Figure 1

Three directed acyclic graphs considered to be plausible models of the relationship of birth weight to infant mortality in response to a stressor (adapted from [6]). Model (a) assumes that the stressor has direct effects on birth weight and mortality, birth weight has direct effects on infant mortality, and an interaction of the stressor and birth weight are assumed to account for the reverse-J-shaped birth weight specific mortality curve. Model (b) also assumes that the stressor has direct effects on birth weight and mortality, birth weight has direct effects on infant mortality, and unobserved covariates U account for the reverse-J shape. Model (c) assumes that the stressor has direct effects on birth weight and mortality, the reverse-J shape is the result of unobserved covariates U, but birth weight does not have direct effects on mortality.