ΔNoise was calculated as the residuals of a LOESS regression of expression noise on median expression, using four different metrics of noise: (A,B) Noise strength, equal to SD2/median (C,D) SD, the standard deviation of expression among genetically identical cells (E,F) CV*, equal to SD/median and (G,H) LogCV*, equal to log2(SD/median). The Pearson’s Correlation Coefficient (PCC) between ΔNoise and ΔFitness was computed for 100 combinations of three parameters for (A,C,E,G) strains with a TDH3 expression level far from optimum and (B,D,F,H) strains with a TDH3 expression level close to optimum. The three parameters that were varied were (1) the smoothing parameter of the LOESS regression used to compute ΔNoise (2/6, 3/6, 4/6, 5/6 and 1), (2) the smoothing parameter of the LOESS regression used to compute ΔFitness (2/6, 3/6, 4/6, 5/6 and 1) and (3) the fitness threshold used to classify strains as far from optimum or close to optimum (0.0025, 0.005, 0.075, 0.01). For each plot, the five boxes represent the variation of PCC between ΔNoise and ΔFitness when parameter one is fixed to one of the values shown on the x-axis and the two other parameters are allowed to vary. The thick horizontal lines represent the median PCC across the 20 combinations of parameters 2 and 3. The bottom and top lines of each box represent the 25th and 75th percentiles. Parameter values used in the main figures are shown in red.