I am pleased (and somewhat relieved) to announce that we have now published the GLA’s 2022-based population projections.

Why are the projections so late this year?

This publication had originally been planned for November 2023 but was pushed back as a result of successive delays to the release of the necessary input data (for more information, see slides from July’s Population Statistics User Group).

We eventually produced the projections early in May – a week after the last set of required input data became available – but then placed the public release of the data on hold to prioritise delivery of the projection services we provide London boroughs to support their school place planning work.

After a busy couple of months in the run-up to the SCAP deadline, we are able to turn our attention back to our public-facing outputs. The long delay means that we are in the unusual situation of publishing our 2022-based projections after official estimates for mid-2023 have already been released.

We still have unfinished work with the 2022-based projections – outputs have been produced for local authorities and electoral wards, but not yet for Middle Super Output Areas (MSOAs). We had planned to wait to until ONS made their rebased small area estimates available to simplify the process of adapting the model for other geographies, but this release is now not expected much before the end of the year. We know that this is the most important output for a number of users who use this data as inputs to their own modelling processes and we are looking at options for bringing this work forward.

What’s new?

As well as advancing the starting point to mid-2022, there are a few changes and updates for the 2022-based projections users should be aware of.

Changes to fertility assumptions

There are two notable changes to how future fertility rates are used in the new projections:

The first is that the three published variants now differ by assumed level of fertility as well as by the period of past migration patterns on which they are based. The span of different fertility rates across the variants is quite conservative, and their inclusion is primarily to improve the utility of the projections for medium-term school place planning work.

The second is that fertility rates are now held fixed at their starting levels rather than changing over the course of the projection.
Previously, future changes in fertility were calculated by applying proportional changes in age-specific rates taken from the from the ONS National Population Projections (NPP) to the starting rates for each local authority.

On review, we found that this approach gave unintuitive results, with some local authorities seeing far larger increases in future fertility rates than others.

Future changes in overall fertility turned out to be sensitive to the starting age distribution of the rates – with areas with fertility rates skewed toward older ages seeing larger increases in total fertility rate.

This unintended behaviour stems from the way that the NPP assumptions (fertility increasing at older ages and falling at younger) were applied in the model as proportional changes to individual rates for each local authority.

International migration

International migration has been especially challenging to account for in the projections since the COVID-19 pandemic triggered a sharp drop in net inflows to the UK. While a rebound had been widely anticipated, a combination of factors have since resulted in a shattering of previous migration records.

Provisional international migration estimates for the UK are typically available at least six months earlier than the mid-year estimates covering the equivalent period. Given that the provisional net inflow of +672 thousand for the year to mid-2023 was so much higher than would have otherwise been projected based on past trends, we made the decision to take this data into account when preparing the 2022-based projections.

The UK estimates do not include any breakdown of flows by geography, age, or sex, and so we disaggregated the data based on the distributions from the previous five years of estimates. This ‘extra’ year of data is included as the assumed flows for the year to mid-2023 but does not contribute to projected trends beyond that point.

Updated housing-led inputs and scenarios

Our previous projections were labelled ‘interim’ because much of the data required to fully update the models were not yet available – particularly for smaller geographies. We have been able to update all of the inputs to the housing-led projections with data from the 2021 census and subsequently published population estimates. These updates should improve the robustness of the ward-level projections and estimates.

For the 2021-based projections we tried something new – releasing three variants of the housing-led projections that were based on a common set of demographic assumptions but different scenarios of housing delivery. The latest projections are again based on a single, common, housing scenarios, with the three variants differentiated by migration and fertility assumptions. There are several reasons for this reversion:

1. To streamline our processes – production was already complex enough this year without the need to manage multiple sets of housing assumptions as well.
2. To better align the published variants with those provided to local authority users through our projection service.
3. Limitations in the housing capacity data available to us meant that the translation of overall London-level delivery assumptions into local housing trajectories had to be cruder than we would have liked.

We haven’t abandoned the idea of providing projections that reflect a range of potential housing scenarios. Hopefully, planned improvements to housing capacity data will provide a new opportunities to take this work forward.

