• 10:30 – Introduction (Sir Bernard Silverman)
• 10:35 – Secretariat and Subgroups Update (Tom Tarling)
• 10:50 – Census and Population Statistics Update & EAP232 – Mid-Year Population Estimates Improvement Plan (Mary Gregory, Nick Taylor)
• 11:45 – EAP231 – Census 2031 Key Population Groups: Identification framework and prioritisation method (Jess Bloomfield, Simon Watkins)
• 12:45 – Break / Lunch
• 13:30 – EAP230 – Geospatial machine learning for intercensal small-area population estimates (Michael Hawkes)e
• 14:30 – EAP227 – Census Question Testing Framework (Karina Williams)
• 15:00 – Any other business and close

• Sir Bernard Silverman
• Ana Basiri
• Oliver Duke-Williams
• Jakob Schneebacher
• Natalie Shlomo

• Mary Gregory (Presenter)
• Nick Taylor (Presenter)
• Jess Bloomfield (Presenter)
• Simon Watkins (Presenter)
• Michael Hawkes (Presenter)
• Steve Denman
• Nigel Henretty
• Karina Williams (Presenter)

• Owen Abbott
• Tom Tarling (Secretariat)
• Dale Williams (Secretariat)
• Susan Williams (Secretariat)

1. Introduction (Sir Bernard Silverman)

1. Sir Bernard welcomed Jakob Schneebacher to the panel. Nik Lomax sends apologies.

2. Secretariat Update

1. ONS presented updates.
2. No actions were open following MARP 41. It was noted the annual report publication would be upcoming.
3. The procedure for submitting and commenting on papers ahead of meetings was reviewed to ensure it was suitable for authors and reviewers.
4. The panel asked for an update on progress with the Labour Force Survey and its transformation, which ONS provided. The panel requested a summary paper of progress and direction for the TLFS be brought to MARP for future review.

Action:

Secretariat to draft the annual report and circulate to members and chair for initial comment. Subgroup activities are to be included in the report.

Action:

ONS to provide TLFS information to the panel, such as via a summary paper.

3. Census and Population Statistics Update

1. ONS presented an update on Census and Population Statistics, covering recent publications on Migration, and Population Projections. ONS also published a decision to publish official data on mid-year estimates using established methods rather than using admin-based population estimates from the Dynamic Population Model (DPM). The DPM MARP subgroup was reviewing their terms of reference and scope.
2. Population Statistics would publish a plan later in 2026 with a focus on continuous improvement with both short and long-term targets. The Recommendation for Census would be getting an update, and mid-year estimates were a high priority for ONS. A Data Maturity Framework discussing suitable and sustainable administrative data was to be developed.
3. The 2027 Census Test was discussed. A research plan for Census would be coming to MARP. The panel urged ONS to deliver the research plan as soon as it could, to maximise time available.

4. EAP232: Mid-Year Population Estimates Improvement Plan

1. ONS presented the paper, and confirmed alternative approaches to producing mid-year populations estimates (MYE) would be brought to MARP for assurance.
2. The panel stated the continued development of the Statistical Population Dataset (SPD) was important, and queried the role of AI in the process. ONS answered it was considering AI for topics including data linkage. The panel commented that a tipping point with AI would come where it would become difficult not to include based on current progression rates.
3. Backseries of data were discussed, with ONS highlighting the impact of revisions to the backseries. Revision analysis would be performed, but there was not always administrative data available for the backseries.
4. Discussion moved onto Key Performance Indicators (KPIs) and Prioritisation. The panel suggested using Red-Amber-Green (RAG) ratings as an alternative. ONS replied it uses RAG for other targets, and could broaden out these KPI to such a system. ONS agreed that thresholds and requirements needed to be discussed.
5. The panel said it was satisfied with the approach, with ONS factoring in their comments. The panel recommended comparing objectives to other National Statistical Institutes (NSIs) and United Nation working groups. It reiterated a need for clear and accessible KPIs.
6. ONS agreed and said next steps included bringing a paper to MARP, and the relevant subgroup on the topic.

Action:

ONS requested a statement on the review of this paper from MARP to accompany publication. Following the meeting the panel agreed this statement:

MARP reviewed the high-level plans for improvements to Mid-year estimates and felt assured that the focus is in the right areas to address the challenges with population estimates. The success criteria sufficiently set out how ONS aims to meet these challenges whilst demonstrating how improvements better meet the standards of Trustworthiness, Quality and Value outlined in the UKSA Code of Practice. MARP expects to play an assurance role in reviewing the data and methodological developments as the plans progress.

