• 10:30 – Introduction
• 10:35 – Secretariat update
• 10:50 – EAP222: Latest Developments on the Longitudinal Population Dataset 
• 11:50 – EAP223: Update Census 2031 Evaluation Criteria: Topic Consultation  
• 12:20 – Break
• 13:00 – EAP224 – Admin-based Household Estimates
• 13:45 – Census 2031 – Innovation and Plans for 2027 Testing
• 14:15 – Statistical Disclosure Control – Powerful Computing
• 14:30 – Any other business and close

Panel members

• Sir Bernard Silverman (Chair)
• Oliver Duke-Williams
• Nik Lomax
• Natalie Shlomo

Other attendees

• Kevin McCafferty (Presenter)
• Claire Bloss (Presenter)
• Louisa Blackwell (Presenter)
• Orlaith Fraser (Presenter)
• Michelle Waters (Presenter)
• Ainslee Woods (Presenter)
• Niamh Walshe (Presenter)
• Iain Dove (Presenter)
• Owen Abbott
• Stephan Tietz
• Ann Blake
• Sally Myles
• Nick Taylor
• Ahmet Aydin
• James Jackson
• Tom Tarling (Secretariat)
• Susan Williams (Secretariat)

1. Introduction

1. Ana Basiri gave her apologies.
2. The chair announced that, following the departure of Carl Emmerson from the panel, approval is being sought from the (acting) National Statistician for an open recruitment for an expert in economic statistics.

2. Secretariat update (Tom Tarling)

1. ONS presented updates on actions, and activities of subgroups since the previous meeting.
2. Action 114: ONS to produce and provide to the panel an explainer on papers covering the topic of Multiple System Estimation.
   • ONS provided the explainer to the panel in September 2025, and it was considered via correspondence. Following discussion and advice, this action can be closed.
3. Action 119: ONS to structure and organise running of subgroups, integrating with the main panel to allow for review and sign-off.
   • Following the previous meeting, Ana Basiri and Nik Lomax have chaired the Dynamic Population Model and Migration subgroups respectively. Processes for the subgroups have been brought in line with the main group. This action can be closed. The panel will review the subgroup terms of reference annually in line with the annual report.
4. ONS then presented a forward agenda of topics for 2026.
5. Action 120: The chair requested the panel discuss the ONS Survey Strategy, adding this topic to the forward agenda.

3. EAP222: Latest Developments on the Longitudinal Population Dataset (Kevin McCafferty, Claire Bloss, Louisa Blackwell)

1. ONS presented slides.
2. The panel suggested that the address and demographic index built for the LPD be compared to the Reference Data Management Framework (RDMF).
3. Regarding births and deaths, the panel agreed that full event date is important to allow enhanced analysis. Identification risk was discussed with the panel noting that more complex methods of confidentiality protection are preferable to simple methods that overly restrict data.
4. ONS highlighted that Census under-enumerates babies under 1 year of age and potentially has over-coverage of people aged over 90. The panel commented that online Census responses could cause respondents to fill in forms ahead of the census date, leading to subsequent births and deaths not being captured. ONS said it would consider data on submission date for the census forms and explore these coverage issues further with a view to further enhancing the coverage estimation methodology.
5. ONS noted some of the quality challenges with using the various admin data. An emerging issue was that of deaths occurring abroad that were not registered, leading to elderly individuals’ administrative records not being updated. The panel suggested researching what proportion of overseas deaths are registered, and how many had officially emigrated.
6. The panel encouraged ONS to be forthright highlighting issues with data collected by other administrative sources, and be aware of potential future changes to administrative data available.
7. The panel also noted the plans for a 2027 census test by ONS.
8. Action 121: The panel suggested the 2027 test online questionnaire is made available and kept for research purposes.
9. Responding to cross-validation work presented by ONS, the panel advised ONS to be very clear with its definition of cross-validation, as the term is used differently across literature.
10. The panel thanked ONS for the presentation and paper.

