The reliability of sexual partnership histories: implications for the measurement of partnership concurrency during surveys.

OBJECTIVES To measure the reliability of sexual partnership histories collected during survey interviews and to assess the impact of measurement error on survey estimates of partnership concurrency. METHODS We used sociocentric data collected on Likoma Island (Malawi). Up to five of the respondents' most recent sexual partners were identified in population rosters. We assessed interpartner agreement (IPA) in reports of sexual partnerships (i.e. whether partners concordantly report that they have had sexual relations with each other) and its association with respondent and partnership characteristics. We estimated the extent of bias in the point prevalence of concurrency and the duration of overlap between concurrent partnerships according to two scenarios: one in which only partnerships reported by both partners were considered as 'true' ('concordant scenario'), and one in which partnerships reported by either partner were included ('complete scenario'). FINDINGS IPA was low in nonmarital relations, but was significantly higher in ongoing than in dissolved nonmarital relations. IPA was further associated with the number of other partners the respondents or their partner(s) had, as well as with the duration of ongoing partnerships. Biases in measurements of the prevalence of concurrent partnerships were large: concurrent partnerships were rare in the concordant scenario, but common in the complete scenario. This was particularly true among never married women. Estimates of the average duration of overlap between concurrent partnerships derived from self-reported survey data were also biased, particularly among married respondents. CONCLUSION Future empirical tests of the 'concurrency hypothesis' and interventions targeting concurrent partnerships should take reporting biases into account.