The use of relational models to estimate age specific disability prevalence schedules

Alan D Marshall, University of Manchester

The relational methodology involves transforming a standard schedule of rates for a particular demographic charactersitic in order to derive a schedule in another population for which direct data is unavailable or unreliable. This technique, originally developed by Brass (1971), to estimate mortality schedules has also been used to derive fertility and migration schedules (Coale and Trussell, 1996. Congdon, 1993). This paper investigates whether relational methodologies are also a valuable tool for the estimation of disability prevalence, and particularly the generation of age-specific disability profiles for a number of disability types at sub-national levels. Such estimates are required by planners of service provision. This paper assesses the accuracy of disability rates predicted by relational models and by more conventional individual-level synthetic regression models for regions in England using data from the Health Survey for England and Census 2001. The results of this analysis indicate that relationals model are very comparable in terms of accuracy over all ages compared with the other models. The evidence suggests that relational models are most accurate for the most common disability types such as locomotor and personal care. For less common disability types there is evidence of a need for a more flexible relational rule to alter the shape of the base profile (Limiting Long Term Illness) to a greater extent. Smoothing of LLTI schedules using an appropriate function and a more complex relational rule are explored as possible solutions.