Comparing subnational population projections to estimates report

The Office for National Statistics produces subnational population projections for England (SNPPs) by age and sex. The projections are based on the mid-year population estimates (MYEs), historical demographic data, and a set of assumptions regarding future fertility, mortality and migration.

The primary purpose of the SNPPs is to provide an indication of the potential future population, which is used for local planning in several important policy areas. It is important to emphasise that the SNPPs are not forecasts and do not attempt to predict the impact of local and government policies, changing economic circumstances, or other factors.

This report assesses the accuracy of the 2012-based, 2014-based, 2016-based and 2018-based SNPPs by comparing the projected populations for 2021 with the 2021 MYEs. The impact of rebasing and revisions that occurred following Census 2021 on these differences is also considered. The terms "accuracy" and "error" throughout this report refer to the differences between the SNPPs and the MYEs. They do not mean that the projections were incorrect.

The report shows: 

• the population in 2021 was lower than that projected by all SNPPs 

• SNPP-MYE differences were larger for smaller geographical areas

• differences were largest for the London region 

• local authorities with the largest differences had their populations revised by a similar amount during the post-2021 Census rebasing

Subnational population projections

The Office for National Statistics (ONS) produces subnational population projections (SNPPs) for geographies below the national level (that is, regions, counties, local authorities and heath geographies) in England. These are based on the mid-year population estimates (MYEs) for that year, and project the population for the next 25 years.

The subnational population projections for England are produced in the same way for all local authorities and use a standard demographic "cohort component" method in which high-quality data sources inform components of population change such as births, deaths and migration.

SNPPs use historical demographic data on fertility, mortality and migration to develop assumptions about what may happen in the future. The projections resulting from the application of these assumptions provide an indication of the possible size and structure of the future population based on the continuation of these past trends.

Additionally, variant subnational population projections were produced alongside the 2016- and 2018-based principal projections. These were based on alternative migration scenarios to show users a range of possible outcomes. All analyses in this report are for the principal projections. 

There are four complete sets of official SNPPs spanning the 2011 to 2021 intercensal period: 

• 2012-based 

• 2014-based 

• 2016-based 

• 2018-based 

In 2021, following user feedback and support requesting that the next release should benefit from Census 2021 and rebased input data, it was decided that the next release of SNPPs would follow census, and 2020-based SNPPs were not produced. The publication of the 2022-based SNPPs has therefore followed a longer interval than usual because of this decision and the need to first develop national population projections (NPPs) to feed into the SNPPs.

Since the 2018-based SNPPs, there have been two sets of interim NPPs: 2020-based and 2021-based, along with a further NPP variant. The most recent SNPPs for which accuracy can be measured are the 2018-based SNPPs, as 2022-based SNPPs were published on the same date as this report.

Mid-year population estimates

The mid-year population estimates (MYEs) provide estimates of the resident population of England and Wales by age and sex at various levels of geography as of 30 June each year. They are produced on a consistent basis across all local authorities.

The MYEs use the cohort component method (as described in Population estimates for England and Wales, mid-2023: methods guide) to roll forward population estimates from the previous year. They undergo a reconciliation and rebasing process every 10 years following a census.

How accuracy is measured

Accuracy is measured by comparing populations projected for a particular year with MYEs for that year. Various approaches can be used to measure accuracy. Our Comparing national population projections to estimates report averaged the differences (projected estimates minus MYEs) for each NPP for a specified number of years, equating the trend length for all NPPs, whereas our previous Subnational population projections accuracy report (PDF, 492KB) compared all SNPPs with 2011 Census-based MYEs as a fixed timepoint.

The approach used in this report is to compare projected populations for 2021 from each SNPP with the 2021 MYEs. The mid-2021 MYEs are considered to be the most reliable population estimates because they are based on Census 2021. Using the 2021 MYEs also allows analysis of the impact of the 2021 rebasing on SNPP accuracy.

It should be noted that the varying durations between SNPP base years and 2021 projections could affect accuracy; earlier SNPPs may have smaller rolled-forward differences in their base year MYEs because of their proximity to 2011 Census but may have accumulated larger differences by 2021 than later SNPPs. The converse is true for later SNPPs. 

