In June this year, the Office for National Statistics held a research conference on the current Census Transformation Programme at the University of Winchester. The conference presented the latest research and current thinking behind the move to a predominantly online census in 2021 and the increased use of administrative data in future.
Plans for the England and Wales Census in 2021 are already well-developed - but it is moving into test phase with questions, outputs and methods still being refined, particularly the use of administrative data to augment counts. There is a sense that the work is going to split between the operational delivery of census 2021, particularly getting the online response, and the Beyond 2021 programme. The long-term project has a clear focus on making a recommendation in 2023 about whether the 2031 census can be achieved just by using administrative data and what form this should take.
The conference began with remarks from former RSS president and National Statistician, John Pullinger, on how welcome and appreciated those from the EU were. Indeed, there were lessons to be learned from international presentations from Italy, Portugal and Estonia: certainly the 75% online response rate target is in line with global trends.
Census primarily online
The design of the census in 2011 used the online system as a translation of the paper form, but in 2021 the census will be online first. This will enable using more exhaustive drop-down lists for topics such as ethnicity. While the full list may not go on paper, there were other mode effects of item level response seen in 2011, so this is a complexity rather than a problem.
The standard way to offer an online completion is to send out an authentication code in the post to the address, and only to provide paper if requested. However, other countries have used identifying numbers and gov.uk is developing the use of an identity service called ‘Verify’ which might be used for census authentication.
The plans to go online seem very simple, as it is just an online form, but there is a curiosity about respondents as individuals and as households. While the legal compulsion is for households to respond, there has always been an option for individuals to request their own separate form and this may increase.
Linking administrative data
Plans are well advanced to test linking in administrative data to estimate incomes. This would get around well-established biases in non-response to income questions in surveys by using tax data, although dependence on this sort of data is already threatened by changes to administrative systems - as seen with child benefit.
Of course, talking about the census being predominantly online misses the fact that the post enumeration survey (called the census coverage survey) will be fieldwork by enumerators. This independent survey checks accuracy and estimates non-response levels to improve population estimates. It is a feature of administrative censuses in many countries.
While there has been great success in matching individuals when their names are pseudonymised, the error rate of 2% is still too high to add into a triple system estimation method. However, if access to all data is granted, it would be possible to use a third source to estimate the dependence between the coverage survey and census, always assumed independent because of model saturation.
The development of an address register, a surprise to anyone who has never seen a census operation, is finally in hand with the establishment of GeoPlace, which brings together existing supplies of address data into a single database. However, that does not mean that all households are identified, a distinction very important to demographers.
The Digital Economy Bill, announced in the Queen’s Speech, offers great potential to access administrative data from other sources. Current proposals mean that ONS could access, for example, energy company data, to help identify vacant properties and number of households at an address.
In 2014, it was suggested that joint funding of academic research with the ESRC might develop census methodology and address questions about the nature of census futures. That has not emerged yet, but it seems more plausible now to think this would support plans Beyond 2021, especially as there is a time when ONS will be substantially occupied with the operational side.
There were repeated questions about whether an administrative data census can replace the enumeration and provide the data that users need in 2031. Administrative records are about individuals and how they engage with services such as health and education, as opposed to census data, which is about relations between people, addresses, housing and households. Thorny questions about this emerged, which other administrative census countries do not solve, despite accurate population counts. When looking beyond allocation of individuals to households, types of households and relations between them and the individuals in them will be more problematic.
On some topics, the census is the only source of data at present: internal migration within the country is shown by the question about where you were living 12 months ago. The number of hours spent doing things such as volunteering, in unpaid caring roles and in self-employment will also be missing. And self-rated health becomes a bit circular if it is estimated by administrative health data, for which services it was collected to plan.
It was also clear that most of the statistical work was focused on quality, ie agreement with census estimates, another circularity which was recognised, of univariate quantities. Similarly, development of administrative data sources was being driven by government departments which were also focused on univariate outcomes unaffected by isolated recording eccentricities, but census utility is in its multivariate tabulations for small areas.
Ultimately, however, there did not seem to be doubts that a predominantly online census can work, and plans to test the response in areas of low connectivity and digital literacy are in place.
The reservations expressed are issues for the future, and represent an area where the UK is placed to lead the world in research. There is considerable potential for collaboration between ONS, data providers, local authorities and academics to focus and answer these questions. But collaboration will be necessary as although all are investing in these skills, recruitment is becoming competitive.
Copies of the presentations given at the conference are now available from the ONS News and Events page under the 'Recent events' heading. You can also follow the discussion on this topic on StatsUserNet.