HRR). These rates describe the propensity for individuals in the population to become the reference person for a household and are defined as: \[{HRR_{age, sex, hhtype}}=\frac{HRP_{age, sex, hhtype}}{PHP_{age, sex}}\]

Household Representative rates generally increase with age and are higher for men than women. These patterns are the consequence of:

Representative rates for one person households are highest for older women because shorter life expectancies for men lead to growing numbers of widows in the population as they age.

4.2 Projecting future Household Representative Rates

Continuation of past trends

Previous official projections have assumed some continuation of past trends in household formation. For the 2016- and 2018-based ONS projections, projected household representative rates for the period 2012 to 2021 were generated by fitting a two-point exponential curve to the 2001 and 2011 census data. Rates were held constant from 2022 onwards. The reasoning for this choice was that it was not good practice for the length of the projection period to exceed the length of the period on which the trends were based.

Because household formation rates fell between 2001 and 2011 - especially for young adults - the projected rates reflected the assumption that rates would continue to fall for a further decade. Projected annual growth in the number of households was much lower than was the case in the 2014-based DCLG projections and far below the government’s stated goal of building 300 thousand homes per year. In the author’s opinion, this is the reason why the government declined to use the 2016- or 2018-based official projections as the basis for the ‘Standard Method’ for assessing local housing need that was introduced to the National Planning Policy Framework (NPPF) in July 2018 - preferring to use the 2014-based projections produced by DCLG, which used data for the period 1971 to 2011 as the basis for projected household formation trends.

For the GLA’s draft household projections, a variant using trended household formation rates has been produced which adopts the same two-point exponential method as employed by ONS, but updated to incorporate the results of the 2021 Census. Trends are based on the period 2001 to 2021 and rates are projected forward 20 years to 2041, before being held constant.

Returning to previous levels of household formation

As an alternative to projecting forward on the basis of a continued worsening of household formation, we can instead assume a reversal of past trends and a return of rates to previous levels. Such an approach was proposed some years ago by members of the wider research community [unfortunately, the author cannot now recall who first made the suggestion].

If the purpose of the household projections is to inform users about future housing need, this approach would appear to have clear advantages that of over trending forward past rates:

  • Simplicity - assuming the ‘target’ rates chosen reflect a situation in which the level of household formation is considered to be acceptable, there is no need to separately attempt to quantify and account for any backlog of need that already exists
  • Stability - projections based on trended rates have been - and likely will continue to be - subject to a high degree of variability over time (such a lack of stability has been cited by the government as one of the reasons for moving away from the use of household projections as a basis for assessing housing need). This approach is inherently more stable as the target rates can remain fixed over successive sets of projections and so the number of households at the end of the projection period will vary only with the input population projection.

Two sets of projected HRR have been produced that assume rates for each age group, sex, and household type will return to at least the same levels as they were in 2001 over a 10-year period (2026 to 2036) or a 20-year period (2026 to 2046), after which rates are held constant.

The scale of improvement to household formation that these scenarios represent could only be made possible through a greatly increased supply of homes and/or lower future population trends.

Throughout this document, the different methods for projecting household representative rates are labelled as follows:

  • Continuation of past trends: hrr 2pt exp 2001-21
  • Returning to previous levels of household formation over 20 years: hrr 20yr return 2001
  • Returning to previous levels of household formation over 20 years: hrr 10yr return 2001
  • Household formation fixed at 2021 rates: hrr constant 2021

These approaches result in markedly different rates over the course of the projection period. The size of the difference is greatest in areas and age groups that previously experienced the largest falls in household formation.

