~ Pergamon 0277-9536(95)00184-0 Sot'. Sci. Med. Vo|. 42, No. 6, pp. 857-869. 1996 Copyright © 1996Elsevier ScienceLtd Printed in Great Britain.All rights reserved 0277-9536/96$15.00+ 0.00 ANALYZING PERCEIVED LIMITING LONG-TERM ILLNESS USING U.K. CENSUS MICRODATA MYLES I. GOULD and KELVYN JONES Department of Geography, University of Portsmouth, Lion Terrace, Portsmouth PO1 3HE, England A~tract--The 1991 Census of England, Wales and Scotland is an improvement on previous censuses in providing fine-graineddetail on the geography of limiting long-term illness.Another innovation of the 1991 Census is the release of a Sample of Anonymized Records (SARs). These provide a considerable sample of detailed data on individuals at the sub-regional level. This paper explores individual and geographical variations in morbidity through a multilevelanalysis of the SARs. Geographical differences in morbidity are found even after allowing for age, sex, ethnicity, housing tenure, social class and car ownership. Key words--limiting long-term illness, morbidity, census, Sample of Anonymized Records (SARs), multilevel modelling, England, Wales and Scotland INTRODUCTION The 1991 U.K. population census provides two novel features for research into the geography of health, a question on morbidity and a large sample of individual records. Previously the U.K. Census contained information about non-availability for work due to permanent and temporary sickness and disability. The sickness information represented a 'residual' category in a question relating to employment, and did not address the question of illness per se. However, in 1991 the following direct question was asked of every person in each household: Does the person have any long-term illness, health problem or handicap which limits his/her activitiesor the work he/she can do? [1] Consequently, the census now provides an unrivalled geographically detailed source of infor- mation on perceived levels of morbidity in the population. Dale [2] notes that pre-census tests on data from the limiting long-term illness question correlated well with other data on GP consultations and in-patient and out-patient visits to hospital. She argues that it provides the only nationally consistent indication of health service needs [2]. Unfortunately, although detailed census data are available for small areas (wards and enumeration districts) [3, 4] the morbidity question is not simultaneously disaggre- gated by more than three socio-demographic variables. It is, therefore, impossible to allow for a broad range of possible confounding factors when undertaking statistical analysis. Much more detailed information are available in the Sample of Anonymized Records (SARs) which were released to academics for the first time by the Census Microdata Unit in Autumn 1993 [5]. The SARs consist of two files, the 2% individual SAR and the 1% household SAR, the first file is considered in this paper. The individual SAR file contains data for 1.1 million individuals who live in households and communal establishments in England, Scotland and Wales. The SARs provide a very large sample of data for investigating variations in self-reported morbidity. They contain individuals' records for all the topics contained in the 1991 census (giving details of economic activity, education and socio-economic characteristics) as well as perceived morbidity [5]. The individual SARs are drawn from the 10% sample of all census forms that are fully processed by the OPCS, and only contain individuals from non-imputed house- holds [6]. Consequently SAR records only contain individuals living in households who fully answered all the questions contained in the 1991 Census form. This paper represents the first example of the statistical modelling of the morbidity data contained in the SARs, and makes use of multilevel logistic models [7-9]. It demonstrates the potential for exploring social and geographical variations in perceived levels of morbidity using the SARs. The paper has three specific aims. First, to compare different sources of U.K. morbidity data paying particular attention to geographical disaggregation and national coverage. Second, to describe how multilevel models can efficiently analyse very large data sets, in this case potentially 1.1 million individual records. Finally to provide some preliminary analysis of the social and geographical variations in illness in the British population. COMPARING THESARS WITH LBS/SAS ANDOTHER SOCIAL SURVEY DATA Table 1 summarizes the strengths and weaknesses of the various sources of morbidity data that are ssM42/~F 857