BACKGROUND: The reported number of sexual partners is a variable used extensively in sexual health research. However, the reliability and consistency of this measure, and the statistical assessment of these attributes, are not well understood. METHODS: Using data at ages 21, 26, and 32 years from a New Zealand birth cohort, we compared responses on the lifetime number of heterosexual sex partners to assess reliability and consistency. Differences by gender and age were considered, and the effect of number of sexual partners. A variety of analytical methods were used to explore statistical challenges of these data including variance estimation, fractional polynomial transformations, and quantile regression. RESULTS: We found some level of discrepancy between reports of the number of sexual partners when assessed at different times is common, driven by those reporting a high number of partners who were disproportionately men. Men reported a higher lifetime number of partners than women at each age, and there were statistically significant differences by gender in (a) consistency between reports at different ages, and (b) reliability of reports as measured by both the Intraclass Correlation Coefficient and the Kappa statistic. CONCLUSIONS: When considering reliability, multiple statistical approaches are necessary or conclusions can be misleading. Variance components should be examined when considering the Intraclass Correlation Coefficient. When modelling, robust methods like fractional polynomials and quantile regression should be employed to accommodate nonlinearity. Sensitivity analyses excluding participants whose partner number is in the upper 5% to 25% are informative, as these were shown to have the highest discrepancies.
BACKGROUND: The reported number of sexual partners is a variable used extensively in sexual health research. However, the reliability and consistency of this measure, and the statistical assessment of these attributes, are not well understood. METHODS: Using data at ages 21, 26, and 32 years from a New Zealand birth cohort, we compared responses on the lifetime number of heterosexual sex partners to assess reliability and consistency. Differences by gender and age were considered, and the effect of number of sexual partners. A variety of analytical methods were used to explore statistical challenges of these data including variance estimation, fractional polynomial transformations, and quantile regression. RESULTS: We found some level of discrepancy between reports of the number of sexual partners when assessed at different times is common, driven by those reporting a high number of partners who were disproportionately men. Men reported a higher lifetime number of partners than women at each age, and there were statistically significant differences by gender in (a) consistency between reports at different ages, and (b) reliability of reports as measured by both the Intraclass Correlation Coefficient and the Kappa statistic. CONCLUSIONS: When considering reliability, multiple statistical approaches are necessary or conclusions can be misleading. Variance components should be examined when considering the Intraclass Correlation Coefficient. When modelling, robust methods like fractional polynomials and quantile regression should be employed to accommodate nonlinearity. Sensitivity analyses excluding participants whose partner number is in the upper 5% to 25% are informative, as these were shown to have the highest discrepancies.
Authors: Krystal Cantos; Molly F Franke; Katherine Tassiopoulos; Paige L Williams; Anna-Barbara Moscicki; Georg R Seage Journal: AIDS Behav Date: 2021-04-24