Accuracy can be measured for overall estimates and for individual components of change such as net migration, births and deaths. There are additional factors that may contribute to differences between the SNPPs and the MYEs. This report provides users of projections with a greater understanding of these factors.

Issues when measuring accuracy

Factors that may have influenced SNPP accuracy between 2011 Census and Census 2021 include:

• methodological differences between MYEs and SNPPs

• methodological improvements to MYEs

• rebasing of the MYEs 

• changes in demographic trends

• subnational boundary changes across the decade 

• census confidence intervals

Methodological differences between the MYEs and SNPPs

SNPPs are produced using historical data and population component trends that are projected into the future, whereas MYEs are produced using changes to the population components measured during the previous year.

The SNPPs do not incorporate any impact on the population of changes in policy, economic circumstances, or significant events, whereas these may affect the MYEs. This means that the SNPPs do not attempt to predict the impact of political circumstances such as the UK’s withdrawal from the European Union, or significant events such as the coronavirus (COVID-19) pandemic, whereas the MYEs will reflect the impact of historical events such as these.

Methodological improvements to the MYEs

Methodological improvements to migration and population statistics between 2012 and 2021 include changes in the way internal migration was measured and the transformation of estimates of long-term international migration, which includes the replacement of most International Passenger Survey data with administrative data. Current official estimates of international migration before 2020 have now been revised to make use of this new methodology, and are reflected in the rebased MYEs.

Rebasing of MYEs

Following Census 2021, the census-based 2021 MYEs were compared with the 2021 MYEs rolled forward from 2011. This reconciliation of mid-year population estimates with Census 2021 found that MYEs for England rolled forward from the 2011 Census were 201,700 higher than those based on Census 2021.

The rebasing of mid-year population estimates following Census 2021 adjusted the MYEs to align with Census 2021, correcting the components of change for each set of MYEs. Most changes made during rebasing can be attributed to the components of change, which are births, deaths and migration. The small amount that cannot be attributed to a specific component is called "unattributable population change" (UPC).

Rebased MYEs replaced the earlier rolled-forward MYEs as the official MYEs for England and Wales. This means that SNPPs produced during this period were based on official MYEs that differ from the rebased MYEs. Some of the differences between projected estimates and MYEs will be because of this rebasing process.

The impact of rebasing on accuracy can be examined both by comparing SNPP base year MYEs (that is, 2012, 2014, 2016, 2018) with rebased MYEs for those years, and by comparing projections for 2021 with 2021 MYEs rolled forward from 2011 and with rebased 2021 MYEs.

Changes in demographic trends

The fertility and mortality rates that inform assumptions used in the projections have changed over time. SNPPs between 2012 and 2021 projected an increase in births as their assumptions were informed by an increase in fertility between 2002 and 2008 and a levelling-off between 2008 and 2012. However, actual births declined between all SNPP base years and 2021.

Life expectancy was projected to increase during this time period following improvements up until 2011, yet the number of deaths per year was higher in 2021 than in all SNPP base years. Importantly, the methods used to produce fertility and mortality rates have not changed, so they are more directly comparable at different time points than migration trends. 

Patterns of international migration have also changed. Net immigration continued to increase until 2016, after which it decreased for several years, dropping to its lowest level in 2020 during the coronavirus (COVID-19) pandemic. Although net immigration increased significantly between 2020 and 2023, it was still at a relatively low point at the beginning of 2021. Differences between projected migration for England in 2021 and the net immigration component of the 2021 MYEs (including international and cross-border) are therefore relatively small (see Figure 3).

Boundary changes

There have been several boundary changes between 2012 and 2021 that affect the comparison of certain local authorities and counties at different points in time. Areas compared over time in this report have the same boundaries but may have been merged into larger local authorities.

Census confidence intervals

The Census 2021 population estimate for England and Wales had a 95% confidence interval of 137,200. This means that the true estimate for the population of England and Wales at Census Day is likely to lie between 59,546,000 and 59,684,000. Some of the differences between projected populations for 2021 and Census 2021-based population estimates could be accounted for by this confidence interval, but it is not possible to quantify the impact of this.

