- 10:30 – Welcome and introductions
- 10:35 – Actions update
- 10:40 – Update on NS23 Recommendation
- 10:50 – Disclosure Control Proposal for Future Population and Migration Statistics
- 11:35 – Future Population and Migration Statistics Statistical Design and System Overview
- 12:15 – Break
- 12:50 – Dynamic Population Model update
- 13:20 – Coverage Estimation Using a PECADO approach
- 13:50 – Any other business
- 14:00 – End of meeting
Panel Members
- Professor Sir Bernard Silverman (Chair)
- Professor Ana Basiri
- Dr Oliver Duke-Williams
- Carl Emmerson
- Professor Nik Lomax
- Professor Natalie Shlomo
Office for National Statistics
- Professor Sir Ian Diamond
- Owen Abbott
- Sarah Henry
- Charlie Wroth-Smith
- Katie O’Farrell
- Camille Harrison
- Eleanor Law
- Amy Large
- Samantha Trace
- Alice White
- Ceejay Hammond
- Robyn Hunt
- Fiona Dawe
- Charlotte Hassell
- Shaun Davies
- Duncan Elliott (ONS presenter)
- Louisa Blackwell (ONS presenter)
- Iain Dove (ONS presenter)
- Mary Gregory (ONS presenter)
- Tom Tarling (ONS Secretariat)
- Cristina Spoiala (ONS Secretariat)
- Susan Williams (ONS Secretariat)
1. Welcome and introductions
Sir Bernard welcomed the group and confirmed those in attendance.
2. Actions Update
In Jan 2024, MARP were briefed on and discussed the National Statistician’s recommendation for future population and migration statistics (FPMS). The panel agreed to provide a statement detailing their previous reviews of methodological practice and highlight methodological issues to be resolved to deliver the vision of the FPMS. This statement was to be reviewed in this MARP meeting on the 8th February. As an action update, the panel provided an agreed response to the NS23 recommendation, with a statement to accompany the report. The panel unanimously agreed the final version. ONS secretariat will arrange for this to be published alongside the upcoming NS23 recommendation.
3. Update on National Statistician’s Recommendation for Future Population and Migration Statistics (NS23)
Following the consultation and all input this will go to the Authority Board. ONS will be submitting a business case to HM Treasury.
ONS gave an update on progress towards the recommendation. The MARP statement on the consultation will be published alongside the recommendation.
4. Disclosure Control Proposal for Future Population and Migration Statistics (FPMS) – EAP204
ONS presented the work. The Panel commented that “secure” is a relative term. Public perception is important. The public may have concerns and so there is a need for good communication and assurance to address this.
The Panel enquired if the ONS legislation covers data linkage. ONS said yes but some aspects are not entirely clear and are currently liaising with the government legal department.
The panel discussed consent. ONS noted that public engagement is ongoing and agreed to come back to MARP with a paper on consent. Public engagement will be an important aspect; emphasising people own their own data. Strong engagement with communities is important for building trust.
The Panel discussed the right to withdraw consent for people whose data is on admin data sets, and how this will affect statistics. ONS replied that it is not for ONS to decide on this yet from a methodological perspective. It is for ONS to bring up a drawn-up programme on different methodological aspects at a future MARP, as well as coming back with more clarity on the different elements, such as legal, ethical, outcomes regarding different aspects of FPMS.
The Panel discussed record swapping. ONS replied that it has carried out a pilot on differential privacy in 2020, which had utility issues arising from perturbing zeros. Epsilon values may also be high if they accumulate from outputs are at multiple geographies. Panel suggested for ONS to look at the US Census Bureau methodology on this and to further investigate perturbation in general.
Distribution of noise was discussed and differencing over time. Where values are static, perturbation being independent over time could lead to removing the protection by averaging.
Panel noted that given that populations move around a lot, dependent on geographic level, the values may not be static and so the longitudinal risk may be low. ONS said it could be possible that there is less of a risk, yet ONS does not yet have any data to confirm. ONS to look into how to deal with continuously moving small time series variables whilst keeping privacy. Where there is natural movement between categories (e.g. age) there are issues. These were raised as issues for further work in the privacy community rather than for the ONS.
