Radical
Statistics
Census Information
3) Response rates and the One Number Census
During the 2001 Census, one aim of Census planners has been to reduce the differences in coverage between areas and population groups. This aim was probably thwarted by the lack of flexibility to re-direct fieldwork resources between census Districts, and the difficulty of focusing on areas of poor response caused by the postback delays. Responses have been relatively hard to gain from locked blocks of flats, student housing, and single-person households, and there are many more than ever before in each of these categories.
ONS are in summer 2001 expecting an overall household response rate of well over 95% (ONS, 2001a). This is likely to be much lower in Inner London and some other areas. The final response rates may well be sufficiently high even if they turn out to be less than in 1991, if the 'One Number Census' procedures for validation and correction of census coverage are successful.
In the 1991 Census it was not so much the level of non-response as the lack of effective measures of its location and characteristics that caused dispute. For 2001 the One Number Census involves a Census Coverage Survey much larger than in 1991, using interviews with 300,000 households sampled in England and Wales (and replicated in Scotland and Northern Ireland). It checked on the Census enumeration, with the aim of directly measuring age-sex specific non-response in 100 areas of the UK, and accurately modelling it for each Local Authority District. A series of quality assurance procedures that are being described to census data users ahead of their use are intended to cope in an agreed manner with a failed coverage survey if that has unfortunately occurred in one or more areas.
The same coverage survey will measure the characteristics of those missed in the enumeration, both whole households and those missed from within enumerated households. Extra census records with those characteristics will be created ('imputed') in appropriate places within the Census database, also in areas that were not included in the coverage survey sample. All tables of output will be based on this augmented database corrected for undercount - thus the term One Number Census. In this way, 2001 census outputs due in late 2002 and in 2003 will be partially estimated but of better quality than could be gained from the Census enumeration alone. Guides to the One Number Census are due to be published by ONS and GRO(S) in September 2001.
The One Number Census procedures are mainly new for 2001. As in 1991, a set of 'hot deck imputation' and validation procedures will also fill in gaps in census forms due to questions omitted by a respondent - or not being captured by scanning procedures.
So long as the coverage survey captures sufficient of those missed by the census to estimate the remainder successfully, the severe difficulties reported by fieldworkers will be lessons to learn from rather than represent a heavy spanner in the works of the census operation and the quality of the resulting data.
