Science and Technology CommitteeWritten evidence submitted by the Royal Statistical Society (Census 30)
1. The Royal Statistical Society (RSS) is the UK’s only professional and learned society devoted to the interests of statistics and statisticians. Founded in 1834, it is one of the most influential and prestigious statistical societies in the world drawing its membership from over 50 countries. It aims to promote public understanding of statistics and provide professional support to users of statistics and to statisticians.
2. The RSS Census Study Group provides a cross-sectoral forum for debate and education among census users, focusing on the broader methodological issues which should concern all users of census data rather than on specific sectoral or topic interests. The group convenes meetings and contributes to the work of other special interest groups, bringing together census users from across academia, business, central and local government and the voluntary sector.
3. The Academy of Social Sciences, which exists to be the voice of the social sciences in the UK for the public benefit and whose membership encompasses 750 individual Academicians and 43 learned societies (with a total reach of 85,000 social scientists), drawn from right across the social science disciplines at the academic, policy and practice levels, is glad to add its support for and endorsement of this submission. The issues that it raises are shared by the wider social science community.
General Points
4. The RSS would stress the importance of the Census in providing more than just national figures; its value and use depends on a key feature – the ability to provide highly disaggregated figures (often for small geographies) that also permit the joint analysis of a range of variables. The Census is also resource that is in constant use as a key information source throughout the ten year inter-censal period. Census outputs take two key forms – highly detailed cross-tabulations for separate locations of interest and anonymised micro-data. Both forms of output must be available, not just to government, but to a much wider user constituency. (In addition there is a third form of output in the longitudinal study.) The current outputs are available to a huge constituency of users outside government and provide enormous economic benefits. If the current Census is to be replaced by something that combines administrative data at the personal level then the issues of access for this wider constituency need to be addressed as an integral aspect of the evaluation.
5. The RSS supports the “Beyond 2011” project and agrees that it is right to investigate whether an alternative to the traditional Census would prove advantageous. Nevertheless the traditional census provides a huge range of information. As well as providing the most vital facts about society at local and national level, this is a crucial underpinning of large parts of the evidence base for government, both central and local, for the NHS, and for business and researchers. It is a major foundation block of the UK’s information system both in respect of the uniquely detailed information it provides and also as it is the benchmark on which numerous sample surveys are founded and against which they are calibrated. We believe it provides substantial value for money.
6. Thus it is crucial that alternatives are fully and properly explored. While it would clearly be desirable to find a lower cost method to replace the census, it would be a false economy to reduce costs if this resulted in overall much lower information content. The current ONS “Beyond 2011” programme of research is to be welcomed as a means of establishing as clear as possible a picture of the genuine potential for assembling a viable census replacement system. It is anticipated that any such system will need to comprise a combination of administrative and survey sources, the latter required to address important variables which are not collected in any comprehensive administrative systems.
7. The work of the ONS covers England and Wales. Separate studies are under way in respect of Scotland and Northern Ireland. It would be a major concern if there were substantial differences in the systems adopted for the different parts of the UK.
Question 1: How do Social Scientists use Census data?
8. Social scientists use census data for an extremely wide range of purposes, reflecting the breadth of the census as a data collection exercise which is designed to meet the needs of many sectors and organizations. It should also be remembered that social scientists are located across all sectors, with significant census use in central and local government, business and academia in addition to the large body of social science researchers and students within the academic sector. Many social scientists are specifically concerned with major thematic research questions in fields such as migration or ageing and will use a range of research data sources, of which the census will be one.
9. A particularly unique strength of the census is its near-comprehensive population coverage which facilitates a unique range of integrated data outputs and cross-tabulations which are available down to a very small area level. No other single source provides detailed tables for small geographical areas. The small area aggregated data provided by the census allow a range of analyses to be undertaken investigating neighbourhood influences on aspects of life such as educational achievement, health and crime. Business users in particular make extensive use of these multivariate small area data for area classification and to inform decisions about target markets and site location. Government users are more likely to be concerned with items such as the assembly of deprivation indicators and denominator datasets which permit the characterisation of areas to inform the delivery of services. These applications demonstrate the enormous importance of the census as a baseline dataset, with which unique analyses are possible due to its unique combination of breadth and depth. The census also provides more specialised research datasets such as flow data permitting the investigation of travel to work and migration patterns, made possible by asking place of residence, place of work and place of residence one year before the census and providing important insights into travel patterns, housing demand and patterns of economic development.
10. A further family of unique analyses is made possible by samples of anonymised records, which permit users, for example, to create bespoke cross tabulations of census variables and to analyse household structures. Although these represent only a small fraction of the census dataset, the near-comprehensive census coverage ensures “sample” sizes far greater than even the largest social surveys. A further most important research linkage is provided by the ONS Longitudinal Study, which now contains a sample of linked records from 1971 to 2011 and permits unique insights into individual life histories, for example understanding the role of past employment and educational history on health and mortality later in life.
Question 2: What impact will the ending of the census have on social science research?
11. The answer to this question depends entirely on the nature of the replacement data sources. Public acceptability will be a major determinant in the success of a potential alternative system although that issue is not addressed in any detail here. There is no certainty that census data can continue to be collected and published while maintaining current quality levels. In 2001 substantial adjustments were necessary to the census estimates for some local authorities due to failings in the census process. Initial feedback suggests that 2011 enumeration has been successful in addressing these challenges, but at increased cost and also by using a variety of different data collection methods (online, post-back, post-back following enumerator visits) which may introduce further biases. It will not be possible to fully assess the implications of these changes until quality assurance of the 2011 census outputs is complete.
