menarche for high versus reference birth weight (data not shown), we further combined those two categories. All polytomous logistic regression models included race-ethnicity, participant’s birth decade, and childhood family income as confounders because they may influence the exposures and age at menarche. Models also included categorical interaction terms for race/ethnicity and birth decade to more fully adjust for confounding by these variables. We did not consider relative height or weight at age 10 as confounders because body size could be on a causal pathway between exposures and age at menarche, and neither factor could be a cause of earlier exposures. We conducted analyses using SAS V9.2 (SAS Institute Inc., Cary, NC).
