Submission to ONS consultation on the census, December 2013

Transcription

1 Submission to ONS consultation on the census, December 2013 The Academy has responded to the Office of National Statistics (ONS) consultation on the census. Two options are proposed: retain the census but require online rather than paper completion by default, or scrap the census and rely on existing administrative data, pooled and supplemented by small sample surveys, to provide annual estimates of population size and characteristics. The final decision, to be taken in 2014, will affect England and Wales. The future of Scottish and Northern Irish censuses remains undecided. The following is the Academy s submission to the consultation. Click here to read a more general statement on the Academy s position. 1. What are your views of the different census approaches described in the consultation document? UK social science draws extensively on the consistent, granular data provided by the decennial census. Evidence-based policy-making, planning and resource allocation depend heavily on censusbased research. While improvements and supplementations to the census would be welcome, the Academy considers it imperative that the full census is maintained and that any changes are subject to robust long-term planning and testing, in order to protect the integrity of the research-policy ecosystem. The Academy s response is informed by the expertise of its individual and learned-society members who have expressed a high degree of unanimity on the future of the census. Approach A: The Academy welcomes the move to encourage online census completion by default, provided that digital exclusion and cyber-security risks are mitigated (see q.8) and electronic records of individual returns are protected in the long term (see q.7). Approach B: We would welcome the benefits of more frequent data and more reliable population estimates towards the end of the decade by making greater use of existing administrative data (see q.5). But we note serious shortcomings to the proposed replacement of the census by administrative and survey data: The loss of data at the output-area level would altogether preclude some kinds of research and data outputs, with implications for policymaking and resource allocation. It would also prevent researchers from producing their own statistics for new areas such as retail catchment areas; areas prone to particular social or environmental risks, and areas whose political or administrative boundaries have 1

2 changed (such as postcodes, Conservation Areas, areas targeted for local economic activities, aircraft noise footprints and Greenbelt zones). (See q.4.1). The effects of over-sampling on small areas and small populations will preclude research based on flow data (such as commuting and migration) and research on minority groups, as the Beyond 2011: Independent Review of Methodology has shown (see q.4.1). Changes to definitions and processes in administrative data from government departments would reduce the consistency of data available, hence also the ability to make reliable comparisons with previous years data. ONS has insufficient authority to control how data is handled by government departments (see q.8). The methodology for population and attributes estimates based on administrative data and sample surveys, promoted as a key advantage to this approach, is not yet fit for purpose (see q.8). This is made clear in the Beyond 2011: Independent Review of Methodology. Extensive legislation and practical measures would be necessary to enable data sharing across departments; ensure consistency in definitions and processes across departments; make survey completion compulsory; ensure survey samples are representative; protect against cyber-security risks. More time is needed to resolve both the methodological and legislative issues. Other European countries and New Zealand take years to implement comparable changes. The issues at stake are too important to rush. Because of these fundamental concerns and a number of associated risks (see again q.8), we are clear that option B must not replace the full census. The Academy recommends: the full census continues (with online completion encouraged, provided the risks of digital exclusion and security are mitigated see q.8); ONS trials the use of administrative data alongside the 2021 Census to cross-check population estimates; if the methodology for producing population estimates from administrative data proves successful, more administrative data is used incrementally over time to supplement not substitute for the decennial census. (See also q.9.) 2. Please specify any significant uses of population and housing statistics that we have not already identified. [Left blank] 2