12. Perhaps the greatest risks of ending the census would result from the lack of integrated data which permit the cross-classification of different variables, with a resulting loss of explanatory power. Administrative data bases do not usually provide detailed cross-tabulations of outcome variables against socio-economic and other characteristics. It is also not certain whether alternative sources will be able to provide all the richer datasets such as flow data, microdata and links to longitudinal studies. The extent to which alternative sources will be able to overcome the weaknesses of the census model, while replacing its combination of geographical and social detail, will be key to evaluating the impact of ending the census. There are some census questions for which there are no alternative sources, or for which alternative sources may take many years to develop for example questions on health and languages.
13. While the use of administrative sources may offer the prospect of more timely data, the overall data infrastructure would become more dependent on the policies of a range of government departments and organizations whose primary objective is not data collection. From a social science perspective this creates a risk that data series may change in unplanned ways and that comparable datasets through time may be difficult to achieve. Government departments supplying information to the statistical system replacing the census should be required to consult with the National Statistician before implementing major changes to their data collections.
14. Despite this, there is no reason why a non-census system could not be developed to continue to create microdata samples and links to longitudinal studies - indeed, the longitudinal studies already demonstrate much of the power of linked administrative systems. However, the lack of a decennial census validation would present real barriers to long-term analyses due to the unknown effect of such a major break in the methodologies employed. International experience suggests that it takes a long time to develop alternative data sources to the point where they can provide replacement data series of equivalent quality to that which has been achieved by recent UK censuses.
15. Even if an alternative to the Census could be developed which would satisfactorily meet the needs of social scientists, this will take time. A main current criticism of the Census is that it is decennial and that there is a long wait for new data – we will continue to use 2001 data until well into 2013 for example. A concern is that if the 2011 census proves to be the last, it may be an unacceptably long time before a suitable alternative can be developed.
Question 3: What alternatives to the Census would provide population and socio-demographic data of equivalent or higher quality?
16. The answer to this question is contingent on the levels of coverage and accuracy which may be achieved by a notional future census. Whereas census data typically appear 18 months to three years (in the case of some of the richest research datasets) after the enumeration, the long time period between each data refresh can be a challenge to many users. The census continues to be challenged in terms of overall response rates, difficulties of identifying and gaining access to properties and a range of hard to reach population groups, including young men, visitors to the UK and those with multiple residences and complex household structures. Initial signs are that the 2011 census has achieved a good result in the face of these challenges, but with a considerable increase in cost. International experience suggests that the difficulties of achieving near-comprehensive coverage will continue to increase. In the 2001 census coverage in some areas fell to levels which were not amenable to satisfactory adjustment using statistical methods.
17. There are thus attractions to the use of a system of alternative sources which might address these difficulties more effectively, but it is unlikely that this can be achieved by use of any single approach. Rather, a combination of methods and alternative sources needs to be evaluated. It seems likely that any alternative would need to include some form of statistical population register or database which would allow the matching of other lists at the individual record level, otherwise it will not be possible to generate multivariate datasets comparable to those obtained from a census.
18. ONS is currently conducting a welcome programme of research on alternative sources, among which the NHS register and DWP/HMRC Customer Information System appear to be the most promising reference frames. Also very welcome is the recent establishment of the GeoPlace national address gazetteer. A substantial amount of additional cross-matching would be necessary in order to overcome some of the current known weaknesses of these lists in terms of under- and over-counting. Against these lists, a wide range of additional registers could be matched to derive additional characteristics for example annual school census data, HESA records, benefits data, DVLA information etc.
19. No system of matched registers would provide the levels of detailed social characteristics available from the census but many of these could be addressed by an integrated and extended social survey system, which drew on the same address gazetteer as a sampling frame and was appropriately stratified. This could in theory address all the current census themes and more, providing greater responsiveness for the inclusion of additional questions, but with data inevitably not available for such detailed geographical areas. The quality of many of these sources remains unknown in detail, which is why the current ONS Beyond 2011 programme of work is to be strongly supported.
Question 4: What other existing sources of population and socio-demographic data could be improved upon?
20. The creation of an integrated non-census population data system will require a range of different, coordinated actions. Firstly, it will be necessary to identify ways of improving the quality of existing sources. For example, reconciliation of addresses from a definitive address register and other sources such as the NHS register will result in improvements to both. The NHS register provides the most promising data source to contribute to population counts and internal migration but it is subject to recognised weaknesses, including both over- and under-counting of specific population groups. Moving to a post-census statistics system would require issues such as this to be seriously addressed and supported to an extent which has not previously been done - and also recognition that complete resolution is most likely unachievable. Secondly, some existing sources would need to be augmented by considering the need for additional data collection, for example whether more systematic ethnicity information could be included in one of the major data series such as the NHS register. Developments in other areas may yield specific new variables of interest: for example the current census question on central heating might in due course be more usefully replaced with information on energy use consumption from utility records.
21. Despite recent improvements, the measurement of international migration still remains a major weakness and the availability of well-structured data from the e-borders system would be an essential element of a workable census replacement system. A third element would be close integration of the current social survey system with the census replacement, almost certainly involving both the integration and expanding of existing surveys so as to provide greater coverage of socioeconomic detail and the continuation of a substantial coverage survey which would allow both user and producers to better understand the characteristics of the post-census population data. Declining response rates to the major government surveys is already a concern and it should not be assumed that using larger or more integrated surveys as part of a census replacement will overcome these challenges without further detailed attention to design and implementation. Increased use of internet-based data collection may offer some advantages in this respect but again its potential role in a greatly changed data environment requires substantial research effort.
Royal Statistical Society
30 November 2011