5. EAP231: Census 2031 Key Population Groups: Identification framework and prioritisation method

1. ONS presented the paper.
2. The panel agreed that Key Population Groups (KPGs) needed bespoke approaches.
3. ONS discussed locations that were difficult to enumerate, and testing new approaches to contact people, particularly students where accommodation did not have traditional post boxes. The panel suggested email be a primary method of contact in such situations. ONS replied there are quality issues with email addresses, and privacy issues with institutional email addresses.
4. The panel commented there would be a continued decrease in importance of post across all population groups as years continued, with research necessary.
5. ONS advised it was using the UN Economic Commission for Europe (UNECE) conceptual framework for hard to reach groups. The panel noted the framework was designed for voluntary surveys rather than for a compulsory census. It recommended ONS review it for differences in approach.
6. ONS agreed, and noted it was meeting with other NSIs to compare approaches.
7. Further discussion covered the possibility of creating new KPGs for overlapping groups and the use of a matrix system to classify group risk.
8. The panel warned there was a risk of misunderstanding groups when using their 2021 data due to altered behaviours during Covid. ONS replied this was a known issue, with gaps in data identified. Other teams in ONS were reviewing these areas.
9. The panel suggested to consider behaviour in longitudinal studies rather than 2011 Census data. ONS stated it was seeking to learn from upcoming censuses in Canada and Ireland. The panel agreed with this but noted trust and behaviour would vary by geography, and so wasn’t directly comparable to the UK.
10. Discussion on prioritisation moved onto cost-benefit analysis, alongside inclusive data needs. The panel emphasised that the elasticity, size and cost for a marginal score change in groups was important for consideration, and that intersectionality would result in improvements for a target group also benefitting others.
11. The panel also noted where ONS identified a lack of evidence, this was evidence for a KPG needing further research.
12. The panel were encouraged by the paper and thanked ONS for their presentation and responses to comments, and requested further information in the future. ONS noted this was only a first step.

6. EAP230: Geospatial machine learning for intercensal small-area population estimates

1. ONS introduced and presented the paper, following up on EAP210 – Geospatial methods for Small Area Population Estimates: Proof of concept brought to the panel on 20 August 2024.
2. ONS asked the panel for their consideration of the integration of geospatial machine learning and the ratio change method. The panel wondered if combining the two models had a risk of compounding their weaknesses. ONS agreed that further tests were needed, demonstrating added value of combinations. It also noted some domain effects between geographic levels.
3. The lack of high-resolution data in prior years was discussed, with the panel noting that resulted in there being no point of comparison for gridded estimates. This caused unknowns with temporal instability.
4. The panel was concerned the methods lacked uncertainty measures. ONS agreed it was a topic requiring further investigation, and should be compared to other statistical models. The panel also noted there were issues with inference for machine learning models.
5. The panel commented the overall approach was sensible, and that it would like to see scenario testing added. The panel recommended the approach was reviewed by independent auditors, as it noted the technical assurance from WorldPop and the University of Southampton wasn’t independent.
6. Next steps were discussed, with ONS explaining further work was occurring in Methodology and Quality within ONS. It clarified this work was not part of Census 2031, as the method was estimating for inter-census years.
7. The panel thanked ONS for the paper.

Action:

The panel believed after reviewing EAP230, that it would be relevant to a subgroup of MARP.  Future developments would be of interest to the subgroup.

Action:

The panel requested some further information on temporal instability. ONS would send on for review via correspondence.

7. EAP227: Census Question Testing Framework

1. ONS presented the paper, covering a questionnaire framework, and a risk register for census question design. The framework discovery phase included focus groups, cognitive interviews, cogability and pop-up testing.
2. Flexibility for new methods was discussed, with ONS responding the framework allowed for newly identified approaches. This allowed emerging topics such as AI to be considered.
3. The panel noted this was a difficult topic – the question of who a ‘user of census’ was would affect what is in scope for the framework. The panel also believed some decisions would be dictated by policy. It suggested the framework offer multiple options for questions, allowing flexibility when high level decisions changed directions and scope.
4. The panel discussed the need for careful design of focus groups and recommended a minimum and maximum size to be set. The panel also commented that concepts and topics might not translate well to some languages and that caution was needed. Discussing mode, ONS confirmed it considered both online and paper responses to questions.
5. The panel commented ONS needed to provide rationale where it deviated from approaches adopted by other NSIs. The panel recommended ONS work with devolved nations, as there were benefits to cohesion in questions. ONS agreed, but recognised the devolved nations were able to make their own decisions.
6. The panel asked what the success criteria for questions would be, and discussed difficulties with comparability over time, evolving understanding of questions, length of questionnaire and response burden. Modernising questions would be a topic for research.
7. The panel thanked ONS for the paper and discussion.