Figure 3.

Tests of normalization methods on simulated data. The top panels plot the Zamar criterion, which is 0 for perfect inference of the true activity. The middle panels plot the 90th percentile ($P_{90}$) of the relative bias. The bottom panels plot the relative median absolute deviation, a measure of precision. (a) Performance of the methods plotted against standard deviation of the error in Renilla luminescence. Mean transfection efficiency: $\overline{t}=0.25$. True activity: $A=10$. Sample size: $N=10$. There were no outliers: ${\sigma }_{12}={\sigma }_{11}$. Firefly errors were scaled with activity: ${\sigma }_{21}={\sigma }_{22}=10{\sigma }_{11}$ (b). Performance plotted against increasing severity of outliers. The standard deviation of the contaminating Renilla errors ${\sigma }_{12}$ was varied to simulate outliers. True activity: $A=3$. Renilla errors: ${\sigma }_{11}=3$. Firefly errors: ${\sigma }_{21}=9$. The standard deviation of firefly contaminating errors was scaled with activity: ${\sigma }_{22}=3{\sigma }_{12}$. The other parameters are the same as in panel (a). The Zamar criterion of the ratiometric method was greater than the upper limit of the y-axis.