3D surfaces of simulation data showing the accuracy (<i>z</i> axis) in the estimated compared to the true subpopulation size using two different methods of population separation: <i>Boxplot1.5</i> and <i>Boxplot3</i>.

<p>Accuracy is shown as a function of different population mixtures (1600 per plot), with subpopulations either varying in mean differences (range: 2–137; n = 40; <i>x</i> axis) and proportions (range: 0.1–40%; n = 40; <i>y</i> axis) at a constant standard deviation (37.7) (A, B), or varying in mean differences (range: 2–137; n = 40; <i>x</i> axis) and standard deviations (range: 10–50; <i>y</i> axis) at a constant proportion (3.2%) (C, D), or with varying standard deviations (range: 10–50; n = 40; <i>x</i> axis) and proportions (range: 0.1–40%; n = 40; <i>y</i> axis) at a constant mean difference (67.8) (E, F). Accuracy is expressed as the percent difference between calculated and real subpopulation size, and therefore indicates the normalized deviance of the calculated subpopulation size from the real subpopulation size. A negative value indicates that the method underestimated the subpopulation size. A positive value indicates an overestimated result. A value of zero indicates absolute accuracy. A smooth surface of the same colour/grey-level indicates a robust separation. NA, missing values.