3 3. Please specify any significant additional benefits of population and housing statistics that we have not already identified. [Left blank] 4. What would the impact be if the most detailed statistics for very small geographic areas and small population groups were no longer available? High, medium, low or no impact? High If medium or high impact, please give further information. OA-level data are essential for research on the smallest geographical areas and minority groups, comparing one local area or minority group with another or with a national average, and monitoring change in small areas over time. This includes studies on urban change (e.g. the impact of a new airport or wind-farm on local households, or the changing demographics of a local area) and research to predict demand for or value placed on services and changes to land use (e.g. health services, parks and Greenbelt land). A study by Ahlfeldt et al (2013), for example, used ward-level and OA-level data on educational level, median income and ethnic composition from the 1991, 2001 and 2011 Censuses to analyse the effects of local demographic changes (gentrification) on the designation of Conservation Areas. 1 Likewise Gibbons and Machin (2005) used population density estimates for Enumeration Districts from the 1991 Census to evaluate the impact of London Underground and DLR improvements on house prices, hence the value of rail access to local residents. These kinds of research would not have been possible without the granularity of the decennial census. The loss of small-area data would also mean the loss of datasets on migration and commuting, with significant impact on local government, transport planners, migration researchers and commercial organisations that use LSOA data to plan new store locations. For example, data on internal migration and ethnicity, only available from the decennial census, have enabled researchers to test empirically the White flight hypothesis the idea that White population groups migrate away from inner-city areas with a high non-white population and to show that it is unfounded (e.g. Simpson 2007), with significant implications for policies aimed at breaking down segregation. 2 Similarly the Eddington Transport Study, commissioned by the Chancellor and Transport Secretary, relied on 1 Ahlfeldt, G.M., Moeller, K., Waights, S. and Wendland, N. (2013). Game of Zones: The Economics of Conservation Areas. Spatial Economics Research Centre Discussion Paper No 143 [pdf] Available online: 2 Simpson, L. (2007). Ghettos of the mind: the empirical behaviour of indices of segregation and integration. Journal of Royal Statistical Society Series A, 170(2):

4 journey-to-work data from the 2001 Census to advise government on the long-term links between transport and the UK s economy. 3 OA data is the building-block of other data outputs such as classifications, indicators, indices and sampling frames. Aggregated statistics would obscure the most granular information, reducing their utility and reliability. Research conducted in 2009 by the Young Foundation, for example, analysed data on households using Output Area Classifications derived from the 2001 Census to investigate who has the most acute needs in Britain. The Big Lottery Fund used the findings to target funding, plan programmes and inform their research and communications strategy. Without census-derived OACs, such precise research would not have been possible and resources could not have been targeted so effectively. 4 The availability of OA-level data also enables researchers to derive their own statistics for new areas and to make valid comparisons between them. Examples include retail catchment areas, areas prone to social or environmental risks and areas whose administrative or political boundaries have changed. Cheshire and Sheppard (2004), for example, aggregated data from the 1991 Census wards on households and ethnicity in order to analyse the relationship between schools quality and houses prices in Reading. 5 Similarly Travel-to-Work Areas, derived from granular commuting data from the census which ONS has shown to be unobtainable from surveys, were the template for Local Enterprise Partnerships and City Deals (HM Government 2010; 2011) What would the additional benefit be if more frequent (i.e. annual) statistics about population characteristics were available for areas like local authorities and electoral wards? High, medium, low or no additional benefit? High. 3 Eddington, R. (2006). The Eddington Transport Study: the case for action. London, H M Treasury on behalf of HMSO. 4 Campaign for Social Science (2013). Britain s unmet needs. Case-study in Making the Case for the Social Sciences: Longitudinal Studies [pdf] Available online: 5 Cheshire, P. and Sheppard, S. (2004). Capitalising the value of free schools: the impact of supply characteristics and uncertainty. The Economic Journal 114(499): F397-F HM Government (2010). Local Enterprise Partnerships Letter from Secretaries of State BIS & CLG to local authority and business leaders [pdf] Available at: HM Government (2011). Unlocking Growth in Cities (Cabinet Office, London) [pdf] Available at: acc.pdf 4

