Mortality by National Statistics Socio-economic Classification (NS-SEC) in 2001-03 and investigation of numerator-denominator bias

Brian Johnson and Yuan Huang Chow, Office for National Statistics

[Project number 20071]

In 2001 ONS introduced NS-SEC which replaced Social Class and Socio-economic Group as the official socio-economic classification. To date, the majority of mortality analyses used Social Class to explore trends in inequality among occupational groups. In the future such analyses will mainly use NS-SEC.

Previous studies by Fitzpatrick have included the calculation of standardised mortality ratios (SMRs) using the 1991 census and a version of NS-SEC based on the 1990 occupational classifications, known as NS-SEC90, to analyse male deaths between 1991 and 1993.

This project aims to both update this analysis by using death registration data from 2001-3 and the 2001 census, and to use the ONS Longitudinal Survey (LS) to analyse and correct for the inherent numerator-denominator bias.

In order to calculate mortality rates by NS-SEC we will use death registration data and population estimates taken from the 2001 census. It is well established that the occupation reported at the census and that reported at death frequently differs and in the former case has a higher likelihood of not being classified to an occupational group, which is a cause of bias. There are also selection effects. This study aims to use longitudinal data to analyse and quantify these sources of bias.

Amongst the outputs will be SMRs by NS-SEC for males and females, and SMRs by major cause of death. The nature and implications of numerator-denominator bias will be discussed.

This study will be based on individuals who were present at the 1991 and 2001 census who died in the eight months following the census.

The LS allows linkage of occupation recorded at death registration and at census which would not be possible using another data source.

Using the LS sample, deaths occurring in the study period could be linked to the census data for the same people in 1991 and 2001. This would identify the proportion of people who were not classified in 2001 but could be classified in 1991, and to which NS-SEC category they belonged. This would allow an estimate of the distribution of the 2001 unclassified group amongst occupational and socio-economic groups. It would also allow an analysis of deaths within eight months of the census and comparison of recorded occupation at census and death registration.

Top of page