“…To assess the robustness of results, two additional statistical approaches were used: fractional polynomials and binarizing age into a discrete variable. Fractional polynomials allow continuous independent variables, such as age, to be flexibly modeled in a non-linear manner without specifying a priori the relationship between the independent and dependent variable [24][25][26] . We used fractional polynomials to find the best-fitting model for age for each diffusivity metric (DTI-FA, TDF-FA, NODDI-ICVF, NODDI-ODI, NODDI-ISOVF) by testing one-and two-term curvilinear models for age using the following possible powers: -2, -1, -0.5, 0, 0.5, 1, 2 and 3, where x 0 corresponds to ln(x); these analyses also included our covariates of non-interest (educational attainment, socioeconomic status, waist-hip ratio, population structure) and participant sex.…”