31 July 2001
Symposium on Global Review of 2000 Round of
Population and Housing Censuses:
Mid-Decade Assessment and Future Prospects
Department of Economic and Social Affairs
United Nations Secretariat
New York, 7-10 August 2001
Statement from United Kingdom *
Andy Teague **
1. The 2001 census in the United Kingdom was carried out on 29 April 2001. Under Census and Devolution legislation there were, in fact, three parallel censuses conducted in the United Kingdom: one in England and Wales by the Office for National Statistics (ONS), one in Scotland by the General Register Office for Scotland and one in Northern Ireland by the Northern Ireland Statistics and Research Agency. The statements below are made with respect to the census in England and Wales, but in many respects they reflect the position in all three parts of the United Kingdom.
2. It is early to draw definitive conclusions of how well the 2001 census went, but nevertheless some important assessments can be made to help shape future censuses in the United Kingdom and elsewhere. The census was conducted on time and early indications are that the coverage (proportion of the population counted) should be at least as high as in 1991 (98 per cent), if not better. This is despite carrying out the census during the height of the foot-and-mouth disease outbreak and in the run-up to a general election. The key measurement will be, however, how well the differential undercount experienced in 1991 has been reduced.
3. Statements are provided with respect to each of the issues on the agenda. Further details on the United Kingdom censuses are available at www.statistics.gov.uk.
4. The census in England and Wales is carried out under the 1920 Census Act and the 1991 Census Confidentiality Act. These acts also apply to the census in Scotland. Similar legislation applies in Northern Ireland. Separate primary legislation for each census was required for censuses conducted in the United Kingdom before 1920.
5. The Census Acts permit the government of the day to hold censuses at intervals of not less than five years. In practice, full censuses have been held every 10 years since 1801 with the exception of 1941.
6. For each census, two pieces of secondary legislation are required to be passed by Parliament, a Census Order and Census Regulations. This legislation sets out the questions to be asked, who has to answer them and the arrangements for conducting the census. They permit a temporary field force of enumerators and field managers to be recruited. For the 2001 census, Census Order and Regulations were made and approved during 2000.
7. The census is compulsory with the exception of the question on religion.
8. The final approval of the census therefore rests with Parliament. However, in leading up to this process, it was important that all known stakeholders were consulted and their views taken into account. This started in 1995, when a series of working groups was established with key census users in central and local government and in the academic and private sectors. Three working groups were established to consider content of the census (questions to be asked), population definitions and output. The first two were key to shaping the census proposals, upon which more formal consultation was carried out with ONS’s established census advisory groups—one for each census-user sector.
9. During 1997 and 1998 formal “business cases” were sought from the key census users as to their needs in respect of the questions to be asked in the census. Other potential census users were also invited to make submissions. This was done so that the final proposals would be based on sound needs—that is, the information was:
§ Vital to the process of resource allocation from central to local government;
§ Key to the development of government policy at a local as well as national level;
§ Needed to help plan local services; and
§ Not available from any other source.
10. Parallel to this was a question-testing programme whereby all new and substantially revised questions were tested with the public. This enabled the most important stakeholders of all to be involved: those who would complete the census questionnaires. Over 40 question tests were carried out in a variety of ways, from cognitive research to a large-scale test of 100,000 households in 1997 and a census rehearsal of 150,000 households in 1999.
11. A key part of the question-testing programme was to assess the burden placed on the public. Despite the fact that the census is compulsory, the census offices need the cooperation of the public. It was vital to this end that questions were seen as acceptable and relevant and were framed in such a way that people were able to understand them and complete them accurately. One of the key findings from this process was that a question on income was not acceptable despite an overwhelming requirement from census users for it to be asked. On the other hand, a question on religion was seen as acceptable and a means of inclusivity.
12. A draft package emerged, which was then put to ministers to make final proposals in the form of a White Paper to Parliament in March 1999; it was subsequently approved by Parliament in 2000.
13. The process of consultation on the whole worked very well, although at times it needed to be stressed that final decisions did not rest with census users but with ministers and Parliament. The final package of questions did all have demonstrated needs. The only question with a strong need that was rejected was that on income, for the reasons stated earlier.
