The results are not homogeneous in these clusters – we find similar spread between intuitive, paradoxical and null correlation between quality of care and risk-adjusted outcomes.
As we were particularly interested in loci under selection between the clusters, we followed up with pairwise comparisons of the clusters (here only including the 'pure' contemporary samples) and non-hierarchical tests among the samples within each cluster (historical and contemporary separately).
To identify evolutionary relationships between these T6SS gene clusters, we followed the approach used by Bingle and coworkers to study phylogeny between T6SS loci [ 1].
At each step, it is necessary to have a definition of the distance between two clusters: we take this distance to be the mean distance of members of different clusters.
Because presence of phylogenetic signal, especially for CV1, seemed to be related to CHC differences between both clusters, we performed an analysis without the D. mojavensis cluster in order to test whether phylogenetic signal would be present in the D. buzzatii cluster only.
As similarity measure between individual objects (diseases/TFs) we used the Euclidean distance, and for distance between clusters we used the average (Euclidean) distance between all pairs of objects.
To limit the risk of contamination between clusters we used census data from Statistics South Africa to demarcate clusters with similar estimated population sizes and suitable borders or natural boundaries (such as roads, rivers, and hills).
To quantify the degree of intergradation between clusters, we classified any mosquito with >10% probability to belong to more than one cluster as an admixed individual (due to hybridization or shared ancestry).
To test whether genetic diversity was significantly different between clusters, we applied a Wilcoxon paired-sample test [ 71] on the 10 single locus values of the statistics of interest.
To test if ω differs significantly between types of clusters, we performed a Mann–Whitney test using R [ 77].
