Assessing the validity of respondents’ reports of their partners’ ages in a rural South African population-based cohort

OBJECTIVES: This study evaluated the validity of using respondents’ reports of age disparity in their sexual relationships (perceived disparity), compared to age disparity based on each partner's report of their own date of birth (actual disparity). SETTING: The study was conducted using data from a longitudinal population-based cohort in rural KwaZulu-Natal, South Africa, between 2005 and 2012. PARTICIPANTS: The study used 13 831 reports of partner age disparity within 7337 unique conjugal relationships. 10 012 (72.4%) reports were made by women. PRIMARY AND SECONDARY OUTCOME MEASURES: The primary outcome was the Lin concordance correlation of perceived and actual age disparities. Secondary outcomes included the sensitivity/specificity of perceived disparities to assess whether the man in the relationship was more than five or more than 10 years older than the woman. RESULTS: Mean relationship age disparity was 6 years. On average, respondents slightly underestimated their partners’ ages (male respondents: 0.50 years; female respondents: 0.85 years). Almost three-quarters (72.3%) of age disparity estimates fell within 2 years of the true values, although a small minority of reports were far from correct. The Lin concordance correlation of perceived and actual age disparities (men: ρ=0.61; women: ρ=0.78), and assessments of whether the man in the relationship was more than five, or more than 10 years older than the woman (sensitivity >60%; specificity >75%), were relatively high. Accuracy was higher for spouses and people living in the same household, but was not affected by relationship duration. CONCLUSIONS: Rural South Africans reported their sexual partners’ ages imperfectly, but with less error than in some other African settings. Further research is required to determine how generalisable these findings are. Self-reported data on age disparity in sexual relationships can be used with caution for research, intervention design, and targeting in this and similar settings.