Applying relational models to the graduation of disability schedules

Age-specific rates of particular disability types are important for planning purposes and are a valuable input to estimates and projections of populations with different disabilities. However, survey estimates of schedules of disability rates display evidence of sampling variability and sub-national disability schedules are often unavailable for reasons of disclosure protection. This paper develops and evaluates a method to smooth sampling variability in national schedules of disability using a technique that has applicability to sub-national estimation of age-specific disability rates. Relational models are used to adjust the limiting long-term illness schedule for England (Census 2001) to represent different disability schedules (Health Survey for England 2000/2001) smoothing sampling fluctuations. For hearing disability a simple Brass relational model involving two parameters provides a good fit. For other disability types a modified version of the Ewbank relational model with three parameters is required. This paper illustrates that relational models can accurately capture the relationship between age-specific rates of limiting long-term illness and various disability types.