Yn yr adran hon
- How do you get definitions off the address register?
- Who will we miss by the strong focus on online response?
- Which are you most worried about – jumping into an unknown future or holding onto an old fashioned past?
- When are you going to do an admin data coverage survey?
- Is a target response of 75% online right? What are the risks and costs if it’s 50%/90%?
- When are you going to set assumptions in stone?
- Will the census colour still be purple?
- How do you handle admin/census data about children who may live in more than one house regularly?
- Can an online census include optional further questions? A small proportion of response from 75% still gives a large sample with known weighting characteristics
- How much local authority held data do you anticipate needing? How can it be used?
- Do households really matter from a census perspective? Is an address proxy good enough?
- If some of the limitations of admin/big data prove insurmountable, do we have to get used to less data in some areas (eg small area)? Is this acceptable?
- What’s so wrong with dependent sampling? If you don’t do it you’re throwing a lot of your most useful information away?
- Which paradata will you collect with the online questionnaire responses? How can it be used to improve census estimates & outputs?
- Does UK harmonisation worry you?
- Is a modular census questionnaire feasible? It would make a lot of sense to randomise some questions ( eg health topics) to keep questionnaire short.
- If you add questions to administrative datasets on statistical grounds won’t this increase respondent burden? Should there be a system that joins up all data collection (admin and surveys) systems in the most efficient way?
- In respect of the new legislation, we will need buy-in from ministers, particularly in terms of resourcing in departments. Will ONS be pushing harder on this?
- Is there an issue with skills and the capability needed? Are the ESRC training students? Are there new types of people/skills needed?
- How will internal migration be measured in the future?
- How can we ensure communities are not excluded from a predominantly online approach?
- Is there one major change in requirement from users you foresee over the next 10 years?
1. How do you get definitions off the address register?
Non-ONS panel response
The real question to address is how do we use the information that is available and understand enough about how it is put together to match the definitions that users need. There are debates to be had around this topic. Rachel Leeser
ONS response
These questions (and others raised verbally during the sessions) touch on a subtle but key issue for census.
We really need a household frame (a list of every space in which we believe a household might be residing) but the best we have is a list of addresses.
In much of the country addresses will be a very good proxy for households - but not everywhere. The use of the National Address Gazetteer which draws upon Local Land and Property Gazetteers will mean most spaces where any type of service is provided will be included - but this is still not a guarantee of finding all household spaces in areas of complex accommodation. Additional households behind single addresses are often very difficult to identify from the street and, while there is always the opportunity for additional households to be added through checking or in the field operation, it is likely that many will be missed.
We are very aware of the potential problem here and are investigating options to get us closer to a true 'household' frame.
There may be opportunities to use admin data to help understand where there are multiple households- and these are being considered. A review of our current definitions and shortcomings and how these impact on statistical quality will be part of the quality definitions for the address index and wider thinking on the statistical design.
Nôl i'r tabl cynnwys2. Who will we miss by the strong focus on online response?
Non-ONS panel response
Possibly those aged 75 and over, international migrants who want to stay off the radar (miss them anyway), 60-74-year -olds who don't trust the internet. Piers Elias
I think this is one of the biggest issues facing the Census in 2021 – the potential bias in the non-response. Some of this may be in the foreseeable category – generally those without online access and/or skills, including some of the more vulnerable members of society. The online focus may also make it easier, or at least more convenient, for those who try to avoid contact with officialdom to either give no information, or potentially misleading information. It may also be that there is a greater (and biased) risk of missing people within addresses, again deliberately or not, particularly within more complex living arrangements and perhaps where more vulnerable people are concerned.
While imputation can compensate to a certain extent, one of the key questions is possibly what is the minimum level of information that is deemed acceptable for a response? Will the online system require all questions to be completed for all individuals or will partial responses be accepted, and if so will this lead to a greater degree of item imputation (and therefore potentially inappropriate targeting of services)? I am not aware of how far ONS have considered this issue. Rachel Leeser.
ONS response
We don't know for sure, but the 2017 test is designed to identify the characteristics of these respondents. We will use this information and information from other online services (for example DVLA services) to help us identify the areas where more support is required to complete online or complete a paper questionnaire.
Nôl i'r tabl cynnwys3. Which are you most worried about – jumping into an unknown future or holding onto an old fashioned past?