Approach to analysis

Most analyses in this report measure accuracy by subtracting the rebased 2021 MYEs from the populations projected for 2021 and calculating these differences as a percentage of the 2021 MYEs. Differences from some analyses of local authorities are reported numerically alongside numeric rebasing adjustments. Analyses assessing the impact of the rebasing additionally compare projections with MYEs rolled forward from the 2011 Census.

The mean percentage error (MPE) is used when comparing differences across different levels of geography and across different types of local authority. This measure uses the mean error (that is, the SNPPs minus the MYE averaged across multiple subnational areas) and expresses this as a percentage of the relevant population MYE. This permits a direct comparison of differences between populations of different sizes. The term "error" is used in this context to describe the difference between the projection and estimates. It does not refer to a mistake in the data. 

Some of the analyses focusing on local authorities with the largest SNPP-MYE differences and reasons for those differences are presented for the 2012-based SNPP only. This SNPP was chosen because it was most representative of changes across the whole 2012 to 2021 analysis period. 

For all analyses, an "overprojection" refers to a projected population that is higher than the MYEs and an "underprojection" refers to a projected population that is lower.

The first set of analyses explore SNPP-MYE differences aggregated to the level of England before and after the post-Census 2021 rebasing. They also assess how differences can be attributed to each component of change. 

Comparison of SNPPs and MYEs for England between 2012 and 2021 

As each new set of subnational population projections (SNPPs) is produced, both the base year and the trend data will change. Methodological improvements will also be incorporated where appropriate. Figure 1 shows each SNPP plotted against the rolled-forward and rebased mid-year population estimates (MYEs) for England between 2012 and 2021.

Figure 1: All SNPPs overprojected the population in 2021

Projected and estimated population from subnational population projections (SNPPs), rolled-forward mid-year estimates (MYEs) and rebased MYEs for England between 2012 and 2021

All SNPPs overprojected the population between 2019 and 2021. These differences increased following the rebasing of MYEs. The rebasing affected each MYE differently, increasing the population between 2012 and 2017, then reducing it between 2018 and 2021. The 2021 MYE for England was reduced by 182,198.

SNPP and MYE differences for England

Projections for 2021 were compared with both the 2021 MYEs rolled forward from 2011 and the rebased 2021 MYEs to assess the impact of the rebasing. Figure 2 shows that all SNPPs overprojected the population for 2021. These overprojections were larger when measured against the rebased MYEs than the rolled-forward MYEs. Overprojections were largest for the 2014-based SNPP and smallest for the 2012- and 2018-based SNPPs.

Cumulative differences for each component of change 

Cumulative SNPP-MYE differences arising from each of the components of change (births, deaths, net migration) between 2012 and 2021 were measured. The rebasing of SNPP base year MYEs was also calculated. Cumulative differences were calculated by subtracting each component of the rebased MYEs from the corresponding component of the SNPPs for each year between 2012 and 2021 and summing these differences.

It should be noted that an underprojection of deaths results in a larger population (and the other way around); the direction of the differences for deaths was therefore reversed to enable meaningful comparison with the other components. The SNPP base year rebasing figure was calculated by subtracting the rebased MYE for the base year from the original rolled-forward MYE used for the base year; this was a single year calculation and not a cumulative measurement. 

The bars in Figure 3 show the cumulative differences for each component and the base year rebasing difference for each SNPP, and the markers show the total SNPP-MYE difference in 2021. The combined differences for the components of change do not equal the total difference because the SNPPs and the rebased MYEs also contain other miscellaneous components. Differences for each component of change were largest for the 2012-based SNPP, but overprojections of births and deaths, and underprojection of net immigration resulted in a smaller total difference (Figure 2).

Overprojections of births were the largest differences for the 2012-based, 2014-based and 2016-based SNPPs. These overprojections reduced with each successive SNPP. An overprojection of net immigration was the largest difference for the 2018-based SNPP. This is because the assumptions used to project migration for this SNPP were informed by a spike in net immigration during the preceding years, but the projection was then compared with relatively low levels of net immigration in early 2021 during the coronavirus (COVID-19) pandemic.

