2.9 Statistical analysis

Significant differences for compartment-specific viral diversity (pairwise genetic distance) between participants were determined using a nonparametric pairwise multiple comparisons analysis based on rank sums (Dunn test package in R) with Bonferroni P-value correction following rejection of the D’Agostino Pearson test of normality. A P-value of ≤0.05 was considered significant. Statistical significance for Critchlow’s correlation coefficient (Slatkin 1989) was determined using a null distribution of permuted sequences (1,000 permutations). A P-value of ≥0.05 was considered significant. For the tip date randomization test, absence of overlap between 95 per cent high posterior density intervals of the mean evolutionary rate of the randomized and the correctly dated run indicated significant temporal signal (Supplementary Fig. S4). With regard to the BSSVS analysis, Bayes factors (BF) were calculated according to Lemey et al. (2009) for participant-specific transition rates between each compartment using a truncated Poisson prior (50% prior probability on the minimal rate configuration). Transition rates with BF >3 were considered to be well supported (Lemey et al. 2009). For the cFEL analysis, a P-value of ≤0.05 was considered significant.