Non-ONS panel response
Jumping into an unknown future; there are many risks, not least once Censuses stop, we can never reset the estimates and errors will multiply. Pier Elias
I don’t believe that anyone is holding on to an old-fashioned past but I think that there are certainly dangers in jumping into an unknown future without looking first. I am more worried about forgetting lessons from the past that may have relevance amid the ambition and pressure, potentially from those who make demands without fully understanding the implications, to follow an untested route. To retain confidence in the statistics that are produced, I believe that an element of caution is essential rather than simple hope. Rachel Leeser.
ONS response
Neither – our (ONS) agile approach means that we can ensure that we will make the best use of new and emerging technology, processes and skills without compromising our ability to deliver.
It is important that we deliver a good quality, value for money census. In order to do this we will have to reuse and/or build on the successes of the 2011 Census whilst maximising the innovations that the future holds. This balance will evolve as the census design is developed and tested.
Nôl i'r tabl cynnwys4. When are you going to do an admin data coverage survey?
ONS response
We are planning to carry out a test of the Population Coverage Survey (PCS) as we call it in 2018. It is early days to set out what the details of the test will be and where. We will discuss details nearer the time with Census Advisory Groups and the LA groups we have set up to discuss the Admin Data Census Research Outputs.
Nôl i'r tabl cynnwys5. Is a target response of 75% online right? What are the risks and costs if it’s 50%/90%?
ONS response
75% is based on international experience and our (ONS) view of digital uptake by 2021.
There will be two main ways the public can make census returns – online or on paper. Estimating the proportion of respondents that will use each route is challenging. At this stage, our planning target of 75% online appears a challenging one; it means we are currently aspiring to achieve the highest internet response rate of the countries we benchmark with.
Access and use of the internet continues to rise; in 2015, 86% of the households in Great Britain had internet access. However in the UK in 2015, 11% of the population had never used the internet. The topic of digital inclusion is a fast-changing one and information can quickly become outdated.
We will revise our assumptions as more information becomes available for example from take up of other digital services, responses to the 2017 Census Test and online response rates from Censuses in other countries. The public’s attitude to digital completion and our ability to drive it up is a key priority for testing over the coming years.
It is impossible to accurately predict the mix of response channels; testing and other evidence can help but this will be determined by public behaviour. This means with a multi-channel response strategy we have to plan for more than 100% capacity between the different channels and build in contingency. Getting high overall response rates is the most critical factor so we will have extra online capacity and extra paper processing capacity to cope with greater or lower numbers of online responses than expected. Our aim is to make it as easy as possible for households to ‘self-respond’.
Nôl i'r tabl cynnwys6. When are you going to set assumptions in stone?
ONS response
As we (ONS) progress with our testing we will replace our planning assumptions with decisions. We need to have a clear view of our plans to inform procurement activity over the next 12 months, and have made many of our decisions (on content and the question set) by Spring 2018 when we will submit the White Paper to Parliament.
Nôl i'r tabl cynnwys7. Will the census colour still be purple?
Non-ONS panel response
Hope so - I think the 2011 promotion material was eye-catching and worked well. Piers Elias.
ONS response
Maybe – we (ONS) will commence branding work on the 2021 Census soon.
Nôl i'r tabl cynnwys8. How do you handle admin/census data about children who may live in more than one house regularly?
Non-ONS panel response
Where they live more of the time. If equal, then where they are on Census night. Piers Elias
I have for a long time advocated a system where people are allowed to be counted in more than one place. The questions that follow from that premise are whether people (both children and adults) are then partially counted in each place so that the total is the sum of the parts, or whether all people are counted wherever they live, and that is likely to mean that the total population is less than the sum of the population in different areas. What really matters is that we have the best numbers for the plans and services that are being provided, both nationally and locally. In an increasingly complex world, we may need different sets of numbers to meet each purpose – different population groups and different population definitions. Rachel Leeser.
ONS response
Census data: we collect the information on where they live, which may include more than one address. The 'usual residence' base is taken to be where they spend the majority of their time, and if this is still judged by the respondent to be split 50/50, usual residence is taken to be where they were staying overnight on Census night. Outputs are created mainly for the usual residence base, but we also output tables relating to the second address.
Admin Data Census: We use a rules based approach around different addresses recorded on admin data. We set out what we did for the research outputs last year in our paper accompanying the research outputs. Essentially we allocated people with different addresses to two admin sources such as the CIS and Patient Register, half to one area and half to another. This year, we will make use of school census data as well, probably using the address supplied there as the best address for children recorded on the School Census. In future we will also look to use the Population Coverage Survey as well as more administrative activity type data but in essence, like the 2021 Census, we are aiming to produce the best usual resident base estimates.