4.3 Projected Household Representative Rates

4.3.1 By age

4.3.1.1 HRR 2021 and 2036

4.3.1.2 Absolute change

4.3.1.3 Proportional change

4.3.2 By household type and age

4.3.2.1 HRR 2021 and 2036

4.3.2.1.1 10yr return

4.3.2.1.2 20yr return

4.3.2.1.3 Trended rates

4.3.2.1.4 Constant rates

4.3.2.2 Absolute change

4.3.2.2.1 10yr return

4.3.2.2.2 20yr return

4.3.2.2.3 Trended rates

4.3.2.2.4 Constant rates

4.3.2.3 Proportional change

4.3.2.3.1 10yr return

4.3.2.3.2 20yr return

4.3.2.3.3 Trended rates

4.3.2.3.4 Constant rates

5 Population projections

The household projections incorporate the trend-based outputs of the GLA’s draft 2023-based population projections.

The trend-based projections are produced using a cohort-component model, with future population being based on a continuation of past patterns of fertility, mortality and migration. The model is similar to that used by ONS for their Subnational Population Projections (SNPP), and results are produced for all local authority districts in England and Wales.

5.1 Projection variants

Three variants of the population projections have been used in preparation of the household projections. These variants have been chosen with the aim of providing users with a plausible range of outcomes that can serve as useful basis for planning. The variants differ in terms of:

  • The assumed level of future fertility - high, low, and principal
  • The assumed level of future improvement in life expectancy – high, low, and principal
  • the length of period of past patterns that are used as the basis for projected future migration trends - 5, 10, or 15 years

The three variants used here have been selected with the aim of providing a reasonable range of potential future outcomes for use in local and strategic planning, and are based on the following combinations of assumptions (note that the 2023-based population projections are not yet finalised for publication and throughout this document are referred to by their working designations):

  • Principal Fertility/Principal Life Expectancy/10-year migration (PPT-10yr)
  • High Fertility/High Life Expectancy/15-year migration (HHT-15yr)
  • Low Fertility/Low Life Expectancy/5-year migration (LLT-5yr)

Additional variants of the population projections - using alternative combinations of the assumptions - have been produced and will also be made available for reference.

5.2 Accounting for recent revisions to international migration estimates

These projections exhibit higher population growth than did the previous round, primarily as a result of recent record levels of international migration from non-EU countries.

In November 2024, ONS revised its estimates of UK Long-Term International Migration (LTIM) for the year ending mid-2023. Estimated net migration for the year rose from an already record high of +672k to +902k. In the same release, ONS published provisional estimates for the year ending mid-2024, with net migration to the UK of +724k.

The 2023 Mid-Year Population Estimates do not reflect the revised international flow figures. ONS have said that they intend to issue an updated set of local authority estimates that are consistent with the revised international migration estimates, but these are not expected to be available before July 2025.

For use in the projections, the GLA has created a set of local authority international migration estimates for 2023 that are consistent with the revised UK LTIM figures. The distribution of the UK-level flows by age and sex were taken from the published outputs of the ONS 2022-based National Population Projections (NPP). The GLA then allocated these flows to local authorities based on the proportion of the overall flow for each combination of age and sex that they had received in the 2023 MYE.

For London as a whole, this reallocation gives international flows for 2023 of:

  • In: 358k (MYE 316k)
  • Out: 135k (MYE 162k)
  • Net: +233k (MYE +154K)

Starting populations in the model have been adjusted to be consistent with the new international migration figures. For London, the mid-2023 population is now assumed to be 9.024 million, 79 thousand higher than the official MYE figure of 8.945 million.

The scale of the revisions to the 2023 UK international migration estimates - and limited information on which to account for these in the model - have the potential to add significant distortions into the projections. Users should anticipate that projections will be revised in late summer/early autumn 2025 following the release of updated estimates by ONS.

0 250 500 750 1000 Jul 2012 Jan 2014 Jul 2015 Jan 2017 Jul 2018 Jan 2020 Jul 2021 Jan 2023 Jul 2024 EU non-EU UK Annual net migration (thousands) UK Long-Term International Migration Source: ONS LTIM estimates November 2024 Note: shaded area denotes provisional estimates

5.3 Differences from current London Plan projections

The latest outputs still show a much lower level of growth in London than did the 2016-based projections that underpin the current London Plan.

