Background: Changes to the design of the General Household Survey (GHS) risk a discontinuity in the ONS Health Expectancy (HE) series. This report compares methodological approaches to ensure consistency of this important National Statistic and reports improvements to the methodology surrounding standard error (SE) estimation.
Methods: A comparison of the use of cross-sectional and longitudinal GHS data in estimating health state prevalence and HE in the UK, GB and constituent countries for the period 2005-07. Incorporation of the GHS sampling design effect into the calculation of standard errors surrounding estimates of HE backdated to 2000-02.
Results: The prevalence of favourable health states was significantly higher in the longitudinal compared to the cross-sectional sample of the GHS. Compared to established trends, the inclusion of longitudinal data in estimates of healthy life expectancy (HLE) led to implausible increases in this metric. However, the prevalence of limiting long-standing illness and estimates of disability-free life expectancy (DFLE) using longitudinal data were broadly comparable and in line with recent trends.Inclusion of the design effect had only a minor impact on the estimation of current and historic HE standard errors. On three occasions, increases in HLE over time were no longer significant using this enhanced method of SE estimation and in a single instance differences in DFLE between countries were no longer significant.
Conclusions: Attrition in the GHS longitudinal dataset introduces significant bias into estimates of health state prevalence and a discontinuity in the established trend of HLE. Utilising the cross-sectional GHS sample alone in 2005-07 represents the most reliable approach to maintain the consistency of this important series. Incorporation of the design effect enhances the precision of estimates of HE.