Purpose: To evaluate two psychometric properties of SF-36, namely unidimensionality and reliability.
Methods: The data are from three cohorts in the HALCyon collaborative research programme into healthy ageing: Aberdeen Birth Cohort 1936 (n = 428), Hertfordshire Ageing Study (n = 358) and Hertfordshire Cohort Study (n = 3,216). The Mokken scaling model was applied to each sub-scale of SF-36 to evaluate unidimensionality as indicated by scalability. The lower bound for internal consistency reliability was determined by Cronbach's alpha.
Results: All six sub-scales of SF-36, with the exception of general health (GH) and mental health (MH), demonstrated strong scalability (0.5 ≤ H < 1). The results were consistent across all 3 cohorts. Both GH and MH showed medium scalability (0.4 ≤ H <0.55), although individual items 'sick easier..', 'as healthy as..' and 'expect to get worse' of the GH sub-scale and 'nervous', 'happy' in the MH sub-scale had low scalability (H < 0.4) in the oldest cohort (aged 73-83). Cronbach's alphas for all sub-scales were between 0.70 and 0.92.
Conclusions: The unidimensionality and reliability of the sub-scales of SF-36 are sufficient to make this a useful measure of health-related quality of life in older people. Caution is needed when interpreting the results for GH and MH in the oldest cohort due to the poor unidimensionality.