The GLA 2016-based central trend projection gave London’s population as:

  • 2036 - 10.46 million (current equivalent variant: 9.83 million)
  • 2041 - 10.78 million (current: 10.02 million)
  • 2046 - 11.04 million (current: 10.17 million)

The 2016-based projections were produced at the end of a decade of rapid growth in London fuelled by:

  • high volumes of international migration, including increased flows from from Eastern Europe, following expansions of the the EU in 2004 and 2007, and from Southern European countries in the aftermath of the 2008 Financial Crisis.
  • reduced domestic outflows from London linked to the impacts of the 2008 Financial Crisis on the UK housing market.
  • a 30 percent increase in annual births between 2002 and 2012

Additionally, the projections were based on official estimates of past population that, for much of London, had become increasingly inflated since 2011, when the results of the census were used to rebase the series. Official estimates at the time gave London’s population as approximately 50 thousand higher than we now think was the case. This inflation was extrapolated forward in the models, leading to artificially high projections of population growth.

Subsequent projections produced by the GLA showed successively lower growth as the impacts of EU referendum, falling housing affordability, declining fertility rates, and COVID-19 pandemic all acted to reduce - and for a period reverse - population growth in London.

-40 0 40 80 120 1983 1988 1993 1998 2003 2008 2013 2018 2023 (thousands) London annual population change Source: ONS Mid-Year Estimates 1981-2023

5.4 Projected population

The projections give a range of London’s total population in 2046 of between 9.9 million and 10.52 million, with the principal projection giving a figure of 10.12 million. This equates to growth of between 870 thousand (10 percent) and 1.50 million (17 percent) over the period 2023 to 2046.

In all variants, natural change - the difference between births and deaths - appears to be the source of the majority of London’s population growth, contributing a total of between 1.06 million and 1.38 million. In comparison, the net contribution of international and domestic migration is between -190 thousand and +120 thousand.

However, the high level of natural change in London exists only because of its young age structure, with a relatively large number of women of childbearing age compared to the population of older adults. This young age structure is the result of complex patterns of domestic and international migration flows. Though London’s overall net migration balance is relatively small, the turnover in its population (the sum of flows into and out of the city) is around a million persons each year.

Overall population growth for England as a whole is projected to be similar across the three variants, ranging from 7.3 million to 7.6 million (both 13%). In contrast to London, the net contribution of natural change is small at between -890 thousand and +670 thousand, with almost all projected growth being the result of migration, which contributes between 6.7 million and 8.5 million over the period.

5.4.1 Population

5.4.1.1 London

5.4.1.2 England

5.4.2 Natural change

5.4.2.1 London

5.4.2.2 England

5.4.3 Migration

5.4.3.1 London

5.4.3.2 England

5.4.4 Cumulative change by component

5.4.4.1 London

5.4.4.2 England

5.5 Projected population by age

Across all variants, the number of children in London is projected to fall back from its recent peak, while the population age 65 and over is set grow rapidly - increasing by between 54 and 63 percent by 2046. London’s population is ageing, with its median age projected to rise from 35.7 years in 2023 to around 39 years by 2046.

However, as the rest of the country is projected to age at a similar rate, London is expected to remain by far the youngest region in England, with its median age in 2046 approximately 4.5 years below that of the next youngest region (the North West - 44 years) and almost 9 years below that of the oldest (the South West - 48 years).

5.5.1 Population by age group

5.5.1.1 London

5.5.1.2 England

5.5.2 Age structure

5.5.2.1 London

5.5.2.2 England

5.5.3 Median age by region

5.6 Projected population growth by region

The Wider South East (WSE) regions (London, South East, East of England) are projected to see the largest absolute increases in population. For each of these regions, projections based on longer periods of past migration give the highest growth. Absolute growth across the variants differs more for London than other regions, reflecting the volatility of migration flows over the last decade.

