Fig. 6

Summary of additive (left column) and dominance (right column) effect sizes from association studies between inversion genotypes and morphological traits (40 estimates = 8 phenotypes × 5 inversions; top row) and of the additive and dominance effect sizes from associations between inversion genotypes and fitness traits (20 [16] estimates = 4 fitness parameters × 5 inversions [minus 4 TguZ dominance effects in females]; bottom row). Empirical effect sizes are shown as the light grey bars overlaid with the null distribution as a black line. Effects that survived strict Bonferroni correction are highlighted in yellow. Power for a given effect size is overlaid in purple with its corresponding y-axis on the right. We estimated the null distribution (and the power values) by permuting the inversion genotypes within sexes (and adding/subtracting the corresponding effect sizes to/from the phenotypic values) and fitting the same mixed models as for the empirical data set (see “Methods” for details). For illustration, the null distribution was scaled to overlap the first bar in the histogram of the empirical estimates completely. Partial regression coefficients of additive and dominance effects are not directly comparable the way we standardized and fitted them and thus their null distributions differ (dominance effects reach higher values than additive effects because their variance is smaller; see also [114, 115])