14. Although the process did identify all main census users, there were some complaints from other potential users and pressure groups that they had not had the opportunity or did not know how to contribute so that their voices could be heard in time. It is always difficult with an exercise such as the census to keep up to date with change, as arrangements need to be made well in advance.
15. A particular example of this was a request very late in the day for a Welsh tick box to be included in the question on ethnic group in Wales. This had not emerged at all during the extensive consultation programme, but with devolution of some powers (but not the census) to the Welsh Assembly, a single “British” category became less acceptable. This proved quite difficult to handle in the months leading up to the census—the census had, after all, already been approved by the Westminster Parliament.
16. One possible approach is to produce a set of initial proposals—a form of “Green Paper”—which would be formally published by the government and upon which reactions and business cases would be sought. It should be noted, however, that in the period between the publication of the White Paper until well after parliamentary approval, there was very little reaction to the government’s proposals. On the contrary, there was support as demonstrated by the ease with which legislation was passed by Parliament.
Of demographic and social statistics and
maintaining census-related activities during intercensal years
17. In 1992, the United Kingdom census offices carried out a policy evaluation and reappraisal. This looked at the needs for census-type information over the coming decades. It concluded that there was still a requirement which would not be met by alternatives, and following a review of the possible options for collecting the information, that a traditional census was the best way to collect the information in 2001. A similar approach will be necessary for the future, and currently work has just begun on whether there is a need for a 2006 census.
18. Longer term, it is vital to assess the needs for information of a type collected by a census at a local level on a comparable basis across the country. This will then need to be looked at in terms of how well these needs might be met. It is still doubtful in the United Kingdom whether administrative records can yet be relied upon to provide the basic numbers of where people are and how old they are, but a new division has been established in ONS to undertake work to see how more use could be made of administrative records. Considerable hurdles—moral, political and legal—will need to be overcome in terms of data sharing, appropriate use of data and ensuring that details are up to date. A further important consideration is that the very groups of the population that the census has most difficulty in counting are likely to be those for whom administrative data are of poor quality or missing.
19. However, there is a growing need for information at the neighbourhood (small-area) level in the United Kingdom to support the government’s neighbourhood renewal policies. A neighbourhood statistics service has been established to make statistics that are already available more widely accessible and to explore new ways of meeting current and future needs.
20. The strategic approach to census taking in the United Kingdom has been to use the resources available in the best possible way. The two main issues facing the United Kingdom in the planning of the management of the census were the expected difficulty (and which fully materialized) in recruiting field staff and the impact on the rest of the statistical office of conducting such a huge operation—the extensive build-up and run-down of staff employed on the census. This is particularly true in the case of the census in the United Kingdom, which has traditionally been held every 10 years.
21. Strategies, therefore, needed to be adopted which minimized these operational risks while targeting our efforts at our goals—to minimize the differential level of undercount in the census and to produce consistent data of sufficient quality to meet needs. The adoption of a postal response method, “one-number” census approach and using a service-contract approach to processing and other census operations all reflected these aims. Further details are given in other parts of this report.
22. The United Kingdom census offices started planning for the next census in 1993. Quality was at the heart of this planning. An extensive research programme was set up to look at various options within the scope of a traditional census: to issue and retrieve a form from every household and person in the country on census night. There were a significant number of drivers for change, including experience with the 1991 census, society, technology and customer needs and expectations.
23. Given the need to change following the 1991 census and developments since, the strategic approach to the census in the United Kingdom was that we should adopt new technology where it was most relevant to meeting needs. This applied particularly in map production, processing and output. New methodologies were also necessary, and they are described below and in the section on data evaluation.
24. In adopting new technology, it became apparent during the testing phase that not all expertise lay within the census offices. The strategy was to use the best available where possible. Processing of the census was identified as a key area where new technology could help but where expertise and flexible access to resources lay elsewhere, outside the census offices.
25. An extensive open procurement exercise was conducted, and Lockheed Martin was awarded the contract to carry out the printing of census forms and the basic data capture and coding of data from the completed forms. The procurement process was deliberately set up in such a way that solutions to the basic job were sought—that is, a service, rather than the supply of systems or software. This enabled the supplier to own the solution and manage it appropriately. The printing of forms was part of the service to enable Lockheed Martin to ensure that they were compatible with their own image-recognition technology.
