Panel:

• Ana Basiri (MARP representative – University of Glasgow) (Chair)
• Jon Forster (University of Warwick)
• Arkadiusz Wisniowski (University of Manchester)
• Fred Piel (Imperial College London)
• George Ploubidis (University College London)

ONS representatives:

• Nick Taylor (ONS)
• Dominic Webber (ONS)
• Joe Southam- Gisby (ONS)
• Eleanor Fordham (ONS)
• Greg Payne (ONS)
• Petya Kozhuharova (ONS)
• Pratibha Vellanki (ONS)
• Salah Merad (ONS)
• Claire Hill (ONS)

Apologies:

• Jon Forster (University of Warwick)

Welcome address and status of prior actions

The Chair welcomed the panel and proceeded to address the currently open actions.

Open actions:

Action DPMSG002: A Comparative Look Before and After DPM Modelling
ONS to revisit how inputs compare to the outputted credible intervals.

Update on action: This action will be covered under Item 2 – LTIM Coherence with DPM Outputs, presented by Eleanor Fordham at today’s panel meeting.

Action DPMSG003: Update on Aggregate Uncertainty and Coverage Adjustments
ONS to investigate why the coefficients of variation of the posterior distribution of population decrease as we move away from census years.

Update on action: ONS is exploring why population uncertainty appears lower further from census years. Early analysis suggests this may be due to model limitations, including sparse covariates and the absence of random draws for migration. To address this, the team is testing additional covariates and applying bootstrap methods to better capture uncertainty. A simulation study is underway to assess the coverage of credible intervals, with results to follow.

LTIM Coherence with DPM Outputs – Presented by Eleanor Fordham

The presentation examined coherence between long-term international migration (LTIM) which are inputs to the dynamic population model (DPM), and admin-based population estimate (ABPE) international migration estimates outputted from the DPM., The presentation identified inconsistencies in geographic coverage, methodology, and data breakdowns. Although most estimates fall within credible intervals, opposing DPM revisions to immigration and emigration figures impact net migration.

Options for addressing incoherence were discussed, including maintaining current publication practices, aligning estimates at specific geographic levels, or transitioning to DPM-based outputs once success criteria are met. The short-term recommendation is to continue publishing as-is, with reassessment planned for early 2027 based on current timescales, once planned methods developments are implemented.

The panel discussed the concept of “fixing” inputs and outputs, noting that small discrepancies can accumulate over time. There are still uncertainties that are not quantified in LTIM and therefore not considered in the DPM. There are also several planned methods developments for LTIM, which does not suggest that DPM international migration outputs should be adjusted to LTIM currently.

Clear communication was emphasised due to the policy sensitivity of migration statistics. LTIM should not be treated as a definitive benchmark. Suggestions included comparing modelled estimates with census data and conducting sensitivity testing to evaluate the impact of LTIM methods changes.

ONS plan to implement improvements to the population stock coverage adjustment in step 2 of the DPM. Limitations and potential improvements to step 3 are also being explored. ONS clarified that the short-term recommendation would be taken to an internal operations group for a decision in December.

Action:

ONS to provide an update on the December Ops group meeting on international migration coherence and discuss options for managing incoherence in more detail at the next sub-MARP meeting.

Action:

ONS to produce and share sensitivity analysis testing the impact of LTIM method changes on differences between international migration DPM inputs and outputs.

Sensitivity analysis – Presented by Joe Southam- Gisby

A five-page executive summary of the first phase of testing was shared ahead of the panel meeting, with the full 68-page report available on request. The sensitivity analysis was commissioned to meet OSR accreditation requirements for the DPM, with ONS committed to completing testing by mid-2026, as outlined in the August 2025 plan for economic statistics.

Phase One included three sensitivity tests: removing SPD stock data (which increased uncertainty in recent years), changing the distribution assumption (resulting in wider credible intervals except in 2021), and simulating a migration spike (excluded from the summary due to quality issues but detailed in the full report).

Phase Two restarted sensitivity testing due to methodological changes in the DPM.  The resource-intensive process involves multiple iterations by a small team, with future tests benchmarked against July 2025 outputs. A threat analysis approach is being used to prioritise high-impact scenarios, and cohort-based testing will reduce production time, focusing on Step 2 of the estimation process.

The long-term aim is to build a scalable sensitivity testing framework that can be repeated iteratively, ensuring the DPM remains robust and reliable across evolving scenarios and geographies.

The panel supported sharing the full report to enhance transparency and welcomed the shift to hierarchical modelling. Members encouraged further testing, with cohort-based approaches seen as practical given resource constraints—provided cohort selection reflects demographic relevance.

The panel emphasised comparing credible intervals over point estimates and highlighted the need to test assumptions around priors, including their informativeness and input accuracy. Suggestions included perturbing priors and developing a simplified “toy model” to benchmark DPM performance, particularly for aggregate vs. local authority estimates.

The panel discussed geographic sensitivity, recommending tests on aggregated estimates for selected local authorities.  Interest was expressed in revisiting simultaneous cohort estimation despite past technical challenges with feasibility to be explored in 2026.

The panel agreed that comparing DPM outputs with mid-year estimates would be valuable, given known drift over time. It was also confirmed that sensitivity analysis should align with ABPE technical requirements, incorporating both internal comparisons and external benchmarking.

Action:

Share the full 68-page report with panel members to improve transparency and understanding.

Update on plan for addressing technical requirements -Presented by Petya Kozhuharova

Petya Kozhuharova provided an overview of ongoing work to meet the Office for Statistical Regulation (OSR) technical requirements for the Dynamic Population Model (DPM). This work is being coordinated across multiple teams and prioritised due to limited resources.

A version-controlled methods guide is being developed to accompany each DPM release, documenting inputs, assumptions, priors, outputs, and code examples. Quality assurance (QA) checks are expanding, with a dashboard scheduled for delivery by March 2026. Uncertainty in Statistical Population Dataset (SPD) and migration inputs has been quantified, and internal migration systems are being refined. Further work is planned to assess errors in linked administrative data sources.

The DPM now uses a two-step Bayesian hierarchical approach. Testing has shown minimal impact from changes in distribution assumptions, and coverage benchmarking is supported through comparisons with census and mid-year estimates. Sensitivity analysis is ongoing, with automated QA and posterior predictive checks already in place.

Future work includes improving models for ages 90+, accounting for measurement error in migration rates, and refining uncertainty patterns. Simulation studies will support these developments. Broader requirements—such as stakeholder engagement, dashboard development, and capacity building—are also being addressed, with further testing and international comparisons considered post-2026.

Action:

Upload presentation slides to Confluence in PDF format to enable easier viewing and commenting.

Action:

Members to review and comment on the direction of work before the next Sub-MARP meeting in January. Any outstanding points not addressed in comments will be discussed at the meeting.

Closing note

The Chair thanked attendees for their time and contributions, confirming that all action points had been captured. She concluded by noting that the next meeting will take place in January and extended early New Year wishes to the group.

Actions:

Action DPMSG004: ONS to provide an update on the December Ops group meeting on international migration coherence and discuss options for managing incoherence in more detail at the next sub-MARP meeting.

Action DPMSG005: ONS to produce and share sensitivity analysis testing the impact of LTIM method changes on differences between international migration DPM inputs and outputs.

Action DPMSG006: Share the full 68-page report with panel members to improve transparency and understanding.

Action DPMSG07: Upload presentation slides to Confluence in PDF format to enable easier viewing and commenting.

Action DPMSG008: Members to review and comment on the direction of work before the next Sub-MARP meeting in January. Any outstanding points not addressed in comments will be discussed at the meeting.