4. EAP223: Update Census 2031 Evaluation Criteria: Topic Consultation (Michelle Waters)

1. ONS presented an update on actions from a previous item, EAP220.
2. ONS reintroduced the approach to topic evaluation – separate scores to be derived for user requirements, operational requirements and other considerations. Relative weighting of each element of user requirements has now been done using Analytic Hierarchy Process (AHP) as per panel recommendation.
3. The panel questioned how the three scores would interact. ONS responded it would score on an impact scale based on user needs, and balance these against operational requirements and other elements.
4. The panel encouraged ONS to be transparent on the decision making process, especially where topics with a high user need are excluded. ONS agreed transparency was important, and noted the 2021 Census had a decision making flowchart, which could be revisited and considered for 2031.

5. EAP224: Admin-based Household Estimates (Ainslie Woods, Niamh Walshe)

1. ONS presented the paper.
2. In addition to the paper, ONS presented a table comparing the Admin-based Household Estimates (ABHE), Labour Force Survey (LFS), Admin-Based Occupied Address Dataset (ABOAD) and Census estimates for households.
3. The definition of a household was discussed with the panel highlighting how the definition from the 1950s is not applicable to modern households. ONS agreed, and noted discussions regarding household definitions in relation to Census were planned. The panel stated the definitional differences could be responsible for some of the differences between census and administrative datasets considered.
4. For the category of households with 5+ occupancy, the panel questioned if the classification of communal establishments (CEs) could be responsible for some of the observed differences. ONS responded that unique property reference numbers (UPRNs) are treated the same way in administrative data as in the census, though the LFS doesn’t include CEs. ONS agreed this assumed correct classification, and there could be cases of very small guest houses, and halls of residence broken into households, not being consistent.
5. ONS outlined planned future work on this topic, including engaging with users to understand needs, work to improve the Admin-Based Living Arrangements Dataset (ABLAD), and use of the Transformed Labour Force Survey (TLFS).
6. The panel thanked ONS for the paper and responses.

5. Census 2031: Innovation and Plans for 2027 Testing (Orlaith Fraser)

1. ONS presented slides and a timeline of plans.
2. The upcoming 2027 test proposed would compare treatment effects on response, such as different questionnaires, SMS reminders, and approaches to field staff training.
3. The panel discussed the length of questionnaire and questioning whether there were differences in drop-out rate for online vs paper based response. Online responses allowed for partial response, whilst paper responses tended to be all or nothing.
4. Agentic AI was discussed, and its possible future evolutions and influence on field staff. The panel noted agentic AI could cover language requirements well.
5. Agentic browsers were also discussed, where respondents may utilise their own agentic AI to respond to the online form. Patterns in responses from such use would need to be understood.

5. Statistical Disclosure Control (SDC): Powerful Computing (Iain Dove)

1. ONS presented slides, discussing plans considering future risks.
2. ONS focussed on the potential of quantum computing to enable usage of algorithms that current computing cannot do. New arithmetic approaches could allow breaking of SDC methods including reconstruction attacks and unpicking perturbation. Repeated tests in 2030 would consider technological advances, allowing for recommendations on 2031 Census SDC methods.
3. The panel agreed it was sensible looking forward at reconstruction attacks, and recommended further exploration of differential privacy and record swapping. It also noted the risk of identity disclosure, which wasn’t linked to reconstruction attacks.
4. It was suggested that AI be considered beyond its ability to combine ONS tables. With a wealth of administrative data available, AI could join administrative data onto census data to break disclosure controls. Much of this data won’t be within ONS control, and protecting with only ONS data in mind is insufficient for disclosure prevention.
5. The panel suggested the risks of external data being joined to census tables should be included on the risk register of ONS.
6. Action 122: ONS to include discussed risk to disclosure control on the risk register.
7. As discussed previously, agentic AI browsers should be considered, as there was a possibility an AI could use data an authorised user has to combine to external datasets and break disclosure controls.

5. Any other business / Close

1. Future meetings were discussed. It was suggested an annual, longer meeting in-person could allow for a mixture of paper items being considered, and workshops and discussion items.
2. The chair closed the meeting.

Actions:

Action 120: The chair requested the panel discuss the ONS Survey Strategy, adding this topic to the forward agenda.

Action 121: The panel suggested the 2027 test questionnaire is made available and kept for research purposes

Action 122: ONS to include discussed risk to disclosure control on the risk register.