5 5.1. If medium or high, please give further information. As public services become increasingly privatised and fragmented, net migration increases and labour markets change, more frequent data is needed to enable timely detection of changes to demand. In theory, annual data at relatively disaggregated levels could help local authorities plan and deliver services that meet contemporary demand e.g. schools places, health services and public transport but the methodology is currently inadequate to deliver this in practice (see q.8). 6. Please specify any significant uses of census information for historical research that we have not already identified. There is increasing interest among urban scholars and economic historians in using individual records, when they enter the public domain, for micro-level analysis of urban and neighbourhood change. See for example Meen and Nygaard (2011) who studied the impact of land use patterns on price-elasticity in local housing supply. In calculating the number of households per MSOA per year, they also noted a counter-intuitive advantage in the lack of annual data, namely that multiplying annual population values by the household representative rate in the 2001 Census removes any endogeneity in household formation. This approach has a long history in housing economics but wouldn t be possible using annual administrative/sample data alone What advantages or disadvantages for genealogical or historical research can you see from a move to a solution based on archiving administrative sources? Historical researchers use individual census returns to derive statistics for units that change over time or no longer exist, such as communities, family structures and occupations. For both technical and legislative reasons this is unlikely to be available for data from surveys or administrative sources. The public provision of anonymised micro-data files may partially mitigate this problem but would require significant additional resources. Consistency over time in the kinds of data available is also vital for historical research. This is unachievable with administrative data since individual departments control which data they collect. ONS has insufficient authority to influence this (see q.8). Period studies on the determinants of mortality, social mobility, racial demographics and women s citizenship are among the kinds of historical research that have relied heavily on the use of census data, and which may not have been possible with inconsistent administrative sources (see the joint submission by the Economic History Society, Royal Historical Society and Social History Society). 7 Meen, G. and Nygaard, C. (2011). Local Housing Supply and the Impact of History and Geography. Urban Studies 48(14):

6 8. What are your views of the risks of each census approach and how they might be managed? Approach A: Requiring online completion risks excluding those without internet access or with unreliable signal coverage, including roaming groups, elderly people, remote communities and those unable to afford the technologies such as the homeless. There is also a small but significant proportion who shun digital technologies. Robust follow-up to non-responses will be essential to ensure the data is representative an issue not fully explored in the consultation document. A centralised digital database of returns would introduce a different set of security risks. We commend ONS s reputation for data security but note that new technical and legislative measures would be needed to minimise and detect fraudulent response, ensure secure data transmission (particularly on mobile devices whose security settings are low to enable advertising), ensure safe storage, prevent data being shared with third parties (including nation-states, businesses, police and intelligence services), and protect against the misuse or misrepresentation of information captured. Approach B: In addition to the fundamental concerns expressed in q.1, we note significant risks to the range of research outputs possible under option B: o Longitudinal studies may be compromised or lost due to the inconsistency of administrative data available since the latter are gathered primarily for other purposes and in accordance with departments changing needs. Where studies could be supplemented by administrative data, their outputs would be less rich than with census data. o Political interests would likely affect which administrative data are continued, integrated and made available publically. ONS will not have a say over what kinds of data are gathered by government departments. There are particular risks in attaining population estimates under option B: o Under-coverage is likely, given the need to be on both the NHS patient register and the Customer Information System to be counted. Using a combination of sources where necessary, e.g. HESA data for students and the School Census, would reduce this risk. o Legislative changes impacting on administrative data, such as privatisation and the new restrictions on child benefit to those on low incomes, would also affect who is counted. o We note also the risks of over-coverage, geographical errors in administrative data and matching errors identified in the Independent Review of Methodology for Beyond Data on attributes of minorities are unreliable with administrative data. The proposed 4% sample survey would help at high aggregate levels but not at the OA-level (see q.4.1). 6

7 Census data is used to calibrate sampling frames (as noted in the Social Research Association s submission). The loss of the census therefore could have significant impact on research methodology. More work is needed to assess the likely impact. Changes to government practice and the need to collaborate across departments to integrate administrative data may be very difficult to achieve in practice. Senior-level central oversight would be needed to ensure effective integration, adding to costs. Anticipated changes to EU legislation on data sharing may also limit the integration of administrative data sources. Possible changes to the administration following the general election may affect planning after As with approach A, the risk of non-response (although mandatory) and digital exclusion means robust follow-up would be essential for non-responses. Changing to an approach based on administrative data and sample surveys would require considerable education to users because of the relative uncertainty compared to statistics based on census data. This could be particularly problematic for local charitable groups and media reporters who are not trained analysts and use census statistics with little regard to confidence intervals. The 4% survey of population characteristics may be vulnerable to cuts. 9. Are there any other issues that you believe we should be taking into account? The transition period for any change will be crucial, although this issue is given little weight in existing Beyond 2011 documents. Between now and 2021 more work is needed on the methodological and legislative issues surrounding the use of administrative data. ONS must also look to strengthen its role in controlling changes to definitions and processes in administrative data. We recommend that ONS conducts a small trial for the online Census, both in areas where we can expect a good response rate and those where we might not, to learn lessons in time for Academy of Social Sciences December