We conducted the analyses using HLM Version 6 (Raudenbush, Bryk, & Congdon, 2004). To evaluate the longitudinal trajectories, we first defined the Level 1 model and then tested possible Level 2 predictors. At Level 1, we fit a quadratic model for BMI because of the established non-linear changes in BMI across the lifespan (Drøyvold et al., 2006; Rissanen, Heliovaara, & Aromaa, 1988). At Level 2, we entered characteristics of the individual as independent variables to explain between-subjects variation in the intercept and the linear slope. Specifically, we tested sex, ethnicity, education, and the first assessment of personality as Level 2 predictors of both the intercept and linear slope. To account for secular trends, we also included the year of each participant’s first BLSA assessment, centered on the grand mean (1991), as a predictor of the intercept. Due to the number of statistical tests, we set p to .01. Finally, we centered age in decades on the grand mean (59.40 years) to minimize the correlation between the linear and quadratic terms. In the present research, the equations for the models were: Level1:BMI=π0+π1(Age)+π2(Age2)+eLevel2:π0=β00+β01(Sex)+β02(Ethnicity[Black])+β03(Ethnicity[Other])+β04(Education)+β05(Time of Measurement)+β06(Personality)+r0π1=β10+β11(Sex)+β12(Ethnicity[Black])+β13(Ethnicity[Other])+β14(Education)+β15(Personality)+r1π2=β20+r2
