Figure 2.

Results of statistical models on the number of malaria infections. A, We fit 3 count data models: a Poisson model, which assumes the same incidence parameter for all individuals; a negative binomial model, which allows for additional variability in the incidence but does not explicitly allow for the clustering of children within pairs; and a hierarchical Poisson model, which allows the incidence parameter to vary between pairs of twins but not within pairs. The y-axis represents the number of infections (black circles) during the follow-up of children, grouped in pairs (x-axis); red stars indicate children with hemoglobin AS genotype. For each child, posterior predictive distributions are presented for the 3 models discussed above; the different degrees of transparency of the colors represent different posterior intervals: 2.5–97.5, 25–75, 40–60. While the Poisson model (orange) does not fit the data well, data are consistent with both the negative binomial (green) and the hierarchical Poisson (purple) models. Of note, we also fit a Poisson model that included as covariate birth before versus after August 2013, when follow-up frequency changed (see Supplementary Figure 4). B, Estimated rate of infection, per month (x-axis; posterior median and 95% interval), based on the hierarchical Poisson model, presented for each of the 25 twin pairs (y-axis). Note that the rate in pair 5 is estimated to be relatively high, owing to infection detection during the relatively short follow-up of this pair. The ordering of this panel, from higher to lower y-axis coordinates, corresponds to the same ordering, from left to right, in A.