26. Other service contracts were let in respect of the payroll for field staff, the public enquiry helpline, distribution and collection of census forms and other material from field staff and printing of other census material.
27. To assist in identifying households and ensuring that enumerators find them all, good maps are essential. Two developments were made in the census in England and Wales. Firstly, enumerator maps were produced using new Geographic Information System (GIS) technology based on Ordnance Survey's product Addresspoint™. This provided enumerators with a customized map on a single sheet of paper, rather than several maps with hand-drawn boundaries of varying scales, as in previous censuses.
28. Secondly, a list of addresses was provided as a starting point for enumerators, but their instructions were still to call at every address in their area to try and make contact to establish how many households and people there were behind the front door. In areas of high levels of multi-occupancy (that is, an address with more than one household), enumerator workloads (that is, the number of addresses an enumerator was expected to cover) were, on average, half the size of those in “easier” areas.
29. It was estimated that the 1991 census counted some 98 per cent of the population. This included some 1.5 per cent of people “imputed” in households known to exist but from which no census form was received. While the overall level of coverage was high in comparison to international standards, the level of coverage varied considerably from area to area and from subgroup to subgroup. This so-called differential underenumeration led to a rethink about the entire approach to conducting the census.
30. There had been significant societal changes—that is, households and people were away from home more frequently, and there had been a significant increase in the number of one-person households and entry-phones into buildings. This made the enumerator’s task of collecting forms that much more difficult.
31. Since 1991, the research programme on data collection focused on the issue of maximizing coverage and particularly on reducing the bias in the resulting count. This meant that our efforts and resources had to be targeted at those households and people with the greatest propensity to not comply with the census—single-person person households, households in multi-occupancy, young men, students, the elderly and ethnic minorities.
32. A number of initiatives were made. Firstly, the 2001 census was conducted on an entirely resident basis. In previous censuses, people were counted where they were on census night and where they were resident if different. People away from home on census night—some 1.5 million in 1991—were required to have information about them supplied twice. However, there was some evidence that people away on census night understandably felt they had already complied with the census and their details were not recorded at their home address. Further, the information on the number of “visitors” (that is, non-residents) to an area was not extensively used. So the first step was to reduce this burden on the public. In 2001, people were asked to supply information only at their home (resident) address.
33. Secondly, the public was asked to post their census forms back. Prepaid return envelopes were provided. Tests had shown that we might expect some 60 per cent to 70 per cent of households to respond by post. In reality, some 88 per cent in 2001 posted their form back, a very pleasing result.
34. It should be emphasized that enumerators still delivered the census forms and “followed up” those households who had not returned a form. This “follow-up” procedure was due to start some four to five days after census day, but in practice it had to be delayed by about a week because of the considerably higher than expected postal response. Tests had shown this to be the best strategic approach, allowing enumerator resources at “follow-up” to be targeted at those areas which were likely to have a lower postback response—inner cities, areas of high elderly population, areas of high ethnic minority population, areas with high levels of multi-occupied buildings and so forth.
35. A community liaison programme was established. This was more than part of the publicity strategy for the 2001 census. The objective was to establish links at national and, more particularly, local levels to “sell” the benefits of the census to different communities and give assistance where needed. This programme was cascaded via the census area managers, the top level of the census field force.
36. Students are another particularly difficult group to enumerate in a census. There were distinct advantages, from a user perspective, in counting students at their term-time (as opposed to vacation) address. It is the term-time population of an area that is used in allocating resources from central to local government. To help achieve an accurate term-time count, the 2001 census was carried out at a time outside vacation periods for the majority of universities and colleges. Students still needed some encouragement, however, and this is where publicity played a role.
37. The publicity strategy for the census was, of course, vital. This raised awareness of the public to the purpose of the census and reminded them to return their form. While the main messages were just this, more specific targeting towards difficult-to-enumerate groups was needed in the form of posters and information leaflets directed towards students, the ethnic minority population (including translations into 24 different languages) and other difficult-to-enumerate groups.