Rebasing increased the base year populations of the 2012-, 2014- and 2016-based SNPPs by relatively small amounts in relation to these SNPPs’ overprojections of the population in 2021. Rebasing decreased the 2018-based SNPP base year population, contributing 12.7% of the total overprojection in 2021. This degree of rebasing will not be the same for each year of the SNPPs. This is because the MYEs were not rebased by the same amount every year but varied by year according to age profiles identified during reconciliation.

Subnational population projections (SNPPs) were then compared with mid-year population estimates (MYEs) at various levels of subnational geography. These analyses were complicated by changes to counties and local authorities between 2012 and 2021. Local authorities (LAs) were reduced in number from 327 to 309 and some counties were discontinued. All SNPP geographies were aligned with 2021 geographies prior to analysis to ensure comparability. 

Comparison of SNPPs with 2021 MYEs by level of geography 

Projections for 2021 were compared with rebased 2021 MYEs for regions, counties and LAs. Counties in the SNPPs that had been replaced by unitary authorities in 2021 MYEs (Buckinghamshire, Dorset and Northamptonshire) were removed from the county level analysis. Absolute differences were calculated so that positive and negative differences within a level of geography would not be hidden when averaged.

The MPE (mean percentage error) for each level of geography was calculated by: 

1. subtracting the MYE from the SNPP for each subnational area (absolute values) 

2. calculating the difference as a percentage of that area's MYE 

3. averaging these percentages across all subnational areas within that level of geography

Figure 4 shows that the MPE was highest for LAs and lowest for regions. It is more difficult to project the population in areas with high levels of internal and international migration, such as people moving to an area to work or study. The majority of internal moves take place within a relatively small locality and are therefore more difficult to project for LAs than for larger counties and regions.

MPEs for all levels of geography generally decreased with the duration between the SNPP base year and the 2021 MYEs, as differences had less time to accumulate. Our analysis by component for England (Figure 3) indicates that this is largely because of smaller SNPP-MYE cumulative differences for births and deaths in the 2016- and 2018-based SNPPs.

Comparison of SNPPs with 2021 MYEs by region 

The projections for 2021 were compared with the rebased 2021 MYEs for each region in England. Positive values in Figure 5 indicate an overprojection higher than the eventual MYE and negative values indicate an underprojection. Note that averaging across regions for each SNPP does not equal the MPE regional values presented in Figure 4. This is because Figure 4 uses absolute error and Figure 5 does not.

The region with the largest SNPP-MYE differences was London, which was overprojected by all four SNPPs. These differences were largest for the 2014-based SNPP. London has high levels of internal and international migration and churn, which also affects fertility and mortality rates. These factors make it more difficult to project the population.

All SNPPs also overprojected Yorkshire and The Humber, and the North East, and underprojected the North West. The size of this underprojection increased with the duration between the SNPP base year and the 2021 MYEs.

Analysis by type of LA using MPE (unitary authority, metropolitan district, non-metropolitan district, London borough) showed that MPE was consistently largest for London boroughs, peaking at 7.5% in the 2014-based SNPP. MPE for unitary authorities, metropolitan districts and non-metropolitan districts was between 2% and 3% for all SNPPs. 

Comparison of SNPPs with 2021 MYEs by size of local authority 

All SNPP LAs were aligned with 2021 geographies then absolute differences were calculated for each LA by subtracting the rebased 2021 MYE from the projection for 2021. These differences were plotted against the 2021 MYEs for each LA. Findings were broadly similar for each SNPP and are shown for the 2012-based SNPP in Figure 6.

A positive correlation was observed for each SNPP, with SNPP-MYE differences increasing with population size. Some LAs consistently featured across all SNPPs as outliers; Camden and Westminster were smaller LAs with large differences, while Birmingham and Leeds were larger LAs with small differences. 

Local authorities with the largest differences 

Differences for all LAs were calculated by subtracting 2021 MYEs from projections for 2021. These are available as numbers and as percentages of the 2021 MYEs for all LAs and for all SNPPs in the Comparison of SNPPs for 2021 with 2021 MYEs for local authorities dataset. Figure 7 shows those LAs with the largest positive percentage differences and those with the largest negative percentage differences for the 2012-based SNPP (Isles of Scilly LA was excluded from Figure 7 because of its small population).