The largest increases for all other regions are seen in projections based on the shortest period of past migration (despite this variant also having the lowest life expectancy and fertility assumptions). This reflects the increase in international migration seen outside of the WSE in recent years. Differences in growth between regions are much less when considered to their relative sizes, but by either measure the North East, and Yorkshire and the Humber, are projected to grow the least.

5.6.1 Annualised growth

5.6.2 Proportional growth

5.7 Population by residence type

The projected population is split into those resident in private households and those in communal establishments, with only the projected private household population contributing to household growth.

The approach used here for splitting the population by residence type is the same as used in previous rounds of official projections:

  • Up to age 74 the number of persons resident in communal establishments is assumed to remain constant at 2021 levels.
  • For ages 75 and over, the proportion of the population resident in communal establishments is assumed to remain constant.

The assumption that the same proportion of persons over the age of 75 will continue to live in communal establishments is one that should be reviewed ahead of production of final projections. The 2021 census showed a fall in both the number and proportion of older persons residing in communal establishments. This may be the result of:

  • Limitations in the capacity or affordability of residential care
  • Short-term impacts of the COVID-19 pandemic
  • Increases in male life expectancy leading to more couples surviving into old age and allowing more people to continue living independently

5.7.1 Residents in communal establishments

5.7.1.1 London

5.7.1.2 England

5.7.2 Proportion resident in communal establishments

5.7.2.1 London

5.7.2.2 England

5.7.3 Projected communal establishment residents by age

5.7.3.1 London

5.7.3.2 England

6 Household Projection Results

6.1 Total households

Despite variants spanning a wide range of future population growth - the results show that the effect of population growth on the number of projected households is far less significant than the choice of assumption about future household formation.

Assuming the central of the three population variants, the number of projected households in 2036 ranges from 3.74 million (trended rates) to 4.51 million (rates returning to at least 2001 levels over a ten-year period) across the different methods for projecting household representative rates. This is represents a increases over the decade from 2026 of between 174 thousand (4.9 percent) and 848 thousand (23.2 percent).

In contrast, assuming household representative rates remain at their 2021 levels, the projected growth in households by 2036 ranges from 313 thousand (8.6 percent) to 412 thousand (11.2 percent).

Across all combinations of HRR method and population variant, the projected growth ranges from 125 thousand (3.5 percent), using the llt_5yr population variant and trended rates, and 905 thousand (24.6 percent), using the hht_15yr variant and assuming a return to past household formation over 10 years.

Note: the household growth figures used in this note are all relative to 2026. Readers should be aware that the 2026 figures vary between outputs because: 1. the population projection variants diverge after 2023; 2. HRRs for the trended method begin changing after 2021, while the rates for all other methods are held constant at 2021 levels until 2026.

Similar patterns hold across all regions, however the scale of the differences across HRR assumptions is greatest in London as this is the region where household formation has been most constrained over the last two decades - meaning both that trending forward past rates leads to the largest future decreases and that assuming a return to past levels gives the largest increases.

For England, projected growth ranges from 1.74 million (7.1 percent), using the hht_15yr population variant and trended rates, and 4.75 million (19.1 percent), using the llt_5yr variant and assuming a return to past household formation over 10 years.

Note: the three population variants used here were selected with the intention of providing a useful range for London - pairing the high fertility and life expectancy assumptions with the 15-year migration scenario (as this gives the highest growth for London) and the low fertility/life expectancy assumptions with the 5-year migration scenario (giving the lowest growth for London). Outside of the Wider South East, the shorter period migration scenario gives higher growth than the longer period one, meaning that results using additional variants (e.g. the llt_15yr and hht_5yr) should be considered to get a fuller understanding of variation across all regions.