38. The level of contract management was considerably increased from previous censuses. While the need to recruit significant numbers of temporary staff to conduct and process the census has been considerably reduced, new skills were required to manage the different levels of contracts and the interfaces between them. This has been demanding, particularly for some contracts. Consultant advisors have been employed to help this process, but one of the key lessons has been the difficulty of managing the interfaces between the different suppliers. Ensuring that census forms were printed to the right timescales so they could be distributed to field staff is one such example.
39. Recruitment of field staff proved a much harder job than previously despite the fact that the numbers required (some 75,000 in total in England and Wales) were reduced by one third from those employed in 1991. Today’s society means that people have less spare time and perhaps less of a community spirit to do the job. In the end, sufficient numbers were recruited, but it was a considerable challenge in London in particular. The number of applicants to posts in some areas was as low as one to one. For future censuses, we will undoubtedly need to look to alternative ways to reduce the need for so many field staff.
40. The new-style maps were well received and enabled enumerators to identify their areas much more easily than hitherto. There was an overwhelming postal response to the census, and it was clear that the right approach had been taken in the method of response for the majority of the public in today’s busy lifestyles.
41. The high level of postal response did, however, cause problems at the local level for field staff in being able to identify those households who had not posted their forms so they could be “followed up”. Problems were encountered in managing this in real time. The management information systems used by census field staff were inadequate in providing information on how the process was going, thus hampering the ability of managers at the centre and regionally to reallocate resources where they were needed. In hindsight, a longer period before follow-up should have been allowed, coupled with the deployment of personal computers to all field managers, rather than relying on a one-way information system using touch-tone technology.
42. The pitfalls of changing methodology were also illustrated, particularly when many of the field staff employed had worked on previous censuses when all forms were collected by hand. Training of field staff was extensive and on the whole well received, but perhaps a more strategic and less procedural approach was needed to make the field staff feel more involved and aware of the need for methodological change.
43. The census was conducted during a major outbreak of foot-and-mouth disease. This necessitated an urgent assessment of whether the census should continue. Plans were well advanced and it was concluded that with the adoption of special measures it was preferable to continue. Special procedures and publicity were organized and distributed, which worked well. There were inevitably scare story reports of enumerators’ transmitting the disease, but these proved unfounded and were dealt with efficiently and sympathetically.
44. Crisis management of this and other issues that arise during the conduct of the census are a key element of managing the census. We developed lines to take on a number of issues that we thought might happen from past experience, such as lost forms. These proved useful in providing an overall line to take, but, as ever, each case is unique and needs an individual response, be it to an individual or the media. This is what happened.
45. Media response to the census was mixed. Tabloid newspapers were, following efforts by the publicity team, supportive, but broadsheet and mainstream newspapers were considerably less so and at times were overly and unfairly critical. This was not helped because of the lead-up to the general election. The strategy taken was to respond to issues that were clearly incorrect. Media management was a short-term and demanding aspect of the census.
46. In an exercise as large and complex as the census, it is inevitable that errors will occur. These are caused mainly at the form-completion stage, through answers’ being missed or inconsistent answers given (such as someone who describes himself as married but is only 10 years old). It is simply not practical, at least in the United Kingdom, to check every form with every household, so it has been traditional for editing routines to be used in processing. Editing and imputation techniques, similar to those used in previous censuses, will be used to “rectify” and estimate such data. The approach to be taken in 2001 builds on the original Fellegi and Holt principles (Fellegi and Holt, 1976).
47. In summary, answers to questions that are considered to be inconsistent with one another are assessed in a series of logical rules. An automatic decision-making process using other information on the form decides which answer is most likely to be incorrect based on the principle of making least change to the data. The answer to the question is then either set to missing or, where there is only one appropriate answer, that answer is substituted.
48. Imputation of missing answers follows by a route whereby the answer to the question is copied from a similar person or household (called the “donor” person or household) in the same or nearby geographical area. Unlike in 1991, it is intended that when a census form contains several missing answers, one “donor” person or household be used. This has been demonstrated to maintain the statistical integrity of the marginal and joint distributions better than using several “donors” to supply several missing answers for one person or household record.
