Fig 4. Inferences for variant fitness r and for the harmonic mean of effective population size $N_e^h\left(t_d\right)$, for the 750 datasets simulated with five virus variants.

(A) Correlation between true r_i (x-axis) and estimated ${\widehat{r}}_i$ (y-axis) (all variants considered together). (B) Correlation between true $N_e^h$ (x-axis) and estimated ${\widehat{N}}_e^h$ (y-axis) (all sampling dates considered together, logarithmic scale). In both panels, the black line is the first bisector and the red dashed line is the best-fitting linear model. In panel A, the 9 points with ${\widehat{r}}_i<0.7$ correspond to datasets in which a highly counterselected variant was observed in only a few plants (5, on average, of the 48 plants) due to an initial low effective population size.