Abstract
Self-rated health (SRH) is commonly used in longitudinal analyses as a repeated outcome measure. This assumes that computed changes in SRH over time truly represent within-individual changes in underlying health. The longitudinal validity of SRH, however, is threatened by ceiling effects (where people reporting the highest level of SRH cannot report subsequent improved health), insensitivity to small changes within SRH categories, reference group effects (where individuals assess their health changes relative to their peers) and stability in SRH even when change in underlying health is occurring. We assessed the longitudinal validity of SRH by comparing computed changes in SRH with a measure of self-assessed change in health (SACH). We used two waves of data (2003-2005) from the New Zealand longitudinal Survey of Family, Income and Employment (SoFIE). Computed change in SRH and SACH were compared directly and also in regression models using an objective measure of health outcome change (hospitalisations within the past year). Computed change in SRH and SACH were not well correlated, consistent with ceiling and/or categorisation effects in SRH. In regression models, SACH was more strongly predictive of hospitalisation than computed change in SRH (worse SACH was associated with an increased odds of hospitalisation of 3.7 compared to 1.8 for decreased computed change in SRH). SACH may be affected by recall bias, but if SRH is used as a repeated outcome measure in longitudinal analyses, results may also be biased, if change in SRH does not occur in response to significant health events.