49. Following the editing and imputation of returned forms comes the estimation of numbers and characteristics of households and people missed entirely by the census. Despite the initiatives at the data-collection stage, it is inevitable that some people and households will not be counted, and some subgroups of the population have a greater propensity to be missed than others. Before I describe the substantial innovations for 2001 in this respect, I will return to the experiences in 1991.
50. It is traditional in many census-taking countries for the census to be “followed up” by a post-enumeration survey (PES) to estimate the number of people missed by the census. This was the approach taken in 1991 in Britain with the Census Validation Survey (as the PES was then known). This survey had a dual purpose: to measure both coverage (the number of people counted) and the quality of answers to census questions. Some 20,000 addresses were sampled.
51. Unfortunately, the Census Validation Survey (CVS) failed to find many of the people missed by the census. When the CVS estimates of the census undercount and the census counts themselves were analysed (particularly by looking at the proportions of males to females), it was evident that a more reliable indicator of the national figure would be provided by the existing series of population estimates based on the 1981 census. Further, the CVS (mainly because of its size) was unable to provide much breakdown of the undercount across the country, although some information was used to “re-base” the local population estimates on the 1991 census, constrained to the national estimate. This was unsatisfactory all round, as insufficient information was available to guide census users on the level and nature of the undercount, particularly at the local level. Only broad guidance could be provided. Further, several figures on the total population of the country were actually produced, thus adding to the confusion.
52. The United Kingdom census offices, in response to the problems in 1991, have tackled this in earnest. A separate research programme was set up in 1996 to evaluate different methodologies and to plan the approach well in advance. The objective of the “One-Number Census” (ONC) programme (as it is known) is to estimate the level of underenumeration and to integrate this with the census counts so that all census outputs sum to one number—the national estimate of the population on census day. The methodological research has included looking at administrative records and whether they could be used to aid the estimation of the census undercount. It was clear, however, that no such records were available of the required quality, supporting the conclusions arrived at in the early 1990s. The approach has been therefore to use a post-enumeration survey but one which concentrates exclusively on coverage and which was much bigger to give the statistical resilience that was not there in the 1991 census.
53. A Census Coverage Survey (CCS) of some 300,000 households was carried out some three to four weeks after the census. It encompassed an intensive re-enumeration of some 20,000 postcode units (average size of 15 households) across the country. The sample was designed to produce direct estimates for some 100 “design groups” (average population size of 500,000 people). The sample was stratified by a “hard-to-count” index so that estimates can be made separately according to the likely level of underenumeration. The survey comprised short doorstep interviews of around 10-15 minutes each.
54. The information from the survey will be combined with that from the census and estimates of underenumeration made at the design-group level using a combination of dual-system and regression-based estimators. These will then be cascaded using synthetic estimation techniques to the local/unitary authority level (average population size of 120,000) to provide the new base of local population estimates by age and sex.
55. The final step of the ONC process is to estimate the probabilities of households and people being missed at the local level by type of household and person. Imputation of households and people according to these probabilities will follow to produce a fully adjusted census database. This final process will be constrained to the estimates produced at the local/unitary authority level. Further information on the methodology for the ONC is available at http://www.ons.gov.uk)
56. The end result of this innovative census-processing stage will be a database comprising both individuals and households counted by the census and synthetic households and people representing those estimated to have been missed. It forms the database to be used to produce the output.
57. Demographic analysis will be used in the United Kingdom to help assure the quality of estimates produced by the Census/CCS dual-system approach. The Census/CCS estimates will not be constrained to the demographic estimates; the latter will purely be used to highlight potential problems with the Census/CCS estimates. For example, sex-ratio analysis based on estimates of the population from the previous census and age distributions will help to identify whether particular sex/age groups have been under- or overestimated.
58. As much as possible will be done at a regional and local level, but more reliable checks can really be carried out only at the national level because of the hitherto unreliability of comparable data at subnational levels. For example, information at a national level from administrative records for pensioners and children will provide further diagnostic checks, but because they are based on benefit and pension records and because payments of child benefit and pensions are made in many cases directly into bank accounts, the address of the recipient is often not strictly necessary.
Fellegi, I.P., and D. Holt (1976). A systematic approach to automatic edit and imputation. Journal of the American Statistical Association, vol. 71, No. 53 (March).