For the 2012-based SNPP, the most overprojected LAs were mainly London boroughs, which have grown at a slower rate than was projected. The largest underprojection was for Cambridge, which was underprojected by all SNPPs. Underprojections were generally smaller in magnitude than overprojections, and the most underprojected LAs were more variable across the four SNPPs. 

Impact of the 2021 rebasing on local authority SNPP-MYE differences 

Figure 8 shows the 2012-based SNPP-MYE differences in 2021 alongside the rebasing correction (that is, rebased 2021 MYEs subtracted from 2021 MYEs rolled forward from 2011) for the most overprojected and underprojected local authorities numerically.

For most LAs, the direction and magnitude of the rebasing correction was similar to the SNPP-MYE differences, indicating that a large proportion of the differences can be attributed to methodological improvements incorporated during the rebasing process. Four of the five most overprojected LAs were London boroughs, whereas most of the underprojected LAs were in the North West and East Midlands.

2012-based SNPP-MYE cumulative differences for each component of change between 2012 and 2021 were then calculated for these LAs and plotted alongside the rebasing UPC (unattributed population change) and SNPP base year rebasing. The net migration component refers to all migration, including internal, international and cross-border migration. Results of this analysis for the most overprojected and underprojected LAs are shown in Figures 9 and 10, respectively.

For Camden, Sheffield and West Northamptonshire, the SNPP components of change contributed relatively small differences to the total, and UPC made the largest contributions.

It is important to note that UPC is not projected as a component of the SNPPs. However, UPC calculated for these LAs during the rebasing largely accounted for the gap between cumulative differences for the combined components of change between 2012 and 2021, and the total difference in 2021. By definition, the origin of UPC cannot be identified. For all other LAs shown in Figure 9 and 10, cumulative differences in net migration made the largest contributions to the total differences in 2021.

This report assessed the accuracy of the subnational population projections (SNPPs) between 2012 and 2021 by comparing these with the 2021 mid-year population estimates (MYEs).

Overall accuracy for England was highest for the 2012- and 2018-based SNPPs and lowest for the 2014-based SNPP. However, differences for each of the components of change were largest for the 2012-based SNPP. The rebasing of the base year MYEs accounted for a small proportion of the total difference for each SNPP, and a larger proportion could be attributed to the reduction of the 2021 MYEs by the rebasing. 

All SNPPs were less accurate for smaller geographical areas, particularly those in London. At a regional level, the largest difference was for London, which was overprojected by all SNPPs. The North West was underprojected by all SNPPs.

Analysis of local authorities showed that those with the largest overprojections generally had their populations reduced by a similar amount during the rebasing. Conversely, those with the largest underprojections had their populations increased by a similar amount during the rebasing. Net migration and rebasing UPC made the most significant contributions to total differences for LAs with the largest overprojections and underprojections.

We note that further analysis could be carried out with regards to the accuracy of the SNPPs (for example, by age and by sex) and we would welcome comments on further analysis that would be valuable, which can be emailed to pop.info@ons.gov.uk

Changes made to subnational geographies between 2012 and 2021 were as follows (in chronological order). 

2018

Shepway local authority was renamed Folkestone and Hythe. 

2019

Suffolk Coastal and Waveney non-metropolitan districts were merged into East Suffolk non-metropolitan district. Forest Heath and St Edmundsbury were merged into West Suffolk non-metropolitan district.

Christchurch non-metropolitan district, Bournemouth unitary authority and Poole unitary authority were merged into Bournemouth, Christchurch and Poole unitary authority.

West Somerset and Taunton Deane non-metropolitan districts were merged into Somerset West and Taunton unitary authority.

North Dorset, Purbeck, West Dorset, Weymouth and Portland non-metropolitan districts were merged into Dorset unitary authority. The county of Dorset was discontinued in 2019. 

2020

Aylesbury Vale district council and Chiltern, South Bucks, and Wycombe non-metropolitan districts were merged into Buckinghamshire unitary authority, which replaced the county of Buckinghamshire.

2021

Corby, East Northamptonshire, Kettering and Wellingborough non-metropolitan districts were merged into North Northamptonshire unitary authority. Daventry, Northampton and South Northamptonshire non-metropolitan districts were merged into West Northamptonshire unitary authority. The county of Northamptonshire was discontinued in 2021.