Nôl i'r tabl cynnwys9. Can an online census include optional further questions? A small proportion of response from 75% still gives a large sample with known weighting characteristics
Non-ONS panel response
Good idea - and extending that, allowing the census record to be used for future updates to record changes beyond 2021 - using hi-tech biometrics for access. Piers Elias.
ONS response
Running an online census presents ONS with a range of opportunities, including those noted above. These also present a set of challenges (including ensuring the burden of the census on the public and the complexity of the census operation, are kept to a minimum) whilst allowing ONS to deliver outputs which meet the identified information requirements and maximise the benefits of the census. Any developments and proposals in this space will need to be included in our proposals in the Census White Paper which will be published in 2018.
Nôl i'r tabl cynnwys11. Do households really matter from a census perspective? Is an address proxy good enough?
Non-ONS panel response
Yes households matter and no an address is not good enough, as you need separate liveable spaces, not mail drop-off points. Splitting up houses will outpace Royal Mail. Piers Elias.
ONS response
The ONS Topic consultation showed that there are strong needs for information on households, and relationships within households. This is not the same as addresses in some cases. It will be in most cases, but as with a lot of census data, it is the richness of information about small populations that meets many user needs that can't be met from alternative sources.
Nôl i'r tabl cynnwys12. If some of the limitations of admin/big data prove insurmountable, do we have to get used to less data in some areas (eg small area)? Is this acceptable?
Non-ONS panel response
Yes, less data is inevitable but it’s not acceptable - not if you want to deal with inequality and do it effectively. Piers Elias.
(See question above on definitions and address register). The real question about this is why households matter from a data user perspective. What purpose does the ‘household’ serve that is different to the family or all people at an address? As when discussing population definitions, the important thing is that the numbers fit the purpose. The reason for needing numbers of households or information about the characteristics of households needs to be identified to make sure that the definitions fit the need, or at least that the differences are understood. It may be that ‘household’ is a redundant concept or it may be that the definition needs to be changed to fit the purpose. Equally, we need to define not just the number, but the characteristics of ‘households’ that are relevant to produce statistical output that are useful. Rachel Leeser.
ONS response
We are aiming to replicate many of the Census outputs using admin data and surveys, to compare with the results from the 2021 Census. An Admin data census will also offer the opportunity to integrate data on topics not traditionally asked in a census (eg housing affordability and fuel poverty), which would be more flexible and responsive to changing needs. Outputs can also be produced much more frequently than the Census.
The key trade-off will still be between the degree of detail that is possible to be produced by an admin data approach and the frequency of outputs as it was during the Beyond 2011 Programme.
By 2022, we are aiming to have made much more use of administrative data than we were able to during the Beyond 2011 Programme, not only in improving the accuracy of admin data based population estimates but also in helping to produce more accurate statistics on households, housing and characteristics of the population using administrative data and surveys. We will be publishing an increasing range of research outputs each year and annual assessments as to where we think we will be by 2023, when we will carry out a consultation before making a recommendation about whether we think we are ready to switch to an admin data based census or not.
Nôl i'r tabl cynnwys13. What’s so wrong with dependent sampling? If you don’t do it you’re throwing a lot of your most useful information away?
ONS response
We will revisit this (including the associated ethical and possible legal) issues in designing the PCS (see response to question 4).
Nôl i'r tabl cynnwys14. Which paradata will you collect with the online questionnaire responses? How can it be used to improve census estimates & outputs?
ONS response
This requirement is still being defined and will be developed following the 2017 Census Test and from international experience. Suggestions on how paradata might be used are welcome.
Nôl i'r tabl cynnwys15. Does UK harmonisation worry you?
Non-ONS panel response
From a Local Authority perspective, not really - lower level data is of more interest than higher level. However, it is much easier to compare if everything is on the same basis. Piers Elias.
ONS response
The National Statistician chairs the UK Census Committee which is charged with ensuring UK harmonisation. We (ONS) are confident that we can deliver a coherent UK Census.
We need to provide key statistics at UK level, and it is important to have harmonisation in methods and outputs as far as possible. The UK Census Offices have published an approach to harmonisation - The Conduct of the 2021 Censuses in the UK
Nôl i'r tabl cynnwys16. Is a modular census questionnaire feasible? It would make a lot of sense to randomise some questions ( eg health topics) to keep questionnaire short.
