The midpoints are strongly correlated, and variability can be decomposed into a correlated and uncorrelated component. (A) Mean midpoints of inactivation μ_i plotted against mean midpoints of activation μ_a for the 150 experiments that reported both. The mean of all points (a mean-of-means) is indicated by a yellow star. A best-fit line is shown as a solid blue line, with its 95% confidence interval indicated by dashed blue lines and a grey shaded area. A second linear regression line with a slope constrained to have a gradient of one is shown in green. For one example experiment (μ_a = −58.2 mV, μ_i = −95.5 mV), we show the vector from the mean-of-means to this point, decomposed into components along the line of best fit (orange, first principal component) and perpendicular to the line of best fit (red, second principal component). The same example point is highlighted in black in panels B and C. (B) The square root of the experiment size as a function of the first principal component, for all points in A. The experiment size is defined as n_a + n_i, where n_a is the number of cells tested for V_a and n_i is the number tested for V_i. (C) The square root of the experiment size as a function of the second principal component.