Fig 2. Illustration of procedure to convert discrete hazard functions for hypothetical children to population-level age-specific trends.
(A) Discrete hazard functions are estimated for each hypothetical child from each mother in the target SBH dataset. Here, we color all children born in the same year with the same color. Only 3 years are shown for simplicity in this example. In real data, the years of birth of hypothetical children would vary by mother based on her age, such that there would be one hypothetical child for each year going back in time from the survey until the mother was 12 years old. (B) Probability of birth distributions is applied to each hypothetical birth from each mother. These are derived from the empirical map distributions from Rajaratnam and colleagues [9], in which a different probability is available by woman’s age, CEB, region of residence, and year prior to the survey. These probabilities are multiplied by each mother’s CEB and carried through to subsequent age bins to estimate the expected number of children entering each age bin (EEB) using estimated survival probabilities. As such, line thicknesses get slightly smaller with each subsequent age bin. The EEB value for each hypothetical child’s age bin represents the number of children entering that age bin that the hypothetical child represents for their given mother. (C) All hypothetical children to mothers are grouped by year of birth. The estimated mortality probabilities for each age bin from all hypothetical children born in the same year are pooled, and EEB is used to calculate a weighted mean. Trends are drawn across for each year, indicated here by a trend in the third age bin. This aggregation procedure can be done for any grouping of women to make estimates for a survey cluster, a district, or a whole country. CEB, children ever born; EEB, expected entering bin; NN, neonatal; PNN1, post-neonatal 1; PNN2, post-neonatal 2; SBH, summary birth history.