Panel raised the issue that the Census table builder worked directly from microdata, rather than hypercubes. Panel suggested producing hypercubes here for the dissemination of the data instead of microdata. ONS noted that it can be an option with there being ongoing research on this.
5. Future Population and Migration Statistics Statistical Design and System Overview
ONS presented an update. The panel commented that characteristics data won’t be owned by ONS and if ONS don’t own data, there will be issues of how much can be done with the data where definitions inconsistent. ONS replies that it is aware of the huge challenges and have the ambition to resolve these. At the same time ONS is aware that survey data in some cases may be the only answer. Worth noting trade-offs, whilst we may lose some quality on a few existing measures, we may gain new items we currently cannot measure and improve quality on measures to achieve greater timeliness and granularity. Further, on data ownership, ONS commented that government is keen to support data owners more and this service will be of help to the matters raised, delivering quality improvements across organisations.
Panel expressed a need to update questions/data on characteristics where they no longer serve the original purpose and they questioned “why” ONS want to ask certain questions, e.g. “heating” question transitioning from being a deprivation measure, to being of interest to net zero; importance of getting nuance. ONS agree and note opportunities here as well as challenges. Panel commented that it can be that users trust current variables yet they themselves do not realise how imperfect those variables are. ONS added that it will be able to react to priorities much more quickly with this proposed system, rather than a census every 10 years, thus being able to identify and work on various needs in a timely manner. A huge challenge being how to prioritise user needs and room for innovation.
Panel and ONS agreed that communicating uncertainty is very important.
Panel commented on overcounts and undercounts in the demographic index and resolving these inconsistencies and these may not all be inconsistencies as they reflect that people may live in multiple locations. ONS noted that a fractional counting approach is still an option.
The panel asked if the previous census be used as a data source? ONS own data that can be used to construct ‘admin’ like data, thus could that be used in the future? ONS replied that for the Longitudinal Population Dataset, the starting point is the census.
ONS mentioned linkage doesn’t need consent or notice under current legislation, you can use it but cannot pass it on. But whilst ONS can use the census act, other government departments might not have that legal right. These are questions to explore.
Panel previously commented on expanding the vision beyond just admin data and asked if that is still in scope. ONS mentioned that this would be up to NS recommendation, but plans in general include novel data beyond admin.
6. Dynamic Population Model (DPM) update
ONS gave an update on developments and future work.
The Panel commented on the proposed publishing schedule given the admin data time lags and asked if there is a possibility for shorter intervals? ONS replied that the more of a squeeze to the schedule the less data ONS has and thus an increase on assumptions. Panel agreed that having a number less often and more accurate is preferable.
ONS clarified it does not intend to publish estimates every month – rather publish twice a year, with publications including monthly breakdowns to capture data of interest for locations such as tourism towns etc. The panel agreed with keeping it running like this for a few years and for ONS to review.
ONS noted an OSR review of DPM is due in June 2024. The ONS noted its intentions to open-source code had been well received.
The panel commented on the public video released, noting it was a good education tool.
Panel asked if ONS has seen evidence that the DPM works. ONS to come back to MARP. Publications have numbers available, but ONS has not published simulation studies, yet would be happy to come back to discuss.
Panel commented on high level benchmarks and net migration. Statistics Norway’s work on population registers would be of interest, using a Hierarchical Bayesian Model. ONS agreed it would in interesting to review.
7. Coverage Estimation Using a PECADO approach update
This item was pushed out to the next MARP meeting to allow for more discussion time and consideration.
8. Any other business
The panel discussed timing for another meeting and agreed that an additional meeting in April would be preferable. The ONS agreed and stated there would be further items coming for review within the future population and migration statistics methodology.
Following this, the chair closed the meeting.
Next meeting is scheduled for 9th of May.
With no other business, the panel closed.
Actions
ONS to respond to and acknowledge comments on the Disclosure Control paper in full.
ONS to evaluate its disclosure control options including Differential Privacy.
The ONS’ FPMS Statistical Design and Systems team to look at the current proposed system diagram as MARP want assurance on all parts of the system as set out in the diagram; breaking the diagram down is important to ensure that the whole picture is seen and that each individual part has been assured.
The ONS’ DPM team and the MARP panel to liaise on Hierarchical Bayesian Modelling.