Non-ONS panel response
I think health topics may be of more interest next census - however, the beauty of a census is that is (in theory) a 100% sample of everything. Piers Elias.
ONS response
See response to question 9.
Nôl i'r tabl cynnwys17. If you add questions to administrative datasets on statistical grounds won’t this increase respondent burden? Should there be a system that joins up all data collection (admin and surveys) systems in the most efficient way?
ONS response
There are things that are badly covered at the moment that should be better covered in admin data (eg ethnicity in health services). The new Data Sharing Legislation (assuming it becomes law), should provide us with the opportunity to better address this, and we should be able to ask questions in a better way - improving both quality and completeness, rather than asking more questions.
Nôl i'r tabl cynnwys18. In respect of the new legislation, we will need buy-in from ministers, particularly in terms of resourcing in departments. Will ONS be pushing harder on this?
Non-ONS panel response
This isn’t just an issue for ONS or statistical institutes. There are also resource barriers in government departments and infrastructure problems, so there needs to be more thought around compatibility and we need to keep working on this. Vanessa Cuthill.
We will need buy-in on a long term basis, not just from the current ministers. Rachel Leeser.
ONS response
There are barriers in respect of legislation, practicalities and cost. This is a step in the right direction to remove the first barrier but ONS will be pushing this further.
John Pullinger and Heather Savory (Director General for Data Capability at ONS, are also providing strong leadership on this not only with ONS but across Government (including with Government Digital Service (GDS).
Nôl i'r tabl cynnwys19. Is there an issue with skills and the capability needed? Are the ESRC training students? Are there new types of people/skills needed?
Non-ONS panel response
ESRC provides training through the ADRN based on the skills needed to access data free of charge. PhD students are using health data and others are using Big Data – there are cohorts of students going through the centres for doctorates. We recognise the need for this and there is more resource going into it. Vanessa Cuthill.
There are lots of data scientists currently working for the Greater London Authority and many LA’s and many people want to learn these new techniques. Rachel Leeser.
ONS response
ONS has lots of researchers with the Census Transformation Programme, but we are struggling to recruit at graduate level at the moment. However, we are working closely with ESRC, ADRN and universities to identify what’s required and to spread the learning. We have good collaboration at a certain level, but we are struggling to recruit.
Nôl i'r tabl cynnwys20. How will internal migration be measured in the future?
Non-ONS panel response
This is broader than just migration statistics. Commuting statistics also cause problems and concern about the use of mobile phone data and the assumptions around data clusters to show where people are working. We cannot assume that because someone goes somewhere once a week that they are working there and we have to be careful about using data that is not designed for purpose, but we must use it where we can. Society now is much more fluid and we need to be clear on what we mean by the term ‘usually resident’ going forward. Rachel Leeser.
ONS response
Our intention is to replicate as many census type outputs as possible, to compare with the 2021 Census outputs but the detail of this is currently unknown. Producing internal migration at LA and LSOA level should be possible in theory, but we will need a coverage survey to make this better. It is also dependent on activity data and whether we will have access to this. There is still no guarantee that we will switch to an Admin Data Census after 2021. We are currently working towards a recommendation in 2023.
Nôl i'r tabl cynnwys21. How can we ensure communities are not excluded from a predominantly online approach?
Non-ONS panel response
There is concern that by going online, certain people will be excluded and it could give people a further opportunity to avoid completion, so how do we get information from them, as these are potentially the people who need the most help. Rachel Leeser.
ONS response
There will always be the issue of people avoiding completion and we have always had this problem. Part of the challenge is to access those populations, but there is a lot of work going on to assist with this. I believe that going online will change the bias, rather than introduce it.
Nôl i'r tabl cynnwys22. Is there one major change in requirement from users you foresee over the next 10 years?
Non-ONS panel response
Academics will be grateful for detailed information on income. Census outputs take too long but a live response online would be good to see. Those working with the data know what’s happening, but for outsiders it’s very difficult for us to find out how the data is being processed, so it would be good to see what’s going on. Humphrey Southall.
We are increasingly under pressure to show value for money so the timeliness of outputs is very important as well as UK comparability. The devolution agenda means that data will be fragmented. Vanessa Cuthill.
There is not just one thing that will change, but lots of things, due to a changing society. Keeping track of changes in real time is key. Rachel Leeser.
The non-resident population. Piers Elias.
Nôl i'r tabl cynnwys