New mortality assumptions method for national population projections

The Office for National Statistics (ONS) undertook an engagement exercise on a new mortality projection methodology for the national population projections (NPPs). It took place from 9 January to 20 February 2023 and was part of our strategy to continuously review and improve our methods.  

This exercise allowed us to gather feedback on the planned use of our statistics that could arise from the prospective change to the new mortality projection methodology. 

As a result of this feedback and with agreement from the NPP committee, which oversees the production of NPPs, we will adopt the new mortality projection methodology and provide guidance to users on its use in the next round of the NPPs, which will incorporate Census 2021 data for England, Wales and Northern Ireland. 

For further details, please see our response summary - new mortality assumptions method for NPPs.  

We would like to thank all respondents for taking part in the engagement exercise and appreciate the time taken to share their views. 

The engagement exercise received 23 responses in total, including six responses by email, from a range of stakeholders. These included businesses, industry bodies, central and local level government, academia, and those responding in a personal capacity. 

When we asked what your overall comments on the prospective mortality projection methodology were, the following main themes emerged. 

Methodology 

• Broad support for the new mortality projection methodology, which users said seems to be transparent and statistically sound. 

• Some users requested that more information be published on the detail of the methodology.  

• Some users valued the transparency and efficiency of using a model, while others were concerned about difficulty replicating a complex model for their own use.  

• There was a request to publish the model code and data to aid users’ understanding. 

• There is support for the added flexibility of the model to weight up more recent data and weight down more historical data; the weighting will need to be reviewed periodically. 

• The model holds up to scrutiny well when compared with similar projection models (such as those used by the Continuous Mortality Investigation | Institute and Faculty of Actuaries). 

• Further detail would be useful on how the initial mortality improvements were derived and how the cohort effects have been incorporated.  

• Discussion of the age ranges where different models are used needs to be presented clearly.  

• There are differences in the treatment of old age mortality compared with other models, but there was support for the proposal to treat mortality at older ages separately.  

Expert opinion  

• The input of expert opinion is valued, especially for mortality shocks and the uncertainty of long-term improvements. 

• Users requested more information on the membership of the expert panel.  

• Any assumptions need to be fully supported by data to show they are unbiased.  

Potential areas for development  

• It would be useful to test how the model would have performed in the past, to consider the accuracy of assumptions from expert opinion and potential for changing the weights or to publish results from past projections run using the new methodology. 

• We need to present mortality variants, for example for more and less optimistic assumptions about mortality improvement and assign probabilities to them. 

• Users told us in their requests for mortality projections for different subgroups, for example, by ethnic group or by deprivation decile, that these groups have been affected differently by the coronavirus (COVID-19) pandemic, and experience different rates of mortality improvement. 

For further details, please see our response summary - new mortality assumptions method for NPPs. 

We committed to publishing this response summary to the engagement exercise by spring 2023.  

We are now using all feedback to refine and develop the new mortality projection methodology further to meet user needs. 

We have concluded from the results of the feedback that the mortality projection methodology is an improvement compared with our current methodology and will produce outputs that will broadly meet user needs. 

We will continue to make expert advisory panel minutes and the membership available to users on request to ensure that the assumption setting process is transparent. This will include information on the evidence to support the assumed long-term rates of mortality improvement by age and sex and the speed of convergence to these long-term rates. 

We will regularly review the specification of the model including weighting and shock adjustments and will be transparent with users about any changes that we make.  

We have noted that users have suggested applying the proposed methodology retrospectively to previous runs of projections. This is a complex task, and we know that any methodology – whether the proposed model or the current method – would not project the slowdown observed in mortality improvements since 2010. Our analysis of 2018-based and 2020-based projections suggests that the changes introduced by the model are no larger than those seen as assumptions change from one projection round to the next. This was detailed in the supporting article Prospective new method for setting mortality assumptions for national population projections, UK: January 2023. The model has the advantage of using the full time series of data on population and deaths, while also having the flexibility to weight up data from more recent years if the expert advisory panel feel this would better reflect likely future trends. Rather than re-running previous projections, we plan to invest in developing better ways to explain the uncertainty around future projections.  

We have noted that users require more information about the model. We will publish detailed guidance and an updated NPP Quality and Methodology Information (QMI) report including information about the strengths and weaknesses of any projections produced and guidance on how to use them. 

We will consider how we can make the model code publicly available. The input data is largely already in the public domain. We plan to run a user engagement exercise to understand users’ needs for future NPPs, Subnational Population Projections and Household projections. This will include questions about needs for variant projections, which will inform our assumption setting for the NPPs. 

We will communicate updates through our Demography newsletter. Please email projections@ons.gov.uk if you would like to sign up to receive these newsletters. The ONS release calendar contains information on population projections releases and we will update it accordingly.