Outputs

The standard output files for the trend-based projections now include results for all local authorities in England and Wales. These are not new outputs as such – the data has been available in each round of projections since the 2015-based – but previously the data was hidden away in a zip archive few were aware of. This change should make it simpler to compare the results for London with those for other regions, as well as making the outputs relevant to a wider audience. [note that we will still be publishing the zip archive – this contains the full detailed outputs from the model].

Some users will no doubt already be aware of (or exasperated by) the lack of update to the Projections Explorer app following the release of the 2021-based projections. We’re aware of how useful a tool this was for many users and work is already underway to update it with the 2022-based outputs.

Some headline results

Total population

London’s population is projected to increase over the coming decades, albeit at a slower rate than in previous results. Unsurprisingly – given the slowdown since 2015 – the variant based on a 15-year period of past migration data with the high fertility rate assumption shows the highest level of future growth, with a projected 2046 London population of 10.2 million. This compares with 9.8 million and 9.3 million for the 10-year central fertility and 5-year low fertility variants, respectively.

For comparison, the results for London’s neighbouring regions span far narrower ranges.

The wide range of outcomes for London is a reflection of just how volatile population trends in London have been in recent years, with annual population change swinging from a high of +150 thousand back in 2011 to -60 thousand a decade later.

Population by age

There are stark differences in the projected trends by age group, with numbers of children projected to fall in all variants (see our recent report looking at this in more detail); working age population rising by up to 12 percent in the most optimistic variant, but barely at all in the 5-year trend variant; and the population age 65 and over projected to increase by sixty percent by 2046.

Comparisons with the 2021-based projections

Projected populations are generally lower than they were for the 2021-based projections. Much of the difference in population occurs early in the projection period and is a result of different short-term migration assumptions.

From 2025 onward, annual growth rates for equivalent variants are similar, with lower levels of natural change (the difference between births and deaths) in the latest projections, being largely offset by lower net migration outflows.

Preparation for the 2023-based projections

For the 2021- and 2022-based projections, our focus was largely on navigating the awkward post-census (and post-pandemic) period – and the myriad gaps and inconsistencies in population data that came with it. For the next round of projections, attention will shift back to the methodologies and assumptions underpinning the projections.

We expect to publish the 2023-based projections by the end of 2024 with development set to begin soon. An overarching goal we have for these projections is to improve the way uncertainty is captured in the models and communicated to users.

We are still in the process of scoping and prioritising the development work, but two items are already pencilled-in:

New methods for modelling fertility

A lack of data has always limited our options for how we include fertility in the projections. Over the last year or so, this situation has improved: with newly commissioned data from ONS, and new methods for processing these into detailed fertility rate estimates that can serve as the basis for developing and testing alternative approaches.

Predicting future changes in fertility is notoriously difficult and there is no guarantee that better data and more sophisticated methods will dramatically improve the accuracy of projections, but we should at least be able to make the process more transparent to users and provide a better sense of the range of uncertainty.

Analysis of retrospective projections

To understand how much uncertainty to expect in future projections we first need to understand how past projections and assumptions have performed. To many readers that may be an obvious statement, and many of you may (very reasonably) expect that this is something that those who produce projections and forecasts do regularly as a matter of course. However, while we do monitor and analyse differences in past projections as new estimates are released, this is often less useful than we would hope.

The problem we face is that, generally, the annual estimates we are comparing past projections to are themselves prone to significant amounts of error. This immediately creates uncertainty about how to interpret differences between projections and estimates. Further complicating things is that: because projections are built on top of these imperfect estimates, a ‘good’ projection may accurately predict future estimates, but be a poor reflection of reality. Even when comparing, e.g., projected births to official estimates (for which we have a high level of confidence) it is often unclear whether differences are a consequence of errors in numbers of potential mothers or in assumed fertility rates.

Right now happens to be a good time to undertake this kind of analysis, because:

• We have a revised series of estimates, consistent with the 2021 census, which provides us with a reliable benchmark for comparing projection results to, as well as the means to isolate and quantify the impacts of past estimate errors on the projections.
• We have detailed data from the 2021 census – the only source that allows us to check many of the assumptions included in the housing-led model, e.g., household representative rates, average household sizes, residents in communal establishments, and ratios of private households to dwellings.

Through this analysis, we hope to better understand the uncertainty in individual assumptions and model parameters, how these contribute to the overall accuracy of the projections, and what we can do to improve the robustness of the projections going forward.