6.1.1 All HRR methods

6.1.1.1 London

6.1.1.2 England

6.1.2 20yr return

6.1.2.1 London

6.1.2.2 England

6.1.3 10yr return

6.1.3.1 London

6.1.3.2 England

6.1.4 Trended rates

6.1.4.1 London

6.1.4.2 England

6.1.5 Constant rates

6.1.5.1 London

6.1.5.2 England

6.2 Annualised change in total households from 2026

6.2.1 By HRR method

6.2.1.1 London

6.2.1.2 England

6.2.2 20yr return

6.2.2.1 London

6.2.2.2 England

6.2.3 10yr return

6.2.3.1 London

6.2.3.2 England

6.2.4 Trended rates

6.2.4.1 London

6.2.4.2 England

6.2.5 Constant rates

6.2.5.1 London

6.2.5.2 England

6.3 Regions - change in households from 2026

6.3.1 By HRR method

6.3.1.1 Annualised

6.3.1.2 Proportional

6.3.2 20yr return

6.3.2.1 Annualised

6.3.2.2 Proportional

6.3.3 10yr return

6.3.3.1 Annualised

6.3.3.2 Proportional

6.3.4 Trended rates

6.3.4.1 Annualised

6.3.4.2 Proportional

6.3.5 Constant rates

6.3.5.1 Annualised

6.3.5.2 Proportional

6.4 Average Household Size

Higher household representative rates result in a larger number of households for a given population and a lower average household size (AHS). Projected average household size therefore depends on both the method used to determine future HRR and the structure of the population (as HRR varies by age and sex).

The ageing of the population and falls in births rates across all regions leads to a tendency for household sizes to decrease as:

  • household formation increases with age

  • children don’t form households and so add to the household population without adding to the number of households

Household sizes are only projected to increase in London when the trended method is used and therefore when household formation is assumed to keep falling. In all other scenarios AHS falls from the recent highs captured by the 2021 census.

6.4.1 By HRR method

6.4.1.1 London

6.4.1.2 England

6.4.2 20yr return

6.4.2.1 London

6.4.2.2 England

6.4.3 10yr return

6.4.3.1 London

6.4.3.2 England

6.4.4 Trended rates

6.4.4.1 London

6.4.4.2 England

6.4.5 Constant rates

6.4.5.1 London

6.4.5.2 England

6.5 Change in households by type

In all variants, the largest proportional increases from 2026 to 2036 in households are for ‘Single family households with all members aged 66 years or older’, ranging from 19.6 percent to 66.0 percent in the ppt_10yr population variant. These increases are driven by the large proportional increases in the population aged 65 and over.

The largest absolute increases are seen in ‘One person households’, with numbers increasing by 416 thousand for the 10yr return HRR method. When the trended rate method is applied, the same household type is also the only one to see an absolute decrease in numbers, falling by 22 thousand. This stark difference reflects the fact that one-person households have seen the largest absolute falls in HRR over the last two decades.

6.5.1 Absolute change

6.5.1.1 London

6.5.1.2 England

6.5.2 Proportional change

6.5.2.1 London

6.5.2.2 England

6.5.3 Broad household types

6.5.3.1 London

6.5.3.2 England

6.6 Change by age of Household Reference Person

Across all projections, numbers of households with an HRP aged 65 and over are set to increase strongly in both absolute and proportional terms over the decade to 2036.

Below the age of 50, there are large differences between projections according to HRR method used - with trended rates giving small increases or absolute falls in numbers, and an assumed return to past levels giving growth across all age groups. As was the case with the results by household type, this pattern reflects the fact that falls in household formation since 2001 have been largest for younger age groups.

6.6.1 Absolute change

6.6.1.1 London

6.6.1.2 England

6.6.2 Proportional change

6.6.2.1 London

6.6.2.2 England

6.6.3 Broad age group of HRP

6.6.3.1 London

6.6.3.2 England

 
 
 
For more information please contact the Demography Team, GLA City Intelligence Unit.
demography@london.gov.uk
 
Copyright © Greater London Authority, 2025