Figure 1. Determining how many cells to... | Rockefeller University Press

Figure 1.

$Determining how many cells to average to achieve reasonable statistical power. (A) Minimum effect detectible with 80% power as a function of averaging error when independent samples are analyzed with Student’s t test. Other errors were (in % CV, from bottom trace to top trace): 1 (black), 2.5 (orange), 5 (light blue), 10 (green), 15 (yellow), and 20 (dark blue). N was 3 replicates. (B) Minimum effect detectible with 80% power as a function of averaging error when samples in replicates share error of 20% CV and are analyzed with paired t tests. Other errors were (in % CV, from bottom trace to top trace): 1 (black), 2.5 (orange), 5 (light blue), and 10 (green). N was 3 replicates. (C) Top row: Plots of estimated within-sample SD as a function of actual within-sample SD when four cells from a single condition are used (left) or 16 cells from each of six replicates are used (right) for the estimate. When six replicates were used, three were simulated with mean of 100 (control) and three with mean of 125 (treated). Blue traces (circles and triangles) were simulated with normally distributed values and red (squares) with log-normally distributed values. Error bars are SD. Blue triangles represent data in which samples shared between-replicate variability of 20% CV, while blue circles represent data simulated without shared variability. The dashed line has an intercept of zero and slope of 1. Middle row: Estimates of the number of cells to average per sample to reach a target of 5% CV for samples that do not share variability (blue triangles for normal data, red squares for log-normal) or 2.5% CV for data simulated to share between-replicate variability (blue circles). Conditions were chosen from A and B to correspond to an effect detectible with 80% power of ∼25 % CV. Bottom row: Power (fraction of P values <0.05) under the conditions described above. The dashed line indicates the expected value of 0.8 for all conditions.$

Determining how many cells to average to achieve reasonable statistical power. (A) Minimum effect detectible with 80% power as a function of averaging error when independent samples are analyzed with Student’s t test. Other errors were (in % CV, from bottom trace to top trace): 1 (black), 2.5 (orange), 5 (light blue), 10 (green), 15 (yellow), and 20 (dark blue). N was 3 replicates. (B) Minimum effect detectible with 80% power as a function of averaging error when samples in replicates share error of 20% CV and are analyzed with paired t tests. Other errors were (in % CV, from bottom trace to top trace): 1 (black), 2.5 (orange), 5 (light blue), and 10 (green). N was 3 replicates. (C) Top row: Plots of estimated within-sample SD as a function of actual within-sample SD when four cells from a single condition are used (left) or 16 cells from each of six replicates are used (right) for the estimate. When six replicates were used, three were simulated with mean of 100 (control) and three with mean of 125 (treated). Blue traces (circles and triangles) were simulated with normally distributed values and red (squares) with log-normally distributed values. Error bars are SD. Blue triangles represent data in which samples shared between-replicate variability of 20% CV, while blue circles represent data simulated without shared variability. The dashed line has an intercept of zero and slope of 1. Middle row: Estimates of the number of cells to average per sample to reach a target of 5% CV for samples that do not share variability (blue triangles for normal data, red squares for log-normal) or 2.5% CV for data simulated to share between-replicate variability (blue circles). Conditions were chosen from A and B to correspond to an effect detectible with 80% power of ∼25 % CV. Bottom row: Power (fraction of P values <0.05) under the conditions described above. The dashed line indicates the expected value of 0.8 for all conditions.